Hi; here's a target-arm pullreq. Apologies for the size: this is because it has all RTH's work to enable emulation of SME2p1 and SVE2p1. thanks -- PMM The following changes since commit c77283dd5d79149f4e7e9edd00f65416c648ee59: Merge tag 'pull-request-2025-07-02' of https://gitlab.com/thuth/qemu into staging (2025-07-03 06:01:41 -0400) are available in the Git repository at: https://gitlab.com/pm215/qemu.git tags/pull-target-arm-20250704 for you to fetch changes up to 083fef73585dfa03f72055ace6de8dec4912d0b0: linux-user/aarch64: Set hwcap bits for SME2p1/SVE2p1 (2025-07-04 15:53:23 +0100) ---------------------------------------------------------------- target-arm queue: * Implement emulation of SME2p1 and SVE2p1 * Correctly enforce alignment checks for v8M loads and stores done via helper functions * Mark the "highbank" and the "midway" machine as deprecated ---------------------------------------------------------------- Peter Maydell (4): target/arm: Rename FMOPA_h to FMOPA_w_h target/arm: Rename BFMOPA to BFMOPA_w target/arm: Implement FMOPA (non-widening) for fp16 target/arm: Implement SME2 BFMOPA (non-widening) Richard Henderson (103): target/arm: Fix SME vs AdvSIMD exception priority target/arm: Fix sve_access_check for SME target/arm: Fix 128-bit element ZIP, UZP, TRN target/arm: Replace @rda_rn_rm_e0 in sve.decode target/arm: Fix FMMLA (64-bit element) for 128-bit VL target/arm: Disable FEAT_F64MM if maximum SVE vector size too small target/arm: Fix PSEL size operands to tcg_gen_gvec_ands target/arm: Fix f16_dotadd vs nan selection target/arm: Fix bfdotadd_ebf vs nan selection target/arm: Remove CPUARMState.vfp.scratch target/arm: Introduce FPST_ZA, FPST_ZA_F16 target/arm: Use FPST_ZA for sme_fmopa_[hsd] target/arm: Rename zarray to za_state.za target/arm: Add isar feature tests for SME2p1, SVE2p1 target/arm: Add ZT0 target/arm: Add zt0_excp_el to DisasContext target/arm: Implement SME2 ZERO ZT0 target/arm: Add alignment argument to gen_sve_{ldr, str} target/arm: Implement SME2 LDR/STR ZT0 target/arm: Implement SME2 MOVT target/arm: Split get_tile_rowcol argument tile_index target/arm: Rename MOVA for translate target/arm: Split out get_zarray target/arm: Introduce ARMCPU.sme_max_vq target/arm: Implement SME2 MOVA to/from tile, multiple registers target/arm: Implement SME2 MOVA to/from array, multiple registers target/arm: Implement SME2 BMOPA target/arm: Implement SME2 SMOPS, UMOPS (2-way) target/arm: Introduce gen_gvec_sve2_sqdmulh target/arm: Implement SME2 Multiple and Single SVE Destructive target/arm: Implement SME2 Multiple Vectors SVE Destructive target/arm: Implement SME2 ADD/SUB (array results, multiple and single vector) target/arm: Implement SME2 ADD/SUB (array results, multiple vectors) target/arm: Pass ZA to helper_sve2_fmlal_zz[zx]w_s target/arm: Add helper_gvec{_ah}_bfmlsl{_nx} target/arm: Implement SME2 FMLAL, BFMLAL target/arm: Implement SME2 FDOT target/arm: Implement SME2 BFDOT target/arm: Implement SME2 FVDOT, BFVDOT target/arm: Rename helper_gvec_*dot_[bh] to *_4[bh] target/arm: Implemement SME2 SDOT, UDOT, USDOT, SUDOT target/arm: Rename SVE SDOT and UDOT patterns target/arm: Tighten USDOT (vectors) decode target/arm: Implement SDOT, UDOT (2-way) for SME2/SVE2p1 target/arm: Implement SME2 SVDOT, UVDOT, SUVDOT, USVDOT target/arm: Implement SME2 SMLAL, SMLSL, UMLAL, UMLSL target/arm: Rename gvec_fml[as]_[hs] with _nf_ infix target/arm: Implement SME2 FMLA, FMLS target/arm: Implement SME2 BFMLA, BFMLS target/arm: Implement SME2 FADD, FSUB, BFADD, BFSUB target/arm: Implement SME2 ADD/SUB (array accumulator) target/arm: Implement SME2 BFCVT, BFCVTN, FCVT, FCVTN target/arm: Implement SME2 FCVT (widening), FCVTL target/arm: Implement SME2 FCVTZS, FCVTZU target/arm: Implement SME2 SCVTF, UCVTF target/arm: Implement SME2 FRINTN, FRINTP, FRINTM, FRINTA target/arm: Introduce do_[us]sat_[bhs] macros target/arm: Use do_[us]sat_[bhs] in sve_helper.c target/arm: Implement SME2 SQCVT, UQCVT, SQCVTU target/arm: Implement SQCVTN, UQCVTN, SQCVTUN for SME2/SVE2p1 target/arm: Implement SME2 SUNPK, UUNPK target/arm: Implement SME2 ZIP, UZP (four registers) target/arm: Move do_urshr, do_srshr to vec_internal.h target/arm: Implement SME2 SQRSHR, UQRSHR, SQRSHRN target/arm: Implement SME2 ZIP, UZP (two registers) target/arm: Implement SME2 FCLAMP, SCLAMP, UCLAMP target/arm: Enable SCLAMP, UCLAMP for SVE2p1 target/arm: Implement FCLAMP for SME2, SVE2p1 target/arm: Implement SME2p1 Multiple Zero target/arm: Introduce pred_count_test target/arm: Fold predtest_ones into helper_sve_brkns target/arm: Expand do_zero inline target/arm: Split out do_whilel from helper_sve_whilel target/arm: Split out do_whileg from helper_sve_whileg target/arm: Move scale by esz into helper_sve_while* target/arm: Split trans_WHILE to lt and gt target/arm: Enable PSEL for SVE2p1 target/arm: Implement SVE2p1 WHILE (predicate pair) target/arm: Implement SVE2p1 WHILE (predicate as counter) target/arm: Implement SVE2p1 PTRUE (predicate as counter) target/arm: Implement {ADD, SMIN, SMAX, UMIN, UMAX}QV for SVE2p1 target/arm: Implement SVE2p1 PEXT target/arm: Implement SME2 SEL target/arm: Implement ANDQV, ORQV, EORQV for SVE2p1 target/arm: Implement FADDQV, F{MIN, MAX}{NM}QV for SVE2p1 target/arm: Implement BFMLSLB{L, T} for SME2/SVE2p1 target/arm: Implement CNTP (predicate as counter) for SME2/SVE2p1 target/arm: Implement DUPQ for SME2p1/SVE2p1 target/arm: Implement EXTQ for SME2p1/SVE2p1 target/arm: Implement PMOV for SME2p1/SVE2p1 target/arm: Implement ZIPQ, UZPQ for SME2p1/SVE2p1 target/arm: Implement TBLQ, TBXQ for SME2p1/SVE2p1 target/arm: Implement SME2 counted predicate register load/store target/arm: Split the ST_zpri and ST_zprr patterns target/arm: Implement {LD1, ST1}{W, D} (128-bit element) for SVE2p1 target/arm: Move ld1qq and st1qq primitives to sve_ldst_internal.h target/arm: Implement {LD, ST}[234]Q for SME2p1/SVE2p1 target/arm: Implement LD1Q, ST1Q for SVE2p1 target/arm: Implement MOVAZ for SME2p1 target/arm: Implement LUTI2, LUTI4 for SME2/SME2p1 target/arm: Support FPCR.AH in SME FMOPS, BFMOPS target/arm: Enable FEAT_SME2p1 on -cpu max linux-user/aarch64: Set hwcap bits for SME2p1/SVE2p1 Thomas Huth (1): hw/arm/highbank: Mark the "highbank" and the "midway" machine as deprecated William Kosasih (11): target/arm: Bring VLSTM/VLLDM helper store/load closer to the ARM pseudocode target/arm: Fix BLXNS helper store alignment checks target/arm: Fix function_return helper load alignment checks target/arm: Fix VLDR helper load alignment checks target/arm: Fix VSTR helper store alignment checks target/arm: Fix VLDR_SG helper load alignment checks target/arm: Fix VSTR_SG helper store alignment checks target/arm: Fix VLD4 helper load alignment checks target/arm: Fix VLD2 helper load alignment checks target/arm: Fix VST4 helper store alignment checks target/arm: Fix VST2 helper store alignment checks docs/about/deprecated.rst | 7 + docs/system/arm/emulation.rst | 6 + target/arm/cpu-features.h | 63 ++ target/arm/cpu.h | 70 +- target/arm/syndrome.h | 1 + target/arm/tcg/helper-sme.h | 215 ++++- target/arm/tcg/helper-sve.h | 212 +++++ target/arm/tcg/helper.h | 91 +- target/arm/tcg/sve_ldst_internal.h | 89 ++ target/arm/tcg/translate-a64.h | 10 +- target/arm/tcg/translate.h | 9 + target/arm/tcg/vec_internal.h | 148 ++++ target/arm/tcg/sme.decode | 937 +++++++++++++++++++- target/arm/tcg/sve.decode | 327 +++++-- hw/arm/highbank.c | 2 + linux-user/aarch64/signal.c | 4 +- linux-user/elfload.c | 8 + target/arm/cpu.c | 11 +- target/arm/cpu64.c | 8 + target/arm/helper.c | 8 +- target/arm/machine.c | 22 +- target/arm/tcg/cpu64.c | 10 +- target/arm/tcg/gengvec64.c | 11 + target/arm/tcg/helper-a64.c | 2 + target/arm/tcg/hflags.c | 34 +- target/arm/tcg/m_helper.c | 33 +- target/arm/tcg/mve_helper.c | 183 ++-- target/arm/tcg/neon_helper.c | 30 + target/arm/tcg/sme_helper.c | 1674 +++++++++++++++++++++++++++++++++--- target/arm/tcg/sve_helper.c | 1201 +++++++++++++++++++++----- target/arm/tcg/translate-a64.c | 45 +- target/arm/tcg/translate-neon.c | 18 +- target/arm/tcg/translate-sme.c | 1480 ++++++++++++++++++++++++++++++- target/arm/tcg/translate-sve.c | 1019 +++++++++++++++++++--- target/arm/tcg/vec_helper.c | 384 +++++++-- target/arm/tcg/vfp_helper.c | 12 +- 36 files changed, 7605 insertions(+), 779 deletions(-) From: Thomas Huth <thuth@redhat.com> We don't have any automatic regression tests for these machines and when asking the usual suspects on the mailing list we came to the conclusion that nobody tests these machines manually, too, so it seems like this is currently just completely unused code. Mark them as depre- cated to see whether anybody still speaks up during the deprecation period, otherwise we can likely remove these two machines in a couple of releases. Signed-off-by: Thomas Huth <thuth@redhat.com> Acked-by: Rob Herring <robh@kernel.org> Acked-by: Guenter Roeck <linux@roeck-us.net> Message-id: 20250702113051.46483-1-thuth@redhat.com Reviewed-by: Peter Maydell <peter.maydell@linaro.org> [PMM: tweaked deprecation.rst text] Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- docs/about/deprecated.rst | 7 +++++++ hw/arm/highbank.c | 2 ++ 2 files changed, 9 insertions(+) diff --git a/docs/about/deprecated.rst b/docs/about/deprecated.rst index XXXXXXX..XXXXXXX 100644 --- a/docs/about/deprecated.rst +++ b/docs/about/deprecated.rst @@ -XXX,XX +XXX,XX @@ they want to use and avoids confusion. Existing users of the ``spike`` machine must ensure that they're setting the ``spike`` machine in the command line (``-M spike``). +Arm ``highbank`` and ``midway`` machines (since 10.1) +''''''''''''''''''''''''''''''''''''''''''''''''''''' + +There are no known users left for these machines (if you still use it, +please write a mail to the qemu-devel mailing list). If you just want to +boot a Cortex-A15 or Cortex-A9 Linux, use the ``virt`` machine instead. + System emulator binaries ------------------------ diff --git a/hw/arm/highbank.c b/hw/arm/highbank.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/highbank.c +++ b/hw/arm/highbank.c @@ -XXX,XX +XXX,XX @@ static void highbank_class_init(ObjectClass *oc, const void *data) mc->max_cpus = 4; mc->ignore_memory_transaction_failures = true; mc->default_ram_id = "highbank.dram"; + mc->deprecation_reason = "no known users left for this machine"; } static const TypeInfo highbank_type = { @@ -XXX,XX +XXX,XX @@ static void midway_class_init(ObjectClass *oc, const void *data) mc->max_cpus = 4; mc->ignore_memory_transaction_failures = true; mc->default_ram_id = "highbank.dram"; + mc->deprecation_reason = "no known users left for this machine"; } static const TypeInfo midway_type = { -- 2.43.0 From: William Kosasih <kosasihwilliam4@gmail.com> This patch brings the VLSTM and VLLDM helper functions closer to the ARM pseudocode by adding MO_ALIGN to the MemOpIdx of the associated store (`cpu_stl_mmu`) operations and load (`cpu_ldl_mmu`) operations. Note that this is not a bug fix: an 8-byte alignment check already exists and remains in place, enforcing stricter alignment than the 4 bytes requirement in the individual loads and stores. This change merely makes the helper implementations closer to the ARM pseudocode. That said, as a side effect, the MMU index is now resolved once instead of on every `cpu_*_data_ra` call, reducing redundant lookups Signed-off-by: William Kosasih <kosasihwilliam4@gmail.com> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250703085604.154449-2-kosasihwilliam4@gmail.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/m_helper.c | 24 +++++++++++++++--------- 1 file changed, 15 insertions(+), 9 deletions(-) diff --git a/target/arm/tcg/m_helper.c b/target/arm/tcg/m_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/m_helper.c +++ b/target/arm/tcg/m_helper.c @@ -XXX,XX +XXX,XX @@ void HELPER(v7m_vlstm)(CPUARMState *env, uint32_t fptr) bool s = env->v7m.fpccr[M_REG_S] & R_V7M_FPCCR_S_MASK; bool lspact = env->v7m.fpccr[s] & R_V7M_FPCCR_LSPACT_MASK; uintptr_t ra = GETPC(); + ARMMMUIdx mmu_idx = arm_mmu_idx(env); + MemOpIdx oi = make_memop_idx(MO_TEUL | MO_ALIGN, + arm_to_core_mmu_idx(mmu_idx)); assert(env->v7m.secure); @@ -XXX,XX +XXX,XX @@ void HELPER(v7m_vlstm)(CPUARMState *env, uint32_t fptr) * Note that we do not use v7m_stack_write() here, because the * accesses should not set the FSR bits for stacking errors if they * fail. (In pseudocode terms, they are AccType_NORMAL, not AccType_STACK - * or AccType_LAZYFP). Faults in cpu_stl_data_ra() will throw exceptions + * or AccType_LAZYFP). Faults in cpu_stl_mmu() will throw exceptions * and longjmp out. */ if (!(env->v7m.fpccr[M_REG_S] & R_V7M_FPCCR_LSPEN_MASK)) { @@ -XXX,XX +XXX,XX @@ void HELPER(v7m_vlstm)(CPUARMState *env, uint32_t fptr) if (i >= 16) { faddr += 8; /* skip the slot for the FPSCR */ } - cpu_stl_data_ra(env, faddr, slo, ra); - cpu_stl_data_ra(env, faddr + 4, shi, ra); + cpu_stl_mmu(env, faddr, slo, oi, ra); + cpu_stl_mmu(env, faddr + 4, shi, oi, ra); } - cpu_stl_data_ra(env, fptr + 0x40, vfp_get_fpscr(env), ra); + cpu_stl_mmu(env, fptr + 0x40, vfp_get_fpscr(env), oi, ra); if (cpu_isar_feature(aa32_mve, cpu)) { - cpu_stl_data_ra(env, fptr + 0x44, env->v7m.vpr, ra); + cpu_stl_mmu(env, fptr + 0x44, env->v7m.vpr, oi, ra); } /* @@ -XXX,XX +XXX,XX @@ void HELPER(v7m_vlldm)(CPUARMState *env, uint32_t fptr) { ARMCPU *cpu = env_archcpu(env); uintptr_t ra = GETPC(); + ARMMMUIdx mmu_idx = arm_mmu_idx(env); + MemOpIdx oi = make_memop_idx(MO_TEUL | MO_ALIGN, + arm_to_core_mmu_idx(mmu_idx)); /* fptr is the value of Rn, the frame pointer we load the FP regs from */ assert(env->v7m.secure); @@ -XXX,XX +XXX,XX @@ void HELPER(v7m_vlldm)(CPUARMState *env, uint32_t fptr) faddr += 8; /* skip the slot for the FPSCR and VPR */ } - slo = cpu_ldl_data_ra(env, faddr, ra); - shi = cpu_ldl_data_ra(env, faddr + 4, ra); + slo = cpu_ldl_mmu(env, faddr, oi, ra); + shi = cpu_ldl_mmu(env, faddr + 4, oi, ra); dn = (uint64_t) shi << 32 | slo; *aa32_vfp_dreg(env, i / 2) = dn; } - fpscr = cpu_ldl_data_ra(env, fptr + 0x40, ra); + fpscr = cpu_ldl_mmu(env, fptr + 0x40, oi, ra); vfp_set_fpscr(env, fpscr); if (cpu_isar_feature(aa32_mve, cpu)) { - env->v7m.vpr = cpu_ldl_data_ra(env, fptr + 0x44, ra); + env->v7m.vpr = cpu_ldl_mmu(env, fptr + 0x44, oi, ra); } } -- 2.43.0 From: William Kosasih <kosasihwilliam4@gmail.com> This patch adds alignment checks in the store operations (when stacking the return pc and psr) in the BLXNS instruction. Resolves: https://gitlab.com/qemu-project/qemu/-/issues/1154 Signed-off-by: William Kosasih <kosasihwilliam4@gmail.com> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250703085604.154449-3-kosasihwilliam4@gmail.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/m_helper.c | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-) diff --git a/target/arm/tcg/m_helper.c b/target/arm/tcg/m_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/m_helper.c +++ b/target/arm/tcg/m_helper.c @@ -XXX,XX +XXX,XX @@ void HELPER(v7m_blxns)(CPUARMState *env, uint32_t dest) } /* Note that these stores can throw exceptions on MPU faults */ - cpu_stl_data_ra(env, sp, nextinst, GETPC()); - cpu_stl_data_ra(env, sp + 4, saved_psr, GETPC()); + ARMMMUIdx mmu_idx = arm_mmu_idx(env); + MemOpIdx oi = make_memop_idx(MO_TEUL | MO_ALIGN, + arm_to_core_mmu_idx(mmu_idx)); + cpu_stl_mmu(env, sp, nextinst, oi, GETPC()); + cpu_stl_mmu(env, sp + 4, saved_psr, oi, GETPC()); env->regs[13] = sp; env->regs[14] = 0xfeffffff; -- 2.43.0 From: William Kosasih <kosasihwilliam4@gmail.com> This patch adds alignment checks in the load operations (when unstacking the return pc and psr) in the FunctionReturn pseudocode. Resolves: https://gitlab.com/qemu-project/qemu/-/issues/1154 Signed-off-by: William Kosasih <kosasihwilliam4@gmail.com> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250703085604.154449-4-kosasihwilliam4@gmail.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/m_helper.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/target/arm/tcg/m_helper.c b/target/arm/tcg/m_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/m_helper.c +++ b/target/arm/tcg/m_helper.c @@ -XXX,XX +XXX,XX @@ static bool do_v7m_function_return(ARMCPU *cpu) * do them as secure, so work out what MMU index that is. */ mmu_idx = arm_v7m_mmu_idx_for_secstate(env, true); - oi = make_memop_idx(MO_LEUL, arm_to_core_mmu_idx(mmu_idx)); + oi = make_memop_idx(MO_LEUL | MO_ALIGN, arm_to_core_mmu_idx(mmu_idx)); newpc = cpu_ldl_mmu(env, frameptr, oi, 0); newpsr = cpu_ldl_mmu(env, frameptr + 4, oi, 0); -- 2.43.0 From: William Kosasih <kosasihwilliam4@gmail.com> This patch adds alignment checks in the load operations in the VLDR instruction. Resolves: https://gitlab.com/qemu-project/qemu/-/issues/1154 Signed-off-by: William Kosasih <kosasihwilliam4@gmail.com> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250703085604.154449-5-kosasihwilliam4@gmail.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/mve_helper.c | 24 +++++++++++++----------- 1 file changed, 13 insertions(+), 11 deletions(-) diff --git a/target/arm/tcg/mve_helper.c b/target/arm/tcg/mve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/mve_helper.c +++ b/target/arm/tcg/mve_helper.c @@ -XXX,XX +XXX,XX @@ static void mve_advance_vpt(CPUARMState *env) } /* For loads, predicated lanes are zeroed instead of keeping their old values */ -#define DO_VLDR(OP, MSIZE, LDTYPE, ESIZE, TYPE) \ +#define DO_VLDR(OP, MFLAG, MSIZE, MTYPE, LDTYPE, ESIZE, TYPE) \ void HELPER(mve_##OP)(CPUARMState *env, void *vd, uint32_t addr) \ { \ TYPE *d = vd; \ uint16_t mask = mve_element_mask(env); \ uint16_t eci_mask = mve_eci_mask(env); \ unsigned b, e; \ + int mmu_idx = arm_to_core_mmu_idx(arm_mmu_idx(env)); \ + MemOpIdx oi = make_memop_idx(MFLAG | MO_ALIGN, mmu_idx); \ /* \ * R_SXTM allows the dest reg to become UNKNOWN for abandoned \ * beats so we don't care if we update part of the dest and \ @@ -XXX,XX +XXX,XX @@ static void mve_advance_vpt(CPUARMState *env) for (b = 0, e = 0; b < 16; b += ESIZE, e++) { \ if (eci_mask & (1 << b)) { \ d[H##ESIZE(e)] = (mask & (1 << b)) ? \ - cpu_##LDTYPE##_data_ra(env, addr, GETPC()) : 0; \ + (MTYPE)cpu_##LDTYPE##_mmu(env, addr, oi, GETPC()) : 0;\ } \ addr += MSIZE; \ } \ @@ -XXX,XX +XXX,XX @@ static void mve_advance_vpt(CPUARMState *env) mve_advance_vpt(env); \ } -DO_VLDR(vldrb, 1, ldub, 1, uint8_t) -DO_VLDR(vldrh, 2, lduw, 2, uint16_t) -DO_VLDR(vldrw, 4, ldl, 4, uint32_t) +DO_VLDR(vldrb, MO_UB, 1, uint8_t, ldb, 1, uint8_t) +DO_VLDR(vldrh, MO_TEUW, 2, uint16_t, ldw, 2, uint16_t) +DO_VLDR(vldrw, MO_TEUL, 4, uint32_t, ldl, 4, uint32_t) DO_VSTR(vstrb, 1, stb, 1, uint8_t) DO_VSTR(vstrh, 2, stw, 2, uint16_t) DO_VSTR(vstrw, 4, stl, 4, uint32_t) -DO_VLDR(vldrb_sh, 1, ldsb, 2, int16_t) -DO_VLDR(vldrb_sw, 1, ldsb, 4, int32_t) -DO_VLDR(vldrb_uh, 1, ldub, 2, uint16_t) -DO_VLDR(vldrb_uw, 1, ldub, 4, uint32_t) -DO_VLDR(vldrh_sw, 2, ldsw, 4, int32_t) -DO_VLDR(vldrh_uw, 2, lduw, 4, uint32_t) +DO_VLDR(vldrb_sh, MO_SB, 1, int8_t, ldb, 2, int16_t) +DO_VLDR(vldrb_sw, MO_SB, 1, int8_t, ldb, 4, int32_t) +DO_VLDR(vldrb_uh, MO_UB, 1, uint8_t, ldb, 2, uint16_t) +DO_VLDR(vldrb_uw, MO_UB, 1, uint8_t, ldb, 4, uint32_t) +DO_VLDR(vldrh_sw, MO_TESW, 2, int16_t, ldw, 4, int32_t) +DO_VLDR(vldrh_uw, MO_TEUW, 2, uint16_t, ldw, 4, uint32_t) DO_VSTR(vstrb_h, 1, stb, 2, int16_t) DO_VSTR(vstrb_w, 1, stb, 4, int32_t) -- 2.43.0 From: William Kosasih <kosasihwilliam4@gmail.com> This patch adds alignment checks in the store operations in the VSTR instruction. Resolves: https://gitlab.com/qemu-project/qemu/-/issues/1154 Signed-off-by: William Kosasih <kosasihwilliam4@gmail.com> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250703085604.154449-6-kosasihwilliam4@gmail.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/mve_helper.c | 18 ++++++++++-------- 1 file changed, 10 insertions(+), 8 deletions(-) diff --git a/target/arm/tcg/mve_helper.c b/target/arm/tcg/mve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/mve_helper.c +++ b/target/arm/tcg/mve_helper.c @@ -XXX,XX +XXX,XX @@ static void mve_advance_vpt(CPUARMState *env) mve_advance_vpt(env); \ } -#define DO_VSTR(OP, MSIZE, STTYPE, ESIZE, TYPE) \ +#define DO_VSTR(OP, MFLAG, MSIZE, STTYPE, ESIZE, TYPE) \ void HELPER(mve_##OP)(CPUARMState *env, void *vd, uint32_t addr) \ { \ TYPE *d = vd; \ uint16_t mask = mve_element_mask(env); \ unsigned b, e; \ + int mmu_idx = arm_to_core_mmu_idx(arm_mmu_idx(env)); \ + MemOpIdx oi = make_memop_idx(MFLAG | MO_ALIGN, mmu_idx); \ for (b = 0, e = 0; b < 16; b += ESIZE, e++) { \ if (mask & (1 << b)) { \ - cpu_##STTYPE##_data_ra(env, addr, d[H##ESIZE(e)], GETPC()); \ + cpu_##STTYPE##_mmu(env, addr, d[H##ESIZE(e)], oi, GETPC()); \ } \ addr += MSIZE; \ } \ @@ -XXX,XX +XXX,XX @@ DO_VLDR(vldrb, MO_UB, 1, uint8_t, ldb, 1, uint8_t) DO_VLDR(vldrh, MO_TEUW, 2, uint16_t, ldw, 2, uint16_t) DO_VLDR(vldrw, MO_TEUL, 4, uint32_t, ldl, 4, uint32_t) -DO_VSTR(vstrb, 1, stb, 1, uint8_t) -DO_VSTR(vstrh, 2, stw, 2, uint16_t) -DO_VSTR(vstrw, 4, stl, 4, uint32_t) +DO_VSTR(vstrb, MO_UB, 1, stb, 1, uint8_t) +DO_VSTR(vstrh, MO_TEUW, 2, stw, 2, uint16_t) +DO_VSTR(vstrw, MO_TEUL, 4, stl, 4, uint32_t) DO_VLDR(vldrb_sh, MO_SB, 1, int8_t, ldb, 2, int16_t) DO_VLDR(vldrb_sw, MO_SB, 1, int8_t, ldb, 4, int32_t) @@ -XXX,XX +XXX,XX @@ DO_VLDR(vldrb_uw, MO_UB, 1, uint8_t, ldb, 4, uint32_t) DO_VLDR(vldrh_sw, MO_TESW, 2, int16_t, ldw, 4, int32_t) DO_VLDR(vldrh_uw, MO_TEUW, 2, uint16_t, ldw, 4, uint32_t) -DO_VSTR(vstrb_h, 1, stb, 2, int16_t) -DO_VSTR(vstrb_w, 1, stb, 4, int32_t) -DO_VSTR(vstrh_w, 2, stw, 4, int32_t) +DO_VSTR(vstrb_h, MO_UB, 1, stb, 2, int16_t) +DO_VSTR(vstrb_w, MO_UB, 1, stb, 4, int32_t) +DO_VSTR(vstrh_w, MO_TEUW, 2, stw, 4, int32_t) #undef DO_VLDR #undef DO_VSTR -- 2.43.0 From: William Kosasih <kosasihwilliam4@gmail.com> This patch adds alignment checks in the load operations in the VLDR_SG instructions. Resolves: https://gitlab.com/qemu-project/qemu/-/issues/1154 Signed-off-by: William Kosasih <kosasihwilliam4@gmail.com> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250703085604.154449-7-kosasihwilliam4@gmail.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/mve_helper.c | 42 ++++++++++++++++++++++--------------- 1 file changed, 25 insertions(+), 17 deletions(-) diff --git a/target/arm/tcg/mve_helper.c b/target/arm/tcg/mve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/mve_helper.c +++ b/target/arm/tcg/mve_helper.c @@ -XXX,XX +XXX,XX @@ DO_VSTR(vstrh_w, MO_TEUW, 2, stw, 4, int32_t) * For loads, predicated lanes are zeroed instead of retaining * their previous values. */ -#define DO_VLDR_SG(OP, LDTYPE, ESIZE, TYPE, OFFTYPE, ADDRFN, WB) \ +#define DO_VLDR_SG(OP, MFLAG, MTYPE, LDTYPE, ESIZE, TYPE, OFFTYPE, ADDRFN, WB)\ void HELPER(mve_##OP)(CPUARMState *env, void *vd, void *vm, \ uint32_t base) \ { \ @@ -XXX,XX +XXX,XX @@ DO_VSTR(vstrh_w, MO_TEUW, 2, stw, 4, int32_t) uint16_t eci_mask = mve_eci_mask(env); \ unsigned e; \ uint32_t addr; \ + int mmu_idx = arm_to_core_mmu_idx(arm_mmu_idx(env)); \ + MemOpIdx oi = make_memop_idx(MFLAG | MO_ALIGN, mmu_idx); \ for (e = 0; e < 16 / ESIZE; e++, mask >>= ESIZE, eci_mask >>= ESIZE) { \ if (!(eci_mask & 1)) { \ continue; \ } \ addr = ADDRFN(base, m[H##ESIZE(e)]); \ d[H##ESIZE(e)] = (mask & 1) ? \ - cpu_##LDTYPE##_data_ra(env, addr, GETPC()) : 0; \ + (MTYPE)cpu_##LDTYPE##_mmu(env, addr, oi, GETPC()) : 0; \ if (WB) { \ m[H##ESIZE(e)] = addr; \ } \ @@ -XXX,XX +XXX,XX @@ DO_VSTR(vstrh_w, MO_TEUW, 2, stw, 4, int32_t) uint16_t eci_mask = mve_eci_mask(env); \ unsigned e; \ uint32_t addr; \ + int mmu_idx = arm_to_core_mmu_idx(arm_mmu_idx(env)); \ + MemOpIdx oi = make_memop_idx(MO_TEUL | MO_ALIGN, mmu_idx); \ for (e = 0; e < 16 / 4; e++, mask >>= 4, eci_mask >>= 4) { \ if (!(eci_mask & 1)) { \ continue; \ } \ addr = ADDRFN(base, m[H4(e & ~1)]); \ addr += 4 * (e & 1); \ - d[H4(e)] = (mask & 1) ? cpu_ldl_data_ra(env, addr, GETPC()) : 0; \ + d[H4(e)] = (mask & 1) ? cpu_ldl_mmu(env, addr, oi, GETPC()) : 0; \ if (WB && (e & 1)) { \ m[H4(e & ~1)] = addr - 4; \ } \ @@ -XXX,XX +XXX,XX @@ DO_VSTR(vstrh_w, MO_TEUW, 2, stw, 4, int32_t) #define ADDR_ADD_OSW(BASE, OFFSET) ((BASE) + ((OFFSET) << 2)) #define ADDR_ADD_OSD(BASE, OFFSET) ((BASE) + ((OFFSET) << 3)) -DO_VLDR_SG(vldrb_sg_sh, ldsb, 2, int16_t, uint16_t, ADDR_ADD, false) -DO_VLDR_SG(vldrb_sg_sw, ldsb, 4, int32_t, uint32_t, ADDR_ADD, false) -DO_VLDR_SG(vldrh_sg_sw, ldsw, 4, int32_t, uint32_t, ADDR_ADD, false) +DO_VLDR_SG(vldrb_sg_sh, MO_SB, int8_t, ldb, 2, int16_t, uint16_t, ADDR_ADD, false) +DO_VLDR_SG(vldrb_sg_sw, MO_SB, int8_t, ldb, 4, int32_t, uint32_t, ADDR_ADD, false) +DO_VLDR_SG(vldrh_sg_sw, MO_TESW, int16_t, ldw, 4, int32_t, uint32_t, ADDR_ADD, false) -DO_VLDR_SG(vldrb_sg_ub, ldub, 1, uint8_t, uint8_t, ADDR_ADD, false) -DO_VLDR_SG(vldrb_sg_uh, ldub, 2, uint16_t, uint16_t, ADDR_ADD, false) -DO_VLDR_SG(vldrb_sg_uw, ldub, 4, uint32_t, uint32_t, ADDR_ADD, false) -DO_VLDR_SG(vldrh_sg_uh, lduw, 2, uint16_t, uint16_t, ADDR_ADD, false) -DO_VLDR_SG(vldrh_sg_uw, lduw, 4, uint32_t, uint32_t, ADDR_ADD, false) -DO_VLDR_SG(vldrw_sg_uw, ldl, 4, uint32_t, uint32_t, ADDR_ADD, false) +DO_VLDR_SG(vldrb_sg_ub, MO_UB, uint8_t, ldb, 1, uint8_t, uint8_t, ADDR_ADD, false) +DO_VLDR_SG(vldrb_sg_uh, MO_UB, uint8_t, ldb, 2, uint16_t, uint16_t, ADDR_ADD, false) +DO_VLDR_SG(vldrb_sg_uw, MO_UB, uint8_t, ldb, 4, uint32_t, uint32_t, ADDR_ADD, false) +DO_VLDR_SG(vldrh_sg_uh, MO_TEUW, uint16_t, ldw, 2, uint16_t, uint16_t, ADDR_ADD, false) +DO_VLDR_SG(vldrh_sg_uw, MO_TEUW, uint16_t, ldw, 4, uint32_t, uint32_t, ADDR_ADD, false) +DO_VLDR_SG(vldrw_sg_uw, MO_TEUL, uint32_t, ldl, 4, uint32_t, uint32_t, ADDR_ADD, false) DO_VLDR64_SG(vldrd_sg_ud, ADDR_ADD, false) -DO_VLDR_SG(vldrh_sg_os_sw, ldsw, 4, int32_t, uint32_t, ADDR_ADD_OSH, false) -DO_VLDR_SG(vldrh_sg_os_uh, lduw, 2, uint16_t, uint16_t, ADDR_ADD_OSH, false) -DO_VLDR_SG(vldrh_sg_os_uw, lduw, 4, uint32_t, uint32_t, ADDR_ADD_OSH, false) -DO_VLDR_SG(vldrw_sg_os_uw, ldl, 4, uint32_t, uint32_t, ADDR_ADD_OSW, false) +DO_VLDR_SG(vldrh_sg_os_sw, MO_TESW, int16_t, ldw, 4, + int32_t, uint32_t, ADDR_ADD_OSH, false) +DO_VLDR_SG(vldrh_sg_os_uh, MO_TEUW, uint16_t, ldw, 2, + uint16_t, uint16_t, ADDR_ADD_OSH, false) +DO_VLDR_SG(vldrh_sg_os_uw, MO_TEUW, uint16_t, ldw, 4, + uint32_t, uint32_t, ADDR_ADD_OSH, false) +DO_VLDR_SG(vldrw_sg_os_uw, MO_TEUL, uint32_t, ldl, 4, + uint32_t, uint32_t, ADDR_ADD_OSW, false) DO_VLDR64_SG(vldrd_sg_os_ud, ADDR_ADD_OSD, false) DO_VSTR_SG(vstrb_sg_ub, stb, 1, uint8_t, ADDR_ADD, false) @@ -XXX,XX +XXX,XX @@ DO_VSTR_SG(vstrh_sg_os_uw, stw, 4, uint32_t, ADDR_ADD_OSH, false) DO_VSTR_SG(vstrw_sg_os_uw, stl, 4, uint32_t, ADDR_ADD_OSW, false) DO_VSTR64_SG(vstrd_sg_os_ud, ADDR_ADD_OSD, false) -DO_VLDR_SG(vldrw_sg_wb_uw, ldl, 4, uint32_t, uint32_t, ADDR_ADD, true) +DO_VLDR_SG(vldrw_sg_wb_uw, MO_TEUL, uint32_t, ldl, 4, uint32_t, uint32_t, ADDR_ADD, true) DO_VLDR64_SG(vldrd_sg_wb_ud, ADDR_ADD, true) DO_VSTR_SG(vstrw_sg_wb_uw, stl, 4, uint32_t, ADDR_ADD, true) DO_VSTR64_SG(vstrd_sg_wb_ud, ADDR_ADD, true) -- 2.43.0 From: William Kosasih <kosasihwilliam4@gmail.com> This patch adds alignment checks in the store operations in the VSTR_SG instructions. Resolves: https://gitlab.com/qemu-project/qemu/-/issues/1154 Signed-off-by: William Kosasih <kosasihwilliam4@gmail.com> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250703085604.154449-8-kosasihwilliam4@gmail.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/mve_helper.c | 30 +++++++++++++++++------------- 1 file changed, 17 insertions(+), 13 deletions(-) diff --git a/target/arm/tcg/mve_helper.c b/target/arm/tcg/mve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/mve_helper.c +++ b/target/arm/tcg/mve_helper.c @@ -XXX,XX +XXX,XX @@ DO_VSTR(vstrh_w, MO_TEUW, 2, stw, 4, int32_t) } /* We know here TYPE is unsigned so always the same as the offset type */ -#define DO_VSTR_SG(OP, STTYPE, ESIZE, TYPE, ADDRFN, WB) \ +#define DO_VSTR_SG(OP, MFLAG, STTYPE, ESIZE, TYPE, ADDRFN, WB) \ void HELPER(mve_##OP)(CPUARMState *env, void *vd, void *vm, \ uint32_t base) \ { \ @@ -XXX,XX +XXX,XX @@ DO_VSTR(vstrh_w, MO_TEUW, 2, stw, 4, int32_t) uint16_t eci_mask = mve_eci_mask(env); \ unsigned e; \ uint32_t addr; \ + int mmu_idx = arm_to_core_mmu_idx(arm_mmu_idx(env)); \ + MemOpIdx oi = make_memop_idx(MFLAG | MO_ALIGN, mmu_idx); \ for (e = 0; e < 16 / ESIZE; e++, mask >>= ESIZE, eci_mask >>= ESIZE) { \ if (!(eci_mask & 1)) { \ continue; \ } \ addr = ADDRFN(base, m[H##ESIZE(e)]); \ if (mask & 1) { \ - cpu_##STTYPE##_data_ra(env, addr, d[H##ESIZE(e)], GETPC()); \ + cpu_##STTYPE##_mmu(env, addr, d[H##ESIZE(e)], oi, GETPC()); \ } \ if (WB) { \ m[H##ESIZE(e)] = addr; \ @@ -XXX,XX +XXX,XX @@ DO_VSTR(vstrh_w, MO_TEUW, 2, stw, 4, int32_t) uint16_t eci_mask = mve_eci_mask(env); \ unsigned e; \ uint32_t addr; \ + int mmu_idx = arm_to_core_mmu_idx(arm_mmu_idx(env)); \ + MemOpIdx oi = make_memop_idx(MO_TEUL | MO_ALIGN, mmu_idx); \ for (e = 0; e < 16 / 4; e++, mask >>= 4, eci_mask >>= 4) { \ if (!(eci_mask & 1)) { \ continue; \ @@ -XXX,XX +XXX,XX @@ DO_VSTR(vstrh_w, MO_TEUW, 2, stw, 4, int32_t) addr = ADDRFN(base, m[H4(e & ~1)]); \ addr += 4 * (e & 1); \ if (mask & 1) { \ - cpu_stl_data_ra(env, addr, d[H4(e)], GETPC()); \ + cpu_stl_mmu(env, addr, d[H4(e)], oi, GETPC()); \ } \ if (WB && (e & 1)) { \ m[H4(e & ~1)] = addr - 4; \ @@ -XXX,XX +XXX,XX @@ DO_VLDR_SG(vldrw_sg_os_uw, MO_TEUL, uint32_t, ldl, 4, uint32_t, uint32_t, ADDR_ADD_OSW, false) DO_VLDR64_SG(vldrd_sg_os_ud, ADDR_ADD_OSD, false) -DO_VSTR_SG(vstrb_sg_ub, stb, 1, uint8_t, ADDR_ADD, false) -DO_VSTR_SG(vstrb_sg_uh, stb, 2, uint16_t, ADDR_ADD, false) -DO_VSTR_SG(vstrb_sg_uw, stb, 4, uint32_t, ADDR_ADD, false) -DO_VSTR_SG(vstrh_sg_uh, stw, 2, uint16_t, ADDR_ADD, false) -DO_VSTR_SG(vstrh_sg_uw, stw, 4, uint32_t, ADDR_ADD, false) -DO_VSTR_SG(vstrw_sg_uw, stl, 4, uint32_t, ADDR_ADD, false) +DO_VSTR_SG(vstrb_sg_ub, MO_UB, stb, 1, uint8_t, ADDR_ADD, false) +DO_VSTR_SG(vstrb_sg_uh, MO_UB, stb, 2, uint16_t, ADDR_ADD, false) +DO_VSTR_SG(vstrb_sg_uw, MO_UB, stb, 4, uint32_t, ADDR_ADD, false) +DO_VSTR_SG(vstrh_sg_uh, MO_TEUW, stw, 2, uint16_t, ADDR_ADD, false) +DO_VSTR_SG(vstrh_sg_uw, MO_TEUW, stw, 4, uint32_t, ADDR_ADD, false) +DO_VSTR_SG(vstrw_sg_uw, MO_TEUL, stl, 4, uint32_t, ADDR_ADD, false) DO_VSTR64_SG(vstrd_sg_ud, ADDR_ADD, false) -DO_VSTR_SG(vstrh_sg_os_uh, stw, 2, uint16_t, ADDR_ADD_OSH, false) -DO_VSTR_SG(vstrh_sg_os_uw, stw, 4, uint32_t, ADDR_ADD_OSH, false) -DO_VSTR_SG(vstrw_sg_os_uw, stl, 4, uint32_t, ADDR_ADD_OSW, false) +DO_VSTR_SG(vstrh_sg_os_uh, MO_TEUW, stw, 2, uint16_t, ADDR_ADD_OSH, false) +DO_VSTR_SG(vstrh_sg_os_uw, MO_TEUW, stw, 4, uint32_t, ADDR_ADD_OSH, false) +DO_VSTR_SG(vstrw_sg_os_uw, MO_TEUL, stl, 4, uint32_t, ADDR_ADD_OSW, false) DO_VSTR64_SG(vstrd_sg_os_ud, ADDR_ADD_OSD, false) DO_VLDR_SG(vldrw_sg_wb_uw, MO_TEUL, uint32_t, ldl, 4, uint32_t, uint32_t, ADDR_ADD, true) DO_VLDR64_SG(vldrd_sg_wb_ud, ADDR_ADD, true) -DO_VSTR_SG(vstrw_sg_wb_uw, stl, 4, uint32_t, ADDR_ADD, true) +DO_VSTR_SG(vstrw_sg_wb_uw, MO_TEUL, stl, 4, uint32_t, ADDR_ADD, true) DO_VSTR64_SG(vstrd_sg_wb_ud, ADDR_ADD, true) /* -- 2.43.0 From: William Kosasih <kosasihwilliam4@gmail.com> This patch adds alignment checks in the load operations in the VLD4 instruction. Resolves: https://gitlab.com/qemu-project/qemu/-/issues/1154 Signed-off-by: William Kosasih <kosasihwilliam4@gmail.com> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250703085604.154449-9-kosasihwilliam4@gmail.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/mve_helper.c | 12 +++++++++--- 1 file changed, 9 insertions(+), 3 deletions(-) diff --git a/target/arm/tcg/mve_helper.c b/target/arm/tcg/mve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/mve_helper.c +++ b/target/arm/tcg/mve_helper.c @@ -XXX,XX +XXX,XX @@ DO_VSTR64_SG(vstrd_sg_wb_ud, ADDR_ADD, true) uint16_t mask = mve_eci_mask(env); \ static const uint8_t off[4] = { O1, O2, O3, O4 }; \ uint32_t addr, data; \ + int mmu_idx = arm_to_core_mmu_idx(arm_mmu_idx(env)); \ + MemOpIdx oi = make_memop_idx(MO_TEUL | MO_ALIGN, mmu_idx); \ for (beat = 0; beat < 4; beat++, mask >>= 4) { \ if ((mask & 1) == 0) { \ /* ECI says skip this beat */ \ continue; \ } \ addr = base + off[beat] * 4; \ - data = cpu_ldl_le_data_ra(env, addr, GETPC()); \ + data = cpu_ldl_mmu(env, addr, oi, GETPC()); \ for (e = 0; e < 4; e++, data >>= 8) { \ uint8_t *qd = (uint8_t *)aa32_vfp_qreg(env, qnidx + e); \ qd[H1(off[beat])] = data; \ @@ -XXX,XX +XXX,XX @@ DO_VSTR64_SG(vstrd_sg_wb_ud, ADDR_ADD, true) uint32_t addr, data; \ int y; /* y counts 0 2 0 2 */ \ uint16_t *qd; \ + int mmu_idx = arm_to_core_mmu_idx(arm_mmu_idx(env)); \ + MemOpIdx oi = make_memop_idx(MO_TEUL | MO_ALIGN, mmu_idx); \ for (beat = 0, y = 0; beat < 4; beat++, mask >>= 4, y ^= 2) { \ if ((mask & 1) == 0) { \ /* ECI says skip this beat */ \ continue; \ } \ addr = base + off[beat] * 8 + (beat & 1) * 4; \ - data = cpu_ldl_le_data_ra(env, addr, GETPC()); \ + data = cpu_ldl_mmu(env, addr, oi, GETPC()); \ qd = (uint16_t *)aa32_vfp_qreg(env, qnidx + y); \ qd[H2(off[beat])] = data; \ data >>= 16; \ @@ -XXX,XX +XXX,XX @@ DO_VSTR64_SG(vstrd_sg_wb_ud, ADDR_ADD, true) uint32_t addr, data; \ uint32_t *qd; \ int y; \ + int mmu_idx = arm_to_core_mmu_idx(arm_mmu_idx(env)); \ + MemOpIdx oi = make_memop_idx(MO_TEUL | MO_ALIGN, mmu_idx); \ for (beat = 0; beat < 4; beat++, mask >>= 4) { \ if ((mask & 1) == 0) { \ /* ECI says skip this beat */ \ continue; \ } \ addr = base + off[beat] * 4; \ - data = cpu_ldl_le_data_ra(env, addr, GETPC()); \ + data = cpu_ldl_mmu(env, addr, oi, GETPC()); \ y = (beat + (O1 & 2)) & 3; \ qd = (uint32_t *)aa32_vfp_qreg(env, qnidx + y); \ qd[H4(off[beat] >> 2)] = data; \ -- 2.43.0 From: William Kosasih <kosasihwilliam4@gmail.com> This patch adds alignment checks in the load operations in the VLD2 instruction. Resolves: https://gitlab.com/qemu-project/qemu/-/issues/1154 Signed-off-by: William Kosasih <kosasihwilliam4@gmail.com> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250703085604.154449-10-kosasihwilliam4@gmail.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/mve_helper.c | 12 +++++++++--- 1 file changed, 9 insertions(+), 3 deletions(-) diff --git a/target/arm/tcg/mve_helper.c b/target/arm/tcg/mve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/mve_helper.c +++ b/target/arm/tcg/mve_helper.c @@ -XXX,XX +XXX,XX @@ DO_VLD4W(vld43w, 6, 7, 8, 9) static const uint8_t off[4] = { O1, O2, O3, O4 }; \ uint32_t addr, data; \ uint8_t *qd; \ + int mmu_idx = arm_to_core_mmu_idx(arm_mmu_idx(env)); \ + MemOpIdx oi = make_memop_idx(MO_TEUL | MO_ALIGN, mmu_idx); \ for (beat = 0; beat < 4; beat++, mask >>= 4) { \ if ((mask & 1) == 0) { \ /* ECI says skip this beat */ \ continue; \ } \ addr = base + off[beat] * 2; \ - data = cpu_ldl_le_data_ra(env, addr, GETPC()); \ + data = cpu_ldl_mmu(env, addr, oi, GETPC()); \ for (e = 0; e < 4; e++, data >>= 8) { \ qd = (uint8_t *)aa32_vfp_qreg(env, qnidx + (e & 1)); \ qd[H1(off[beat] + (e >> 1))] = data; \ @@ -XXX,XX +XXX,XX @@ DO_VLD4W(vld43w, 6, 7, 8, 9) uint32_t addr, data; \ int e; \ uint16_t *qd; \ + int mmu_idx = arm_to_core_mmu_idx(arm_mmu_idx(env)); \ + MemOpIdx oi = make_memop_idx(MO_TEUL | MO_ALIGN, mmu_idx); \ for (beat = 0; beat < 4; beat++, mask >>= 4) { \ if ((mask & 1) == 0) { \ /* ECI says skip this beat */ \ continue; \ } \ addr = base + off[beat] * 4; \ - data = cpu_ldl_le_data_ra(env, addr, GETPC()); \ + data = cpu_ldl_mmu(env, addr, oi, GETPC()); \ for (e = 0; e < 2; e++, data >>= 16) { \ qd = (uint16_t *)aa32_vfp_qreg(env, qnidx + e); \ qd[H2(off[beat])] = data; \ @@ -XXX,XX +XXX,XX @@ DO_VLD4W(vld43w, 6, 7, 8, 9) static const uint8_t off[4] = { O1, O2, O3, O4 }; \ uint32_t addr, data; \ uint32_t *qd; \ + int mmu_idx = arm_to_core_mmu_idx(arm_mmu_idx(env)); \ + MemOpIdx oi = make_memop_idx(MO_TEUL | MO_ALIGN, mmu_idx); \ for (beat = 0; beat < 4; beat++, mask >>= 4) { \ if ((mask & 1) == 0) { \ /* ECI says skip this beat */ \ continue; \ } \ addr = base + off[beat]; \ - data = cpu_ldl_le_data_ra(env, addr, GETPC()); \ + data = cpu_ldl_mmu(env, addr, oi, GETPC()); \ qd = (uint32_t *)aa32_vfp_qreg(env, qnidx + (beat & 1)); \ qd[H4(off[beat] >> 3)] = data; \ } \ -- 2.43.0 From: William Kosasih <kosasihwilliam4@gmail.com> This patch adds alignment checks in the store operations in the VST4 instruction. Resolves: https://gitlab.com/qemu-project/qemu/-/issues/1154 Signed-off-by: William Kosasih <kosasihwilliam4@gmail.com> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250703085604.154449-11-kosasihwilliam4@gmail.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/mve_helper.c | 12 +++++++++--- 1 file changed, 9 insertions(+), 3 deletions(-) diff --git a/target/arm/tcg/mve_helper.c b/target/arm/tcg/mve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/mve_helper.c +++ b/target/arm/tcg/mve_helper.c @@ -XXX,XX +XXX,XX @@ DO_VLD2W(vld21w, 8, 12, 16, 20) uint16_t mask = mve_eci_mask(env); \ static const uint8_t off[4] = { O1, O2, O3, O4 }; \ uint32_t addr, data; \ + int mmu_idx = arm_to_core_mmu_idx(arm_mmu_idx(env)); \ + MemOpIdx oi = make_memop_idx(MO_TEUL | MO_ALIGN, mmu_idx); \ for (beat = 0; beat < 4; beat++, mask >>= 4) { \ if ((mask & 1) == 0) { \ /* ECI says skip this beat */ \ @@ -XXX,XX +XXX,XX @@ DO_VLD2W(vld21w, 8, 12, 16, 20) uint8_t *qd = (uint8_t *)aa32_vfp_qreg(env, qnidx + e); \ data = (data << 8) | qd[H1(off[beat])]; \ } \ - cpu_stl_le_data_ra(env, addr, data, GETPC()); \ + cpu_stl_mmu(env, addr, data, oi, GETPC()); \ } \ } @@ -XXX,XX +XXX,XX @@ DO_VLD2W(vld21w, 8, 12, 16, 20) uint32_t addr, data; \ int y; /* y counts 0 2 0 2 */ \ uint16_t *qd; \ + int mmu_idx = arm_to_core_mmu_idx(arm_mmu_idx(env)); \ + MemOpIdx oi = make_memop_idx(MO_TEUL | MO_ALIGN, mmu_idx); \ for (beat = 0, y = 0; beat < 4; beat++, mask >>= 4, y ^= 2) { \ if ((mask & 1) == 0) { \ /* ECI says skip this beat */ \ @@ -XXX,XX +XXX,XX @@ DO_VLD2W(vld21w, 8, 12, 16, 20) data = qd[H2(off[beat])]; \ qd = (uint16_t *)aa32_vfp_qreg(env, qnidx + y + 1); \ data |= qd[H2(off[beat])] << 16; \ - cpu_stl_le_data_ra(env, addr, data, GETPC()); \ + cpu_stl_mmu(env, addr, data, oi, GETPC()); \ } \ } @@ -XXX,XX +XXX,XX @@ DO_VLD2W(vld21w, 8, 12, 16, 20) uint32_t addr, data; \ uint32_t *qd; \ int y; \ + int mmu_idx = arm_to_core_mmu_idx(arm_mmu_idx(env)); \ + MemOpIdx oi = make_memop_idx(MO_TEUL | MO_ALIGN, mmu_idx); \ for (beat = 0; beat < 4; beat++, mask >>= 4) { \ if ((mask & 1) == 0) { \ /* ECI says skip this beat */ \ @@ -XXX,XX +XXX,XX @@ DO_VLD2W(vld21w, 8, 12, 16, 20) y = (beat + (O1 & 2)) & 3; \ qd = (uint32_t *)aa32_vfp_qreg(env, qnidx + y); \ data = qd[H4(off[beat] >> 2)]; \ - cpu_stl_le_data_ra(env, addr, data, GETPC()); \ + cpu_stl_mmu(env, addr, data, oi, GETPC()); \ } \ } -- 2.43.0 From: William Kosasih <kosasihwilliam4@gmail.com> This patch adds alignment checks in the store operations in the VST2 instruction. Resolves: https://gitlab.com/qemu-project/qemu/-/issues/1154 Signed-off-by: William Kosasih <kosasihwilliam4@gmail.com> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250703085604.154449-12-kosasihwilliam4@gmail.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/mve_helper.c | 12 +++++++++--- 1 file changed, 9 insertions(+), 3 deletions(-) diff --git a/target/arm/tcg/mve_helper.c b/target/arm/tcg/mve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/mve_helper.c +++ b/target/arm/tcg/mve_helper.c @@ -XXX,XX +XXX,XX @@ DO_VST4W(vst43w, 6, 7, 8, 9) static const uint8_t off[4] = { O1, O2, O3, O4 }; \ uint32_t addr, data; \ uint8_t *qd; \ + int mmu_idx = arm_to_core_mmu_idx(arm_mmu_idx(env)); \ + MemOpIdx oi = make_memop_idx(MO_TEUL | MO_ALIGN, mmu_idx); \ for (beat = 0; beat < 4; beat++, mask >>= 4) { \ if ((mask & 1) == 0) { \ /* ECI says skip this beat */ \ @@ -XXX,XX +XXX,XX @@ DO_VST4W(vst43w, 6, 7, 8, 9) qd = (uint8_t *)aa32_vfp_qreg(env, qnidx + (e & 1)); \ data = (data << 8) | qd[H1(off[beat] + (e >> 1))]; \ } \ - cpu_stl_le_data_ra(env, addr, data, GETPC()); \ + cpu_stl_mmu(env, addr, data, oi, GETPC()); \ } \ } @@ -XXX,XX +XXX,XX @@ DO_VST4W(vst43w, 6, 7, 8, 9) uint32_t addr, data; \ int e; \ uint16_t *qd; \ + int mmu_idx = arm_to_core_mmu_idx(arm_mmu_idx(env)); \ + MemOpIdx oi = make_memop_idx(MO_TEUL | MO_ALIGN, mmu_idx); \ for (beat = 0; beat < 4; beat++, mask >>= 4) { \ if ((mask & 1) == 0) { \ /* ECI says skip this beat */ \ @@ -XXX,XX +XXX,XX @@ DO_VST4W(vst43w, 6, 7, 8, 9) qd = (uint16_t *)aa32_vfp_qreg(env, qnidx + e); \ data = (data << 16) | qd[H2(off[beat])]; \ } \ - cpu_stl_le_data_ra(env, addr, data, GETPC()); \ + cpu_stl_mmu(env, addr, data, oi, GETPC()); \ } \ } @@ -XXX,XX +XXX,XX @@ DO_VST4W(vst43w, 6, 7, 8, 9) static const uint8_t off[4] = { O1, O2, O3, O4 }; \ uint32_t addr, data; \ uint32_t *qd; \ + int mmu_idx = arm_to_core_mmu_idx(arm_mmu_idx(env)); \ + MemOpIdx oi = make_memop_idx(MO_TEUL | MO_ALIGN, mmu_idx); \ for (beat = 0; beat < 4; beat++, mask >>= 4) { \ if ((mask & 1) == 0) { \ /* ECI says skip this beat */ \ @@ -XXX,XX +XXX,XX @@ DO_VST4W(vst43w, 6, 7, 8, 9) addr = base + off[beat]; \ qd = (uint32_t *)aa32_vfp_qreg(env, qnidx + (beat & 1)); \ data = qd[H4(off[beat] >> 3)]; \ - cpu_stl_le_data_ra(env, addr, data, GETPC()); \ + cpu_stl_mmu(env, addr, data, oi, GETPC()); \ } \ } -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> We failed to raise an exception when sme_excp_el == 0 and fp_excp_el == 1. Cc: qemu-stable@nongnu.org Fixes: 3d74825f4d6 ("target/arm: Add SME enablement checks") Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-2-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/translate-a64.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/target/arm/tcg/translate-a64.c b/target/arm/tcg/translate-a64.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-a64.c +++ b/target/arm/tcg/translate-a64.c @@ -XXX,XX +XXX,XX @@ bool sme_enabled_check(DisasContext *s) * to be zero when fp_excp_el has priority. This is because we need * sme_excp_el by itself for cpregs access checks. */ - if (!s->fp_excp_el || s->sme_excp_el < s->fp_excp_el) { + if (s->sme_excp_el + && (!s->fp_excp_el || s->sme_excp_el <= s->fp_excp_el)) { bool ret = sme_access_check(s); s->fp_access_checked = (ret ? 1 : -1); return ret; -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Do not assume SME implies SVE. Ensure that the non-streaming check is present along the SME path, since it is not implied by sme_*_enabled_check. Cc: qemu-stable@nongnu.org Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-3-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/translate-a64.c | 26 +++++++++++++++++++------- 1 file changed, 19 insertions(+), 7 deletions(-) diff --git a/target/arm/tcg/translate-a64.c b/target/arm/tcg/translate-a64.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-a64.c +++ b/target/arm/tcg/translate-a64.c @@ -XXX,XX +XXX,XX @@ static bool fp_access_check_only(DisasContext *s) return true; } -static bool fp_access_check(DisasContext *s) +static bool nonstreaming_check(DisasContext *s) { - if (!fp_access_check_only(s)) { - return false; - } if (s->sme_trap_nonstreaming && s->is_nonstreaming) { gen_exception_insn(s, 0, EXCP_UDEF, syn_smetrap(SME_ET_Streaming, false)); @@ -XXX,XX +XXX,XX @@ static bool fp_access_check(DisasContext *s) return true; } +static bool fp_access_check(DisasContext *s) +{ + return fp_access_check_only(s) && nonstreaming_check(s); +} + /* * Return <0 for non-supported element sizes, with MO_16 controlled by * FEAT_FP16; return 0 for fp disabled; otherwise return >0 for success. @@ -XXX,XX +XXX,XX @@ static int fp_access_check_vector_hsd(DisasContext *s, bool is_q, MemOp esz) */ bool sve_access_check(DisasContext *s) { - if (s->pstate_sm || !dc_isar_feature(aa64_sve, s)) { + if (dc_isar_feature(aa64_sme, s)) { bool ret; - assert(dc_isar_feature(aa64_sme, s)); - ret = sme_sm_enabled_check(s); + if (s->pstate_sm) { + ret = sme_enabled_check(s); + } else if (dc_isar_feature(aa64_sve, s)) { + goto continue_sve; + } else { + ret = sme_sm_enabled_check(s); + } + if (ret) { + ret = nonstreaming_check(s); + } s->sve_access_checked = (ret ? 1 : -1); return ret; } + + continue_sve: if (s->sve_excp_el) { /* Assert that we only raise one exception per instruction. */ assert(!s->sve_access_checked); -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> We missed the instructions UDEF when the vector size is too small. We missed marking the instructions non-streaming with SME. Cc: qemu-stable@nongnu.org Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-4-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/translate-sve.c | 43 ++++++++++++++++++++++++---------- 1 file changed, 30 insertions(+), 13 deletions(-) diff --git a/target/arm/tcg/translate-sve.c b/target/arm/tcg/translate-sve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sve.c +++ b/target/arm/tcg/translate-sve.c @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(PUNPKHI, aa64_sve, do_perm_pred2, a, 1, gen_helper_sve_punpk_p) *** SVE Permute - Interleaving Group */ +static bool do_interleave_q(DisasContext *s, gen_helper_gvec_3 *fn, + arg_rrr_esz *a, int data) +{ + if (sve_access_check(s)) { + unsigned vsz = vec_full_reg_size(s); + if (vsz < 32) { + unallocated_encoding(s); + } else { + tcg_gen_gvec_3_ool(vec_full_reg_offset(s, a->rd), + vec_full_reg_offset(s, a->rn), + vec_full_reg_offset(s, a->rm), + vsz, vsz, data, fn); + } + } + return true; +} + static gen_helper_gvec_3 * const zip_fns[4] = { gen_helper_sve_zip_b, gen_helper_sve_zip_h, gen_helper_sve_zip_s, gen_helper_sve_zip_d, @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(ZIP1_z, aa64_sve, gen_gvec_ool_arg_zzz, TRANS_FEAT(ZIP2_z, aa64_sve, gen_gvec_ool_arg_zzz, zip_fns[a->esz], a, vec_full_reg_size(s) / 2) -TRANS_FEAT(ZIP1_q, aa64_sve_f64mm, gen_gvec_ool_arg_zzz, - gen_helper_sve2_zip_q, a, 0) -TRANS_FEAT(ZIP2_q, aa64_sve_f64mm, gen_gvec_ool_arg_zzz, - gen_helper_sve2_zip_q, a, - QEMU_ALIGN_DOWN(vec_full_reg_size(s), 32) / 2) +TRANS_FEAT_NONSTREAMING(ZIP1_q, aa64_sve_f64mm, do_interleave_q, + gen_helper_sve2_zip_q, a, 0) +TRANS_FEAT_NONSTREAMING(ZIP2_q, aa64_sve_f64mm, do_interleave_q, + gen_helper_sve2_zip_q, a, + QEMU_ALIGN_DOWN(vec_full_reg_size(s), 32) / 2) static gen_helper_gvec_3 * const uzp_fns[4] = { gen_helper_sve_uzp_b, gen_helper_sve_uzp_h, @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(UZP1_z, aa64_sve, gen_gvec_ool_arg_zzz, TRANS_FEAT(UZP2_z, aa64_sve, gen_gvec_ool_arg_zzz, uzp_fns[a->esz], a, 1 << a->esz) -TRANS_FEAT(UZP1_q, aa64_sve_f64mm, gen_gvec_ool_arg_zzz, - gen_helper_sve2_uzp_q, a, 0) -TRANS_FEAT(UZP2_q, aa64_sve_f64mm, gen_gvec_ool_arg_zzz, - gen_helper_sve2_uzp_q, a, 16) +TRANS_FEAT_NONSTREAMING(UZP1_q, aa64_sve_f64mm, do_interleave_q, + gen_helper_sve2_uzp_q, a, 0) +TRANS_FEAT_NONSTREAMING(UZP2_q, aa64_sve_f64mm, do_interleave_q, + gen_helper_sve2_uzp_q, a, 16) static gen_helper_gvec_3 * const trn_fns[4] = { gen_helper_sve_trn_b, gen_helper_sve_trn_h, @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(TRN1_z, aa64_sve, gen_gvec_ool_arg_zzz, TRANS_FEAT(TRN2_z, aa64_sve, gen_gvec_ool_arg_zzz, trn_fns[a->esz], a, 1 << a->esz) -TRANS_FEAT(TRN1_q, aa64_sve_f64mm, gen_gvec_ool_arg_zzz, - gen_helper_sve2_trn_q, a, 0) -TRANS_FEAT(TRN2_q, aa64_sve_f64mm, gen_gvec_ool_arg_zzz, - gen_helper_sve2_trn_q, a, 16) +TRANS_FEAT_NONSTREAMING(TRN1_q, aa64_sve_f64mm, do_interleave_q, + gen_helper_sve2_trn_q, a, 0) +TRANS_FEAT_NONSTREAMING(TRN2_q, aa64_sve_f64mm, do_interleave_q, + gen_helper_sve2_trn_q, a, 16) /* *** SVE Permute Vector - Predicated Group -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Replace @rda_rn_rm_e0 with @rda_rn_rm_ex, and require users to supply an explicit esz. Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-5-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/sve.decode | 48 +++++++++++++++++++-------------------- 1 file changed, 24 insertions(+), 24 deletions(-) diff --git a/target/arm/tcg/sve.decode b/target/arm/tcg/sve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve.decode +++ b/target/arm/tcg/sve.decode @@ -XXX,XX +XXX,XX @@ @rda_rn_rm ........ esz:2 . rm:5 ... ... rn:5 rd:5 \ &rrrr_esz ra=%reg_movprfx -# Four operand with unused vector element size -@rda_rn_rm_e0 ........ ... rm:5 ... ... rn:5 rd:5 \ - &rrrr_esz esz=0 ra=%reg_movprfx -@rdn_ra_rm_e0 ........ ... rm:5 ... ... ra:5 rd:5 \ - &rrrr_esz esz=0 rn=%reg_movprfx +# Four operand with explicit vector element size +@rda_rn_rm_ex ........ ... rm:5 ... ... rn:5 rd:5 \ + &rrrr_esz ra=%reg_movprfx +@rdn_ra_rm_ex ........ ... rm:5 ... ... ra:5 rd:5 \ + &rrrr_esz rn=%reg_movprfx # Three operand with "memory" size, aka immediate left shift @rd_rn_msz_rm ........ ... rm:5 .... imm:2 rn:5 rd:5 &rrri @@ -XXX,XX +XXX,XX @@ XAR 00000100 .. 1 ..... 001 101 rm:5 rd:5 &rrri_esz \ rn=%reg_movprfx esz=%tszimm16_esz imm=%tszimm16_shr # SVE2 bitwise ternary operations -EOR3 00000100 00 1 ..... 001 110 ..... ..... @rdn_ra_rm_e0 -BSL 00000100 00 1 ..... 001 111 ..... ..... @rdn_ra_rm_e0 -BCAX 00000100 01 1 ..... 001 110 ..... ..... @rdn_ra_rm_e0 -BSL1N 00000100 01 1 ..... 001 111 ..... ..... @rdn_ra_rm_e0 -BSL2N 00000100 10 1 ..... 001 111 ..... ..... @rdn_ra_rm_e0 -NBSL 00000100 11 1 ..... 001 111 ..... ..... @rdn_ra_rm_e0 +EOR3 00000100 00 1 ..... 001 110 ..... ..... @rdn_ra_rm_ex esz=0 +BSL 00000100 00 1 ..... 001 111 ..... ..... @rdn_ra_rm_ex esz=0 +BCAX 00000100 01 1 ..... 001 110 ..... ..... @rdn_ra_rm_ex esz=0 +BSL1N 00000100 01 1 ..... 001 111 ..... ..... @rdn_ra_rm_ex esz=0 +BSL2N 00000100 10 1 ..... 001 111 ..... ..... @rdn_ra_rm_ex esz=0 +NBSL 00000100 11 1 ..... 001 111 ..... ..... @rdn_ra_rm_ex esz=0 ### SVE Index Generation Group @@ -XXX,XX +XXX,XX @@ EORTB 01000101 .. 0 ..... 10010 1 ..... ..... @rd_rn_rm ## SVE integer matrix multiply accumulate -SMMLA 01000101 00 0 ..... 10011 0 ..... ..... @rda_rn_rm_e0 -USMMLA 01000101 10 0 ..... 10011 0 ..... ..... @rda_rn_rm_e0 -UMMLA 01000101 11 0 ..... 10011 0 ..... ..... @rda_rn_rm_e0 +SMMLA 01000101 00 0 ..... 10011 0 ..... ..... @rda_rn_rm_ex esz=2 +USMMLA 01000101 10 0 ..... 10011 0 ..... ..... @rda_rn_rm_ex esz=2 +UMMLA 01000101 11 0 ..... 10011 0 ..... ..... @rda_rn_rm_ex esz=2 ## SVE2 bitwise permute @@ -XXX,XX +XXX,XX @@ SQRDCMLAH_zzzz 01000100 esz:2 0 rm:5 0011 rot:2 rn:5 rd:5 ra=%reg_movprfx USDOT_zzzz 01000100 .. 0 ..... 011 110 ..... ..... @rda_rn_rm ### SVE2 floating point matrix multiply accumulate -BFMMLA 01100100 01 1 ..... 111 001 ..... ..... @rda_rn_rm_e0 -FMMLA_s 01100100 10 1 ..... 111 001 ..... ..... @rda_rn_rm_e0 -FMMLA_d 01100100 11 1 ..... 111 001 ..... ..... @rda_rn_rm_e0 +BFMMLA 01100100 01 1 ..... 111 001 ..... ..... @rda_rn_rm_ex esz=1 +FMMLA_s 01100100 10 1 ..... 111 001 ..... ..... @rda_rn_rm_ex esz=2 +FMMLA_d 01100100 11 1 ..... 111 001 ..... ..... @rda_rn_rm_ex esz=3 ### SVE2 Memory Gather Load Group @@ -XXX,XX +XXX,XX @@ FCVTLT_sd 01100100 11 0010 11 101 ... ..... ..... @rd_pg_rn_e0 FLOGB 01100101 00 011 esz:2 0101 pg:3 rn:5 rd:5 &rpr_esz ### SVE2 floating-point multiply-add long (vectors) -FMLALB_zzzw 01100100 10 1 ..... 10 0 00 0 ..... ..... @rda_rn_rm_e0 -FMLALT_zzzw 01100100 10 1 ..... 10 0 00 1 ..... ..... @rda_rn_rm_e0 -FMLSLB_zzzw 01100100 10 1 ..... 10 1 00 0 ..... ..... @rda_rn_rm_e0 -FMLSLT_zzzw 01100100 10 1 ..... 10 1 00 1 ..... ..... @rda_rn_rm_e0 +FMLALB_zzzw 01100100 10 1 ..... 10 0 00 0 ..... ..... @rda_rn_rm_ex esz=2 +FMLALT_zzzw 01100100 10 1 ..... 10 0 00 1 ..... ..... @rda_rn_rm_ex esz=2 +FMLSLB_zzzw 01100100 10 1 ..... 10 1 00 0 ..... ..... @rda_rn_rm_ex esz=2 +FMLSLT_zzzw 01100100 10 1 ..... 10 1 00 1 ..... ..... @rda_rn_rm_ex esz=2 -BFMLALB_zzzw 01100100 11 1 ..... 10 0 00 0 ..... ..... @rda_rn_rm_e0 -BFMLALT_zzzw 01100100 11 1 ..... 10 0 00 1 ..... ..... @rda_rn_rm_e0 +BFMLALB_zzzw 01100100 11 1 ..... 10 0 00 0 ..... ..... @rda_rn_rm_ex esz=2 +BFMLALT_zzzw 01100100 11 1 ..... 10 0 00 1 ..... ..... @rda_rn_rm_ex esz=2 ### SVE2 floating-point bfloat16 dot-product -BFDOT_zzzz 01100100 01 1 ..... 10 0 00 0 ..... ..... @rda_rn_rm_e0 +BFDOT_zzzz 01100100 01 1 ..... 10 0 00 0 ..... ..... @rda_rn_rm_ex esz=2 ### SVE2 floating-point multiply-add long (indexed) FMLALB_zzxw 01100100 10 1 ..... 0100.0 ..... ..... @rrxr_3a esz=2 -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-6-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/translate-sve.c | 23 ++++++++++++++++------- 1 file changed, 16 insertions(+), 7 deletions(-) diff --git a/target/arm/tcg/translate-sve.c b/target/arm/tcg/translate-sve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sve.c +++ b/target/arm/tcg/translate-sve.c @@ -XXX,XX +XXX,XX @@ DO_ZPZZ_FP(FMINNMP, aa64_sve2, sve2_fminnmp_zpzz) DO_ZPZZ_FP(FMAXP, aa64_sve2, sve2_fmaxp_zpzz) DO_ZPZZ_FP(FMINP, aa64_sve2, sve2_fminp_zpzz) +static bool do_fmmla(DisasContext *s, arg_rrrr_esz *a, + gen_helper_gvec_4_ptr *fn) +{ + if (sve_access_check(s)) { + if (vec_full_reg_size(s) < 4 * memop_size(a->esz)) { + unallocated_encoding(s); + } else { + gen_gvec_fpst_zzzz(s, fn, a->rd, a->rn, a->rm, a->ra, 0, FPST_A64); + } + } + return true; +} + +TRANS_FEAT_NONSTREAMING(FMMLA_s, aa64_sve_f32mm, do_fmmla, a, gen_helper_fmmla_s) +TRANS_FEAT_NONSTREAMING(FMMLA_d, aa64_sve_f64mm, do_fmmla, a, gen_helper_fmmla_d) + /* * SVE Integer Multiply-Add (unpredicated) */ -TRANS_FEAT_NONSTREAMING(FMMLA_s, aa64_sve_f32mm, gen_gvec_fpst_zzzz, - gen_helper_fmmla_s, a->rd, a->rn, a->rm, a->ra, - 0, FPST_A64) -TRANS_FEAT_NONSTREAMING(FMMLA_d, aa64_sve_f64mm, gen_gvec_fpst_zzzz, - gen_helper_fmmla_d, a->rd, a->rn, a->rm, a->ra, - 0, FPST_A64) - static gen_helper_gvec_4 * const sqdmlal_zzzw_fns[] = { NULL, gen_helper_sve2_sqdmlal_zzzw_h, gen_helper_sve2_sqdmlal_zzzw_s, gen_helper_sve2_sqdmlal_zzzw_d, -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> All F64MM instructions operate on a 256-bit vector. If only 128-bit vectors is supported by the cpu, then the cpu cannot enable F64MM. Suggested-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-7-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/cpu64.c | 7 +++++++ 1 file changed, 7 insertions(+) diff --git a/target/arm/cpu64.c b/target/arm/cpu64.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/cpu64.c +++ b/target/arm/cpu64.c @@ -XXX,XX +XXX,XX @@ void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp) /* From now on sve_max_vq is the actual maximum supported length. */ cpu->sve_max_vq = max_vq; cpu->sve_vq.map = vq_map; + + /* FEAT_F64MM requires the existence of a 256-bit vector size. */ + if (max_vq < 2) { + uint64_t t = GET_IDREG(&cpu->isar, ID_AA64ZFR0); + t = FIELD_DP64(t, ID_AA64ZFR0, F64MM, 0); + SET_IDREG(&cpu->isar, ID_AA64ZFR0, t); + } } /* -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Gvec only operates on size 8 and multiples of 16. Predicates may be any multiple of 2. Round up the size using the appropriate function. Cc: qemu-stable@nongnu.org Fixes: 598ab0b24c0 ("target/arm: Implement PSEL") Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-8-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/translate-sve.c | 1 + 1 file changed, 1 insertion(+) diff --git a/target/arm/tcg/translate-sve.c b/target/arm/tcg/translate-sve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sve.c +++ b/target/arm/tcg/translate-sve.c @@ -XXX,XX +XXX,XX @@ static bool trans_PSEL(DisasContext *s, arg_psel *a) tcg_gen_neg_i64(tmp, tmp); /* Apply to either copy the source, or write zeros. */ + pl = size_for_gvec(pl); tcg_gen_gvec_ands(MO_64, pred_full_reg_offset(s, a->pd), pred_full_reg_offset(s, a->pn), tmp, pl, pl); return true; -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Implement FPProcessNaNs4 within f16_dotadd, rather than simply letting NaNs propagate through the function. Cc: qemu-stable@nongnu.org Fixes: 3916841ac75 ("target/arm: Implement FMOPA, FMOPS (widening)") Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-9-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/sme_helper.c | 62 +++++++++++++++++++++++++++---------- 1 file changed, 46 insertions(+), 16 deletions(-) diff --git a/target/arm/tcg/sme_helper.c b/target/arm/tcg/sme_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme_helper.c +++ b/target/arm/tcg/sme_helper.c @@ -XXX,XX +XXX,XX @@ static float32 f16_dotadd(float32 sum, uint32_t e1, uint32_t e2, * - we have pre-set-up copy of s_std which is set to round-to-odd, * for the multiply (see below) */ - float64 e1r = float16_to_float64(e1 & 0xffff, true, s_f16); - float64 e1c = float16_to_float64(e1 >> 16, true, s_f16); - float64 e2r = float16_to_float64(e2 & 0xffff, true, s_f16); - float64 e2c = float16_to_float64(e2 >> 16, true, s_f16); - float64 t64; + float16 h1r = e1 & 0xffff; + float16 h1c = e1 >> 16; + float16 h2r = e2 & 0xffff; + float16 h2c = e2 >> 16; float32 t32; - /* - * The ARM pseudocode function FPDot performs both multiplies - * and the add with a single rounding operation. Emulate this - * by performing the first multiply in round-to-odd, then doing - * the second multiply as fused multiply-add, and rounding to - * float32 all in one step. - */ - t64 = float64_mul(e1r, e2r, s_odd); - t64 = float64r32_muladd(e1c, e2c, t64, 0, s_std); + /* C.f. FPProcessNaNs4 */ + if (float16_is_any_nan(h1r) || float16_is_any_nan(h1c) || + float16_is_any_nan(h2r) || float16_is_any_nan(h2c)) { + float16 t16; - /* This conversion is exact, because we've already rounded. */ - t32 = float64_to_float32(t64, s_std); + if (float16_is_signaling_nan(h1r, s_f16)) { + t16 = h1r; + } else if (float16_is_signaling_nan(h1c, s_f16)) { + t16 = h1c; + } else if (float16_is_signaling_nan(h2r, s_f16)) { + t16 = h2r; + } else if (float16_is_signaling_nan(h2c, s_f16)) { + t16 = h2c; + } else if (float16_is_any_nan(h1r)) { + t16 = h1r; + } else if (float16_is_any_nan(h1c)) { + t16 = h1c; + } else if (float16_is_any_nan(h2r)) { + t16 = h2r; + } else { + t16 = h2c; + } + t32 = float16_to_float32(t16, true, s_f16); + } else { + float64 e1r = float16_to_float64(h1r, true, s_f16); + float64 e1c = float16_to_float64(h1c, true, s_f16); + float64 e2r = float16_to_float64(h2r, true, s_f16); + float64 e2c = float16_to_float64(h2c, true, s_f16); + float64 t64; + + /* + * The ARM pseudocode function FPDot performs both multiplies + * and the add with a single rounding operation. Emulate this + * by performing the first multiply in round-to-odd, then doing + * the second multiply as fused multiply-add, and rounding to + * float32 all in one step. + */ + t64 = float64_mul(e1r, e2r, s_odd); + t64 = float64r32_muladd(e1c, e2c, t64, 0, s_std); + + /* This conversion is exact, because we've already rounded. */ + t32 = float64_to_float32(t64, s_std); + } /* The final accumulation step is not fused. */ return float32_add(sum, t32, s_std); -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Implement FPProcessNaNs4 within bfdotadd_ebf, rather than simply letting NaNs propagate through the function. Cc: qemu-stable@nongnu.org Fixes: 0e1850182a1 ("target/arm: Implement FPCR.EBF=1 semantics for bfdotadd()") Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Message-id: 20250704142112.1018902-10-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/vec_helper.c | 75 ++++++++++++++++++++++++++----------- 1 file changed, 53 insertions(+), 22 deletions(-) diff --git a/target/arm/tcg/vec_helper.c b/target/arm/tcg/vec_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/vec_helper.c +++ b/target/arm/tcg/vec_helper.c @@ -XXX,XX +XXX,XX @@ float32 bfdotadd(float32 sum, uint32_t e1, uint32_t e2, float_status *fpst) float32 bfdotadd_ebf(float32 sum, uint32_t e1, uint32_t e2, float_status *fpst, float_status *fpst_odd) { - /* - * Compare f16_dotadd() in sme_helper.c, but here we have - * bfloat16 inputs. In particular that means that we do not - * want the FPCR.FZ16 flush semantics, so we use the normal - * float_status for the input handling here. - */ - float64 e1r = float32_to_float64(e1 << 16, fpst); - float64 e1c = float32_to_float64(e1 & 0xffff0000u, fpst); - float64 e2r = float32_to_float64(e2 << 16, fpst); - float64 e2c = float32_to_float64(e2 & 0xffff0000u, fpst); - float64 t64; + float32 s1r = e1 << 16; + float32 s1c = e1 & 0xffff0000u; + float32 s2r = e2 << 16; + float32 s2c = e2 & 0xffff0000u; float32 t32; - /* - * The ARM pseudocode function FPDot performs both multiplies - * and the add with a single rounding operation. Emulate this - * by performing the first multiply in round-to-odd, then doing - * the second multiply as fused multiply-add, and rounding to - * float32 all in one step. - */ - t64 = float64_mul(e1r, e2r, fpst_odd); - t64 = float64r32_muladd(e1c, e2c, t64, 0, fpst); + /* C.f. FPProcessNaNs4 */ + if (float32_is_any_nan(s1r) || float32_is_any_nan(s1c) || + float32_is_any_nan(s2r) || float32_is_any_nan(s2c)) { + if (float32_is_signaling_nan(s1r, fpst)) { + t32 = s1r; + } else if (float32_is_signaling_nan(s1c, fpst)) { + t32 = s1c; + } else if (float32_is_signaling_nan(s2r, fpst)) { + t32 = s2r; + } else if (float32_is_signaling_nan(s2c, fpst)) { + t32 = s2c; + } else if (float32_is_any_nan(s1r)) { + t32 = s1r; + } else if (float32_is_any_nan(s1c)) { + t32 = s1c; + } else if (float32_is_any_nan(s2r)) { + t32 = s2r; + } else { + t32 = s2c; + } + /* + * FPConvertNaN(FPProcessNaN(t32)) will be done as part + * of the final addition below. + */ + } else { + /* + * Compare f16_dotadd() in sme_helper.c, but here we have + * bfloat16 inputs. In particular that means that we do not + * want the FPCR.FZ16 flush semantics, so we use the normal + * float_status for the input handling here. + */ + float64 e1r = float32_to_float64(s1r, fpst); + float64 e1c = float32_to_float64(s1c, fpst); + float64 e2r = float32_to_float64(s2r, fpst); + float64 e2c = float32_to_float64(s2c, fpst); + float64 t64; - /* This conversion is exact, because we've already rounded. */ - t32 = float64_to_float32(t64, fpst); + /* + * The ARM pseudocode function FPDot performs both multiplies + * and the add with a single rounding operation. Emulate this + * by performing the first multiply in round-to-odd, then doing + * the second multiply as fused multiply-add, and rounding to + * float32 all in one step. + */ + t64 = float64_mul(e1r, e2r, fpst_odd); + t64 = float64r32_muladd(e1c, e2c, t64, 0, fpst); + + /* This conversion is exact, because we've already rounded. */ + t32 = float64_to_float32(t64, fpst); + } /* The final accumulation step is not fused. */ return float32_add(sum, t32, fpst); -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> The last use of this field was removed in b2fc7be972b9. Reviewed-by: Alex Bennée <alex.bennee@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-11-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/cpu.h | 3 --- 1 file changed, 3 deletions(-) diff --git a/target/arm/cpu.h b/target/arm/cpu.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/cpu.h +++ b/target/arm/cpu.h @@ -XXX,XX +XXX,XX @@ typedef struct CPUArchState { uint32_t xregs[16]; - /* Scratch space for aa32 neon expansion. */ - uint32_t scratch[8]; - /* There are a number of distinct float control structures. */ float_status fp_status[FPST_COUNT]; -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Rather than repeatedly copying FPST_FPCR to locals and setting default nan mode, create dedicated float_status. Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-12-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/cpu.h | 12 +++++++++++- target/arm/cpu.c | 4 ++++ target/arm/tcg/vfp_helper.c | 12 +++++++++++- 3 files changed, 26 insertions(+), 2 deletions(-) diff --git a/target/arm/cpu.h b/target/arm/cpu.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/cpu.h +++ b/target/arm/cpu.h @@ -XXX,XX +XXX,XX @@ typedef struct NVICState NVICState; * when FPCR.AH == 1 (bfloat16 conversions and multiplies, * and the reciprocal and square root estimate/step insns); * for half-precision + * ZA: the "streaming sve" fp status. + * ZA_F16: likewise for half-precision. * * Half-precision operations are governed by a separate * flush-to-zero control bit in FPSCR:FZ16. We pass a separate @@ -XXX,XX +XXX,XX @@ typedef struct NVICState NVICState; * they ignore FPCR.RMode. But they don't ignore FPCR.FZ16, * which means we need an FPST_AH_F16 as well. * + * The "ZA" float_status are for Streaming SVE operations which use + * default-NaN and do not generate fp exceptions, which means that they + * do not accumulate exception bits back into FPCR. + * See e.g. FPAdd vs FPAdd_ZA pseudocode functions, and the setting + * of fpcr.DN and fpexec parameters. + * * To avoid having to transfer exception bits around, we simply * say that the FPSCR cumulative exception flags are the logical * OR of the flags in the four fp statuses. This relies on the @@ -XXX,XX +XXX,XX @@ typedef enum ARMFPStatusFlavour { FPST_A64_F16, FPST_AH, FPST_AH_F16, + FPST_ZA, + FPST_ZA_F16, FPST_STD, FPST_STD_F16, } ARMFPStatusFlavour; -#define FPST_COUNT 8 +#define FPST_COUNT 10 typedef struct CPUArchState { /* Regs for current mode. */ diff --git a/target/arm/cpu.c b/target/arm/cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/cpu.c +++ b/target/arm/cpu.c @@ -XXX,XX +XXX,XX @@ static void arm_cpu_reset_hold(Object *obj, ResetType type) set_flush_inputs_to_zero(1, &env->vfp.fp_status[FPST_STD]); set_default_nan_mode(1, &env->vfp.fp_status[FPST_STD]); set_default_nan_mode(1, &env->vfp.fp_status[FPST_STD_F16]); + set_default_nan_mode(1, &env->vfp.fp_status[FPST_ZA]); + set_default_nan_mode(1, &env->vfp.fp_status[FPST_ZA_F16]); arm_set_default_fp_behaviours(&env->vfp.fp_status[FPST_A32]); arm_set_default_fp_behaviours(&env->vfp.fp_status[FPST_A64]); + arm_set_default_fp_behaviours(&env->vfp.fp_status[FPST_ZA]); arm_set_default_fp_behaviours(&env->vfp.fp_status[FPST_STD]); arm_set_default_fp_behaviours(&env->vfp.fp_status[FPST_A32_F16]); arm_set_default_fp_behaviours(&env->vfp.fp_status[FPST_A64_F16]); + arm_set_default_fp_behaviours(&env->vfp.fp_status[FPST_ZA_F16]); arm_set_default_fp_behaviours(&env->vfp.fp_status[FPST_STD_F16]); arm_set_ah_fp_behaviours(&env->vfp.fp_status[FPST_AH]); set_flush_to_zero(1, &env->vfp.fp_status[FPST_AH]); diff --git a/target/arm/tcg/vfp_helper.c b/target/arm/tcg/vfp_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/vfp_helper.c +++ b/target/arm/tcg/vfp_helper.c @@ -XXX,XX +XXX,XX @@ uint32_t vfp_get_fpsr_from_host(CPUARMState *env) a64_flags |= (get_float_exception_flags(&env->vfp.fp_status[FPST_A64_F16]) & ~(float_flag_input_denormal_flushed | float_flag_input_denormal_used)); /* - * We do not merge in flags from FPST_AH or FPST_AH_F16, because + * We do not merge in flags from FPST_{AH,ZA} or FPST_{AH,ZA}_F16, because * they are used for insns that must not set the cumulative exception bits. */ @@ -XXX,XX +XXX,XX @@ void vfp_set_fpcr_to_host(CPUARMState *env, uint32_t val, uint32_t mask) set_float_rounding_mode(i, &env->vfp.fp_status[FPST_A64]); set_float_rounding_mode(i, &env->vfp.fp_status[FPST_A32_F16]); set_float_rounding_mode(i, &env->vfp.fp_status[FPST_A64_F16]); + set_float_rounding_mode(i, &env->vfp.fp_status[FPST_ZA]); + set_float_rounding_mode(i, &env->vfp.fp_status[FPST_ZA_F16]); } if (changed & FPCR_FZ16) { bool ftz_enabled = val & FPCR_FZ16; @@ -XXX,XX +XXX,XX @@ void vfp_set_fpcr_to_host(CPUARMState *env, uint32_t val, uint32_t mask) set_flush_to_zero(ftz_enabled, &env->vfp.fp_status[FPST_A64_F16]); set_flush_to_zero(ftz_enabled, &env->vfp.fp_status[FPST_STD_F16]); set_flush_to_zero(ftz_enabled, &env->vfp.fp_status[FPST_AH_F16]); + set_flush_to_zero(ftz_enabled, &env->vfp.fp_status[FPST_ZA_F16]); set_flush_inputs_to_zero(ftz_enabled, &env->vfp.fp_status[FPST_A32_F16]); set_flush_inputs_to_zero(ftz_enabled, &env->vfp.fp_status[FPST_A64_F16]); set_flush_inputs_to_zero(ftz_enabled, &env->vfp.fp_status[FPST_STD_F16]); set_flush_inputs_to_zero(ftz_enabled, &env->vfp.fp_status[FPST_AH_F16]); + set_flush_inputs_to_zero(ftz_enabled, &env->vfp.fp_status[FPST_ZA_F16]); } if (changed & FPCR_FZ) { bool ftz_enabled = val & FPCR_FZ; set_flush_to_zero(ftz_enabled, &env->vfp.fp_status[FPST_A32]); set_flush_to_zero(ftz_enabled, &env->vfp.fp_status[FPST_A64]); + set_flush_to_zero(ftz_enabled, &env->vfp.fp_status[FPST_ZA]); /* FIZ is A64 only so FZ always makes A32 code flush inputs to zero */ set_flush_inputs_to_zero(ftz_enabled, &env->vfp.fp_status[FPST_A32]); } @@ -XXX,XX +XXX,XX @@ void vfp_set_fpcr_to_host(CPUARMState *env, uint32_t val, uint32_t mask) bool fitz_enabled = (val & FPCR_FIZ) || (val & (FPCR_FZ | FPCR_AH)) == FPCR_FZ; set_flush_inputs_to_zero(fitz_enabled, &env->vfp.fp_status[FPST_A64]); + set_flush_inputs_to_zero(fitz_enabled, &env->vfp.fp_status[FPST_ZA]); } if (changed & FPCR_DN) { bool dnan_enabled = val & FPCR_DN; @@ -XXX,XX +XXX,XX @@ void vfp_set_fpcr_to_host(CPUARMState *env, uint32_t val, uint32_t mask) /* Change behaviours for A64 FP operations */ arm_set_ah_fp_behaviours(&env->vfp.fp_status[FPST_A64]); arm_set_ah_fp_behaviours(&env->vfp.fp_status[FPST_A64_F16]); + arm_set_ah_fp_behaviours(&env->vfp.fp_status[FPST_ZA]); + arm_set_ah_fp_behaviours(&env->vfp.fp_status[FPST_ZA_F16]); } else { arm_set_default_fp_behaviours(&env->vfp.fp_status[FPST_A64]); arm_set_default_fp_behaviours(&env->vfp.fp_status[FPST_A64_F16]); + arm_set_default_fp_behaviours(&env->vfp.fp_status[FPST_ZA]); + arm_set_default_fp_behaviours(&env->vfp.fp_status[FPST_ZA_F16]); } } /* -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-13-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/sme_helper.c | 37 ++++++++-------------------------- target/arm/tcg/translate-sme.c | 4 ++-- 2 files changed, 10 insertions(+), 31 deletions(-) diff --git a/target/arm/tcg/sme_helper.c b/target/arm/tcg/sme_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme_helper.c +++ b/target/arm/tcg/sme_helper.c @@ -XXX,XX +XXX,XX @@ void HELPER(sme_addva_d)(void *vzda, void *vzn, void *vpn, } void HELPER(sme_fmopa_s)(void *vza, void *vzn, void *vzm, void *vpn, - void *vpm, float_status *fpst_in, uint32_t desc) + void *vpm, float_status *fpst, uint32_t desc) { intptr_t row, col, oprsz = simd_maxsz(desc); uint32_t neg = simd_data(desc) << 31; uint16_t *pn = vpn, *pm = vpm; - float_status fpst; - - /* - * Make a copy of float_status because this operation does not - * update the cumulative fp exception status. It also produces - * default nans. - */ - fpst = *fpst_in; - set_default_nan_mode(true, &fpst); for (row = 0; row < oprsz; ) { uint16_t pa = pn[H2(row >> 4)]; @@ -XXX,XX +XXX,XX @@ void HELPER(sme_fmopa_s)(void *vza, void *vzn, void *vzm, void *vpn, if (pb & 1) { uint32_t *a = vza_row + H1_4(col); uint32_t *m = vzm + H1_4(col); - *a = float32_muladd(n, *m, *a, 0, &fpst); + *a = float32_muladd(n, *m, *a, 0, fpst); } col += 4; pb >>= 4; @@ -XXX,XX +XXX,XX @@ void HELPER(sme_fmopa_s)(void *vza, void *vzn, void *vzm, void *vpn, } void HELPER(sme_fmopa_d)(void *vza, void *vzn, void *vzm, void *vpn, - void *vpm, float_status *fpst_in, uint32_t desc) + void *vpm, float_status *fpst, uint32_t desc) { intptr_t row, col, oprsz = simd_oprsz(desc) / 8; uint64_t neg = (uint64_t)simd_data(desc) << 63; uint64_t *za = vza, *zn = vzn, *zm = vzm; uint8_t *pn = vpn, *pm = vpm; - float_status fpst = *fpst_in; - - set_default_nan_mode(true, &fpst); for (row = 0; row < oprsz; ++row) { if (pn[H1(row)] & 1) { @@ -XXX,XX +XXX,XX @@ void HELPER(sme_fmopa_d)(void *vza, void *vzn, void *vzm, void *vpn, for (col = 0; col < oprsz; ++col) { if (pm[H1(col)] & 1) { uint64_t *a = &za_row[col]; - *a = float64_muladd(n, zm[col], *a, 0, &fpst); + *a = float64_muladd(n, zm[col], *a, 0, fpst); } } } @@ -XXX,XX +XXX,XX @@ void HELPER(sme_fmopa_h)(void *vza, void *vzn, void *vzm, void *vpn, intptr_t row, col, oprsz = simd_maxsz(desc); uint32_t neg = simd_data(desc) * 0x80008000u; uint16_t *pn = vpn, *pm = vpm; - float_status fpst_odd, fpst_std, fpst_f16; + float_status fpst_odd = env->vfp.fp_status[FPST_ZA]; - /* - * Make copies of the fp status fields we use, because this operation - * does not update the cumulative fp exception status. It also - * produces default NaNs. We also need a second copy of fp_status with - * round-to-odd -- see above. - */ - fpst_f16 = env->vfp.fp_status[FPST_A64_F16]; - fpst_std = env->vfp.fp_status[FPST_A64]; - set_default_nan_mode(true, &fpst_std); - set_default_nan_mode(true, &fpst_f16); - fpst_odd = fpst_std; set_float_rounding_mode(float_round_to_odd, &fpst_odd); for (row = 0; row < oprsz; ) { @@ -XXX,XX +XXX,XX @@ void HELPER(sme_fmopa_h)(void *vza, void *vzn, void *vzm, void *vpn, m = f16mop_adj_pair(m, pcol, 0); *a = f16_dotadd(*a, n, m, - &fpst_f16, &fpst_std, &fpst_odd); + &env->vfp.fp_status[FPST_ZA_F16], + &env->vfp.fp_status[FPST_ZA], + &fpst_odd); } col += 4; pcol >>= 4; diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ static bool do_outprod_env(DisasContext *s, arg_op *a, MemOp esz, TRANS_FEAT(FMOPA_h, aa64_sme, do_outprod_env, a, MO_32, gen_helper_sme_fmopa_h) TRANS_FEAT(FMOPA_s, aa64_sme, do_outprod_fpst, a, - MO_32, FPST_A64, gen_helper_sme_fmopa_s) + MO_32, FPST_ZA, gen_helper_sme_fmopa_s) TRANS_FEAT(FMOPA_d, aa64_sme_f64f64, do_outprod_fpst, a, - MO_64, FPST_A64, gen_helper_sme_fmopa_d) + MO_64, FPST_ZA, gen_helper_sme_fmopa_d) TRANS_FEAT(BFMOPA, aa64_sme, do_outprod_env, a, MO_32, gen_helper_sme_bfmopa) -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> The whole ZA state will also contain ZT0. Make things easier in aarch64_set_svcr to zero both by wrapping them in a common structure. Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-14-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/cpu.h | 48 +++++++++++++++++++--------------- linux-user/aarch64/signal.c | 4 +-- target/arm/cpu.c | 4 +-- target/arm/helper.c | 2 +- target/arm/machine.c | 2 +- target/arm/tcg/sme_helper.c | 6 ++--- target/arm/tcg/translate-sme.c | 4 +-- 7 files changed, 38 insertions(+), 32 deletions(-) diff --git a/target/arm/cpu.h b/target/arm/cpu.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/cpu.h +++ b/target/arm/cpu.h @@ -XXX,XX +XXX,XX @@ typedef struct CPUArchState { uint64_t scxtnum_el[4]; - /* - * SME ZA storage -- 256 x 256 byte array, with bytes in host word order, - * as we do with vfp.zregs[]. This corresponds to the architectural ZA - * array, where ZA[N] is in the least-significant bytes of env->zarray[N]. - * When SVL is less than the architectural maximum, the accessible - * storage is restricted, such that if the SVL is X bytes the guest can - * see only the bottom X elements of zarray[], and only the least - * significant X bytes of each element of the array. (In other words, - * the observable part is always square.) - * - * The ZA storage can also be considered as a set of square tiles of - * elements of different sizes. The mapping from tiles to the ZA array - * is architecturally defined, such that for tiles of elements of esz - * bytes, the Nth row (or "horizontal slice") of tile T is in - * ZA[T + N * esz]. Note that this means that each tile is not contiguous - * in the ZA storage, because its rows are striped through the ZA array. - * - * Because this is so large, keep this toward the end of the reset area, - * to keep the offsets into the rest of the structure smaller. - */ - ARMVectorReg zarray[ARM_MAX_VQ * 16]; + struct { + /* + * SME ZA storage -- 256 x 256 byte array, with bytes in host + * word order, as we do with vfp.zregs[]. This corresponds to + * the architectural ZA array, where ZA[N] is in the least + * significant bytes of env->za_state.za[N]. + * + * When SVL is less than the architectural maximum, the accessible + * storage is restricted, such that if the SVL is X bytes the guest + * can see only the bottom X elements of zarray[], and only the least + * significant X bytes of each element of the array. (In other words, + * the observable part is always square.) + * + * The ZA storage can also be considered as a set of square tiles of + * elements of different sizes. The mapping from tiles to the ZA array + * is architecturally defined, such that for tiles of elements of esz + * bytes, the Nth row (or "horizontal slice") of tile T is in + * ZA[T + N * esz]. Note that this means that each tile is not + * contiguous in the ZA storage, because its rows are striped through + * the ZA array. + * + * Because this is so large, keep this toward the end of the + * reset area, to keep the offsets into the rest of the structure + * smaller. + */ + ARMVectorReg za[ARM_MAX_VQ * 16]; + } za_state; struct CPUBreakpoint *cpu_breakpoint[16]; struct CPUWatchpoint *cpu_watchpoint[16]; diff --git a/linux-user/aarch64/signal.c b/linux-user/aarch64/signal.c index XXXXXXX..XXXXXXX 100644 --- a/linux-user/aarch64/signal.c +++ b/linux-user/aarch64/signal.c @@ -XXX,XX +XXX,XX @@ static void target_setup_za_record(struct target_za_context *za, for (i = 0; i < vl; ++i) { uint64_t *z = (void *)za + TARGET_ZA_SIG_ZAV_OFFSET(vq, i); for (j = 0; j < vq * 2; ++j) { - __put_user_e(env->zarray[i].d[j], z + j, le); + __put_user_e(env->za_state.za[i].d[j], z + j, le); } } } @@ -XXX,XX +XXX,XX @@ static bool target_restore_za_record(CPUARMState *env, for (i = 0; i < vl; ++i) { uint64_t *z = (void *)za + TARGET_ZA_SIG_ZAV_OFFSET(vq, i); for (j = 0; j < vq * 2; ++j) { - __get_user_e(env->zarray[i].d[j], z + j, le); + __get_user_e(env->za_state.za[i].d[j], z + j, le); } } return true; diff --git a/target/arm/cpu.c b/target/arm/cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/cpu.c +++ b/target/arm/cpu.c @@ -XXX,XX +XXX,XX @@ static void aarch64_cpu_dump_state(CPUState *cs, FILE *f, int flags) qemu_fprintf(f, "ZA[%0*d]=", svl_lg10, i); for (j = zcr_len; j >= 0; --j) { qemu_fprintf(f, "%016" PRIx64 ":%016" PRIx64 "%c", - env->zarray[i].d[2 * j + 1], - env->zarray[i].d[2 * j], + env->za_state.za[i].d[2 * j + 1], + env->za_state.za[i].d[2 * j], j ? ':' : '\n'); } } diff --git a/target/arm/helper.c b/target/arm/helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/helper.c +++ b/target/arm/helper.c @@ -XXX,XX +XXX,XX @@ void aarch64_set_svcr(CPUARMState *env, uint64_t new, uint64_t mask) * when disabled either. */ if (change & new & R_SVCR_ZA_MASK) { - memset(env->zarray, 0, sizeof(env->zarray)); + memset(&env->za_state, 0, sizeof(env->za_state)); } if (tcg_enabled()) { diff --git a/target/arm/machine.c b/target/arm/machine.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/machine.c +++ b/target/arm/machine.c @@ -XXX,XX +XXX,XX @@ static const VMStateDescription vmstate_za = { .minimum_version_id = 1, .needed = za_needed, .fields = (const VMStateField[]) { - VMSTATE_STRUCT_ARRAY(env.zarray, ARMCPU, ARM_MAX_VQ * 16, 0, + VMSTATE_STRUCT_ARRAY(env.za_state.za, ARMCPU, ARM_MAX_VQ * 16, 0, vmstate_vreg, ARMVectorReg), VMSTATE_END_OF_LIST() } diff --git a/target/arm/tcg/sme_helper.c b/target/arm/tcg/sme_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme_helper.c +++ b/target/arm/tcg/sme_helper.c @@ -XXX,XX +XXX,XX @@ void helper_sme_zero(CPUARMState *env, uint32_t imm, uint32_t svl) uint32_t i; /* - * Special case clearing the entire ZA space. + * Special case clearing the entire ZArray. * This falls into the CONSTRAINED UNPREDICTABLE zeroing of any * parts of the ZA storage outside of SVL. */ if (imm == 0xff) { - memset(env->zarray, 0, sizeof(env->zarray)); + memset(env->za_state.za, 0, sizeof(env->za_state.za)); return; } @@ -XXX,XX +XXX,XX @@ void helper_sme_zero(CPUARMState *env, uint32_t imm, uint32_t svl) */ for (i = 0; i < svl; i++) { if (imm & (1 << (i % 8))) { - memset(&env->zarray[i], 0, svl); + memset(&env->za_state.za[i], 0, svl); } } } diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ static TCGv_ptr get_tile_rowcol(DisasContext *s, int esz, int rs, offset = tile * sizeof(ARMVectorReg); /* Include the byte offset of zarray to make this relative to env. */ - offset += offsetof(CPUARMState, zarray); + offset += offsetof(CPUARMState, za_state.za); tcg_gen_addi_i32(tmp, tmp, offset); /* Add the byte offset to env to produce the final pointer. */ @@ -XXX,XX +XXX,XX @@ static TCGv_ptr get_tile(DisasContext *s, int esz, int tile) TCGv_ptr addr = tcg_temp_new_ptr(); int offset; - offset = tile * sizeof(ARMVectorReg) + offsetof(CPUARMState, zarray); + offset = tile * sizeof(ARMVectorReg) + offsetof(CPUARMState, za_state.za); tcg_gen_addi_ptr(addr, tcg_env, offset); return addr; -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-15-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/cpu-features.h | 63 +++++++++++++++++++++++++++++++++++++++ target/arm/cpu.h | 1 + 2 files changed, 64 insertions(+) diff --git a/target/arm/cpu-features.h b/target/arm/cpu-features.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/cpu-features.h +++ b/target/arm/cpu-features.h @@ -XXX,XX +XXX,XX @@ static inline bool isar_feature_aa64_rpres(const ARMISARegisters *id) return FIELD_EX64_IDREG(id, ID_AA64ISAR2, RPRES); } +static inline bool isar_feature_aa64_lut(const ARMISARegisters *id) +{ + return FIELD_EX64_IDREG(id, ID_AA64ISAR2, LUT); +} + static inline bool isar_feature_aa64_fp_simd(const ARMISARegisters *id) { /* We always set the AdvSIMD and FP fields identically. */ @@ -XXX,XX +XXX,XX @@ static inline bool isar_feature_aa64_sve2(const ARMISARegisters *id) return FIELD_EX64_IDREG(id, ID_AA64ZFR0, SVEVER) != 0; } +static inline bool isar_feature_aa64_sve2p1(const ARMISARegisters *id) +{ + return FIELD_EX64_IDREG(id, ID_AA64ZFR0, SVEVER) >=2; +} + static inline bool isar_feature_aa64_sve2_aes(const ARMISARegisters *id) { return FIELD_EX64_IDREG(id, ID_AA64ZFR0, AES) != 0; @@ -XXX,XX +XXX,XX @@ static inline bool isar_feature_aa64_sve_f64mm(const ARMISARegisters *id) return FIELD_EX64_IDREG(id, ID_AA64ZFR0, F64MM) != 0; } +static inline bool isar_feature_aa64_sve_b16b16(const ARMISARegisters *id) +{ + return FIELD_EX64_IDREG(id, ID_AA64ZFR0, B16B16); +} + +static inline bool isar_feature_aa64_sme_b16b16(const ARMISARegisters *id) +{ + return FIELD_EX64_IDREG(id, ID_AA64SMFR0, B16B16); +} + +static inline bool isar_feature_aa64_sme_f16f16(const ARMISARegisters *id) +{ + return FIELD_EX64_IDREG(id, ID_AA64SMFR0, F16F16); +} + static inline bool isar_feature_aa64_sme_f64f64(const ARMISARegisters *id) { return FIELD_EX64_IDREG(id, ID_AA64SMFR0, F64F64); @@ -XXX,XX +XXX,XX @@ static inline bool isar_feature_aa64_sme_fa64(const ARMISARegisters *id) return FIELD_EX64_IDREG(id, ID_AA64SMFR0, FA64); } +static inline bool isar_feature_aa64_sme2(const ARMISARegisters *id) +{ + return FIELD_EX64_IDREG(id, ID_AA64SMFR0, SMEVER) != 0; +} + +static inline bool isar_feature_aa64_sme2p1(const ARMISARegisters *id) +{ + return FIELD_EX64_IDREG(id, ID_AA64SMFR0, SMEVER) >= 2; +} + +/* + * Combinations of feature tests, for ease of use with TRANS_FEAT. + */ +static inline bool isar_feature_aa64_sme_or_sve2p1(const ARMISARegisters *id) +{ + return isar_feature_aa64_sme(id) || isar_feature_aa64_sve2p1(id); +} + +static inline bool isar_feature_aa64_sme2_or_sve2p1(const ARMISARegisters *id) +{ + return isar_feature_aa64_sme2(id) || isar_feature_aa64_sve2p1(id); +} + +static inline bool isar_feature_aa64_sme2p1_or_sve2p1(const ARMISARegisters *id) +{ + return isar_feature_aa64_sme2p1(id) || isar_feature_aa64_sve2p1(id); +} + +static inline bool isar_feature_aa64_sme2_i16i64(const ARMISARegisters *id) +{ + return isar_feature_aa64_sme2(id) && isar_feature_aa64_sme_i16i64(id); +} + +static inline bool isar_feature_aa64_sme2_f64f64(const ARMISARegisters *id) +{ + return isar_feature_aa64_sme2(id) && isar_feature_aa64_sme_f64f64(id); +} + /* * Feature tests for "does this exist in either 32-bit or 64-bit?" */ diff --git a/target/arm/cpu.h b/target/arm/cpu.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/cpu.h +++ b/target/arm/cpu.h @@ -XXX,XX +XXX,XX @@ FIELD(ID_AA64ISAR2, SYSINSTR_128, 36, 4) FIELD(ID_AA64ISAR2, PRFMSLC, 40, 4) FIELD(ID_AA64ISAR2, RPRFM, 48, 4) FIELD(ID_AA64ISAR2, CSSC, 52, 4) +FIELD(ID_AA64ISAR2, LUT, 56, 4) FIELD(ID_AA64ISAR2, ATS1A, 60, 4) FIELD(ID_AA64PFR0, EL0, 0, 4) -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> This is a 512-bit array introduced with SME2. Save it only when ZA is in use. Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-16-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/cpu.h | 3 +++ target/arm/machine.c | 20 ++++++++++++++++++++ 2 files changed, 23 insertions(+) diff --git a/target/arm/cpu.h b/target/arm/cpu.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/cpu.h +++ b/target/arm/cpu.h @@ -XXX,XX +XXX,XX @@ typedef struct CPUArchState { uint64_t scxtnum_el[4]; struct { + /* SME2 ZT0 -- 512 bit array, with data ordered like ARMVectorReg. */ + uint64_t zt0[512 / 64] QEMU_ALIGNED(16); + /* * SME ZA storage -- 256 x 256 byte array, with bytes in host * word order, as we do with vfp.zregs[]. This corresponds to diff --git a/target/arm/machine.c b/target/arm/machine.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/machine.c +++ b/target/arm/machine.c @@ -XXX,XX +XXX,XX @@ static const VMStateDescription vmstate_za = { } }; +static bool zt0_needed(void *opaque) +{ + ARMCPU *cpu = opaque; + + return za_needed(cpu) && cpu_isar_feature(aa64_sme2, cpu); +} + +static const VMStateDescription vmstate_zt0 = { + .name = "cpu/zt0", + .version_id = 1, + .minimum_version_id = 1, + .needed = zt0_needed, + .fields = (VMStateField[]) { + VMSTATE_UINT64_ARRAY(env.za_state.zt0, ARMCPU, + ARRAY_SIZE(((CPUARMState *)0)->za_state.zt0)), + VMSTATE_END_OF_LIST() + } +}; + static bool serror_needed(void *opaque) { ARMCPU *cpu = opaque; @@ -XXX,XX +XXX,XX @@ const VMStateDescription vmstate_arm_cpu = { &vmstate_m_security, &vmstate_sve, &vmstate_za, + &vmstate_zt0, &vmstate_serror, &vmstate_irq_line_state, &vmstate_wfxt_timer, -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Pipe the value through from SMCR_ELx through hflags and into the disassembly context. Enable EZT0 in smcr_write. Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-17-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/cpu.h | 2 ++ target/arm/tcg/translate.h | 1 + target/arm/cpu.c | 3 +++ target/arm/helper.c | 6 +++++- target/arm/tcg/hflags.c | 34 +++++++++++++++++++++++++++++++++- target/arm/tcg/translate-a64.c | 1 + 6 files changed, 45 insertions(+), 2 deletions(-) diff --git a/target/arm/cpu.h b/target/arm/cpu.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/cpu.h +++ b/target/arm/cpu.h @@ -XXX,XX +XXX,XX @@ FIELD(SVCR, ZA, 1, 1) /* Fields for SMCR_ELx. */ FIELD(SMCR, LEN, 0, 4) +FIELD(SMCR, EZT0, 30, 1) FIELD(SMCR, FA64, 31, 1) /* Write a new value to v7m.exception, thus transitioning into or out @@ -XXX,XX +XXX,XX @@ FIELD(TBFLAG_A64, NV2_MEM_E20, 35, 1) FIELD(TBFLAG_A64, NV2_MEM_BE, 36, 1) FIELD(TBFLAG_A64, AH, 37, 1) /* FPCR.AH */ FIELD(TBFLAG_A64, NEP, 38, 1) /* FPCR.NEP */ +FIELD(TBFLAG_A64, ZT0EXC_EL, 39, 2) /* * Helpers for using the above. Note that only the A64 accessors use diff --git a/target/arm/tcg/translate.h b/target/arm/tcg/translate.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate.h +++ b/target/arm/tcg/translate.h @@ -XXX,XX +XXX,XX @@ typedef struct DisasContext { int fp_excp_el; /* FP exception EL or 0 if enabled */ int sve_excp_el; /* SVE exception EL or 0 if enabled */ int sme_excp_el; /* SME exception EL or 0 if enabled */ + int zt0_excp_el; /* ZT0 exception EL or 0 if enabled */ int vl; /* current vector length in bytes */ int svl; /* current streaming vector length in bytes */ bool vfp_enabled; /* FP enabled via FPSCR.EN */ diff --git a/target/arm/cpu.c b/target/arm/cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/cpu.c +++ b/target/arm/cpu.c @@ -XXX,XX +XXX,XX @@ void arm_emulate_firmware_reset(CPUState *cpustate, int target_el) env->cp15.cptr_el[3] |= R_CPTR_EL3_ESM_MASK; env->cp15.scr_el3 |= SCR_ENTP2; env->vfp.smcr_el[3] = 0xf; + if (cpu_isar_feature(aa64_sme2, cpu)) { + env->vfp.smcr_el[3] |= R_SMCR_EZT0_MASK; + } } if (cpu_isar_feature(aa64_hcx, cpu)) { env->cp15.scr_el3 |= SCR_HXEN; diff --git a/target/arm/helper.c b/target/arm/helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/helper.c +++ b/target/arm/helper.c @@ -XXX,XX +XXX,XX @@ static void smcr_write(CPUARMState *env, const ARMCPRegInfo *ri, { int cur_el = arm_current_el(env); int old_len = sve_vqm1_for_el(env, cur_el); + uint64_t valid_mask = R_SMCR_LEN_MASK | R_SMCR_FA64_MASK; int new_len; QEMU_BUILD_BUG_ON(ARM_MAX_VQ > R_SMCR_LEN_MASK + 1); - value &= R_SMCR_LEN_MASK | R_SMCR_FA64_MASK; + if (cpu_isar_feature(aa64_sme2, env_archcpu(env))) { + valid_mask |= R_SMCR_EZT0_MASK; + } + value &= valid_mask; raw_write(env, ri, value); /* diff --git a/target/arm/tcg/hflags.c b/target/arm/tcg/hflags.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/hflags.c +++ b/target/arm/tcg/hflags.c @@ -XXX,XX +XXX,XX @@ static CPUARMTBFlags rebuild_hflags_a32(CPUARMState *env, int fp_el, return rebuild_hflags_common_32(env, fp_el, mmu_idx, flags); } +/* + * Return the exception level to which exceptions should be taken for ZT0. + * C.f. the ARM pseudocode function CheckSMEZT0Enabled, after the ZA check. + */ +static int zt0_exception_el(CPUARMState *env, int el) +{ +#ifndef CONFIG_USER_ONLY + if (el <= 1 + && !el_is_in_host(env, el) + && !FIELD_EX64(env->vfp.smcr_el[1], SMCR, EZT0)) { + return 1; + } + if (el <= 2 + && arm_is_el2_enabled(env) + && !FIELD_EX64(env->vfp.smcr_el[2], SMCR, EZT0)) { + return 2; + } + if (arm_feature(env, ARM_FEATURE_EL3) + && !FIELD_EX64(env->vfp.smcr_el[3], SMCR, EZT0)) { + return 3; + } +#endif + return 0; +} + static CPUARMTBFlags rebuild_hflags_a64(CPUARMState *env, int el, int fp_el, ARMMMUIdx mmu_idx) { @@ -XXX,XX +XXX,XX @@ static CPUARMTBFlags rebuild_hflags_a64(CPUARMState *env, int el, int fp_el, DP_TBFLAG_A64(flags, PSTATE_SM, 1); DP_TBFLAG_A64(flags, SME_TRAP_NONSTREAMING, !sme_fa64(env, el)); } - DP_TBFLAG_A64(flags, PSTATE_ZA, FIELD_EX64(env->svcr, SVCR, ZA)); + + if (FIELD_EX64(env->svcr, SVCR, ZA)) { + DP_TBFLAG_A64(flags, PSTATE_ZA, 1); + if (cpu_isar_feature(aa64_sme2, env_archcpu(env))) { + int zt0_el = zt0_exception_el(env, el); + DP_TBFLAG_A64(flags, ZT0EXC_EL, zt0_el); + } + } } sctlr = regime_sctlr(env, stage1); diff --git a/target/arm/tcg/translate-a64.c b/target/arm/tcg/translate-a64.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-a64.c +++ b/target/arm/tcg/translate-a64.c @@ -XXX,XX +XXX,XX @@ static void aarch64_tr_init_disas_context(DisasContextBase *dcbase, dc->trap_eret = EX_TBFLAG_A64(tb_flags, TRAP_ERET); dc->sve_excp_el = EX_TBFLAG_A64(tb_flags, SVEEXC_EL); dc->sme_excp_el = EX_TBFLAG_A64(tb_flags, SMEEXC_EL); + dc->zt0_excp_el = EX_TBFLAG_A64(tb_flags, ZT0EXC_EL); dc->vl = (EX_TBFLAG_A64(tb_flags, VL) + 1) * 16; dc->svl = (EX_TBFLAG_A64(tb_flags, SVL) + 1) * 16; dc->pauth_active = EX_TBFLAG_A64(tb_flags, PAUTH_ACTIVE); -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-18-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/syndrome.h | 1 + target/arm/tcg/sme.decode | 1 + target/arm/tcg/translate-sme.c | 26 ++++++++++++++++++++++++++ 3 files changed, 28 insertions(+) diff --git a/target/arm/syndrome.h b/target/arm/syndrome.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/syndrome.h +++ b/target/arm/syndrome.h @@ -XXX,XX +XXX,XX @@ typedef enum { SME_ET_Streaming, SME_ET_NotStreaming, SME_ET_InactiveZA, + SME_ET_InaccessibleZT0, } SMEExceptionType; #define ARM_EL_EC_LENGTH 6 diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ ### SME Misc ZERO 11000000 00 001 00000000000 imm:8 +ZERO_zt0 11000000 01 001 00000000000 00000001 ### SME Move into/from Array diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ #include "decode-sme.c.inc" +static bool sme2_zt0_enabled_check(DisasContext *s) +{ + if (!sme_za_enabled_check(s)) { + return false; + } + if (s->zt0_excp_el) { + gen_exception_insn_el(s, 0, EXCP_UDEF, + syn_smetrap(SME_ET_InaccessibleZT0, false), + s->zt0_excp_el); + return false; + } + return true; +} /* * Resolve tile.size[index] to a host pointer, where tile and index @@ -XXX,XX +XXX,XX @@ static bool trans_ZERO(DisasContext *s, arg_ZERO *a) return true; } +static bool trans_ZERO_zt0(DisasContext *s, arg_ZERO_zt0 *a) +{ + if (!dc_isar_feature(aa64_sme2, s)) { + return false; + } + if (sme_enabled_check(s) && sme2_zt0_enabled_check(s)) { + tcg_gen_gvec_dup_imm(MO_64, offsetof(CPUARMState, za_state.zt0), + sizeof_field(CPUARMState, za_state.zt0), + sizeof_field(CPUARMState, za_state.zt0), 0); + } + return true; +} + static bool trans_MOVA(DisasContext *s, arg_MOVA *a) { static gen_helper_gvec_4 * const h_fns[5] = { -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Honor AlignmentEnforced() for LDR/STR (vector), (predicate), and (array vector). Within the expansion functions, clear @align when we're done emitting loads at the largest size. Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-19-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/translate-a64.h | 6 ++-- target/arm/tcg/translate-sme.c | 5 ++-- target/arm/tcg/translate-sve.c | 50 ++++++++++++++++++++++++---------- 3 files changed, 42 insertions(+), 19 deletions(-) diff --git a/target/arm/tcg/translate-a64.h b/target/arm/tcg/translate-a64.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-a64.h +++ b/target/arm/tcg/translate-a64.h @@ -XXX,XX +XXX,XX @@ void gen_gvec_usqadd_qc(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs, uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz); -void gen_sve_ldr(DisasContext *s, TCGv_ptr, int vofs, int len, int rn, int imm); -void gen_sve_str(DisasContext *s, TCGv_ptr, int vofs, int len, int rn, int imm); +void gen_sve_ldr(DisasContext *s, TCGv_ptr, int vofs, + int len, int rn, int imm, MemOp align); +void gen_sve_str(DisasContext *s, TCGv_ptr, int vofs, + int len, int rn, int imm, MemOp align); #endif /* TARGET_ARM_TRANSLATE_A64_H */ diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ static bool trans_LDST1(DisasContext *s, arg_LDST1 *a) return true; } -typedef void GenLdStR(DisasContext *, TCGv_ptr, int, int, int, int); +typedef void GenLdStR(DisasContext *, TCGv_ptr, int, int, int, int, MemOp); static bool do_ldst_r(DisasContext *s, arg_ldstr *a, GenLdStR *fn) { @@ -XXX,XX +XXX,XX @@ static bool do_ldst_r(DisasContext *s, arg_ldstr *a, GenLdStR *fn) /* ZA[n] equates to ZA0H.B[n]. */ base = get_tile_rowcol(s, MO_8, a->rv, imm, false); - fn(s, base, 0, svl, a->rn, imm * svl); + fn(s, base, 0, svl, a->rn, imm * svl, + s->align_mem ? MO_ALIGN_16 : MO_UNALN); return true; } diff --git a/target/arm/tcg/translate-sve.c b/target/arm/tcg/translate-sve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sve.c +++ b/target/arm/tcg/translate-sve.c @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(UCVTF_dd, aa64_sve, gen_gvec_fpst_arg_zpz, */ void gen_sve_ldr(DisasContext *s, TCGv_ptr base, int vofs, - int len, int rn, int imm) + int len, int rn, int imm, MemOp align) { int len_align = QEMU_ALIGN_DOWN(len, 16); int len_remain = len % 16; @@ -XXX,XX +XXX,XX @@ void gen_sve_ldr(DisasContext *s, TCGv_ptr base, int vofs, for (i = 0; i < len_align; i += 16) { tcg_gen_qemu_ld_i128(t16, clean_addr, midx, - MO_LE | MO_128 | MO_ATOM_NONE); + MO_LE | MO_128 | MO_ATOM_NONE | align); tcg_gen_extr_i128_i64(t0, t1, t16); tcg_gen_st_i64(t0, base, vofs + i); tcg_gen_st_i64(t1, base, vofs + i + 8); tcg_gen_addi_i64(clean_addr, clean_addr, 16); } + if (len_align) { + align = MO_UNALN; + } } else { TCGLabel *loop = gen_new_label(); TCGv_ptr tp, i = tcg_temp_new_ptr(); @@ -XXX,XX +XXX,XX @@ void gen_sve_ldr(DisasContext *s, TCGv_ptr base, int vofs, t16 = tcg_temp_new_i128(); tcg_gen_qemu_ld_i128(t16, clean_addr, midx, - MO_LE | MO_128 | MO_ATOM_NONE); + MO_LE | MO_128 | MO_ATOM_NONE | align); tcg_gen_addi_i64(clean_addr, clean_addr, 16); tp = tcg_temp_new_ptr(); @@ -XXX,XX +XXX,XX @@ void gen_sve_ldr(DisasContext *s, TCGv_ptr base, int vofs, tcg_gen_st_i64(t1, tp, vofs + 8); tcg_gen_brcondi_ptr(TCG_COND_LTU, i, len_align, loop); + align = MO_UNALN; } /* @@ -XXX,XX +XXX,XX @@ void gen_sve_ldr(DisasContext *s, TCGv_ptr base, int vofs, */ if (len_remain >= 8) { t0 = tcg_temp_new_i64(); - tcg_gen_qemu_ld_i64(t0, clean_addr, midx, MO_LEUQ | MO_ATOM_NONE); + tcg_gen_qemu_ld_i64(t0, clean_addr, midx, + MO_LEUQ | MO_ATOM_NONE | align); + align = MO_UNALN; tcg_gen_st_i64(t0, base, vofs + len_align); len_remain -= 8; len_align += 8; @@ -XXX,XX +XXX,XX @@ void gen_sve_ldr(DisasContext *s, TCGv_ptr base, int vofs, case 4: case 8: tcg_gen_qemu_ld_i64(t0, clean_addr, midx, - MO_LE | ctz32(len_remain) | MO_ATOM_NONE); + MO_LE | ctz32(len_remain) + | MO_ATOM_NONE | align); break; case 6: t1 = tcg_temp_new_i64(); - tcg_gen_qemu_ld_i64(t0, clean_addr, midx, MO_LEUL | MO_ATOM_NONE); + tcg_gen_qemu_ld_i64(t0, clean_addr, midx, + MO_LEUL | MO_ATOM_NONE | align); tcg_gen_addi_i64(clean_addr, clean_addr, 4); tcg_gen_qemu_ld_i64(t1, clean_addr, midx, MO_LEUW | MO_ATOM_NONE); tcg_gen_deposit_i64(t0, t0, t1, 32, 32); @@ -XXX,XX +XXX,XX @@ void gen_sve_ldr(DisasContext *s, TCGv_ptr base, int vofs, /* Similarly for stores. */ void gen_sve_str(DisasContext *s, TCGv_ptr base, int vofs, - int len, int rn, int imm) + int len, int rn, int imm, MemOp align) { int len_align = QEMU_ALIGN_DOWN(len, 16); int len_remain = len % 16; @@ -XXX,XX +XXX,XX @@ void gen_sve_str(DisasContext *s, TCGv_ptr base, int vofs, tcg_gen_ld_i64(t1, base, vofs + i + 8); tcg_gen_concat_i64_i128(t16, t0, t1); tcg_gen_qemu_st_i128(t16, clean_addr, midx, - MO_LE | MO_128 | MO_ATOM_NONE); + MO_LE | MO_128 | MO_ATOM_NONE | align); tcg_gen_addi_i64(clean_addr, clean_addr, 16); } + if (len_align) { + align = MO_UNALN; + } } else { TCGLabel *loop = gen_new_label(); TCGv_ptr tp, i = tcg_temp_new_ptr(); @@ -XXX,XX +XXX,XX @@ void gen_sve_str(DisasContext *s, TCGv_ptr base, int vofs, tcg_gen_addi_i64(clean_addr, clean_addr, 16); tcg_gen_brcondi_ptr(TCG_COND_LTU, i, len_align, loop); + align = MO_UNALN; } /* Predicate register stores can be any multiple of 2. */ if (len_remain >= 8) { t0 = tcg_temp_new_i64(); tcg_gen_ld_i64(t0, base, vofs + len_align); - tcg_gen_qemu_st_i64(t0, clean_addr, midx, MO_LEUQ | MO_ATOM_NONE); + tcg_gen_qemu_st_i64(t0, clean_addr, midx, + MO_LEUQ | MO_ATOM_NONE | align); + align = MO_UNALN; len_remain -= 8; len_align += 8; if (len_remain) { @@ -XXX,XX +XXX,XX @@ void gen_sve_str(DisasContext *s, TCGv_ptr base, int vofs, case 4: case 8: tcg_gen_qemu_st_i64(t0, clean_addr, midx, - MO_LE | ctz32(len_remain) | MO_ATOM_NONE); + MO_LE | ctz32(len_remain) + | MO_ATOM_NONE | align); break; case 6: - tcg_gen_qemu_st_i64(t0, clean_addr, midx, MO_LEUL | MO_ATOM_NONE); + tcg_gen_qemu_st_i64(t0, clean_addr, midx, + MO_LEUL | MO_ATOM_NONE | align); tcg_gen_addi_i64(clean_addr, clean_addr, 4); tcg_gen_shri_i64(t0, t0, 32); tcg_gen_qemu_st_i64(t0, clean_addr, midx, MO_LEUW | MO_ATOM_NONE); @@ -XXX,XX +XXX,XX @@ static bool trans_LDR_zri(DisasContext *s, arg_rri *a) if (sve_access_check(s)) { int size = vec_full_reg_size(s); int off = vec_full_reg_offset(s, a->rd); - gen_sve_ldr(s, tcg_env, off, size, a->rn, a->imm * size); + gen_sve_ldr(s, tcg_env, off, size, a->rn, a->imm * size, + s->align_mem ? MO_ALIGN_16 : MO_UNALN); } return true; } @@ -XXX,XX +XXX,XX @@ static bool trans_LDR_pri(DisasContext *s, arg_rri *a) if (sve_access_check(s)) { int size = pred_full_reg_size(s); int off = pred_full_reg_offset(s, a->rd); - gen_sve_ldr(s, tcg_env, off, size, a->rn, a->imm * size); + gen_sve_ldr(s, tcg_env, off, size, a->rn, a->imm * size, + s->align_mem ? MO_ALIGN_2 : MO_UNALN); } return true; } @@ -XXX,XX +XXX,XX @@ static bool trans_STR_zri(DisasContext *s, arg_rri *a) if (sve_access_check(s)) { int size = vec_full_reg_size(s); int off = vec_full_reg_offset(s, a->rd); - gen_sve_str(s, tcg_env, off, size, a->rn, a->imm * size); + gen_sve_str(s, tcg_env, off, size, a->rn, a->imm * size, + s->align_mem ? MO_ALIGN_16 : MO_UNALN); } return true; } @@ -XXX,XX +XXX,XX @@ static bool trans_STR_pri(DisasContext *s, arg_rri *a) if (sve_access_check(s)) { int size = pred_full_reg_size(s); int off = pred_full_reg_offset(s, a->rd); - gen_sve_str(s, tcg_env, off, size, a->rn, a->imm * size); + gen_sve_str(s, tcg_env, off, size, a->rn, a->imm * size, + s->align_mem ? MO_ALIGN_2 : MO_UNALN); } return true; } -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-20-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/sme.decode | 6 ++++++ target/arm/tcg/translate-sme.c | 13 +++++++++++++ 2 files changed, 19 insertions(+) diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ LDST1 1110000 111 st:1 rm:5 v:1 .. pg:3 rn:5 0 za_imm:4 \ LDR 1110000 100 0 000000 .. 000 ..... 0 .... @ldstr STR 1110000 100 1 000000 .. 000 ..... 0 .... @ldstr +&ldstzt0 rn +@ldstzt0 ....... ... . ...... .. ... rn:5 ..... &ldstzt0 + +LDR_zt0 1110000 100 0 111111 00 000 ..... 00000 @ldstzt0 +STR_zt0 1110000 100 1 111111 00 000 ..... 00000 @ldstzt0 + ### SME Add Vector to Array &adda zad zn pm pn diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ static bool do_ldst_r(DisasContext *s, arg_ldstr *a, GenLdStR *fn) TRANS_FEAT(LDR, aa64_sme, do_ldst_r, a, gen_sve_ldr) TRANS_FEAT(STR, aa64_sme, do_ldst_r, a, gen_sve_str) +static bool do_ldst_zt0(DisasContext *s, arg_ldstzt0 *a, GenLdStR *fn) +{ + if (sme2_zt0_enabled_check(s)) { + fn(s, tcg_env, offsetof(CPUARMState, za_state.zt0), + sizeof_field(CPUARMState, za_state.zt0), a->rn, 0, + s->align_mem ? MO_ALIGN_16 : MO_UNALN); + } + return true; +} + +TRANS_FEAT(LDR_zt0, aa64_sme2, do_ldst_zt0, a, gen_sve_ldr) +TRANS_FEAT(STR_zt0, aa64_sme2, do_ldst_zt0, a, gen_sve_str) + static bool do_adda(DisasContext *s, arg_adda *a, MemOp esz, gen_helper_gvec_4 *fn) { -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-21-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/sme.decode | 5 +++++ target/arm/tcg/translate-sme.c | 13 +++++++++++++ 2 files changed, 18 insertions(+) diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ MOVA 11000000 esz:2 00001 0 v:1 .. pg:3 0 za_imm:4 zr:5 \ MOVA 11000000 11 00001 1 v:1 .. pg:3 0 za_imm:4 zr:5 \ &mova to_vec=1 rs=%mova_rs esz=4 +### SME Move into/from ZT0 + +MOVT_rzt 1100 0000 0100 1100 0 off:3 00 11111 rt:5 +MOVT_ztr 1100 0000 0100 1110 0 off:3 00 11111 rt:5 + ### SME Memory &ldst esz rs pg rn rm za_imm v:bool st:bool diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ static bool trans_MOVA(DisasContext *s, arg_MOVA *a) return true; } +static bool do_movt(DisasContext *s, arg_MOVT_rzt *a, + void (*func)(TCGv_i64, TCGv_ptr, tcg_target_long)) +{ + if (sme2_zt0_enabled_check(s)) { + func(cpu_reg(s, a->rt), tcg_env, + offsetof(CPUARMState, za_state.zt0) + a->off * 8); + } + return true; +} + +TRANS_FEAT(MOVT_rzt, aa64_sme2, do_movt, a, tcg_gen_ld_i64) +TRANS_FEAT(MOVT_ztr, aa64_sme2, do_movt, a, tcg_gen_st_i64) + static bool trans_LDST1(DisasContext *s, arg_LDST1 *a) { typedef void GenLdSt1(TCGv_env, TCGv_ptr, TCGv_ptr, TCGv, TCGv_i32); -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Decode tile number and index offset beforehand and separately. Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-22-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/sme.decode | 46 +++++++++++++++++++++++----------- target/arm/tcg/translate-sme.c | 17 +++++-------- 2 files changed, 38 insertions(+), 25 deletions(-) diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ ZERO_zt0 11000000 01 001 00000000000 00000001 ### SME Move into/from Array %mova_rs 13:2 !function=plus_12 -&mova esz rs pg zr za_imm v:bool to_vec:bool +&mova esz rs pg zr za off v:bool to_vec:bool -MOVA 11000000 esz:2 00000 0 v:1 .. pg:3 zr:5 0 za_imm:4 \ - &mova to_vec=0 rs=%mova_rs -MOVA 11000000 11 00000 1 v:1 .. pg:3 zr:5 0 za_imm:4 \ - &mova to_vec=0 rs=%mova_rs esz=4 +MOVA 11000000 00 00000 0 v:1 .. pg:3 zr:5 0 off:4 \ + &mova to_vec=0 rs=%mova_rs esz=0 za=0 +MOVA 11000000 01 00000 0 v:1 .. pg:3 zr:5 0 za:1 off:3 \ + &mova to_vec=0 rs=%mova_rs esz=1 +MOVA 11000000 10 00000 0 v:1 .. pg:3 zr:5 0 za:2 off:2 \ + &mova to_vec=0 rs=%mova_rs esz=2 +MOVA 11000000 11 00000 0 v:1 .. pg:3 zr:5 0 za:3 off:1 \ + &mova to_vec=0 rs=%mova_rs esz=3 +MOVA 11000000 11 00000 1 v:1 .. pg:3 zr:5 0 za:4 \ + &mova to_vec=0 rs=%mova_rs esz=4 off=0 -MOVA 11000000 esz:2 00001 0 v:1 .. pg:3 0 za_imm:4 zr:5 \ - &mova to_vec=1 rs=%mova_rs -MOVA 11000000 11 00001 1 v:1 .. pg:3 0 za_imm:4 zr:5 \ - &mova to_vec=1 rs=%mova_rs esz=4 +MOVA 11000000 00 00001 0 v:1 .. pg:3 0 off:4 zr:5 \ + &mova to_vec=1 rs=%mova_rs esz=0 za=0 +MOVA 11000000 01 00001 0 v:1 .. pg:3 0 za:1 off:3 zr:5 \ + &mova to_vec=1 rs=%mova_rs esz=1 +MOVA 11000000 10 00001 0 v:1 .. pg:3 0 za:2 off:2 zr:5 \ + &mova to_vec=1 rs=%mova_rs esz=2 +MOVA 11000000 11 00001 0 v:1 .. pg:3 0 za:3 off:1 zr:5 \ + &mova to_vec=1 rs=%mova_rs esz=3 +MOVA 11000000 11 00001 1 v:1 .. pg:3 0 za:4 zr:5 \ + &mova to_vec=1 rs=%mova_rs esz=4 off=0 ### SME Move into/from ZT0 @@ -XXX,XX +XXX,XX @@ MOVT_ztr 1100 0000 0100 1110 0 off:3 00 11111 rt:5 ### SME Memory -&ldst esz rs pg rn rm za_imm v:bool st:bool +&ldst esz rs pg rn rm za off v:bool st:bool -LDST1 1110000 0 esz:2 st:1 rm:5 v:1 .. pg:3 rn:5 0 za_imm:4 \ - &ldst rs=%mova_rs -LDST1 1110000 111 st:1 rm:5 v:1 .. pg:3 rn:5 0 za_imm:4 \ - &ldst esz=4 rs=%mova_rs +LDST1 1110000 0 00 st:1 rm:5 v:1 .. pg:3 rn:5 0 off:4 \ + &ldst rs=%mova_rs esz=0 za=0 +LDST1 1110000 0 01 st:1 rm:5 v:1 .. pg:3 rn:5 0 za:1 off:3 \ + &ldst rs=%mova_rs esz=1 +LDST1 1110000 0 10 st:1 rm:5 v:1 .. pg:3 rn:5 0 za:2 off:2 \ + &ldst rs=%mova_rs esz=2 +LDST1 1110000 0 11 st:1 rm:5 v:1 .. pg:3 rn:5 0 za:3 off:1 \ + &ldst rs=%mova_rs esz=3 +LDST1 1110000 1 11 st:1 rm:5 v:1 .. pg:3 rn:5 0 za:4 \ + &ldst rs=%mova_rs esz=4 off=0 &ldstr rv rn imm @ldstr ....... ... . ...... .. ... rn:5 . imm:4 \ diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ static bool sme2_zt0_enabled_check(DisasContext *s) return true; } -/* - * Resolve tile.size[index] to a host pointer, where tile and index - * are always decoded together, dependent on the element size. - */ +/* Resolve tile.size[rs+imm] to a host pointer. */ static TCGv_ptr get_tile_rowcol(DisasContext *s, int esz, int rs, - int tile_index, bool vertical) + int tile, int imm, bool vertical) { - int tile = tile_index >> (4 - esz); - int index = esz == MO_128 ? 0 : extract32(tile_index, 0, 4 - esz); int pos, len, offset; TCGv_i32 tmp; TCGv_ptr addr; @@ -XXX,XX +XXX,XX @@ static TCGv_ptr get_tile_rowcol(DisasContext *s, int esz, int rs, /* Compute the final index, which is Rs+imm. */ tmp = tcg_temp_new_i32(); tcg_gen_trunc_tl_i32(tmp, cpu_reg(s, rs)); - tcg_gen_addi_i32(tmp, tmp, index); + tcg_gen_addi_i32(tmp, tmp, imm); /* Prepare a power-of-two modulo via extraction of @len bits. */ len = ctz32(streaming_vec_reg_size(s)) - esz; @@ -XXX,XX +XXX,XX @@ static bool trans_MOVA(DisasContext *s, arg_MOVA *a) return true; } - t_za = get_tile_rowcol(s, a->esz, a->rs, a->za_imm, a->v); + t_za = get_tile_rowcol(s, a->esz, a->rs, a->za, a->off, a->v); t_zr = vec_full_reg_ptr(s, a->zr); t_pg = pred_full_reg_ptr(s, a->pg); @@ -XXX,XX +XXX,XX @@ static bool trans_LDST1(DisasContext *s, arg_LDST1 *a) return true; } - t_za = get_tile_rowcol(s, a->esz, a->rs, a->za_imm, a->v); + t_za = get_tile_rowcol(s, a->esz, a->rs, a->za, a->off, a->v); t_pg = pred_full_reg_ptr(s, a->pg); addr = tcg_temp_new_i64(); @@ -XXX,XX +XXX,XX @@ static bool do_ldst_r(DisasContext *s, arg_ldstr *a, GenLdStR *fn) } /* ZA[n] equates to ZA0H.B[n]. */ - base = get_tile_rowcol(s, MO_8, a->rv, imm, false); + base = get_tile_rowcol(s, MO_8, a->rv, 0, imm, false); fn(s, base, 0, svl, a->rn, imm * svl, s->align_mem ? MO_ALIGN_16 : MO_UNALN); -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Prepare for more kinds of MOVA from SME2 by renaming the existing SME1 MOVA to indicate tile to/from vector. Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-23-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/sme.decode | 42 +++++++++++++++++----------------- target/arm/tcg/translate-sme.c | 12 +++++----- 2 files changed, 27 insertions(+), 27 deletions(-) diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ ZERO_zt0 11000000 01 001 00000000000 00000001 ### SME Move into/from Array %mova_rs 13:2 !function=plus_12 -&mova esz rs pg zr za off v:bool to_vec:bool +&mova_p esz rs pg zr za off v:bool -MOVA 11000000 00 00000 0 v:1 .. pg:3 zr:5 0 off:4 \ - &mova to_vec=0 rs=%mova_rs esz=0 za=0 -MOVA 11000000 01 00000 0 v:1 .. pg:3 zr:5 0 za:1 off:3 \ - &mova to_vec=0 rs=%mova_rs esz=1 -MOVA 11000000 10 00000 0 v:1 .. pg:3 zr:5 0 za:2 off:2 \ - &mova to_vec=0 rs=%mova_rs esz=2 -MOVA 11000000 11 00000 0 v:1 .. pg:3 zr:5 0 za:3 off:1 \ - &mova to_vec=0 rs=%mova_rs esz=3 -MOVA 11000000 11 00000 1 v:1 .. pg:3 zr:5 0 za:4 \ - &mova to_vec=0 rs=%mova_rs esz=4 off=0 +MOVA_tz 11000000 00 00000 0 v:1 .. pg:3 zr:5 0 off:4 \ + &mova_p rs=%mova_rs esz=0 za=0 +MOVA_tz 11000000 01 00000 0 v:1 .. pg:3 zr:5 0 za:1 off:3 \ + &mova_p rs=%mova_rs esz=1 +MOVA_tz 11000000 10 00000 0 v:1 .. pg:3 zr:5 0 za:2 off:2 \ + &mova_p rs=%mova_rs esz=2 +MOVA_tz 11000000 11 00000 0 v:1 .. pg:3 zr:5 0 za:3 off:1 \ + &mova_p rs=%mova_rs esz=3 +MOVA_tz 11000000 11 00000 1 v:1 .. pg:3 zr:5 0 za:4 \ + &mova_p rs=%mova_rs esz=4 off=0 -MOVA 11000000 00 00001 0 v:1 .. pg:3 0 off:4 zr:5 \ - &mova to_vec=1 rs=%mova_rs esz=0 za=0 -MOVA 11000000 01 00001 0 v:1 .. pg:3 0 za:1 off:3 zr:5 \ - &mova to_vec=1 rs=%mova_rs esz=1 -MOVA 11000000 10 00001 0 v:1 .. pg:3 0 za:2 off:2 zr:5 \ - &mova to_vec=1 rs=%mova_rs esz=2 -MOVA 11000000 11 00001 0 v:1 .. pg:3 0 za:3 off:1 zr:5 \ - &mova to_vec=1 rs=%mova_rs esz=3 -MOVA 11000000 11 00001 1 v:1 .. pg:3 0 za:4 zr:5 \ - &mova to_vec=1 rs=%mova_rs esz=4 off=0 +MOVA_zt 11000000 00 00001 0 v:1 .. pg:3 0 off:4 zr:5 \ + &mova_p rs=%mova_rs esz=0 za=0 +MOVA_zt 11000000 01 00001 0 v:1 .. pg:3 0 za:1 off:3 zr:5 \ + &mova_p rs=%mova_rs esz=1 +MOVA_zt 11000000 10 00001 0 v:1 .. pg:3 0 za:2 off:2 zr:5 \ + &mova_p rs=%mova_rs esz=2 +MOVA_zt 11000000 11 00001 0 v:1 .. pg:3 0 za:3 off:1 zr:5 \ + &mova_p rs=%mova_rs esz=3 +MOVA_zt 11000000 11 00001 1 v:1 .. pg:3 0 za:4 zr:5 \ + &mova_p rs=%mova_rs esz=4 off=0 ### SME Move into/from ZT0 diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ static bool trans_ZERO_zt0(DisasContext *s, arg_ZERO_zt0 *a) return true; } -static bool trans_MOVA(DisasContext *s, arg_MOVA *a) +static bool do_mova_tile(DisasContext *s, arg_mova_p *a, bool to_vec) { static gen_helper_gvec_4 * const h_fns[5] = { gen_helper_sve_sel_zpzz_b, gen_helper_sve_sel_zpzz_h, @@ -XXX,XX +XXX,XX @@ static bool trans_MOVA(DisasContext *s, arg_MOVA *a) TCGv_i32 t_desc; int svl; - if (!dc_isar_feature(aa64_sme, s)) { - return false; - } if (!sme_smza_enabled_check(s)) { return true; } @@ -XXX,XX +XXX,XX @@ static bool trans_MOVA(DisasContext *s, arg_MOVA *a) if (a->v) { /* Vertical slice -- use sme mova helpers. */ - if (a->to_vec) { + if (to_vec) { zc_fns[a->esz](t_zr, t_za, t_pg, t_desc); } else { cz_fns[a->esz](t_za, t_zr, t_pg, t_desc); } } else { /* Horizontal slice -- reuse sve sel helpers. */ - if (a->to_vec) { + if (to_vec) { h_fns[a->esz](t_zr, t_za, t_zr, t_pg, t_desc); } else { h_fns[a->esz](t_za, t_zr, t_za, t_pg, t_desc); @@ -XXX,XX +XXX,XX @@ static bool trans_MOVA(DisasContext *s, arg_MOVA *a) return true; } +TRANS_FEAT(MOVA_tz, aa64_sme, do_mova_tile, a, false) +TRANS_FEAT(MOVA_zt, aa64_sme, do_mova_tile, a, true) + static bool do_movt(DisasContext *s, arg_MOVT_rzt *a, void (*func)(TCGv_i64, TCGv_ptr, tcg_target_long)) { -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Prepare for MOVA array to/from vector with multiple registers by adding a div_len parameter, herein always 1, and a vec_mod parameter, herein always 0. Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-24-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/translate-sme.c | 47 +++++++++++++++++++++++----------- 1 file changed, 32 insertions(+), 15 deletions(-) diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ static bool sme2_zt0_enabled_check(DisasContext *s) /* Resolve tile.size[rs+imm] to a host pointer. */ static TCGv_ptr get_tile_rowcol(DisasContext *s, int esz, int rs, - int tile, int imm, bool vertical) + int tile, int imm, int div_len, + int vec_mod, bool vertical) { int pos, len, offset; TCGv_i32 tmp; @@ -XXX,XX +XXX,XX @@ static TCGv_ptr get_tile_rowcol(DisasContext *s, int esz, int rs, /* Compute the final index, which is Rs+imm. */ tmp = tcg_temp_new_i32(); tcg_gen_trunc_tl_i32(tmp, cpu_reg(s, rs)); + /* + * Round the vector index down to a multiple of vec_mod if necessary. + * We do this before adding the offset, to handle cases like + * MOVA (tile to vector, 2 registers) where we want to call this + * several times in a loop with an increasing offset. We rely on + * the instruction encodings always forcing the initial offset in + * [rs + offset] to be a multiple of vec_mod. The pseudocode usually + * does the round-down after adding the offset rather than before, + * but MOVA is an exception. + */ + if (vec_mod > 1) { + tcg_gen_andc_i32(tmp, tmp, tcg_constant_i32(vec_mod - 1)); + } tcg_gen_addi_i32(tmp, tmp, imm); /* Prepare a power-of-two modulo via extraction of @len bits. */ - len = ctz32(streaming_vec_reg_size(s)) - esz; + len = ctz32(streaming_vec_reg_size(s) / div_len) - esz; if (!len) { /* @@ -XXX,XX +XXX,XX @@ static TCGv_ptr get_tile_rowcol(DisasContext *s, int esz, int rs, return addr; } +/* Resolve ZArray[rs+imm] to a host pointer. */ +static TCGv_ptr get_zarray(DisasContext *s, int rs, int imm, + int div_len, int vec_mod) +{ + /* ZA[n] equates to ZA0H.B[n]. */ + return get_tile_rowcol(s, MO_8, rs, 0, imm, div_len, vec_mod, false); +} + /* * Resolve tile.size[0] to a host pointer. * Used by e.g. outer product insns where we require the entire tile. @@ -XXX,XX +XXX,XX @@ static bool do_mova_tile(DisasContext *s, arg_mova_p *a, bool to_vec) return true; } - t_za = get_tile_rowcol(s, a->esz, a->rs, a->za, a->off, a->v); + t_za = get_tile_rowcol(s, a->esz, a->rs, a->za, a->off, 1, 0, a->v); t_zr = vec_full_reg_ptr(s, a->zr); t_pg = pred_full_reg_ptr(s, a->pg); @@ -XXX,XX +XXX,XX @@ static bool trans_LDST1(DisasContext *s, arg_LDST1 *a) return true; } - t_za = get_tile_rowcol(s, a->esz, a->rs, a->za, a->off, a->v); + t_za = get_tile_rowcol(s, a->esz, a->rs, a->za, a->off, 1, 0, a->v); t_pg = pred_full_reg_ptr(s, a->pg); addr = tcg_temp_new_i64(); @@ -XXX,XX +XXX,XX @@ typedef void GenLdStR(DisasContext *, TCGv_ptr, int, int, int, int, MemOp); static bool do_ldst_r(DisasContext *s, arg_ldstr *a, GenLdStR *fn) { - int svl = streaming_vec_reg_size(s); - int imm = a->imm; - TCGv_ptr base; + if (sme_za_enabled_check(s)) { + int svl = streaming_vec_reg_size(s); + int imm = a->imm; + TCGv_ptr base = get_zarray(s, a->rv, imm, 1, 0); - if (!sme_za_enabled_check(s)) { - return true; + fn(s, base, 0, svl, a->rn, imm * svl, + s->align_mem ? MO_ALIGN_16 : MO_UNALN); } - - /* ZA[n] equates to ZA0H.B[n]. */ - base = get_tile_rowcol(s, MO_8, a->rv, 0, imm, false); - - fn(s, base, 0, svl, a->rn, imm * svl, - s->align_mem ? MO_ALIGN_16 : MO_UNALN); return true; } -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-25-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/cpu.h | 1 + target/arm/cpu64.c | 1 + 2 files changed, 2 insertions(+) diff --git a/target/arm/cpu.h b/target/arm/cpu.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/cpu.h +++ b/target/arm/cpu.h @@ -XXX,XX +XXX,XX @@ struct ArchCPU { /* Used to set the maximum vector length the cpu will support. */ uint32_t sve_max_vq; + uint32_t sme_max_vq; #ifdef CONFIG_USER_ONLY /* Used to set the default vector length at process start. */ diff --git a/target/arm/cpu64.c b/target/arm/cpu64.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/cpu64.c +++ b/target/arm/cpu64.c @@ -XXX,XX +XXX,XX @@ void arm_cpu_sme_finalize(ARMCPU *cpu, Error **errp) } cpu->sme_vq.map = vq_map; + cpu->sme_max_vq = 32 - clz32(vq_map); } static bool cpu_arm_get_sme(Object *obj, Error **errp) -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-26-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper-sme.h | 9 +++++ target/arm/tcg/translate.h | 1 + target/arm/tcg/sme.decode | 37 ++++++++++++++++++ target/arm/tcg/sme_helper.c | 64 ++++++++++++++++++++++++++++++ target/arm/tcg/translate-a64.c | 1 + target/arm/tcg/translate-sme.c | 71 ++++++++++++++++++++++++++++++++++ 6 files changed, 183 insertions(+) diff --git a/target/arm/tcg/helper-sme.h b/target/arm/tcg/helper-sme.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-sme.h +++ b/target/arm/tcg/helper-sme.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_4(sme_mova_zc_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_4(sme_mova_cz_q, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_4(sme_mova_zc_q, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_mova_cz_b, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_mova_zc_b, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_mova_cz_h, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_mova_zc_h, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_mova_cz_s, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_mova_zc_s, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_mova_cz_d, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_mova_zc_d, TCG_CALL_NO_RWG, void, ptr, ptr, i32) + DEF_HELPER_FLAGS_5(sme_ld1b_h, TCG_CALL_NO_WG, void, env, ptr, ptr, tl, i32) DEF_HELPER_FLAGS_5(sme_ld1b_v, TCG_CALL_NO_WG, void, env, ptr, ptr, tl, i32) DEF_HELPER_FLAGS_5(sme_ld1b_h_mte, TCG_CALL_NO_WG, void, env, ptr, ptr, tl, i32) diff --git a/target/arm/tcg/translate.h b/target/arm/tcg/translate.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate.h +++ b/target/arm/tcg/translate.h @@ -XXX,XX +XXX,XX @@ typedef struct DisasContext { int zt0_excp_el; /* ZT0 exception EL or 0 if enabled */ int vl; /* current vector length in bytes */ int svl; /* current streaming vector length in bytes */ + int max_svl; /* maximum implemented streaming vector length */ bool vfp_enabled; /* FP enabled via FPSCR.EN */ int vec_len; int vec_stride; diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ ZERO_zt0 11000000 01 001 00000000000 00000001 %mova_rs 13:2 !function=plus_12 &mova_p esz rs pg zr za off v:bool +&mova_t esz rs zr za off v:bool MOVA_tz 11000000 00 00000 0 v:1 .. pg:3 zr:5 0 off:4 \ &mova_p rs=%mova_rs esz=0 za=0 @@ -XXX,XX +XXX,XX @@ MOVA_zt 11000000 11 00001 0 v:1 .. pg:3 0 za:3 off:1 zr:5 \ MOVA_zt 11000000 11 00001 1 v:1 .. pg:3 0 za:4 zr:5 \ &mova_p rs=%mova_rs esz=4 off=0 +MOVA_tz2 11000000 00 00010 0 v:1 .. 000 zr:4 0 00 off:3 \ + &mova_t rs=%mova_rs esz=0 za=0 +MOVA_tz2 11000000 01 00010 0 v:1 .. 000 zr:4 0 00 za:1 off:2 \ + &mova_t rs=%mova_rs esz=1 +MOVA_tz2 11000000 10 00010 0 v:1 .. 000 zr:4 0 00 za:2 off:1 \ + &mova_t rs=%mova_rs esz=2 +MOVA_tz2 11000000 11 00010 0 v:1 .. 000 zr:4 0 00 za:3 \ + &mova_t rs=%mova_rs esz=3 off=0 + +MOVA_zt2 11000000 00 00011 0 v:1 .. 000 00 off:3 zr:4 0 \ + &mova_t rs=%mova_rs esz=0 za=0 +MOVA_zt2 11000000 01 00011 0 v:1 .. 000 00 za:1 off:2 zr:4 0 \ + &mova_t rs=%mova_rs esz=1 +MOVA_zt2 11000000 10 00011 0 v:1 .. 000 00 za:2 off:1 zr:4 0 \ + &mova_t rs=%mova_rs esz=2 +MOVA_zt2 11000000 11 00011 0 v:1 .. 000 00 za:3 zr:4 0 \ + &mova_t rs=%mova_rs esz=3 off=0 + +MOVA_tz4 11000000 00 00010 0 v:1 .. 001 zr:3 00 000 off:2 \ + &mova_t rs=%mova_rs esz=0 za=0 +MOVA_tz4 11000000 01 00010 0 v:1 .. 001 zr:3 00 000 za:1 off:1 \ + &mova_t rs=%mova_rs esz=1 +MOVA_tz4 11000000 10 00010 0 v:1 .. 001 zr:3 00 000 za:2 \ + &mova_t rs=%mova_rs esz=2 off=0 +MOVA_tz4 11000000 11 00010 0 v:1 .. 001 zr:3 00 00 za:3 \ + &mova_t rs=%mova_rs esz=3 off=0 + +MOVA_zt4 11000000 00 00011 0 v:1 .. 001 000 off:2 zr:3 00 \ + &mova_t rs=%mova_rs esz=0 za=0 +MOVA_zt4 11000000 01 00011 0 v:1 .. 001 000 za:1 off:1 zr:3 00 \ + &mova_t rs=%mova_rs esz=1 +MOVA_zt4 11000000 10 00011 0 v:1 .. 001 000 za:2 zr:3 00 \ + &mova_t rs=%mova_rs esz=2 off=0 +MOVA_zt4 11000000 11 00011 0 v:1 .. 001 00 za:3 zr:3 00 \ + &mova_t rs=%mova_rs esz=3 off=0 + ### SME Move into/from ZT0 MOVT_rzt 1100 0000 0100 1100 0 off:3 00 11111 rt:5 diff --git a/target/arm/tcg/sme_helper.c b/target/arm/tcg/sme_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme_helper.c +++ b/target/arm/tcg/sme_helper.c @@ -XXX,XX +XXX,XX @@ void HELPER(sme_mova_zc_q)(void *vd, void *za, void *vg, uint32_t desc) #undef DO_MOVA_Z +void HELPER(sme2_mova_zc_b)(void *vdst, void *vsrc, uint32_t desc) +{ + const uint8_t *src = vsrc; + uint8_t *dst = vdst; + size_t i, n = simd_oprsz(desc); + + for (i = 0; i < n; ++i) { + dst[i] = src[tile_vslice_index(i)]; + } +} + +void HELPER(sme2_mova_zc_h)(void *vdst, void *vsrc, uint32_t desc) +{ + const uint16_t *src = vsrc; + uint16_t *dst = vdst; + size_t i, n = simd_oprsz(desc) / 2; + + for (i = 0; i < n; ++i) { + dst[i] = src[tile_vslice_index(i)]; + } +} + +void HELPER(sme2_mova_zc_s)(void *vdst, void *vsrc, uint32_t desc) +{ + const uint32_t *src = vsrc; + uint32_t *dst = vdst; + size_t i, n = simd_oprsz(desc) / 4; + + for (i = 0; i < n; ++i) { + dst[i] = src[tile_vslice_index(i)]; + } +} + +void HELPER(sme2_mova_zc_d)(void *vdst, void *vsrc, uint32_t desc) +{ + const uint64_t *src = vsrc; + uint64_t *dst = vdst; + size_t i, n = simd_oprsz(desc) / 8; + + for (i = 0; i < n; ++i) { + dst[i] = src[tile_vslice_index(i)]; + } +} + /* * Clear elements in a tile slice comprising len bytes. */ @@ -XXX,XX +XXX,XX @@ static void copy_vertical_q(void *vdst, const void *vsrc, size_t len) } } +void HELPER(sme2_mova_cz_b)(void *vdst, void *vsrc, uint32_t desc) +{ + copy_vertical_b(vdst, vsrc, simd_oprsz(desc)); +} + +void HELPER(sme2_mova_cz_h)(void *vdst, void *vsrc, uint32_t desc) +{ + copy_vertical_h(vdst, vsrc, simd_oprsz(desc)); +} + +void HELPER(sme2_mova_cz_s)(void *vdst, void *vsrc, uint32_t desc) +{ + copy_vertical_s(vdst, vsrc, simd_oprsz(desc)); +} + +void HELPER(sme2_mova_cz_d)(void *vdst, void *vsrc, uint32_t desc) +{ + copy_vertical_d(vdst, vsrc, simd_oprsz(desc)); +} + /* * Host and TLB primitives for vertical tile slice addressing. */ diff --git a/target/arm/tcg/translate-a64.c b/target/arm/tcg/translate-a64.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-a64.c +++ b/target/arm/tcg/translate-a64.c @@ -XXX,XX +XXX,XX @@ static void aarch64_tr_init_disas_context(DisasContextBase *dcbase, dc->zt0_excp_el = EX_TBFLAG_A64(tb_flags, ZT0EXC_EL); dc->vl = (EX_TBFLAG_A64(tb_flags, VL) + 1) * 16; dc->svl = (EX_TBFLAG_A64(tb_flags, SVL) + 1) * 16; + dc->max_svl = arm_cpu->sme_max_vq * 16; dc->pauth_active = EX_TBFLAG_A64(tb_flags, PAUTH_ACTIVE); dc->bt = EX_TBFLAG_A64(tb_flags, BT); dc->btype = EX_TBFLAG_A64(tb_flags, BTYPE); diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ static bool do_mova_tile(DisasContext *s, arg_mova_p *a, bool to_vec) TRANS_FEAT(MOVA_tz, aa64_sme, do_mova_tile, a, false) TRANS_FEAT(MOVA_zt, aa64_sme, do_mova_tile, a, true) +static bool do_mova_tile_n(DisasContext *s, arg_mova_t *a, int n, bool to_vec) +{ + static gen_helper_gvec_2 * const cz_fns[] = { + gen_helper_sme2_mova_cz_b, gen_helper_sme2_mova_cz_h, + gen_helper_sme2_mova_cz_s, gen_helper_sme2_mova_cz_d, + }; + static gen_helper_gvec_2 * const zc_fns[] = { + gen_helper_sme2_mova_zc_b, gen_helper_sme2_mova_zc_h, + gen_helper_sme2_mova_zc_s, gen_helper_sme2_mova_zc_d, + }; + TCGv_ptr t_za; + int svl, bytes_per_op = n << a->esz; + + /* + * The MaxImplementedSVL check happens in the decode pseudocode, + * before the SM+ZA enabled check in the operation pseudocode. + * This will (currently) only fail for NREG=4, ESZ=MO_64. + */ + if (s->max_svl < bytes_per_op) { + unallocated_encoding(s); + return true; + } + + if (!sme_smza_enabled_check(s)) { + return true; + } + + svl = streaming_vec_reg_size(s); + + /* + * The CurrentVL check happens in the operation pseudocode, + * after the SM+ZA enabled check. + */ + if (svl < bytes_per_op) { + unallocated_encoding(s); + return true; + } + + if (a->v) { + TCGv_i32 t_desc = tcg_constant_i32(simd_desc(svl, svl, 0)); + + for (int i = 0; i < n; ++i) { + TCGv_ptr t_zr = vec_full_reg_ptr(s, a->zr * n + i); + t_za = get_tile_rowcol(s, a->esz, a->rs, a->za, + a->off * n + i, 1, n, a->v); + if (to_vec) { + zc_fns[a->esz](t_zr, t_za, t_desc); + } else { + cz_fns[a->esz](t_za, t_zr, t_desc); + } + } + } else { + for (int i = 0; i < n; ++i) { + int o_zr = vec_full_reg_offset(s, a->zr * n + i); + t_za = get_tile_rowcol(s, a->esz, a->rs, a->za, + a->off * n + i, 1, n, a->v); + if (to_vec) { + tcg_gen_gvec_mov_var(MO_8, tcg_env, o_zr, t_za, 0, svl, svl); + } else { + tcg_gen_gvec_mov_var(MO_8, t_za, 0, tcg_env, o_zr, svl, svl); + } + } + } + return true; +} + +TRANS_FEAT(MOVA_tz2, aa64_sme2, do_mova_tile_n, a, 2, false) +TRANS_FEAT(MOVA_tz4, aa64_sme2, do_mova_tile_n, a, 4, false) +TRANS_FEAT(MOVA_zt2, aa64_sme2, do_mova_tile_n, a, 2, true) +TRANS_FEAT(MOVA_zt4, aa64_sme2, do_mova_tile_n, a, 4, true) + static bool do_movt(DisasContext *s, arg_MOVT_rzt *a, void (*func)(TCGv_i64, TCGv_ptr, tcg_target_long)) { -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-27-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/translate.h | 5 +++++ target/arm/tcg/sme.decode | 12 ++++++++++++ target/arm/tcg/translate-sme.c | 30 ++++++++++++++++++++++++++++++ 3 files changed, 47 insertions(+) diff --git a/target/arm/tcg/translate.h b/target/arm/tcg/translate.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate.h +++ b/target/arm/tcg/translate.h @@ -XXX,XX +XXX,XX @@ static inline int plus_2(DisasContext *s, int x) return x + 2; } +static inline int plus_8(DisasContext *s, int x) +{ + return x + 8; +} + static inline int plus_12(DisasContext *s, int x) { return x + 12; diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ ZERO_zt0 11000000 01 001 00000000000 00000001 ### SME Move into/from Array %mova_rs 13:2 !function=plus_12 +%mova_rv 13:2 !function=plus_8 +&mova_a rv zr off &mova_p esz rs pg zr za off v:bool &mova_t esz rs zr za off v:bool @@ -XXX,XX +XXX,XX @@ MOVA_zt4 11000000 10 00011 0 v:1 .. 001 000 za:2 zr:3 00 \ MOVA_zt4 11000000 11 00011 0 v:1 .. 001 00 za:3 zr:3 00 \ &mova_t rs=%mova_rs esz=3 off=0 +MOVA_az2 11000000 00 00010 00 .. 010 zr:4 000 off:3 \ + &mova_a rv=%mova_rv +MOVA_az4 11000000 00 00010 00 .. 011 zr:3 0000 off:3 \ + &mova_a rv=%mova_rv + +MOVA_za2 11000000 00 00011 00 .. 010 00 off:3 zr:4 0 \ + &mova_a rv=%mova_rv +MOVA_za4 11000000 00 00011 00 .. 011 00 off:3 zr:3 00 \ + &mova_a rv=%mova_rv + ### SME Move into/from ZT0 MOVT_rzt 1100 0000 0100 1100 0 off:3 00 11111 rt:5 diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(MOVA_tz4, aa64_sme2, do_mova_tile_n, a, 4, false) TRANS_FEAT(MOVA_zt2, aa64_sme2, do_mova_tile_n, a, 2, true) TRANS_FEAT(MOVA_zt4, aa64_sme2, do_mova_tile_n, a, 4, true) +static bool do_mova_array_n(DisasContext *s, arg_mova_a *a, int n, bool to_vec) +{ + TCGv_ptr t_za; + int svl; + + if (!sme_smza_enabled_check(s)) { + return true; + } + + svl = streaming_vec_reg_size(s); + t_za = get_zarray(s, a->rv, a->off, n, 0); + + for (int i = 0; i < n; ++i) { + int o_za = (svl / n * sizeof(ARMVectorReg)) * i; + int o_zr = vec_full_reg_offset(s, a->zr * n + i); + + if (to_vec) { + tcg_gen_gvec_mov_var(MO_8, tcg_env, o_zr, t_za, o_za, svl, svl); + } else { + tcg_gen_gvec_mov_var(MO_8, t_za, o_za, tcg_env, o_zr, svl, svl); + } + } + return true; +} + +TRANS_FEAT(MOVA_az2, aa64_sme2, do_mova_array_n, a, 2, false) +TRANS_FEAT(MOVA_az4, aa64_sme2, do_mova_array_n, a, 4, false) +TRANS_FEAT(MOVA_za2, aa64_sme2, do_mova_array_n, a, 2, true) +TRANS_FEAT(MOVA_za4, aa64_sme2, do_mova_array_n, a, 4, true) + static bool do_movt(DisasContext *s, arg_MOVT_rzt *a, void (*func)(TCGv_i64, TCGv_ptr, tcg_target_long)) { -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-28-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper-sme.h | 3 +++ target/arm/tcg/sme.decode | 2 ++ target/arm/tcg/sme_helper.c | 34 ++++++++++++++++++++++++---------- target/arm/tcg/translate-sme.c | 2 ++ 4 files changed, 31 insertions(+), 10 deletions(-) diff --git a/target/arm/tcg/helper-sme.h b/target/arm/tcg/helper-sme.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-sme.h +++ b/target/arm/tcg/helper-sme.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_6(sme_sumopa_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_6(sme_usmopa_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, ptr, i32) + +DEF_HELPER_FLAGS_6(sme2_bmopa_s, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, ptr, i32) diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ SMOPA_d 1010000 0 11 0 ..... ... ... ..... . 0 ... @op_64 SUMOPA_d 1010000 0 11 1 ..... ... ... ..... . 0 ... @op_64 USMOPA_d 1010000 1 11 0 ..... ... ... ..... . 0 ... @op_64 UMOPA_d 1010000 1 11 1 ..... ... ... ..... . 0 ... @op_64 + +BMOPA 1000000 0 10 0 ..... ... ... ..... . 10 .. @op_32 diff --git a/target/arm/tcg/sme_helper.c b/target/arm/tcg/sme_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme_helper.c +++ b/target/arm/tcg/sme_helper.c @@ -XXX,XX +XXX,XX @@ DEF_IMOP_64(umopa_d, uint16_t, uint16_t) DEF_IMOP_64(sumopa_d, int16_t, uint16_t) DEF_IMOP_64(usmopa_d, uint16_t, int16_t) -#define DEF_IMOPH(NAME, S) \ - void HELPER(sme_##NAME##_##S)(void *vza, void *vzn, void *vzm, \ +#define DEF_IMOPH(P, NAME, S) \ + void HELPER(P##_##NAME##_##S)(void *vza, void *vzn, void *vzm, \ void *vpn, void *vpm, uint32_t desc) \ { do_imopa_##S(vza, vzn, vzm, vpn, vpm, desc, NAME##_##S); } -DEF_IMOPH(smopa, s) -DEF_IMOPH(umopa, s) -DEF_IMOPH(sumopa, s) -DEF_IMOPH(usmopa, s) +DEF_IMOPH(sme, smopa, s) +DEF_IMOPH(sme, umopa, s) +DEF_IMOPH(sme, sumopa, s) +DEF_IMOPH(sme, usmopa, s) -DEF_IMOPH(smopa, d) -DEF_IMOPH(umopa, d) -DEF_IMOPH(sumopa, d) -DEF_IMOPH(usmopa, d) +DEF_IMOPH(sme, smopa, d) +DEF_IMOPH(sme, umopa, d) +DEF_IMOPH(sme, sumopa, d) +DEF_IMOPH(sme, usmopa, d) + +static uint32_t bmopa_s(uint32_t n, uint32_t m, uint32_t a, uint8_t p, bool neg) +{ + uint32_t sum = ctpop32(~(n ^ m)); + if (neg) { + sum = -sum; + } + if (!(p & 1)) { + sum = 0; + } + return a + sum; +} + +DEF_IMOPH(sme2, bmopa, s) diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(SMOPA_d, aa64_sme_i16i64, do_outprod, a, MO_64, gen_helper_sme_smopa_ TRANS_FEAT(UMOPA_d, aa64_sme_i16i64, do_outprod, a, MO_64, gen_helper_sme_umopa_d) TRANS_FEAT(SUMOPA_d, aa64_sme_i16i64, do_outprod, a, MO_64, gen_helper_sme_sumopa_d) TRANS_FEAT(USMOPA_d, aa64_sme_i16i64, do_outprod, a, MO_64, gen_helper_sme_usmopa_d) + +TRANS_FEAT(BMOPA, aa64_sme2, do_outprod, a, MO_32, gen_helper_sme2_bmopa_s) -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-29-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper-sme.h | 4 ++++ target/arm/tcg/sme.decode | 2 ++ target/arm/tcg/sme_helper.c | 37 +++++++++++++++++++++++++--------- target/arm/tcg/translate-sme.c | 2 ++ 4 files changed, 35 insertions(+), 10 deletions(-) diff --git a/target/arm/tcg/helper-sme.h b/target/arm/tcg/helper-sme.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-sme.h +++ b/target/arm/tcg/helper-sme.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_6(sme_usmopa_d, TCG_CALL_NO_RWG, DEF_HELPER_FLAGS_6(sme2_bmopa_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_6(sme2_smopa2_s, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_6(sme2_umopa2_s, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, ptr, i32) diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ USMOPA_d 1010000 1 11 0 ..... ... ... ..... . 0 ... @op_64 UMOPA_d 1010000 1 11 1 ..... ... ... ..... . 0 ... @op_64 BMOPA 1000000 0 10 0 ..... ... ... ..... . 10 .. @op_32 +SMOPA2_s 1010000 0 10 0 ..... ... ... ..... . 10 .. @op_32 +UMOPA2_s 1010000 1 10 0 ..... ... ... ..... . 10 .. @op_32 diff --git a/target/arm/tcg/sme_helper.c b/target/arm/tcg/sme_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme_helper.c +++ b/target/arm/tcg/sme_helper.c @@ -XXX,XX +XXX,XX @@ static inline void do_imopa_d(uint64_t *za, uint64_t *zn, uint64_t *zm, } } -#define DEF_IMOP_32(NAME, NTYPE, MTYPE) \ +#define DEF_IMOP_8x4_32(NAME, NTYPE, MTYPE) \ static uint32_t NAME(uint32_t n, uint32_t m, uint32_t a, uint8_t p, bool neg) \ { \ uint32_t sum = 0; \ @@ -XXX,XX +XXX,XX @@ static uint32_t NAME(uint32_t n, uint32_t m, uint32_t a, uint8_t p, bool neg) \ return neg ? a - sum : a + sum; \ } -#define DEF_IMOP_64(NAME, NTYPE, MTYPE) \ +#define DEF_IMOP_16x4_64(NAME, NTYPE, MTYPE) \ static uint64_t NAME(uint64_t n, uint64_t m, uint64_t a, uint8_t p, bool neg) \ { \ uint64_t sum = 0; \ @@ -XXX,XX +XXX,XX @@ static uint64_t NAME(uint64_t n, uint64_t m, uint64_t a, uint8_t p, bool neg) \ return neg ? a - sum : a + sum; \ } -DEF_IMOP_32(smopa_s, int8_t, int8_t) -DEF_IMOP_32(umopa_s, uint8_t, uint8_t) -DEF_IMOP_32(sumopa_s, int8_t, uint8_t) -DEF_IMOP_32(usmopa_s, uint8_t, int8_t) +DEF_IMOP_8x4_32(smopa_s, int8_t, int8_t) +DEF_IMOP_8x4_32(umopa_s, uint8_t, uint8_t) +DEF_IMOP_8x4_32(sumopa_s, int8_t, uint8_t) +DEF_IMOP_8x4_32(usmopa_s, uint8_t, int8_t) -DEF_IMOP_64(smopa_d, int16_t, int16_t) -DEF_IMOP_64(umopa_d, uint16_t, uint16_t) -DEF_IMOP_64(sumopa_d, int16_t, uint16_t) -DEF_IMOP_64(usmopa_d, uint16_t, int16_t) +DEF_IMOP_16x4_64(smopa_d, int16_t, int16_t) +DEF_IMOP_16x4_64(umopa_d, uint16_t, uint16_t) +DEF_IMOP_16x4_64(sumopa_d, int16_t, uint16_t) +DEF_IMOP_16x4_64(usmopa_d, uint16_t, int16_t) #define DEF_IMOPH(P, NAME, S) \ void HELPER(P##_##NAME##_##S)(void *vza, void *vzn, void *vzm, \ @@ -XXX,XX +XXX,XX @@ static uint32_t bmopa_s(uint32_t n, uint32_t m, uint32_t a, uint8_t p, bool neg) } DEF_IMOPH(sme2, bmopa, s) + +#define DEF_IMOP_16x2_32(NAME, NTYPE, MTYPE) \ +static uint32_t NAME(uint32_t n, uint32_t m, uint32_t a, uint8_t p, bool neg) \ +{ \ + uint32_t sum = 0; \ + /* Apply P to N as a mask, making the inactive elements 0. */ \ + n &= expand_pred_h(p); \ + sum += (NTYPE)(n >> 0) * (MTYPE)(m >> 0); \ + sum += (NTYPE)(n >> 16) * (MTYPE)(m >> 16); \ + return neg ? a - sum : a + sum; \ +} + +DEF_IMOP_16x2_32(smopa2_s, int16_t, int16_t) +DEF_IMOP_16x2_32(umopa2_s, uint16_t, uint16_t) + +DEF_IMOPH(sme2, smopa2, s) +DEF_IMOPH(sme2, umopa2, s) diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(SUMOPA_d, aa64_sme_i16i64, do_outprod, a, MO_64, gen_helper_sme_sumop TRANS_FEAT(USMOPA_d, aa64_sme_i16i64, do_outprod, a, MO_64, gen_helper_sme_usmopa_d) TRANS_FEAT(BMOPA, aa64_sme2, do_outprod, a, MO_32, gen_helper_sme2_bmopa_s) +TRANS_FEAT(SMOPA2_s, aa64_sme2, do_outprod, a, MO_32, gen_helper_sme2_smopa2_s) +TRANS_FEAT(UMOPA2_s, aa64_sme2, do_outprod, a, MO_32, gen_helper_sme2_umopa2_s) -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> To be used by both SVE2 and SME2. Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-30-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/translate-a64.h | 4 ++++ target/arm/tcg/gengvec64.c | 11 +++++++++++ target/arm/tcg/translate-sve.c | 8 +------- 3 files changed, 16 insertions(+), 7 deletions(-) diff --git a/target/arm/tcg/translate-a64.h b/target/arm/tcg/translate-a64.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-a64.h +++ b/target/arm/tcg/translate-a64.h @@ -XXX,XX +XXX,XX @@ void gen_gvec_usqadd_qc(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs, uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz); +void gen_gvec_sve2_sqdmulh(unsigned vece, uint32_t rd_ofs, + uint32_t rn_ofs, uint32_t rm_ofs, + uint32_t opr_sz, uint32_t max_sz); + void gen_sve_ldr(DisasContext *s, TCGv_ptr, int vofs, int len, int rn, int imm, MemOp align); void gen_sve_str(DisasContext *s, TCGv_ptr, int vofs, diff --git a/target/arm/tcg/gengvec64.c b/target/arm/tcg/gengvec64.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/gengvec64.c +++ b/target/arm/tcg/gengvec64.c @@ -XXX,XX +XXX,XX @@ void gen_gvec_usqadd_qc(unsigned vece, uint32_t rd_ofs, tcg_gen_gvec_4(rd_ofs, offsetof(CPUARMState, vfp.qc), rn_ofs, rm_ofs, opr_sz, max_sz, &ops[vece]); } + +void gen_gvec_sve2_sqdmulh(unsigned vece, uint32_t rd_ofs, + uint32_t rn_ofs, uint32_t rm_ofs, + uint32_t opr_sz, uint32_t max_sz) +{ + static gen_helper_gvec_3 * const fns[4] = { + gen_helper_sve2_sqdmulh_b, gen_helper_sve2_sqdmulh_h, + gen_helper_sve2_sqdmulh_s, gen_helper_sve2_sqdmulh_d, + }; + tcg_gen_gvec_3_ool(rd_ofs, rn_ofs, rm_ofs, opr_sz, max_sz, 0, fns[vece]); +} diff --git a/target/arm/tcg/translate-sve.c b/target/arm/tcg/translate-sve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sve.c +++ b/target/arm/tcg/translate-sve.c @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(MOVPRFX_z, aa64_sve, do_movz_zpz, a->rd, a->rn, a->pg, a->esz, false) */ TRANS_FEAT(MUL_zzz, aa64_sve2, gen_gvec_fn_arg_zzz, tcg_gen_gvec_mul, a) +TRANS_FEAT(SQDMULH_zzz, aa64_sve2, gen_gvec_fn_arg_zzz, gen_gvec_sve2_sqdmulh, a) static gen_helper_gvec_3 * const smulh_zzz_fns[4] = { gen_helper_gvec_smulh_b, gen_helper_gvec_smulh_h, @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(UMULH_zzz, aa64_sve2, gen_gvec_ool_arg_zzz, TRANS_FEAT(PMUL_zzz, aa64_sve2, gen_gvec_ool_arg_zzz, gen_helper_gvec_pmul_b, a, 0) -static gen_helper_gvec_3 * const sqdmulh_zzz_fns[4] = { - gen_helper_sve2_sqdmulh_b, gen_helper_sve2_sqdmulh_h, - gen_helper_sve2_sqdmulh_s, gen_helper_sve2_sqdmulh_d, -}; -TRANS_FEAT(SQDMULH_zzz, aa64_sve2, gen_gvec_ool_arg_zzz, - sqdmulh_zzz_fns[a->esz], a, 0) - static gen_helper_gvec_3 * const sqrdmulh_zzz_fns[4] = { gen_helper_sve2_sqrdmulh_b, gen_helper_sve2_sqrdmulh_h, gen_helper_sve2_sqrdmulh_s, gen_helper_sve2_sqrdmulh_d, -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-31-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper-sme.h | 13 ++++ target/arm/tcg/helper.h | 8 ++ target/arm/tcg/vec_internal.h | 4 + target/arm/tcg/sme.decode | 40 ++++++++++ target/arm/tcg/helper-a64.c | 2 + target/arm/tcg/neon_helper.c | 30 ++++++++ target/arm/tcg/translate-sme.c | 137 +++++++++++++++++++++++++++++++++ target/arm/tcg/vec_helper.c | 7 ++ 8 files changed, 241 insertions(+) diff --git a/target/arm/tcg/helper-sme.h b/target/arm/tcg/helper-sme.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-sme.h +++ b/target/arm/tcg/helper-sme.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_6(sme2_smopa2_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_6(sme2_umopa2_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, ptr, i32) + +DEF_HELPER_FLAGS_5(gvec_fmax_b16, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_5(gvec_fmin_b16, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_5(gvec_ah_fmax_b16, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_5(gvec_ah_fmin_b16, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_5(gvec_fmaxnum_b16, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_5(gvec_fminnum_b16, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, fpst, i32) diff --git a/target/arm/tcg/helper.h b/target/arm/tcg/helper.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper.h +++ b/target/arm/tcg/helper.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_4(gvec_urshl_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_4(gvec_urshl_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_4(gvec_urshl_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sme2_srshl_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sme2_srshl_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sme2_srshl_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) + +DEF_HELPER_FLAGS_4(sme2_urshl_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sme2_urshl_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sme2_urshl_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) + DEF_HELPER_2(neon_add_u8, i32, i32, i32) DEF_HELPER_2(neon_add_u16, i32, i32, i32) DEF_HELPER_2(neon_sub_u8, i32, i32, i32) diff --git a/target/arm/tcg/vec_internal.h b/target/arm/tcg/vec_internal.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/vec_internal.h +++ b/target/arm/tcg/vec_internal.h @@ -XXX,XX +XXX,XX @@ static inline float64 float64_maybe_ah_chs(float64 a, bool fpcr_ah) return fpcr_ah && float64_is_any_nan(a) ? a : float64_chs(a); } +/* Not actually called directly as a helper, but uses similar machinery. */ +bfloat16 helper_sme2_ah_fmax_b16(bfloat16 a, bfloat16 b, float_status *fpst); +bfloat16 helper_sme2_ah_fmin_b16(bfloat16 a, bfloat16 b, float_status *fpst); + #endif /* TARGET_ARM_VEC_INTERNAL_H */ diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ UMOPA_d 1010000 1 11 1 ..... ... ... ..... . 0 ... @op_64 BMOPA 1000000 0 10 0 ..... ... ... ..... . 10 .. @op_32 SMOPA2_s 1010000 0 10 0 ..... ... ... ..... . 10 .. @op_32 UMOPA2_s 1010000 1 10 0 ..... ... ... ..... . 10 .. @op_32 + +### SME2 Multi-vector Multiple and Single SVE Destructive + +%zd_ax2 1:4 !function=times_2 +%zd_ax4 2:3 !function=times_4 + +&z2z_en zdn zm esz n +@z2z_2x1 ....... . esz:2 .. zm:4 ....0. ..... .... . \ + &z2z_en n=2 zdn=%zd_ax2 +@z2z_4x1 ....... . esz:2 .. zm:4 ....1. ..... ...0 . \ + &z2z_en n=4 zdn=%zd_ax4 + +SMAX_n1 1100000 1 .. 10 .... 1010.0 00000 .... 0 @z2z_2x1 +SMAX_n1 1100000 1 .. 10 .... 1010.0 00000 .... 0 @z2z_4x1 +UMAX_n1 1100000 1 .. 10 .... 1010.0 00000 .... 1 @z2z_2x1 +UMAX_n1 1100000 1 .. 10 .... 1010.0 00000 .... 1 @z2z_4x1 +SMIN_n1 1100000 1 .. 10 .... 1010.0 00001 .... 0 @z2z_2x1 +SMIN_n1 1100000 1 .. 10 .... 1010.0 00001 .... 0 @z2z_4x1 +UMIN_n1 1100000 1 .. 10 .... 1010.0 00001 .... 1 @z2z_2x1 +UMIN_n1 1100000 1 .. 10 .... 1010.0 00001 .... 1 @z2z_4x1 + +FMAX_n1 1100000 1 .. 10 .... 1010.0 01000 .... 0 @z2z_2x1 +FMAX_n1 1100000 1 .. 10 .... 1010.0 01000 .... 0 @z2z_4x1 +FMIN_n1 1100000 1 .. 10 .... 1010.0 01000 .... 1 @z2z_2x1 +FMIN_n1 1100000 1 .. 10 .... 1010.0 01000 .... 1 @z2z_4x1 +FMAXNM_n1 1100000 1 .. 10 .... 1010.0 01001 .... 0 @z2z_2x1 +FMAXNM_n1 1100000 1 .. 10 .... 1010.0 01001 .... 0 @z2z_4x1 +FMINNM_n1 1100000 1 .. 10 .... 1010.0 01001 .... 1 @z2z_2x1 +FMINNM_n1 1100000 1 .. 10 .... 1010.0 01001 .... 1 @z2z_4x1 + +SRSHL_n1 1100000 1 .. 10 .... 1010.0 10001 .... 0 @z2z_2x1 +SRSHL_n1 1100000 1 .. 10 .... 1010.0 10001 .... 0 @z2z_4x1 +URSHL_n1 1100000 1 .. 10 .... 1010.0 10001 .... 1 @z2z_2x1 +URSHL_n1 1100000 1 .. 10 .... 1010.0 10001 .... 1 @z2z_4x1 + +ADD_n1 1100000 1 .. 10 .... 1010.0 11000 .... 0 @z2z_2x1 +ADD_n1 1100000 1 .. 10 .... 1010.0 11000 .... 0 @z2z_4x1 + +SQDMULH_n1 1100000 1 .. 10 .... 1010.1 00000 .... 0 @z2z_2x1 +SQDMULH_n1 1100000 1 .. 10 .... 1010.1 00000 .... 0 @z2z_4x1 diff --git a/target/arm/tcg/helper-a64.c b/target/arm/tcg/helper-a64.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-a64.c +++ b/target/arm/tcg/helper-a64.c @@ -XXX,XX +XXX,XX @@ AH_MINMAX_HELPER(vfp_ah_mind, float64, float64, min) AH_MINMAX_HELPER(vfp_ah_maxh, dh_ctype_f16, float16, max) AH_MINMAX_HELPER(vfp_ah_maxs, float32, float32, max) AH_MINMAX_HELPER(vfp_ah_maxd, float64, float64, max) +AH_MINMAX_HELPER(sme2_ah_fmax_b16, bfloat16, bfloat16, max) +AH_MINMAX_HELPER(sme2_ah_fmin_b16, bfloat16, bfloat16, min) /* 64-bit versions of the CRC helpers. Note that although the operation * (and the prototypes of crc32c() and crc32() mean that only the bottom diff --git a/target/arm/tcg/neon_helper.c b/target/arm/tcg/neon_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/neon_helper.c +++ b/target/arm/tcg/neon_helper.c @@ -XXX,XX +XXX,XX @@ NEON_VOP(rshl_s16, neon_s16, 2) NEON_GVEC_VOP2(gvec_srshl_h, int16_t) #undef NEON_FN +#define NEON_FN(dest, src1, src2) \ + (dest = do_sqrshl_bhs(src1, src2, 16, true, NULL)) +NEON_GVEC_VOP2(sme2_srshl_h, int16_t) +#undef NEON_FN + #define NEON_FN(dest, src1, src2) \ (dest = do_sqrshl_bhs(src1, (int8_t)src2, 32, true, NULL)) NEON_GVEC_VOP2(gvec_srshl_s, int32_t) #undef NEON_FN +#define NEON_FN(dest, src1, src2) \ + (dest = do_sqrshl_bhs(src1, src2, 32, true, NULL)) +NEON_GVEC_VOP2(sme2_srshl_s, int32_t) +#undef NEON_FN + #define NEON_FN(dest, src1, src2) \ (dest = do_sqrshl_d(src1, (int8_t)src2, true, NULL)) NEON_GVEC_VOP2(gvec_srshl_d, int64_t) #undef NEON_FN +#define NEON_FN(dest, src1, src2) \ + (dest = do_sqrshl_d(src1, src2, true, NULL)) +NEON_GVEC_VOP2(sme2_srshl_d, int64_t) +#undef NEON_FN + uint32_t HELPER(neon_rshl_s32)(uint32_t val, uint32_t shift) { return do_sqrshl_bhs(val, (int8_t)shift, 32, true, NULL); @@ -XXX,XX +XXX,XX @@ NEON_VOP(rshl_u16, neon_u16, 2) NEON_GVEC_VOP2(gvec_urshl_h, uint16_t) #undef NEON_FN +#define NEON_FN(dest, src1, src2) \ + (dest = do_uqrshl_bhs(src1, (int16_t)src2, 16, true, NULL)) +NEON_GVEC_VOP2(sme2_urshl_h, uint16_t) +#undef NEON_FN + #define NEON_FN(dest, src1, src2) \ (dest = do_uqrshl_bhs(src1, (int8_t)src2, 32, true, NULL)) NEON_GVEC_VOP2(gvec_urshl_s, int32_t) #undef NEON_FN +#define NEON_FN(dest, src1, src2) \ + (dest = do_uqrshl_bhs(src1, src2, 32, true, NULL)) +NEON_GVEC_VOP2(sme2_urshl_s, int32_t) +#undef NEON_FN + #define NEON_FN(dest, src1, src2) \ (dest = do_uqrshl_d(src1, (int8_t)src2, true, NULL)) NEON_GVEC_VOP2(gvec_urshl_d, int64_t) #undef NEON_FN +#define NEON_FN(dest, src1, src2) \ + (dest = do_uqrshl_d(src1, src2, true, NULL)) +NEON_GVEC_VOP2(sme2_urshl_d, int64_t) +#undef NEON_FN + uint32_t HELPER(neon_rshl_u32)(uint32_t val, uint32_t shift) { return do_uqrshl_bhs(val, (int8_t)shift, 32, true, NULL); diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(USMOPA_d, aa64_sme_i16i64, do_outprod, a, MO_64, gen_helper_sme_usmop TRANS_FEAT(BMOPA, aa64_sme2, do_outprod, a, MO_32, gen_helper_sme2_bmopa_s) TRANS_FEAT(SMOPA2_s, aa64_sme2, do_outprod, a, MO_32, gen_helper_sme2_smopa2_s) TRANS_FEAT(UMOPA2_s, aa64_sme2, do_outprod, a, MO_32, gen_helper_sme2_umopa2_s) + +static bool do_z2z_n1(DisasContext *s, arg_z2z_en *a, GVecGen3Fn *fn) +{ + int esz, dn, vsz, mofs, n; + bool overlap = false; + + if (!sme_sm_enabled_check(s)) { + return true; + } + + esz = a->esz; + n = a->n; + dn = a->zdn; + mofs = vec_full_reg_offset(s, a->zm); + vsz = streaming_vec_reg_size(s); + + for (int i = 0; i < n; i++) { + int dofs = vec_full_reg_offset(s, dn + i); + if (dofs == mofs) { + overlap = true; + } else { + fn(esz, dofs, dofs, mofs, vsz, vsz); + } + } + if (overlap) { + fn(esz, mofs, mofs, mofs, vsz, vsz); + } + return true; +} + +static void gen_sme2_srshl(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs, + uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz) +{ + static gen_helper_gvec_3 * const fns[] = { + gen_helper_gvec_srshl_b, gen_helper_sme2_srshl_h, + gen_helper_sme2_srshl_s, gen_helper_sme2_srshl_d, + }; + tcg_debug_assert(vece <= MO_64); + tcg_gen_gvec_3_ool(rd_ofs, rn_ofs, rm_ofs, opr_sz, max_sz, 0, fns[vece]); +} + +static void gen_sme2_urshl(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs, + uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz) +{ + static gen_helper_gvec_3 * const fns[] = { + gen_helper_gvec_urshl_b, gen_helper_sme2_urshl_h, + gen_helper_sme2_urshl_s, gen_helper_sme2_urshl_d, + }; + tcg_debug_assert(vece <= MO_64); + tcg_gen_gvec_3_ool(rd_ofs, rn_ofs, rm_ofs, opr_sz, max_sz, 0, fns[vece]); +} + +TRANS_FEAT(ADD_n1, aa64_sme2, do_z2z_n1, a, tcg_gen_gvec_add) +TRANS_FEAT(SMAX_n1, aa64_sme2, do_z2z_n1, a, tcg_gen_gvec_smax) +TRANS_FEAT(SMIN_n1, aa64_sme2, do_z2z_n1, a, tcg_gen_gvec_smin) +TRANS_FEAT(UMAX_n1, aa64_sme2, do_z2z_n1, a, tcg_gen_gvec_umax) +TRANS_FEAT(UMIN_n1, aa64_sme2, do_z2z_n1, a, tcg_gen_gvec_umin) +TRANS_FEAT(SRSHL_n1, aa64_sme2, do_z2z_n1, a, gen_sme2_srshl) +TRANS_FEAT(URSHL_n1, aa64_sme2, do_z2z_n1, a, gen_sme2_urshl) +TRANS_FEAT(SQDMULH_n1, aa64_sme2, do_z2z_n1, a, gen_gvec_sve2_sqdmulh) + +static bool do_z2z_n1_fpst(DisasContext *s, arg_z2z_en *a, + gen_helper_gvec_3_ptr * const fns[4]) +{ + int esz = a->esz, n, dn, vsz, mofs; + bool overlap = false; + gen_helper_gvec_3_ptr *fn; + TCGv_ptr fpst; + + /* These insns use MO_8 to encode BFloat16. */ + if (esz == MO_8 && !dc_isar_feature(aa64_sme_b16b16, s)) { + return false; + } + if (!sme_sm_enabled_check(s)) { + return true; + } + + fpst = fpstatus_ptr(esz == MO_16 ? FPST_A64_F16 : FPST_A64); + fn = fns[esz]; + n = a->n; + dn = a->zdn; + mofs = vec_full_reg_offset(s, a->zm); + vsz = streaming_vec_reg_size(s); + + for (int i = 0; i < n; i++) { + int dofs = vec_full_reg_offset(s, dn + i); + if (dofs == mofs) { + overlap = true; + } else { + tcg_gen_gvec_3_ptr(dofs, dofs, mofs, fpst, vsz, vsz, 0, fn); + } + } + if (overlap) { + tcg_gen_gvec_3_ptr(mofs, mofs, mofs, fpst, vsz, vsz, 0, fn); + } + return true; +} + +static gen_helper_gvec_3_ptr * const f_vector_fmax[2][4] = { + { gen_helper_gvec_fmax_b16, + gen_helper_gvec_fmax_h, + gen_helper_gvec_fmax_s, + gen_helper_gvec_fmax_d }, + { gen_helper_gvec_ah_fmax_b16, + gen_helper_gvec_ah_fmax_h, + gen_helper_gvec_ah_fmax_s, + gen_helper_gvec_ah_fmax_d }, +}; +TRANS_FEAT(FMAX_n1, aa64_sme2, do_z2z_n1_fpst, a, f_vector_fmax[s->fpcr_ah]) + +static gen_helper_gvec_3_ptr * const f_vector_fmin[2][4] = { + { gen_helper_gvec_fmin_b16, + gen_helper_gvec_fmin_h, + gen_helper_gvec_fmin_s, + gen_helper_gvec_fmin_d }, + { gen_helper_gvec_ah_fmin_b16, + gen_helper_gvec_ah_fmin_h, + gen_helper_gvec_ah_fmin_s, + gen_helper_gvec_ah_fmin_d }, +}; +TRANS_FEAT(FMIN_n1, aa64_sme2, do_z2z_n1_fpst, a, f_vector_fmin[s->fpcr_ah]) + +static gen_helper_gvec_3_ptr * const f_vector_fmaxnm[4] = { + gen_helper_gvec_fmaxnum_b16, + gen_helper_gvec_fmaxnum_h, + gen_helper_gvec_fmaxnum_s, + gen_helper_gvec_fmaxnum_d, +}; +TRANS_FEAT(FMAXNM_n1, aa64_sme2, do_z2z_n1_fpst, a, f_vector_fmaxnm) + +static gen_helper_gvec_3_ptr * const f_vector_fminnm[4] = { + gen_helper_gvec_fminnum_b16, + gen_helper_gvec_fminnum_h, + gen_helper_gvec_fminnum_s, + gen_helper_gvec_fminnum_d, +}; +TRANS_FEAT(FMINNM_n1, aa64_sme2, do_z2z_n1_fpst, a, f_vector_fminnm) diff --git a/target/arm/tcg/vec_helper.c b/target/arm/tcg/vec_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/vec_helper.c +++ b/target/arm/tcg/vec_helper.c @@ -XXX,XX +XXX,XX @@ DO_3OP(gvec_ah_fmin_h, helper_vfp_ah_minh, float16) DO_3OP(gvec_ah_fmin_s, helper_vfp_ah_mins, float32) DO_3OP(gvec_ah_fmin_d, helper_vfp_ah_mind, float64) +DO_3OP(gvec_fmax_b16, bfloat16_max, bfloat16) +DO_3OP(gvec_fmin_b16, bfloat16_min, bfloat16) +DO_3OP(gvec_fmaxnum_b16, bfloat16_maxnum, bfloat16) +DO_3OP(gvec_fminnum_b16, bfloat16_minnum, bfloat16) +DO_3OP(gvec_ah_fmax_b16, helper_sme2_ah_fmax_b16, bfloat16) +DO_3OP(gvec_ah_fmin_b16, helper_sme2_ah_fmin_b16, bfloat16) + #endif #undef DO_3OP -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-32-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/sme.decode | 36 +++++++++++++++++++ target/arm/tcg/translate-sme.c | 65 ++++++++++++++++++++++++++++++++++ 2 files changed, 101 insertions(+) diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ ADD_n1 1100000 1 .. 10 .... 1010.0 11000 .... 0 @z2z_4x1 SQDMULH_n1 1100000 1 .. 10 .... 1010.1 00000 .... 0 @z2z_2x1 SQDMULH_n1 1100000 1 .. 10 .... 1010.1 00000 .... 0 @z2z_4x1 + +### SME2 Multi-vector Multiple Vectors SVE Destructive + +%zm_ax2 17:4 !function=times_2 +%zm_ax4 18:3 !function=times_4 + +@z2z_2x2 ....... . esz:2 . ....0 ....0. ..... .... . \ + &z2z_en n=2 zdn=%zd_ax2 zm=%zm_ax2 +@z2z_4x4 ....... . esz:2 . ...00 ....1. ..... ...0 . \ + &z2z_en n=4 zdn=%zd_ax4 zm=%zm_ax4 + +SMAX_nn 1100000 1 .. 1 ..... 1011.0 00000 .... 0 @z2z_2x2 +SMAX_nn 1100000 1 .. 1 ..... 1011.0 00000 .... 0 @z2z_4x4 +UMAX_nn 1100000 1 .. 1 ..... 1011.0 00000 .... 1 @z2z_2x2 +UMAX_nn 1100000 1 .. 1 ..... 1011.0 00000 .... 1 @z2z_4x4 +SMIN_nn 1100000 1 .. 1 ..... 1011.0 00001 .... 0 @z2z_2x2 +SMIN_nn 1100000 1 .. 1 ..... 1011.0 00001 .... 0 @z2z_4x4 +UMIN_nn 1100000 1 .. 1 ..... 1011.0 00001 .... 1 @z2z_2x2 +UMIN_nn 1100000 1 .. 1 ..... 1011.0 00001 .... 1 @z2z_4x4 + +FMAX_nn 1100000 1 .. 1 ..... 1011.0 01000 .... 0 @z2z_2x2 +FMAX_nn 1100000 1 .. 1 ..... 1011.0 01000 .... 0 @z2z_4x4 +FMIN_nn 1100000 1 .. 1 ..... 1011.0 01000 .... 1 @z2z_2x2 +FMIN_nn 1100000 1 .. 1 ..... 1011.0 01000 .... 1 @z2z_4x4 +FMAXNM_nn 1100000 1 .. 1 ..... 1011.0 01001 .... 0 @z2z_2x2 +FMAXNM_nn 1100000 1 .. 1 ..... 1011.0 01001 .... 0 @z2z_4x4 +FMINNM_nn 1100000 1 .. 1 ..... 1011.0 01001 .... 1 @z2z_2x2 +FMINNM_nn 1100000 1 .. 1 ..... 1011.0 01001 .... 1 @z2z_4x4 + +SRSHL_nn 1100000 1 .. 1 ..... 1011.0 10001 .... 0 @z2z_2x2 +SRSHL_nn 1100000 1 .. 1 ..... 1011.0 10001 .... 0 @z2z_4x4 +URSHL_nn 1100000 1 .. 1 ..... 1011.0 10001 .... 1 @z2z_2x2 +URSHL_nn 1100000 1 .. 1 ..... 1011.0 10001 .... 1 @z2z_4x4 + +SQDMULH_nn 1100000 1 .. 1 ..... 1011.1 00000 .... 0 @z2z_2x2 +SQDMULH_nn 1100000 1 .. 1 ..... 1011.1 00000 .... 0 @z2z_4x4 diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(SRSHL_n1, aa64_sme2, do_z2z_n1, a, gen_sme2_srshl) TRANS_FEAT(URSHL_n1, aa64_sme2, do_z2z_n1, a, gen_sme2_urshl) TRANS_FEAT(SQDMULH_n1, aa64_sme2, do_z2z_n1, a, gen_gvec_sve2_sqdmulh) +static bool do_z2z_nn(DisasContext *s, arg_z2z_en *a, GVecGen3Fn *fn) +{ + int esz, dn, dm, vsz, n; + + if (!sme_sm_enabled_check(s)) { + return true; + } + + esz = a->esz; + n = a->n; + dn = a->zdn; + dm = a->zm; + vsz = streaming_vec_reg_size(s); + + for (int i = 0; i < n; i++) { + int dofs = vec_full_reg_offset(s, dn + i); + int mofs = vec_full_reg_offset(s, dm + i); + + fn(esz, dofs, dofs, mofs, vsz, vsz); + } + return true; +} + +TRANS_FEAT(SMAX_nn, aa64_sme2, do_z2z_nn, a, tcg_gen_gvec_smax) +TRANS_FEAT(SMIN_nn, aa64_sme2, do_z2z_nn, a, tcg_gen_gvec_smin) +TRANS_FEAT(UMAX_nn, aa64_sme2, do_z2z_nn, a, tcg_gen_gvec_umax) +TRANS_FEAT(UMIN_nn, aa64_sme2, do_z2z_nn, a, tcg_gen_gvec_umin) +TRANS_FEAT(SRSHL_nn, aa64_sme2, do_z2z_nn, a, gen_sme2_srshl) +TRANS_FEAT(URSHL_nn, aa64_sme2, do_z2z_nn, a, gen_sme2_urshl) +TRANS_FEAT(SQDMULH_nn, aa64_sme2, do_z2z_nn, a, gen_gvec_sve2_sqdmulh) + static bool do_z2z_n1_fpst(DisasContext *s, arg_z2z_en *a, gen_helper_gvec_3_ptr * const fns[4]) { @@ -XXX,XX +XXX,XX @@ static bool do_z2z_n1_fpst(DisasContext *s, arg_z2z_en *a, return true; } +static bool do_z2z_nn_fpst(DisasContext *s, arg_z2z_en *a, + gen_helper_gvec_3_ptr * const fns[4]) +{ + int esz = a->esz, n, dn, dm, vsz; + gen_helper_gvec_3_ptr *fn; + TCGv_ptr fpst; + + if (esz == MO_8 && !dc_isar_feature(aa64_sme_b16b16, s)) { + return false; + } + if (!sme_sm_enabled_check(s)) { + return true; + } + + fpst = fpstatus_ptr(esz == MO_16 ? FPST_A64_F16 : FPST_A64); + fn = fns[esz]; + n = a->n; + dn = a->zdn; + dm = a->zm; + vsz = streaming_vec_reg_size(s); + + for (int i = 0; i < n; i++) { + int dofs = vec_full_reg_offset(s, dn + i); + int mofs = vec_full_reg_offset(s, dm + i); + + tcg_gen_gvec_3_ptr(dofs, dofs, mofs, fpst, vsz, vsz, 0, fn); + } + return true; +} + static gen_helper_gvec_3_ptr * const f_vector_fmax[2][4] = { { gen_helper_gvec_fmax_b16, gen_helper_gvec_fmax_h, @@ -XXX,XX +XXX,XX @@ static gen_helper_gvec_3_ptr * const f_vector_fmax[2][4] = { gen_helper_gvec_ah_fmax_d }, }; TRANS_FEAT(FMAX_n1, aa64_sme2, do_z2z_n1_fpst, a, f_vector_fmax[s->fpcr_ah]) +TRANS_FEAT(FMAX_nn, aa64_sme2, do_z2z_nn_fpst, a, f_vector_fmax[s->fpcr_ah]) static gen_helper_gvec_3_ptr * const f_vector_fmin[2][4] = { { gen_helper_gvec_fmin_b16, @@ -XXX,XX +XXX,XX @@ static gen_helper_gvec_3_ptr * const f_vector_fmin[2][4] = { gen_helper_gvec_ah_fmin_d }, }; TRANS_FEAT(FMIN_n1, aa64_sme2, do_z2z_n1_fpst, a, f_vector_fmin[s->fpcr_ah]) +TRANS_FEAT(FMIN_nn, aa64_sme2, do_z2z_nn_fpst, a, f_vector_fmin[s->fpcr_ah]) static gen_helper_gvec_3_ptr * const f_vector_fmaxnm[4] = { gen_helper_gvec_fmaxnum_b16, @@ -XXX,XX +XXX,XX @@ static gen_helper_gvec_3_ptr * const f_vector_fmaxnm[4] = { gen_helper_gvec_fmaxnum_d, }; TRANS_FEAT(FMAXNM_n1, aa64_sme2, do_z2z_n1_fpst, a, f_vector_fmaxnm) +TRANS_FEAT(FMAXNM_nn, aa64_sme2, do_z2z_nn_fpst, a, f_vector_fmaxnm) static gen_helper_gvec_3_ptr * const f_vector_fminnm[4] = { gen_helper_gvec_fminnum_b16, @@ -XXX,XX +XXX,XX @@ static gen_helper_gvec_3_ptr * const f_vector_fminnm[4] = { gen_helper_gvec_fminnum_d, }; TRANS_FEAT(FMINNM_n1, aa64_sme2, do_z2z_n1_fpst, a, f_vector_fminnm) +TRANS_FEAT(FMINNM_nn, aa64_sme2, do_z2z_nn_fpst, a, f_vector_fminnm) -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-33-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/translate.h | 2 ++ target/arm/tcg/sme.decode | 15 +++++++++++++++ target/arm/tcg/translate-sme.c | 30 ++++++++++++++++++++++++++++++ 3 files changed, 47 insertions(+) diff --git a/target/arm/tcg/translate.h b/target/arm/tcg/translate.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate.h +++ b/target/arm/tcg/translate.h @@ -XXX,XX +XXX,XX @@ typedef void GVecGen3Fn(unsigned, uint32_t, uint32_t, uint32_t, uint32_t, uint32_t); typedef void GVecGen4Fn(unsigned, uint32_t, uint32_t, uint32_t, uint32_t, uint32_t, uint32_t); +typedef void GVecGen3FnVar(unsigned, TCGv_ptr, uint32_t, TCGv_ptr, uint32_t, + TCGv_ptr, uint32_t, uint32_t, uint32_t); /* Function prototype for gen_ functions for calling Neon helpers */ typedef void NeonGenOneOpFn(TCGv_i32, TCGv_i32); diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ URSHL_nn 1100000 1 .. 1 ..... 1011.0 10001 .... 1 @z2z_4x4 SQDMULH_nn 1100000 1 .. 1 ..... 1011.1 00000 .... 0 @z2z_2x2 SQDMULH_nn 1100000 1 .. 1 ..... 1011.1 00000 .... 0 @z2z_4x4 + +### SME2 Multi-vector Multiple and Single Array Vectors + +&azz_n n off rv zn zm +@azz_nx1_o3 ........ .... zm:4 ...... zn:5 .. off:3 &azz_n rv=%mova_rv + +ADD_azz_n1_s 11000001 0010 .... 0 .. 110 ..... 10 ... @azz_nx1_o3 n=2 +ADD_azz_n1_s 11000001 0011 .... 0 .. 110 ..... 10 ... @azz_nx1_o3 n=4 +ADD_azz_n1_d 11000001 0110 .... 0 .. 110 ..... 10 ... @azz_nx1_o3 n=2 +ADD_azz_n1_d 11000001 0111 .... 0 .. 110 ..... 10 ... @azz_nx1_o3 n=4 + +SUB_azz_n1_s 11000001 0010 .... 0 .. 110 ..... 11 ... @azz_nx1_o3 n=2 +SUB_azz_n1_s 11000001 0011 .... 0 .. 110 ..... 11 ... @azz_nx1_o3 n=4 +SUB_azz_n1_d 11000001 0110 .... 0 .. 110 ..... 11 ... @azz_nx1_o3 n=2 +SUB_azz_n1_d 11000001 0111 .... 0 .. 110 ..... 11 ... @azz_nx1_o3 n=4 diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ static gen_helper_gvec_3_ptr * const f_vector_fminnm[4] = { }; TRANS_FEAT(FMINNM_n1, aa64_sme2, do_z2z_n1_fpst, a, f_vector_fminnm) TRANS_FEAT(FMINNM_nn, aa64_sme2, do_z2z_nn_fpst, a, f_vector_fminnm) + +/* Add/Sub vector Z[m] to each Z[n*N] with result in ZA[d*N]. */ +static bool do_azz_n1(DisasContext *s, arg_azz_n *a, int esz, + GVecGen3FnVar *fn) +{ + TCGv_ptr t_za; + int svl, n, o_zm; + + if (!sme_smza_enabled_check(s)) { + return true; + } + + n = a->n; + t_za = get_zarray(s, a->rv, a->off, n, 0); + o_zm = vec_full_reg_offset(s, a->zm); + svl = streaming_vec_reg_size(s); + + for (int i = 0; i < n; ++i) { + int o_za = (svl / n * sizeof(ARMVectorReg)) * i; + int o_zn = vec_full_reg_offset(s, (a->zn + i) % 32); + + fn(esz, t_za, o_za, tcg_env, o_zn, tcg_env, o_zm, svl, svl); + } + return true; +} + +TRANS_FEAT(ADD_azz_n1_s, aa64_sme2, do_azz_n1, a, MO_32, tcg_gen_gvec_add_var) +TRANS_FEAT(SUB_azz_n1_s, aa64_sme2, do_azz_n1, a, MO_32, tcg_gen_gvec_sub_var) +TRANS_FEAT(ADD_azz_n1_d, aa64_sme2_i16i64, do_azz_n1, a, MO_64, tcg_gen_gvec_add_var) +TRANS_FEAT(SUB_azz_n1_d, aa64_sme2_i16i64, do_azz_n1, a, MO_64, tcg_gen_gvec_sub_var) -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-34-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/sme.decode | 20 ++++++++++++++++++++ target/arm/tcg/translate-sme.c | 30 ++++++++++++++++++++++++++++++ 2 files changed, 50 insertions(+) diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ SUB_azz_n1_s 11000001 0010 .... 0 .. 110 ..... 11 ... @azz_nx1_o3 n=2 SUB_azz_n1_s 11000001 0011 .... 0 .. 110 ..... 11 ... @azz_nx1_o3 n=4 SUB_azz_n1_d 11000001 0110 .... 0 .. 110 ..... 11 ... @azz_nx1_o3 n=2 SUB_azz_n1_d 11000001 0111 .... 0 .. 110 ..... 11 ... @azz_nx1_o3 n=4 + +### SME2 Multi-vector Multiple Array Vectors + +%zn_ax2 6:4 !function=times_2 +%zn_ax4 7:3 !function=times_4 + +@azz_2x2_o3 ........ ... ..... . .. ... ..... .. off:3 \ + &azz_n n=2 rv=%mova_rv zn=%zn_ax2 zm=%zm_ax2 +@azz_4x4_o3 ........ ... ..... . .. ... ..... .. off:3 \ + &azz_n n=4 rv=%mova_rv zn=%zn_ax4 zm=%zm_ax4 + +ADD_azz_nn_s 11000001 101 ....0 0 .. 110 ....0 10 ... @azz_2x2_o3 +ADD_azz_nn_s 11000001 101 ...01 0 .. 110 ...00 10 ... @azz_4x4_o3 +ADD_azz_nn_d 11000001 111 ....0 0 .. 110 ....0 10 ... @azz_2x2_o3 +ADD_azz_nn_d 11000001 111 ...01 0 .. 110 ...00 10 ... @azz_4x4_o3 + +SUB_azz_nn_s 11000001 101 ....0 0 .. 110 ....0 11 ... @azz_2x2_o3 +SUB_azz_nn_s 11000001 101 ...01 0 .. 110 ...00 11 ... @azz_4x4_o3 +SUB_azz_nn_d 11000001 111 ....0 0 .. 110 ....0 11 ... @azz_2x2_o3 +SUB_azz_nn_d 11000001 111 ...01 0 .. 110 ...00 11 ... @azz_4x4_o3 diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(ADD_azz_n1_s, aa64_sme2, do_azz_n1, a, MO_32, tcg_gen_gvec_add_var) TRANS_FEAT(SUB_azz_n1_s, aa64_sme2, do_azz_n1, a, MO_32, tcg_gen_gvec_sub_var) TRANS_FEAT(ADD_azz_n1_d, aa64_sme2_i16i64, do_azz_n1, a, MO_64, tcg_gen_gvec_add_var) TRANS_FEAT(SUB_azz_n1_d, aa64_sme2_i16i64, do_azz_n1, a, MO_64, tcg_gen_gvec_sub_var) + +/* Add/Sub each vector Z[m*N] to each Z[n*N] with result in ZA[d*N]. */ +static bool do_azz_nn(DisasContext *s, arg_azz_n *a, int esz, + GVecGen3FnVar *fn) +{ + TCGv_ptr t_za; + int svl, n; + + if (!sme_smza_enabled_check(s)) { + return true; + } + + n = a->n; + t_za = get_zarray(s, a->rv, a->off, n, 1); + svl = streaming_vec_reg_size(s); + + for (int i = 0; i < n; ++i) { + int o_za = (svl / n * sizeof(ARMVectorReg)) * i; + int o_zn = vec_full_reg_offset(s, a->zn + i); + int o_zm = vec_full_reg_offset(s, a->zm + i); + + fn(esz, t_za, o_za, tcg_env, o_zn, tcg_env, o_zm, svl, svl); + } + return true; +} + +TRANS_FEAT(ADD_azz_nn_s, aa64_sme2, do_azz_nn, a, MO_32, tcg_gen_gvec_add_var) +TRANS_FEAT(SUB_azz_nn_s, aa64_sme2, do_azz_nn, a, MO_32, tcg_gen_gvec_sub_var) +TRANS_FEAT(ADD_azz_nn_d, aa64_sme2_i16i64, do_azz_nn, a, MO_64, tcg_gen_gvec_add_var) +TRANS_FEAT(SUB_azz_nn_d, aa64_sme2_i16i64, do_azz_nn, a, MO_64, tcg_gen_gvec_sub_var) -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Indicate whether to use FPST_FPCR or FPST_ZA via bit 2 of simd_data(desc). For SVE, this bit remains zero. For do_FMLAL_zzzw, this requires no change. For do_FMLAL_zzxw, move the index up one bit. Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-35-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/translate-sve.c | 2 +- target/arm/tcg/vec_helper.c | 8 +++++--- 2 files changed, 6 insertions(+), 4 deletions(-) diff --git a/target/arm/tcg/translate-sve.c b/target/arm/tcg/translate-sve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sve.c +++ b/target/arm/tcg/translate-sve.c @@ -XXX,XX +XXX,XX @@ static bool do_FMLAL_zzxw(DisasContext *s, arg_rrxr_esz *a, bool sub, bool sel) { return gen_gvec_ptr_zzzz(s, gen_helper_sve2_fmlal_zzxw_s, a->rd, a->rn, a->rm, a->ra, - (a->index << 2) | (sel << 1) | sub, tcg_env); + (a->index << 3) | (sel << 1) | sub, tcg_env); } TRANS_FEAT(FMLALB_zzxw, aa64_sve2, do_FMLAL_zzxw, a, false, false) diff --git a/target/arm/tcg/vec_helper.c b/target/arm/tcg/vec_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/vec_helper.c +++ b/target/arm/tcg/vec_helper.c @@ -XXX,XX +XXX,XX @@ void HELPER(sve2_fmlal_zzzw_s)(void *vd, void *vn, void *vm, void *va, intptr_t i, oprsz = simd_oprsz(desc); bool is_s = extract32(desc, SIMD_DATA_SHIFT, 1); intptr_t sel = extract32(desc, SIMD_DATA_SHIFT + 1, 1) * sizeof(float16); - float_status *status = &env->vfp.fp_status[FPST_A64]; + bool za = extract32(desc, SIMD_DATA_SHIFT + 2, 1); + float_status *status = &env->vfp.fp_status[za ? FPST_ZA : FPST_A64]; bool fz16 = env->vfp.fpcr & FPCR_FZ16; int negx = 0, negf = 0; @@ -XXX,XX +XXX,XX @@ void HELPER(sve2_fmlal_zzxw_s)(void *vd, void *vn, void *vm, void *va, intptr_t i, j, oprsz = simd_oprsz(desc); bool is_s = extract32(desc, SIMD_DATA_SHIFT, 1); intptr_t sel = extract32(desc, SIMD_DATA_SHIFT + 1, 1) * sizeof(float16); - intptr_t idx = extract32(desc, SIMD_DATA_SHIFT + 2, 3) * sizeof(float16); - float_status *status = &env->vfp.fp_status[FPST_A64]; + bool za = extract32(desc, SIMD_DATA_SHIFT + 2, 1); + intptr_t idx = extract32(desc, SIMD_DATA_SHIFT + 3, 3) * sizeof(float16); + float_status *status = &env->vfp.fp_status[za ? FPST_ZA : FPST_A64]; bool fz16 = env->vfp.fpcr & FPCR_FZ16; int negx = 0, negf = 0; -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-36-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper.h | 8 +++++ target/arm/tcg/vec_helper.c | 58 ++++++++++++++++++++++++++++--------- 2 files changed, 53 insertions(+), 13 deletions(-) diff --git a/target/arm/tcg/helper.h b/target/arm/tcg/helper.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper.h +++ b/target/arm/tcg/helper.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_6(gvec_bfmmla, TCG_CALL_NO_RWG, DEF_HELPER_FLAGS_6(gvec_bfmlal, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_6(gvec_bfmlsl, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_6(gvec_ah_bfmlsl, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_6(gvec_bfmlal_idx, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_6(gvec_bfmlsl_idx, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_6(gvec_ah_bfmlsl_idx, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_5(gvec_sclamp_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) diff --git a/target/arm/tcg/vec_helper.c b/target/arm/tcg/vec_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/vec_helper.c +++ b/target/arm/tcg/vec_helper.c @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_bfmmla)(void *vd, void *vn, void *vm, void *va, clear_tail(d, opr_sz, simd_maxsz(desc)); } -void HELPER(gvec_bfmlal)(void *vd, void *vn, void *vm, void *va, - float_status *stat, uint32_t desc) +static void do_bfmlal(float32 *d, bfloat16 *n, bfloat16 *m, float32 *a, + float_status *stat, uint32_t desc, int negx, int negf) { intptr_t i, opr_sz = simd_oprsz(desc); - intptr_t sel = simd_data(desc); - float32 *d = vd, *a = va; - bfloat16 *n = vn, *m = vm; + intptr_t sel = extract32(desc, SIMD_DATA_SHIFT, 1); for (i = 0; i < opr_sz / 4; ++i) { - float32 nn = n[H2(i * 2 + sel)] << 16; + float32 nn = (negx ^ n[H2(i * 2 + sel)]) << 16; float32 mm = m[H2(i * 2 + sel)] << 16; - d[H4(i)] = float32_muladd(nn, mm, a[H4(i)], 0, stat); + d[H4(i)] = float32_muladd(nn, mm, a[H4(i)], negf, stat); } clear_tail(d, opr_sz, simd_maxsz(desc)); } -void HELPER(gvec_bfmlal_idx)(void *vd, void *vn, void *vm, - void *va, float_status *stat, uint32_t desc) +void HELPER(gvec_bfmlal)(void *vd, void *vn, void *vm, void *va, + float_status *stat, uint32_t desc) +{ + do_bfmlal(vd, vn, vm, va, stat, desc, 0, 0); +} + +void HELPER(gvec_bfmlsl)(void *vd, void *vn, void *vm, void *va, + float_status *stat, uint32_t desc) +{ + do_bfmlal(vd, vn, vm, va, stat, desc, 0x8000, 0); +} + +void HELPER(gvec_ah_bfmlsl)(void *vd, void *vn, void *vm, void *va, + float_status *stat, uint32_t desc) +{ + do_bfmlal(vd, vn, vm, va, stat, desc, 0, float_muladd_negate_product); +} + +static void do_bfmlal_idx(float32 *d, bfloat16 *n, bfloat16 *m, float32 *a, + float_status *stat, uint32_t desc, int negx, int negf) { intptr_t i, j, opr_sz = simd_oprsz(desc); intptr_t sel = extract32(desc, SIMD_DATA_SHIFT, 1); intptr_t index = extract32(desc, SIMD_DATA_SHIFT + 1, 3); intptr_t elements = opr_sz / 4; intptr_t eltspersegment = MIN(16 / 4, elements); - float32 *d = vd, *a = va; - bfloat16 *n = vn, *m = vm; for (i = 0; i < elements; i += eltspersegment) { float32 m_idx = m[H2(2 * i + index)] << 16; for (j = i; j < i + eltspersegment; j++) { - float32 n_j = n[H2(2 * j + sel)] << 16; - d[H4(j)] = float32_muladd(n_j, m_idx, a[H4(j)], 0, stat); + float32 n_j = (negx ^ n[H2(2 * j + sel)]) << 16; + d[H4(j)] = float32_muladd(n_j, m_idx, a[H4(j)], negf, stat); } } clear_tail(d, opr_sz, simd_maxsz(desc)); } +void HELPER(gvec_bfmlal_idx)(void *vd, void *vn, void *vm, void *va, + float_status *stat, uint32_t desc) +{ + do_bfmlal_idx(vd, vn, vm, va, stat, desc, 0, 0); +} + +void HELPER(gvec_bfmlsl_idx)(void *vd, void *vn, void *vm, void *va, + float_status *stat, uint32_t desc) +{ + do_bfmlal_idx(vd, vn, vm, va, stat, desc, 0x8000, 0); +} + +void HELPER(gvec_ah_bfmlsl_idx)(void *vd, void *vn, void *vm, void *va, + float_status *stat, uint32_t desc) +{ + do_bfmlal_idx(vd, vn, vm, va, stat, desc, 0, float_muladd_negate_product); +} + #define DO_CLAMP(NAME, TYPE) \ void HELPER(NAME)(void *d, void *n, void *m, void *a, uint32_t desc) \ { \ -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-37-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/sme.decode | 71 ++++++++++++++++++++++++ target/arm/tcg/translate-sme.c | 98 ++++++++++++++++++++++++++++++++++ 2 files changed, 169 insertions(+) diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ SUB_azz_n1_s 11000001 0011 .... 0 .. 110 ..... 11 ... @azz_nx1_o3 n=4 SUB_azz_n1_d 11000001 0110 .... 0 .. 110 ..... 11 ... @azz_nx1_o3 n=2 SUB_azz_n1_d 11000001 0111 .... 0 .. 110 ..... 11 ... @azz_nx1_o3 n=4 +%off3_x2 0:3 !function=times_2 +%off2_x2 0:2 !function=times_2 + +@azz_nx1_o3x2 ........ ... . zm:4 . .. ... zn:5 .. ... \ + &azz_n off=%off3_x2 rv=%mova_rv +@azz_nx1_o2x2 ........ ... . zm:4 . .. ... zn:5 ... .. \ + &azz_n off=%off2_x2 rv=%mova_rv + +FMLAL_n1 11000001 001 0 .... 0 .. 011 ..... 00 ... @azz_nx1_o3x2 n=1 +FMLAL_n1 11000001 001 0 .... 0 .. 010 ..... 000 .. @azz_nx1_o2x2 n=2 +FMLAL_n1 11000001 001 1 .... 0 .. 010 ..... 000 .. @azz_nx1_o2x2 n=4 + +FMLSL_n1 11000001 001 0 .... 0 .. 011 ..... 01 ... @azz_nx1_o3x2 n=1 +FMLSL_n1 11000001 001 0 .... 0 .. 010 ..... 010 .. @azz_nx1_o2x2 n=2 +FMLSL_n1 11000001 001 1 .... 0 .. 010 ..... 010 .. @azz_nx1_o2x2 n=4 + +BFMLAL_n1 11000001 001 0 .... 0 .. 011 ..... 10 ... @azz_nx1_o3x2 n=1 +BFMLAL_n1 11000001 001 0 .... 0 .. 010 ..... 100 .. @azz_nx1_o2x2 n=2 +BFMLAL_n1 11000001 001 1 .... 0 .. 010 ..... 100 .. @azz_nx1_o2x2 n=4 + +BFMLSL_n1 11000001 001 0 .... 0 .. 011 ..... 11 ... @azz_nx1_o3x2 n=1 +BFMLSL_n1 11000001 001 0 .... 0 .. 010 ..... 110 .. @azz_nx1_o2x2 n=2 +BFMLSL_n1 11000001 001 1 .... 0 .. 010 ..... 110 .. @azz_nx1_o2x2 n=4 + ### SME2 Multi-vector Multiple Array Vectors %zn_ax2 6:4 !function=times_2 @@ -XXX,XX +XXX,XX @@ SUB_azz_nn_s 11000001 101 ....0 0 .. 110 ....0 11 ... @azz_2x2_o3 SUB_azz_nn_s 11000001 101 ...01 0 .. 110 ...00 11 ... @azz_4x4_o3 SUB_azz_nn_d 11000001 111 ....0 0 .. 110 ....0 11 ... @azz_2x2_o3 SUB_azz_nn_d 11000001 111 ...01 0 .. 110 ...00 11 ... @azz_4x4_o3 + +@azz_2x2_o2x2 ........ ... ..... . .. ... ..... ... .. \ + &azz_n n=2 rv=%mova_rv zn=%zn_ax2 zm=%zm_ax2 off=%off2_x2 +@azz_4x4_o2x2 ........ ... ..... . .. ... ..... ... .. \ + &azz_n n=4 rv=%mova_rv zn=%zn_ax4 zm=%zm_ax4 off=%off2_x2 + +FMLAL_nn 11000001 101 ....0 0 .. 010 ....0 000 .. @azz_2x2_o2x2 +FMLAL_nn 11000001 101 ...01 0 .. 010 ...00 000 .. @azz_4x4_o2x2 + +FMLSL_nn 11000001 101 ....0 0 .. 010 ....0 010 .. @azz_2x2_o2x2 +FMLSL_nn 11000001 101 ...01 0 .. 010 ...00 010 .. @azz_4x4_o2x2 + +BFMLAL_nn 11000001 101 ....0 0 .. 010 ....0 100 .. @azz_2x2_o2x2 +BFMLAL_nn 11000001 101 ...01 0 .. 010 ...00 100 .. @azz_4x4_o2x2 + +BFMLSL_nn 11000001 101 ....0 0 .. 010 ....0 110 .. @azz_2x2_o2x2 +BFMLSL_nn 11000001 101 ...01 0 .. 010 ...00 110 .. @azz_4x4_o2x2 + +### SME2 Multi-vector Indexed + +&azx_n n off rv zn zm idx + +%idx3_15_10 15:1 10:2 +%idx2_10_2 10:2 2:1 + +@azx_1x1_o3x2 ........ .... zm:4 . .. . .. zn:5 .. ... \ + &azx_n n=1 rv=%mova_rv off=%off3_x2 idx=%idx3_15_10 +@azx_2x1_o2x2 ........ .... zm:4 . .. . .. ..... .. ... \ + &azx_n n=2 rv=%mova_rv off=%off2_x2 zn=%zn_ax2 idx=%idx2_10_2 +@azx_4x1_o2x2 ........ .... zm:4 . .. . .. ..... .. ... \ + &azx_n n=4 rv=%mova_rv off=%off2_x2 zn=%zn_ax4 idx=%idx2_10_2 + +FMLAL_nx 11000001 1000 .... . .. 1 .. ..... 00 ... @azx_1x1_o3x2 +FMLAL_nx 11000001 1001 .... 0 .. 1 .. ....0 00 ... @azx_2x1_o2x2 +FMLAL_nx 11000001 1001 .... 1 .. 1 .. ...00 00 ... @azx_4x1_o2x2 + +FMLSL_nx 11000001 1000 .... . .. 1 .. ..... 01 ... @azx_1x1_o3x2 +FMLSL_nx 11000001 1001 .... 0 .. 1 .. ....0 01 ... @azx_2x1_o2x2 +FMLSL_nx 11000001 1001 .... 1 .. 1 .. ...00 01 ... @azx_4x1_o2x2 + +BFMLAL_nx 11000001 1000 .... . .. 1 .. ..... 10 ... @azx_1x1_o3x2 +BFMLAL_nx 11000001 1001 .... 0 .. 1 .. ....0 10 ... @azx_2x1_o2x2 +BFMLAL_nx 11000001 1001 .... 1 .. 1 .. ...00 10 ... @azx_4x1_o2x2 + +BFMLSL_nx 11000001 1000 .... . .. 1 .. ..... 11 ... @azx_1x1_o3x2 +BFMLSL_nx 11000001 1001 .... 0 .. 1 .. ....0 11 ... @azx_2x1_o2x2 +BFMLSL_nx 11000001 1001 .... 1 .. 1 .. ...00 11 ... @azx_4x1_o2x2 diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(ADD_azz_nn_s, aa64_sme2, do_azz_nn, a, MO_32, tcg_gen_gvec_add_var) TRANS_FEAT(SUB_azz_nn_s, aa64_sme2, do_azz_nn, a, MO_32, tcg_gen_gvec_sub_var) TRANS_FEAT(ADD_azz_nn_d, aa64_sme2_i16i64, do_azz_nn, a, MO_64, tcg_gen_gvec_add_var) TRANS_FEAT(SUB_azz_nn_d, aa64_sme2_i16i64, do_azz_nn, a, MO_64, tcg_gen_gvec_sub_var) + +/* + * Expand array multi-vector single (n1), array multi-vector (nn), + * and array multi-vector indexed (nx), for floating-point accumulate. + * multi: true for nn, false for n1. + * fpst: >= 0 to set ptr argument for FPST_*, < 0 for ENV. + * data: stuff for simd_data, including any index. + */ +#define FPST_ENV -1 + +static bool do_azz_acc_fp(DisasContext *s, int nreg, int nsel, + int rv, int off, int zn, int zm, + int data, int shsel, bool multi, int fpst, + gen_helper_gvec_4_ptr *fn) +{ + if (sme_smza_enabled_check(s)) { + int svl = streaming_vec_reg_size(s); + int vstride = svl / nreg; + TCGv_ptr t_za = get_zarray(s, rv, off, nreg, nsel); + TCGv_ptr t, ptr; + + if (fpst >= 0) { + ptr = fpstatus_ptr(fpst); + } else { + ptr = tcg_env; + } + t = tcg_temp_new_ptr(); + + for (int r = 0; r < nreg; ++r) { + TCGv_ptr t_zn = vec_full_reg_ptr(s, zn); + TCGv_ptr t_zm = vec_full_reg_ptr(s, zm); + + for (int i = 0; i < nsel; ++i) { + int o_za = (r * vstride + i) * sizeof(ARMVectorReg); + int desc = simd_desc(svl, svl, data | (i << shsel)); + + tcg_gen_addi_ptr(t, t_za, o_za); + fn(t, t_zn, t_zm, t, ptr, tcg_constant_i32(desc)); + } + + /* + * For multiple-and-single vectors, Zn may wrap. + * For multiple vectors, both Zn and Zm are aligned. + */ + zn = (zn + 1) % 32; + zm += multi; + } + } + return true; +} + +static bool do_fmlal(DisasContext *s, arg_azz_n *a, bool sub, bool multi) +{ + return do_azz_acc_fp(s, a->n, 2, a->rv, a->off, a->zn, a->zm, + (1 << 2) | sub, 1, + multi, FPST_ENV, gen_helper_sve2_fmlal_zzzw_s); +} + +TRANS_FEAT(FMLAL_n1, aa64_sme2, do_fmlal, a, false, false) +TRANS_FEAT(FMLSL_n1, aa64_sme2, do_fmlal, a, true, false) +TRANS_FEAT(FMLAL_nn, aa64_sme2, do_fmlal, a, false, true) +TRANS_FEAT(FMLSL_nn, aa64_sme2, do_fmlal, a, true, true) + +static bool do_fmlal_nx(DisasContext *s, arg_azx_n *a, bool sub) +{ + return do_azz_acc_fp(s, a->n, 2, a->rv, a->off, a->zn, a->zm, + (a->idx << 3) | (1 << 2) | sub, 1, + false, FPST_ENV, gen_helper_sve2_fmlal_zzxw_s); +} + +TRANS_FEAT(FMLAL_nx, aa64_sme2, do_fmlal_nx, a, false) +TRANS_FEAT(FMLSL_nx, aa64_sme2, do_fmlal_nx, a, true) + +static bool do_bfmlal(DisasContext *s, arg_azz_n *a, bool sub, bool multi) +{ + return do_azz_acc_fp(s, a->n, 2, a->rv, a->off, a->zn, a->zm, + 0, 0, multi, FPST_ZA, + (!sub ? gen_helper_gvec_bfmlal + : s->fpcr_ah ? gen_helper_gvec_ah_bfmlsl + : gen_helper_gvec_bfmlsl)); +} + +TRANS_FEAT(BFMLAL_n1, aa64_sme2, do_bfmlal, a, false, false) +TRANS_FEAT(BFMLSL_n1, aa64_sme2, do_bfmlal, a, true, false) +TRANS_FEAT(BFMLAL_nn, aa64_sme2, do_bfmlal, a, false, true) +TRANS_FEAT(BFMLSL_nn, aa64_sme2, do_bfmlal, a, true, true) + +static bool do_bfmlal_nx(DisasContext *s, arg_azx_n *a, bool sub) +{ + return do_azz_acc_fp(s, a->n, 2, a->rv, a->off, a->zn, a->zm, + a->idx << 1, 0, false, FPST_ZA, + !sub ? gen_helper_gvec_bfmlal_idx + : s->fpcr_ah ? gen_helper_gvec_ah_bfmlsl_idx + : gen_helper_gvec_bfmlsl_idx); +} + +TRANS_FEAT(BFMLAL_nx, aa64_sme2, do_bfmlal_nx, a, false) +TRANS_FEAT(BFMLSL_nx, aa64_sme2, do_bfmlal_nx, a, true) -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-38-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper-sme.h | 5 ++++ target/arm/tcg/sme.decode | 14 +++++++++++ target/arm/tcg/sve.decode | 7 ++++-- target/arm/tcg/sme_helper.c | 44 ++++++++++++++++++++++++++++++++++ target/arm/tcg/translate-sme.c | 18 ++++++++++++++ target/arm/tcg/translate-sve.c | 5 ++++ 6 files changed, 91 insertions(+), 2 deletions(-) diff --git a/target/arm/tcg/helper-sme.h b/target/arm/tcg/helper-sme.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-sme.h +++ b/target/arm/tcg/helper-sme.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_5(gvec_fmaxnum_b16, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_5(gvec_fminnum_b16, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) + +DEF_HELPER_FLAGS_6(sme2_fdot_h, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, env, i32) +DEF_HELPER_FLAGS_6(sme2_fdot_idx_h, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, env, i32) diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ BFMLSL_n1 11000001 001 0 .... 0 .. 011 ..... 11 ... @azz_nx1_o3x2 n=1 BFMLSL_n1 11000001 001 0 .... 0 .. 010 ..... 110 .. @azz_nx1_o2x2 n=2 BFMLSL_n1 11000001 001 1 .... 0 .. 010 ..... 110 .. @azz_nx1_o2x2 n=4 +FDOT_n1 11000001 001 0 .... 0 .. 100 ..... 00 ... @azz_nx1_o3 n=2 +FDOT_n1 11000001 001 1 .... 0 .. 100 ..... 00 ... @azz_nx1_o3 n=4 + ### SME2 Multi-vector Multiple Array Vectors %zn_ax2 6:4 !function=times_2 @@ -XXX,XX +XXX,XX @@ BFMLAL_nn 11000001 101 ...01 0 .. 010 ...00 100 .. @azz_4x4_o2x2 BFMLSL_nn 11000001 101 ....0 0 .. 010 ....0 110 .. @azz_2x2_o2x2 BFMLSL_nn 11000001 101 ...01 0 .. 010 ...00 110 .. @azz_4x4_o2x2 +FDOT_nn 11000001 101 ....0 0 .. 100 ....0 00 ... @azz_2x2_o3 +FDOT_nn 11000001 101 ...01 0 .. 100 ...00 00 ... @azz_4x4_o3 + ### SME2 Multi-vector Indexed &azx_n n off rv zn zm idx @@ -XXX,XX +XXX,XX @@ BFMLAL_nx 11000001 1001 .... 1 .. 1 .. ...00 10 ... @azx_4x1_o2x2 BFMLSL_nx 11000001 1000 .... . .. 1 .. ..... 11 ... @azx_1x1_o3x2 BFMLSL_nx 11000001 1001 .... 0 .. 1 .. ....0 11 ... @azx_2x1_o2x2 BFMLSL_nx 11000001 1001 .... 1 .. 1 .. ...00 11 ... @azx_4x1_o2x2 + +@azx_2x1_i2_o3 ........ .... zm:4 . .. . idx:2 .... ... off:3 \ + &azx_n n=2 rv=%mova_rv zn=%zn_ax2 +@azx_4x1_i2_o3 ........ .... zm:4 . .. . idx:2 .... ... off:3 \ + &azx_n n=2 rv=%mova_rv zn=%zn_ax4 + +FDOT_nx 11000001 0101 .... 0 .. 1 .. ....0 01 ... @azx_2x1_i2_o3 +FDOT_nx 11000001 0101 .... 1 .. 1 .. ...00 01 ... @azx_4x1_i2_o3 diff --git a/target/arm/tcg/sve.decode b/target/arm/tcg/sve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve.decode +++ b/target/arm/tcg/sve.decode @@ -XXX,XX +XXX,XX @@ FMLSLT_zzzw 01100100 10 1 ..... 10 1 00 1 ..... ..... @rda_rn_rm_ex esz=2 BFMLALB_zzzw 01100100 11 1 ..... 10 0 00 0 ..... ..... @rda_rn_rm_ex esz=2 BFMLALT_zzzw 01100100 11 1 ..... 10 0 00 1 ..... ..... @rda_rn_rm_ex esz=2 -### SVE2 floating-point bfloat16 dot-product +### SVE2 floating-point dot-product +FDOT_zzzz 01100100 00 1 ..... 10 0 00 0 ..... ..... @rda_rn_rm_ex esz=2 BFDOT_zzzz 01100100 01 1 ..... 10 0 00 0 ..... ..... @rda_rn_rm_ex esz=2 ### SVE2 floating-point multiply-add long (indexed) @@ -XXX,XX +XXX,XX @@ FMLSLT_zzxw 01100100 10 1 ..... 0110.1 ..... ..... @rrxr_3a esz=2 BFMLALB_zzxw 01100100 11 1 ..... 0100.0 ..... ..... @rrxr_3a esz=2 BFMLALT_zzxw 01100100 11 1 ..... 0100.1 ..... ..... @rrxr_3a esz=2 -### SVE2 floating-point bfloat16 dot-product (indexed) +### SVE2 floating-point dot-product (indexed) + +FDOT_zzxz 01100100 00 1 ..... 010000 ..... ..... @rrxr_2 esz=2 BFDOT_zzxz 01100100 01 1 ..... 010000 ..... ..... @rrxr_2 esz=2 ### SVE broadcast predicate element diff --git a/target/arm/tcg/sme_helper.c b/target/arm/tcg/sme_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme_helper.c +++ b/target/arm/tcg/sme_helper.c @@ -XXX,XX +XXX,XX @@ void HELPER(sme_fmopa_h)(void *vza, void *vzn, void *vzm, void *vpn, } } +void HELPER(sme2_fdot_h)(void *vd, void *vn, void *vm, void *va, + CPUARMState *env, uint32_t desc) +{ + intptr_t i, oprsz = simd_maxsz(desc); + bool za = extract32(desc, SIMD_DATA_SHIFT, 1); + float_status *fpst_std = &env->vfp.fp_status[za ? FPST_ZA : FPST_A64]; + float_status *fpst_f16 = &env->vfp.fp_status[za ? FPST_ZA_F16 : FPST_A64_F16]; + float_status fpst_odd = *fpst_std; + float32 *d = vd, *a = va; + uint32_t *n = vn, *m = vm; + + set_float_rounding_mode(float_round_to_odd, &fpst_odd); + + for (i = 0; i < oprsz / sizeof(float32); ++i) { + d[H4(i)] = f16_dotadd(a[H4(i)], n[H4(i)], m[H4(i)], + fpst_f16, fpst_std, &fpst_odd); + } +} + +void HELPER(sme2_fdot_idx_h)(void *vd, void *vn, void *vm, void *va, + CPUARMState *env, uint32_t desc) +{ + intptr_t i, j, oprsz = simd_maxsz(desc); + intptr_t elements = oprsz / sizeof(float32); + intptr_t eltspersegment = MIN(4, elements); + int idx = extract32(desc, SIMD_DATA_SHIFT, 2); + bool za = extract32(desc, SIMD_DATA_SHIFT + 2, 1); + float_status *fpst_std = &env->vfp.fp_status[za ? FPST_ZA : FPST_A64]; + float_status *fpst_f16 = &env->vfp.fp_status[za ? FPST_ZA_F16 : FPST_A64_F16]; + float_status fpst_odd = *fpst_std; + float32 *d = vd, *a = va; + uint32_t *n = vn, *m = (uint32_t *)vm + H4(idx); + + set_float_rounding_mode(float_round_to_odd, &fpst_odd); + + for (i = 0; i < elements; i += eltspersegment) { + uint32_t mm = m[i]; + for (j = 0; j < eltspersegment; ++j) { + d[H4(i + j)] = f16_dotadd(a[H4(i + j)], n[H4(i + j)], mm, + fpst_f16, fpst_std, &fpst_odd); + } + } +} + void HELPER(sme_bfmopa)(void *vza, void *vzn, void *vzm, void *vpn, void *vpm, CPUARMState *env, uint32_t desc) { diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ static bool do_bfmlal_nx(DisasContext *s, arg_azx_n *a, bool sub) TRANS_FEAT(BFMLAL_nx, aa64_sme2, do_bfmlal_nx, a, false) TRANS_FEAT(BFMLSL_nx, aa64_sme2, do_bfmlal_nx, a, true) + +static bool do_fdot(DisasContext *s, arg_azz_n *a, bool multi) +{ + return do_azz_acc_fp(s, a->n, 1, a->rv, a->off, a->zn, a->zm, 1, 0, + multi, FPST_ENV, gen_helper_sme2_fdot_h); +} + +TRANS_FEAT(FDOT_n1, aa64_sme2, do_fdot, a, false) +TRANS_FEAT(FDOT_nn, aa64_sme2, do_fdot, a, true) + +static bool do_fdot_nx(DisasContext *s, arg_azx_n *a) +{ + return do_azz_acc_fp(s, a->n, 1, a->rv, a->off, a->zn, a->zm, + a->idx | (1 << 2), 0, false, FPST_ENV, + gen_helper_sme2_fdot_idx_h); +} + +TRANS_FEAT(FDOT_nx, aa64_sme2, do_fdot_nx, a) diff --git a/target/arm/tcg/translate-sve.c b/target/arm/tcg/translate-sve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sve.c +++ b/target/arm/tcg/translate-sve.c @@ -XXX,XX +XXX,XX @@ TRANS_FEAT_NONSTREAMING(USMMLA, aa64_sve_i8mm, gen_gvec_ool_arg_zzzz, TRANS_FEAT_NONSTREAMING(UMMLA, aa64_sve_i8mm, gen_gvec_ool_arg_zzzz, gen_helper_gvec_ummla_b, a, 0) +TRANS_FEAT(FDOT_zzzz, aa64_sme2_or_sve2p1, gen_gvec_env_arg_zzzz, + gen_helper_sme2_fdot_h, a, 0) +TRANS_FEAT(FDOT_zzxz, aa64_sme2_or_sve2p1, gen_gvec_env_arg_zzxz, + gen_helper_sme2_fdot_idx_h, a) + TRANS_FEAT(BFDOT_zzzz, aa64_sve_bf16, gen_gvec_env_arg_zzzz, gen_helper_gvec_bfdot, a, 0) TRANS_FEAT(BFDOT_zzxz, aa64_sve_bf16, gen_gvec_env_arg_zzxz, -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-39-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/sme.decode | 9 +++++++++ target/arm/tcg/translate-sme.c | 17 +++++++++++++++++ 2 files changed, 26 insertions(+) diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ BFMLSL_n1 11000001 001 1 .... 0 .. 010 ..... 110 .. @azz_nx1_o2x2 n=4 FDOT_n1 11000001 001 0 .... 0 .. 100 ..... 00 ... @azz_nx1_o3 n=2 FDOT_n1 11000001 001 1 .... 0 .. 100 ..... 00 ... @azz_nx1_o3 n=4 +BFDOT_n1 11000001 001 0 .... 0 .. 100 ..... 10 ... @azz_nx1_o3 n=2 +BFDOT_n1 11000001 001 1 .... 0 .. 100 ..... 10 ... @azz_nx1_o3 n=4 + ### SME2 Multi-vector Multiple Array Vectors %zn_ax2 6:4 !function=times_2 @@ -XXX,XX +XXX,XX @@ BFMLSL_nn 11000001 101 ...01 0 .. 010 ...00 110 .. @azz_4x4_o2x2 FDOT_nn 11000001 101 ....0 0 .. 100 ....0 00 ... @azz_2x2_o3 FDOT_nn 11000001 101 ...01 0 .. 100 ...00 00 ... @azz_4x4_o3 +BFDOT_nn 11000001 101 ....0 0 .. 100 ....0 10 ... @azz_2x2_o3 +BFDOT_nn 11000001 101 ...01 0 .. 100 ...00 10 ... @azz_4x4_o3 + ### SME2 Multi-vector Indexed &azx_n n off rv zn zm idx @@ -XXX,XX +XXX,XX @@ BFMLSL_nx 11000001 1001 .... 1 .. 1 .. ...00 11 ... @azx_4x1_o2x2 FDOT_nx 11000001 0101 .... 0 .. 1 .. ....0 01 ... @azx_2x1_i2_o3 FDOT_nx 11000001 0101 .... 1 .. 1 .. ...00 01 ... @azx_4x1_i2_o3 + +BFDOT_nx 11000001 0101 .... 0 .. 1 .. ....0 11 ... @azx_2x1_i2_o3 +BFDOT_nx 11000001 0101 .... 1 .. 1 .. ...00 11 ... @azx_4x1_i2_o3 diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ static bool do_fdot_nx(DisasContext *s, arg_azx_n *a) } TRANS_FEAT(FDOT_nx, aa64_sme2, do_fdot_nx, a) + +static bool do_bfdot(DisasContext *s, arg_azz_n *a, bool multi) +{ + return do_azz_acc_fp(s, a->n, 1, a->rv, a->off, a->zn, a->zm, 0, 0, + multi, FPST_ENV, gen_helper_gvec_bfdot); +} + +TRANS_FEAT(BFDOT_n1, aa64_sme2, do_bfdot, a, false) +TRANS_FEAT(BFDOT_nn, aa64_sme2, do_bfdot, a, true) + +static bool do_bfdot_nx(DisasContext *s, arg_azx_n *a) +{ + return do_azz_acc_fp(s, a->n, 1, a->rv, a->off, a->zn, a->zm, a->idx, 0, + false, FPST_ENV, gen_helper_gvec_bfdot_idx); +} + +TRANS_FEAT(BFDOT_nx, aa64_sme2, do_bfdot_nx, a) -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-40-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper-sme.h | 2 ++ target/arm/tcg/helper.h | 2 ++ target/arm/tcg/sme.decode | 3 +++ target/arm/tcg/sme_helper.c | 30 ++++++++++++++++++++++++++ target/arm/tcg/translate-sme.c | 24 +++++++++++++++++++++ target/arm/tcg/vec_helper.c | 39 ++++++++++++++++++++++++++++++++++ 6 files changed, 100 insertions(+) diff --git a/target/arm/tcg/helper-sme.h b/target/arm/tcg/helper-sme.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-sme.h +++ b/target/arm/tcg/helper-sme.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_6(sme2_fdot_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, env, i32) DEF_HELPER_FLAGS_6(sme2_fdot_idx_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, env, i32) +DEF_HELPER_FLAGS_6(sme2_fvdot_idx_h, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, env, i32) diff --git a/target/arm/tcg/helper.h b/target/arm/tcg/helper.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper.h +++ b/target/arm/tcg/helper.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_6(gvec_bfdot, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, env, i32) DEF_HELPER_FLAGS_6(gvec_bfdot_idx, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, env, i32) +DEF_HELPER_FLAGS_6(sme2_bfvdot_idx, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, env, i32) DEF_HELPER_FLAGS_6(gvec_bfmmla, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, env, i32) diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ FDOT_nx 11000001 0101 .... 1 .. 1 .. ...00 01 ... @azx_4x1_i2_o3 BFDOT_nx 11000001 0101 .... 0 .. 1 .. ....0 11 ... @azx_2x1_i2_o3 BFDOT_nx 11000001 0101 .... 1 .. 1 .. ...00 11 ... @azx_4x1_i2_o3 + +FVDOT 11000001 0101 .... 0 .. 0 .. ....0 01 ... @azx_2x1_i2_o3 +BFVDOT 11000001 0101 .... 0 .. 0 .. ....0 11 ... @azx_2x1_i2_o3 diff --git a/target/arm/tcg/sme_helper.c b/target/arm/tcg/sme_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme_helper.c +++ b/target/arm/tcg/sme_helper.c @@ -XXX,XX +XXX,XX @@ void HELPER(sme2_fdot_idx_h)(void *vd, void *vn, void *vm, void *va, } } +void HELPER(sme2_fvdot_idx_h)(void *vd, void *vn, void *vm, void *va, + CPUARMState *env, uint32_t desc) +{ + intptr_t i, j, oprsz = simd_maxsz(desc); + intptr_t elements = oprsz / sizeof(float32); + intptr_t eltspersegment = MIN(4, elements); + int idx = extract32(desc, SIMD_DATA_SHIFT, 2); + int sel = extract32(desc, SIMD_DATA_SHIFT + 2, 1); + float_status fpst_odd, *fpst_std, *fpst_f16; + float32 *d = vd, *a = va; + uint16_t *n0 = vn; + uint16_t *n1 = vn + sizeof(ARMVectorReg); + uint32_t *m = (uint32_t *)vm + H4(idx); + + fpst_std = &env->vfp.fp_status[FPST_ZA]; + fpst_f16 = &env->vfp.fp_status[FPST_ZA_F16]; + fpst_odd = *fpst_std; + set_float_rounding_mode(float_round_to_odd, &fpst_odd); + + for (i = 0; i < elements; i += eltspersegment) { + uint32_t mm = m[i]; + for (j = 0; j < eltspersegment; ++j) { + uint32_t nn = (n0[H2(2 * (i + j) + sel)]) + | (n1[H2(2 * (i + j) + sel)] << 16); + d[i + H4(j)] = f16_dotadd(a[i + H4(j)], nn, mm, + fpst_f16, fpst_std, &fpst_odd); + } + } +} + void HELPER(sme_bfmopa)(void *vza, void *vzn, void *vzm, void *vpn, void *vpm, CPUARMState *env, uint32_t desc) { diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ static bool do_bfdot_nx(DisasContext *s, arg_azx_n *a) } TRANS_FEAT(BFDOT_nx, aa64_sme2, do_bfdot_nx, a) + +static bool do_vdot(DisasContext *s, arg_azx_n *a, gen_helper_gvec_4_ptr *fn) +{ + if (sme_smza_enabled_check(s)) { + int svl = streaming_vec_reg_size(s); + int vstride = svl / 2; + TCGv_ptr t_za = get_zarray(s, a->rv, a->off, 2, 1); + TCGv_ptr t_zn = vec_full_reg_ptr(s, a->zn); + TCGv_ptr t_zm = vec_full_reg_ptr(s, a->zm); + TCGv_ptr t = tcg_temp_new_ptr(); + + for (int i = 0; i < 2; ++i) { + int o_za = i * vstride * sizeof(ARMVectorReg); + int desc = simd_desc(svl, svl, a->idx | (i << 2)); + + tcg_gen_addi_ptr(t, t_za, o_za); + fn(t, t_zn, t_zm, t, tcg_env, tcg_constant_i32(desc)); + } + } + return true; +} + +TRANS_FEAT(FVDOT, aa64_sme, do_vdot, a, gen_helper_sme2_fvdot_idx_h) +TRANS_FEAT(BFVDOT, aa64_sme, do_vdot, a, gen_helper_sme2_bfvdot_idx) diff --git a/target/arm/tcg/vec_helper.c b/target/arm/tcg/vec_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/vec_helper.c +++ b/target/arm/tcg/vec_helper.c @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_bfdot_idx)(void *vd, void *vn, void *vm, clear_tail(d, opr_sz, simd_maxsz(desc)); } +void HELPER(sme2_bfvdot_idx)(void *vd, void *vn, void *vm, + void *va, CPUARMState *env, uint32_t desc) +{ + intptr_t i, j, opr_sz = simd_oprsz(desc); + intptr_t idx = extract32(desc, SIMD_DATA_SHIFT, 2); + intptr_t sel = extract32(desc, SIMD_DATA_SHIFT + 2, 1); + intptr_t elements = opr_sz / 4; + intptr_t eltspersegment = MIN(16 / 4, elements); + float32 *d = vd, *a = va; + uint16_t *n0 = vn; + uint16_t *n1 = vn + sizeof(ARMVectorReg); + uint32_t *m = vm; + float_status fpst, fpst_odd; + + if (is_ebf(env, &fpst, &fpst_odd)) { + for (i = 0; i < elements; i += eltspersegment) { + uint32_t m_idx = m[i + H4(idx)]; + + for (j = 0; j < eltspersegment; j++) { + uint32_t nn = (n0[H2(2 * (i + j) + sel)]) + | (n1[H2(2 * (i + j) + sel)] << 16); + d[i + H4(j)] = bfdotadd_ebf(a[i + H4(j)], nn, m_idx, + &fpst, &fpst_odd); + } + } + } else { + for (i = 0; i < elements; i += eltspersegment) { + uint32_t m_idx = m[i + H4(idx)]; + + for (j = 0; j < eltspersegment; j++) { + uint32_t nn = (n0[H2(2 * (i + j) + sel)]) + | (n1[H2(2 * (i + j) + sel)] << 16); + d[i + H4(j)] = bfdotadd(a[i + H4(j)], nn, m_idx, &fpst); + } + } + } + clear_tail(d, opr_sz, simd_maxsz(desc)); +} + void HELPER(gvec_bfmmla)(void *vd, void *vn, void *vm, void *va, CPUARMState *env, uint32_t desc) { -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Emphasize that these are 4-way dot products. Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-41-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper.h | 22 +++++++++++----------- target/arm/tcg/translate-a64.c | 14 +++++++------- target/arm/tcg/translate-neon.c | 14 +++++++------- target/arm/tcg/translate-sve.c | 18 +++++++++--------- target/arm/tcg/vec_helper.c | 22 +++++++++++----------- 5 files changed, 45 insertions(+), 45 deletions(-) diff --git a/target/arm/tcg/helper.h b/target/arm/tcg/helper.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper.h +++ b/target/arm/tcg/helper.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_5(sve2_sqrdmlah_d, TCG_CALL_NO_RWG, DEF_HELPER_FLAGS_5(sve2_sqrdmlsh_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) -DEF_HELPER_FLAGS_5(gvec_sdot_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) -DEF_HELPER_FLAGS_5(gvec_udot_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) -DEF_HELPER_FLAGS_5(gvec_sdot_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) -DEF_HELPER_FLAGS_5(gvec_udot_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) -DEF_HELPER_FLAGS_5(gvec_usdot_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(gvec_sdot_4b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(gvec_udot_4b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(gvec_sdot_4h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(gvec_udot_4h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(gvec_usdot_4b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) -DEF_HELPER_FLAGS_5(gvec_sdot_idx_b, TCG_CALL_NO_RWG, +DEF_HELPER_FLAGS_5(gvec_sdot_idx_4b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) -DEF_HELPER_FLAGS_5(gvec_udot_idx_b, TCG_CALL_NO_RWG, +DEF_HELPER_FLAGS_5(gvec_udot_idx_4b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) -DEF_HELPER_FLAGS_5(gvec_sdot_idx_h, TCG_CALL_NO_RWG, +DEF_HELPER_FLAGS_5(gvec_sdot_idx_4h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) -DEF_HELPER_FLAGS_5(gvec_udot_idx_h, TCG_CALL_NO_RWG, +DEF_HELPER_FLAGS_5(gvec_udot_idx_4h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) -DEF_HELPER_FLAGS_5(gvec_sudot_idx_b, TCG_CALL_NO_RWG, +DEF_HELPER_FLAGS_5(gvec_sudot_idx_4b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) -DEF_HELPER_FLAGS_5(gvec_usdot_idx_b, TCG_CALL_NO_RWG, +DEF_HELPER_FLAGS_5(gvec_usdot_idx_4b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_5(gvec_fcaddh, TCG_CALL_NO_RWG, diff --git a/target/arm/tcg/translate-a64.c b/target/arm/tcg/translate-a64.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-a64.c +++ b/target/arm/tcg/translate-a64.c @@ -XXX,XX +XXX,XX @@ static bool do_dot_vector_env(DisasContext *s, arg_qrrr_e *a, return true; } -TRANS_FEAT(SDOT_v, aa64_dp, do_dot_vector, a, gen_helper_gvec_sdot_b) -TRANS_FEAT(UDOT_v, aa64_dp, do_dot_vector, a, gen_helper_gvec_udot_b) -TRANS_FEAT(USDOT_v, aa64_i8mm, do_dot_vector, a, gen_helper_gvec_usdot_b) +TRANS_FEAT(SDOT_v, aa64_dp, do_dot_vector, a, gen_helper_gvec_sdot_4b) +TRANS_FEAT(UDOT_v, aa64_dp, do_dot_vector, a, gen_helper_gvec_udot_4b) +TRANS_FEAT(USDOT_v, aa64_i8mm, do_dot_vector, a, gen_helper_gvec_usdot_4b) TRANS_FEAT(BFDOT_v, aa64_bf16, do_dot_vector_env, a, gen_helper_gvec_bfdot) TRANS_FEAT(BFMMLA, aa64_bf16, do_dot_vector_env, a, gen_helper_gvec_bfmmla) TRANS_FEAT(SMMLA, aa64_i8mm, do_dot_vector, a, gen_helper_gvec_smmla_b) @@ -XXX,XX +XXX,XX @@ static bool do_dot_vector_idx_env(DisasContext *s, arg_qrrx_e *a, return true; } -TRANS_FEAT(SDOT_vi, aa64_dp, do_dot_vector_idx, a, gen_helper_gvec_sdot_idx_b) -TRANS_FEAT(UDOT_vi, aa64_dp, do_dot_vector_idx, a, gen_helper_gvec_udot_idx_b) +TRANS_FEAT(SDOT_vi, aa64_dp, do_dot_vector_idx, a, gen_helper_gvec_sdot_idx_4b) +TRANS_FEAT(UDOT_vi, aa64_dp, do_dot_vector_idx, a, gen_helper_gvec_udot_idx_4b) TRANS_FEAT(SUDOT_vi, aa64_i8mm, do_dot_vector_idx, a, - gen_helper_gvec_sudot_idx_b) + gen_helper_gvec_sudot_idx_4b) TRANS_FEAT(USDOT_vi, aa64_i8mm, do_dot_vector_idx, a, - gen_helper_gvec_usdot_idx_b) + gen_helper_gvec_usdot_idx_4b) TRANS_FEAT(BFDOT_vi, aa64_bf16, do_dot_vector_idx_env, a, gen_helper_gvec_bfdot_idx) diff --git a/target/arm/tcg/translate-neon.c b/target/arm/tcg/translate-neon.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-neon.c +++ b/target/arm/tcg/translate-neon.c @@ -XXX,XX +XXX,XX @@ static bool trans_VSDOT(DisasContext *s, arg_VSDOT *a) return false; } return do_neon_ddda(s, a->q * 7, a->vd, a->vn, a->vm, 0, - gen_helper_gvec_sdot_b); + gen_helper_gvec_sdot_4b); } static bool trans_VUDOT(DisasContext *s, arg_VUDOT *a) @@ -XXX,XX +XXX,XX @@ static bool trans_VUDOT(DisasContext *s, arg_VUDOT *a) return false; } return do_neon_ddda(s, a->q * 7, a->vd, a->vn, a->vm, 0, - gen_helper_gvec_udot_b); + gen_helper_gvec_udot_4b); } static bool trans_VUSDOT(DisasContext *s, arg_VUSDOT *a) @@ -XXX,XX +XXX,XX @@ static bool trans_VUSDOT(DisasContext *s, arg_VUSDOT *a) return false; } return do_neon_ddda(s, a->q * 7, a->vd, a->vn, a->vm, 0, - gen_helper_gvec_usdot_b); + gen_helper_gvec_usdot_4b); } static bool trans_VDOT_b16(DisasContext *s, arg_VDOT_b16 *a) @@ -XXX,XX +XXX,XX @@ static bool trans_VSDOT_scalar(DisasContext *s, arg_VSDOT_scalar *a) return false; } return do_neon_ddda(s, a->q * 6, a->vd, a->vn, a->vm, a->index, - gen_helper_gvec_sdot_idx_b); + gen_helper_gvec_sdot_idx_4b); } static bool trans_VUDOT_scalar(DisasContext *s, arg_VUDOT_scalar *a) @@ -XXX,XX +XXX,XX @@ static bool trans_VUDOT_scalar(DisasContext *s, arg_VUDOT_scalar *a) return false; } return do_neon_ddda(s, a->q * 6, a->vd, a->vn, a->vm, a->index, - gen_helper_gvec_udot_idx_b); + gen_helper_gvec_udot_idx_4b); } static bool trans_VUSDOT_scalar(DisasContext *s, arg_VUSDOT_scalar *a) @@ -XXX,XX +XXX,XX @@ static bool trans_VUSDOT_scalar(DisasContext *s, arg_VUSDOT_scalar *a) return false; } return do_neon_ddda(s, a->q * 6, a->vd, a->vn, a->vm, a->index, - gen_helper_gvec_usdot_idx_b); + gen_helper_gvec_usdot_idx_4b); } static bool trans_VSUDOT_scalar(DisasContext *s, arg_VSUDOT_scalar *a) @@ -XXX,XX +XXX,XX @@ static bool trans_VSUDOT_scalar(DisasContext *s, arg_VSUDOT_scalar *a) return false; } return do_neon_ddda(s, a->q * 6, a->vd, a->vn, a->vm, a->index, - gen_helper_gvec_sudot_idx_b); + gen_helper_gvec_sudot_idx_4b); } static bool trans_VDOT_b16_scal(DisasContext *s, arg_VDOT_b16_scal *a) diff --git a/target/arm/tcg/translate-sve.c b/target/arm/tcg/translate-sve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sve.c +++ b/target/arm/tcg/translate-sve.c @@ -XXX,XX +XXX,XX @@ DO_ZZI(UMIN, umin) #undef DO_ZZI static gen_helper_gvec_4 * const dot_fns[2][2] = { - { gen_helper_gvec_sdot_b, gen_helper_gvec_sdot_h }, - { gen_helper_gvec_udot_b, gen_helper_gvec_udot_h } + { gen_helper_gvec_sdot_4b, gen_helper_gvec_sdot_4h }, + { gen_helper_gvec_udot_4b, gen_helper_gvec_udot_4h } }; TRANS_FEAT(DOT_zzzz, aa64_sve, gen_gvec_ool_zzzz, dot_fns[a->u][a->sz], a->rd, a->rn, a->rm, a->ra, 0) @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(DOT_zzzz, aa64_sve, gen_gvec_ool_zzzz, */ TRANS_FEAT(SDOT_zzxw_s, aa64_sve, gen_gvec_ool_arg_zzxz, - gen_helper_gvec_sdot_idx_b, a) + gen_helper_gvec_sdot_idx_4b, a) TRANS_FEAT(SDOT_zzxw_d, aa64_sve, gen_gvec_ool_arg_zzxz, - gen_helper_gvec_sdot_idx_h, a) + gen_helper_gvec_sdot_idx_4h, a) TRANS_FEAT(UDOT_zzxw_s, aa64_sve, gen_gvec_ool_arg_zzxz, - gen_helper_gvec_udot_idx_b, a) + gen_helper_gvec_udot_idx_4b, a) TRANS_FEAT(UDOT_zzxw_d, aa64_sve, gen_gvec_ool_arg_zzxz, - gen_helper_gvec_udot_idx_h, a) + gen_helper_gvec_udot_idx_4h, a) TRANS_FEAT(SUDOT_zzxw_s, aa64_sve_i8mm, gen_gvec_ool_arg_zzxz, - gen_helper_gvec_sudot_idx_b, a) + gen_helper_gvec_sudot_idx_4b, a) TRANS_FEAT(USDOT_zzxw_s, aa64_sve_i8mm, gen_gvec_ool_arg_zzxz, - gen_helper_gvec_usdot_idx_b, a) + gen_helper_gvec_usdot_idx_4b, a) #define DO_SVE2_RRX(NAME, FUNC) \ TRANS_FEAT(NAME, aa64_sve, gen_gvec_ool_zzz, FUNC, \ @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(SQRDCMLAH_zzzz, aa64_sve2, gen_gvec_ool_zzzz, sqrdcmlah_fns[a->esz], a->rd, a->rn, a->rm, a->ra, a->rot) TRANS_FEAT(USDOT_zzzz, aa64_sve_i8mm, gen_gvec_ool_arg_zzzz, - a->esz == 2 ? gen_helper_gvec_usdot_b : NULL, a, 0) + a->esz == 2 ? gen_helper_gvec_usdot_4b : NULL, a, 0) TRANS_FEAT_NONSTREAMING(AESMC, aa64_sve2_aes, gen_gvec_ool_zz, gen_helper_crypto_aesmc, a->rd, a->rd, 0) diff --git a/target/arm/tcg/vec_helper.c b/target/arm/tcg/vec_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/vec_helper.c +++ b/target/arm/tcg/vec_helper.c @@ -XXX,XX +XXX,XX @@ void HELPER(NAME)(void *vd, void *vn, void *vm, void *va, uint32_t desc) \ clear_tail(d, opr_sz, simd_maxsz(desc)); \ } -DO_DOT(gvec_sdot_b, int32_t, int8_t, int8_t) -DO_DOT(gvec_udot_b, uint32_t, uint8_t, uint8_t) -DO_DOT(gvec_usdot_b, uint32_t, uint8_t, int8_t) -DO_DOT(gvec_sdot_h, int64_t, int16_t, int16_t) -DO_DOT(gvec_udot_h, uint64_t, uint16_t, uint16_t) +DO_DOT(gvec_sdot_4b, int32_t, int8_t, int8_t) +DO_DOT(gvec_udot_4b, uint32_t, uint8_t, uint8_t) +DO_DOT(gvec_usdot_4b, uint32_t, uint8_t, int8_t) +DO_DOT(gvec_sdot_4h, int64_t, int16_t, int16_t) +DO_DOT(gvec_udot_4h, uint64_t, uint16_t, uint16_t) #define DO_DOT_IDX(NAME, TYPED, TYPEN, TYPEM, HD) \ void HELPER(NAME)(void *vd, void *vn, void *vm, void *va, uint32_t desc) \ @@ -XXX,XX +XXX,XX @@ void HELPER(NAME)(void *vd, void *vn, void *vm, void *va, uint32_t desc) \ clear_tail(d, opr_sz, simd_maxsz(desc)); \ } -DO_DOT_IDX(gvec_sdot_idx_b, int32_t, int8_t, int8_t, H4) -DO_DOT_IDX(gvec_udot_idx_b, uint32_t, uint8_t, uint8_t, H4) -DO_DOT_IDX(gvec_sudot_idx_b, int32_t, int8_t, uint8_t, H4) -DO_DOT_IDX(gvec_usdot_idx_b, int32_t, uint8_t, int8_t, H4) -DO_DOT_IDX(gvec_sdot_idx_h, int64_t, int16_t, int16_t, H8) -DO_DOT_IDX(gvec_udot_idx_h, uint64_t, uint16_t, uint16_t, H8) +DO_DOT_IDX(gvec_sdot_idx_4b, int32_t, int8_t, int8_t, H4) +DO_DOT_IDX(gvec_udot_idx_4b, uint32_t, uint8_t, uint8_t, H4) +DO_DOT_IDX(gvec_sudot_idx_4b, int32_t, int8_t, uint8_t, H4) +DO_DOT_IDX(gvec_usdot_idx_4b, int32_t, uint8_t, int8_t, H4) +DO_DOT_IDX(gvec_sdot_idx_4h, int64_t, int16_t, int16_t, H8) +DO_DOT_IDX(gvec_udot_idx_4h, uint64_t, uint16_t, uint16_t, H8) void HELPER(gvec_fcaddh)(void *vd, void *vn, void *vm, float_status *fpst, uint32_t desc) -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Message-id: 20250704142112.1018902-42-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper.h | 8 ++++ target/arm/tcg/sme.decode | 63 ++++++++++++++++++++++++- target/arm/tcg/translate-sme.c | 85 ++++++++++++++++++++++++++++++++++ target/arm/tcg/vec_helper.c | 51 ++++++++++++++++++++ 4 files changed, 206 insertions(+), 1 deletion(-) diff --git a/target/arm/tcg/helper.h b/target/arm/tcg/helper.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper.h +++ b/target/arm/tcg/helper.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_5(gvec_sdot_4h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_5(gvec_udot_4h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_5(gvec_usdot_4b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(gvec_sdot_2h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(gvec_udot_2h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) + DEF_HELPER_FLAGS_5(gvec_sdot_idx_4b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_5(gvec_udot_idx_4b, TCG_CALL_NO_RWG, @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_5(gvec_sudot_idx_4b, TCG_CALL_NO_RWG, DEF_HELPER_FLAGS_5(gvec_usdot_idx_4b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(gvec_sdot_idx_2h, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(gvec_udot_idx_2h, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) + DEF_HELPER_FLAGS_5(gvec_fcaddh, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_5(gvec_fcadds, TCG_CALL_NO_RWG, diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ FDOT_n1 11000001 001 1 .... 0 .. 100 ..... 00 ... @azz_nx1_o3 n=4 BFDOT_n1 11000001 001 0 .... 0 .. 100 ..... 10 ... @azz_nx1_o3 n=2 BFDOT_n1 11000001 001 1 .... 0 .. 100 ..... 10 ... @azz_nx1_o3 n=4 +USDOT_n1 11000001 001 0 .... 0 .. 101 ..... 01 ... @azz_nx1_o3 n=2 +USDOT_n1 11000001 001 1 .... 0 .. 101 ..... 01 ... @azz_nx1_o3 n=4 + +SUDOT_n1 11000001 001 0 .... 0 .. 101 ..... 11 ... @azz_nx1_o3 n=2 +SUDOT_n1 11000001 001 1 .... 0 .. 101 ..... 11 ... @azz_nx1_o3 n=4 + +SDOT_n1_4b 11000001 001 0 .... 0 .. 101 ..... 00 ... @azz_nx1_o3 n=2 +SDOT_n1_4b 11000001 001 1 .... 0 .. 101 ..... 00 ... @azz_nx1_o3 n=4 +SDOT_n1_4h 11000001 011 0 .... 0 .. 101 ..... 00 ... @azz_nx1_o3 n=2 +SDOT_n1_4h 11000001 011 1 .... 0 .. 101 ..... 00 ... @azz_nx1_o3 n=4 +SDOT_n1_2h 11000001 011 0 .... 0 .. 101 ..... 01 ... @azz_nx1_o3 n=2 +SDOT_n1_2h 11000001 011 1 .... 0 .. 101 ..... 01 ... @azz_nx1_o3 n=4 + +UDOT_n1_4b 11000001 001 0 .... 0 .. 101 ..... 10 ... @azz_nx1_o3 n=2 +UDOT_n1_4b 11000001 001 1 .... 0 .. 101 ..... 10 ... @azz_nx1_o3 n=4 +UDOT_n1_4h 11000001 011 0 .... 0 .. 101 ..... 10 ... @azz_nx1_o3 n=2 +UDOT_n1_4h 11000001 011 1 .... 0 .. 101 ..... 10 ... @azz_nx1_o3 n=4 +UDOT_n1_2h 11000001 011 0 .... 0 .. 101 ..... 11 ... @azz_nx1_o3 n=2 +UDOT_n1_2h 11000001 011 1 .... 0 .. 101 ..... 11 ... @azz_nx1_o3 n=4 + ### SME2 Multi-vector Multiple Array Vectors %zn_ax2 6:4 !function=times_2 @@ -XXX,XX +XXX,XX @@ FDOT_nn 11000001 101 ...01 0 .. 100 ...00 00 ... @azz_4x4_o3 BFDOT_nn 11000001 101 ....0 0 .. 100 ....0 10 ... @azz_2x2_o3 BFDOT_nn 11000001 101 ...01 0 .. 100 ...00 10 ... @azz_4x4_o3 +USDOT_nn 11000001 101 ....0 0 .. 101 ....0 01 ... @azz_2x2_o3 +USDOT_nn 11000001 101 ...01 0 .. 101 ...00 01 ... @azz_4x4_o3 + +SDOT_nn_4b 11000001 101 ....0 0 .. 101 ....0 00 ... @azz_2x2_o3 +SDOT_nn_4b 11000001 101 ...01 0 .. 101 ...00 00 ... @azz_4x4_o3 +SDOT_nn_4h 11000001 111 ....0 0 .. 101 ....0 00 ... @azz_2x2_o3 +SDOT_nn_4h 11000001 111 ...01 0 .. 101 ...00 00 ... @azz_4x4_o3 +SDOT_nn_2h 11000001 111 ....0 0 .. 101 ....0 01 ... @azz_2x2_o3 +SDOT_nn_2h 11000001 111 ...01 0 .. 101 ...00 01 ... @azz_4x4_o3 + +UDOT_nn_4b 11000001 101 ....0 0 .. 101 ....0 10 ... @azz_2x2_o3 +UDOT_nn_4b 11000001 101 ...01 0 .. 101 ...00 10 ... @azz_4x4_o3 +UDOT_nn_4h 11000001 111 ....0 0 .. 101 ....0 10 ... @azz_2x2_o3 +UDOT_nn_4h 11000001 111 ...01 0 .. 101 ...00 10 ... @azz_4x4_o3 +UDOT_nn_2h 11000001 111 ....0 0 .. 101 ....0 11 ... @azz_2x2_o3 +UDOT_nn_2h 11000001 111 ...01 0 .. 101 ...00 11 ... @azz_4x4_o3 + ### SME2 Multi-vector Indexed &azx_n n off rv zn zm idx @@ -XXX,XX +XXX,XX @@ BFMLSL_nx 11000001 1001 .... 1 .. 1 .. ...00 11 ... @azx_4x1_o2x2 @azx_2x1_i2_o3 ........ .... zm:4 . .. . idx:2 .... ... off:3 \ &azx_n n=2 rv=%mova_rv zn=%zn_ax2 @azx_4x1_i2_o3 ........ .... zm:4 . .. . idx:2 .... ... off:3 \ - &azx_n n=2 rv=%mova_rv zn=%zn_ax4 + &azx_n n=4 rv=%mova_rv zn=%zn_ax4 +@azx_2x1_i1_o3 ........ .... zm:4 . .. .. idx:1 .... ... off:3 \ + &azx_n n=2 rv=%mova_rv zn=%zn_ax2 +@azx_4x1_i1_o3 ........ .... zm:4 . .. .. idx:1 .... ... off:3 \ + &azx_n n=4 rv=%mova_rv zn=%zn_ax4 FDOT_nx 11000001 0101 .... 0 .. 1 .. ....0 01 ... @azx_2x1_i2_o3 FDOT_nx 11000001 0101 .... 1 .. 1 .. ...00 01 ... @azx_4x1_i2_o3 @@ -XXX,XX +XXX,XX @@ BFDOT_nx 11000001 0101 .... 1 .. 1 .. ...00 11 ... @azx_4x1_i2_o3 FVDOT 11000001 0101 .... 0 .. 0 .. ....0 01 ... @azx_2x1_i2_o3 BFVDOT 11000001 0101 .... 0 .. 0 .. ....0 11 ... @azx_2x1_i2_o3 + +SDOT_nx_2h 11000001 0101 .... 0 .. 1 .. ....0 00 ... @azx_2x1_i2_o3 +SDOT_nx_2h 11000001 0101 .... 1 .. 1 .. ...00 00 ... @azx_4x1_i2_o3 +SDOT_nx_4b 11000001 0101 .... 0 .. 1 .. ....1 00 ... @azx_2x1_i2_o3 +SDOT_nx_4b 11000001 0101 .... 1 .. 1 .. ...01 00 ... @azx_4x1_i2_o3 +SDOT_nx_4h 11000001 1101 .... 0 .. 00 . ....0 01 ... @azx_2x1_i1_o3 +SDOT_nx_4h 11000001 1101 .... 1 .. 00 . ...00 01 ... @azx_4x1_i1_o3 + +UDOT_nx_2h 11000001 0101 .... 0 .. 1 .. ....0 10 ... @azx_2x1_i2_o3 +UDOT_nx_2h 11000001 0101 .... 1 .. 1 .. ...00 10 ... @azx_4x1_i2_o3 +UDOT_nx_4b 11000001 0101 .... 0 .. 1 .. ....1 10 ... @azx_2x1_i2_o3 +UDOT_nx_4b 11000001 0101 .... 1 .. 1 .. ...01 10 ... @azx_4x1_i2_o3 +UDOT_nx_4h 11000001 1101 .... 0 .. 00 . ....0 11 ... @azx_2x1_i1_o3 +UDOT_nx_4h 11000001 1101 .... 1 .. 00 . ...00 11 ... @azx_4x1_i1_o3 + +USDOT_nx 11000001 0101 .... 0 .. 1 .. ....1 01 ... @azx_2x1_i2_o3 +USDOT_nx 11000001 0101 .... 1 .. 1 .. ...01 01 ... @azx_4x1_i2_o3 + +SUDOT_nx 11000001 0101 .... 0 .. 1 .. ....1 11 ... @azx_2x1_i2_o3 +SUDOT_nx 11000001 0101 .... 1 .. 1 .. ...01 11 ... @azx_4x1_i2_o3 diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ static bool do_vdot(DisasContext *s, arg_azx_n *a, gen_helper_gvec_4_ptr *fn) TRANS_FEAT(FVDOT, aa64_sme, do_vdot, a, gen_helper_sme2_fvdot_idx_h) TRANS_FEAT(BFVDOT, aa64_sme, do_vdot, a, gen_helper_sme2_bfvdot_idx) + +/* + * Expand array multi-vector single (n1), array multi-vector (nn), + * and array multi-vector indexed (nx), for integer accumulate. + * multi: true for nn, false for n1. + * data: stuff for simd_data, including any index. + */ +static bool do_azz_acc(DisasContext *s, int nreg, int nsel, + int rv, int off, int zn, int zm, + int data, int shsel, bool multi, + gen_helper_gvec_4 *fn) +{ + if (sme_smza_enabled_check(s)) { + int svl = streaming_vec_reg_size(s); + int vstride = svl / nreg; + TCGv_ptr t_za = get_zarray(s, rv, off, nreg, nsel); + TCGv_ptr t = tcg_temp_new_ptr(); + + for (int r = 0; r < nreg; ++r) { + TCGv_ptr t_zn = vec_full_reg_ptr(s, zn); + TCGv_ptr t_zm = vec_full_reg_ptr(s, zm); + + for (int i = 0; i < nsel; ++i) { + int o_za = (r * vstride + i) * sizeof(ARMVectorReg); + int desc = simd_desc(svl, svl, data | (i << shsel)); + + tcg_gen_addi_ptr(t, t_za, o_za); + fn(t, t_zn, t_zm, t, tcg_constant_i32(desc)); + } + + /* + * For multiple-and-single vectors, Zn may wrap. + * For multiple vectors, both Zn and Zm are aligned. + */ + zn = (zn + 1) % 32; + zm += multi; + } + } + return true; +} + +static bool do_dot(DisasContext *s, arg_azz_n *a, bool multi, + gen_helper_gvec_4 *fn) +{ + return do_azz_acc(s, a->n, 1, a->rv, a->off, a->zn, a->zm, + 0, 0, multi, fn); +} + +static void gen_helper_gvec_sudot_4b(TCGv_ptr d, TCGv_ptr n, TCGv_ptr m, + TCGv_ptr a, TCGv_i32 desc) +{ + gen_helper_gvec_usdot_4b(d, m, n, a, desc); +} + +TRANS_FEAT(USDOT_n1, aa64_sme2, do_dot, a, false, gen_helper_gvec_usdot_4b) +TRANS_FEAT(SUDOT_n1, aa64_sme2, do_dot, a, false, gen_helper_gvec_sudot_4b) +TRANS_FEAT(SDOT_n1_2h, aa64_sme2, do_dot, a, false, gen_helper_gvec_sdot_2h) +TRANS_FEAT(UDOT_n1_2h, aa64_sme2, do_dot, a, false, gen_helper_gvec_udot_2h) +TRANS_FEAT(SDOT_n1_4b, aa64_sme2, do_dot, a, false, gen_helper_gvec_sdot_4b) +TRANS_FEAT(UDOT_n1_4b, aa64_sme2, do_dot, a, false, gen_helper_gvec_udot_4b) +TRANS_FEAT(SDOT_n1_4h, aa64_sme2_i16i64, do_dot, a, false, gen_helper_gvec_sdot_4h) +TRANS_FEAT(UDOT_n1_4h, aa64_sme2_i16i64, do_dot, a, false, gen_helper_gvec_udot_4h) + +TRANS_FEAT(USDOT_nn, aa64_sme2, do_dot, a, true, gen_helper_gvec_usdot_4b) +TRANS_FEAT(SDOT_nn_2h, aa64_sme2, do_dot, a, true, gen_helper_gvec_sdot_2h) +TRANS_FEAT(UDOT_nn_2h, aa64_sme2, do_dot, a, true, gen_helper_gvec_udot_2h) +TRANS_FEAT(SDOT_nn_4b, aa64_sme2, do_dot, a, true, gen_helper_gvec_sdot_4b) +TRANS_FEAT(UDOT_nn_4b, aa64_sme2, do_dot, a, true, gen_helper_gvec_udot_4b) +TRANS_FEAT(SDOT_nn_4h, aa64_sme2_i16i64, do_dot, a, true, gen_helper_gvec_sdot_4h) +TRANS_FEAT(UDOT_nn_4h, aa64_sme2_i16i64, do_dot, a, true, gen_helper_gvec_udot_4h) + +static bool do_dot_nx(DisasContext *s, arg_azx_n *a, gen_helper_gvec_4 *fn) +{ + return do_azz_acc(s, a->n, 1, a->rv, a->off, a->zn, a->zm, + a->idx, 0, false, fn); +} + +TRANS_FEAT(USDOT_nx, aa64_sme2, do_dot_nx, a, gen_helper_gvec_usdot_idx_4b) +TRANS_FEAT(SUDOT_nx, aa64_sme2, do_dot_nx, a, gen_helper_gvec_sudot_idx_4b) +TRANS_FEAT(SDOT_nx_2h, aa64_sme2, do_dot_nx, a, gen_helper_gvec_sdot_idx_2h) +TRANS_FEAT(UDOT_nx_2h, aa64_sme2, do_dot_nx, a, gen_helper_gvec_udot_idx_2h) +TRANS_FEAT(SDOT_nx_4b, aa64_sme2, do_dot_nx, a, gen_helper_gvec_sdot_idx_4b) +TRANS_FEAT(UDOT_nx_4b, aa64_sme2, do_dot_nx, a, gen_helper_gvec_udot_idx_4b) +TRANS_FEAT(SDOT_nx_4h, aa64_sme2_i16i64, do_dot_nx, a, gen_helper_gvec_sdot_idx_4h) +TRANS_FEAT(UDOT_nx_4h, aa64_sme2_i16i64, do_dot_nx, a, gen_helper_gvec_udot_idx_4h) diff --git a/target/arm/tcg/vec_helper.c b/target/arm/tcg/vec_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/vec_helper.c +++ b/target/arm/tcg/vec_helper.c @@ -XXX,XX +XXX,XX @@ DO_DOT_IDX(gvec_usdot_idx_4b, int32_t, uint8_t, int8_t, H4) DO_DOT_IDX(gvec_sdot_idx_4h, int64_t, int16_t, int16_t, H8) DO_DOT_IDX(gvec_udot_idx_4h, uint64_t, uint16_t, uint16_t, H8) +#undef DO_DOT +#undef DO_DOT_IDX + +/* Similar for 2-way dot product */ +#define DO_DOT(NAME, TYPED, TYPEN, TYPEM) \ +void HELPER(NAME)(void *vd, void *vn, void *vm, void *va, uint32_t desc) \ +{ \ + intptr_t i, opr_sz = simd_oprsz(desc); \ + TYPED *d = vd, *a = va; \ + TYPEN *n = vn; \ + TYPEM *m = vm; \ + for (i = 0; i < opr_sz / sizeof(TYPED); ++i) { \ + d[i] = (a[i] + \ + (TYPED)n[i * 2 + 0] * m[i * 2 + 0] + \ + (TYPED)n[i * 2 + 1] * m[i * 2 + 1]); \ + } \ + clear_tail(d, opr_sz, simd_maxsz(desc)); \ +} + +#define DO_DOT_IDX(NAME, TYPED, TYPEN, TYPEM, HD) \ +void HELPER(NAME)(void *vd, void *vn, void *vm, void *va, uint32_t desc) \ +{ \ + intptr_t i = 0, opr_sz = simd_oprsz(desc); \ + intptr_t opr_sz_n = opr_sz / sizeof(TYPED); \ + intptr_t segend = MIN(16 / sizeof(TYPED), opr_sz_n); \ + intptr_t index = simd_data(desc); \ + TYPED *d = vd, *a = va; \ + TYPEN *n = vn; \ + TYPEM *m_indexed = (TYPEM *)vm + HD(index) * 2; \ + do { \ + TYPED m0 = m_indexed[i * 2 + 0]; \ + TYPED m1 = m_indexed[i * 2 + 1]; \ + do { \ + d[i] = (a[i] + \ + n[i * 2 + 0] * m0 + \ + n[i * 2 + 1] * m1); \ + } while (++i < segend); \ + segend = i + (16 / sizeof(TYPED)); \ + } while (i < opr_sz_n); \ + clear_tail(d, opr_sz, simd_maxsz(desc)); \ +} + +DO_DOT(gvec_sdot_2h, int32_t, int16_t, int16_t) +DO_DOT(gvec_udot_2h, uint32_t, uint16_t, uint16_t) + +DO_DOT_IDX(gvec_sdot_idx_2h, int32_t, int16_t, int16_t, H4) +DO_DOT_IDX(gvec_udot_idx_2h, uint32_t, uint16_t, uint16_t, H4) + +#undef DO_DOT +#undef DO_DOT_IDX + void HELPER(gvec_fcaddh)(void *vd, void *vn, void *vm, float_status *fpst, uint32_t desc) { -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Emphasize the 4-way nature of these dot products. Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-43-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/sve.decode | 12 ++++++------ target/arm/tcg/translate-sve.c | 12 ++++++------ 2 files changed, 12 insertions(+), 12 deletions(-) diff --git a/target/arm/tcg/sve.decode b/target/arm/tcg/sve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve.decode +++ b/target/arm/tcg/sve.decode @@ -XXX,XX +XXX,XX @@ CDOT_zzzz 01000100 esz:2 0 rm:5 0001 rot:2 rn:5 rd:5 ra=%reg_movprfx #### SVE Multiply - Indexed # SVE integer dot product (indexed) -SDOT_zzxw_s 01000100 10 1 ..... 000000 ..... ..... @rrxr_2 esz=2 -SDOT_zzxw_d 01000100 11 1 ..... 000000 ..... ..... @rrxr_1 esz=3 -UDOT_zzxw_s 01000100 10 1 ..... 000001 ..... ..... @rrxr_2 esz=2 -UDOT_zzxw_d 01000100 11 1 ..... 000001 ..... ..... @rrxr_1 esz=3 +SDOT_zzxw_4s 01000100 10 1 ..... 000000 ..... ..... @rrxr_2 esz=2 +SDOT_zzxw_4d 01000100 11 1 ..... 000000 ..... ..... @rrxr_1 esz=3 +UDOT_zzxw_4s 01000100 10 1 ..... 000001 ..... ..... @rrxr_2 esz=2 +UDOT_zzxw_4d 01000100 11 1 ..... 000001 ..... ..... @rrxr_1 esz=3 # SVE2 integer multiply-add (indexed) MLA_zzxz_h 01000100 0. 1 ..... 000010 ..... ..... @rrxr_3 esz=1 @@ -XXX,XX +XXX,XX @@ SQRDMLSH_zzxz_s 01000100 10 1 ..... 000101 ..... ..... @rrxr_2 esz=2 SQRDMLSH_zzxz_d 01000100 11 1 ..... 000101 ..... ..... @rrxr_1 esz=3 # SVE mixed sign dot product (indexed) -USDOT_zzxw_s 01000100 10 1 ..... 000110 ..... ..... @rrxr_2 esz=2 -SUDOT_zzxw_s 01000100 10 1 ..... 000111 ..... ..... @rrxr_2 esz=2 +USDOT_zzxw_4s 01000100 10 1 ..... 000110 ..... ..... @rrxr_2 esz=2 +SUDOT_zzxw_4s 01000100 10 1 ..... 000111 ..... ..... @rrxr_2 esz=2 # SVE2 saturating multiply-add (indexed) SQDMLALB_zzxw_s 01000100 10 1 ..... 0010.0 ..... ..... @rrxr_3a esz=2 diff --git a/target/arm/tcg/translate-sve.c b/target/arm/tcg/translate-sve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sve.c +++ b/target/arm/tcg/translate-sve.c @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(DOT_zzzz, aa64_sve, gen_gvec_ool_zzzz, * SVE Multiply - Indexed */ -TRANS_FEAT(SDOT_zzxw_s, aa64_sve, gen_gvec_ool_arg_zzxz, +TRANS_FEAT(SDOT_zzxw_4s, aa64_sve, gen_gvec_ool_arg_zzxz, gen_helper_gvec_sdot_idx_4b, a) -TRANS_FEAT(SDOT_zzxw_d, aa64_sve, gen_gvec_ool_arg_zzxz, +TRANS_FEAT(SDOT_zzxw_4d, aa64_sve, gen_gvec_ool_arg_zzxz, gen_helper_gvec_sdot_idx_4h, a) -TRANS_FEAT(UDOT_zzxw_s, aa64_sve, gen_gvec_ool_arg_zzxz, +TRANS_FEAT(UDOT_zzxw_4s, aa64_sve, gen_gvec_ool_arg_zzxz, gen_helper_gvec_udot_idx_4b, a) -TRANS_FEAT(UDOT_zzxw_d, aa64_sve, gen_gvec_ool_arg_zzxz, +TRANS_FEAT(UDOT_zzxw_4d, aa64_sve, gen_gvec_ool_arg_zzxz, gen_helper_gvec_udot_idx_4h, a) -TRANS_FEAT(SUDOT_zzxw_s, aa64_sve_i8mm, gen_gvec_ool_arg_zzxz, +TRANS_FEAT(SUDOT_zzxw_4s, aa64_sve_i8mm, gen_gvec_ool_arg_zzxz, gen_helper_gvec_sudot_idx_4b, a) -TRANS_FEAT(USDOT_zzxw_s, aa64_sve_i8mm, gen_gvec_ool_arg_zzxz, +TRANS_FEAT(USDOT_zzxw_4s, aa64_sve_i8mm, gen_gvec_ool_arg_zzxz, gen_helper_gvec_usdot_idx_4b, a) #define DO_SVE2_RRX(NAME, FUNC) \ -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Rename to USDOT_zzzz_4s and force size=2 during decode. Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-44-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/sve.decode | 2 +- target/arm/tcg/translate-sve.c | 4 ++-- 2 files changed, 3 insertions(+), 3 deletions(-) diff --git a/target/arm/tcg/sve.decode b/target/arm/tcg/sve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve.decode +++ b/target/arm/tcg/sve.decode @@ -XXX,XX +XXX,XX @@ SQRDCMLAH_zzzz 01000100 esz:2 0 rm:5 0011 rot:2 rn:5 rd:5 ra=%reg_movprfx ## SVE mixed sign dot product -USDOT_zzzz 01000100 .. 0 ..... 011 110 ..... ..... @rda_rn_rm +USDOT_zzzz_4s 01000100 10 0 ..... 011 110 ..... ..... @rda_rn_rm_ex esz=2 ### SVE2 floating point matrix multiply accumulate BFMMLA 01100100 01 1 ..... 111 001 ..... ..... @rda_rn_rm_ex esz=1 diff --git a/target/arm/tcg/translate-sve.c b/target/arm/tcg/translate-sve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sve.c +++ b/target/arm/tcg/translate-sve.c @@ -XXX,XX +XXX,XX @@ static gen_helper_gvec_4 * const sqrdcmlah_fns[] = { TRANS_FEAT(SQRDCMLAH_zzzz, aa64_sve2, gen_gvec_ool_zzzz, sqrdcmlah_fns[a->esz], a->rd, a->rn, a->rm, a->ra, a->rot) -TRANS_FEAT(USDOT_zzzz, aa64_sve_i8mm, gen_gvec_ool_arg_zzzz, - a->esz == 2 ? gen_helper_gvec_usdot_4b : NULL, a, 0) +TRANS_FEAT(USDOT_zzzz_4s, aa64_sve_i8mm, gen_gvec_ool_arg_zzzz, + gen_helper_gvec_usdot_4b, a, 0) TRANS_FEAT_NONSTREAMING(AESMC, aa64_sve2_aes, gen_gvec_ool_zz, gen_helper_crypto_aesmc, a->rd, a->rd, 0) -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-45-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/sve.decode | 8 +++++++- target/arm/tcg/translate-sve.c | 10 ++++++++++ 2 files changed, 17 insertions(+), 1 deletion(-) diff --git a/target/arm/tcg/sve.decode b/target/arm/tcg/sve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve.decode +++ b/target/arm/tcg/sve.decode @@ -XXX,XX +XXX,XX @@ SDOT_zzxw_4d 01000100 11 1 ..... 000000 ..... ..... @rrxr_1 esz=3 UDOT_zzxw_4s 01000100 10 1 ..... 000001 ..... ..... @rrxr_2 esz=2 UDOT_zzxw_4d 01000100 11 1 ..... 000001 ..... ..... @rrxr_1 esz=3 +SDOT_zzxw_2s 01000100 10 0 ..... 110010 ..... ..... @rrxr_2 esz=2 +UDOT_zzxw_2s 01000100 10 0 ..... 110011 ..... ..... @rrxr_2 esz=2 + # SVE2 integer multiply-add (indexed) MLA_zzxz_h 01000100 0. 1 ..... 000010 ..... ..... @rrxr_3 esz=1 MLA_zzxz_s 01000100 10 1 ..... 000010 ..... ..... @rrxr_2 esz=2 @@ -XXX,XX +XXX,XX @@ UMLSLT_zzzw 01000100 .. 0 ..... 010 111 ..... ..... @rda_rn_rm CMLA_zzzz 01000100 esz:2 0 rm:5 0010 rot:2 rn:5 rd:5 ra=%reg_movprfx SQRDCMLAH_zzzz 01000100 esz:2 0 rm:5 0011 rot:2 rn:5 rd:5 ra=%reg_movprfx -## SVE mixed sign dot product +## SVE dot product + +SDOT_zzzz_2s 01000100 00 0 ..... 110 010 ..... ..... @rda_rn_rm_ex esz=2 +UDOT_zzzz_2s 01000100 00 0 ..... 110 011 ..... ..... @rda_rn_rm_ex esz=2 USDOT_zzzz_4s 01000100 10 0 ..... 011 110 ..... ..... @rda_rn_rm_ex esz=2 diff --git a/target/arm/tcg/translate-sve.c b/target/arm/tcg/translate-sve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sve.c +++ b/target/arm/tcg/translate-sve.c @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(SUDOT_zzxw_4s, aa64_sve_i8mm, gen_gvec_ool_arg_zzxz, TRANS_FEAT(USDOT_zzxw_4s, aa64_sve_i8mm, gen_gvec_ool_arg_zzxz, gen_helper_gvec_usdot_idx_4b, a) +TRANS_FEAT(SDOT_zzxw_2s, aa64_sme2_or_sve2p1, gen_gvec_ool_arg_zzxz, + gen_helper_gvec_sdot_idx_2h, a) +TRANS_FEAT(UDOT_zzxw_2s, aa64_sme2_or_sve2p1, gen_gvec_ool_arg_zzxz, + gen_helper_gvec_udot_idx_2h, a) + #define DO_SVE2_RRX(NAME, FUNC) \ TRANS_FEAT(NAME, aa64_sve, gen_gvec_ool_zzz, FUNC, \ a->rd, a->rn, a->rm, a->index) @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(SQRDCMLAH_zzzz, aa64_sve2, gen_gvec_ool_zzzz, TRANS_FEAT(USDOT_zzzz_4s, aa64_sve_i8mm, gen_gvec_ool_arg_zzzz, gen_helper_gvec_usdot_4b, a, 0) +TRANS_FEAT(SDOT_zzzz_2s, aa64_sme2_or_sve2p1, gen_gvec_ool_arg_zzzz, + gen_helper_gvec_sdot_2h, a, 0) +TRANS_FEAT(UDOT_zzzz_2s, aa64_sme2_or_sve2p1, gen_gvec_ool_arg_zzzz, + gen_helper_gvec_udot_2h, a, 0) + TRANS_FEAT_NONSTREAMING(AESMC, aa64_sve2_aes, gen_gvec_ool_zz, gen_helper_crypto_aesmc, a->rd, a->rd, 0) TRANS_FEAT_NONSTREAMING(AESIMC, aa64_sve2_aes, gen_gvec_ool_zz, -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-46-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper-sme.h | 11 +++++++++ target/arm/tcg/sme.decode | 11 +++++++++ target/arm/tcg/sme_helper.c | 42 ++++++++++++++++++++++++++++++++++ target/arm/tcg/translate-sme.c | 23 +++++++++++++++++++ 4 files changed, 87 insertions(+) diff --git a/target/arm/tcg/helper-sme.h b/target/arm/tcg/helper-sme.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-sme.h +++ b/target/arm/tcg/helper-sme.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_6(sme2_fdot_idx_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, env, i32) DEF_HELPER_FLAGS_6(sme2_fvdot_idx_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, env, i32) + +DEF_HELPER_FLAGS_4(sme2_svdot_idx_4b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sme2_uvdot_idx_4b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sme2_suvdot_idx_4b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sme2_usvdot_idx_4b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) + +DEF_HELPER_FLAGS_4(sme2_svdot_idx_4h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sme2_uvdot_idx_4h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) + +DEF_HELPER_FLAGS_4(sme2_svdot_idx_2h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sme2_uvdot_idx_2h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ USDOT_nx 11000001 0101 .... 1 .. 1 .. ...01 01 ... @azx_4x1_i2_o3 SUDOT_nx 11000001 0101 .... 0 .. 1 .. ....1 11 ... @azx_2x1_i2_o3 SUDOT_nx 11000001 0101 .... 1 .. 1 .. ...01 11 ... @azx_4x1_i2_o3 + +SVDOT_nx_2h 11000001 0101 .... 0 .. 0 .. ....1 00 ... @azx_2x1_i2_o3 +SVDOT_nx_4b 11000001 0101 .... 1 .. 0 .. ...01 00 ... @azx_4x1_i2_o3 +SVDOT_nx_4h 11000001 1101 .... 1 .. 01 . ...00 01 ... @azx_4x1_i1_o3 + +UVDOT_nx_2h 11000001 0101 .... 0 .. 0 .. ....1 10 ... @azx_2x1_i2_o3 +UVDOT_nx_4b 11000001 0101 .... 1 .. 0 .. ...01 10 ... @azx_4x1_i2_o3 +UVDOT_nx_4h 11000001 1101 .... 1 .. 01 . ...00 11 ... @azx_4x1_i1_o3 + +SUVDOT_nx_4b 11000001 0101 .... 1 .. 0 .. ...01 11 ... @azx_4x1_i2_o3 +USVDOT_nx_4b 11000001 0101 .... 1 .. 0 .. ...01 01 ... @azx_4x1_i2_o3 diff --git a/target/arm/tcg/sme_helper.c b/target/arm/tcg/sme_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme_helper.c +++ b/target/arm/tcg/sme_helper.c @@ -XXX,XX +XXX,XX @@ DEF_IMOP_16x2_32(umopa2_s, uint16_t, uint16_t) DEF_IMOPH(sme2, smopa2, s) DEF_IMOPH(sme2, umopa2, s) + +#define DO_VDOT_IDX(NAME, TYPED, TYPEN, TYPEM, HD, HN) \ +void HELPER(NAME)(void *vd, void *vn, void *vm, uint32_t desc) \ +{ \ + intptr_t svl = simd_oprsz(desc); \ + intptr_t elements = svl / sizeof(TYPED); \ + intptr_t eltperseg = 16 / sizeof(TYPED); \ + intptr_t nreg = sizeof(TYPED) / sizeof(TYPEN); \ + intptr_t vstride = (svl / nreg) * sizeof(ARMVectorReg); \ + intptr_t zstride = sizeof(ARMVectorReg) / sizeof(TYPEN); \ + intptr_t idx = extract32(desc, SIMD_DATA_SHIFT, 2); \ + TYPEN *n = vn; \ + TYPEM *m = vm; \ + for (intptr_t r = 0; r < nreg; r++) { \ + TYPED *d = vd + r * vstride; \ + for (intptr_t seg = 0; seg < elements; seg += eltperseg) { \ + intptr_t s = seg + idx; \ + for (intptr_t e = seg; e < seg + eltperseg; e++) { \ + TYPED sum = d[HD(e)]; \ + for (intptr_t i = 0; i < nreg; i++) { \ + TYPED nn = n[i * zstride + HN(nreg * e + r)]; \ + TYPED mm = m[HN(nreg * s + i)]; \ + sum += nn * mm; \ + } \ + d[HD(e)] = sum; \ + } \ + } \ + } \ +} + +DO_VDOT_IDX(sme2_svdot_idx_4b, int32_t, int8_t, int8_t, H4, H1) +DO_VDOT_IDX(sme2_uvdot_idx_4b, uint32_t, uint8_t, uint8_t, H4, H1) +DO_VDOT_IDX(sme2_suvdot_idx_4b, int32_t, int8_t, uint8_t, H4, H1) +DO_VDOT_IDX(sme2_usvdot_idx_4b, int32_t, uint8_t, int8_t, H4, H1) + +DO_VDOT_IDX(sme2_svdot_idx_4h, int64_t, int16_t, int16_t, H8, H2) +DO_VDOT_IDX(sme2_uvdot_idx_4h, uint64_t, uint16_t, uint16_t, H8, H2) + +DO_VDOT_IDX(sme2_svdot_idx_2h, int32_t, int16_t, int16_t, H4, H2) +DO_VDOT_IDX(sme2_uvdot_idx_2h, uint32_t, uint16_t, uint16_t, H4, H2) + +#undef DO_VDOT_IDX diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(SDOT_nx_4b, aa64_sme2, do_dot_nx, a, gen_helper_gvec_sdot_idx_4b) TRANS_FEAT(UDOT_nx_4b, aa64_sme2, do_dot_nx, a, gen_helper_gvec_udot_idx_4b) TRANS_FEAT(SDOT_nx_4h, aa64_sme2_i16i64, do_dot_nx, a, gen_helper_gvec_sdot_idx_4h) TRANS_FEAT(UDOT_nx_4h, aa64_sme2_i16i64, do_dot_nx, a, gen_helper_gvec_udot_idx_4h) + +static bool do_vdot_nx(DisasContext *s, arg_azx_n *a, gen_helper_gvec_3 *fn) +{ + if (sme_smza_enabled_check(s)) { + int svl = streaming_vec_reg_size(s); + fn(get_zarray(s, a->rv, a->off, a->n, 0), + vec_full_reg_ptr(s, a->zn), + vec_full_reg_ptr(s, a->zm), + tcg_constant_i32(simd_desc(svl, svl, a->idx))); + } + return true; +} + +TRANS_FEAT(SVDOT_nx_2h, aa64_sme2, do_vdot_nx, a, gen_helper_sme2_svdot_idx_2h) +TRANS_FEAT(SVDOT_nx_4b, aa64_sme2, do_vdot_nx, a, gen_helper_sme2_svdot_idx_4b) +TRANS_FEAT(SVDOT_nx_4h, aa64_sme2, do_vdot_nx, a, gen_helper_sme2_svdot_idx_4h) + +TRANS_FEAT(UVDOT_nx_2h, aa64_sme2, do_vdot_nx, a, gen_helper_sme2_uvdot_idx_2h) +TRANS_FEAT(UVDOT_nx_4b, aa64_sme2, do_vdot_nx, a, gen_helper_sme2_uvdot_idx_4b) +TRANS_FEAT(UVDOT_nx_4h, aa64_sme2, do_vdot_nx, a, gen_helper_sme2_uvdot_idx_4h) + +TRANS_FEAT(SUVDOT_nx_4b, aa64_sme2, do_vdot_nx, a, gen_helper_sme2_suvdot_idx_4b) +TRANS_FEAT(USVDOT_nx_4b, aa64_sme2, do_vdot_nx, a, gen_helper_sme2_usvdot_idx_4b) -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-47-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper-sme.h | 21 +++++ target/arm/tcg/sme.decode | 168 +++++++++++++++++++++++++++++++++ target/arm/tcg/sme_helper.c | 59 ++++++++++++ target/arm/tcg/translate-sme.c | 84 +++++++++++++++++ 4 files changed, 332 insertions(+) diff --git a/target/arm/tcg/helper-sme.h b/target/arm/tcg/helper-sme.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-sme.h +++ b/target/arm/tcg/helper-sme.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_4(sme2_uvdot_idx_4h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_4(sme2_svdot_idx_2h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_4(sme2_uvdot_idx_2h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) + +DEF_HELPER_FLAGS_5(sme2_smlall_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sme2_smlall_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sme2_smlsll_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sme2_smlsll_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sme2_umlall_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sme2_umlall_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sme2_umlsll_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sme2_umlsll_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sme2_usmlall_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) + +DEF_HELPER_FLAGS_5(sme2_smlall_idx_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sme2_smlall_idx_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sme2_smlsll_idx_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sme2_smlsll_idx_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sme2_umlall_idx_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sme2_umlall_idx_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sme2_umlsll_idx_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sme2_umlsll_idx_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sme2_usmlall_idx_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(sme2_sumlall_idx_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ UDOT_n1_4h 11000001 011 1 .... 0 .. 101 ..... 10 ... @azz_nx1_o3 n=4 UDOT_n1_2h 11000001 011 0 .... 0 .. 101 ..... 11 ... @azz_nx1_o3 n=2 UDOT_n1_2h 11000001 011 1 .... 0 .. 101 ..... 11 ... @azz_nx1_o3 n=4 +SMLAL_n1 11000001 011 0 .... 0 .. 011 ..... 00 ... @azz_nx1_o3x2 n=1 +SMLAL_n1 11000001 011 0 .... 0 .. 010 ..... 000 .. @azz_nx1_o2x2 n=2 +SMLAL_n1 11000001 011 1 .... 0 .. 010 ..... 000 .. @azz_nx1_o2x2 n=4 + +SMLSL_n1 11000001 011 0 .... 0 .. 011 ..... 01 ... @azz_nx1_o3x2 n=1 +SMLSL_n1 11000001 011 0 .... 0 .. 010 ..... 010 .. @azz_nx1_o2x2 n=2 +SMLSL_n1 11000001 011 1 .... 0 .. 010 ..... 010 .. @azz_nx1_o2x2 n=4 + +UMLAL_n1 11000001 011 0 .... 0 .. 011 ..... 10 ... @azz_nx1_o3x2 n=1 +UMLAL_n1 11000001 011 0 .... 0 .. 010 ..... 100 .. @azz_nx1_o2x2 n=2 +UMLAL_n1 11000001 011 1 .... 0 .. 010 ..... 100 .. @azz_nx1_o2x2 n=4 + +UMLSL_n1 11000001 011 0 .... 0 .. 011 ..... 11 ... @azz_nx1_o3x2 n=1 +UMLSL_n1 11000001 011 0 .... 0 .. 010 ..... 110 .. @azz_nx1_o2x2 n=2 +UMLSL_n1 11000001 011 1 .... 0 .. 010 ..... 110 .. @azz_nx1_o2x2 n=4 + +%off2_x4 0:2 !function=times_4 +%off1_x4 0:1 !function=times_4 + +@azz_nx1_o2x4 ........ ... . zm:4 . .. ... zn:5 ... .. \ + &azz_n off=%off2_x4 rv=%mova_rv +@azz_nx1_o1x4 ........ ... . zm:4 . .. ... zn:5 .... . \ + &azz_n off=%off1_x4 rv=%mova_rv + +SMLALL_n1_s 11000001 001 0 .... 0 .. 001 ..... 000 .. @azz_nx1_o2x4 n=1 +SMLALL_n1_d 11000001 011 0 .... 0 .. 001 ..... 000 .. @azz_nx1_o2x4 n=1 +SMLALL_n1_s 11000001 001 0 .... 0 .. 000 ..... 0000 . @azz_nx1_o1x4 n=2 +SMLALL_n1_d 11000001 011 0 .... 0 .. 000 ..... 0000 . @azz_nx1_o1x4 n=2 +SMLALL_n1_s 11000001 001 1 .... 0 .. 000 ..... 0000 . @azz_nx1_o1x4 n=4 +SMLALL_n1_d 11000001 011 1 .... 0 .. 000 ..... 0000 . @azz_nx1_o1x4 n=4 + +SMLSLL_n1_s 11000001 001 0 .... 0 .. 001 ..... 010 .. @azz_nx1_o2x4 n=1 +SMLSLL_n1_d 11000001 011 0 .... 0 .. 001 ..... 010 .. @azz_nx1_o2x4 n=1 +SMLSLL_n1_s 11000001 001 0 .... 0 .. 000 ..... 0100 . @azz_nx1_o1x4 n=2 +SMLSLL_n1_d 11000001 011 0 .... 0 .. 000 ..... 0100 . @azz_nx1_o1x4 n=2 +SMLSLL_n1_s 11000001 001 1 .... 0 .. 000 ..... 0100 . @azz_nx1_o1x4 n=4 +SMLSLL_n1_d 11000001 011 1 .... 0 .. 000 ..... 0100 . @azz_nx1_o1x4 n=4 + +UMLALL_n1_s 11000001 001 0 .... 0 .. 001 ..... 100 .. @azz_nx1_o2x4 n=1 +UMLALL_n1_d 11000001 011 0 .... 0 .. 001 ..... 100 .. @azz_nx1_o2x4 n=1 +UMLALL_n1_s 11000001 001 0 .... 0 .. 000 ..... 1000 . @azz_nx1_o1x4 n=2 +UMLALL_n1_d 11000001 011 0 .... 0 .. 000 ..... 1000 . @azz_nx1_o1x4 n=2 +UMLALL_n1_s 11000001 001 1 .... 0 .. 000 ..... 1000 . @azz_nx1_o1x4 n=4 +UMLALL_n1_d 11000001 011 1 .... 0 .. 000 ..... 1000 . @azz_nx1_o1x4 n=4 + +UMLSLL_n1_s 11000001 001 0 .... 0 .. 001 ..... 110 .. @azz_nx1_o2x4 n=1 +UMLSLL_n1_d 11000001 011 0 .... 0 .. 001 ..... 110 .. @azz_nx1_o2x4 n=1 +UMLSLL_n1_s 11000001 001 0 .... 0 .. 000 ..... 1100 . @azz_nx1_o1x4 n=2 +UMLSLL_n1_d 11000001 011 0 .... 0 .. 000 ..... 1100 . @azz_nx1_o1x4 n=2 +UMLSLL_n1_s 11000001 001 1 .... 0 .. 000 ..... 1100 . @azz_nx1_o1x4 n=4 +UMLSLL_n1_d 11000001 011 1 .... 0 .. 000 ..... 1100 . @azz_nx1_o1x4 n=4 + +USMLALL_n1_s 11000001 001 0 .... 0 .. 001 ..... 001 .. @azz_nx1_o2x4 n=1 +USMLALL_n1_s 11000001 001 0 .... 0 .. 000 ..... 0010 . @azz_nx1_o1x4 n=2 +USMLALL_n1_s 11000001 001 1 .... 0 .. 000 ..... 0010 . @azz_nx1_o1x4 n=4 + +SUMLALL_n1_s 11000001 001 0 .... 0 .. 000 ..... 1010 . @azz_nx1_o1x4 n=2 +SUMLALL_n1_s 11000001 001 1 .... 0 .. 000 ..... 1010 . @azz_nx1_o1x4 n=4 + ### SME2 Multi-vector Multiple Array Vectors %zn_ax2 6:4 !function=times_2 @@ -XXX,XX +XXX,XX @@ UDOT_nn_4h 11000001 111 ...01 0 .. 101 ...00 10 ... @azz_4x4_o3 UDOT_nn_2h 11000001 111 ....0 0 .. 101 ....0 11 ... @azz_2x2_o3 UDOT_nn_2h 11000001 111 ...01 0 .. 101 ...00 11 ... @azz_4x4_o3 +SMLAL_nn 11000001 111 ....0 0 .. 010 ....0 000 .. @azz_2x2_o2x2 +SMLAL_nn 11000001 111 ...01 0 .. 010 ...00 000 .. @azz_4x4_o2x2 + +SMLSL_nn 11000001 111 ....0 0 .. 010 ....0 010 .. @azz_2x2_o2x2 +SMLSL_nn 11000001 111 ...01 0 .. 010 ...00 010 .. @azz_4x4_o2x2 + +UMLAL_nn 11000001 111 ....0 0 .. 010 ....0 100 .. @azz_2x2_o2x2 +UMLAL_nn 11000001 111 ...01 0 .. 010 ...00 100 .. @azz_4x4_o2x2 + +UMLSL_nn 11000001 111 ....0 0 .. 010 ....0 110 .. @azz_2x2_o2x2 +UMLSL_nn 11000001 111 ...01 0 .. 010 ...00 110 .. @azz_4x4_o2x2 + +@azz_2x2_o1x4 ........ ... ..... . .. ... ..... ... .. \ + &azz_n n=2 rv=%mova_rv zn=%zn_ax2 zm=%zm_ax2 off=%off1_x4 +@azz_4x4_o1x4 ........ ... ..... . .. ... ..... ... .. \ + &azz_n n=4 rv=%mova_rv zn=%zn_ax4 zm=%zm_ax4 off=%off1_x4 + +SMLALL_nn_s 11000001 101 ....0 0 .. 000 ....0 0000 . @azz_2x2_o1x4 +SMLALL_nn_d 11000001 111 ....0 0 .. 000 ....0 0000 . @azz_2x2_o1x4 +SMLALL_nn_s 11000001 101 ...01 0 .. 000 ...00 0000 . @azz_4x4_o1x4 +SMLALL_nn_d 11000001 111 ...01 0 .. 000 ...00 0000 . @azz_4x4_o1x4 + +SMLSLL_nn_s 11000001 101 ....0 0 .. 000 ....0 0100 . @azz_2x2_o1x4 +SMLSLL_nn_d 11000001 111 ....0 0 .. 000 ....0 0100 . @azz_2x2_o1x4 +SMLSLL_nn_s 11000001 101 ...01 0 .. 000 ...00 0100 . @azz_4x4_o1x4 +SMLSLL_nn_d 11000001 111 ...01 0 .. 000 ...00 0100 . @azz_4x4_o1x4 + +UMLALL_nn_s 11000001 101 ....0 0 .. 000 ....0 1000 . @azz_2x2_o1x4 +UMLALL_nn_d 11000001 111 ....0 0 .. 000 ....0 1000 . @azz_2x2_o1x4 +UMLALL_nn_s 11000001 101 ...01 0 .. 000 ...00 1000 . @azz_4x4_o1x4 +UMLALL_nn_d 11000001 111 ...01 0 .. 000 ...00 1000 . @azz_4x4_o1x4 + +UMLSLL_nn_s 11000001 101 ....0 0 .. 000 ....0 1100 . @azz_2x2_o1x4 +UMLSLL_nn_d 11000001 111 ....0 0 .. 000 ....0 1100 . @azz_2x2_o1x4 +UMLSLL_nn_s 11000001 101 ...01 0 .. 000 ...00 1100 . @azz_4x4_o1x4 +UMLSLL_nn_d 11000001 111 ...01 0 .. 000 ...00 1100 . @azz_4x4_o1x4 + +USMLALL_nn_s 11000001 101 ....0 0 .. 000 ....0 0010 . @azz_2x2_o1x4 +USMLALL_nn_s 11000001 101 ...01 0 .. 000 ...00 0010 . @azz_4x4_o1x4 + ### SME2 Multi-vector Indexed &azx_n n off rv zn zm idx @@ -XXX,XX +XXX,XX @@ UVDOT_nx_4h 11000001 1101 .... 1 .. 01 . ...00 11 ... @azx_4x1_i1_o3 SUVDOT_nx_4b 11000001 0101 .... 1 .. 0 .. ...01 11 ... @azx_4x1_i2_o3 USVDOT_nx_4b 11000001 0101 .... 1 .. 0 .. ...01 01 ... @azx_4x1_i2_o3 + +SMLAL_nx 11000001 1100 .... . .. 1 .. ..... 00 ... @azx_1x1_o3x2 +SMLAL_nx 11000001 1101 .... 0 .. 1 .. ....0 00 ... @azx_2x1_o2x2 +SMLAL_nx 11000001 1101 .... 1 .. 1 .. ...00 00 ... @azx_4x1_o2x2 + +SMLSL_nx 11000001 1100 .... . .. 1 .. ..... 01 ... @azx_1x1_o3x2 +SMLSL_nx 11000001 1101 .... 0 .. 1 .. ....0 01 ... @azx_2x1_o2x2 +SMLSL_nx 11000001 1101 .... 1 .. 1 .. ...00 01 ... @azx_4x1_o2x2 + +UMLAL_nx 11000001 1100 .... . .. 1 .. ..... 10 ... @azx_1x1_o3x2 +UMLAL_nx 11000001 1101 .... 0 .. 1 .. ....0 10 ... @azx_2x1_o2x2 +UMLAL_nx 11000001 1101 .... 1 .. 1 .. ...00 10 ... @azx_4x1_o2x2 + +UMLSL_nx 11000001 1100 .... . .. 1 .. ..... 11 ... @azx_1x1_o3x2 +UMLSL_nx 11000001 1101 .... 0 .. 1 .. ....0 11 ... @azx_2x1_o2x2 +UMLSL_nx 11000001 1101 .... 1 .. 1 .. ...00 11 ... @azx_4x1_o2x2 + +%idx4_15_10 15:1 10:3 +%idx4_10_1 10:2 1:2 +%idx3_10_1 10:1 1:2 + +@azx_1x1_i4_o2 ........ .... zm:4 . .. ... zn:5 ... .. \ + &azx_n n=1 rv=%mova_rv off=%off2_x4 idx=%idx4_15_10 +@azx_1x1_i3_o2 ........ .... zm:4 . .. ... zn:5 ... .. \ + &azx_n n=1 rv=%mova_rv off=%off2_x4 idx=%idx3_15_10 +@azx_2x1_i4_o1 ........ .... zm:4 . .. ... ..... ... .. \ + &azx_n n=2 rv=%mova_rv off=%off1_x4 zn=%zn_ax2 idx=%idx4_10_1 +@azx_2x1_i3_o1 ........ .... zm:4 . .. ... ..... ... .. \ + &azx_n n=2 rv=%mova_rv off=%off1_x4 zn=%zn_ax2 idx=%idx3_10_1 +@azx_4x1_i4_o1 ........ .... zm:4 . .. ... ..... ... .. \ + &azx_n n=4 rv=%mova_rv off=%off1_x4 zn=%zn_ax4 idx=%idx4_10_1 +@azx_4x1_i3_o1 ........ .... zm:4 . .. ... ..... ... .. \ + &azx_n n=4 rv=%mova_rv off=%off1_x4 zn=%zn_ax4 idx=%idx3_10_1 + +SMLALL_nx_s 11000001 0000 .... . .. ... ..... 000 .. @azx_1x1_i4_o2 +SMLALL_nx_d 11000001 1000 .... . .. 0.. ..... 000 .. @azx_1x1_i3_o2 +SMLALL_nx_s 11000001 0001 .... 0 .. 0.. ....0 00 ... @azx_2x1_i4_o1 +SMLALL_nx_d 11000001 1001 .... 0 .. 00. ....0 00 ... @azx_2x1_i3_o1 +SMLALL_nx_s 11000001 0001 .... 1 .. 0.. ...00 00 ... @azx_4x1_i4_o1 +SMLALL_nx_d 11000001 1001 .... 1 .. 00. ...00 00 ... @azx_4x1_i3_o1 + +SMLSLL_nx_s 11000001 0000 .... . .. ... ..... 010 .. @azx_1x1_i4_o2 +SMLSLL_nx_d 11000001 1000 .... . .. 0.. ..... 010 .. @azx_1x1_i3_o2 +SMLSLL_nx_s 11000001 0001 .... 0 .. 0.. ....0 01 ... @azx_2x1_i4_o1 +SMLSLL_nx_d 11000001 1001 .... 0 .. 00. ....0 01 ... @azx_2x1_i3_o1 +SMLSLL_nx_s 11000001 0001 .... 1 .. 0.. ...00 01 ... @azx_4x1_i4_o1 +SMLSLL_nx_d 11000001 1001 .... 1 .. 00. ...00 01 ... @azx_4x1_i3_o1 + +UMLALL_nx_s 11000001 0000 .... . .. ... ..... 100 .. @azx_1x1_i4_o2 +UMLALL_nx_d 11000001 1000 .... . .. 0.. ..... 100 .. @azx_1x1_i3_o2 +UMLALL_nx_s 11000001 0001 .... 0 .. 0.. ....0 10 ... @azx_2x1_i4_o1 +UMLALL_nx_d 11000001 1001 .... 0 .. 00. ....0 10 ... @azx_2x1_i3_o1 +UMLALL_nx_s 11000001 0001 .... 1 .. 0.. ...00 10 ... @azx_4x1_i4_o1 +UMLALL_nx_d 11000001 1001 .... 1 .. 00. ...00 10 ... @azx_4x1_i3_o1 + +UMLSLL_nx_s 11000001 0000 .... . .. ... ..... 110 .. @azx_1x1_i4_o2 +UMLSLL_nx_d 11000001 1000 .... . .. 0.. ..... 110 .. @azx_1x1_i3_o2 +UMLSLL_nx_s 11000001 0001 .... 0 .. 0.. ....0 11 ... @azx_2x1_i4_o1 +UMLSLL_nx_d 11000001 1001 .... 0 .. 00. ....0 11 ... @azx_2x1_i3_o1 +UMLSLL_nx_s 11000001 0001 .... 1 .. 0.. ...00 11 ... @azx_4x1_i4_o1 +UMLSLL_nx_d 11000001 1001 .... 1 .. 00. ...00 11 ... @azx_4x1_i3_o1 + +USMLALL_nx_s 11000001 0000 .... . .. ... ..... 001 .. @azx_1x1_i4_o2 +USMLALL_nx_s 11000001 0001 .... 0 .. 0.. ....1 00 ... @azx_2x1_i4_o1 +USMLALL_nx_s 11000001 0001 .... 1 .. 0.. ...01 00 ... @azx_4x1_i4_o1 + +SUMLALL_nx_s 11000001 0000 .... . .. ... ..... 101 .. @azx_1x1_i4_o2 +SUMLALL_nx_s 11000001 0001 .... 0 .. 0.. ....1 10 ... @azx_2x1_i4_o1 +SUMLALL_nx_s 11000001 0001 .... 1 .. 0.. ...01 10 ... @azx_4x1_i4_o1 diff --git a/target/arm/tcg/sme_helper.c b/target/arm/tcg/sme_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme_helper.c +++ b/target/arm/tcg/sme_helper.c @@ -XXX,XX +XXX,XX @@ DO_VDOT_IDX(sme2_svdot_idx_2h, int32_t, int16_t, int16_t, H4, H2) DO_VDOT_IDX(sme2_uvdot_idx_2h, uint32_t, uint16_t, uint16_t, H4, H2) #undef DO_VDOT_IDX + +#define DO_MLALL(NAME, TYPEW, TYPEN, TYPEM, HW, HN, OP) \ +void HELPER(NAME)(void *vd, void *vn, void *vm, void *va, uint32_t desc) \ +{ \ + intptr_t elements = simd_oprsz(desc) / sizeof(TYPEW); \ + intptr_t sel = extract32(desc, SIMD_DATA_SHIFT, 2); \ + TYPEW *d = vd, *a = va; TYPEN *n = vn; TYPEM *m = vm; \ + for (intptr_t i = 0; i < elements; ++i) { \ + TYPEW nn = n[HN(i * 4 + sel)]; \ + TYPEM mm = m[HN(i * 4 + sel)]; \ + d[HW(i)] = a[HW(i)] OP (nn * mm); \ + } \ +} + +DO_MLALL(sme2_smlall_s, int32_t, int8_t, int8_t, H4, H1, +) +DO_MLALL(sme2_smlall_d, int64_t, int16_t, int16_t, H8, H2, +) +DO_MLALL(sme2_smlsll_s, int32_t, int8_t, int8_t, H4, H1, -) +DO_MLALL(sme2_smlsll_d, int64_t, int16_t, int16_t, H8, H2, -) + +DO_MLALL(sme2_umlall_s, uint32_t, uint8_t, uint8_t, H4, H1, +) +DO_MLALL(sme2_umlall_d, uint64_t, uint16_t, uint16_t, H8, H2, +) +DO_MLALL(sme2_umlsll_s, uint32_t, uint8_t, uint8_t, H4, H1, -) +DO_MLALL(sme2_umlsll_d, uint64_t, uint16_t, uint16_t, H8, H2, -) + +DO_MLALL(sme2_usmlall_s, uint32_t, uint8_t, int8_t, H4, H1, +) + +#undef DO_MLALL + +#define DO_MLALL_IDX(NAME, TYPEW, TYPEN, TYPEM, HW, HN, OP) \ +void HELPER(NAME)(void *vd, void *vn, void *vm, void *va, uint32_t desc) \ +{ \ + intptr_t elements = simd_oprsz(desc) / sizeof(TYPEW); \ + intptr_t eltspersegment = 16 / sizeof(TYPEW); \ + intptr_t sel = extract32(desc, SIMD_DATA_SHIFT, 2); \ + intptr_t idx = extract32(desc, SIMD_DATA_SHIFT + 2, 4); \ + TYPEW *d = vd, *a = va; TYPEN *n = vn; TYPEM *m = vm; \ + for (intptr_t i = 0; i < elements; i += eltspersegment) { \ + TYPEW mm = m[HN(i * 4 + idx)]; \ + for (intptr_t j = 0; j < eltspersegment; ++j) { \ + TYPEN nn = n[HN((i + j) * 4 + sel)]; \ + d[HW(i + j)] = a[HW(i + j)] OP (nn * mm); \ + } \ + } \ +} + +DO_MLALL_IDX(sme2_smlall_idx_s, int32_t, int8_t, int8_t, H4, H1, +) +DO_MLALL_IDX(sme2_smlall_idx_d, int64_t, int16_t, int16_t, H8, H2, +) +DO_MLALL_IDX(sme2_smlsll_idx_s, int32_t, int8_t, int8_t, H4, H1, -) +DO_MLALL_IDX(sme2_smlsll_idx_d, int64_t, int16_t, int16_t, H8, H2, -) + +DO_MLALL_IDX(sme2_umlall_idx_s, uint32_t, uint8_t, uint8_t, H4, H1, +) +DO_MLALL_IDX(sme2_umlall_idx_d, uint64_t, uint16_t, uint16_t, H8, H2, +) +DO_MLALL_IDX(sme2_umlsll_idx_s, uint32_t, uint8_t, uint8_t, H4, H1, -) +DO_MLALL_IDX(sme2_umlsll_idx_d, uint64_t, uint16_t, uint16_t, H8, H2, -) + +DO_MLALL_IDX(sme2_usmlall_idx_s, uint32_t, uint8_t, int8_t, H4, H1, +) +DO_MLALL_IDX(sme2_sumlall_idx_s, uint32_t, int8_t, uint8_t, H4, H1, +) + +#undef DO_MLALL_IDX diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(UVDOT_nx_4h, aa64_sme2, do_vdot_nx, a, gen_helper_sme2_uvdot_idx_4h) TRANS_FEAT(SUVDOT_nx_4b, aa64_sme2, do_vdot_nx, a, gen_helper_sme2_suvdot_idx_4b) TRANS_FEAT(USVDOT_nx_4b, aa64_sme2, do_vdot_nx, a, gen_helper_sme2_usvdot_idx_4b) + +static bool do_smlal(DisasContext *s, arg_azz_n *a, bool multi, + gen_helper_gvec_4 *fn) +{ + return do_azz_acc(s, a->n, 2, a->rv, a->off, a->zn, a->zm, + 0, 0, multi, fn); +} + +TRANS_FEAT(SMLAL_n1, aa64_sme2, do_smlal, a, false, gen_helper_sve2_smlal_zzzw_s) +TRANS_FEAT(SMLSL_n1, aa64_sme2, do_smlal, a, false, gen_helper_sve2_smlsl_zzzw_s) +TRANS_FEAT(UMLAL_n1, aa64_sme2, do_smlal, a, false, gen_helper_sve2_umlal_zzzw_s) +TRANS_FEAT(UMLSL_n1, aa64_sme2, do_smlal, a, false, gen_helper_sve2_umlsl_zzzw_s) + +TRANS_FEAT(SMLAL_nn, aa64_sme2, do_smlal, a, true, gen_helper_sve2_smlal_zzzw_s) +TRANS_FEAT(SMLSL_nn, aa64_sme2, do_smlal, a, true, gen_helper_sve2_smlsl_zzzw_s) +TRANS_FEAT(UMLAL_nn, aa64_sme2, do_smlal, a, true, gen_helper_sve2_umlal_zzzw_s) +TRANS_FEAT(UMLSL_nn, aa64_sme2, do_smlal, a, true, gen_helper_sve2_umlsl_zzzw_s) + +static bool do_smlal_nx(DisasContext *s, arg_azx_n *a, + gen_helper_gvec_4 *fn) +{ + return do_azz_acc(s, a->n, 2, a->rv, a->off, a->zn, a->zm, + a->idx << 1, 0, false, fn); +} + +TRANS_FEAT(SMLAL_nx, aa64_sme2, do_smlal_nx, a, gen_helper_sve2_smlal_idx_s) +TRANS_FEAT(SMLSL_nx, aa64_sme2, do_smlal_nx, a, gen_helper_sve2_smlsl_idx_s) +TRANS_FEAT(UMLAL_nx, aa64_sme2, do_smlal_nx, a, gen_helper_sve2_umlal_idx_s) +TRANS_FEAT(UMLSL_nx, aa64_sme2, do_smlal_nx, a, gen_helper_sve2_umlsl_idx_s) + +static bool do_smlall(DisasContext *s, arg_azz_n *a, bool multi, + gen_helper_gvec_4 *fn) +{ + return do_azz_acc(s, a->n, 4, a->rv, a->off, a->zn, a->zm, + 0, 0, multi, fn); +} + +static void gen_helper_sme2_sumlall_s(TCGv_ptr d, TCGv_ptr n, TCGv_ptr m, + TCGv_ptr a, TCGv_i32 desc) +{ + gen_helper_sme2_usmlall_s(d, m, n, a, desc); +} + +TRANS_FEAT(SMLALL_n1_s, aa64_sme2, do_smlall, a, false, gen_helper_sme2_smlall_s) +TRANS_FEAT(SMLSLL_n1_s, aa64_sme2, do_smlall, a, false, gen_helper_sme2_smlsll_s) +TRANS_FEAT(UMLALL_n1_s, aa64_sme2, do_smlall, a, false, gen_helper_sme2_umlall_s) +TRANS_FEAT(UMLSLL_n1_s, aa64_sme2, do_smlall, a, false, gen_helper_sme2_umlsll_s) +TRANS_FEAT(USMLALL_n1_s, aa64_sme2, do_smlall, a, false, gen_helper_sme2_usmlall_s) +TRANS_FEAT(SUMLALL_n1_s, aa64_sme2, do_smlall, a, false, gen_helper_sme2_sumlall_s) + +TRANS_FEAT(SMLALL_n1_d, aa64_sme2_i16i64, do_smlall, a, false, gen_helper_sme2_smlall_d) +TRANS_FEAT(SMLSLL_n1_d, aa64_sme2_i16i64, do_smlall, a, false, gen_helper_sme2_smlsll_d) +TRANS_FEAT(UMLALL_n1_d, aa64_sme2_i16i64, do_smlall, a, false, gen_helper_sme2_umlall_d) +TRANS_FEAT(UMLSLL_n1_d, aa64_sme2_i16i64, do_smlall, a, false, gen_helper_sme2_umlsll_d) + +TRANS_FEAT(SMLALL_nn_s, aa64_sme2, do_smlall, a, true, gen_helper_sme2_smlall_s) +TRANS_FEAT(SMLSLL_nn_s, aa64_sme2, do_smlall, a, true, gen_helper_sme2_smlsll_s) +TRANS_FEAT(UMLALL_nn_s, aa64_sme2, do_smlall, a, true, gen_helper_sme2_umlall_s) +TRANS_FEAT(UMLSLL_nn_s, aa64_sme2, do_smlall, a, true, gen_helper_sme2_umlsll_s) +TRANS_FEAT(USMLALL_nn_s, aa64_sme2, do_smlall, a, true, gen_helper_sme2_usmlall_s) + +TRANS_FEAT(SMLALL_nn_d, aa64_sme2_i16i64, do_smlall, a, true, gen_helper_sme2_smlall_d) +TRANS_FEAT(SMLSLL_nn_d, aa64_sme2_i16i64, do_smlall, a, true, gen_helper_sme2_smlsll_d) +TRANS_FEAT(UMLALL_nn_d, aa64_sme2_i16i64, do_smlall, a, true, gen_helper_sme2_umlall_d) +TRANS_FEAT(UMLSLL_nn_d, aa64_sme2_i16i64, do_smlall, a, true, gen_helper_sme2_umlsll_d) + +static bool do_smlall_nx(DisasContext *s, arg_azx_n *a, + gen_helper_gvec_4 *fn) +{ + return do_azz_acc(s, a->n, 4, a->rv, a->off, a->zn, a->zm, + a->idx << 2, 0, false, fn); +} + +TRANS_FEAT(SMLALL_nx_s, aa64_sme2, do_smlall_nx, a, gen_helper_sme2_smlall_idx_s) +TRANS_FEAT(SMLSLL_nx_s, aa64_sme2, do_smlall_nx, a, gen_helper_sme2_smlsll_idx_s) +TRANS_FEAT(UMLALL_nx_s, aa64_sme2, do_smlall_nx, a, gen_helper_sme2_umlall_idx_s) +TRANS_FEAT(UMLSLL_nx_s, aa64_sme2, do_smlall_nx, a, gen_helper_sme2_umlsll_idx_s) +TRANS_FEAT(USMLALL_nx_s, aa64_sme2, do_smlall_nx, a, gen_helper_sme2_usmlall_idx_s) +TRANS_FEAT(SUMLALL_nx_s, aa64_sme2, do_smlall_nx, a, gen_helper_sme2_sumlall_idx_s) + +TRANS_FEAT(SMLALL_nx_d, aa64_sme2_i16i64, do_smlall_nx, a, gen_helper_sme2_smlall_idx_d) +TRANS_FEAT(SMLSLL_nx_d, aa64_sme2_i16i64, do_smlall_nx, a, gen_helper_sme2_smlsll_idx_d) +TRANS_FEAT(UMLALL_nx_d, aa64_sme2_i16i64, do_smlall_nx, a, gen_helper_sme2_umlall_idx_d) +TRANS_FEAT(UMLSLL_nx_d, aa64_sme2_i16i64, do_smlall_nx, a, gen_helper_sme2_umlsll_idx_d) -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Emphasize the non-fused nature of these multiply-add. Matches other helpers such as gvec_rsqrts_nf_[hs]. Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-48-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper.h | 8 ++++---- target/arm/tcg/translate-neon.c | 4 ++-- target/arm/tcg/vec_helper.c | 8 ++++---- 3 files changed, 10 insertions(+), 10 deletions(-) diff --git a/target/arm/tcg/helper.h b/target/arm/tcg/helper.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper.h +++ b/target/arm/tcg/helper.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_5(gvec_recps_nf_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, DEF_HELPER_FLAGS_5(gvec_rsqrts_nf_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_5(gvec_rsqrts_nf_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) -DEF_HELPER_FLAGS_5(gvec_fmla_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) -DEF_HELPER_FLAGS_5(gvec_fmla_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_5(gvec_fmla_nf_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_5(gvec_fmla_nf_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) -DEF_HELPER_FLAGS_5(gvec_fmls_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) -DEF_HELPER_FLAGS_5(gvec_fmls_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_5(gvec_fmls_nf_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_5(gvec_fmls_nf_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_5(gvec_vfma_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_5(gvec_vfma_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) diff --git a/target/arm/tcg/translate-neon.c b/target/arm/tcg/translate-neon.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-neon.c +++ b/target/arm/tcg/translate-neon.c @@ -XXX,XX +XXX,XX @@ DO_3S_FP_GVEC(VACGE, gen_helper_gvec_facge_s, gen_helper_gvec_facge_h) DO_3S_FP_GVEC(VACGT, gen_helper_gvec_facgt_s, gen_helper_gvec_facgt_h) DO_3S_FP_GVEC(VMAX, gen_helper_gvec_fmax_s, gen_helper_gvec_fmax_h) DO_3S_FP_GVEC(VMIN, gen_helper_gvec_fmin_s, gen_helper_gvec_fmin_h) -DO_3S_FP_GVEC(VMLA, gen_helper_gvec_fmla_s, gen_helper_gvec_fmla_h) -DO_3S_FP_GVEC(VMLS, gen_helper_gvec_fmls_s, gen_helper_gvec_fmls_h) +DO_3S_FP_GVEC(VMLA, gen_helper_gvec_fmla_nf_s, gen_helper_gvec_fmla_nf_h) +DO_3S_FP_GVEC(VMLS, gen_helper_gvec_fmls_nf_s, gen_helper_gvec_fmls_nf_h) DO_3S_FP_GVEC(VFMA, gen_helper_gvec_vfma_s, gen_helper_gvec_vfma_h) DO_3S_FP_GVEC(VFMS, gen_helper_gvec_vfms_s, gen_helper_gvec_vfms_h) DO_3S_FP_GVEC(VRECPS, gen_helper_gvec_recps_nf_s, gen_helper_gvec_recps_nf_h) diff --git a/target/arm/tcg/vec_helper.c b/target/arm/tcg/vec_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/vec_helper.c +++ b/target/arm/tcg/vec_helper.c @@ -XXX,XX +XXX,XX @@ void HELPER(NAME)(void *vd, void *vn, void *vm, \ clear_tail(d, oprsz, simd_maxsz(desc)); \ } -DO_MULADD(gvec_fmla_h, float16_muladd_nf, float16) -DO_MULADD(gvec_fmla_s, float32_muladd_nf, float32) +DO_MULADD(gvec_fmla_nf_h, float16_muladd_nf, float16) +DO_MULADD(gvec_fmla_nf_s, float32_muladd_nf, float32) -DO_MULADD(gvec_fmls_h, float16_mulsub_nf, float16) -DO_MULADD(gvec_fmls_s, float32_mulsub_nf, float32) +DO_MULADD(gvec_fmls_nf_h, float16_mulsub_nf, float16) +DO_MULADD(gvec_fmls_nf_s, float32_mulsub_nf, float32) DO_MULADD(gvec_vfma_h, float16_muladd_f, float16) DO_MULADD(gvec_vfma_s, float32_muladd_f, float32) -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-49-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/sme.decode | 48 +++++++++++++++++ target/arm/tcg/translate-sme.c | 95 ++++++++++++++++++++++++++++++++++ 2 files changed, 143 insertions(+) diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ USMLALL_n1_s 11000001 001 1 .... 0 .. 000 ..... 0010 . @azz_nx1_o1x4 n=4 SUMLALL_n1_s 11000001 001 0 .... 0 .. 000 ..... 1010 . @azz_nx1_o1x4 n=2 SUMLALL_n1_s 11000001 001 1 .... 0 .. 000 ..... 1010 . @azz_nx1_o1x4 n=4 +FMLA_n1_h 11000001 001 0 .... 0 .. 111 ..... 00 ... @azz_nx1_o3 n=2 +FMLA_n1_s 11000001 001 0 .... 0 .. 110 ..... 00 ... @azz_nx1_o3 n=2 +FMLA_n1_d 11000001 011 0 .... 0 .. 110 ..... 00 ... @azz_nx1_o3 n=2 +FMLA_n1_h 11000001 001 1 .... 0 .. 111 ..... 00 ... @azz_nx1_o3 n=4 +FMLA_n1_s 11000001 001 1 .... 0 .. 110 ..... 00 ... @azz_nx1_o3 n=4 +FMLA_n1_d 11000001 011 1 .... 0 .. 110 ..... 00 ... @azz_nx1_o3 n=4 + +FMLS_n1_h 11000001 001 0 .... 0 .. 111 ..... 01 ... @azz_nx1_o3 n=2 +FMLS_n1_s 11000001 001 0 .... 0 .. 110 ..... 01 ... @azz_nx1_o3 n=2 +FMLS_n1_d 11000001 011 0 .... 0 .. 110 ..... 01 ... @azz_nx1_o3 n=2 +FMLS_n1_h 11000001 001 1 .... 0 .. 111 ..... 01 ... @azz_nx1_o3 n=4 +FMLS_n1_s 11000001 001 1 .... 0 .. 110 ..... 01 ... @azz_nx1_o3 n=4 +FMLS_n1_d 11000001 011 1 .... 0 .. 110 ..... 01 ... @azz_nx1_o3 n=4 + ### SME2 Multi-vector Multiple Array Vectors %zn_ax2 6:4 !function=times_2 @@ -XXX,XX +XXX,XX @@ UMLSLL_nn_d 11000001 111 ...01 0 .. 000 ...00 1100 . @azz_4x4_o1x4 USMLALL_nn_s 11000001 101 ....0 0 .. 000 ....0 0010 . @azz_2x2_o1x4 USMLALL_nn_s 11000001 101 ...01 0 .. 000 ...00 0010 . @azz_4x4_o1x4 +FMLA_nn_h 11000001 101 ....0 0 .. 100 ....0 01 ... @azz_2x2_o3 +FMLA_nn_s 11000001 101 ....0 0 .. 110 ....0 00 ... @azz_2x2_o3 +FMLA_nn_d 11000001 111 ....0 0 .. 110 ....0 00 ... @azz_2x2_o3 +FMLA_nn_h 11000001 101 ...01 0 .. 100 ...00 01 ... @azz_4x4_o3 +FMLA_nn_s 11000001 101 ...01 0 .. 110 ...00 00 ... @azz_4x4_o3 +FMLA_nn_d 11000001 111 ...01 0 .. 110 ...00 00 ... @azz_4x4_o3 + +FMLS_nn_h 11000001 101 ....0 0 .. 100 ....0 11 ... @azz_2x2_o3 +FMLS_nn_s 11000001 101 ....0 0 .. 110 ....0 01 ... @azz_2x2_o3 +FMLS_nn_d 11000001 111 ....0 0 .. 110 ....0 01 ... @azz_2x2_o3 +FMLS_nn_h 11000001 101 ...01 0 .. 100 ...00 11 ... @azz_4x4_o3 +FMLS_nn_s 11000001 101 ...01 0 .. 110 ...00 01 ... @azz_4x4_o3 +FMLS_nn_d 11000001 111 ...01 0 .. 110 ...00 01 ... @azz_4x4_o3 + ### SME2 Multi-vector Indexed &azx_n n off rv zn zm idx @@ -XXX,XX +XXX,XX @@ USMLALL_nx_s 11000001 0001 .... 1 .. 0.. ...01 00 ... @azx_4x1_i4_o1 SUMLALL_nx_s 11000001 0000 .... . .. ... ..... 101 .. @azx_1x1_i4_o2 SUMLALL_nx_s 11000001 0001 .... 0 .. 0.. ....1 10 ... @azx_2x1_i4_o1 SUMLALL_nx_s 11000001 0001 .... 1 .. 0.. ...01 10 ... @azx_4x1_i4_o1 + +%idx3_10_3 10:2 3:1 +@azx_2x1_i3_o3 ........ .... zm:4 . .. ... ..... .. off:3 \ + &azx_n n=2 rv=%mova_rv zn=%zn_ax2 idx=%idx3_10_3 +@azx_4x1_i3_o3 ........ .... zm:4 . .. ... ..... .. off:3 \ + &azx_n n=4 rv=%mova_rv zn=%zn_ax4 idx=%idx3_10_3 + +FMLA_nx_h 11000001 0001 .... 0 .. 1.. ....0 0 .... @azx_2x1_i3_o3 +FMLA_nx_s 11000001 0101 .... 0 .. 0.. ....0 00 ... @azx_2x1_i2_o3 +FMLA_nx_d 11000001 1101 .... 0 .. 00. ....0 00 ... @azx_2x1_i1_o3 +FMLA_nx_h 11000001 0001 .... 1 .. 1.. ...00 0 .... @azx_4x1_i3_o3 +FMLA_nx_s 11000001 0101 .... 1 .. 0.. ...00 00 ... @azx_4x1_i2_o3 +FMLA_nx_d 11000001 1101 .... 1 .. 00. ...00 00 ... @azx_4x1_i1_o3 + +FMLS_nx_h 11000001 0001 .... 0 .. 1.. ....0 1 .... @azx_2x1_i3_o3 +FMLS_nx_s 11000001 0101 .... 0 .. 0.. ....0 10 ... @azx_2x1_i2_o3 +FMLS_nx_d 11000001 1101 .... 0 .. 00. ....0 10 ... @azx_2x1_i1_o3 +FMLS_nx_h 11000001 0001 .... 1 .. 1.. ...00 1 .... @azx_4x1_i3_o3 +FMLS_nx_s 11000001 0101 .... 1 .. 0.. ...00 10 ... @azx_4x1_i2_o3 +FMLS_nx_d 11000001 1101 .... 1 .. 00. ...00 10 ... @azx_4x1_i1_o3 diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(SUB_azz_nn_d, aa64_sme2_i16i64, do_azz_nn, a, MO_64, tcg_gen_gvec_sub */ #define FPST_ENV -1 +static bool do_azz_fp(DisasContext *s, int nreg, int nsel, + int rv, int off, int zn, int zm, + int data, int shsel, bool multi, int fpst, + gen_helper_gvec_3_ptr *fn) +{ + if (sme_smza_enabled_check(s)) { + int svl = streaming_vec_reg_size(s); + int vstride = svl / nreg; + TCGv_ptr t_za = get_zarray(s, rv, off, nreg, nsel); + TCGv_ptr t, ptr; + + if (fpst >= 0) { + ptr = fpstatus_ptr(fpst); + } else { + ptr = tcg_env; + } + t = tcg_temp_new_ptr(); + + for (int r = 0; r < nreg; ++r) { + TCGv_ptr t_zn = vec_full_reg_ptr(s, zn); + TCGv_ptr t_zm = vec_full_reg_ptr(s, zm); + + for (int i = 0; i < nsel; ++i) { + int o_za = (r * vstride + i) * sizeof(ARMVectorReg); + int desc = simd_desc(svl, svl, data | (i << shsel)); + + tcg_gen_addi_ptr(t, t_za, o_za); + fn(t, t_zn, t_zm, ptr, tcg_constant_i32(desc)); + } + + /* + * For multiple-and-single vectors, Zn may wrap. + * For multiple vectors, both Zn and Zm are aligned. + */ + zn = (zn + 1) % 32; + zm += multi; + } + } + return true; +} + static bool do_azz_acc_fp(DisasContext *s, int nreg, int nsel, int rv, int off, int zn, int zm, int data, int shsel, bool multi, int fpst, @@ -XXX,XX +XXX,XX @@ static bool do_vdot(DisasContext *s, arg_azx_n *a, gen_helper_gvec_4_ptr *fn) TRANS_FEAT(FVDOT, aa64_sme, do_vdot, a, gen_helper_sme2_fvdot_idx_h) TRANS_FEAT(BFVDOT, aa64_sme, do_vdot, a, gen_helper_sme2_bfvdot_idx) +static bool do_fmla(DisasContext *s, arg_azz_n *a, bool multi, + ARMFPStatusFlavour fpst, gen_helper_gvec_3_ptr *fn) +{ + return do_azz_fp(s, a->n, 1, a->rv, a->off, a->zn, a->zm, + 0, 0, multi, fpst, fn); +} + +TRANS_FEAT(FMLA_n1_h, aa64_sme_f16f16, do_fmla, a, false, FPST_ZA_F16, + gen_helper_gvec_vfma_h) +TRANS_FEAT(FMLS_n1_h, aa64_sme_f16f16, do_fmla, a, false, FPST_ZA_F16, + s->fpcr_ah ? gen_helper_gvec_ah_vfms_h : gen_helper_gvec_vfms_h) +TRANS_FEAT(FMLA_nn_h, aa64_sme_f16f16, do_fmla, a, true, FPST_ZA_F16, + gen_helper_gvec_vfma_h) +TRANS_FEAT(FMLS_nn_h, aa64_sme_f16f16, do_fmla, a, true, FPST_ZA_F16, + s->fpcr_ah ? gen_helper_gvec_ah_vfms_h : gen_helper_gvec_vfms_h) + +TRANS_FEAT(FMLA_n1_s, aa64_sme2, do_fmla, a, false, FPST_ZA, + gen_helper_gvec_vfma_s) +TRANS_FEAT(FMLS_n1_s, aa64_sme2, do_fmla, a, false, FPST_ZA, + s->fpcr_ah ? gen_helper_gvec_ah_vfms_s : gen_helper_gvec_vfms_s) +TRANS_FEAT(FMLA_nn_s, aa64_sme2, do_fmla, a, true, FPST_ZA, + gen_helper_gvec_vfma_s) +TRANS_FEAT(FMLS_nn_s, aa64_sme2, do_fmla, a, true, FPST_ZA, + s->fpcr_ah ? gen_helper_gvec_ah_vfms_s : gen_helper_gvec_vfms_s) + +TRANS_FEAT(FMLA_n1_d, aa64_sme2_f64f64, do_fmla, a, false, FPST_ZA, + gen_helper_gvec_vfma_d) +TRANS_FEAT(FMLS_n1_d, aa64_sme2_f64f64, do_fmla, a, false, FPST_ZA, + s->fpcr_ah ? gen_helper_gvec_ah_vfms_d : gen_helper_gvec_vfms_d) +TRANS_FEAT(FMLA_nn_d, aa64_sme2_f64f64, do_fmla, a, true, FPST_ZA, + gen_helper_gvec_vfma_d) +TRANS_FEAT(FMLS_nn_d, aa64_sme2_f64f64, do_fmla, a, true, FPST_ZA, + s->fpcr_ah ? gen_helper_gvec_ah_vfms_d : gen_helper_gvec_vfms_d) + +static bool do_fmla_nx(DisasContext *s, arg_azx_n *a, + ARMFPStatusFlavour fpst, gen_helper_gvec_4_ptr *fn) +{ + return do_azz_acc_fp(s, a->n, 1, a->rv, a->off, a->zn, a->zm, + a->idx, 0, false, fpst, fn); +} + +TRANS_FEAT(FMLA_nx_h, aa64_sme_f16f16, do_fmla_nx, a, FPST_ZA_F16, + gen_helper_gvec_fmla_idx_h) +TRANS_FEAT(FMLS_nx_h, aa64_sme_f16f16, do_fmla_nx, a, FPST_ZA_F16, + s->fpcr_ah ? gen_helper_gvec_ah_fmls_idx_h : gen_helper_gvec_fmls_idx_h) +TRANS_FEAT(FMLA_nx_s, aa64_sme2, do_fmla_nx, a, FPST_ZA, + gen_helper_gvec_fmla_idx_s) +TRANS_FEAT(FMLS_nx_s, aa64_sme2, do_fmla_nx, a, FPST_ZA, + s->fpcr_ah ? gen_helper_gvec_ah_fmls_idx_s : gen_helper_gvec_fmls_idx_s) +TRANS_FEAT(FMLA_nx_d, aa64_sme2_f64f64, do_fmla_nx, a, FPST_ZA, + gen_helper_gvec_fmla_idx_d) +TRANS_FEAT(FMLS_nx_d, aa64_sme2_f64f64, do_fmla_nx, a, FPST_ZA, + s->fpcr_ah ? gen_helper_gvec_ah_fmls_idx_d : gen_helper_gvec_fmls_idx_d) + /* * Expand array multi-vector single (n1), array multi-vector (nn), * and array multi-vector indexed (nx), for integer accumulate. -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-50-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper.h | 9 +++++++++ target/arm/tcg/sme.decode | 18 ++++++++++++++++++ target/arm/tcg/translate-sme.c | 14 ++++++++++++++ target/arm/tcg/vec_helper.c | 26 ++++++++++++++++++++++++++ 4 files changed, 67 insertions(+) diff --git a/target/arm/tcg/helper.h b/target/arm/tcg/helper.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper.h +++ b/target/arm/tcg/helper.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_5(gvec_fmls_nf_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i DEF_HELPER_FLAGS_5(gvec_vfma_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_5(gvec_vfma_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_5(gvec_vfma_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_5(gvec_bfmla, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_5(gvec_vfms_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_5(gvec_vfms_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_5(gvec_vfms_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_5(gvec_bfmls, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_5(gvec_ah_vfms_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_5(gvec_ah_vfms_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_5(gvec_ah_vfms_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_5(gvec_ah_bfmls, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_5(gvec_ftsmul_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_6(gvec_fmla_idx_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_6(gvec_fmla_idx_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_6(gvec_bfmla_idx, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_6(gvec_fmls_idx_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, fpst, i32) @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_6(gvec_fmls_idx_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_6(gvec_fmls_idx_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_6(gvec_bfmls_idx, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_6(gvec_ah_fmls_idx_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, fpst, i32) @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_6(gvec_ah_fmls_idx_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_6(gvec_ah_fmls_idx_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_6(gvec_ah_bfmls_idx, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_5(gvec_uqadd_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ USMLALL_n1_s 11000001 001 1 .... 0 .. 000 ..... 0010 . @azz_nx1_o1x4 n=4 SUMLALL_n1_s 11000001 001 0 .... 0 .. 000 ..... 1010 . @azz_nx1_o1x4 n=2 SUMLALL_n1_s 11000001 001 1 .... 0 .. 000 ..... 1010 . @azz_nx1_o1x4 n=4 +BFMLA_n1 11000001 011 0 .... 0 .. 111 ..... 00 ... @azz_nx1_o3 n=2 FMLA_n1_h 11000001 001 0 .... 0 .. 111 ..... 00 ... @azz_nx1_o3 n=2 FMLA_n1_s 11000001 001 0 .... 0 .. 110 ..... 00 ... @azz_nx1_o3 n=2 FMLA_n1_d 11000001 011 0 .... 0 .. 110 ..... 00 ... @azz_nx1_o3 n=2 + +BFMLA_n1 11000001 011 1 .... 0 .. 111 ..... 00 ... @azz_nx1_o3 n=4 FMLA_n1_h 11000001 001 1 .... 0 .. 111 ..... 00 ... @azz_nx1_o3 n=4 FMLA_n1_s 11000001 001 1 .... 0 .. 110 ..... 00 ... @azz_nx1_o3 n=4 FMLA_n1_d 11000001 011 1 .... 0 .. 110 ..... 00 ... @azz_nx1_o3 n=4 +BFMLS_n1 11000001 011 0 .... 0 .. 111 ..... 01 ... @azz_nx1_o3 n=2 FMLS_n1_h 11000001 001 0 .... 0 .. 111 ..... 01 ... @azz_nx1_o3 n=2 FMLS_n1_s 11000001 001 0 .... 0 .. 110 ..... 01 ... @azz_nx1_o3 n=2 FMLS_n1_d 11000001 011 0 .... 0 .. 110 ..... 01 ... @azz_nx1_o3 n=2 + +BFMLS_n1 11000001 011 1 .... 0 .. 111 ..... 01 ... @azz_nx1_o3 n=4 FMLS_n1_h 11000001 001 1 .... 0 .. 111 ..... 01 ... @azz_nx1_o3 n=4 FMLS_n1_s 11000001 001 1 .... 0 .. 110 ..... 01 ... @azz_nx1_o3 n=4 FMLS_n1_d 11000001 011 1 .... 0 .. 110 ..... 01 ... @azz_nx1_o3 n=4 @@ -XXX,XX +XXX,XX @@ UMLSLL_nn_d 11000001 111 ...01 0 .. 000 ...00 1100 . @azz_4x4_o1x4 USMLALL_nn_s 11000001 101 ....0 0 .. 000 ....0 0010 . @azz_2x2_o1x4 USMLALL_nn_s 11000001 101 ...01 0 .. 000 ...00 0010 . @azz_4x4_o1x4 +BFMLA_nn 11000001 111 ....0 0 .. 100 ....0 01 ... @azz_2x2_o3 FMLA_nn_h 11000001 101 ....0 0 .. 100 ....0 01 ... @azz_2x2_o3 FMLA_nn_s 11000001 101 ....0 0 .. 110 ....0 00 ... @azz_2x2_o3 FMLA_nn_d 11000001 111 ....0 0 .. 110 ....0 00 ... @azz_2x2_o3 + +BFMLA_nn 11000001 111 ...01 0 .. 100 ...00 01 ... @azz_4x4_o3 FMLA_nn_h 11000001 101 ...01 0 .. 100 ...00 01 ... @azz_4x4_o3 FMLA_nn_s 11000001 101 ...01 0 .. 110 ...00 00 ... @azz_4x4_o3 FMLA_nn_d 11000001 111 ...01 0 .. 110 ...00 00 ... @azz_4x4_o3 +BFMLS_nn 11000001 111 ....0 0 .. 100 ....0 11 ... @azz_2x2_o3 FMLS_nn_h 11000001 101 ....0 0 .. 100 ....0 11 ... @azz_2x2_o3 FMLS_nn_s 11000001 101 ....0 0 .. 110 ....0 01 ... @azz_2x2_o3 FMLS_nn_d 11000001 111 ....0 0 .. 110 ....0 01 ... @azz_2x2_o3 + +BFMLS_nn 11000001 111 ...01 0 .. 100 ...00 11 ... @azz_4x4_o3 FMLS_nn_h 11000001 101 ...01 0 .. 100 ...00 11 ... @azz_4x4_o3 FMLS_nn_s 11000001 101 ...01 0 .. 110 ...00 01 ... @azz_4x4_o3 FMLS_nn_d 11000001 111 ...01 0 .. 110 ...00 01 ... @azz_4x4_o3 @@ -XXX,XX +XXX,XX @@ SUMLALL_nx_s 11000001 0001 .... 1 .. 0.. ...01 10 ... @azx_4x1_i4_o1 @azx_4x1_i3_o3 ........ .... zm:4 . .. ... ..... .. off:3 \ &azx_n n=4 rv=%mova_rv zn=%zn_ax4 idx=%idx3_10_3 +BFMLA_nx 11000001 0001 .... 0 .. 1.. ....1 0 .... @azx_2x1_i3_o3 FMLA_nx_h 11000001 0001 .... 0 .. 1.. ....0 0 .... @azx_2x1_i3_o3 FMLA_nx_s 11000001 0101 .... 0 .. 0.. ....0 00 ... @azx_2x1_i2_o3 FMLA_nx_d 11000001 1101 .... 0 .. 00. ....0 00 ... @azx_2x1_i1_o3 + +BFMLA_nx 11000001 0001 .... 1 .. 1.. ...01 0 .... @azx_4x1_i3_o3 FMLA_nx_h 11000001 0001 .... 1 .. 1.. ...00 0 .... @azx_4x1_i3_o3 FMLA_nx_s 11000001 0101 .... 1 .. 0.. ...00 00 ... @azx_4x1_i2_o3 FMLA_nx_d 11000001 1101 .... 1 .. 00. ...00 00 ... @azx_4x1_i1_o3 +BFMLS_nx 11000001 0001 .... 0 .. 1.. ....1 1 .... @azx_2x1_i3_o3 FMLS_nx_h 11000001 0001 .... 0 .. 1.. ....0 1 .... @azx_2x1_i3_o3 FMLS_nx_s 11000001 0101 .... 0 .. 0.. ....0 10 ... @azx_2x1_i2_o3 FMLS_nx_d 11000001 1101 .... 0 .. 00. ....0 10 ... @azx_2x1_i1_o3 + +BFMLS_nx 11000001 0001 .... 1 .. 1.. ...01 1 .... @azx_4x1_i3_o3 FMLS_nx_h 11000001 0001 .... 1 .. 1.. ...00 1 .... @azx_4x1_i3_o3 FMLS_nx_s 11000001 0101 .... 1 .. 0.. ...00 10 ... @azx_4x1_i2_o3 FMLS_nx_d 11000001 1101 .... 1 .. 00. ...00 10 ... @azx_4x1_i1_o3 diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(FMLA_nn_d, aa64_sme2_f64f64, do_fmla, a, true, FPST_ZA, TRANS_FEAT(FMLS_nn_d, aa64_sme2_f64f64, do_fmla, a, true, FPST_ZA, s->fpcr_ah ? gen_helper_gvec_ah_vfms_d : gen_helper_gvec_vfms_d) +TRANS_FEAT(BFMLA_n1, aa64_sme_b16b16, do_fmla, a, false, FPST_ZA, + gen_helper_gvec_bfmla) +TRANS_FEAT(BFMLS_n1, aa64_sme_b16b16, do_fmla, a, false, FPST_ZA, + s->fpcr_ah ? gen_helper_gvec_ah_bfmls : gen_helper_gvec_bfmls) +TRANS_FEAT(BFMLA_nn, aa64_sme_b16b16, do_fmla, a, true, FPST_ZA, + gen_helper_gvec_bfmla) +TRANS_FEAT(BFMLS_nn, aa64_sme_b16b16, do_fmla, a, true, FPST_ZA, + s->fpcr_ah ? gen_helper_gvec_ah_bfmls : gen_helper_gvec_bfmls) + static bool do_fmla_nx(DisasContext *s, arg_azx_n *a, ARMFPStatusFlavour fpst, gen_helper_gvec_4_ptr *fn) { @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(FMLA_nx_d, aa64_sme2_f64f64, do_fmla_nx, a, FPST_ZA, TRANS_FEAT(FMLS_nx_d, aa64_sme2_f64f64, do_fmla_nx, a, FPST_ZA, s->fpcr_ah ? gen_helper_gvec_ah_fmls_idx_d : gen_helper_gvec_fmls_idx_d) +TRANS_FEAT(BFMLA_nx, aa64_sme_b16b16, do_fmla_nx, a, FPST_ZA, + gen_helper_gvec_bfmla_idx) +TRANS_FEAT(BFMLS_nx, aa64_sme_b16b16, do_fmla_nx, a, FPST_ZA, + s->fpcr_ah ? gen_helper_gvec_ah_bfmls_idx : gen_helper_gvec_bfmls_idx) + /* * Expand array multi-vector single (n1), array multi-vector (nn), * and array multi-vector indexed (nx), for integer accumulate. diff --git a/target/arm/tcg/vec_helper.c b/target/arm/tcg/vec_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/vec_helper.c +++ b/target/arm/tcg/vec_helper.c @@ -XXX,XX +XXX,XX @@ static float16 float16_muladd_f(float16 dest, float16 op1, float16 op2, return float16_muladd(op1, op2, dest, 0, stat); } +static bfloat16 bfloat16_muladd_f(bfloat16 dest, bfloat16 op1, bfloat16 op2, + float_status *stat) +{ + return bfloat16_muladd(op1, op2, dest, 0, stat); +} + static float32 float32_muladd_f(float32 dest, float32 op1, float32 op2, float_status *stat) { @@ -XXX,XX +XXX,XX @@ static float16 float16_mulsub_f(float16 dest, float16 op1, float16 op2, return float16_muladd(float16_chs(op1), op2, dest, 0, stat); } +static bfloat16 bfloat16_mulsub_f(bfloat16 dest, bfloat16 op1, bfloat16 op2, + float_status *stat) +{ + return bfloat16_muladd(bfloat16_chs(op1), op2, dest, 0, stat); +} + static float32 float32_mulsub_f(float32 dest, float32 op1, float32 op2, float_status *stat) { @@ -XXX,XX +XXX,XX @@ static float16 float16_ah_mulsub_f(float16 dest, float16 op1, float16 op2, return float16_muladd(op1, op2, dest, float_muladd_negate_product, stat); } +static bfloat16 bfloat16_ah_mulsub_f(bfloat16 dest, bfloat16 op1, bfloat16 op2, + float_status *stat) +{ + return bfloat16_muladd(op1, op2, dest, float_muladd_negate_product, stat); +} + static float32 float32_ah_mulsub_f(float32 dest, float32 op1, float32 op2, float_status *stat) { @@ -XXX,XX +XXX,XX @@ DO_MULADD(gvec_fmls_nf_s, float32_mulsub_nf, float32) DO_MULADD(gvec_vfma_h, float16_muladd_f, float16) DO_MULADD(gvec_vfma_s, float32_muladd_f, float32) DO_MULADD(gvec_vfma_d, float64_muladd_f, float64) +DO_MULADD(gvec_bfmla, bfloat16_muladd_f, bfloat16) DO_MULADD(gvec_vfms_h, float16_mulsub_f, float16) DO_MULADD(gvec_vfms_s, float32_mulsub_f, float32) DO_MULADD(gvec_vfms_d, float64_mulsub_f, float64) +DO_MULADD(gvec_bfmls, bfloat16_mulsub_f, bfloat16) DO_MULADD(gvec_ah_vfms_h, float16_ah_mulsub_f, float16) DO_MULADD(gvec_ah_vfms_s, float32_ah_mulsub_f, float32) DO_MULADD(gvec_ah_vfms_d, float64_ah_mulsub_f, float64) +DO_MULADD(gvec_ah_bfmls, bfloat16_ah_mulsub_f, bfloat16) + +#undef DO_MULADD /* For the indexed ops, SVE applies the index per 128-bit vector segment. * For AdvSIMD, there is of course only one such vector segment. @@ -XXX,XX +XXX,XX @@ void HELPER(NAME)(void *vd, void *vn, void *vm, void *va, \ DO_FMLA_IDX(gvec_fmla_idx_h, float16, H2, 0, 0) DO_FMLA_IDX(gvec_fmla_idx_s, float32, H4, 0, 0) DO_FMLA_IDX(gvec_fmla_idx_d, float64, H8, 0, 0) +DO_FMLA_IDX(gvec_bfmla_idx, bfloat16, H2, 0, 0) DO_FMLA_IDX(gvec_fmls_idx_h, float16, H2, INT16_MIN, 0) DO_FMLA_IDX(gvec_fmls_idx_s, float32, H4, INT32_MIN, 0) DO_FMLA_IDX(gvec_fmls_idx_d, float64, H8, INT64_MIN, 0) +DO_FMLA_IDX(gvec_bfmls_idx, bfloat16, H2, INT16_MIN, 0) DO_FMLA_IDX(gvec_ah_fmls_idx_h, float16, H2, 0, float_muladd_negate_product) DO_FMLA_IDX(gvec_ah_fmls_idx_s, float32, H4, 0, float_muladd_negate_product) DO_FMLA_IDX(gvec_ah_fmls_idx_d, float64, H8, 0, float_muladd_negate_product) +DO_FMLA_IDX(gvec_ah_bfmls_idx, bfloat16, H2, 0, float_muladd_negate_product) #undef DO_FMLA_IDX -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-51-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper.h | 2 ++ target/arm/tcg/sme.decode | 25 +++++++++++++++++++ target/arm/tcg/translate-sme.c | 44 ++++++++++++++++++++++++++++++++++ target/arm/tcg/vec_helper.c | 2 ++ 4 files changed, 73 insertions(+) diff --git a/target/arm/tcg/helper.h b/target/arm/tcg/helper.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper.h +++ b/target/arm/tcg/helper.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_4(gvec_fclt0_d, TCG_CALL_NO_RWG, void, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_5(gvec_fadd_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_5(gvec_fadd_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_5(gvec_fadd_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_5(gvec_bfadd, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_5(gvec_fsub_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_5(gvec_fsub_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_5(gvec_fsub_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_5(gvec_bfsub, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_5(gvec_fmul_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_5(gvec_fmul_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ FMLS_nn_h 11000001 101 ...01 0 .. 100 ...00 11 ... @azz_4x4_o3 FMLS_nn_s 11000001 101 ...01 0 .. 110 ...00 01 ... @azz_4x4_o3 FMLS_nn_d 11000001 111 ...01 0 .. 110 ...00 01 ... @azz_4x4_o3 +&az_n n off rv zm +@az_2x2_o3 ........ ... ..... . .. ... ..... .. off:3 \ + &az_n n=2 rv=%mova_rv zm=%zn_ax2 +@az_4x4_o3 ........ ... ..... . .. ... ..... .. off:3 \ + &az_n n=4 rv=%mova_rv zm=%zn_ax4 + +FADD_nn_h 11000001 101 00100 0 .. 111 ....0 00 ... @az_2x2_o3 +FADD_nn_s 11000001 101 00000 0 .. 111 ....0 00 ... @az_2x2_o3 +FADD_nn_d 11000001 111 00000 0 .. 111 ....0 00 ... @az_2x2_o3 +FADD_nn_h 11000001 101 00101 0 .. 111 ...00 00 ... @az_4x4_o3 +FADD_nn_s 11000001 101 00001 0 .. 111 ...00 00 ... @az_4x4_o3 +FADD_nn_d 11000001 111 00001 0 .. 111 ...00 00 ... @az_4x4_o3 + +FSUB_nn_h 11000001 101 00100 0 .. 111 ....0 01 ... @az_2x2_o3 +FSUB_nn_s 11000001 101 00000 0 .. 111 ....0 01 ... @az_2x2_o3 +FSUB_nn_d 11000001 111 00000 0 .. 111 ....0 01 ... @az_2x2_o3 +FSUB_nn_h 11000001 101 00101 0 .. 111 ...00 01 ... @az_4x4_o3 +FSUB_nn_s 11000001 101 00001 0 .. 111 ...00 01 ... @az_4x4_o3 +FSUB_nn_d 11000001 111 00001 0 .. 111 ...00 01 ... @az_4x4_o3 + +BFADD_nn 11000001 111 00100 0 .. 111 ....0 00 ... @az_2x2_o3 +BFADD_nn 11000001 111 00101 0 .. 111 ...00 00 ... @az_4x4_o3 +BFSUB_nn 11000001 111 00100 0 .. 111 ....0 01 ... @az_2x2_o3 +BFSUB_nn 11000001 111 00101 0 .. 111 ...00 01 ... @az_4x4_o3 + ### SME2 Multi-vector Indexed &azx_n n off rv zn zm idx diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(BFMLA_nx, aa64_sme_b16b16, do_fmla_nx, a, FPST_ZA, TRANS_FEAT(BFMLS_nx, aa64_sme_b16b16, do_fmla_nx, a, FPST_ZA, s->fpcr_ah ? gen_helper_gvec_ah_bfmls_idx : gen_helper_gvec_bfmls_idx) +static bool do_faddsub(DisasContext *s, arg_az_n *a, ARMFPStatusFlavour fpst, + gen_helper_gvec_3_ptr *fn) +{ + if (sme_smza_enabled_check(s)) { + int svl = streaming_vec_reg_size(s); + int n = a->n; + int zm = a->zm; + int vstride = svl / n; + TCGv_ptr t_za = get_zarray(s, a->rv, a->off, n, 0); + TCGv_ptr ptr = fpstatus_ptr(fpst); + TCGv_ptr t = tcg_temp_new_ptr(); + + for (int r = 0; r < n; ++r) { + TCGv_ptr t_zm = vec_full_reg_ptr(s, zm + r); + int o_za = r * vstride * sizeof(ARMVectorReg); + int desc = simd_desc(svl, svl, 0); + + tcg_gen_addi_ptr(t, t_za, o_za); + fn(t, t, t_zm, ptr, tcg_constant_i32(desc)); + } + } + return true; +} + +TRANS_FEAT(FADD_nn_h, aa64_sme_f16f16, do_faddsub, a, + FPST_ZA_F16, gen_helper_gvec_fadd_h) +TRANS_FEAT(FSUB_nn_h, aa64_sme_f16f16, do_faddsub, a, + FPST_ZA_F16, gen_helper_gvec_fsub_h) + +TRANS_FEAT(FADD_nn_s, aa64_sme2, do_faddsub, a, + FPST_ZA, gen_helper_gvec_fadd_s) +TRANS_FEAT(FSUB_nn_s, aa64_sme2, do_faddsub, a, + FPST_ZA, gen_helper_gvec_fsub_s) + +TRANS_FEAT(FADD_nn_d, aa64_sme2_f64f64, do_faddsub, a, + FPST_ZA, gen_helper_gvec_fadd_d) +TRANS_FEAT(FSUB_nn_d, aa64_sme2_f64f64, do_faddsub, a, + FPST_ZA, gen_helper_gvec_fsub_d) + +TRANS_FEAT(BFADD_nn, aa64_sme_b16b16, do_faddsub, a, + FPST_ZA, gen_helper_gvec_bfadd) +TRANS_FEAT(BFSUB_nn, aa64_sme_b16b16, do_faddsub, a, + FPST_ZA, gen_helper_gvec_bfsub) + /* * Expand array multi-vector single (n1), array multi-vector (nn), * and array multi-vector indexed (nx), for integer accumulate. diff --git a/target/arm/tcg/vec_helper.c b/target/arm/tcg/vec_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/vec_helper.c +++ b/target/arm/tcg/vec_helper.c @@ -XXX,XX +XXX,XX @@ void HELPER(NAME)(void *vd, void *vn, void *vm, \ DO_3OP(gvec_fadd_h, float16_add, float16) DO_3OP(gvec_fadd_s, float32_add, float32) DO_3OP(gvec_fadd_d, float64_add, float64) +DO_3OP(gvec_bfadd, bfloat16_add, bfloat16) DO_3OP(gvec_fsub_h, float16_sub, float16) DO_3OP(gvec_fsub_s, float32_sub, float32) DO_3OP(gvec_fsub_d, float64_sub, float64) +DO_3OP(gvec_bfsub, bfloat16_sub, bfloat16) DO_3OP(gvec_fmul_h, float16_mul, float16) DO_3OP(gvec_fmul_s, float32_mul, float32) -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-52-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/sme.decode | 12 ++++++++++++ target/arm/tcg/translate-sme.c | 28 ++++++++++++++++++++++++++++ 2 files changed, 40 insertions(+) diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ BFMLS_nx 11000001 0001 .... 1 .. 1.. ...01 1 .... @azx_4x1_i3_o3 FMLS_nx_h 11000001 0001 .... 1 .. 1.. ...00 1 .... @azx_4x1_i3_o3 FMLS_nx_s 11000001 0101 .... 1 .. 0.. ...00 10 ... @azx_4x1_i2_o3 FMLS_nx_d 11000001 1101 .... 1 .. 00. ...00 10 ... @azx_4x1_i1_o3 + +### SME2 Add / Sub array accumulators + +ADD_aaz_s 11000001 101 000000 .. 111 ....0 10 ... @az_2x2_o3 +ADD_aaz_s 11000001 101 000010 .. 111 ...00 10 ... @az_4x4_o3 +ADD_aaz_d 11000001 111 000000 .. 111 ....0 10 ... @az_2x2_o3 +ADD_aaz_d 11000001 111 000010 .. 111 ...00 10 ... @az_4x4_o3 + +SUB_aaz_s 11000001 101 000000 .. 111 ....0 11 ... @az_2x2_o3 +SUB_aaz_s 11000001 101 000010 .. 111 ...00 11 ... @az_4x4_o3 +SUB_aaz_d 11000001 111 000000 .. 111 ....0 11 ... @az_2x2_o3 +SUB_aaz_d 11000001 111 000010 .. 111 ...00 11 ... @az_4x4_o3 diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(SUB_azz_nn_s, aa64_sme2, do_azz_nn, a, MO_32, tcg_gen_gvec_sub_var) TRANS_FEAT(ADD_azz_nn_d, aa64_sme2_i16i64, do_azz_nn, a, MO_64, tcg_gen_gvec_add_var) TRANS_FEAT(SUB_azz_nn_d, aa64_sme2_i16i64, do_azz_nn, a, MO_64, tcg_gen_gvec_sub_var) +/* Add/Sub each ZA[d*N] += Z[m*N] */ +static bool do_aaz(DisasContext *s, arg_az_n *a, int esz, GVecGen3FnVar *fn) +{ + TCGv_ptr t_za; + int svl, n; + + if (!sme_smza_enabled_check(s)) { + return true; + } + + n = a->n; + t_za = get_zarray(s, a->rv, a->off, n, 0); + svl = streaming_vec_reg_size(s); + + for (int i = 0; i < n; ++i) { + int o_za = (svl / n * sizeof(ARMVectorReg)) * i; + int o_zm = vec_full_reg_offset(s, a->zm + i); + + fn(esz, t_za, o_za, t_za, o_za, tcg_env, o_zm, svl, svl); + } + return true; +} + +TRANS_FEAT(ADD_aaz_s, aa64_sme2, do_aaz, a, MO_32, tcg_gen_gvec_add_var) +TRANS_FEAT(SUB_aaz_s, aa64_sme2, do_aaz, a, MO_32, tcg_gen_gvec_sub_var) +TRANS_FEAT(ADD_aaz_d, aa64_sme2_i16i64, do_aaz, a, MO_64, tcg_gen_gvec_add_var) +TRANS_FEAT(SUB_aaz_d, aa64_sme2_i16i64, do_aaz, a, MO_64, tcg_gen_gvec_sub_var) + /* * Expand array multi-vector single (n1), array multi-vector (nn), * and array multi-vector indexed (nx), for floating-point accumulate. -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-53-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper-sme.h | 5 +++ target/arm/tcg/vec_internal.h | 2 + target/arm/tcg/sme.decode | 12 ++++++ target/arm/tcg/sme_helper.c | 74 ++++++++++++++++++++++++++++++++++ target/arm/tcg/sve_helper.c | 2 +- target/arm/tcg/translate-sme.c | 25 ++++++++++++ 6 files changed, 119 insertions(+), 1 deletion(-) diff --git a/target/arm/tcg/helper-sme.h b/target/arm/tcg/helper-sme.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-sme.h +++ b/target/arm/tcg/helper-sme.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_5(sme2_umlsll_idx_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, DEF_HELPER_FLAGS_5(sme2_umlsll_idx_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_5(sme2_usmlall_idx_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_5(sme2_sumlall_idx_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) + +DEF_HELPER_FLAGS_4(sme2_bfcvt, TCG_CALL_NO_RWG, void, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_4(sme2_bfcvtn, TCG_CALL_NO_RWG, void, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_4(sme2_fcvt_n, TCG_CALL_NO_RWG, void, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_4(sme2_fcvtn, TCG_CALL_NO_RWG, void, ptr, ptr, fpst, i32) diff --git a/target/arm/tcg/vec_internal.h b/target/arm/tcg/vec_internal.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/vec_internal.h +++ b/target/arm/tcg/vec_internal.h @@ -XXX,XX +XXX,XX @@ static inline float64 float64_maybe_ah_chs(float64 a, bool fpcr_ah) bfloat16 helper_sme2_ah_fmax_b16(bfloat16 a, bfloat16 b, float_status *fpst); bfloat16 helper_sme2_ah_fmin_b16(bfloat16 a, bfloat16 b, float_status *fpst); +float16 sve_f32_to_f16(float32 f, float_status *fpst); + #endif /* TARGET_ARM_VEC_INTERNAL_H */ diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ SUB_aaz_s 11000001 101 000000 .. 111 ....0 11 ... @az_2x2_o3 SUB_aaz_s 11000001 101 000010 .. 111 ...00 11 ... @az_4x4_o3 SUB_aaz_d 11000001 111 000000 .. 111 ....0 11 ... @az_2x2_o3 SUB_aaz_d 11000001 111 000010 .. 111 ...00 11 ... @az_4x4_o3 + +### SME2 Multi-vector SVE Constructive Unary + +&zz_n zd zn n +@zz_1x2 ........ ... ..... ...... ..... zd:5 \ + &zz_n n=1 zn=%zn_ax2 + +BFCVT 11000001 011 00000 111000 ....0 ..... @zz_1x2 +BFCVTN 11000001 011 00000 111000 ....1 ..... @zz_1x2 + +FCVT_n 11000001 001 00000 111000 ....0 ..... @zz_1x2 +FCVTN 11000001 001 00000 111000 ....1 ..... @zz_1x2 diff --git a/target/arm/tcg/sme_helper.c b/target/arm/tcg/sme_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme_helper.c +++ b/target/arm/tcg/sme_helper.c @@ -XXX,XX +XXX,XX @@ DO_MLALL_IDX(sme2_usmlall_idx_s, uint32_t, uint8_t, int8_t, H4, H1, +) DO_MLALL_IDX(sme2_sumlall_idx_s, uint32_t, int8_t, uint8_t, H4, H1, +) #undef DO_MLALL_IDX + +/* Convert and compress */ +void HELPER(sme2_bfcvt)(void *vd, void *vs, float_status *fpst, uint32_t desc) +{ + ARMVectorReg scratch; + size_t oprsz = simd_oprsz(desc); + size_t i, n = oprsz / 4; + float32 *s0 = vs; + float32 *s1 = vs + sizeof(ARMVectorReg); + bfloat16 *d = vd; + + if (vd == s1) { + s1 = memcpy(&scratch, s1, oprsz); + } + + for (i = 0; i < n; ++i) { + d[H2(i)] = float32_to_bfloat16(s0[H4(i)], fpst); + } + for (i = 0; i < n; ++i) { + d[H2(i) + n] = float32_to_bfloat16(s1[H4(i)], fpst); + } +} + +void HELPER(sme2_fcvt_n)(void *vd, void *vs, float_status *fpst, uint32_t desc) +{ + ARMVectorReg scratch; + size_t oprsz = simd_oprsz(desc); + size_t i, n = oprsz / 4; + float32 *s0 = vs; + float32 *s1 = vs + sizeof(ARMVectorReg); + float16 *d = vd; + + if (vd == s1) { + s1 = memcpy(&scratch, s1, oprsz); + } + + for (i = 0; i < n; ++i) { + d[H2(i)] = sve_f32_to_f16(s0[H4(i)], fpst); + } + for (i = 0; i < n; ++i) { + d[H2(i) + n] = sve_f32_to_f16(s1[H4(i)], fpst); + } +} + +/* Convert and interleave */ +void HELPER(sme2_bfcvtn)(void *vd, void *vs, float_status *fpst, uint32_t desc) +{ + size_t i, n = simd_oprsz(desc) / 4; + float32 *s0 = vs; + float32 *s1 = vs + sizeof(ARMVectorReg); + bfloat16 *d = vd; + + for (i = 0; i < n; ++i) { + bfloat16 d0 = float32_to_bfloat16(s0[H4(i)], fpst); + bfloat16 d1 = float32_to_bfloat16(s1[H4(i)], fpst); + d[H2(i * 2 + 0)] = d0; + d[H2(i * 2 + 1)] = d1; + } +} + +void HELPER(sme2_fcvtn)(void *vd, void *vs, float_status *fpst, uint32_t desc) +{ + size_t i, n = simd_oprsz(desc) / 4; + float32 *s0 = vs; + float32 *s1 = vs + sizeof(ARMVectorReg); + bfloat16 *d = vd; + + for (i = 0; i < n; ++i) { + bfloat16 d0 = sve_f32_to_f16(s0[H4(i)], fpst); + bfloat16 d1 = sve_f32_to_f16(s1[H4(i)], fpst); + d[H2(i * 2 + 0)] = d0; + d[H2(i * 2 + 1)] = d1; + } +} diff --git a/target/arm/tcg/sve_helper.c b/target/arm/tcg/sve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve_helper.c +++ b/target/arm/tcg/sve_helper.c @@ -XXX,XX +XXX,XX @@ static inline float64 sve_f16_to_f64(float16 f, float_status *fpst) return ret; } -static inline float16 sve_f32_to_f16(float32 f, float_status *fpst) +float16 sve_f32_to_f16(float32 f, float_status *fpst) { bool save = get_flush_to_zero(fpst); float16 ret; diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(SMLALL_nx_d, aa64_sme2_i16i64, do_smlall_nx, a, gen_helper_sme2_smlal TRANS_FEAT(SMLSLL_nx_d, aa64_sme2_i16i64, do_smlall_nx, a, gen_helper_sme2_smlsll_idx_d) TRANS_FEAT(UMLALL_nx_d, aa64_sme2_i16i64, do_smlall_nx, a, gen_helper_sme2_umlall_idx_d) TRANS_FEAT(UMLSLL_nx_d, aa64_sme2_i16i64, do_smlall_nx, a, gen_helper_sme2_umlsll_idx_d) + +static bool do_zz_fpst(DisasContext *s, arg_zz_n *a, int data, + ARMFPStatusFlavour type, gen_helper_gvec_2_ptr *fn) +{ + if (sme_sm_enabled_check(s)) { + int svl = streaming_vec_reg_size(s); + TCGv_ptr fpst = fpstatus_ptr(type); + + for (int i = 0, n = a->n; i < n; ++i) { + tcg_gen_gvec_2_ptr(vec_full_reg_offset(s, a->zd + i), + vec_full_reg_offset(s, a->zn + i), + fpst, svl, svl, data, fn); + } + } + return true; +} + +TRANS_FEAT(BFCVT, aa64_sme2, do_zz_fpst, a, 0, + FPST_A64, gen_helper_sme2_bfcvt) +TRANS_FEAT(BFCVTN, aa64_sme2, do_zz_fpst, a, 0, + FPST_A64, gen_helper_sme2_bfcvtn) +TRANS_FEAT(FCVT_n, aa64_sme2, do_zz_fpst, a, 0, + FPST_A64, gen_helper_sme2_fcvt_n) +TRANS_FEAT(FCVTN, aa64_sme2, do_zz_fpst, a, 0, + FPST_A64, gen_helper_sme2_fcvtn) -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-54-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper-sme.h | 2 ++ target/arm/tcg/vec_internal.h | 1 + target/arm/tcg/sme.decode | 5 ++++ target/arm/tcg/sme_helper.c | 45 ++++++++++++++++++++++++++++++++++ target/arm/tcg/sve_helper.c | 2 +- target/arm/tcg/translate-sme.c | 5 ++++ 6 files changed, 59 insertions(+), 1 deletion(-) diff --git a/target/arm/tcg/helper-sme.h b/target/arm/tcg/helper-sme.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-sme.h +++ b/target/arm/tcg/helper-sme.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_4(sme2_bfcvt, TCG_CALL_NO_RWG, void, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_4(sme2_bfcvtn, TCG_CALL_NO_RWG, void, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_4(sme2_fcvt_n, TCG_CALL_NO_RWG, void, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_4(sme2_fcvtn, TCG_CALL_NO_RWG, void, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_4(sme2_fcvt_w, TCG_CALL_NO_RWG, void, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_4(sme2_fcvtl, TCG_CALL_NO_RWG, void, ptr, ptr, fpst, i32) diff --git a/target/arm/tcg/vec_internal.h b/target/arm/tcg/vec_internal.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/vec_internal.h +++ b/target/arm/tcg/vec_internal.h @@ -XXX,XX +XXX,XX @@ static inline float64 float64_maybe_ah_chs(float64 a, bool fpcr_ah) bfloat16 helper_sme2_ah_fmax_b16(bfloat16 a, bfloat16 b, float_status *fpst); bfloat16 helper_sme2_ah_fmin_b16(bfloat16 a, bfloat16 b, float_status *fpst); +float32 sve_f16_to_f32(float16 f, float_status *fpst); float16 sve_f32_to_f16(float32 f, float_status *fpst); #endif /* TARGET_ARM_VEC_INTERNAL_H */ diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ SUB_aaz_d 11000001 111 000010 .. 111 ...00 11 ... @az_4x4_o3 &zz_n zd zn n @zz_1x2 ........ ... ..... ...... ..... zd:5 \ &zz_n n=1 zn=%zn_ax2 +@zz_2x1 ........ ... ..... ...... zn:5 ..... \ + &zz_n n=1 zd=%zd_ax2 BFCVT 11000001 011 00000 111000 ....0 ..... @zz_1x2 BFCVTN 11000001 011 00000 111000 ....1 ..... @zz_1x2 FCVT_n 11000001 001 00000 111000 ....0 ..... @zz_1x2 FCVTN 11000001 001 00000 111000 ....1 ..... @zz_1x2 + +FCVT_w 11000001 101 00000 111000 ..... ....0 @zz_2x1 +FCVTL 11000001 101 00000 111000 ..... ....1 @zz_2x1 diff --git a/target/arm/tcg/sme_helper.c b/target/arm/tcg/sme_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme_helper.c +++ b/target/arm/tcg/sme_helper.c @@ -XXX,XX +XXX,XX @@ #include "vec_internal.h" #include "sve_ldst_internal.h" + +static bool vectors_overlap(ARMVectorReg *x, unsigned nx, + ARMVectorReg *y, unsigned ny) +{ + return !(x + nx <= y || y + ny <= x); +} + void helper_set_svcr(CPUARMState *env, uint32_t val, uint32_t mask) { aarch64_set_svcr(env, val, mask); @@ -XXX,XX +XXX,XX @@ void HELPER(sme2_fcvtn)(void *vd, void *vs, float_status *fpst, uint32_t desc) d[H2(i * 2 + 1)] = d1; } } + +/* Expand and convert */ +void HELPER(sme2_fcvt_w)(void *vd, void *vs, float_status *fpst, uint32_t desc) +{ + ARMVectorReg scratch; + size_t oprsz = simd_oprsz(desc); + size_t i, n = oprsz / 4; + float16 *s = vs; + float32 *d0 = vd; + float32 *d1 = vd + sizeof(ARMVectorReg); + + if (vectors_overlap(vd, 1, vs, 2)) { + s = memcpy(&scratch, s, oprsz); + } + + for (i = 0; i < n; ++i) { + d0[H4(i)] = sve_f16_to_f32(s[H2(i)], fpst); + } + for (i = 0; i < n; ++i) { + d1[H4(i)] = sve_f16_to_f32(s[H2(n + i)], fpst); + } +} + +/* Deinterleave and convert. */ +void HELPER(sme2_fcvtl)(void *vd, void *vs, float_status *fpst, uint32_t desc) +{ + size_t i, n = simd_oprsz(desc) / 4; + float16 *s = vs; + float32 *d0 = vd; + float32 *d1 = vd + sizeof(ARMVectorReg); + + for (i = 0; i < n; ++i) { + float32 v0 = sve_f16_to_f32(s[H2(i * 2 + 0)], fpst); + float32 v1 = sve_f16_to_f32(s[H2(i * 2 + 1)], fpst); + d0[H4(i)] = v0; + d1[H4(i)] = v1; + } +} diff --git a/target/arm/tcg/sve_helper.c b/target/arm/tcg/sve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve_helper.c +++ b/target/arm/tcg/sve_helper.c @@ -XXX,XX +XXX,XX @@ void HELPER(NAME)(void *vd, void *vn, void *vg, \ * FZ16. When converting from fp16, this affects flushing input denormals; * when converting to fp16, this affects flushing output denormals. */ -static inline float32 sve_f16_to_f32(float16 f, float_status *fpst) +float32 sve_f16_to_f32(float16 f, float_status *fpst) { bool save = get_flush_inputs_to_zero(fpst); float32 ret; diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(FCVT_n, aa64_sme2, do_zz_fpst, a, 0, FPST_A64, gen_helper_sme2_fcvt_n) TRANS_FEAT(FCVTN, aa64_sme2, do_zz_fpst, a, 0, FPST_A64, gen_helper_sme2_fcvtn) + +TRANS_FEAT(FCVT_w, aa64_sme_f16f16, do_zz_fpst, a, 0, + FPST_A64_F16, gen_helper_sme2_fcvt_w) +TRANS_FEAT(FCVTL, aa64_sme_f16f16, do_zz_fpst, a, 0, + FPST_A64_F16, gen_helper_sme2_fcvtl) -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-55-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/sme.decode | 9 +++++++++ target/arm/tcg/translate-sme.c | 5 +++++ 2 files changed, 14 insertions(+) diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ SUB_aaz_d 11000001 111 000010 .. 111 ...00 11 ... @az_4x4_o3 &zz_n n=1 zn=%zn_ax2 @zz_2x1 ........ ... ..... ...... zn:5 ..... \ &zz_n n=1 zd=%zd_ax2 +@zz_2x2 ........ ... ..... ...... .... . ..... \ + &zz_n n=2 zd=%zd_ax2 zn=%zn_ax2 +@zz_4x4 ........ ... ..... ...... .... . ..... \ + &zz_n n=4 zd=%zd_ax4 zn=%zn_ax4 BFCVT 11000001 011 00000 111000 ....0 ..... @zz_1x2 BFCVTN 11000001 011 00000 111000 ....1 ..... @zz_1x2 @@ -XXX,XX +XXX,XX @@ FCVTN 11000001 001 00000 111000 ....1 ..... @zz_1x2 FCVT_w 11000001 101 00000 111000 ..... ....0 @zz_2x1 FCVTL 11000001 101 00000 111000 ..... ....1 @zz_2x1 + +FCVTZS 11000001 001 00001 111000 ....0 ....0 @zz_2x2 +FCVTZS 11000001 001 10001 111000 ...00 ...00 @zz_4x4 +FCVTZU 11000001 001 00001 111000 ....1 ....0 @zz_2x2 +FCVTZU 11000001 001 10001 111000 ...01 ...00 @zz_4x4 diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(FCVT_w, aa64_sme_f16f16, do_zz_fpst, a, 0, FPST_A64_F16, gen_helper_sme2_fcvt_w) TRANS_FEAT(FCVTL, aa64_sme_f16f16, do_zz_fpst, a, 0, FPST_A64_F16, gen_helper_sme2_fcvtl) + +TRANS_FEAT(FCVTZS, aa64_sme2, do_zz_fpst, a, 0, + FPST_A64, gen_helper_gvec_vcvt_rz_fs) +TRANS_FEAT(FCVTZU, aa64_sme2, do_zz_fpst, a, 0, + FPST_A64, gen_helper_gvec_vcvt_rz_fu) -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-56-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper-sme.h | 2 ++ target/arm/tcg/sme.decode | 5 +++++ target/arm/tcg/sme_helper.c | 22 ++++++++++++++++++++++ target/arm/tcg/translate-sme.c | 5 +++++ 4 files changed, 34 insertions(+) diff --git a/target/arm/tcg/helper-sme.h b/target/arm/tcg/helper-sme.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-sme.h +++ b/target/arm/tcg/helper-sme.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_4(sme2_fcvt_n, TCG_CALL_NO_RWG, void, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_4(sme2_fcvtn, TCG_CALL_NO_RWG, void, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_4(sme2_fcvt_w, TCG_CALL_NO_RWG, void, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_4(sme2_fcvtl, TCG_CALL_NO_RWG, void, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_4(sme2_scvtf, TCG_CALL_NO_RWG, void, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_4(sme2_ucvtf, TCG_CALL_NO_RWG, void, ptr, ptr, fpst, i32) diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ FCVTZS 11000001 001 00001 111000 ....0 ....0 @zz_2x2 FCVTZS 11000001 001 10001 111000 ...00 ...00 @zz_4x4 FCVTZU 11000001 001 00001 111000 ....1 ....0 @zz_2x2 FCVTZU 11000001 001 10001 111000 ...01 ...00 @zz_4x4 + +SCVTF 11000001 001 00010 111000 ....0 ....0 @zz_2x2 +SCVTF 11000001 001 10010 111000 ...00 ...00 @zz_4x4 +UCVTF 11000001 001 00010 111000 ....1 ....0 @zz_2x2 +UCVTF 11000001 001 10010 111000 ...01 ...00 @zz_4x4 diff --git a/target/arm/tcg/sme_helper.c b/target/arm/tcg/sme_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme_helper.c +++ b/target/arm/tcg/sme_helper.c @@ -XXX,XX +XXX,XX @@ void HELPER(sme2_fcvtl)(void *vd, void *vs, float_status *fpst, uint32_t desc) d1[H4(i)] = v1; } } + +void HELPER(sme2_scvtf)(void *vd, void *vs, float_status *fpst, uint32_t desc) +{ + size_t i, n = simd_oprsz(desc) / 4; + int32_t *d = vd; + float32 *s = vs; + + for (i = 0; i < n; ++i) { + d[i] = int32_to_float32(s[i], fpst); + } +} + +void HELPER(sme2_ucvtf)(void *vd, void *vs, float_status *fpst, uint32_t desc) +{ + size_t i, n = simd_oprsz(desc) / 4; + uint32_t *d = vd; + float32 *s = vs; + + for (i = 0; i < n; ++i) { + d[i] = uint32_to_float32(s[i], fpst); + } +} diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(FCVTZS, aa64_sme2, do_zz_fpst, a, 0, FPST_A64, gen_helper_gvec_vcvt_rz_fs) TRANS_FEAT(FCVTZU, aa64_sme2, do_zz_fpst, a, 0, FPST_A64, gen_helper_gvec_vcvt_rz_fu) + +TRANS_FEAT(SCVTF, aa64_sme2, do_zz_fpst, a, 0, + FPST_A64, gen_helper_sme2_scvtf) +TRANS_FEAT(UCVTF, aa64_sme2, do_zz_fpst, a, 0, + FPST_A64, gen_helper_sme2_ucvtf) -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-57-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/sme.decode | 9 +++++++++ target/arm/tcg/translate-sme.c | 9 +++++++++ 2 files changed, 18 insertions(+) diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ SCVTF 11000001 001 00010 111000 ....0 ....0 @zz_2x2 SCVTF 11000001 001 10010 111000 ...00 ...00 @zz_4x4 UCVTF 11000001 001 00010 111000 ....1 ....0 @zz_2x2 UCVTF 11000001 001 10010 111000 ...01 ...00 @zz_4x4 + +FRINTN 11000001 101 01000 111000 ....0 ....0 @zz_2x2 +FRINTN 11000001 101 11000 111000 ...00 ...00 @zz_4x4 +FRINTP 11000001 101 01001 111000 ....0 ....0 @zz_2x2 +FRINTP 11000001 101 11001 111000 ...00 ...00 @zz_4x4 +FRINTM 11000001 101 01010 111000 ....0 ....0 @zz_2x2 +FRINTM 11000001 101 11010 111000 ...00 ...00 @zz_4x4 +FRINTA 11000001 101 01100 111000 ....0 ....0 @zz_2x2 +FRINTA 11000001 101 11100 111000 ...00 ...00 @zz_4x4 diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(SCVTF, aa64_sme2, do_zz_fpst, a, 0, FPST_A64, gen_helper_sme2_scvtf) TRANS_FEAT(UCVTF, aa64_sme2, do_zz_fpst, a, 0, FPST_A64, gen_helper_sme2_ucvtf) + +TRANS_FEAT(FRINTN, aa64_sme2, do_zz_fpst, a, float_round_nearest_even, + FPST_A64, gen_helper_gvec_vrint_rm_s) +TRANS_FEAT(FRINTP, aa64_sme2, do_zz_fpst, a, float_round_up, + FPST_A64, gen_helper_gvec_vrint_rm_s) +TRANS_FEAT(FRINTM, aa64_sme2, do_zz_fpst, a, float_round_down, + FPST_A64, gen_helper_gvec_vrint_rm_s) +TRANS_FEAT(FRINTA, aa64_sme2, do_zz_fpst, a, float_round_ties_away, + FPST_A64, gen_helper_gvec_vrint_rm_s) -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Inputs are a wider type of indeterminate sign. Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-58-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/vec_internal.h | 7 +++++++ 1 file changed, 7 insertions(+) diff --git a/target/arm/tcg/vec_internal.h b/target/arm/tcg/vec_internal.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/vec_internal.h +++ b/target/arm/tcg/vec_internal.h @@ -XXX,XX +XXX,XX @@ int16_t do_sqrdmlah_h(int16_t, int16_t, int16_t, bool, bool, uint32_t *); int32_t do_sqrdmlah_s(int32_t, int32_t, int32_t, bool, bool, uint32_t *); int64_t do_sqrdmlah_d(int64_t, int64_t, int64_t, bool, bool); +#define do_ssat_b(val) MIN(MAX(val, INT8_MIN), INT8_MAX) +#define do_ssat_h(val) MIN(MAX(val, INT16_MIN), INT16_MAX) +#define do_ssat_s(val) MIN(MAX(val, INT32_MIN), INT32_MAX) +#define do_usat_b(val) MIN(MAX(val, 0), UINT8_MAX) +#define do_usat_h(val) MIN(MAX(val, 0), UINT16_MAX) +#define do_usat_s(val) MIN(MAX(val, 0), UINT32_MAX) + /** * bfdotadd: * @sum: addend -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Replace and remove do_sat_bhs. This avoids multiple repetitions of INT*_MIN/MAX. Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-59-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/sve_helper.c | 116 +++++++++++++++--------------------- 1 file changed, 48 insertions(+), 68 deletions(-) diff --git a/target/arm/tcg/sve_helper.c b/target/arm/tcg/sve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve_helper.c +++ b/target/arm/tcg/sve_helper.c @@ -XXX,XX +XXX,XX @@ DO_ZPZZ(sve2_uhsub_zpzz_h, uint16_t, H1_2, DO_HSUB_BHS) DO_ZPZZ(sve2_uhsub_zpzz_s, uint32_t, H1_4, DO_HSUB_BHS) DO_ZPZZ_D(sve2_uhsub_zpzz_d, uint64_t, DO_HSUB_D) -static inline int32_t do_sat_bhs(int64_t val, int64_t min, int64_t max) -{ - return val >= max ? max : val <= min ? min : val; -} - -#define DO_SQADD_B(n, m) do_sat_bhs((int64_t)n + m, INT8_MIN, INT8_MAX) -#define DO_SQADD_H(n, m) do_sat_bhs((int64_t)n + m, INT16_MIN, INT16_MAX) -#define DO_SQADD_S(n, m) do_sat_bhs((int64_t)n + m, INT32_MIN, INT32_MAX) +#define DO_SQADD_B(n, m) do_ssat_b((int64_t)n + m) +#define DO_SQADD_H(n, m) do_ssat_h((int64_t)n + m) +#define DO_SQADD_S(n, m) do_ssat_s((int64_t)n + m) static inline int64_t do_sqadd_d(int64_t n, int64_t m) { @@ -XXX,XX +XXX,XX @@ DO_ZPZZ(sve2_sqadd_zpzz_h, int16_t, H1_2, DO_SQADD_H) DO_ZPZZ(sve2_sqadd_zpzz_s, int32_t, H1_4, DO_SQADD_S) DO_ZPZZ_D(sve2_sqadd_zpzz_d, int64_t, do_sqadd_d) -#define DO_UQADD_B(n, m) do_sat_bhs((int64_t)n + m, 0, UINT8_MAX) -#define DO_UQADD_H(n, m) do_sat_bhs((int64_t)n + m, 0, UINT16_MAX) -#define DO_UQADD_S(n, m) do_sat_bhs((int64_t)n + m, 0, UINT32_MAX) +#define DO_UQADD_B(n, m) do_usat_b((int64_t)n + m) +#define DO_UQADD_H(n, m) do_usat_h((int64_t)n + m) +#define DO_UQADD_S(n, m) do_usat_s((int64_t)n + m) static inline uint64_t do_uqadd_d(uint64_t n, uint64_t m) { @@ -XXX,XX +XXX,XX @@ DO_ZPZZ(sve2_uqadd_zpzz_h, uint16_t, H1_2, DO_UQADD_H) DO_ZPZZ(sve2_uqadd_zpzz_s, uint32_t, H1_4, DO_UQADD_S) DO_ZPZZ_D(sve2_uqadd_zpzz_d, uint64_t, do_uqadd_d) -#define DO_SQSUB_B(n, m) do_sat_bhs((int64_t)n - m, INT8_MIN, INT8_MAX) -#define DO_SQSUB_H(n, m) do_sat_bhs((int64_t)n - m, INT16_MIN, INT16_MAX) -#define DO_SQSUB_S(n, m) do_sat_bhs((int64_t)n - m, INT32_MIN, INT32_MAX) +#define DO_SQSUB_B(n, m) do_ssat_b((int64_t)n - m) +#define DO_SQSUB_H(n, m) do_ssat_h((int64_t)n - m) +#define DO_SQSUB_S(n, m) do_ssat_s((int64_t)n - m) static inline int64_t do_sqsub_d(int64_t n, int64_t m) { @@ -XXX,XX +XXX,XX @@ DO_ZPZZ(sve2_sqsub_zpzz_h, int16_t, H1_2, DO_SQSUB_H) DO_ZPZZ(sve2_sqsub_zpzz_s, int32_t, H1_4, DO_SQSUB_S) DO_ZPZZ_D(sve2_sqsub_zpzz_d, int64_t, do_sqsub_d) -#define DO_UQSUB_B(n, m) do_sat_bhs((int64_t)n - m, 0, UINT8_MAX) -#define DO_UQSUB_H(n, m) do_sat_bhs((int64_t)n - m, 0, UINT16_MAX) -#define DO_UQSUB_S(n, m) do_sat_bhs((int64_t)n - m, 0, UINT32_MAX) +#define DO_UQSUB_B(n, m) do_usat_b((int64_t)n - m) +#define DO_UQSUB_H(n, m) do_usat_h((int64_t)n - m) +#define DO_UQSUB_S(n, m) do_usat_s((int64_t)n - m) static inline uint64_t do_uqsub_d(uint64_t n, uint64_t m) { @@ -XXX,XX +XXX,XX @@ DO_ZPZZ(sve2_uqsub_zpzz_h, uint16_t, H1_2, DO_UQSUB_H) DO_ZPZZ(sve2_uqsub_zpzz_s, uint32_t, H1_4, DO_UQSUB_S) DO_ZPZZ_D(sve2_uqsub_zpzz_d, uint64_t, do_uqsub_d) -#define DO_SUQADD_B(n, m) \ - do_sat_bhs((int64_t)(int8_t)n + m, INT8_MIN, INT8_MAX) -#define DO_SUQADD_H(n, m) \ - do_sat_bhs((int64_t)(int16_t)n + m, INT16_MIN, INT16_MAX) -#define DO_SUQADD_S(n, m) \ - do_sat_bhs((int64_t)(int32_t)n + m, INT32_MIN, INT32_MAX) +#define DO_SUQADD_B(n, m) do_ssat_b((int64_t)(int8_t)n + m) +#define DO_SUQADD_H(n, m) do_ssat_h((int64_t)(int16_t)n + m) +#define DO_SUQADD_S(n, m) do_ssat_s((int64_t)(int32_t)n + m) static inline int64_t do_suqadd_d(int64_t n, uint64_t m) { @@ -XXX,XX +XXX,XX @@ DO_ZPZZ(sve2_suqadd_zpzz_h, uint16_t, H1_2, DO_SUQADD_H) DO_ZPZZ(sve2_suqadd_zpzz_s, uint32_t, H1_4, DO_SUQADD_S) DO_ZPZZ_D(sve2_suqadd_zpzz_d, uint64_t, do_suqadd_d) -#define DO_USQADD_B(n, m) \ - do_sat_bhs((int64_t)n + (int8_t)m, 0, UINT8_MAX) -#define DO_USQADD_H(n, m) \ - do_sat_bhs((int64_t)n + (int16_t)m, 0, UINT16_MAX) -#define DO_USQADD_S(n, m) \ - do_sat_bhs((int64_t)n + (int32_t)m, 0, UINT32_MAX) +#define DO_USQADD_B(n, m) do_usat_b((int64_t)n + (int8_t)m) +#define DO_USQADD_H(n, m) do_usat_h((int64_t)n + (int16_t)m) +#define DO_USQADD_S(n, m) do_usat_s((int64_t)n + (int32_t)m) static inline uint64_t do_usqadd_d(uint64_t n, int64_t m) { @@ -XXX,XX +XXX,XX @@ void HELPER(NAME)(void *vd, void *vn, uint32_t desc) \ } \ } -#define DO_SQXTN_H(n) do_sat_bhs(n, INT8_MIN, INT8_MAX) -#define DO_SQXTN_S(n) do_sat_bhs(n, INT16_MIN, INT16_MAX) -#define DO_SQXTN_D(n) do_sat_bhs(n, INT32_MIN, INT32_MAX) +DO_XTNB(sve2_sqxtnb_h, int16_t, do_ssat_b) +DO_XTNB(sve2_sqxtnb_s, int32_t, do_ssat_h) +DO_XTNB(sve2_sqxtnb_d, int64_t, do_ssat_s) -DO_XTNB(sve2_sqxtnb_h, int16_t, DO_SQXTN_H) -DO_XTNB(sve2_sqxtnb_s, int32_t, DO_SQXTN_S) -DO_XTNB(sve2_sqxtnb_d, int64_t, DO_SQXTN_D) +DO_XTNT(sve2_sqxtnt_h, int16_t, int8_t, H1, do_ssat_b) +DO_XTNT(sve2_sqxtnt_s, int32_t, int16_t, H1_2, do_ssat_h) +DO_XTNT(sve2_sqxtnt_d, int64_t, int32_t, H1_4, do_ssat_s) -DO_XTNT(sve2_sqxtnt_h, int16_t, int8_t, H1, DO_SQXTN_H) -DO_XTNT(sve2_sqxtnt_s, int32_t, int16_t, H1_2, DO_SQXTN_S) -DO_XTNT(sve2_sqxtnt_d, int64_t, int32_t, H1_4, DO_SQXTN_D) +DO_XTNB(sve2_uqxtnb_h, uint16_t, do_usat_b) +DO_XTNB(sve2_uqxtnb_s, uint32_t, do_usat_h) +DO_XTNB(sve2_uqxtnb_d, uint64_t, do_usat_s) -#define DO_UQXTN_H(n) do_sat_bhs(n, 0, UINT8_MAX) -#define DO_UQXTN_S(n) do_sat_bhs(n, 0, UINT16_MAX) -#define DO_UQXTN_D(n) do_sat_bhs(n, 0, UINT32_MAX) +DO_XTNT(sve2_uqxtnt_h, uint16_t, uint8_t, H1, do_usat_b) +DO_XTNT(sve2_uqxtnt_s, uint32_t, uint16_t, H1_2, do_usat_h) +DO_XTNT(sve2_uqxtnt_d, uint64_t, uint32_t, H1_4, do_usat_s) -DO_XTNB(sve2_uqxtnb_h, uint16_t, DO_UQXTN_H) -DO_XTNB(sve2_uqxtnb_s, uint32_t, DO_UQXTN_S) -DO_XTNB(sve2_uqxtnb_d, uint64_t, DO_UQXTN_D) +DO_XTNB(sve2_sqxtunb_h, int16_t, do_usat_b) +DO_XTNB(sve2_sqxtunb_s, int32_t, do_usat_h) +DO_XTNB(sve2_sqxtunb_d, int64_t, do_usat_s) -DO_XTNT(sve2_uqxtnt_h, uint16_t, uint8_t, H1, DO_UQXTN_H) -DO_XTNT(sve2_uqxtnt_s, uint32_t, uint16_t, H1_2, DO_UQXTN_S) -DO_XTNT(sve2_uqxtnt_d, uint64_t, uint32_t, H1_4, DO_UQXTN_D) - -DO_XTNB(sve2_sqxtunb_h, int16_t, DO_UQXTN_H) -DO_XTNB(sve2_sqxtunb_s, int32_t, DO_UQXTN_S) -DO_XTNB(sve2_sqxtunb_d, int64_t, DO_UQXTN_D) - -DO_XTNT(sve2_sqxtunt_h, int16_t, int8_t, H1, DO_UQXTN_H) -DO_XTNT(sve2_sqxtunt_s, int32_t, int16_t, H1_2, DO_UQXTN_S) -DO_XTNT(sve2_sqxtunt_d, int64_t, int32_t, H1_4, DO_UQXTN_D) +DO_XTNT(sve2_sqxtunt_h, int16_t, int8_t, H1, do_usat_b) +DO_XTNT(sve2_sqxtunt_s, int32_t, int16_t, H1_2, do_usat_h) +DO_XTNT(sve2_sqxtunt_d, int64_t, int32_t, H1_4, do_usat_s) #undef DO_XTNB #undef DO_XTNT @@ -XXX,XX +XXX,XX @@ DO_SHRNT(sve2_rshrnt_h, uint16_t, uint8_t, H1_2, H1, do_urshr) DO_SHRNT(sve2_rshrnt_s, uint32_t, uint16_t, H1_4, H1_2, do_urshr) DO_SHRNT(sve2_rshrnt_d, uint64_t, uint32_t, H1_8, H1_4, do_urshr) -#define DO_SQSHRUN_H(x, sh) do_sat_bhs((int64_t)(x) >> sh, 0, UINT8_MAX) -#define DO_SQSHRUN_S(x, sh) do_sat_bhs((int64_t)(x) >> sh, 0, UINT16_MAX) -#define DO_SQSHRUN_D(x, sh) \ - do_sat_bhs((int64_t)(x) >> (sh < 64 ? sh : 63), 0, UINT32_MAX) +#define DO_SQSHRUN_H(x, sh) do_usat_b((int64_t)(x) >> sh) +#define DO_SQSHRUN_S(x, sh) do_usat_h((int64_t)(x) >> sh) +#define DO_SQSHRUN_D(x, sh) do_usat_s((int64_t)(x) >> (sh < 64 ? sh : 63)) DO_SHRNB(sve2_sqshrunb_h, int16_t, uint8_t, DO_SQSHRUN_H) DO_SHRNB(sve2_sqshrunb_s, int32_t, uint16_t, DO_SQSHRUN_S) @@ -XXX,XX +XXX,XX @@ DO_SHRNT(sve2_sqshrunt_h, int16_t, uint8_t, H1_2, H1, DO_SQSHRUN_H) DO_SHRNT(sve2_sqshrunt_s, int32_t, uint16_t, H1_4, H1_2, DO_SQSHRUN_S) DO_SHRNT(sve2_sqshrunt_d, int64_t, uint32_t, H1_8, H1_4, DO_SQSHRUN_D) -#define DO_SQRSHRUN_H(x, sh) do_sat_bhs(do_srshr(x, sh), 0, UINT8_MAX) -#define DO_SQRSHRUN_S(x, sh) do_sat_bhs(do_srshr(x, sh), 0, UINT16_MAX) -#define DO_SQRSHRUN_D(x, sh) do_sat_bhs(do_srshr(x, sh), 0, UINT32_MAX) +#define DO_SQRSHRUN_H(x, sh) do_usat_b(do_srshr(x, sh)) +#define DO_SQRSHRUN_S(x, sh) do_usat_h(do_srshr(x, sh)) +#define DO_SQRSHRUN_D(x, sh) do_usat_s(do_srshr(x, sh)) DO_SHRNB(sve2_sqrshrunb_h, int16_t, uint8_t, DO_SQRSHRUN_H) DO_SHRNB(sve2_sqrshrunb_s, int32_t, uint16_t, DO_SQRSHRUN_S) @@ -XXX,XX +XXX,XX @@ DO_SHRNT(sve2_sqrshrunt_h, int16_t, uint8_t, H1_2, H1, DO_SQRSHRUN_H) DO_SHRNT(sve2_sqrshrunt_s, int32_t, uint16_t, H1_4, H1_2, DO_SQRSHRUN_S) DO_SHRNT(sve2_sqrshrunt_d, int64_t, uint32_t, H1_8, H1_4, DO_SQRSHRUN_D) -#define DO_SQSHRN_H(x, sh) do_sat_bhs(x >> sh, INT8_MIN, INT8_MAX) -#define DO_SQSHRN_S(x, sh) do_sat_bhs(x >> sh, INT16_MIN, INT16_MAX) -#define DO_SQSHRN_D(x, sh) do_sat_bhs(x >> sh, INT32_MIN, INT32_MAX) +#define DO_SQSHRN_H(x, sh) do_ssat_b(x >> sh) +#define DO_SQSHRN_S(x, sh) do_ssat_h(x >> sh) +#define DO_SQSHRN_D(x, sh) do_ssat_s(x >> sh) DO_SHRNB(sve2_sqshrnb_h, int16_t, uint8_t, DO_SQSHRN_H) DO_SHRNB(sve2_sqshrnb_s, int32_t, uint16_t, DO_SQSHRN_S) @@ -XXX,XX +XXX,XX @@ DO_SHRNT(sve2_sqshrnt_h, int16_t, uint8_t, H1_2, H1, DO_SQSHRN_H) DO_SHRNT(sve2_sqshrnt_s, int32_t, uint16_t, H1_4, H1_2, DO_SQSHRN_S) DO_SHRNT(sve2_sqshrnt_d, int64_t, uint32_t, H1_8, H1_4, DO_SQSHRN_D) -#define DO_SQRSHRN_H(x, sh) do_sat_bhs(do_srshr(x, sh), INT8_MIN, INT8_MAX) -#define DO_SQRSHRN_S(x, sh) do_sat_bhs(do_srshr(x, sh), INT16_MIN, INT16_MAX) -#define DO_SQRSHRN_D(x, sh) do_sat_bhs(do_srshr(x, sh), INT32_MIN, INT32_MAX) +#define DO_SQRSHRN_H(x, sh) do_ssat_b(do_srshr(x, sh)) +#define DO_SQRSHRN_S(x, sh) do_ssat_h(do_srshr(x, sh)) +#define DO_SQRSHRN_D(x, sh) do_ssat_s(do_srshr(x, sh)) DO_SHRNB(sve2_sqrshrnb_h, int16_t, uint8_t, DO_SQRSHRN_H) DO_SHRNB(sve2_sqrshrnb_s, int32_t, uint16_t, DO_SQRSHRN_S) -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Message-id: 20250704142112.1018902-60-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper-sme.h | 20 ++++++ target/arm/tcg/sme.decode | 22 +++++++ target/arm/tcg/sme_helper.c | 116 +++++++++++++++++++++++++++++++++ target/arm/tcg/translate-sme.c | 35 ++++++++++ 4 files changed, 193 insertions(+) diff --git a/target/arm/tcg/helper-sme.h b/target/arm/tcg/helper-sme.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-sme.h +++ b/target/arm/tcg/helper-sme.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_4(sme2_fcvt_w, TCG_CALL_NO_RWG, void, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_4(sme2_fcvtl, TCG_CALL_NO_RWG, void, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_4(sme2_scvtf, TCG_CALL_NO_RWG, void, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_4(sme2_ucvtf, TCG_CALL_NO_RWG, void, ptr, ptr, fpst, i32) + +DEF_HELPER_FLAGS_3(sme2_sqcvt_sb, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_uqcvt_sb, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_sqcvtu_sb, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_sqcvt_sh, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_uqcvt_sh, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_sqcvtu_sh, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_sqcvt_dh, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_uqcvt_dh, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_sqcvtu_dh, TCG_CALL_NO_RWG, void, ptr, ptr, i32) + +DEF_HELPER_FLAGS_3(sme2_sqcvtn_sb, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_uqcvtn_sb, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_sqcvtun_sb, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_sqcvtn_sh, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_uqcvtn_sh, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_sqcvtun_sh, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_sqcvtn_dh, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_uqcvtn_dh, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_sqcvtun_dh, TCG_CALL_NO_RWG, void, ptr, ptr, i32) diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ SUB_aaz_d 11000001 111 000010 .. 111 ...00 11 ... @az_4x4_o3 &zz_n zd zn n @zz_1x2 ........ ... ..... ...... ..... zd:5 \ &zz_n n=1 zn=%zn_ax2 +@zz_1x4 ........ ... ..... ...... ..... zd:5 \ + &zz_n n=1 zn=%zn_ax4 @zz_2x1 ........ ... ..... ...... zn:5 ..... \ &zz_n n=1 zd=%zd_ax2 @zz_2x2 ........ ... ..... ...... .... . ..... \ @@ -XXX,XX +XXX,XX @@ FRINTM 11000001 101 01010 111000 ....0 ....0 @zz_2x2 FRINTM 11000001 101 11010 111000 ...00 ...00 @zz_4x4 FRINTA 11000001 101 01100 111000 ....0 ....0 @zz_2x2 FRINTA 11000001 101 11100 111000 ...00 ...00 @zz_4x4 + +SQCVT_sh 11000001 001 00011 111000 ....0 ..... @zz_1x2 +UQCVT_sh 11000001 001 00011 111000 ....1 ..... @zz_1x2 +SQCVTU_sh 11000001 011 00011 111000 ....0 ..... @zz_1x2 + +SQCVT_sb 11000001 001 10011 111000 ...00 ..... @zz_1x4 +UQCVT_sb 11000001 001 10011 111000 ...01 ..... @zz_1x4 +SQCVTU_sb 11000001 011 10011 111000 ...00 ..... @zz_1x4 + +SQCVT_dh 11000001 101 10011 111000 ...00 ..... @zz_1x4 +UQCVT_dh 11000001 101 10011 111000 ...01 ..... @zz_1x4 +SQCVTU_dh 11000001 111 10011 111000 ...00 ..... @zz_1x4 + +SQCVTN_sb 11000001 001 10011 111000 ...10 ..... @zz_1x4 +UQCVTN_sb 11000001 001 10011 111000 ...11 ..... @zz_1x4 +SQCVTUN_sb 11000001 011 10011 111000 ...10 ..... @zz_1x4 + +SQCVTN_dh 11000001 101 10011 111000 ...10 ..... @zz_1x4 +UQCVTN_dh 11000001 101 10011 111000 ...11 ..... @zz_1x4 +SQCVTUN_dh 11000001 111 10011 111000 ...10 ..... @zz_1x4 diff --git a/target/arm/tcg/sme_helper.c b/target/arm/tcg/sme_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme_helper.c +++ b/target/arm/tcg/sme_helper.c @@ -XXX,XX +XXX,XX @@ void HELPER(sme2_fcvt_n)(void *vd, void *vs, float_status *fpst, uint32_t desc) } } +#define SQCVT2(NAME, TW, TN, HW, HN, SAT) \ +void HELPER(NAME)(void *vd, void *vs, uint32_t desc) \ +{ \ + ARMVectorReg scratch; \ + size_t oprsz = simd_oprsz(desc), n = oprsz / sizeof(TW); \ + TW *s0 = vs, *s1 = vs + sizeof(ARMVectorReg); \ + TN *d = vd; \ + if (vectors_overlap(vd, 1, vs, 2)) { \ + d = (TN *)&scratch; \ + } \ + for (size_t i = 0; i < n; ++i) { \ + d[HN(i)] = SAT(s0[HW(i)]); \ + d[HN(i + n)] = SAT(s1[HW(i)]); \ + } \ + if (d != vd) { \ + memcpy(vd, d, oprsz); \ + } \ +} + +SQCVT2(sme2_sqcvt_sh, int32_t, int16_t, H4, H2, do_ssat_h) +SQCVT2(sme2_uqcvt_sh, uint32_t, uint16_t, H4, H2, do_usat_h) +SQCVT2(sme2_sqcvtu_sh, int32_t, uint16_t, H4, H2, do_usat_h) + +#undef SQCVT2 + +#define SQCVT4(NAME, TW, TN, HW, HN, SAT) \ +void HELPER(NAME)(void *vd, void *vs, uint32_t desc) \ +{ \ + ARMVectorReg scratch; \ + size_t oprsz = simd_oprsz(desc), n = oprsz / sizeof(TW); \ + TW *s0 = vs, *s1 = vs + sizeof(ARMVectorReg); \ + TW *s2 = vs + 2 * sizeof(ARMVectorReg); \ + TW *s3 = vs + 3 * sizeof(ARMVectorReg); \ + TN *d = vd; \ + if (vectors_overlap(vd, 1, vs, 4)) { \ + d = (TN *)&scratch; \ + } \ + for (size_t i = 0; i < n; ++i) { \ + d[HN(i)] = SAT(s0[HW(i)]); \ + d[HN(i + n)] = SAT(s1[HW(i)]); \ + d[HN(i + 2 * n)] = SAT(s2[HW(i)]); \ + d[HN(i + 3 * n)] = SAT(s3[HW(i)]); \ + } \ + if (d != vd) { \ + memcpy(vd, d, oprsz); \ + } \ +} + +SQCVT4(sme2_sqcvt_sb, int32_t, int8_t, H4, H2, do_ssat_b) +SQCVT4(sme2_uqcvt_sb, uint32_t, uint8_t, H4, H2, do_usat_b) +SQCVT4(sme2_sqcvtu_sb, int32_t, uint8_t, H4, H2, do_usat_b) + +SQCVT4(sme2_sqcvt_dh, int64_t, int16_t, H8, H2, do_ssat_h) +SQCVT4(sme2_uqcvt_dh, uint64_t, uint16_t, H8, H2, do_usat_h) +SQCVT4(sme2_sqcvtu_dh, int64_t, uint16_t, H8, H2, do_usat_h) + +#undef SQCVT4 + /* Convert and interleave */ void HELPER(sme2_bfcvtn)(void *vd, void *vs, float_status *fpst, uint32_t desc) { @@ -XXX,XX +XXX,XX @@ void HELPER(sme2_fcvtn)(void *vd, void *vs, float_status *fpst, uint32_t desc) } } +#define SQCVTN2(NAME, TW, TN, HW, HN, SAT) \ +void HELPER(NAME)(void *vd, void *vs, uint32_t desc) \ +{ \ + ARMVectorReg scratch; \ + size_t oprsz = simd_oprsz(desc), n = oprsz / sizeof(TW); \ + TW *s0 = vs, *s1 = vs + sizeof(ARMVectorReg); \ + TN *d = vd; \ + if (vectors_overlap(vd, 1, vs, 2)) { \ + d = (TN *)&scratch; \ + } \ + for (size_t i = 0; i < n; ++i) { \ + d[HN(2 * i + 0)] = SAT(s0[HW(i)]); \ + d[HN(2 * i + 1)] = SAT(s1[HW(i)]); \ + } \ + if (d != vd) { \ + memcpy(vd, d, oprsz); \ + } \ +} + +SQCVTN2(sme2_sqcvtn_sh, int32_t, int16_t, H4, H2, do_ssat_h) +SQCVTN2(sme2_uqcvtn_sh, uint32_t, uint16_t, H4, H2, do_usat_h) +SQCVTN2(sme2_sqcvtun_sh, int32_t, uint16_t, H4, H2, do_usat_h) + +#undef SQCVTN2 + +#define SQCVTN4(NAME, TW, TN, HW, HN, SAT) \ +void HELPER(NAME)(void *vd, void *vs, uint32_t desc) \ +{ \ + ARMVectorReg scratch; \ + size_t oprsz = simd_oprsz(desc), n = oprsz / sizeof(TW); \ + TW *s0 = vs, *s1 = vs + sizeof(ARMVectorReg); \ + TW *s2 = vs + 2 * sizeof(ARMVectorReg); \ + TW *s3 = vs + 3 * sizeof(ARMVectorReg); \ + TN *d = vd; \ + if (vectors_overlap(vd, 1, vs, 4)) { \ + d = (TN *)&scratch; \ + } \ + for (size_t i = 0; i < n; ++i) { \ + d[HN(4 * i + 0)] = SAT(s0[HW(i)]); \ + d[HN(4 * i + 1)] = SAT(s1[HW(i)]); \ + d[HN(4 * i + 2)] = SAT(s2[HW(i)]); \ + d[HN(4 * i + 3)] = SAT(s3[HW(i)]); \ + } \ + if (d != vd) { \ + memcpy(vd, d, oprsz); \ + } \ +} + +SQCVTN4(sme2_sqcvtn_sb, int32_t, int8_t, H4, H1, do_ssat_b) +SQCVTN4(sme2_uqcvtn_sb, uint32_t, uint8_t, H4, H1, do_usat_b) +SQCVTN4(sme2_sqcvtun_sb, int32_t, uint8_t, H4, H1, do_usat_b) + +SQCVTN4(sme2_sqcvtn_dh, int64_t, int16_t, H8, H2, do_ssat_h) +SQCVTN4(sme2_uqcvtn_dh, uint64_t, uint16_t, H8, H2, do_usat_h) +SQCVTN4(sme2_sqcvtun_dh, int64_t, uint16_t, H8, H2, do_usat_h) + +#undef SQCVTN4 + /* Expand and convert */ void HELPER(sme2_fcvt_w)(void *vd, void *vs, float_status *fpst, uint32_t desc) { diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(FRINTM, aa64_sme2, do_zz_fpst, a, float_round_down, FPST_A64, gen_helper_gvec_vrint_rm_s) TRANS_FEAT(FRINTA, aa64_sme2, do_zz_fpst, a, float_round_ties_away, FPST_A64, gen_helper_gvec_vrint_rm_s) + +static bool do_zz(DisasContext *s, arg_zz_n *a, int data, + gen_helper_gvec_2 *fn) +{ + if (sme_sm_enabled_check(s)) { + int svl = streaming_vec_reg_size(s); + + for (int i = 0, n = a->n; i < n; ++i) { + tcg_gen_gvec_2_ool(vec_full_reg_offset(s, a->zd + i), + vec_full_reg_offset(s, a->zn + i), + svl, svl, data, fn); + } + } + return true; +} + +TRANS_FEAT(SQCVT_sh, aa64_sme2, do_zz, a, 0, gen_helper_sme2_sqcvt_sh) +TRANS_FEAT(UQCVT_sh, aa64_sme2, do_zz, a, 0, gen_helper_sme2_uqcvt_sh) +TRANS_FEAT(SQCVTU_sh, aa64_sme2, do_zz, a, 0, gen_helper_sme2_sqcvtu_sh) + +TRANS_FEAT(SQCVT_sb, aa64_sme2, do_zz, a, 0, gen_helper_sme2_sqcvt_sb) +TRANS_FEAT(UQCVT_sb, aa64_sme2, do_zz, a, 0, gen_helper_sme2_uqcvt_sb) +TRANS_FEAT(SQCVTU_sb, aa64_sme2, do_zz, a, 0, gen_helper_sme2_sqcvtu_sb) + +TRANS_FEAT(SQCVT_dh, aa64_sme2, do_zz, a, 0, gen_helper_sme2_sqcvt_dh) +TRANS_FEAT(UQCVT_dh, aa64_sme2, do_zz, a, 0, gen_helper_sme2_uqcvt_dh) +TRANS_FEAT(SQCVTU_dh, aa64_sme2, do_zz, a, 0, gen_helper_sme2_sqcvtu_dh) + +TRANS_FEAT(SQCVTN_sb, aa64_sme2, do_zz, a, 0, gen_helper_sme2_sqcvtn_sb) +TRANS_FEAT(UQCVTN_sb, aa64_sme2, do_zz, a, 0, gen_helper_sme2_uqcvtn_sb) +TRANS_FEAT(SQCVTUN_sb, aa64_sme2, do_zz, a, 0, gen_helper_sme2_sqcvtun_sb) + +TRANS_FEAT(SQCVTN_dh, aa64_sme2, do_zz, a, 0, gen_helper_sme2_sqcvtn_dh) +TRANS_FEAT(UQCVTN_dh, aa64_sme2, do_zz, a, 0, gen_helper_sme2_uqcvtn_dh) +TRANS_FEAT(SQCVTUN_dh, aa64_sme2, do_zz, a, 0, gen_helper_sme2_sqcvtun_dh) -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-61-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/sve.decode | 20 ++++++++++++++++---- target/arm/tcg/translate-sve.c | 7 +++++++ 2 files changed, 23 insertions(+), 4 deletions(-) diff --git a/target/arm/tcg/sve.decode b/target/arm/tcg/sve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve.decode +++ b/target/arm/tcg/sve.decode @@ -XXX,XX +XXX,XX @@ # as propagated via the MOVPRFX instruction. %reg_movprfx 0:5 +%rn_ax2 6:4 !function=times_2 + ########################################################################### # Named attribute sets. These are used to make nice(er) names # when creating helpers common to those for the individual @@ -XXX,XX +XXX,XX @@ # Two operand @pd_pn ........ esz:2 .. .... ....... rn:4 . rd:4 &rr_esz @rd_rn ........ esz:2 ...... ...... rn:5 rd:5 &rr_esz +@rd_rnx2 ........ ... ..... ...... ..... rd:5 &rr_esz rn=%rn_ax2 # Two operand with governing predicate, flags setting @pd_pg_pn_s ........ . s:1 ...... .. pg:4 . rn:4 . rd:4 &rpr_s @@ -XXX,XX +XXX,XX @@ UABA 01000101 .. 0 ..... 11111 1 ..... ..... @rd_rn_rm #### SVE2 Narrowing ## SVE2 saturating extract narrow - # Bits 23, 18-16 are zero, limited in the translator via esz < 3 & imm == 0. -SQXTNB 01000101 .. 1 ..... 010 000 ..... ..... @rd_rn_tszimm_shl + +{ + SQCVTN_sh 01000101 00 1 10001 010 000 ....0 ..... @rd_rnx2 esz=1 + SQXTNB 01000101 .. 1 ..... 010 000 ..... ..... @rd_rn_tszimm_shl +} SQXTNT 01000101 .. 1 ..... 010 001 ..... ..... @rd_rn_tszimm_shl -UQXTNB 01000101 .. 1 ..... 010 010 ..... ..... @rd_rn_tszimm_shl +{ + UQCVTN_sh 01000101 00 1 10001 010 010 ....0 ..... @rd_rnx2 esz=1 + UQXTNB 01000101 .. 1 ..... 010 010 ..... ..... @rd_rn_tszimm_shl +} UQXTNT 01000101 .. 1 ..... 010 011 ..... ..... @rd_rn_tszimm_shl -SQXTUNB 01000101 .. 1 ..... 010 100 ..... ..... @rd_rn_tszimm_shl +{ + SQCVTUN_sh 01000101 00 1 10001 010 100 ....0 ..... @rd_rnx2 esz=1 + SQXTUNB 01000101 .. 1 ..... 010 100 ..... ..... @rd_rn_tszimm_shl +} SQXTUNT 01000101 .. 1 ..... 010 101 ..... ..... @rd_rn_tszimm_shl ## SVE2 bitwise shift right narrow diff --git a/target/arm/tcg/translate-sve.c b/target/arm/tcg/translate-sve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sve.c +++ b/target/arm/tcg/translate-sve.c @@ -XXX,XX +XXX,XX @@ static void gen_uclamp(unsigned vece, uint32_t d, uint32_t n, uint32_t m, } TRANS_FEAT(UCLAMP, aa64_sme, gen_gvec_fn_arg_zzzz, gen_uclamp, a) + +TRANS_FEAT(SQCVTN_sh, aa64_sme2_or_sve2p1, gen_gvec_ool_zz, + gen_helper_sme2_sqcvtn_sh, a->rd, a->rn, 0) +TRANS_FEAT(UQCVTN_sh, aa64_sme2_or_sve2p1, gen_gvec_ool_zz, + gen_helper_sme2_uqcvtn_sh, a->rd, a->rn, 0) +TRANS_FEAT(SQCVTUN_sh, aa64_sme2_or_sve2p1, gen_gvec_ool_zz, + gen_helper_sme2_sqcvtun_sh, a->rd, a->rn, 0) -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Message-id: 20250704142112.1018902-62-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper-sme.h | 13 ++++++++++++ target/arm/tcg/sme.decode | 18 ++++++++++++++++ target/arm/tcg/sme_helper.c | 38 ++++++++++++++++++++++++++++++++++ target/arm/tcg/translate-sme.c | 16 ++++++++++++++ 4 files changed, 85 insertions(+) diff --git a/target/arm/tcg/helper-sme.h b/target/arm/tcg/helper-sme.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-sme.h +++ b/target/arm/tcg/helper-sme.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_3(sme2_sqcvtun_sh, TCG_CALL_NO_RWG, void, ptr, ptr, i32) DEF_HELPER_FLAGS_3(sme2_sqcvtn_dh, TCG_CALL_NO_RWG, void, ptr, ptr, i32) DEF_HELPER_FLAGS_3(sme2_uqcvtn_dh, TCG_CALL_NO_RWG, void, ptr, ptr, i32) DEF_HELPER_FLAGS_3(sme2_sqcvtun_dh, TCG_CALL_NO_RWG, void, ptr, ptr, i32) + +DEF_HELPER_FLAGS_3(sme2_sunpk2_bh, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_sunpk2_hs, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_sunpk2_sd, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_sunpk4_bh, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_sunpk4_hs, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_sunpk4_sd, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_uunpk2_bh, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_uunpk2_hs, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_uunpk2_sd, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_uunpk4_bh, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_uunpk4_hs, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_uunpk4_sd, TCG_CALL_NO_RWG, void, ptr, ptr, i32) diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ SUB_aaz_d 11000001 111 000010 .. 111 ...00 11 ... @az_4x4_o3 &zz_n n=2 zd=%zd_ax2 zn=%zn_ax2 @zz_4x4 ........ ... ..... ...... .... . ..... \ &zz_n n=4 zd=%zd_ax4 zn=%zn_ax4 +@zz_4x2_n1 ........ ... ..... ...... .... . ..... \ + &zz_n n=1 zd=%zd_ax4 zn=%zn_ax2 BFCVT 11000001 011 00000 111000 ....0 ..... @zz_1x2 BFCVTN 11000001 011 00000 111000 ....1 ..... @zz_1x2 @@ -XXX,XX +XXX,XX @@ SQCVTUN_sb 11000001 011 10011 111000 ...10 ..... @zz_1x4 SQCVTN_dh 11000001 101 10011 111000 ...10 ..... @zz_1x4 UQCVTN_dh 11000001 101 10011 111000 ...11 ..... @zz_1x4 SQCVTUN_dh 11000001 111 10011 111000 ...10 ..... @zz_1x4 + +SUNPK_2bh 11000001 011 00101 111000 ..... ....0 @zz_2x1 +SUNPK_2hs 11000001 101 00101 111000 ..... ....0 @zz_2x1 +SUNPK_2sd 11000001 111 00101 111000 ..... ....0 @zz_2x1 + +UUNPK_2bh 11000001 011 00101 111000 ..... ....1 @zz_2x1 +UUNPK_2hs 11000001 101 00101 111000 ..... ....1 @zz_2x1 +UUNPK_2sd 11000001 111 00101 111000 ..... ....1 @zz_2x1 + +SUNPK_4bh 11000001 011 10101 111000 ....0 ...00 @zz_4x2_n1 +SUNPK_4hs 11000001 101 10101 111000 ....0 ...00 @zz_4x2_n1 +SUNPK_4sd 11000001 111 10101 111000 ....0 ...00 @zz_4x2_n1 + +UUNPK_4bh 11000001 011 10101 111000 ....0 ...01 @zz_4x2_n1 +UUNPK_4hs 11000001 101 10101 111000 ....0 ...01 @zz_4x2_n1 +UUNPK_4sd 11000001 111 10101 111000 ....0 ...01 @zz_4x2_n1 diff --git a/target/arm/tcg/sme_helper.c b/target/arm/tcg/sme_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme_helper.c +++ b/target/arm/tcg/sme_helper.c @@ -XXX,XX +XXX,XX @@ void HELPER(sme2_fcvt_w)(void *vd, void *vs, float_status *fpst, uint32_t desc) } } +#define UNPK(NAME, SREG, TW, TN, HW, HN) \ +void HELPER(NAME)(void *vd, void *vs, uint32_t desc) \ +{ \ + ARMVectorReg scratch[SREG]; \ + size_t oprsz = simd_oprsz(desc); \ + size_t n = oprsz / sizeof(TW); \ + if (vectors_overlap(vd, 2 * SREG, vs, SREG)) { \ + vs = memcpy(scratch, vs, sizeof(scratch)); \ + } \ + for (size_t r = 0; r < SREG; ++r) { \ + TN *s = vs + r * sizeof(ARMVectorReg); \ + for (size_t i = 0; i < 2; ++i) { \ + TW *d = vd + (2 * r + i) * sizeof(ARMVectorReg); \ + for (size_t e = 0; e < n; ++e) { \ + d[HW(e)] = s[HN(i * n + e)]; \ + } \ + } \ + } \ +} + +UNPK(sme2_sunpk2_bh, 1, int16_t, int8_t, H2, H1) +UNPK(sme2_sunpk2_hs, 1, int32_t, int16_t, H4, H2) +UNPK(sme2_sunpk2_sd, 1, int64_t, int32_t, H8, H4) + +UNPK(sme2_sunpk4_bh, 2, int16_t, int8_t, H2, H1) +UNPK(sme2_sunpk4_hs, 2, int32_t, int16_t, H4, H2) +UNPK(sme2_sunpk4_sd, 2, int64_t, int32_t, H8, H4) + +UNPK(sme2_uunpk2_bh, 1, uint16_t, uint8_t, H2, H1) +UNPK(sme2_uunpk2_hs, 1, uint32_t, uint16_t, H4, H2) +UNPK(sme2_uunpk2_sd, 1, uint64_t, uint32_t, H8, H4) + +UNPK(sme2_uunpk4_bh, 2, uint16_t, uint8_t, H2, H1) +UNPK(sme2_uunpk4_hs, 2, uint32_t, uint16_t, H4, H2) +UNPK(sme2_uunpk4_sd, 2, uint64_t, uint32_t, H8, H4) + +#undef UNPK + /* Deinterleave and convert. */ void HELPER(sme2_fcvtl)(void *vd, void *vs, float_status *fpst, uint32_t desc) { diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(SQCVTUN_sb, aa64_sme2, do_zz, a, 0, gen_helper_sme2_sqcvtun_sb) TRANS_FEAT(SQCVTN_dh, aa64_sme2, do_zz, a, 0, gen_helper_sme2_sqcvtn_dh) TRANS_FEAT(UQCVTN_dh, aa64_sme2, do_zz, a, 0, gen_helper_sme2_uqcvtn_dh) TRANS_FEAT(SQCVTUN_dh, aa64_sme2, do_zz, a, 0, gen_helper_sme2_sqcvtun_dh) + +TRANS_FEAT(SUNPK_2bh, aa64_sme2, do_zz, a, 0, gen_helper_sme2_sunpk2_bh) +TRANS_FEAT(SUNPK_2hs, aa64_sme2, do_zz, a, 0, gen_helper_sme2_sunpk2_hs) +TRANS_FEAT(SUNPK_2sd, aa64_sme2, do_zz, a, 0, gen_helper_sme2_sunpk2_sd) + +TRANS_FEAT(SUNPK_4bh, aa64_sme2, do_zz, a, 0, gen_helper_sme2_sunpk4_bh) +TRANS_FEAT(SUNPK_4hs, aa64_sme2, do_zz, a, 0, gen_helper_sme2_sunpk4_hs) +TRANS_FEAT(SUNPK_4sd, aa64_sme2, do_zz, a, 0, gen_helper_sme2_sunpk4_sd) + +TRANS_FEAT(UUNPK_2bh, aa64_sme2, do_zz, a, 0, gen_helper_sme2_uunpk2_bh) +TRANS_FEAT(UUNPK_2hs, aa64_sme2, do_zz, a, 0, gen_helper_sme2_uunpk2_hs) +TRANS_FEAT(UUNPK_2sd, aa64_sme2, do_zz, a, 0, gen_helper_sme2_uunpk2_sd) + +TRANS_FEAT(UUNPK_4bh, aa64_sme2, do_zz, a, 0, gen_helper_sme2_uunpk4_bh) +TRANS_FEAT(UUNPK_4hs, aa64_sme2, do_zz, a, 0, gen_helper_sme2_uunpk4_hs) +TRANS_FEAT(UUNPK_4sd, aa64_sme2, do_zz, a, 0, gen_helper_sme2_uunpk4_sd) -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-63-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper-sme.h | 12 ++++++ target/arm/tcg/sme.decode | 11 ++++++ target/arm/tcg/sme_helper.c | 68 ++++++++++++++++++++++++++++++++++ target/arm/tcg/translate-sme.c | 39 +++++++++++++++++++ 4 files changed, 130 insertions(+) diff --git a/target/arm/tcg/helper-sme.h b/target/arm/tcg/helper-sme.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-sme.h +++ b/target/arm/tcg/helper-sme.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_3(sme2_uunpk2_sd, TCG_CALL_NO_RWG, void, ptr, ptr, i32) DEF_HELPER_FLAGS_3(sme2_uunpk4_bh, TCG_CALL_NO_RWG, void, ptr, ptr, i32) DEF_HELPER_FLAGS_3(sme2_uunpk4_hs, TCG_CALL_NO_RWG, void, ptr, ptr, i32) DEF_HELPER_FLAGS_3(sme2_uunpk4_sd, TCG_CALL_NO_RWG, void, ptr, ptr, i32) + +DEF_HELPER_FLAGS_3(sme2_zip4_b, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_zip4_h, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_zip4_s, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_zip4_d, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_zip4_q, TCG_CALL_NO_RWG, void, ptr, ptr, i32) + +DEF_HELPER_FLAGS_3(sme2_uzp4_b, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_uzp4_h, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_uzp4_s, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_uzp4_d, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_uzp4_q, TCG_CALL_NO_RWG, void, ptr, ptr, i32) diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ SUB_aaz_d 11000001 111 000010 .. 111 ...00 11 ... @az_4x4_o3 ### SME2 Multi-vector SVE Constructive Unary +&zz_e zd zn esz &zz_n zd zn n @zz_1x2 ........ ... ..... ...... ..... zd:5 \ &zz_n n=1 zn=%zn_ax2 @@ -XXX,XX +XXX,XX @@ SUNPK_4sd 11000001 111 10101 111000 ....0 ...00 @zz_4x2_n1 UUNPK_4bh 11000001 011 10101 111000 ....0 ...01 @zz_4x2_n1 UUNPK_4hs 11000001 101 10101 111000 ....0 ...01 @zz_4x2_n1 UUNPK_4sd 11000001 111 10101 111000 ....0 ...01 @zz_4x2_n1 + +ZIP_4 11000001 esz:2 1 10110 111000 ...00 ... 00 \ + &zz_e zd=%zd_ax4 zn=%zn_ax4 +ZIP_4 11000001 001 10111 111000 ...00 ... 00 \ + &zz_e esz=4 zd=%zd_ax4 zn=%zn_ax4 + +UZP_4 11000001 esz:2 1 10110 111000 ...00 ... 10 \ + &zz_e zd=%zd_ax4 zn=%zn_ax4 +UZP_4 11000001 001 10111 111000 ...00 ... 10 \ + &zz_e esz=4 zd=%zd_ax4 zn=%zn_ax4 diff --git a/target/arm/tcg/sme_helper.c b/target/arm/tcg/sme_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme_helper.c +++ b/target/arm/tcg/sme_helper.c @@ -XXX,XX +XXX,XX @@ void HELPER(sme2_ucvtf)(void *vd, void *vs, float_status *fpst, uint32_t desc) d[i] = uint32_to_float32(s[i], fpst); } } + +#define ZIP4(NAME, TYPE, H) \ +void HELPER(NAME)(void *vd, void *vs, uint32_t desc) \ +{ \ + ARMVectorReg scratch[4]; \ + size_t oprsz = simd_oprsz(desc); \ + size_t quads = oprsz / (sizeof(TYPE) * 4); \ + TYPE *s0, *s1, *s2, *s3; \ + if (vs == vd) { \ + vs = memcpy(scratch, vs, sizeof(scratch)); \ + } \ + s0 = vs; \ + s1 = vs + sizeof(ARMVectorReg); \ + s2 = vs + 2 * sizeof(ARMVectorReg); \ + s3 = vs + 3 * sizeof(ARMVectorReg); \ + for (size_t r = 0; r < 4; ++r) { \ + TYPE *d = vd + r * sizeof(ARMVectorReg); \ + size_t base = r * quads; \ + for (size_t q = 0; q < quads; ++q) { \ + d[H(4 * q + 0)] = s0[base + H(q)]; \ + d[H(4 * q + 1)] = s1[base + H(q)]; \ + d[H(4 * q + 2)] = s2[base + H(q)]; \ + d[H(4 * q + 3)] = s3[base + H(q)]; \ + } \ + } \ +} + +ZIP4(sme2_zip4_b, uint8_t, H1) +ZIP4(sme2_zip4_h, uint16_t, H2) +ZIP4(sme2_zip4_s, uint32_t, H4) +ZIP4(sme2_zip4_d, uint64_t, ) +ZIP4(sme2_zip4_q, Int128, ) + +#undef ZIP4 + +#define UZP4(NAME, TYPE, H) \ +void HELPER(NAME)(void *vd, void *vs, uint32_t desc) \ +{ \ + ARMVectorReg scratch[4]; \ + size_t oprsz = simd_oprsz(desc); \ + size_t quads = oprsz / (sizeof(TYPE) * 4); \ + TYPE *d0, *d1, *d2, *d3; \ + if (vs == vd) { \ + vs = memcpy(scratch, vs, sizeof(scratch)); \ + } \ + d0 = vd; \ + d1 = vd + sizeof(ARMVectorReg); \ + d2 = vd + 2 * sizeof(ARMVectorReg); \ + d3 = vd + 3 * sizeof(ARMVectorReg); \ + for (size_t r = 0; r < 4; ++r) { \ + TYPE *s = vs + r * sizeof(ARMVectorReg); \ + size_t base = r * quads; \ + for (size_t q = 0; q < quads; ++q) { \ + d0[base + H(q)] = s[H(4 * q + 0)]; \ + d1[base + H(q)] = s[H(4 * q + 1)]; \ + d2[base + H(q)] = s[H(4 * q + 2)]; \ + d3[base + H(q)] = s[H(4 * q + 3)]; \ + } \ + } \ +} + +UZP4(sme2_uzp4_b, uint8_t, H1) +UZP4(sme2_uzp4_h, uint16_t, H2) +UZP4(sme2_uzp4_s, uint32_t, H4) +UZP4(sme2_uzp4_d, uint64_t, ) +UZP4(sme2_uzp4_q, Int128, ) + +#undef UZP4 diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(UUNPK_2sd, aa64_sme2, do_zz, a, 0, gen_helper_sme2_uunpk2_sd) TRANS_FEAT(UUNPK_4bh, aa64_sme2, do_zz, a, 0, gen_helper_sme2_uunpk4_bh) TRANS_FEAT(UUNPK_4hs, aa64_sme2, do_zz, a, 0, gen_helper_sme2_uunpk4_hs) TRANS_FEAT(UUNPK_4sd, aa64_sme2, do_zz, a, 0, gen_helper_sme2_uunpk4_sd) + +static bool do_zipuzp_4(DisasContext *s, arg_zz_e *a, + gen_helper_gvec_2 * const fn[5]) +{ + int bytes_per_op = 4 << a->esz; + + /* Both MO_64 and MO_128 can fail the size test. */ + if (s->max_svl < bytes_per_op) { + unallocated_encoding(s); + } else if (sme_sm_enabled_check(s)) { + int svl = streaming_vec_reg_size(s); + if (svl < bytes_per_op) { + unallocated_encoding(s); + } else { + tcg_gen_gvec_2_ool(vec_full_reg_offset(s, a->zd), + vec_full_reg_offset(s, a->zn), + svl, svl, 0, fn[a->esz]); + } + } + return true; +} + +static gen_helper_gvec_2 * const zip4_fns[] = { + gen_helper_sme2_zip4_b, + gen_helper_sme2_zip4_h, + gen_helper_sme2_zip4_s, + gen_helper_sme2_zip4_d, + gen_helper_sme2_zip4_q, +}; +TRANS_FEAT(ZIP_4, aa64_sme2, do_zipuzp_4, a, zip4_fns) + +static gen_helper_gvec_2 * const uzp4_fns[] = { + gen_helper_sme2_uzp4_b, + gen_helper_sme2_uzp4_h, + gen_helper_sme2_uzp4_s, + gen_helper_sme2_uzp4_d, + gen_helper_sme2_uzp4_q, +}; +TRANS_FEAT(UZP_4, aa64_sme2, do_zipuzp_4, a, uzp4_fns) -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Unify two copies of these inline functions. Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-64-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/vec_internal.h | 21 +++++++++++++++++++++ target/arm/tcg/mve_helper.c | 21 --------------------- target/arm/tcg/sve_helper.c | 21 --------------------- 3 files changed, 21 insertions(+), 42 deletions(-) diff --git a/target/arm/tcg/vec_internal.h b/target/arm/tcg/vec_internal.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/vec_internal.h +++ b/target/arm/tcg/vec_internal.h @@ -XXX,XX +XXX,XX @@ int64_t do_sqrdmlah_d(int64_t, int64_t, int64_t, bool, bool); #define do_usat_h(val) MIN(MAX(val, 0), UINT16_MAX) #define do_usat_s(val) MIN(MAX(val, 0), UINT32_MAX) +static inline uint64_t do_urshr(uint64_t x, unsigned sh) +{ + if (likely(sh < 64)) { + return (x >> sh) + ((x >> (sh - 1)) & 1); + } else if (sh == 64) { + return x >> 63; + } else { + return 0; + } +} + +static inline int64_t do_srshr(int64_t x, unsigned sh) +{ + if (likely(sh < 64)) { + return (x >> sh) + ((x >> (sh - 1)) & 1); + } else { + /* Rounding the sign bit always produces 0. */ + return 0; + } +} + /** * bfdotadd: * @sum: addend diff --git a/target/arm/tcg/mve_helper.c b/target/arm/tcg/mve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/mve_helper.c +++ b/target/arm/tcg/mve_helper.c @@ -XXX,XX +XXX,XX @@ DO_VSHLL_ALL(vshllt, true) DO_VSHRN(OP##tb, true, 1, uint8_t, 2, uint16_t, FN) \ DO_VSHRN(OP##th, true, 2, uint16_t, 4, uint32_t, FN) -static inline uint64_t do_urshr(uint64_t x, unsigned sh) -{ - if (likely(sh < 64)) { - return (x >> sh) + ((x >> (sh - 1)) & 1); - } else if (sh == 64) { - return x >> 63; - } else { - return 0; - } -} - -static inline int64_t do_srshr(int64_t x, unsigned sh) -{ - if (likely(sh < 64)) { - return (x >> sh) + ((x >> (sh - 1)) & 1); - } else { - /* Rounding the sign bit always produces 0. */ - return 0; - } -} - DO_VSHRN_ALL(vshrn, DO_SHR) DO_VSHRN_ALL(vrshrn, do_urshr) diff --git a/target/arm/tcg/sve_helper.c b/target/arm/tcg/sve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve_helper.c +++ b/target/arm/tcg/sve_helper.c @@ -XXX,XX +XXX,XX @@ void HELPER(NAME)(void *vd, void *vn, void *vg, uint32_t desc) \ when N is negative, add 2**M-1. */ #define DO_ASRD(N, M) ((N + (N < 0 ? ((__typeof(N))1 << M) - 1 : 0)) >> M) -static inline uint64_t do_urshr(uint64_t x, unsigned sh) -{ - if (likely(sh < 64)) { - return (x >> sh) + ((x >> (sh - 1)) & 1); - } else if (sh == 64) { - return x >> 63; - } else { - return 0; - } -} - -static inline int64_t do_srshr(int64_t x, unsigned sh) -{ - if (likely(sh < 64)) { - return (x >> sh) + ((x >> (sh - 1)) & 1); - } else { - /* Rounding the sign bit always produces 0. */ - return 0; - } -} - DO_ZPZI(sve_asr_zpzi_b, int8_t, H1, DO_SHR) DO_ZPZI(sve_asr_zpzi_h, int16_t, H1_2, DO_SHR) DO_ZPZI(sve_asr_zpzi_s, int32_t, H1_4, DO_SHR) -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Message-id: 20250704142112.1018902-65-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper-sme.h | 20 ++++++ target/arm/tcg/sme.decode | 37 ++++++++++ target/arm/tcg/sme_helper.c | 120 +++++++++++++++++++++++++++++++++ target/arm/tcg/translate-sme.c | 33 +++++++++ 4 files changed, 210 insertions(+) diff --git a/target/arm/tcg/helper-sme.h b/target/arm/tcg/helper-sme.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-sme.h +++ b/target/arm/tcg/helper-sme.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_3(sme2_uzp4_h, TCG_CALL_NO_RWG, void, ptr, ptr, i32) DEF_HELPER_FLAGS_3(sme2_uzp4_s, TCG_CALL_NO_RWG, void, ptr, ptr, i32) DEF_HELPER_FLAGS_3(sme2_uzp4_d, TCG_CALL_NO_RWG, void, ptr, ptr, i32) DEF_HELPER_FLAGS_3(sme2_uzp4_q, TCG_CALL_NO_RWG, void, ptr, ptr, i32) + +DEF_HELPER_FLAGS_3(sme2_sqrshr_sh, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_uqrshr_sh, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_sqrshru_sh, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_sqrshr_sb, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_uqrshr_sb, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_sqrshru_sb, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_sqrshr_dh, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_uqrshr_dh, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_sqrshru_dh, TCG_CALL_NO_RWG, void, ptr, ptr, i32) + +DEF_HELPER_FLAGS_3(sme2_sqrshrn_sh, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_uqrshrn_sh, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_sqrshrun_sh, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_sqrshrn_sb, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_uqrshrn_sb, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_sqrshrun_sb, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_sqrshrn_dh, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_uqrshrn_dh, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2_sqrshrun_dh, TCG_CALL_NO_RWG, void, ptr, ptr, i32) diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ UZP_4 11000001 esz:2 1 10110 111000 ...00 ... 10 \ &zz_e zd=%zd_ax4 zn=%zn_ax4 UZP_4 11000001 001 10111 111000 ...00 ... 10 \ &zz_e esz=4 zd=%zd_ax4 zn=%zn_ax4 + +### SME2 Multi-vector SVE Constructive Binary + +&rshr zd zn shift + +%rshr_sh_shift 16:4 !function=rsub_16 +%rshr_sb_shift 16:5 !function=rsub_32 +%rshr_dh_shift 22:1 16:5 !function=rsub_64 + +@rshr_sh ........ .... .... ...... ..... zd:5 \ + &rshr zn=%zn_ax2 shift=%rshr_sh_shift +@rshr_sb ........ ... ..... ...... ..... zd:5 \ + &rshr zn=%zn_ax4 shift=%rshr_sb_shift +@rshr_dh ........ ... ..... ...... ..... zd:5 \ + &rshr zn=%zn_ax4 shift=%rshr_dh_shift + +SQRSHR_sh 11000001 1110 .... 110101 ....0 ..... @rshr_sh +UQRSHR_sh 11000001 1110 .... 110101 ....1 ..... @rshr_sh +SQRSHRU_sh 11000001 1111 .... 110101 ....0 ..... @rshr_sh + +SQRSHR_sb 11000001 011 ..... 110110 ...00 ..... @rshr_sb +SQRSHR_dh 11000001 1.1 ..... 110110 ...00 ..... @rshr_dh +UQRSHR_sb 11000001 011 ..... 110110 ...01 ..... @rshr_sb +UQRSHR_dh 11000001 1.1 ..... 110110 ...01 ..... @rshr_dh +SQRSHRU_sb 11000001 011 ..... 110110 ...10 ..... @rshr_sb +SQRSHRU_dh 11000001 1.1 ..... 110110 ...10 ..... @rshr_dh + +SQRSHRN_sh 01000101 1011 .... 001010 ....0 ..... @rshr_sh +UQRSHRN_sh 01000101 1011 .... 001110 ....0 ..... @rshr_sh +SQRSHRUN_sh 01000101 1011 .... 000010 ....0 ..... @rshr_sh + +SQRSHRN_sb 11000001 011 ..... 110111 ...00 ..... @rshr_sb +SQRSHRN_dh 11000001 1.1 ..... 110111 ...00 ..... @rshr_dh +UQRSHRN_sb 11000001 011 ..... 110111 ...01 ..... @rshr_sb +UQRSHRN_dh 11000001 1.1 ..... 110111 ...01 ..... @rshr_dh +SQRSHRUN_sb 11000001 011 ..... 110111 ...10 ..... @rshr_sb +SQRSHRUN_dh 11000001 1.1 ..... 110111 ...10 ..... @rshr_dh diff --git a/target/arm/tcg/sme_helper.c b/target/arm/tcg/sme_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme_helper.c +++ b/target/arm/tcg/sme_helper.c @@ -XXX,XX +XXX,XX @@ SQCVT4(sme2_sqcvtu_dh, int64_t, uint16_t, H8, H2, do_usat_h) #undef SQCVT4 +#define SQRSHR2(NAME, TW, TN, HW, HN, RSHR, SAT) \ +void HELPER(NAME)(void *vd, void *vs, uint32_t desc) \ +{ \ + ARMVectorReg scratch; \ + size_t oprsz = simd_oprsz(desc), n = oprsz / sizeof(TW); \ + int shift = simd_data(desc); \ + TW *s0 = vs, *s1 = vs + sizeof(ARMVectorReg); \ + TN *d = vd; \ + if (vectors_overlap(vd, 1, vs, 2)) { \ + d = (TN *)&scratch; \ + } \ + for (size_t i = 0; i < n; ++i) { \ + d[HN(i)] = SAT(RSHR(s0[HW(i)], shift)); \ + d[HN(i + n)] = SAT(RSHR(s1[HW(i)], shift)); \ + } \ + if (d != vd) { \ + memcpy(vd, d, oprsz); \ + } \ +} + +SQRSHR2(sme2_sqrshr_sh, int32_t, int16_t, H4, H2, do_srshr, do_ssat_h) +SQRSHR2(sme2_uqrshr_sh, uint32_t, uint16_t, H4, H2, do_urshr, do_usat_h) +SQRSHR2(sme2_sqrshru_sh, int32_t, uint16_t, H4, H2, do_srshr, do_usat_h) + +#undef SQRSHR2 + +#define SQRSHR4(NAME, TW, TN, HW, HN, RSHR, SAT) \ +void HELPER(NAME)(void *vd, void *vs, uint32_t desc) \ +{ \ + ARMVectorReg scratch; \ + size_t oprsz = simd_oprsz(desc), n = oprsz / sizeof(TW); \ + int shift = simd_data(desc); \ + TW *s0 = vs, *s1 = vs + sizeof(ARMVectorReg); \ + TW *s2 = vs + 2 * sizeof(ARMVectorReg); \ + TW *s3 = vs + 3 * sizeof(ARMVectorReg); \ + TN *d = vd; \ + if (vectors_overlap(vd, 1, vs, 4)) { \ + d = (TN *)&scratch; \ + } \ + for (size_t i = 0; i < n; ++i) { \ + d[HN(i)] = SAT(RSHR(s0[HW(i)], shift)); \ + d[HN(i + n)] = SAT(RSHR(s1[HW(i)], shift)); \ + d[HN(i + 2 * n)] = SAT(RSHR(s2[HW(i)], shift)); \ + d[HN(i + 3 * n)] = SAT(RSHR(s3[HW(i)], shift)); \ + } \ + if (d != vd) { \ + memcpy(vd, d, oprsz); \ + } \ +} + +SQRSHR4(sme2_sqrshr_sb, int32_t, int8_t, H4, H2, do_srshr, do_ssat_b) +SQRSHR4(sme2_uqrshr_sb, uint32_t, uint8_t, H4, H2, do_urshr, do_usat_b) +SQRSHR4(sme2_sqrshru_sb, int32_t, uint8_t, H4, H2, do_srshr, do_usat_b) + +SQRSHR4(sme2_sqrshr_dh, int64_t, int16_t, H8, H2, do_srshr, do_ssat_h) +SQRSHR4(sme2_uqrshr_dh, uint64_t, uint16_t, H8, H2, do_urshr, do_usat_h) +SQRSHR4(sme2_sqrshru_dh, int64_t, uint16_t, H8, H2, do_srshr, do_usat_h) + +#undef SQRSHR4 + /* Convert and interleave */ void HELPER(sme2_bfcvtn)(void *vd, void *vs, float_status *fpst, uint32_t desc) { @@ -XXX,XX +XXX,XX @@ SQCVTN4(sme2_sqcvtun_dh, int64_t, uint16_t, H8, H2, do_usat_h) #undef SQCVTN4 +#define SQRSHRN2(NAME, TW, TN, HW, HN, RSHR, SAT) \ +void HELPER(NAME)(void *vd, void *vs, uint32_t desc) \ +{ \ + ARMVectorReg scratch; \ + size_t oprsz = simd_oprsz(desc), n = oprsz / sizeof(TW); \ + int shift = simd_data(desc); \ + TW *s0 = vs, *s1 = vs + sizeof(ARMVectorReg); \ + TN *d = vd; \ + if (vectors_overlap(vd, 1, vs, 2)) { \ + d = (TN *)&scratch; \ + } \ + for (size_t i = 0; i < n; ++i) { \ + d[HN(2 * i + 0)] = SAT(RSHR(s0[HW(i)], shift)); \ + d[HN(2 * i + 1)] = SAT(RSHR(s1[HW(i)], shift)); \ + } \ + if (d != vd) { \ + memcpy(vd, d, oprsz); \ + } \ +} + +SQRSHRN2(sme2_sqrshrn_sh, int32_t, int16_t, H4, H2, do_srshr, do_ssat_h) +SQRSHRN2(sme2_uqrshrn_sh, uint32_t, uint16_t, H4, H2, do_urshr, do_usat_h) +SQRSHRN2(sme2_sqrshrun_sh, int32_t, uint16_t, H4, H2, do_srshr, do_usat_h) + +#undef SQRSHRN2 + +#define SQRSHRN4(NAME, TW, TN, HW, HN, RSHR, SAT) \ +void HELPER(NAME)(void *vd, void *vs, uint32_t desc) \ +{ \ + ARMVectorReg scratch; \ + size_t oprsz = simd_oprsz(desc), n = oprsz / sizeof(TW); \ + int shift = simd_data(desc); \ + TW *s0 = vs, *s1 = vs + sizeof(ARMVectorReg); \ + TW *s2 = vs + 2 * sizeof(ARMVectorReg); \ + TW *s3 = vs + 3 * sizeof(ARMVectorReg); \ + TN *d = vd; \ + if (vectors_overlap(vd, 1, vs, 4)) { \ + d = (TN *)&scratch; \ + } \ + for (size_t i = 0; i < n; ++i) { \ + d[HN(4 * i + 0)] = SAT(RSHR(s0[HW(i)], shift)); \ + d[HN(4 * i + 1)] = SAT(RSHR(s1[HW(i)], shift)); \ + d[HN(4 * i + 2)] = SAT(RSHR(s2[HW(i)], shift)); \ + d[HN(4 * i + 3)] = SAT(RSHR(s3[HW(i)], shift)); \ + } \ + if (d != vd) { \ + memcpy(vd, d, oprsz); \ + } \ +} + +SQRSHRN4(sme2_sqrshrn_sb, int32_t, int8_t, H4, H1, do_srshr, do_ssat_b) +SQRSHRN4(sme2_uqrshrn_sb, uint32_t, uint8_t, H4, H1, do_urshr, do_usat_b) +SQRSHRN4(sme2_sqrshrun_sb, int32_t, uint8_t, H4, H1, do_srshr, do_usat_b) + +SQRSHRN4(sme2_sqrshrn_dh, int64_t, int16_t, H8, H2, do_srshr, do_ssat_h) +SQRSHRN4(sme2_uqrshrn_dh, uint64_t, uint16_t, H8, H2, do_urshr, do_usat_h) +SQRSHRN4(sme2_sqrshrun_dh, int64_t, uint16_t, H8, H2, do_srshr, do_usat_h) + +#undef SQRSHRN4 + /* Expand and convert */ void HELPER(sme2_fcvt_w)(void *vd, void *vs, float_status *fpst, uint32_t desc) { diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ static gen_helper_gvec_2 * const uzp4_fns[] = { gen_helper_sme2_uzp4_q, }; TRANS_FEAT(UZP_4, aa64_sme2, do_zipuzp_4, a, uzp4_fns) + +static bool do_zz_rshr(DisasContext *s, arg_rshr *a, gen_helper_gvec_2 *fn) +{ + if (sve_access_check(s)) { + int vl = vec_full_reg_size(s); + tcg_gen_gvec_2_ool(vec_full_reg_offset(s, a->zd), + vec_full_reg_offset(s, a->zn), + vl, vl, a->shift, fn); + } + return true; +} + +TRANS_FEAT(SQRSHR_sh, aa64_sme2, do_zz_rshr, a, gen_helper_sme2_sqrshr_sh) +TRANS_FEAT(UQRSHR_sh, aa64_sme2, do_zz_rshr, a, gen_helper_sme2_uqrshr_sh) +TRANS_FEAT(SQRSHRU_sh, aa64_sme2, do_zz_rshr, a, gen_helper_sme2_sqrshru_sh) + +TRANS_FEAT(SQRSHR_sb, aa64_sme2, do_zz_rshr, a, gen_helper_sme2_sqrshr_sb) +TRANS_FEAT(SQRSHR_dh, aa64_sme2, do_zz_rshr, a, gen_helper_sme2_sqrshr_dh) +TRANS_FEAT(UQRSHR_sb, aa64_sme2, do_zz_rshr, a, gen_helper_sme2_uqrshr_sb) +TRANS_FEAT(UQRSHR_dh, aa64_sme2, do_zz_rshr, a, gen_helper_sme2_uqrshr_dh) +TRANS_FEAT(SQRSHRU_sb, aa64_sme2, do_zz_rshr, a, gen_helper_sme2_sqrshru_sb) +TRANS_FEAT(SQRSHRU_dh, aa64_sme2, do_zz_rshr, a, gen_helper_sme2_sqrshru_dh) + +TRANS_FEAT(SQRSHRN_sh, aa64_sme2_or_sve2p1, do_zz_rshr, a, gen_helper_sme2_sqrshrn_sh) +TRANS_FEAT(UQRSHRN_sh, aa64_sme2_or_sve2p1, do_zz_rshr, a, gen_helper_sme2_uqrshrn_sh) +TRANS_FEAT(SQRSHRUN_sh, aa64_sme2_or_sve2p1, do_zz_rshr, a, gen_helper_sme2_sqrshrun_sh) + +TRANS_FEAT(SQRSHRN_sb, aa64_sme2, do_zz_rshr, a, gen_helper_sme2_sqrshrn_sb) +TRANS_FEAT(SQRSHRN_dh, aa64_sme2, do_zz_rshr, a, gen_helper_sme2_sqrshrn_dh) +TRANS_FEAT(UQRSHRN_sb, aa64_sme2, do_zz_rshr, a, gen_helper_sme2_uqrshrn_sb) +TRANS_FEAT(UQRSHRN_dh, aa64_sme2, do_zz_rshr, a, gen_helper_sme2_uqrshrn_dh) +TRANS_FEAT(SQRSHRUN_sb, aa64_sme2, do_zz_rshr, a, gen_helper_sme2_sqrshrun_sb) +TRANS_FEAT(SQRSHRUN_dh, aa64_sme2, do_zz_rshr, a, gen_helper_sme2_sqrshrun_dh) -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-66-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper-sme.h | 12 +++++++ target/arm/tcg/sme.decode | 12 +++++++ target/arm/tcg/sme_helper.c | 62 ++++++++++++++++++++++++++++++++++ target/arm/tcg/translate-sme.c | 40 ++++++++++++++++++++++ 4 files changed, 126 insertions(+) diff --git a/target/arm/tcg/helper-sme.h b/target/arm/tcg/helper-sme.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-sme.h +++ b/target/arm/tcg/helper-sme.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_3(sme2_uunpk4_bh, TCG_CALL_NO_RWG, void, ptr, ptr, i32) DEF_HELPER_FLAGS_3(sme2_uunpk4_hs, TCG_CALL_NO_RWG, void, ptr, ptr, i32) DEF_HELPER_FLAGS_3(sme2_uunpk4_sd, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sme2_zip2_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sme2_zip2_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sme2_zip2_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sme2_zip2_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sme2_zip2_q, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) + +DEF_HELPER_FLAGS_4(sme2_uzp2_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sme2_uzp2_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sme2_uzp2_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sme2_uzp2_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sme2_uzp2_q, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) + DEF_HELPER_FLAGS_3(sme2_zip4_b, TCG_CALL_NO_RWG, void, ptr, ptr, i32) DEF_HELPER_FLAGS_3(sme2_zip4_h, TCG_CALL_NO_RWG, void, ptr, ptr, i32) DEF_HELPER_FLAGS_3(sme2_zip4_s, TCG_CALL_NO_RWG, void, ptr, ptr, i32) diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ UQRSHRN_sb 11000001 011 ..... 110111 ...01 ..... @rshr_sb UQRSHRN_dh 11000001 1.1 ..... 110111 ...01 ..... @rshr_dh SQRSHRUN_sb 11000001 011 ..... 110111 ...10 ..... @rshr_sb SQRSHRUN_dh 11000001 1.1 ..... 110111 ...10 ..... @rshr_dh + +&zzz_e zd zn zm esz + +ZIP_2 11000001 esz:2 1 zm:5 110100 zn:5 .... 0 \ + &zzz_e zd=%zd_ax2 +ZIP_2 11000001 00 1 zm:5 110101 zn:5 .... 0 \ + &zzz_e zd=%zd_ax2 esz=4 + +UZP_2 11000001 esz:2 1 zm:5 110100 zn:5 .... 1 \ + &zzz_e zd=%zd_ax2 +UZP_2 11000001 00 1 zm:5 110101 zn:5 .... 1 \ + &zzz_e zd=%zd_ax2 esz=4 diff --git a/target/arm/tcg/sme_helper.c b/target/arm/tcg/sme_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme_helper.c +++ b/target/arm/tcg/sme_helper.c @@ -XXX,XX +XXX,XX @@ void HELPER(sme2_ucvtf)(void *vd, void *vs, float_status *fpst, uint32_t desc) } } +#define ZIP2(NAME, TYPE, H) \ +void HELPER(NAME)(void *vd, void *vn, void *vm, uint32_t desc) \ +{ \ + ARMVectorReg scratch[2]; \ + size_t oprsz = simd_oprsz(desc); \ + size_t pairs = oprsz / (sizeof(TYPE) * 2); \ + TYPE *n = vn, *m = vm; \ + if (vectors_overlap(vd, 2, vn, 1)) { \ + n = memcpy(&scratch[0], vn, oprsz); \ + } \ + if (vectors_overlap(vd, 2, vm, 1)) { \ + m = memcpy(&scratch[1], vm, oprsz); \ + } \ + for (size_t r = 0; r < 2; ++r) { \ + TYPE *d = vd + r * sizeof(ARMVectorReg); \ + size_t base = r * pairs; \ + for (size_t p = 0; p < pairs; ++p) { \ + d[H(2 * p + 0)] = n[base + H(p)]; \ + d[H(2 * p + 1)] = m[base + H(p)]; \ + } \ + } \ +} + +ZIP2(sme2_zip2_b, uint8_t, H1) +ZIP2(sme2_zip2_h, uint16_t, H2) +ZIP2(sme2_zip2_s, uint32_t, H4) +ZIP2(sme2_zip2_d, uint64_t, ) +ZIP2(sme2_zip2_q, Int128, ) + +#undef ZIP2 + #define ZIP4(NAME, TYPE, H) \ void HELPER(NAME)(void *vd, void *vs, uint32_t desc) \ { \ @@ -XXX,XX +XXX,XX @@ ZIP4(sme2_zip4_q, Int128, ) #undef ZIP4 +#define UZP2(NAME, TYPE, H) \ +void HELPER(NAME)(void *vd, void *vn, void *vm, uint32_t desc) \ +{ \ + ARMVectorReg scratch[2]; \ + size_t oprsz = simd_oprsz(desc); \ + size_t pairs = oprsz / (sizeof(TYPE) * 2); \ + TYPE *d0 = vd, *d1 = vd + sizeof(ARMVectorReg); \ + if (vectors_overlap(vd, 2, vn, 1)) { \ + vn = memcpy(&scratch[0], vn, oprsz); \ + } \ + if (vectors_overlap(vd, 2, vm, 1)) { \ + vm = memcpy(&scratch[1], vm, oprsz); \ + } \ + for (size_t r = 0; r < 2; ++r) { \ + TYPE *s = r ? vm : vn; \ + size_t base = r * pairs; \ + for (size_t p = 0; p < pairs; ++p) { \ + d0[base + H(p)] = s[H(2 * p + 0)]; \ + d1[base + H(p)] = s[H(2 * p + 1)]; \ + } \ + } \ +} + +UZP2(sme2_uzp2_b, uint8_t, H1) +UZP2(sme2_uzp2_h, uint16_t, H2) +UZP2(sme2_uzp2_s, uint32_t, H4) +UZP2(sme2_uzp2_d, uint64_t, ) +UZP2(sme2_uzp2_q, Int128, ) + +#undef UZP2 + #define UZP4(NAME, TYPE, H) \ void HELPER(NAME)(void *vd, void *vs, uint32_t desc) \ { \ diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(UQRSHRN_sb, aa64_sme2, do_zz_rshr, a, gen_helper_sme2_uqrshrn_sb) TRANS_FEAT(UQRSHRN_dh, aa64_sme2, do_zz_rshr, a, gen_helper_sme2_uqrshrn_dh) TRANS_FEAT(SQRSHRUN_sb, aa64_sme2, do_zz_rshr, a, gen_helper_sme2_sqrshrun_sb) TRANS_FEAT(SQRSHRUN_dh, aa64_sme2, do_zz_rshr, a, gen_helper_sme2_sqrshrun_dh) + +static bool do_zipuzp_2(DisasContext *s, arg_zzz_e *a, + gen_helper_gvec_3 * const fn[5]) +{ + int bytes_per_op = 2 << a->esz; + + /* MO_128 can fail the size test. */ + if (s->max_svl < bytes_per_op) { + unallocated_encoding(s); + } else if (sme_sm_enabled_check(s)) { + int svl = streaming_vec_reg_size(s); + if (svl < bytes_per_op) { + unallocated_encoding(s); + } else { + tcg_gen_gvec_3_ool(vec_full_reg_offset(s, a->zd), + vec_full_reg_offset(s, a->zn), + vec_full_reg_offset(s, a->zm), + svl, svl, 0, fn[a->esz]); + } + } + return true; +} + +static gen_helper_gvec_3 * const zip2_fns[] = { + gen_helper_sme2_zip2_b, + gen_helper_sme2_zip2_h, + gen_helper_sme2_zip2_s, + gen_helper_sme2_zip2_d, + gen_helper_sme2_zip2_q, +}; +TRANS_FEAT(ZIP_2, aa64_sme2, do_zipuzp_2, a, zip2_fns) + +static gen_helper_gvec_3 * const uzp2_fns[] = { + gen_helper_sme2_uzp2_b, + gen_helper_sme2_uzp2_h, + gen_helper_sme2_uzp2_s, + gen_helper_sme2_uzp2_d, + gen_helper_sme2_uzp2_q, +}; +TRANS_FEAT(UZP_2, aa64_sme2, do_zipuzp_2, a, uzp2_fns) -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-67-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper-sme.h | 15 +++++++ target/arm/tcg/sme.decode | 17 ++++++++ target/arm/tcg/sme_helper.c | 56 +++++++++++++++++++++++++ target/arm/tcg/translate-sme.c | 75 ++++++++++++++++++++++++++++++++++ 4 files changed, 163 insertions(+) diff --git a/target/arm/tcg/helper-sme.h b/target/arm/tcg/helper-sme.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-sme.h +++ b/target/arm/tcg/helper-sme.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_3(sme2_sqrshrun_sb, TCG_CALL_NO_RWG, void, ptr, ptr, i32) DEF_HELPER_FLAGS_3(sme2_sqrshrn_dh, TCG_CALL_NO_RWG, void, ptr, ptr, i32) DEF_HELPER_FLAGS_3(sme2_uqrshrn_dh, TCG_CALL_NO_RWG, void, ptr, ptr, i32) DEF_HELPER_FLAGS_3(sme2_sqrshrun_dh, TCG_CALL_NO_RWG, void, ptr, ptr, i32) + +DEF_HELPER_FLAGS_4(sme2_sclamp_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sme2_sclamp_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sme2_sclamp_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sme2_sclamp_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) + +DEF_HELPER_FLAGS_4(sme2_uclamp_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sme2_uclamp_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sme2_uclamp_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sme2_uclamp_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) + +DEF_HELPER_FLAGS_5(sme2_fclamp_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_5(sme2_fclamp_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_5(sme2_fclamp_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_5(sme2_bfclamp, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ UZP_2 11000001 esz:2 1 zm:5 110100 zn:5 .... 1 \ &zzz_e zd=%zd_ax2 UZP_2 11000001 00 1 zm:5 110101 zn:5 .... 1 \ &zzz_e zd=%zd_ax2 esz=4 + +&zzz_en zd zn zm esz n + +FCLAMP 11000001 esz:2 1 zm:5 110000 zn:5 .... 0 \ + &zzz_en zd=%zd_ax2 n=2 +FCLAMP 11000001 esz:2 1 zm:5 110010 zn:5 ...0 0 \ + &zzz_en zd=%zd_ax4 n=4 + +SCLAMP 11000001 esz:2 1 zm:5 110001 zn:5 .... 0 \ + &zzz_en zd=%zd_ax2 n=2 +SCLAMP 11000001 esz:2 1 zm:5 110011 zn:5 ...0 0 \ + &zzz_en zd=%zd_ax4 n=4 + +UCLAMP 11000001 esz:2 1 zm:5 110001 zn:5 .... 1 \ + &zzz_en zd=%zd_ax2 n=2 +UCLAMP 11000001 esz:2 1 zm:5 110011 zn:5 ...0 1 \ + &zzz_en zd=%zd_ax4 n=4 diff --git a/target/arm/tcg/sme_helper.c b/target/arm/tcg/sme_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme_helper.c +++ b/target/arm/tcg/sme_helper.c @@ -XXX,XX +XXX,XX @@ UZP4(sme2_uzp4_d, uint64_t, ) UZP4(sme2_uzp4_q, Int128, ) #undef UZP4 + +#define ICLAMP(NAME, TYPE, H) \ +void HELPER(NAME)(void *vd, void *vn, void *vm, uint32_t desc) \ +{ \ + size_t stride = sizeof(ARMVectorReg) / sizeof(TYPE); \ + size_t elements = simd_oprsz(desc) / sizeof(TYPE); \ + size_t nreg = simd_data(desc); \ + TYPE *d = vd, *n = vn, *m = vm; \ + for (size_t e = 0; e < elements; e++) { \ + TYPE nn = n[H(e)], mm = m[H(e)]; \ + for (size_t r = 0; r < nreg; r++) { \ + TYPE *dd = &d[r * stride + H(e)]; \ + *dd = MIN(MAX(*dd, nn), mm); \ + } \ + } \ +} + +ICLAMP(sme2_sclamp_b, int8_t, H1) +ICLAMP(sme2_sclamp_h, int16_t, H2) +ICLAMP(sme2_sclamp_s, int32_t, H4) +ICLAMP(sme2_sclamp_d, int64_t, H8) + +ICLAMP(sme2_uclamp_b, uint8_t, H1) +ICLAMP(sme2_uclamp_h, uint16_t, H2) +ICLAMP(sme2_uclamp_s, uint32_t, H4) +ICLAMP(sme2_uclamp_d, uint64_t, H8) + +#undef ICLAMP + +/* + * Note the argument ordering to minnum and maxnum must match + * the ARM pseudocode so that NaNs are propagated properly. + */ +#define FCLAMP(NAME, TYPE, H) \ +void HELPER(NAME)(void *vd, void *vn, void *vm, \ + float_status *fpst, uint32_t desc) \ +{ \ + size_t stride = sizeof(ARMVectorReg) / sizeof(TYPE); \ + size_t elements = simd_oprsz(desc) / sizeof(TYPE); \ + size_t nreg = simd_data(desc); \ + TYPE *d = vd, *n = vn, *m = vm; \ + for (size_t e = 0; e < elements; e++) { \ + TYPE nn = n[H(e)], mm = m[H(e)]; \ + for (size_t r = 0; r < nreg; r++) { \ + TYPE *dd = &d[r * stride + H(e)]; \ + *dd = TYPE##_minnum(TYPE##_maxnum(nn, *dd, fpst), mm, fpst); \ + } \ + } \ +} + +FCLAMP(sme2_fclamp_h, float16, H2) +FCLAMP(sme2_fclamp_s, float32, H4) +FCLAMP(sme2_fclamp_d, float64, H8) +FCLAMP(sme2_bfclamp, bfloat16, H2) + +#undef FCLAMP diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ static gen_helper_gvec_3 * const uzp2_fns[] = { gen_helper_sme2_uzp2_q, }; TRANS_FEAT(UZP_2, aa64_sme2, do_zipuzp_2, a, uzp2_fns) + +static bool trans_FCLAMP(DisasContext *s, arg_zzz_en *a) +{ + static gen_helper_gvec_3_ptr * const fn[] = { + gen_helper_sme2_bfclamp, + gen_helper_sme2_fclamp_h, + gen_helper_sme2_fclamp_s, + gen_helper_sme2_fclamp_d, + }; + TCGv_ptr fpst; + int vl; + + if (!dc_isar_feature(aa64_sme2, s)) { + return false; + } + /* This insn uses MO_8 to encode BFloat16. */ + if (a->esz == MO_8 && !dc_isar_feature(aa64_sme_b16b16, s)) { + return false; + } + if (!sme_sm_enabled_check(s)) { + return true; + } + + fpst = fpstatus_ptr(a->esz == MO_16 ? FPST_A64_F16 : FPST_A64); + vl = vec_full_reg_size(s); + + tcg_gen_gvec_3_ptr(vec_full_reg_offset(s, a->zd), + vec_full_reg_offset(s, a->zn), + vec_full_reg_offset(s, a->zm), + fpst, vl, vl, a->n, fn[a->esz]); + return true; +} + +static bool do_clamp(DisasContext *s, arg_zzz_en *a, + gen_helper_gvec_3 * const fn[4]) +{ + int vl; + + if (!dc_isar_feature(aa64_sme2, s)) { + return false; + } + if (!sme_sm_enabled_check(s)) { + return true; + } + + /* + * Clamp is just a min+max, easily supported by most host + * vector operations -- we already have such an expansion in + * translate-sve.c for a single output. + * TODO: Add support in gvec for multiple simultaneous output, + * and/or copy to temporary upon overlap. + */ + vl = vec_full_reg_size(s); + tcg_gen_gvec_3_ool(vec_full_reg_offset(s, a->zd), + vec_full_reg_offset(s, a->zn), + vec_full_reg_offset(s, a->zm), + vl, vl, a->n, fn[a->esz]); + return true; +} + +static gen_helper_gvec_3 * const sclamp_fns[] = { + gen_helper_sme2_sclamp_b, + gen_helper_sme2_sclamp_h, + gen_helper_sme2_sclamp_s, + gen_helper_sme2_sclamp_d, +}; +TRANS(SCLAMP, do_clamp, a, sclamp_fns) + +static gen_helper_gvec_3 * const uclamp_fns[] = { + gen_helper_sme2_uclamp_b, + gen_helper_sme2_uclamp_h, + gen_helper_sme2_uclamp_s, + gen_helper_sme2_uclamp_d, +}; +TRANS(UCLAMP, do_clamp, a, uclamp_fns) -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> These instructions are present in both SME(1) and SVE2.1 extensions. Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-68-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/translate-sve.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/target/arm/tcg/translate-sve.c b/target/arm/tcg/translate-sve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sve.c +++ b/target/arm/tcg/translate-sve.c @@ -XXX,XX +XXX,XX @@ static void gen_sclamp(unsigned vece, uint32_t d, uint32_t n, uint32_t m, tcg_gen_gvec_4(d, n, m, a, oprsz, maxsz, &ops[vece]); } -TRANS_FEAT(SCLAMP, aa64_sme, gen_gvec_fn_arg_zzzz, gen_sclamp, a) +TRANS_FEAT(SCLAMP, aa64_sme_or_sve2p1, gen_gvec_fn_arg_zzzz, gen_sclamp, a) static void gen_uclamp_i32(TCGv_i32 d, TCGv_i32 n, TCGv_i32 m, TCGv_i32 a) { @@ -XXX,XX +XXX,XX @@ static void gen_uclamp(unsigned vece, uint32_t d, uint32_t n, uint32_t m, tcg_gen_gvec_4(d, n, m, a, oprsz, maxsz, &ops[vece]); } -TRANS_FEAT(UCLAMP, aa64_sme, gen_gvec_fn_arg_zzzz, gen_uclamp, a) +TRANS_FEAT(UCLAMP, aa64_sme_or_sve2p1, gen_gvec_fn_arg_zzzz, gen_uclamp, a) TRANS_FEAT(SQCVTN_sh, aa64_sme2_or_sve2p1, gen_gvec_ool_zz, gen_helper_sme2_sqcvtn_sh, a->rd, a->rn, 0) -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> This is the single vector version within SVE decode space. Tested-by: Alex Bennée <alex.bennee@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-69-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/sve.decode | 2 ++ target/arm/tcg/translate-sve.c | 22 ++++++++++++++++++++++ 2 files changed, 24 insertions(+) diff --git a/target/arm/tcg/sve.decode b/target/arm/tcg/sve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve.decode +++ b/target/arm/tcg/sve.decode @@ -XXX,XX +XXX,XX @@ PSEL 00100101 .1 1 000 .. 01 .... 0 .... 0 .... \ SCLAMP 01000100 .. 0 ..... 110000 ..... ..... @rda_rn_rm UCLAMP 01000100 .. 0 ..... 110001 ..... ..... @rda_rn_rm + +FCLAMP 01100100 .. 1 ..... 001001 ..... ..... @rda_rn_rm diff --git a/target/arm/tcg/translate-sve.c b/target/arm/tcg/translate-sve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sve.c +++ b/target/arm/tcg/translate-sve.c @@ -XXX,XX +XXX,XX @@ static void gen_uclamp(unsigned vece, uint32_t d, uint32_t n, uint32_t m, TRANS_FEAT(UCLAMP, aa64_sme_or_sve2p1, gen_gvec_fn_arg_zzzz, gen_uclamp, a) +static bool trans_FCLAMP(DisasContext *s, arg_FCLAMP *a) +{ + static gen_helper_gvec_3_ptr * const fn[] = { + gen_helper_sme2_bfclamp, + gen_helper_sme2_fclamp_h, + gen_helper_sme2_fclamp_s, + gen_helper_sme2_fclamp_d, + }; + + /* This insn uses MO_8 to encode BFloat16. */ + if (a->esz == MO_8 + ? !dc_isar_feature(aa64_sve_b16b16, s) + : !dc_isar_feature(aa64_sme2_or_sve2p1, s)) { + return false; + } + + /* So far we never optimize rda with MOVPRFX */ + assert(a->rd == a->ra); + return gen_gvec_fpst_zzz(s, fn[a->esz], a->rd, a->rn, a->rm, 1, + a->esz == MO_16 ? FPST_A64_F16 : FPST_A64); +} + TRANS_FEAT(SQCVTN_sh, aa64_sme2_or_sve2p1, gen_gvec_ool_zz, gen_helper_sme2_sqcvtn_sh, a->rd, a->rn, 0) TRANS_FEAT(UQCVTN_sh, aa64_sme2_or_sve2p1, gen_gvec_ool_zz, -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-70-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/sme.decode | 23 +++++++++++++++++++++++ target/arm/tcg/translate-sme.c | 20 ++++++++++++++++++++ 2 files changed, 43 insertions(+) diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ UCLAMP 11000001 esz:2 1 zm:5 110001 zn:5 .... 1 \ &zzz_en zd=%zd_ax2 n=2 UCLAMP 11000001 esz:2 1 zm:5 110011 zn:5 ...0 1 \ &zzz_en zd=%zd_ax4 n=4 + +### SME Multiple Zero + +&zero_za rv off ngrp nvec + +ZERO_za 11000000 000011 000 .. 0000000000 off:3 \ + &zero_za ngrp=2 nvec=1 rv=%mova_rv +ZERO_za 11000000 000011 100 .. 0000000000 off:3 \ + &zero_za ngrp=4 nvec=1 rv=%mova_rv + +ZERO_za 11000000 000011 001 .. 0000000000 ... \ + &zero_za ngrp=1 nvec=2 rv=%mova_rv off=%off3_x2 +ZERO_za 11000000 000011 010 .. 0000000000 0.. \ + &zero_za ngrp=2 nvec=2 rv=%mova_rv off=%off2_x2 +ZERO_za 11000000 000011 011 .. 0000000000 0.. \ + &zero_za ngrp=4 nvec=2 rv=%mova_rv off=%off2_x2 + +ZERO_za 11000000 000011 101 .. 0000000000 0.. \ + &zero_za ngrp=1 nvec=4 rv=%mova_rv off=%off2_x4 +ZERO_za 11000000 000011 110 .. 0000000000 00. \ + &zero_za ngrp=2 nvec=4 rv=%mova_rv off=%off1_x4 +ZERO_za 11000000 000011 111 .. 0000000000 00. \ + &zero_za ngrp=4 nvec=4 rv=%mova_rv off=%off1_x4 diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ static bool trans_ZERO_zt0(DisasContext *s, arg_ZERO_zt0 *a) return true; } +static bool trans_ZERO_za(DisasContext *s, arg_ZERO_za *a) +{ + if (!dc_isar_feature(aa64_sme2p1, s)) { + return false; + } + if (sme_smza_enabled_check(s)) { + int svl = streaming_vec_reg_size(s); + int vstride = svl / a->ngrp; + TCGv_ptr t_za = get_zarray(s, a->rv, a->off, a->ngrp, a->nvec); + + for (int r = 0; r < a->ngrp; ++r) { + for (int i = 0; i < a->nvec; ++i) { + int o_za = (r * vstride + i) * sizeof(ARMVectorReg); + tcg_gen_gvec_dup_imm_var(MO_64, t_za, o_za, svl, svl, 0); + } + } + } + return true; +} + static bool do_mova_tile(DisasContext *s, arg_mova_p *a, bool to_vec) { static gen_helper_gvec_4 * const h_fns[5] = { -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> For WHILE, we have the count of enabled predicates, so we don't need to search to compute the PredTest result. Reuse the logic that will shortly be required for counted predicates. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Message-id: 20250704142112.1018902-71-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/sve_helper.c | 81 +++++++++++++++++++++---------------- 1 file changed, 46 insertions(+), 35 deletions(-) diff --git a/target/arm/tcg/sve_helper.c b/target/arm/tcg/sve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve_helper.c +++ b/target/arm/tcg/sve_helper.c @@ -XXX,XX +XXX,XX @@ uint64_t HELPER(sve_cntp)(void *vn, void *vg, uint32_t pred_desc) return sum; } +/* C.f. Arm pseudocode PredCountTest */ +static uint32_t pred_count_test(uint32_t elements, uint32_t count, bool invert) +{ + uint32_t flags; + + if (count == 0) { + flags = 1; /* !N, Z, C */ + } else if (!invert) { + flags = (1u << 31) | 2; /* N, !Z */ + flags |= count != elements; /* C */ + } else { + flags = 2; /* !Z, !C */ + flags |= (count == elements) << 31; /* N */ + } + return flags; +} + uint32_t HELPER(sve_whilel)(void *vd, uint32_t count, uint32_t pred_desc) { intptr_t oprsz = FIELD_EX32(pred_desc, PREDDESC, OPRSZ); intptr_t esz = FIELD_EX32(pred_desc, PREDDESC, ESZ); uint64_t esz_mask = pred_esz_masks[esz]; ARMPredicateReg *d = vd; - uint32_t flags; - intptr_t i; + intptr_t i, oprbits = oprsz * 8; + + tcg_debug_assert(count <= oprbits); /* Begin with a zero predicate register. */ - flags = do_zero(d, oprsz); - if (count == 0) { - return flags; + do_zero(d, oprsz); + if (count) { + /* Set all of the requested bits. */ + for (i = 0; i < count / 64; ++i) { + d->p[i] = esz_mask; + } + if (count & 63) { + d->p[i] = MAKE_64BIT_MASK(0, count & 63) & esz_mask; + } } - /* Set all of the requested bits. */ - for (i = 0; i < count / 64; ++i) { - d->p[i] = esz_mask; - } - if (count & 63) { - d->p[i] = MAKE_64BIT_MASK(0, count & 63) & esz_mask; - } - - return predtest_ones(d, oprsz, esz_mask); + return pred_count_test(oprbits, count, false); } uint32_t HELPER(sve_whileg)(void *vd, uint32_t count, uint32_t pred_desc) @@ -XXX,XX +XXX,XX @@ uint32_t HELPER(sve_whileg)(void *vd, uint32_t count, uint32_t pred_desc) intptr_t esz = FIELD_EX32(pred_desc, PREDDESC, ESZ); uint64_t esz_mask = pred_esz_masks[esz]; ARMPredicateReg *d = vd; - intptr_t i, invcount, oprbits; + intptr_t i, invcount, oprbits = oprsz * 8; uint64_t bits; - if (count == 0) { - return do_zero(d, oprsz); - } - - oprbits = oprsz * 8; tcg_debug_assert(count <= oprbits); - bits = esz_mask; - if (oprbits & 63) { - bits &= MAKE_64BIT_MASK(0, oprbits & 63); - } - - invcount = oprbits - count; - for (i = (oprsz - 1) / 8; i > invcount / 64; --i) { - d->p[i] = bits; + /* Begin with a zero predicate register. */ + do_zero(d, oprsz); + if (count) { + /* Set all of the requested bits. */ bits = esz_mask; + if (oprbits & 63) { + bits &= MAKE_64BIT_MASK(0, oprbits & 63); + } + + invcount = oprbits - count; + for (i = (oprsz - 1) / 8; i > invcount / 64; --i) { + d->p[i] = bits; + bits = esz_mask; + } + d->p[i] = bits & MAKE_64BIT_MASK(invcount & 63, 64); } - d->p[i] = bits & MAKE_64BIT_MASK(invcount & 63, 64); - - while (--i >= 0) { - d->p[i] = 0; - } - - return predtest_ones(d, oprsz, esz_mask); + return pred_count_test(oprbits, count, true); } /* Recursive reduction on a function; -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Merge predtest_ones into its only caller. Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-72-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/sve_helper.c | 34 ++++++++++++++-------------------- 1 file changed, 14 insertions(+), 20 deletions(-) diff --git a/target/arm/tcg/sve_helper.c b/target/arm/tcg/sve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve_helper.c +++ b/target/arm/tcg/sve_helper.c @@ -XXX,XX +XXX,XX @@ void HELPER(sve_brkn)(void *vd, void *vn, void *vg, uint32_t pred_desc) } } -/* As if PredTest(Ones(PL), D, esz). */ -static uint32_t predtest_ones(ARMPredicateReg *d, intptr_t oprsz, - uint64_t esz_mask) -{ - uint32_t flags = PREDTEST_INIT; - intptr_t i; - - for (i = 0; i < oprsz / 8; i++) { - flags = iter_predtest_fwd(d->p[i], esz_mask, flags); - } - if (oprsz & 7) { - uint64_t mask = ~(-1ULL << (8 * (oprsz & 7))); - flags = iter_predtest_fwd(d->p[i], esz_mask & mask, flags); - } - return flags; -} - uint32_t HELPER(sve_brkns)(void *vd, void *vn, void *vg, uint32_t pred_desc) { intptr_t oprsz = FIELD_EX32(pred_desc, PREDDESC, OPRSZ); if (last_active_pred(vn, vg, oprsz)) { - return predtest_ones(vd, oprsz, -1); - } else { - return do_zero(vd, oprsz); + ARMPredicateReg *d = vd; + uint32_t flags = PREDTEST_INIT; + intptr_t i; + + /* As if PredTest(Ones(PL), D, MO_8). */ + for (i = 0; i < oprsz / 8; i++) { + flags = iter_predtest_fwd(d->p[i], -1, flags); + } + if (oprsz & 7) { + uint64_t mask = ~(-1ULL << (8 * (oprsz & 7))); + flags = iter_predtest_fwd(d->p[i], mask, flags); + } + return flags; } + return do_zero(vd, oprsz); } uint64_t HELPER(sve_cntp)(void *vn, void *vg, uint32_t pred_desc) -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Expand to memset plus the return value, when used. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Message-id: 20250704142112.1018902-73-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/sve_helper.c | 28 +++++++++++----------------- 1 file changed, 11 insertions(+), 17 deletions(-) diff --git a/target/arm/tcg/sve_helper.c b/target/arm/tcg/sve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve_helper.c +++ b/target/arm/tcg/sve_helper.c @@ -XXX,XX +XXX,XX @@ static uint32_t compute_brks_m(uint64_t *d, uint64_t *n, uint64_t *g, return flags; } -static uint32_t do_zero(ARMPredicateReg *d, intptr_t oprsz) -{ - /* It is quicker to zero the whole predicate than loop on OPRSZ. - * The compiler should turn this into 4 64-bit integer stores. - */ - memset(d, 0, sizeof(ARMPredicateReg)); - return PREDTEST_INIT; -} - void HELPER(sve_brkpa)(void *vd, void *vn, void *vm, void *vg, uint32_t pred_desc) { @@ -XXX,XX +XXX,XX @@ void HELPER(sve_brkpa)(void *vd, void *vn, void *vm, void *vg, if (last_active_pred(vn, vg, oprsz)) { compute_brk_z(vd, vm, vg, oprsz, true); } else { - do_zero(vd, oprsz); + memset(vd, 0, sizeof(ARMPredicateReg)); } } @@ -XXX,XX +XXX,XX @@ uint32_t HELPER(sve_brkpas)(void *vd, void *vn, void *vm, void *vg, if (last_active_pred(vn, vg, oprsz)) { return compute_brks_z(vd, vm, vg, oprsz, true); } else { - return do_zero(vd, oprsz); + memset(vd, 0, sizeof(ARMPredicateReg)); + return PREDTEST_INIT; } } @@ -XXX,XX +XXX,XX @@ void HELPER(sve_brkpb)(void *vd, void *vn, void *vm, void *vg, if (last_active_pred(vn, vg, oprsz)) { compute_brk_z(vd, vm, vg, oprsz, false); } else { - do_zero(vd, oprsz); + memset(vd, 0, sizeof(ARMPredicateReg)); } } @@ -XXX,XX +XXX,XX @@ uint32_t HELPER(sve_brkpbs)(void *vd, void *vn, void *vm, void *vg, if (last_active_pred(vn, vg, oprsz)) { return compute_brks_z(vd, vm, vg, oprsz, false); } else { - return do_zero(vd, oprsz); + memset(vd, 0, sizeof(ARMPredicateReg)); + return PREDTEST_INIT; } } @@ -XXX,XX +XXX,XX @@ void HELPER(sve_brkn)(void *vd, void *vn, void *vg, uint32_t pred_desc) { intptr_t oprsz = FIELD_EX32(pred_desc, PREDDESC, OPRSZ); if (!last_active_pred(vn, vg, oprsz)) { - do_zero(vd, oprsz); + memset(vd, 0, sizeof(ARMPredicateReg)); } } @@ -XXX,XX +XXX,XX @@ uint32_t HELPER(sve_brkns)(void *vd, void *vn, void *vg, uint32_t pred_desc) } return flags; } - return do_zero(vd, oprsz); + memset(vd, 0, sizeof(ARMPredicateReg)); + return PREDTEST_INIT; } uint64_t HELPER(sve_cntp)(void *vn, void *vg, uint32_t pred_desc) @@ -XXX,XX +XXX,XX @@ uint32_t HELPER(sve_whilel)(void *vd, uint32_t count, uint32_t pred_desc) tcg_debug_assert(count <= oprbits); /* Begin with a zero predicate register. */ - do_zero(d, oprsz); + memset(d, 0, sizeof(*d)); if (count) { /* Set all of the requested bits. */ for (i = 0; i < count / 64; ++i) { @@ -XXX,XX +XXX,XX @@ uint32_t HELPER(sve_whileg)(void *vd, uint32_t count, uint32_t pred_desc) tcg_debug_assert(count <= oprbits); /* Begin with a zero predicate register. */ - do_zero(d, oprsz); + memset(d, 0, sizeof(*d)); if (count) { /* Set all of the requested bits. */ bits = esz_mask; -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-74-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/sve_helper.c | 26 ++++++++++++++++---------- 1 file changed, 16 insertions(+), 10 deletions(-) diff --git a/target/arm/tcg/sve_helper.c b/target/arm/tcg/sve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve_helper.c +++ b/target/arm/tcg/sve_helper.c @@ -XXX,XX +XXX,XX @@ static uint32_t pred_count_test(uint32_t elements, uint32_t count, bool invert) return flags; } -uint32_t HELPER(sve_whilel)(void *vd, uint32_t count, uint32_t pred_desc) +/* D must be cleared on entry. */ +static void do_whilel(ARMPredicateReg *d, uint64_t esz_mask, + uint32_t count, uint32_t oprbits) { - intptr_t oprsz = FIELD_EX32(pred_desc, PREDDESC, OPRSZ); - intptr_t esz = FIELD_EX32(pred_desc, PREDDESC, ESZ); - uint64_t esz_mask = pred_esz_masks[esz]; - ARMPredicateReg *d = vd; - intptr_t i, oprbits = oprsz * 8; - tcg_debug_assert(count <= oprbits); - - /* Begin with a zero predicate register. */ - memset(d, 0, sizeof(*d)); if (count) { + uint32_t i; + /* Set all of the requested bits. */ for (i = 0; i < count / 64; ++i) { d->p[i] = esz_mask; @@ -XXX,XX +XXX,XX @@ uint32_t HELPER(sve_whilel)(void *vd, uint32_t count, uint32_t pred_desc) d->p[i] = MAKE_64BIT_MASK(0, count & 63) & esz_mask; } } +} +uint32_t HELPER(sve_whilel)(void *vd, uint32_t count, uint32_t pred_desc) +{ + uint32_t oprsz = FIELD_EX32(pred_desc, PREDDESC, OPRSZ); + uint32_t esz = FIELD_EX32(pred_desc, PREDDESC, ESZ); + uint32_t oprbits = oprsz * 8; + uint64_t esz_mask = pred_esz_masks[esz]; + ARMPredicateReg *d = vd; + + memset(d, 0, sizeof(*d)); + do_whilel(d, esz_mask, count, oprbits); return pred_count_test(oprbits, count, false); } -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Message-id: 20250704142112.1018902-75-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/sve_helper.c | 39 +++++++++++++++++++------------------ 1 file changed, 20 insertions(+), 19 deletions(-) diff --git a/target/arm/tcg/sve_helper.c b/target/arm/tcg/sve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve_helper.c +++ b/target/arm/tcg/sve_helper.c @@ -XXX,XX +XXX,XX @@ uint32_t HELPER(sve_whilel)(void *vd, uint32_t count, uint32_t pred_desc) return pred_count_test(oprbits, count, false); } -uint32_t HELPER(sve_whileg)(void *vd, uint32_t count, uint32_t pred_desc) +/* D must be cleared on entry. */ +static void do_whileg(ARMPredicateReg *d, uint64_t esz_mask, + uint32_t count, uint32_t oprbits) { - intptr_t oprsz = FIELD_EX32(pred_desc, PREDDESC, OPRSZ); - intptr_t esz = FIELD_EX32(pred_desc, PREDDESC, ESZ); - uint64_t esz_mask = pred_esz_masks[esz]; - ARMPredicateReg *d = vd; - intptr_t i, invcount, oprbits = oprsz * 8; - uint64_t bits; - tcg_debug_assert(count <= oprbits); - - /* Begin with a zero predicate register. */ - memset(d, 0, sizeof(*d)); if (count) { - /* Set all of the requested bits. */ - bits = esz_mask; - if (oprbits & 63) { - bits &= MAKE_64BIT_MASK(0, oprbits & 63); - } + uint32_t i, invcount = oprbits - count; + uint64_t bits = esz_mask & MAKE_64BIT_MASK(invcount & 63, 64); - invcount = oprbits - count; - for (i = (oprsz - 1) / 8; i > invcount / 64; --i) { + for (i = invcount / 64; i < oprbits / 64; ++i) { d->p[i] = bits; bits = esz_mask; } - d->p[i] = bits & MAKE_64BIT_MASK(invcount & 63, 64); + if (oprbits & 63) { + d->p[i] = bits & MAKE_64BIT_MASK(0, oprbits & 63); + } } +} +uint32_t HELPER(sve_whileg)(void *vd, uint32_t count, uint32_t pred_desc) +{ + uint32_t oprsz = FIELD_EX32(pred_desc, PREDDESC, OPRSZ); + uint32_t esz = FIELD_EX32(pred_desc, PREDDESC, ESZ); + uint32_t oprbits = oprsz * 8; + uint64_t esz_mask = pred_esz_masks[esz]; + ARMPredicateReg *d = vd; + + memset(d, 0, sizeof(*d)); + do_whileg(d, esz_mask, count, oprbits); return pred_count_test(oprbits, count, true); } -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Change the API to pass element count rather than bit count. This will be helpful later for predicate as counter. Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-76-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/sve_helper.c | 2 ++ target/arm/tcg/translate-sve.c | 13 +++++-------- 2 files changed, 7 insertions(+), 8 deletions(-) diff --git a/target/arm/tcg/sve_helper.c b/target/arm/tcg/sve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve_helper.c +++ b/target/arm/tcg/sve_helper.c @@ -XXX,XX +XXX,XX @@ uint32_t HELPER(sve_whilel)(void *vd, uint32_t count, uint32_t pred_desc) uint64_t esz_mask = pred_esz_masks[esz]; ARMPredicateReg *d = vd; + count <<= esz; memset(d, 0, sizeof(*d)); do_whilel(d, esz_mask, count, oprbits); return pred_count_test(oprbits, count, false); @@ -XXX,XX +XXX,XX @@ uint32_t HELPER(sve_whileg)(void *vd, uint32_t count, uint32_t pred_desc) uint64_t esz_mask = pred_esz_masks[esz]; ARMPredicateReg *d = vd; + count <<= esz; memset(d, 0, sizeof(*d)); do_whileg(d, esz_mask, count, oprbits); return pred_count_test(oprbits, count, true); diff --git a/target/arm/tcg/translate-sve.c b/target/arm/tcg/translate-sve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sve.c +++ b/target/arm/tcg/translate-sve.c @@ -XXX,XX +XXX,XX @@ static bool trans_WHILE(DisasContext *s, arg_WHILE *a) t2 = tcg_temp_new_i32(); tcg_gen_extrl_i64_i32(t2, t0); - /* Scale elements to bits. */ - tcg_gen_shli_i32(t2, t2, a->esz); - desc = FIELD_DP32(desc, PREDDESC, OPRSZ, vsz / 8); desc = FIELD_DP32(desc, PREDDESC, ESZ, a->esz); @@ -XXX,XX +XXX,XX @@ static bool trans_WHILE_ptr(DisasContext *s, arg_WHILE_ptr *a) op0 = read_cpu_reg(s, a->rn, 1); op1 = read_cpu_reg(s, a->rm, 1); - tmax = tcg_constant_i64(vsz); + tmax = tcg_constant_i64(vsz >> a->esz); diff = tcg_temp_new_i64(); if (a->rw) { @@ -XXX,XX +XXX,XX @@ static bool trans_WHILE_ptr(DisasContext *s, arg_WHILE_ptr *a) tcg_gen_sub_i64(diff, op0, op1); tcg_gen_sub_i64(t1, op1, op0); tcg_gen_movcond_i64(TCG_COND_GEU, diff, op0, op1, diff, t1); - /* Round down to a multiple of ESIZE. */ - tcg_gen_andi_i64(diff, diff, -1 << a->esz); + /* Divide, rounding down, by ESIZE. */ + tcg_gen_shri_i64(diff, diff, a->esz); /* If op1 == op0, diff == 0, and the condition is always true. */ tcg_gen_movcond_i64(TCG_COND_EQ, diff, op0, op1, tmax, diff); } else { /* WHILEWR */ tcg_gen_sub_i64(diff, op1, op0); - /* Round down to a multiple of ESIZE. */ - tcg_gen_andi_i64(diff, diff, -1 << a->esz); + /* Divide, rounding down, by ESIZE. */ + tcg_gen_shri_i64(diff, diff, a->esz); /* If op0 >= op1, diff <= 0, the condition is always true. */ tcg_gen_movcond_i64(TCG_COND_GEU, diff, op0, op1, tmax, diff); } -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Use TRANS_FEAT to select the correct predicate. Pass the helper and a boolean to do_WHILE. Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-77-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/sve.decode | 4 +++- target/arm/tcg/translate-sve.c | 23 +++++++++-------------- 2 files changed, 12 insertions(+), 15 deletions(-) diff --git a/target/arm/tcg/sve.decode b/target/arm/tcg/sve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve.decode +++ b/target/arm/tcg/sve.decode @@ -XXX,XX +XXX,XX @@ SINCDECP_z 00100101 .. 1010 d:1 u:1 10000 00 .... ..... @incdec2_pred CTERM 00100101 1 sf:1 1 rm:5 001000 rn:5 ne:1 0000 # SVE integer compare scalar count and limit -WHILE 00100101 esz:2 1 rm:5 000 sf:1 u:1 lt:1 rn:5 eq:1 rd:4 +&while esz rd rn rm sf u eq +WHILE_lt 00100101 esz:2 1 rm:5 000 sf:1 u:1 1 rn:5 eq:1 rd:4 &while +WHILE_gt 00100101 esz:2 1 rm:5 000 sf:1 u:1 0 rn:5 eq:1 rd:4 &while # SVE2 pointer conflict compare WHILE_ptr 00100101 esz:2 1 rm:5 001 100 rn:5 rw:1 rd:4 diff --git a/target/arm/tcg/translate-sve.c b/target/arm/tcg/translate-sve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sve.c +++ b/target/arm/tcg/translate-sve.c @@ -XXX,XX +XXX,XX @@ static bool trans_CTERM(DisasContext *s, arg_CTERM *a) return true; } -static bool trans_WHILE(DisasContext *s, arg_WHILE *a) +typedef void gen_while_fn(TCGv_i32, TCGv_ptr, TCGv_i32, TCGv_i32); +static bool do_WHILE(DisasContext *s, arg_while *a, bool lt, gen_while_fn *fn) { TCGv_i64 op0, op1, t0, t1, tmax; TCGv_i32 t2; @@ -XXX,XX +XXX,XX @@ static bool trans_WHILE(DisasContext *s, arg_WHILE *a) TCGCond cond; uint64_t maxval; /* Note that GE/HS has a->eq == 0 and GT/HI has a->eq == 1. */ - bool eq = a->eq == a->lt; + bool eq = a->eq == lt; - /* The greater-than conditions are all SVE2. */ - if (a->lt - ? !dc_isar_feature(aa64_sve, s) - : !dc_isar_feature(aa64_sve2, s)) { - return false; - } if (!sve_access_check(s)) { return true; } @@ -XXX,XX +XXX,XX @@ static bool trans_WHILE(DisasContext *s, arg_WHILE *a) t0 = tcg_temp_new_i64(); t1 = tcg_temp_new_i64(); - if (a->lt) { + if (lt) { tcg_gen_sub_i64(t0, op1, op0); if (a->u) { maxval = a->sf ? UINT64_MAX : UINT32_MAX; @@ -XXX,XX +XXX,XX @@ static bool trans_WHILE(DisasContext *s, arg_WHILE *a) ptr = tcg_temp_new_ptr(); tcg_gen_addi_ptr(ptr, tcg_env, pred_full_reg_offset(s, a->rd)); - if (a->lt) { - gen_helper_sve_whilel(t2, ptr, t2, tcg_constant_i32(desc)); - } else { - gen_helper_sve_whileg(t2, ptr, t2, tcg_constant_i32(desc)); - } + fn(t2, ptr, t2, tcg_constant_i32(desc)); + do_pred_flags(t2); return true; } +TRANS_FEAT(WHILE_lt, aa64_sve, do_WHILE, a, true, gen_helper_sve_whilel) +TRANS_FEAT(WHILE_gt, aa64_sve2, do_WHILE, a, false, gen_helper_sve_whileg) + static bool trans_WHILE_ptr(DisasContext *s, arg_WHILE_ptr *a) { TCGv_i64 op0, op1, diff, t1, tmax; -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-78-richard.henderson@linaro.org This instruction is present in both SME(1) and SVE2.1 extensions. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/translate-sve.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/target/arm/tcg/translate-sve.c b/target/arm/tcg/translate-sve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sve.c +++ b/target/arm/tcg/translate-sve.c @@ -XXX,XX +XXX,XX @@ static bool trans_PSEL(DisasContext *s, arg_psel *a) TCGv_i64 tmp, didx, dbit; TCGv_ptr ptr; - if (!dc_isar_feature(aa64_sme, s)) { + if (!dc_isar_feature(aa64_sme_or_sve2p1, s)) { return false; } if (!sve_access_check(s)) { -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-79-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper-sve.h | 3 +++ target/arm/tcg/sve.decode | 8 +++++++ target/arm/tcg/sve_helper.c | 40 ++++++++++++++++++++++++++++++++++ target/arm/tcg/translate-sve.c | 14 ++++++++---- 4 files changed, 61 insertions(+), 4 deletions(-) diff --git a/target/arm/tcg/helper-sve.h b/target/arm/tcg/helper-sve.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-sve.h +++ b/target/arm/tcg/helper-sve.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_3(sve_cntp, TCG_CALL_NO_RWG, i64, ptr, ptr, i32) DEF_HELPER_FLAGS_3(sve_whilel, TCG_CALL_NO_RWG, i32, ptr, i32, i32) DEF_HELPER_FLAGS_3(sve_whileg, TCG_CALL_NO_RWG, i32, ptr, i32, i32) +DEF_HELPER_FLAGS_3(sve_while2l, TCG_CALL_NO_RWG, i32, ptr, i32, i32) +DEF_HELPER_FLAGS_3(sve_while2g, TCG_CALL_NO_RWG, i32, ptr, i32, i32) + DEF_HELPER_FLAGS_4(sve_subri_b, TCG_CALL_NO_RWG, void, ptr, ptr, i64, i32) DEF_HELPER_FLAGS_4(sve_subri_h, TCG_CALL_NO_RWG, void, ptr, ptr, i64, i32) DEF_HELPER_FLAGS_4(sve_subri_s, TCG_CALL_NO_RWG, void, ptr, ptr, i64, i32) diff --git a/target/arm/tcg/sve.decode b/target/arm/tcg/sve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve.decode +++ b/target/arm/tcg/sve.decode @@ -XXX,XX +XXX,XX @@ WHILE_gt 00100101 esz:2 1 rm:5 000 sf:1 u:1 0 rn:5 eq:1 rd:4 &while # SVE2 pointer conflict compare WHILE_ptr 00100101 esz:2 1 rm:5 001 100 rn:5 rw:1 rd:4 +# SVE2.1 predicate pair +%pd_pair 1:3 !function=times_2 +@while_pair ........ esz:2 . rm:5 .... u:1 . rn:5 . ... eq:1 \ + &while rd=%pd_pair sf=1 + +WHILE_lt_pair 00100101 .. 1 ..... 0101 . 1 ..... 1 ... . @while_pair +WHILE_gt_pair 00100101 .. 1 ..... 0101 . 0 ..... 1 ... . @while_pair + ### SVE Integer Wide Immediate - Unpredicated Group # SVE broadcast floating-point immediate (unpredicated) diff --git a/target/arm/tcg/sve_helper.c b/target/arm/tcg/sve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve_helper.c +++ b/target/arm/tcg/sve_helper.c @@ -XXX,XX +XXX,XX @@ uint32_t HELPER(sve_whilel)(void *vd, uint32_t count, uint32_t pred_desc) return pred_count_test(oprbits, count, false); } +uint32_t HELPER(sve_while2l)(void *vd, uint32_t count, uint32_t pred_desc) +{ + uint32_t oprsz = FIELD_EX32(pred_desc, PREDDESC, OPRSZ); + uint32_t esz = FIELD_EX32(pred_desc, PREDDESC, ESZ); + uint32_t oprbits = oprsz * 8; + uint64_t esz_mask = pred_esz_masks[esz]; + ARMPredicateReg *d = vd; + + count <<= esz; + memset(d, 0, 2 * sizeof(*d)); + if (count <= oprbits) { + do_whilel(&d[0], esz_mask, count, oprbits); + } else { + do_whilel(&d[0], esz_mask, oprbits, oprbits); + do_whilel(&d[1], esz_mask, count - oprbits, oprbits); + } + + return pred_count_test(2 * oprbits, count, false); +} + /* D must be cleared on entry. */ static void do_whileg(ARMPredicateReg *d, uint64_t esz_mask, uint32_t count, uint32_t oprbits) @@ -XXX,XX +XXX,XX @@ uint32_t HELPER(sve_whileg)(void *vd, uint32_t count, uint32_t pred_desc) return pred_count_test(oprbits, count, true); } +uint32_t HELPER(sve_while2g)(void *vd, uint32_t count, uint32_t pred_desc) +{ + uint32_t oprsz = FIELD_EX32(pred_desc, PREDDESC, OPRSZ); + uint32_t esz = FIELD_EX32(pred_desc, PREDDESC, ESZ); + uint32_t oprbits = oprsz * 8; + uint64_t esz_mask = pred_esz_masks[esz]; + ARMPredicateReg *d = vd; + + count <<= esz; + memset(d, 0, 2 * sizeof(*d)); + if (count <= oprbits) { + do_whileg(&d[1], esz_mask, count, oprbits); + } else { + do_whilel(&d[1], esz_mask, oprbits, oprbits); + do_whileg(&d[0], esz_mask, count - oprbits, oprbits); + } + + return pred_count_test(2 * oprbits, count, true); +} + /* Recursive reduction on a function; * C.f. the ARM ARM function ReducePredicated. * diff --git a/target/arm/tcg/translate-sve.c b/target/arm/tcg/translate-sve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sve.c +++ b/target/arm/tcg/translate-sve.c @@ -XXX,XX +XXX,XX @@ static bool trans_CTERM(DisasContext *s, arg_CTERM *a) } typedef void gen_while_fn(TCGv_i32, TCGv_ptr, TCGv_i32, TCGv_i32); -static bool do_WHILE(DisasContext *s, arg_while *a, bool lt, gen_while_fn *fn) +static bool do_WHILE(DisasContext *s, arg_while *a, + bool lt, int scale, gen_while_fn *fn) { TCGv_i64 op0, op1, t0, t1, tmax; TCGv_i32 t2; @@ -XXX,XX +XXX,XX @@ static bool do_WHILE(DisasContext *s, arg_while *a, bool lt, gen_while_fn *fn) } } - tmax = tcg_constant_i64(vsz >> a->esz); + tmax = tcg_constant_i64((vsz << scale) >> a->esz); if (eq) { /* Equality means one more iteration. */ tcg_gen_addi_i64(t0, t0, 1); @@ -XXX,XX +XXX,XX @@ static bool do_WHILE(DisasContext *s, arg_while *a, bool lt, gen_while_fn *fn) return true; } -TRANS_FEAT(WHILE_lt, aa64_sve, do_WHILE, a, true, gen_helper_sve_whilel) -TRANS_FEAT(WHILE_gt, aa64_sve2, do_WHILE, a, false, gen_helper_sve_whileg) +TRANS_FEAT(WHILE_lt, aa64_sve, do_WHILE, a, true, 0, gen_helper_sve_whilel) +TRANS_FEAT(WHILE_gt, aa64_sve2, do_WHILE, a, false, 0, gen_helper_sve_whileg) + +TRANS_FEAT(WHILE_lt_pair, aa64_sme2_or_sve2p1, do_WHILE, + a, true, 1, gen_helper_sve_while2l) +TRANS_FEAT(WHILE_gt_pair, aa64_sme2_or_sve2p1, do_WHILE, + a, false, 1, gen_helper_sve_while2g) static bool trans_WHILE_ptr(DisasContext *s, arg_WHILE_ptr *a) { -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-80-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper-sve.h | 3 ++ target/arm/tcg/sve.decode | 11 +++++++ target/arm/tcg/sve_helper.c | 53 ++++++++++++++++++++++++++++++++++ target/arm/tcg/translate-sve.c | 22 ++++++++++---- 4 files changed, 84 insertions(+), 5 deletions(-) diff --git a/target/arm/tcg/helper-sve.h b/target/arm/tcg/helper-sve.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-sve.h +++ b/target/arm/tcg/helper-sve.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_3(sve_whileg, TCG_CALL_NO_RWG, i32, ptr, i32, i32) DEF_HELPER_FLAGS_3(sve_while2l, TCG_CALL_NO_RWG, i32, ptr, i32, i32) DEF_HELPER_FLAGS_3(sve_while2g, TCG_CALL_NO_RWG, i32, ptr, i32, i32) +DEF_HELPER_FLAGS_3(sve_whilecl, TCG_CALL_NO_RWG, i32, ptr, i32, i32) +DEF_HELPER_FLAGS_3(sve_whilecg, TCG_CALL_NO_RWG, i32, ptr, i32, i32) + DEF_HELPER_FLAGS_4(sve_subri_b, TCG_CALL_NO_RWG, void, ptr, ptr, i64, i32) DEF_HELPER_FLAGS_4(sve_subri_h, TCG_CALL_NO_RWG, void, ptr, ptr, i64, i32) DEF_HELPER_FLAGS_4(sve_subri_s, TCG_CALL_NO_RWG, void, ptr, ptr, i64, i32) diff --git a/target/arm/tcg/sve.decode b/target/arm/tcg/sve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve.decode +++ b/target/arm/tcg/sve.decode @@ -XXX,XX +XXX,XX @@ %rn_ax2 6:4 !function=times_2 +%pnd 0:3 !function=plus_8 + ########################################################################### # Named attribute sets. These are used to make nice(er) names # when creating helpers common to those for the individual @@ -XXX,XX +XXX,XX @@ WHILE_ptr 00100101 esz:2 1 rm:5 001 100 rn:5 rw:1 rd:4 WHILE_lt_pair 00100101 .. 1 ..... 0101 . 1 ..... 1 ... . @while_pair WHILE_gt_pair 00100101 .. 1 ..... 0101 . 0 ..... 1 ... . @while_pair +# SVE2.1 predicate as count +@while_cnt ........ esz:2 . rm:5 .... u:1 . rn:5 . eq:1 ... \ + &while rd=%pnd sf=1 + +WHILE_lt_cnt2 00100101 .. 1 ..... 0100 . 1 ..... 1 . ... @while_cnt +WHILE_lt_cnt4 00100101 .. 1 ..... 0110 . 1 ..... 1 . ... @while_cnt +WHILE_gt_cnt2 00100101 .. 1 ..... 0100 . 0 ..... 1 . ... @while_cnt +WHILE_gt_cnt4 00100101 .. 1 ..... 0110 . 0 ..... 1 . ... @while_cnt + ### SVE Integer Wide Immediate - Unpredicated Group # SVE broadcast floating-point immediate (unpredicated) diff --git a/target/arm/tcg/sve_helper.c b/target/arm/tcg/sve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve_helper.c +++ b/target/arm/tcg/sve_helper.c @@ -XXX,XX +XXX,XX @@ uint64_t HELPER(sve_cntp)(void *vn, void *vg, uint32_t pred_desc) return sum; } +/* C.f. Arm pseudocode EncodePredCount */ +static uint64_t encode_pred_count(uint32_t elements, uint32_t count, + uint32_t esz, bool invert) +{ + uint32_t pred; + + if (count == 0) { + return 0; + } + if (invert) { + count = elements - count; + } else if (count == elements) { + count = 0; + invert = true; + } + + pred = (count << 1) | 1; + pred <<= esz; + pred |= invert << 15; + + return pred; +} + /* C.f. Arm pseudocode PredCountTest */ static uint32_t pred_count_test(uint32_t elements, uint32_t count, bool invert) { @@ -XXX,XX +XXX,XX @@ uint32_t HELPER(sve_while2l)(void *vd, uint32_t count, uint32_t pred_desc) return pred_count_test(2 * oprbits, count, false); } +uint32_t HELPER(sve_whilecl)(void *vd, uint32_t count, uint32_t pred_desc) +{ + uint32_t pl = FIELD_EX32(pred_desc, PREDDESC, OPRSZ); + uint32_t esz = FIELD_EX32(pred_desc, PREDDESC, ESZ); + uint32_t scale = FIELD_EX32(pred_desc, PREDDESC, DATA); + uint32_t vl = pl * 8; + uint32_t elements = (vl >> esz) << scale; + ARMPredicateReg *d = vd; + + *d = (ARMPredicateReg) { + .p[0] = encode_pred_count(elements, count, esz, false) + }; + return pred_count_test(elements, count, false); +} + /* D must be cleared on entry. */ static void do_whileg(ARMPredicateReg *d, uint64_t esz_mask, uint32_t count, uint32_t oprbits) @@ -XXX,XX +XXX,XX @@ uint32_t HELPER(sve_while2g)(void *vd, uint32_t count, uint32_t pred_desc) return pred_count_test(2 * oprbits, count, true); } +uint32_t HELPER(sve_whilecg)(void *vd, uint32_t count, uint32_t pred_desc) +{ + uint32_t pl = FIELD_EX32(pred_desc, PREDDESC, OPRSZ); + uint32_t esz = FIELD_EX32(pred_desc, PREDDESC, ESZ); + uint32_t scale = FIELD_EX32(pred_desc, PREDDESC, DATA); + uint32_t vl = pl * 8; + uint32_t elements = (vl >> esz) << scale; + ARMPredicateReg *d = vd; + + *d = (ARMPredicateReg) { + .p[0] = encode_pred_count(elements, count, esz, true) + }; + return pred_count_test(elements, count, true); +} + /* Recursive reduction on a function; * C.f. the ARM ARM function ReducePredicated. * diff --git a/target/arm/tcg/translate-sve.c b/target/arm/tcg/translate-sve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sve.c +++ b/target/arm/tcg/translate-sve.c @@ -XXX,XX +XXX,XX @@ static bool trans_CTERM(DisasContext *s, arg_CTERM *a) typedef void gen_while_fn(TCGv_i32, TCGv_ptr, TCGv_i32, TCGv_i32); static bool do_WHILE(DisasContext *s, arg_while *a, - bool lt, int scale, gen_while_fn *fn) + bool lt, int scale, int data, gen_while_fn *fn) { TCGv_i64 op0, op1, t0, t1, tmax; TCGv_i32 t2; @@ -XXX,XX +XXX,XX @@ static bool do_WHILE(DisasContext *s, arg_while *a, desc = FIELD_DP32(desc, PREDDESC, OPRSZ, vsz / 8); desc = FIELD_DP32(desc, PREDDESC, ESZ, a->esz); + desc = FIELD_DP32(desc, PREDDESC, DATA, data); ptr = tcg_temp_new_ptr(); tcg_gen_addi_ptr(ptr, tcg_env, pred_full_reg_offset(s, a->rd)); @@ -XXX,XX +XXX,XX @@ static bool do_WHILE(DisasContext *s, arg_while *a, return true; } -TRANS_FEAT(WHILE_lt, aa64_sve, do_WHILE, a, true, 0, gen_helper_sve_whilel) -TRANS_FEAT(WHILE_gt, aa64_sve2, do_WHILE, a, false, 0, gen_helper_sve_whileg) +TRANS_FEAT(WHILE_lt, aa64_sve, do_WHILE, + a, true, 0, 0, gen_helper_sve_whilel) +TRANS_FEAT(WHILE_gt, aa64_sve2, do_WHILE, + a, false, 0, 0, gen_helper_sve_whileg) TRANS_FEAT(WHILE_lt_pair, aa64_sme2_or_sve2p1, do_WHILE, - a, true, 1, gen_helper_sve_while2l) + a, true, 1, 0, gen_helper_sve_while2l) TRANS_FEAT(WHILE_gt_pair, aa64_sme2_or_sve2p1, do_WHILE, - a, false, 1, gen_helper_sve_while2g) + a, false, 1, 0, gen_helper_sve_while2g) + +TRANS_FEAT(WHILE_lt_cnt2, aa64_sme2_or_sve2p1, do_WHILE, + a, true, 1, 1, gen_helper_sve_whilecl) +TRANS_FEAT(WHILE_lt_cnt4, aa64_sme2_or_sve2p1, do_WHILE, + a, true, 2, 2, gen_helper_sve_whilecl) +TRANS_FEAT(WHILE_gt_cnt2, aa64_sme2_or_sve2p1, do_WHILE, + a, false, 1, 1, gen_helper_sve_whilecg) +TRANS_FEAT(WHILE_gt_cnt4, aa64_sme2_or_sve2p1, do_WHILE, + a, false, 2, 2, gen_helper_sve_whilecg) static bool trans_WHILE_ptr(DisasContext *s, arg_WHILE_ptr *a) { -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-81-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/sve.decode | 1 + target/arm/tcg/translate-sve.c | 16 ++++++++++++++++ 2 files changed, 17 insertions(+) diff --git a/target/arm/tcg/sve.decode b/target/arm/tcg/sve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve.decode +++ b/target/arm/tcg/sve.decode @@ -XXX,XX +XXX,XX @@ PTEST 00100101 01 010000 11 pg:4 0 rn:4 0 0000 # SVE predicate initialize PTRUE 00100101 esz:2 01100 s:1 111000 pat:5 0 rd:4 +PTRUE_cnt 00100101 esz:2 1000000111100000010 ... rd=%pnd # SVE initialize FFR SETFFR 00100101 0010 1100 1001 0000 0000 0000 diff --git a/target/arm/tcg/translate-sve.c b/target/arm/tcg/translate-sve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sve.c +++ b/target/arm/tcg/translate-sve.c @@ -XXX,XX +XXX,XX @@ static bool do_predset(DisasContext *s, int esz, int rd, int pat, bool setflag) TRANS_FEAT(PTRUE, aa64_sve, do_predset, a->esz, a->rd, a->pat, a->s) +static bool trans_PTRUE_cnt(DisasContext *s, arg_PTRUE_cnt *a) +{ + if (!dc_isar_feature(aa64_sme2_or_sve2p1, s)) { + return false; + } + if (sve_access_check(s)) { + /* Canonical TRUE is 0 count, invert bit, plus element size. */ + int val = (1 << 15) | (1 << a->esz); + + /* Write val to the first uint64_t; clear all of the rest. */ + tcg_gen_gvec_dup_imm(MO_64, pred_full_reg_offset(s, a->rd), + 8, size_for_gvec(pred_full_reg_size(s)), val); + } + return true; +} + /* Note pat == 31 is #all, to set all elements. */ TRANS_FEAT_NONSTREAMING(SETFFR, aa64_sve, do_predset, 0, FFR_PRED_NUM, 31, false) -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-82-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper-sve.h | 25 ++++++++++++++++++ target/arm/tcg/sve.decode | 7 ++++++ target/arm/tcg/sve_helper.c | 46 ++++++++++++++++++++++++++++++++++ target/arm/tcg/translate-sve.c | 35 ++++++++++++++++++++++++++ 4 files changed, 113 insertions(+) diff --git a/target/arm/tcg/helper-sve.h b/target/arm/tcg/helper-sve.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-sve.h +++ b/target/arm/tcg/helper-sve.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_4(sve2_sqshlu_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_4(sve2_sqshlu_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_4(sve2_sqshlu_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_4(sve2_sqshlu_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) + +DEF_HELPER_FLAGS_4(sve2p1_addqv_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_addqv_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_addqv_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_addqv_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) + +DEF_HELPER_FLAGS_4(sve2p1_smaxqv_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_smaxqv_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_smaxqv_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_smaxqv_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) + +DEF_HELPER_FLAGS_4(sve2p1_sminqv_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_sminqv_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_sminqv_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_sminqv_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) + +DEF_HELPER_FLAGS_4(sve2p1_umaxqv_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_umaxqv_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_umaxqv_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_umaxqv_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) + +DEF_HELPER_FLAGS_4(sve2p1_uminqv_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_uminqv_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_uminqv_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_uminqv_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) diff --git a/target/arm/tcg/sve.decode b/target/arm/tcg/sve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve.decode +++ b/target/arm/tcg/sve.decode @@ -XXX,XX +XXX,XX @@ UMAXV 00000100 .. 001 001 001 ... ..... ..... @rd_pg_rn SMINV 00000100 .. 001 010 001 ... ..... ..... @rd_pg_rn UMINV 00000100 .. 001 011 001 ... ..... ..... @rd_pg_rn +# SVE2.1 segment reduction +ADDQV 00000100 .. 000 101 001 ... ..... ..... @rd_pg_rn +SMAXQV 00000100 .. 001 100 001 ... ..... ..... @rd_pg_rn +SMINQV 00000100 .. 001 110 001 ... ..... ..... @rd_pg_rn +UMAXQV 00000100 .. 001 101 001 ... ..... ..... @rd_pg_rn +UMINQV 00000100 .. 001 111 001 ... ..... ..... @rd_pg_rn + ### SVE Shift by Immediate - Predicated Group # SVE bitwise shift by immediate (predicated) diff --git a/target/arm/tcg/sve_helper.c b/target/arm/tcg/sve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve_helper.c +++ b/target/arm/tcg/sve_helper.c @@ -XXX,XX +XXX,XX @@ DO_VPZ_D(sve_uminv_d, uint64_t, uint64_t, -1, DO_MIN) #undef DO_VPZ #undef DO_VPZ_D +#define DO_VPQ(NAME, TYPE, H, INIT, OP) \ +void HELPER(NAME)(void *vd, void *vn, void *vg, uint32_t desc) \ +{ \ + TYPE tmp[16 / sizeof(TYPE)] = { [0 ... 16 / sizeof(TYPE) - 1] = INIT }; \ + TYPE *n = vn; uint16_t *g = vg; \ + uintptr_t oprsz = simd_oprsz(desc); \ + uintptr_t nseg = oprsz / 16, nsegelt = 16 / sizeof(TYPE); \ + for (uintptr_t s = 0; s < nseg; s++) { \ + uint16_t pg = g[H2(s)]; \ + for (uintptr_t e = 0; e < nsegelt; e++, pg >>= sizeof(TYPE)) { \ + if (pg & 1) { \ + tmp[e] = OP(tmp[H(e)], n[s * nsegelt + H(e)]); \ + } \ + } \ + } \ + memcpy(vd, tmp, 16); \ + clear_tail(vd, 16, simd_maxsz(desc)); \ +} + +DO_VPQ(sve2p1_addqv_b, uint8_t, H1, 0, DO_ADD) +DO_VPQ(sve2p1_addqv_h, uint16_t, H2, 0, DO_ADD) +DO_VPQ(sve2p1_addqv_s, uint32_t, H4, 0, DO_ADD) +DO_VPQ(sve2p1_addqv_d, uint64_t, H8, 0, DO_ADD) + +DO_VPQ(sve2p1_smaxqv_b, int8_t, H1, INT8_MIN, DO_MAX) +DO_VPQ(sve2p1_smaxqv_h, int16_t, H2, INT16_MIN, DO_MAX) +DO_VPQ(sve2p1_smaxqv_s, int32_t, H4, INT32_MIN, DO_MAX) +DO_VPQ(sve2p1_smaxqv_d, int64_t, H8, INT64_MIN, DO_MAX) + +DO_VPQ(sve2p1_sminqv_b, int8_t, H1, INT8_MAX, DO_MIN) +DO_VPQ(sve2p1_sminqv_h, int16_t, H2, INT16_MAX, DO_MIN) +DO_VPQ(sve2p1_sminqv_s, int32_t, H4, INT32_MAX, DO_MIN) +DO_VPQ(sve2p1_sminqv_d, int64_t, H8, INT64_MAX, DO_MIN) + +DO_VPQ(sve2p1_umaxqv_b, uint8_t, H1, 0, DO_MAX) +DO_VPQ(sve2p1_umaxqv_h, uint16_t, H2, 0, DO_MAX) +DO_VPQ(sve2p1_umaxqv_s, uint32_t, H4, 0, DO_MAX) +DO_VPQ(sve2p1_umaxqv_d, uint64_t, H8, 0, DO_MAX) + +DO_VPQ(sve2p1_uminqv_b, uint8_t, H1, -1, DO_MIN) +DO_VPQ(sve2p1_uminqv_h, uint16_t, H2, -1, DO_MIN) +DO_VPQ(sve2p1_uminqv_s, uint32_t, H4, -1, DO_MIN) +DO_VPQ(sve2p1_uminqv_d, uint64_t, H8, -1, DO_MIN) + +#undef DO_VPQ + /* Two vector operand, one scalar operand, unpredicated. */ #define DO_ZZI(NAME, TYPE, OP) \ void HELPER(NAME)(void *vd, void *vn, uint64_t s64, uint32_t desc) \ diff --git a/target/arm/tcg/translate-sve.c b/target/arm/tcg/translate-sve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sve.c +++ b/target/arm/tcg/translate-sve.c @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(SXTW, aa64_sve, gen_gvec_ool_arg_zpz, TRANS_FEAT(UXTW, aa64_sve, gen_gvec_ool_arg_zpz, a->esz == 3 ? gen_helper_sve_uxtw_d : NULL, a, 0) +static gen_helper_gvec_3 * const addqv_fns[4] = { + gen_helper_sve2p1_addqv_b, gen_helper_sve2p1_addqv_h, + gen_helper_sve2p1_addqv_s, gen_helper_sve2p1_addqv_d, +}; +TRANS_FEAT(ADDQV, aa64_sme2p1_or_sve2p1, + gen_gvec_ool_arg_zpz, addqv_fns[a->esz], a, 0) + +static gen_helper_gvec_3 * const smaxqv_fns[4] = { + gen_helper_sve2p1_smaxqv_b, gen_helper_sve2p1_smaxqv_h, + gen_helper_sve2p1_smaxqv_s, gen_helper_sve2p1_smaxqv_d, +}; +TRANS_FEAT(SMAXQV, aa64_sme2p1_or_sve2p1, + gen_gvec_ool_arg_zpz, smaxqv_fns[a->esz], a, 0) + +static gen_helper_gvec_3 * const sminqv_fns[4] = { + gen_helper_sve2p1_sminqv_b, gen_helper_sve2p1_sminqv_h, + gen_helper_sve2p1_sminqv_s, gen_helper_sve2p1_sminqv_d, +}; +TRANS_FEAT(SMINQV, aa64_sme2p1_or_sve2p1, + gen_gvec_ool_arg_zpz, sminqv_fns[a->esz], a, 0) + +static gen_helper_gvec_3 * const umaxqv_fns[4] = { + gen_helper_sve2p1_umaxqv_b, gen_helper_sve2p1_umaxqv_h, + gen_helper_sve2p1_umaxqv_s, gen_helper_sve2p1_umaxqv_d, +}; +TRANS_FEAT(UMAXQV, aa64_sme2p1_or_sve2p1, + gen_gvec_ool_arg_zpz, umaxqv_fns[a->esz], a, 0) + +static gen_helper_gvec_3 * const uminqv_fns[4] = { + gen_helper_sve2p1_uminqv_b, gen_helper_sve2p1_uminqv_h, + gen_helper_sve2p1_uminqv_s, gen_helper_sve2p1_uminqv_d, +}; +TRANS_FEAT(UMINQV, aa64_sme2p1_or_sve2p1, + gen_gvec_ool_arg_zpz, uminqv_fns[a->esz], a, 0) + /* *** SVE Integer Reduction Group */ -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-83-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper-sve.h | 2 + target/arm/tcg/vec_internal.h | 74 ++++++++++++++++++++++++++++++++++ target/arm/tcg/sve.decode | 6 +++ target/arm/tcg/sve_helper.c | 28 +++++++++++++ target/arm/tcg/translate-sve.c | 36 +++++++++++++++++ 5 files changed, 146 insertions(+) diff --git a/target/arm/tcg/helper-sve.h b/target/arm/tcg/helper-sve.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-sve.h +++ b/target/arm/tcg/helper-sve.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_4(sve2p1_uminqv_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_4(sve2p1_uminqv_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_4(sve2p1_uminqv_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_4(sve2p1_uminqv_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) + +DEF_HELPER_FLAGS_3(pext, TCG_CALL_NO_RWG, void, ptr, i32, i32) diff --git a/target/arm/tcg/vec_internal.h b/target/arm/tcg/vec_internal.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/vec_internal.h +++ b/target/arm/tcg/vec_internal.h @@ -XXX,XX +XXX,XX @@ bfloat16 helper_sme2_ah_fmin_b16(bfloat16 a, bfloat16 b, float_status *fpst); float32 sve_f16_to_f32(float16 f, float_status *fpst); float16 sve_f32_to_f16(float32 f, float_status *fpst); +/* + * Decode helper functions for predicate as counter. + */ + +typedef struct { + unsigned count; + unsigned lg2_stride; + bool invert; +} DecodeCounter; + +static inline DecodeCounter +decode_counter(unsigned png, unsigned vl, unsigned v_esz) +{ + DecodeCounter ret = { }; + + /* C.f. Arm pseudocode CounterToPredicate. */ + if (likely(png & 0xf)) { + unsigned p_esz = ctz32(png); + + /* + * maxbit = log2(pl(bits) * 4) + * = log2(vl(bytes) * 4) + * = log2(vl) + 2 + * maxbit_mask = ones<maxbit:0> + * = (1 << (maxbit + 1)) - 1 + * = (1 << (log2(vl) + 2 + 1)) - 1 + * = (1 << (log2(vl) + 3)) - 1 + * = (pow2ceil(vl) << 3) - 1 + */ + ret.count = png & (((unsigned)pow2ceil(vl) << 3) - 1); + ret.count >>= p_esz + 1; + + ret.invert = (png >> 15) & 1; + + /* + * The Arm pseudocode for CounterToPredicate expands the count to + * a set of bits, and then the operation proceeds as for the original + * interpretation of predicates as a set of bits. + * + * We can avoid the expansion by adjusting the count and supplying + * an element stride. + */ + if (unlikely(p_esz != v_esz)) { + if (p_esz < v_esz) { + /* + * For predicate esz < vector esz, the expanded predicate + * will have more bits set than will be consumed. + * Adjust the count down, rounding up. + * Consider p_esz = MO_8, v_esz = MO_64, count 14: + * The expanded predicate would be + * 0011 1111 1111 1111 + * The significant bits are + * ...1 ...1 ...1 ...1 + */ + unsigned shift = v_esz - p_esz; + unsigned trunc = ret.count >> shift; + ret.count = trunc + (ret.count != (trunc << shift)); + } else { + /* + * For predicate esz > vector esz, the expanded predicate + * will have bits set only at power-of-two multiples of + * the vector esz. Bits at other multiples will all be + * false. Adjust the count up, and supply the caller + * with a stride of elements to skip. + */ + unsigned shift = p_esz - v_esz; + ret.count <<= shift; + ret.lg2_stride = shift; + } + } + } + return ret; +} + #endif /* TARGET_ARM_VEC_INTERNAL_H */ diff --git a/target/arm/tcg/sve.decode b/target/arm/tcg/sve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve.decode +++ b/target/arm/tcg/sve.decode @@ -XXX,XX +XXX,XX @@ %rn_ax2 6:4 !function=times_2 %pnd 0:3 !function=plus_8 +%pnn 5:3 !function=plus_8 ########################################################################### # Named attribute sets. These are used to make nice(er) names @@ -XXX,XX +XXX,XX @@ WHILE_lt_cnt4 00100101 .. 1 ..... 0110 . 1 ..... 1 . ... @while_cnt WHILE_gt_cnt2 00100101 .. 1 ..... 0100 . 0 ..... 1 . ... @while_cnt WHILE_gt_cnt4 00100101 .. 1 ..... 0110 . 0 ..... 1 . ... @while_cnt +# SVE2.1 extract mask predicate from predicate-as-counter +&pext rd rn esz imm +PEXT_1 00100101 esz:2 1 00000 0111 00 imm:2 ... 1 rd:4 &pext rn=%pnn +PEXT_2 00100101 esz:2 1 00000 0111 010 imm:1 ... 1 rd:4 &pext rn=%pnn + ### SVE Integer Wide Immediate - Unpredicated Group # SVE broadcast floating-point immediate (unpredicated) diff --git a/target/arm/tcg/sve_helper.c b/target/arm/tcg/sve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve_helper.c +++ b/target/arm/tcg/sve_helper.c @@ -XXX,XX +XXX,XX @@ DO_FCVTLT(sve2_fcvtlt_sd, uint64_t, uint32_t, H1_8, H1_4, float32_to_float64) #undef DO_FCVTLT #undef DO_FCVTNT + +void HELPER(pext)(void *vd, uint32_t png, uint32_t desc) +{ + int pl = FIELD_EX32(desc, PREDDESC, OPRSZ); + int vl = pl * 8; + unsigned v_esz = FIELD_EX32(desc, PREDDESC, ESZ); + int part = FIELD_EX32(desc, PREDDESC, DATA); + DecodeCounter p = decode_counter(png, vl, v_esz); + uint64_t mask = pred_esz_masks[v_esz + p.lg2_stride]; + ARMPredicateReg *d = vd; + + /* + * Convert from element count to byte count and adjust + * for the portion of the 4*VL counter to be extracted. + */ + int b_count = (p.count << v_esz) - vl * part; + + memset(d, 0, sizeof(*d)); + if (p.invert) { + if (b_count <= 0) { + do_whilel(vd, mask, vl, vl); + } else if (b_count < vl) { + do_whileg(vd, mask, vl - b_count, vl); + } + } else if (b_count > 0) { + do_whilel(vd, mask, MIN(b_count, vl), vl); + } +} diff --git a/target/arm/tcg/translate-sve.c b/target/arm/tcg/translate-sve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sve.c +++ b/target/arm/tcg/translate-sve.c @@ -XXX,XX +XXX,XX @@ static bool trans_WHILE_ptr(DisasContext *s, arg_WHILE_ptr *a) return true; } +static bool do_pext(DisasContext *s, arg_pext *a, int n) +{ + TCGv_i32 t_png; + TCGv_ptr t_pd; + int pl; + + if (!sve_access_check(s)) { + return true; + } + + t_png = tcg_temp_new_i32(); + tcg_gen_ld16u_i32(t_png, tcg_env, + pred_full_reg_offset(s, a->rn) ^ + (HOST_BIG_ENDIAN ? 6 : 0)); + + t_pd = tcg_temp_new_ptr(); + pl = pred_full_reg_size(s); + + for (int i = 0; i < n; ++i) { + int rd = (a->rd + i) % 16; + int part = a->imm * n + i; + unsigned desc = 0; + + desc = FIELD_DP32(desc, PREDDESC, OPRSZ, pl); + desc = FIELD_DP32(desc, PREDDESC, ESZ, a->esz); + desc = FIELD_DP32(desc, PREDDESC, DATA, part); + + tcg_gen_addi_ptr(t_pd, tcg_env, pred_full_reg_offset(s, rd)); + gen_helper_pext(t_pd, t_png, tcg_constant_i32(desc)); + } + return true; +} + +TRANS_FEAT(PEXT_1, aa64_sme2_or_sve2p1, do_pext, a, 1) +TRANS_FEAT(PEXT_2, aa64_sme2_or_sve2p1, do_pext, a, 2) + /* *** SVE Integer Wide Immediate - Unpredicated Group */ -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-84-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper-sme.h | 5 + target/arm/tcg/sme.decode | 9 + target/arm/tcg/sme_helper.c | 317 +++++++++++++++++++++++++++++++++ target/arm/tcg/translate-sme.c | 31 ++++ 4 files changed, 362 insertions(+) diff --git a/target/arm/tcg/helper-sme.h b/target/arm/tcg/helper-sme.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-sme.h +++ b/target/arm/tcg/helper-sme.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_5(sme2_fclamp_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i3 DEF_HELPER_FLAGS_5(sme2_fclamp_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_5(sme2_fclamp_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_5(sme2_bfclamp, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, fpst, i32) + +DEF_HELPER_FLAGS_5(sme2_sel_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32, i32) +DEF_HELPER_FLAGS_5(sme2_sel_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32, i32) +DEF_HELPER_FLAGS_5(sme2_sel_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32, i32) +DEF_HELPER_FLAGS_5(sme2_sel_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32, i32) diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ UCLAMP 11000001 esz:2 1 zm:5 110001 zn:5 .... 1 \ UCLAMP 11000001 esz:2 1 zm:5 110011 zn:5 ...0 1 \ &zzz_en zd=%zd_ax4 n=4 +### SME2 Multi-vector SVE Select + +%sel_pg 10:3 !function=plus_8 + +SEL 11000001 esz:2 1 ....0 100 ... ....0 ....0 \ + n=2 zd=%zd_ax2 zn=%zn_ax2 zm=%zm_ax2 pg=%sel_pg +SEL 11000001 esz:2 1 ...01 100 ... ...00 ...00 \ + n=4 zd=%zd_ax4 zn=%zn_ax4 zm=%zm_ax4 pg=%sel_pg + ### SME Multiple Zero &zero_za rv off ngrp nvec diff --git a/target/arm/tcg/sme_helper.c b/target/arm/tcg/sme_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme_helper.c +++ b/target/arm/tcg/sme_helper.c @@ -XXX,XX +XXX,XX @@ FCLAMP(sme2_fclamp_d, float64, H8) FCLAMP(sme2_bfclamp, bfloat16, H2) #undef FCLAMP + +void HELPER(sme2_sel_b)(void *vd, void *vn, void *vm, + uint32_t png, uint32_t desc) +{ + int vl = simd_oprsz(desc); + int nreg = simd_data(desc); + int elements = vl / sizeof(uint8_t); + DecodeCounter p = decode_counter(png, vl, MO_8); + + if (p.lg2_stride == 0) { + if (p.invert) { + for (int r = 0; r < nreg; r++) { + uint8_t *d = vd + r * sizeof(ARMVectorReg); + uint8_t *n = vn + r * sizeof(ARMVectorReg); + uint8_t *m = vm + r * sizeof(ARMVectorReg); + int split = p.count - r * elements; + + if (split <= 0) { + memcpy(d, n, vl); /* all true */ + } else if (elements <= split) { + memcpy(d, m, vl); /* all false */ + } else { + for (int e = 0; e < split; e++) { + d[H1(e)] = m[H1(e)]; + } + for (int e = split; e < elements; e++) { + d[H1(e)] = n[H1(e)]; + } + } + } + } else { + for (int r = 0; r < nreg; r++) { + uint8_t *d = vd + r * sizeof(ARMVectorReg); + uint8_t *n = vn + r * sizeof(ARMVectorReg); + uint8_t *m = vm + r * sizeof(ARMVectorReg); + int split = p.count - r * elements; + + if (split <= 0) { + memcpy(d, m, vl); /* all false */ + } else if (elements <= split) { + memcpy(d, n, vl); /* all true */ + } else { + for (int e = 0; e < split; e++) { + d[H1(e)] = n[H1(e)]; + } + for (int e = split; e < elements; e++) { + d[H1(e)] = m[H1(e)]; + } + } + } + } + } else { + int estride = 1 << p.lg2_stride; + if (p.invert) { + for (int r = 0; r < nreg; r++) { + uint8_t *d = vd + r * sizeof(ARMVectorReg); + uint8_t *n = vn + r * sizeof(ARMVectorReg); + uint8_t *m = vm + r * sizeof(ARMVectorReg); + int split = p.count - r * elements; + int e = 0; + + for (; e < MIN(split, elements); e++) { + d[H1(e)] = m[H1(e)]; + } + for (; e < elements; e += estride) { + d[H1(e)] = n[H1(e)]; + for (int i = 1; i < estride; i++) { + d[H1(e + i)] = m[H1(e + i)]; + } + } + } + } else { + for (int r = 0; r < nreg; r++) { + uint8_t *d = vd + r * sizeof(ARMVectorReg); + uint8_t *n = vn + r * sizeof(ARMVectorReg); + uint8_t *m = vm + r * sizeof(ARMVectorReg); + int split = p.count - r * elements; + int e = 0; + + for (; e < MIN(split, elements); e += estride) { + d[H1(e)] = n[H1(e)]; + for (int i = 1; i < estride; i++) { + d[H1(e + i)] = m[H1(e + i)]; + } + } + for (; e < elements; e++) { + d[H1(e)] = m[H1(e)]; + } + } + } + } +} + +void HELPER(sme2_sel_h)(void *vd, void *vn, void *vm, + uint32_t png, uint32_t desc) +{ + int vl = simd_oprsz(desc); + int nreg = simd_data(desc); + int elements = vl / sizeof(uint16_t); + DecodeCounter p = decode_counter(png, vl, MO_16); + + if (p.lg2_stride == 0) { + if (p.invert) { + for (int r = 0; r < nreg; r++) { + uint16_t *d = vd + r * sizeof(ARMVectorReg); + uint16_t *n = vn + r * sizeof(ARMVectorReg); + uint16_t *m = vm + r * sizeof(ARMVectorReg); + int split = p.count - r * elements; + + if (split <= 0) { + memcpy(d, n, vl); /* all true */ + } else if (elements <= split) { + memcpy(d, m, vl); /* all false */ + } else { + for (int e = 0; e < split; e++) { + d[H2(e)] = m[H2(e)]; + } + for (int e = split; e < elements; e++) { + d[H2(e)] = n[H2(e)]; + } + } + } + } else { + for (int r = 0; r < nreg; r++) { + uint16_t *d = vd + r * sizeof(ARMVectorReg); + uint16_t *n = vn + r * sizeof(ARMVectorReg); + uint16_t *m = vm + r * sizeof(ARMVectorReg); + int split = p.count - r * elements; + + if (split <= 0) { + memcpy(d, m, vl); /* all false */ + } else if (elements <= split) { + memcpy(d, n, vl); /* all true */ + } else { + for (int e = 0; e < split; e++) { + d[H2(e)] = n[H2(e)]; + } + for (int e = split; e < elements; e++) { + d[H2(e)] = m[H2(e)]; + } + } + } + } + } else { + int estride = 1 << p.lg2_stride; + if (p.invert) { + for (int r = 0; r < nreg; r++) { + uint16_t *d = vd + r * sizeof(ARMVectorReg); + uint16_t *n = vn + r * sizeof(ARMVectorReg); + uint16_t *m = vm + r * sizeof(ARMVectorReg); + int split = p.count - r * elements; + int e = 0; + + for (; e < MIN(split, elements); e++) { + d[H2(e)] = m[H2(e)]; + } + for (; e < elements; e += estride) { + d[H2(e)] = n[H2(e)]; + for (int i = 1; i < estride; i++) { + d[H2(e + i)] = m[H2(e + i)]; + } + } + } + } else { + for (int r = 0; r < nreg; r++) { + uint16_t *d = vd + r * sizeof(ARMVectorReg); + uint16_t *n = vn + r * sizeof(ARMVectorReg); + uint16_t *m = vm + r * sizeof(ARMVectorReg); + int split = p.count - r * elements; + int e = 0; + + for (; e < MIN(split, elements); e += estride) { + d[H2(e)] = n[H2(e)]; + for (int i = 1; i < estride; i++) { + d[H2(e + i)] = m[H2(e + i)]; + } + } + for (; e < elements; e++) { + d[H2(e)] = m[H2(e)]; + } + } + } + } +} + +void HELPER(sme2_sel_s)(void *vd, void *vn, void *vm, + uint32_t png, uint32_t desc) +{ + int vl = simd_oprsz(desc); + int nreg = simd_data(desc); + int elements = vl / sizeof(uint32_t); + DecodeCounter p = decode_counter(png, vl, MO_32); + + if (p.lg2_stride == 0) { + if (p.invert) { + for (int r = 0; r < nreg; r++) { + uint32_t *d = vd + r * sizeof(ARMVectorReg); + uint32_t *n = vn + r * sizeof(ARMVectorReg); + uint32_t *m = vm + r * sizeof(ARMVectorReg); + int split = p.count - r * elements; + + if (split <= 0) { + memcpy(d, n, vl); /* all true */ + } else if (elements <= split) { + memcpy(d, m, vl); /* all false */ + } else { + for (int e = 0; e < split; e++) { + d[H4(e)] = m[H4(e)]; + } + for (int e = split; e < elements; e++) { + d[H4(e)] = n[H4(e)]; + } + } + } + } else { + for (int r = 0; r < nreg; r++) { + uint32_t *d = vd + r * sizeof(ARMVectorReg); + uint32_t *n = vn + r * sizeof(ARMVectorReg); + uint32_t *m = vm + r * sizeof(ARMVectorReg); + int split = p.count - r * elements; + + if (split <= 0) { + memcpy(d, m, vl); /* all false */ + } else if (elements <= split) { + memcpy(d, n, vl); /* all true */ + } else { + for (int e = 0; e < split; e++) { + d[H4(e)] = n[H4(e)]; + } + for (int e = split; e < elements; e++) { + d[H4(e)] = m[H4(e)]; + } + } + } + } + } else { + /* p.esz must be MO_64, so stride must be 2. */ + if (p.invert) { + for (int r = 0; r < nreg; r++) { + uint32_t *d = vd + r * sizeof(ARMVectorReg); + uint32_t *n = vn + r * sizeof(ARMVectorReg); + uint32_t *m = vm + r * sizeof(ARMVectorReg); + int split = p.count - r * elements; + int e = 0; + + for (; e < MIN(split, elements); e++) { + d[H4(e)] = m[H4(e)]; + } + for (; e < elements; e += 2) { + d[H4(e)] = n[H4(e)]; + d[H4(e + 1)] = m[H4(e + 1)]; + } + } + } else { + for (int r = 0; r < nreg; r++) { + uint32_t *d = vd + r * sizeof(ARMVectorReg); + uint32_t *n = vn + r * sizeof(ARMVectorReg); + uint32_t *m = vm + r * sizeof(ARMVectorReg); + int split = p.count - r * elements; + int e = 0; + + for (; e < MIN(split, elements); e += 2) { + d[H4(e)] = n[H4(e)]; + d[H4(e + 1)] = m[H4(e + 1)]; + } + for (; e < elements; e++) { + d[H4(e)] = m[H4(e)]; + } + } + } + } +} + +void HELPER(sme2_sel_d)(void *vd, void *vn, void *vm, + uint32_t png, uint32_t desc) +{ + int vl = simd_oprsz(desc); + int nreg = simd_data(desc); + int elements = vl / sizeof(uint64_t); + DecodeCounter p = decode_counter(png, vl, MO_64); + + if (p.invert) { + for (int r = 0; r < nreg; r++) { + uint64_t *d = vd + r * sizeof(ARMVectorReg); + uint64_t *n = vn + r * sizeof(ARMVectorReg); + uint64_t *m = vm + r * sizeof(ARMVectorReg); + int split = p.count - r * elements; + + if (split <= 0) { + memcpy(d, n, vl); /* all true */ + } else if (elements <= split) { + memcpy(d, m, vl); /* all false */ + } else { + memcpy(d, m, split * sizeof(uint64_t)); + memcpy(d + split, n + split, + (elements - split) * sizeof(uint64_t)); + } + } + } else { + for (int r = 0; r < nreg; r++) { + uint64_t *d = vd + r * sizeof(ARMVectorReg); + uint64_t *n = vn + r * sizeof(ARMVectorReg); + uint64_t *m = vm + r * sizeof(ARMVectorReg); + int split = p.count - r * elements; + + if (split <= 0) { + memcpy(d, m, vl); /* all false */ + } else if (elements <= split) { + memcpy(d, n, vl); /* all true */ + } else { + memcpy(d, n, split * sizeof(uint64_t)); + memcpy(d + split, m + split, + (elements - split) * sizeof(uint64_t)); + } + } + } +} diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ static gen_helper_gvec_3 * const uclamp_fns[] = { gen_helper_sme2_uclamp_d, }; TRANS(UCLAMP, do_clamp, a, uclamp_fns) + +static bool trans_SEL(DisasContext *s, arg_SEL *a) +{ + typedef void sme_sel_fn(TCGv_ptr, TCGv_ptr, TCGv_ptr, TCGv_i32, TCGv_i32); + static sme_sel_fn * const fns[4] = { + gen_helper_sme2_sel_b, gen_helper_sme2_sel_h, + gen_helper_sme2_sel_s, gen_helper_sme2_sel_d + }; + + if (!dc_isar_feature(aa64_sme2, s)) { + return false; + } + if (sme_sm_enabled_check(s)) { + int svl = streaming_vec_reg_size(s); + uint32_t desc = simd_desc(svl, svl, a->n); + TCGv_ptr t_d = tcg_temp_new_ptr(); + TCGv_ptr t_n = tcg_temp_new_ptr(); + TCGv_ptr t_m = tcg_temp_new_ptr(); + TCGv_i32 png = tcg_temp_new_i32(); + + tcg_gen_addi_ptr(t_d, tcg_env, vec_full_reg_offset(s, a->zd)); + tcg_gen_addi_ptr(t_n, tcg_env, vec_full_reg_offset(s, a->zn)); + tcg_gen_addi_ptr(t_m, tcg_env, vec_full_reg_offset(s, a->zm)); + + tcg_gen_ld16u_i32(png, tcg_env, pred_full_reg_offset(s, a->pg) + ^ (HOST_BIG_ENDIAN ? 6 : 0)); + + fns[a->esz](t_d, t_n, t_m, png, tcg_constant_i32(desc)); + } + return true; +} -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-85-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper-sve.h | 15 ++++++++++++ target/arm/tcg/sve.decode | 5 ++++ target/arm/tcg/sve_helper.c | 42 ++++++++++++++++++++++++++++++++++ target/arm/tcg/translate-sve.c | 3 +++ 4 files changed, 65 insertions(+) diff --git a/target/arm/tcg/helper-sve.h b/target/arm/tcg/helper-sve.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-sve.h +++ b/target/arm/tcg/helper-sve.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_4(sve2p1_uminqv_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_4(sve2p1_uminqv_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_3(pext, TCG_CALL_NO_RWG, void, ptr, i32, i32) + +DEF_HELPER_FLAGS_4(sve2p1_orqv_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_orqv_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_orqv_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_orqv_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) + +DEF_HELPER_FLAGS_4(sve2p1_eorqv_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_eorqv_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_eorqv_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_eorqv_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) + +DEF_HELPER_FLAGS_4(sve2p1_andqv_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_andqv_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_andqv_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_andqv_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) diff --git a/target/arm/tcg/sve.decode b/target/arm/tcg/sve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve.decode +++ b/target/arm/tcg/sve.decode @@ -XXX,XX +XXX,XX @@ ORV 00000100 .. 011 000 001 ... ..... ..... @rd_pg_rn EORV 00000100 .. 011 001 001 ... ..... ..... @rd_pg_rn ANDV 00000100 .. 011 010 001 ... ..... ..... @rd_pg_rn +# SVE2.1 bitwise logical reduction (quadwords) +ORQV 00000100 .. 011 100 001 ... ..... ..... @rd_pg_rn +EORQV 00000100 .. 011 101 001 ... ..... ..... @rd_pg_rn +ANDQV 00000100 .. 011 110 001 ... ..... ..... @rd_pg_rn + # SVE constructive prefix (predicated) MOVPRFX_z 00000100 .. 010 000 001 ... ..... ..... @rd_pg_rn MOVPRFX_m 00000100 .. 010 001 001 ... ..... ..... @rd_pg_rn diff --git a/target/arm/tcg/sve_helper.c b/target/arm/tcg/sve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve_helper.c +++ b/target/arm/tcg/sve_helper.c @@ -XXX,XX +XXX,XX @@ static inline uint64_t expand_pred_s(uint8_t byte) return word[byte & 0x11]; } +static inline uint64_t expand_pred_d(uint8_t byte) +{ + return -(uint64_t)(byte & 1); +} + #define LOGICAL_PPPP(NAME, FUNC) \ void HELPER(NAME)(void *vd, void *vn, void *vm, void *vg, uint32_t desc) \ { \ @@ -XXX,XX +XXX,XX @@ void HELPER(NAME)(void *vd, void *vn, void *vm, void *vg, uint32_t desc) \ #define DO_EOR(N, M) (N ^ M) #define DO_ORR(N, M) (N | M) #define DO_BIC(N, M) (N & ~M) +#define DO_ORC(N, M) (N | ~M) #define DO_ADD(N, M) (N + M) #define DO_SUB(N, M) (N - M) #define DO_MAX(N, M) ((N) >= (M) ? (N) : (M)) @@ -XXX,XX +XXX,XX @@ DO_ZZI(sve_umini_d, uint64_t, DO_MIN) #undef DO_ZZI +#define DO_LOGIC_QV(NAME, SUFF, INIT, VOP, POP) \ +void HELPER(NAME ## _ ## SUFF)(void *vd, void *vn, void *vg, uint32_t desc) \ +{ \ + unsigned seg = simd_oprsz(desc) / 16; \ + uint64_t r0 = INIT, r1 = INIT; \ + for (unsigned s = 0; s < seg; s++) { \ + uint64_t p0 = expand_pred_##SUFF(*(uint8_t *)(vg + H1(s * 2))); \ + uint64_t p1 = expand_pred_##SUFF(*(uint8_t *)(vg + H1(s * 2 + 1))); \ + uint64_t v0 = *(uint64_t *)(vn + s * 16); \ + uint64_t v1 = *(uint64_t *)(vn + s * 16 + 8); \ + v0 = POP(v0, p0), v1 = POP(v1, p1); \ + r0 = VOP(r0, v0), r1 = VOP(r1, v1); \ + } \ + *(uint64_t *)(vd + 0) = r0; \ + *(uint64_t *)(vd + 8) = r1; \ + clear_tail(vd, 16, simd_maxsz(desc)); \ +} + +DO_LOGIC_QV(sve2p1_orqv, b, 0, DO_ORR, DO_AND) +DO_LOGIC_QV(sve2p1_orqv, h, 0, DO_ORR, DO_AND) +DO_LOGIC_QV(sve2p1_orqv, s, 0, DO_ORR, DO_AND) +DO_LOGIC_QV(sve2p1_orqv, d, 0, DO_ORR, DO_AND) + +DO_LOGIC_QV(sve2p1_eorqv, b, 0, DO_EOR, DO_AND) +DO_LOGIC_QV(sve2p1_eorqv, h, 0, DO_EOR, DO_AND) +DO_LOGIC_QV(sve2p1_eorqv, s, 0, DO_EOR, DO_AND) +DO_LOGIC_QV(sve2p1_eorqv, d, 0, DO_EOR, DO_AND) + +DO_LOGIC_QV(sve2p1_andqv, b, -1, DO_AND, DO_ORC) +DO_LOGIC_QV(sve2p1_andqv, h, -1, DO_AND, DO_ORC) +DO_LOGIC_QV(sve2p1_andqv, s, -1, DO_AND, DO_ORC) +DO_LOGIC_QV(sve2p1_andqv, d, -1, DO_AND, DO_ORC) + +#undef DO_LOGIC_QV + #undef DO_AND #undef DO_ORR #undef DO_EOR #undef DO_BIC +#undef DO_ORC #undef DO_ADD #undef DO_SUB #undef DO_MAX diff --git a/target/arm/tcg/translate-sve.c b/target/arm/tcg/translate-sve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sve.c +++ b/target/arm/tcg/translate-sve.c @@ -XXX,XX +XXX,XX @@ DO_ZPZ(NOT_zpz, aa64_sve, sve_not_zpz) DO_ZPZ(ABS, aa64_sve, sve_abs) DO_ZPZ(NEG, aa64_sve, sve_neg) DO_ZPZ(RBIT, aa64_sve, sve_rbit) +DO_ZPZ(ORQV, aa64_sme2p1_or_sve2p1, sve2p1_orqv) +DO_ZPZ(EORQV, aa64_sme2p1_or_sve2p1, sve2p1_eorqv) +DO_ZPZ(ANDQV, aa64_sme2p1_or_sve2p1, sve2p1_andqv) static gen_helper_gvec_3 * const fabs_fns[4] = { NULL, gen_helper_sve_fabs_h, -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-86-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper-sve.h | 49 ++++++++++++++++++++++++ target/arm/tcg/sve.decode | 8 ++++ target/arm/tcg/sve_helper.c | 70 +++++++++++++++++++++------------- target/arm/tcg/translate-sve.c | 48 +++++++++++++++++++++++ 4 files changed, 148 insertions(+), 27 deletions(-) diff --git a/target/arm/tcg/helper-sve.h b/target/arm/tcg/helper-sve.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-sve.h +++ b/target/arm/tcg/helper-sve.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_4(sve_ah_fminv_s, TCG_CALL_NO_RWG, DEF_HELPER_FLAGS_4(sve_ah_fminv_d, TCG_CALL_NO_RWG, i64, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_5(sve2p1_faddqv_h, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_5(sve2p1_faddqv_s, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_5(sve2p1_faddqv_d, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, fpst, i32) + +DEF_HELPER_FLAGS_5(sve2p1_fmaxnmqv_h, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_5(sve2p1_fmaxnmqv_s, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_5(sve2p1_fmaxnmqv_d, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, fpst, i32) + +DEF_HELPER_FLAGS_5(sve2p1_fminnmqv_h, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_5(sve2p1_fminnmqv_s, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_5(sve2p1_fminnmqv_d, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, fpst, i32) + +DEF_HELPER_FLAGS_5(sve2p1_fmaxqv_h, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_5(sve2p1_fmaxqv_s, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_5(sve2p1_fmaxqv_d, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, fpst, i32) + +DEF_HELPER_FLAGS_5(sve2p1_fminqv_h, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_5(sve2p1_fminqv_s, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_5(sve2p1_fminqv_d, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, fpst, i32) + +DEF_HELPER_FLAGS_5(sve2p1_ah_fmaxqv_h, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_5(sve2p1_ah_fmaxqv_s, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_5(sve2p1_ah_fmaxqv_d, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, fpst, i32) + +DEF_HELPER_FLAGS_5(sve2p1_ah_fminqv_h, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_5(sve2p1_ah_fminqv_s, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_5(sve2p1_ah_fminqv_d, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, fpst, i32) + DEF_HELPER_FLAGS_5(sve_fadda_h, TCG_CALL_NO_RWG, i64, i64, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_5(sve_fadda_s, TCG_CALL_NO_RWG, diff --git a/target/arm/tcg/sve.decode b/target/arm/tcg/sve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve.decode +++ b/target/arm/tcg/sve.decode @@ -XXX,XX +XXX,XX @@ FMINNMV 01100101 .. 000 101 001 ... ..... ..... @rd_pg_rn FMAXV 01100101 .. 000 110 001 ... ..... ..... @rd_pg_rn FMINV 01100101 .. 000 111 001 ... ..... ..... @rd_pg_rn +### SVE FP recursive reduction (quadwords) + +FADDQV 01100100 .. 010 000 101 ... ..... ..... @rd_pg_rn +FMAXNMQV 01100100 .. 010 100 101 ... ..... ..... @rd_pg_rn +FMINNMQV 01100100 .. 010 101 101 ... ..... ..... @rd_pg_rn +FMAXQV 01100100 .. 010 110 101 ... ..... ..... @rd_pg_rn +FMINQV 01100100 .. 010 111 101 ... ..... ..... @rd_pg_rn + ## SVE Floating Point Unary Operations - Unpredicated Group FRECPE 01100101 .. 001 110 001100 ..... ..... @rd_rn diff --git a/target/arm/tcg/sve_helper.c b/target/arm/tcg/sve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve_helper.c +++ b/target/arm/tcg/sve_helper.c @@ -XXX,XX +XXX,XX @@ uint32_t HELPER(sve_whilecg)(void *vd, uint32_t count, uint32_t pred_desc) * The recursion is bounded to depth 7 (128 fp16 elements), so there's * little to gain with a more complex non-recursive form. */ -#define DO_REDUCE(NAME, TYPE, H, FUNC, IDENT) \ -static TYPE NAME##_reduce(TYPE *data, float_status *status, uintptr_t n) \ +#define DO_REDUCE(NAME, SUF, TYPE, H, FUNC, IDENT) \ +static TYPE FUNC##_reduce(TYPE *data, float_status *status, uintptr_t n) \ { \ if (n == 1) { \ return *data; \ } else { \ uintptr_t half = n / 2; \ - TYPE lo = NAME##_reduce(data, status, half); \ - TYPE hi = NAME##_reduce(data + half, status, half); \ + TYPE lo = FUNC##_reduce(data, status, half); \ + TYPE hi = FUNC##_reduce(data + half, status, half); \ return FUNC(lo, hi, status); \ } \ } \ -uint64_t HELPER(NAME)(void *vn, void *vg, float_status *s, uint32_t desc) \ +uint64_t helper_sve_##NAME##v_##SUF(void *vn, void *vg, \ + float_status *s, uint32_t desc) \ { \ uintptr_t i, oprsz = simd_oprsz(desc), maxsz = simd_data(desc); \ TYPE data[sizeof(ARMVectorReg) / sizeof(TYPE)]; \ @@ -XXX,XX +XXX,XX @@ uint64_t HELPER(NAME)(void *vn, void *vg, float_status *s, uint32_t desc) \ for (; i < maxsz; i += sizeof(TYPE)) { \ *(TYPE *)((void *)data + i) = IDENT; \ } \ - return NAME##_reduce(data, s, maxsz / sizeof(TYPE)); \ + return FUNC##_reduce(data, s, maxsz / sizeof(TYPE)); \ +} \ +void helper_sve2p1_##NAME##qv_##SUF(void *vd, void *vn, void *vg, \ + float_status *status, uint32_t desc) \ +{ \ + unsigned oprsz = simd_oprsz(desc), segments = oprsz / 16; \ + for (unsigned e = 0; e < 16; e += sizeof(TYPE)) { \ + TYPE data[ARM_MAX_VQ]; \ + for (unsigned s = 0; s < segments; s++) { \ + uint16_t pg = *(uint16_t *)(vg + H1_2(s * 2)); \ + TYPE nn = *(TYPE *)(vn + H(s * 16 + H(e))); \ + data[s] = (pg >> e) & 1 ? nn : IDENT; \ + } \ + *(TYPE *)(vd + H(e)) = FUNC##_reduce(data, status, segments); \ + } \ + clear_tail(vd, 16, simd_maxsz(desc)); \ } -DO_REDUCE(sve_faddv_h, float16, H1_2, float16_add, float16_zero) -DO_REDUCE(sve_faddv_s, float32, H1_4, float32_add, float32_zero) -DO_REDUCE(sve_faddv_d, float64, H1_8, float64_add, float64_zero) +DO_REDUCE(fadd,h, float16, H1_2, float16_add, float16_zero) +DO_REDUCE(fadd,s, float32, H1_4, float32_add, float32_zero) +DO_REDUCE(fadd,d, float64, H1_8, float64_add, float64_zero) /* Identity is floatN_default_nan, without the function call. */ -DO_REDUCE(sve_fminnmv_h, float16, H1_2, float16_minnum, 0x7E00) -DO_REDUCE(sve_fminnmv_s, float32, H1_4, float32_minnum, 0x7FC00000) -DO_REDUCE(sve_fminnmv_d, float64, H1_8, float64_minnum, 0x7FF8000000000000ULL) +DO_REDUCE(fminnm,h, float16, H1_2, float16_minnum, 0x7E00) +DO_REDUCE(fminnm,s, float32, H1_4, float32_minnum, 0x7FC00000) +DO_REDUCE(fminnm,d, float64, H1_8, float64_minnum, 0x7FF8000000000000ULL) -DO_REDUCE(sve_fmaxnmv_h, float16, H1_2, float16_maxnum, 0x7E00) -DO_REDUCE(sve_fmaxnmv_s, float32, H1_4, float32_maxnum, 0x7FC00000) -DO_REDUCE(sve_fmaxnmv_d, float64, H1_8, float64_maxnum, 0x7FF8000000000000ULL) +DO_REDUCE(fmaxnm,h, float16, H1_2, float16_maxnum, 0x7E00) +DO_REDUCE(fmaxnm,s, float32, H1_4, float32_maxnum, 0x7FC00000) +DO_REDUCE(fmaxnm,d, float64, H1_8, float64_maxnum, 0x7FF8000000000000ULL) -DO_REDUCE(sve_fminv_h, float16, H1_2, float16_min, float16_infinity) -DO_REDUCE(sve_fminv_s, float32, H1_4, float32_min, float32_infinity) -DO_REDUCE(sve_fminv_d, float64, H1_8, float64_min, float64_infinity) +DO_REDUCE(fmin,h, float16, H1_2, float16_min, float16_infinity) +DO_REDUCE(fmin,s, float32, H1_4, float32_min, float32_infinity) +DO_REDUCE(fmin,d, float64, H1_8, float64_min, float64_infinity) -DO_REDUCE(sve_fmaxv_h, float16, H1_2, float16_max, float16_chs(float16_infinity)) -DO_REDUCE(sve_fmaxv_s, float32, H1_4, float32_max, float32_chs(float32_infinity)) -DO_REDUCE(sve_fmaxv_d, float64, H1_8, float64_max, float64_chs(float64_infinity)) +DO_REDUCE(fmax,h, float16, H1_2, float16_max, float16_chs(float16_infinity)) +DO_REDUCE(fmax,s, float32, H1_4, float32_max, float32_chs(float32_infinity)) +DO_REDUCE(fmax,d, float64, H1_8, float64_max, float64_chs(float64_infinity)) -DO_REDUCE(sve_ah_fminv_h, float16, H1_2, helper_vfp_ah_minh, float16_infinity) -DO_REDUCE(sve_ah_fminv_s, float32, H1_4, helper_vfp_ah_mins, float32_infinity) -DO_REDUCE(sve_ah_fminv_d, float64, H1_8, helper_vfp_ah_mind, float64_infinity) +DO_REDUCE(ah_fmin,h, float16, H1_2, helper_vfp_ah_minh, float16_infinity) +DO_REDUCE(ah_fmin,s, float32, H1_4, helper_vfp_ah_mins, float32_infinity) +DO_REDUCE(ah_fmin,d, float64, H1_8, helper_vfp_ah_mind, float64_infinity) -DO_REDUCE(sve_ah_fmaxv_h, float16, H1_2, helper_vfp_ah_maxh, +DO_REDUCE(ah_fmax,h, float16, H1_2, helper_vfp_ah_maxh, float16_chs(float16_infinity)) -DO_REDUCE(sve_ah_fmaxv_s, float32, H1_4, helper_vfp_ah_maxs, +DO_REDUCE(ah_fmax,s, float32, H1_4, helper_vfp_ah_maxs, float32_chs(float32_infinity)) -DO_REDUCE(sve_ah_fmaxv_d, float64, H1_8, helper_vfp_ah_maxd, +DO_REDUCE(ah_fmax,d, float64, H1_8, helper_vfp_ah_maxd, float64_chs(float64_infinity)) #undef DO_REDUCE diff --git a/target/arm/tcg/translate-sve.c b/target/arm/tcg/translate-sve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sve.c +++ b/target/arm/tcg/translate-sve.c @@ -XXX,XX +XXX,XX @@ DO_VPZ_AH(FMAXV, fmaxv) #undef DO_VPZ +static gen_helper_gvec_3_ptr * const faddqv_fns[4] = { + NULL, gen_helper_sve2p1_faddqv_h, + gen_helper_sve2p1_faddqv_s, gen_helper_sve2p1_faddqv_d, +}; +TRANS_FEAT(FADDQV, aa64_sme2p1_or_sve2p1, gen_gvec_fpst_arg_zpz, + faddqv_fns[a->esz], a, 0, + a->esz == MO_16 ? FPST_A64_F16 : FPST_A64) + +static gen_helper_gvec_3_ptr * const fmaxnmqv_fns[4] = { + NULL, gen_helper_sve2p1_fmaxnmqv_h, + gen_helper_sve2p1_fmaxnmqv_s, gen_helper_sve2p1_fmaxnmqv_d, +}; +TRANS_FEAT(FMAXNMQV, aa64_sme2p1_or_sve2p1, gen_gvec_fpst_arg_zpz, + fmaxnmqv_fns[a->esz], a, 0, + a->esz == MO_16 ? FPST_A64_F16 : FPST_A64) + +static gen_helper_gvec_3_ptr * const fminnmqv_fns[4] = { + NULL, gen_helper_sve2p1_fminnmqv_h, + gen_helper_sve2p1_fminnmqv_s, gen_helper_sve2p1_fminnmqv_d, +}; +TRANS_FEAT(FMINNMQV, aa64_sme2p1_or_sve2p1, gen_gvec_fpst_arg_zpz, + fminnmqv_fns[a->esz], a, 0, + a->esz == MO_16 ? FPST_A64_F16 : FPST_A64) + +static gen_helper_gvec_3_ptr * const fmaxqv_fns[4] = { + NULL, gen_helper_sve2p1_fmaxqv_h, + gen_helper_sve2p1_fmaxqv_s, gen_helper_sve2p1_fmaxqv_d, +}; +static gen_helper_gvec_3_ptr * const fmaxqv_ah_fns[4] = { + NULL, gen_helper_sve2p1_ah_fmaxqv_h, + gen_helper_sve2p1_ah_fmaxqv_s, gen_helper_sve2p1_ah_fmaxqv_d, +}; +TRANS_FEAT(FMAXQV, aa64_sme2p1_or_sve2p1, gen_gvec_fpst_arg_zpz, + (s->fpcr_ah ? fmaxqv_fns : fmaxqv_ah_fns)[a->esz], a, 0, + a->esz == MO_16 ? FPST_A64_F16 : FPST_A64) + +static gen_helper_gvec_3_ptr * const fminqv_fns[4] = { + NULL, gen_helper_sve2p1_fminqv_h, + gen_helper_sve2p1_fminqv_s, gen_helper_sve2p1_fminqv_d, +}; +static gen_helper_gvec_3_ptr * const fminqv_ah_fns[4] = { + NULL, gen_helper_sve2p1_ah_fminqv_h, + gen_helper_sve2p1_ah_fminqv_s, gen_helper_sve2p1_ah_fminqv_d, +}; +TRANS_FEAT(FMINQV, aa64_sme2p1_or_sve2p1, gen_gvec_fpst_arg_zpz, + (s->fpcr_ah ? fminqv_fns : fminqv_ah_fns)[a->esz], a, 0, + a->esz == MO_16 ? FPST_A64_F16 : FPST_A64) + /* *** SVE Floating Point Unary Operations - Unpredicated Group */ -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-87-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/sve.decode | 6 ++++++ target/arm/tcg/translate-sve.c | 30 ++++++++++++++++++++++++++++++ 2 files changed, 36 insertions(+) diff --git a/target/arm/tcg/sve.decode b/target/arm/tcg/sve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve.decode +++ b/target/arm/tcg/sve.decode @@ -XXX,XX +XXX,XX @@ FMLSLT_zzzw 01100100 10 1 ..... 10 1 00 1 ..... ..... @rda_rn_rm_ex esz=2 BFMLALB_zzzw 01100100 11 1 ..... 10 0 00 0 ..... ..... @rda_rn_rm_ex esz=2 BFMLALT_zzzw 01100100 11 1 ..... 10 0 00 1 ..... ..... @rda_rn_rm_ex esz=2 +BFMLSLB_zzzw 01100100 11 1 ..... 10 1 00 0 ..... ..... @rda_rn_rm_ex esz=2 +BFMLSLT_zzzw 01100100 11 1 ..... 10 1 00 1 ..... ..... @rda_rn_rm_ex esz=2 ### SVE2 floating-point dot-product FDOT_zzzz 01100100 00 1 ..... 10 0 00 0 ..... ..... @rda_rn_rm_ex esz=2 BFDOT_zzzz 01100100 01 1 ..... 10 0 00 0 ..... ..... @rda_rn_rm_ex esz=2 ### SVE2 floating-point multiply-add long (indexed) + FMLALB_zzxw 01100100 10 1 ..... 0100.0 ..... ..... @rrxr_3a esz=2 FMLALT_zzxw 01100100 10 1 ..... 0100.1 ..... ..... @rrxr_3a esz=2 FMLSLB_zzxw 01100100 10 1 ..... 0110.0 ..... ..... @rrxr_3a esz=2 FMLSLT_zzxw 01100100 10 1 ..... 0110.1 ..... ..... @rrxr_3a esz=2 + BFMLALB_zzxw 01100100 11 1 ..... 0100.0 ..... ..... @rrxr_3a esz=2 BFMLALT_zzxw 01100100 11 1 ..... 0100.1 ..... ..... @rrxr_3a esz=2 +BFMLSLB_zzxw 01100100 11 1 ..... 0110.0 ..... ..... @rrxr_3a esz=2 +BFMLSLT_zzxw 01100100 11 1 ..... 0110.1 ..... ..... @rrxr_3a esz=2 ### SVE2 floating-point dot-product (indexed) diff --git a/target/arm/tcg/translate-sve.c b/target/arm/tcg/translate-sve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sve.c +++ b/target/arm/tcg/translate-sve.c @@ -XXX,XX +XXX,XX @@ static bool do_BFMLAL_zzxw(DisasContext *s, arg_rrxr_esz *a, bool sel) TRANS_FEAT(BFMLALB_zzxw, aa64_sve_bf16, do_BFMLAL_zzxw, a, false) TRANS_FEAT(BFMLALT_zzxw, aa64_sve_bf16, do_BFMLAL_zzxw, a, true) +static bool do_BFMLSL_zzzw(DisasContext *s, arg_rrrr_esz *a, bool sel) +{ + if (s->fpcr_ah) { + return gen_gvec_fpst_zzzz(s, gen_helper_gvec_ah_bfmlsl, + a->rd, a->rn, a->rm, a->ra, sel, FPST_AH); + } else { + return gen_gvec_fpst_zzzz(s, gen_helper_gvec_bfmlsl, + a->rd, a->rn, a->rm, a->ra, sel, FPST_A64); + } +} + +TRANS_FEAT(BFMLSLB_zzzw, aa64_sme2_or_sve2p1, do_BFMLSL_zzzw, a, false) +TRANS_FEAT(BFMLSLT_zzzw, aa64_sme2_or_sve2p1, do_BFMLSL_zzzw, a, true) + +static bool do_BFMLSL_zzxw(DisasContext *s, arg_rrxr_esz *a, bool sel) +{ + if (s->fpcr_ah) { + return gen_gvec_fpst_zzzz(s, gen_helper_gvec_ah_bfmlsl_idx, + a->rd, a->rn, a->rm, a->ra, + (a->index << 1) | sel, FPST_AH); + } else { + return gen_gvec_fpst_zzzz(s, gen_helper_gvec_bfmlsl_idx, + a->rd, a->rn, a->rm, a->ra, + (a->index << 1) | sel, FPST_A64); + } +} + +TRANS_FEAT(BFMLSLB_zzxw, aa64_sme2_or_sve2p1, do_BFMLSL_zzxw, a, false) +TRANS_FEAT(BFMLSLT_zzxw, aa64_sme2_or_sve2p1, do_BFMLSL_zzxw, a, true) + static bool trans_PSEL(DisasContext *s, arg_psel *a) { int vl = vec_full_reg_size(s); -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-88-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper-sve.h | 1 + target/arm/tcg/sve.decode | 3 ++- target/arm/tcg/sve_helper.c | 21 +++++++++++++++++++++ target/arm/tcg/translate-sve.c | 30 ++++++++++++++++++++++++++++++ 4 files changed, 54 insertions(+), 1 deletion(-) diff --git a/target/arm/tcg/helper-sve.h b/target/arm/tcg/helper-sve.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-sve.h +++ b/target/arm/tcg/helper-sve.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_4(sve_brkn, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_4(sve_brkns, TCG_CALL_NO_RWG, i32, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_3(sve_cntp, TCG_CALL_NO_RWG, i64, ptr, ptr, i32) +DEF_HELPER_FLAGS_2(sve2p1_cntp_c, TCG_CALL_NO_RWG_SE, i64, i32, i32) DEF_HELPER_FLAGS_3(sve_whilel, TCG_CALL_NO_RWG, i32, ptr, i32, i32) DEF_HELPER_FLAGS_3(sve_whileg, TCG_CALL_NO_RWG, i32, ptr, i32, i32) diff --git a/target/arm/tcg/sve.decode b/target/arm/tcg/sve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve.decode +++ b/target/arm/tcg/sve.decode @@ -XXX,XX +XXX,XX @@ BRKN 00100101 0. 01100001 .... 0 .... 0 .... @pd_pg_pn_s ### SVE Predicate Count Group # SVE predicate count -CNTP 00100101 .. 100 000 10 .... 0 .... ..... @rd_pg4_pn +CNTP 00100101 .. 100 000 10 .... 0 .... ..... @rd_pg4_pn +CNTP_c 00100101 esz:2 100 000 10 000 vl:1 1 rn:4 rd:5 # SVE inc/dec register by predicate count INCDECP_r 00100101 .. 10110 d:1 10001 00 .... ..... @incdec_pred u=1 diff --git a/target/arm/tcg/sve_helper.c b/target/arm/tcg/sve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve_helper.c +++ b/target/arm/tcg/sve_helper.c @@ -XXX,XX +XXX,XX @@ uint64_t HELPER(sve_cntp)(void *vn, void *vg, uint32_t pred_desc) return sum; } +uint64_t HELPER(sve2p1_cntp_c)(uint32_t png, uint32_t desc) +{ + int pl = FIELD_EX32(desc, PREDDESC, OPRSZ); + int vl = pl * 8; + unsigned v_esz = FIELD_EX32(desc, PREDDESC, ESZ); + int lg2_width = FIELD_EX32(desc, PREDDESC, DATA) + 1; + DecodeCounter p = decode_counter(png, vl, v_esz); + unsigned maxelem = (vl << lg2_width) >> v_esz; + unsigned count = p.count; + + if (p.invert) { + if (count >= maxelem) { + return 0; + } + count = maxelem - count; + } else { + count = MIN(count, maxelem); + } + return count >> p.lg2_stride; +} + /* C.f. Arm pseudocode EncodePredCount */ static uint64_t encode_pred_count(uint32_t elements, uint32_t count, uint32_t esz, bool invert) diff --git a/target/arm/tcg/translate-sve.c b/target/arm/tcg/translate-sve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sve.c +++ b/target/arm/tcg/translate-sve.c @@ -XXX,XX +XXX,XX @@ static bool trans_CNTP(DisasContext *s, arg_CNTP *a) return true; } +static bool trans_CNTP_c(DisasContext *s, arg_CNTP_c *a) +{ + TCGv_i32 t_png; + uint32_t desc = 0; + + if (dc_isar_feature(aa64_sve2p1, s)) { + if (!sve_access_check(s)) { + return true; + } + } else if (dc_isar_feature(aa64_sme2, s)) { + if (!sme_sm_enabled_check(s)) { + return true; + } + } else { + return false; + } + + t_png = tcg_temp_new_i32(); + tcg_gen_ld16u_i32(t_png, tcg_env, + pred_full_reg_offset(s, a->rn) ^ + (HOST_BIG_ENDIAN ? 6 : 0)); + + desc = FIELD_DP32(desc, PREDDESC, OPRSZ, pred_full_reg_size(s)); + desc = FIELD_DP32(desc, PREDDESC, ESZ, a->esz); + desc = FIELD_DP32(desc, PREDDESC, DATA, a->vl); + + gen_helper_sve2p1_cntp_c(cpu_reg(s, a->rd), t_png, tcg_constant_i32(desc)); + return true; +} + static bool trans_INCDECP_r(DisasContext *s, arg_incdec_pred *a) { if (!dc_isar_feature(aa64_sve, s)) { -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-89-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/sve.decode | 6 ++++++ target/arm/tcg/translate-sve.c | 21 +++++++++++++++++++++ 2 files changed, 27 insertions(+) diff --git a/target/arm/tcg/sve.decode b/target/arm/tcg/sve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve.decode +++ b/target/arm/tcg/sve.decode @@ -XXX,XX +XXX,XX @@ DUP_s 00000101 .. 1 00000 001110 ..... ..... @rd_rn DUP_x 00000101 .. 1 ..... 001000 rn:5 rd:5 \ &rri imm=%imm7_22_16 +# SVE Permute Vector - one source quadwords +DUPQ 00000101 001 imm:4 1 001001 rn:5 rd:5 &rri_esz esz=0 +DUPQ 00000101 001 imm:3 10 001001 rn:5 rd:5 &rri_esz esz=1 +DUPQ 00000101 001 imm:2 100 001001 rn:5 rd:5 &rri_esz esz=2 +DUPQ 00000101 001 imm:1 1000 001001 rn:5 rd:5 &rri_esz esz=3 + # SVE insert SIMD&FP scalar register INSR_f 00000101 .. 1 10100 001110 ..... ..... @rdn_rm diff --git a/target/arm/tcg/translate-sve.c b/target/arm/tcg/translate-sve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sve.c +++ b/target/arm/tcg/translate-sve.c @@ -XXX,XX +XXX,XX @@ static bool trans_DUP_x(DisasContext *s, arg_DUP_x *a) return true; } +static bool trans_DUPQ(DisasContext *s, arg_DUPQ *a) +{ + unsigned vl, dofs, nofs; + + if (!dc_isar_feature(aa64_sme2p1_or_sve2p1, s)) { + return false; + } + if (!sve_access_check(s)) { + return true; + } + + vl = vec_full_reg_size(s); + dofs = vec_full_reg_offset(s, a->rd); + nofs = vec_reg_offset(s, a->rn, a->imm, a->esz); + + for (unsigned i = 0; i < vl; i += 16) { + tcg_gen_gvec_dup_mem(a->esz, dofs + i, nofs + i, 16, 16); + } + return true; +} + static void do_insr_i64(DisasContext *s, arg_rrr_esz *a, TCGv_i64 val) { typedef void gen_insr(TCGv_ptr, TCGv_ptr, TCGv_i64, TCGv_i32); -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-90-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/sve.decode | 2 ++ target/arm/tcg/translate-sve.c | 49 ++++++++++++++++++++++++++++++++++ 2 files changed, 51 insertions(+) diff --git a/target/arm/tcg/sve.decode b/target/arm/tcg/sve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve.decode +++ b/target/arm/tcg/sve.decode @@ -XXX,XX +XXX,XX @@ DUPQ 00000101 001 imm:3 10 001001 rn:5 rd:5 &rri_esz esz=1 DUPQ 00000101 001 imm:2 100 001001 rn:5 rd:5 &rri_esz esz=2 DUPQ 00000101 001 imm:1 1000 001001 rn:5 rd:5 &rri_esz esz=3 +EXTQ 00000101 0110 imm:4 001001 rn:5 rd:5 &rri + # SVE insert SIMD&FP scalar register INSR_f 00000101 .. 1 10100 001110 ..... ..... @rdn_rm diff --git a/target/arm/tcg/translate-sve.c b/target/arm/tcg/translate-sve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sve.c +++ b/target/arm/tcg/translate-sve.c @@ -XXX,XX +XXX,XX @@ static bool do_EXT(DisasContext *s, int rd, int rn, int rm, int imm) TRANS_FEAT(EXT, aa64_sve, do_EXT, a->rd, a->rn, a->rm, a->imm) TRANS_FEAT(EXT_sve2, aa64_sve2, do_EXT, a->rd, a->rn, (a->rn + 1) % 32, a->imm) +static bool trans_EXTQ(DisasContext *s, arg_EXTQ *a) +{ + unsigned vl, dofs, sofs0, sofs1, sofs2, imm; + + if (!dc_isar_feature(aa64_sme2p1_or_sve2p1, s)) { + return false; + } + if (!sve_access_check(s)) { + return true; + } + + imm = a->imm; + if (imm == 0) { + /* So far we never optimize Zdn with MOVPRFX, so zd = zn is a nop. */ + return true; + } + + vl = vec_full_reg_size(s); + dofs = vec_full_reg_offset(s, a->rd); + sofs2 = vec_full_reg_offset(s, a->rn); + + if (imm & 8) { + sofs0 = dofs + 8; + sofs1 = sofs2; + sofs2 += 8; + } else { + sofs0 = dofs; + sofs1 = dofs + 8; + } + imm = (imm & 7) << 3; + + for (unsigned i = 0; i < vl; i += 16) { + TCGv_i64 s0 = tcg_temp_new_i64(); + TCGv_i64 s1 = tcg_temp_new_i64(); + TCGv_i64 s2 = tcg_temp_new_i64(); + + tcg_gen_ld_i64(s0, tcg_env, sofs0 + i); + tcg_gen_ld_i64(s1, tcg_env, sofs1 + i); + tcg_gen_ld_i64(s2, tcg_env, sofs2 + i); + + tcg_gen_extract2_i64(s0, s0, s1, imm); + tcg_gen_extract2_i64(s1, s1, s2, imm); + + tcg_gen_st_i64(s0, tcg_env, dofs + i); + tcg_gen_st_i64(s1, tcg_env, dofs + i + 8); + } + return true; +} + /* *** SVE Permute - Unpredicated Group */ -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-91-richard.henderson@linaro.org Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper-sve.h | 8 +++ target/arm/tcg/vec_internal.h | 34 ++++++++++++ target/arm/tcg/sve.decode | 17 ++++++ target/arm/tcg/sve_helper.c | 50 +++++++++++++++++ target/arm/tcg/translate-sve.c | 98 ++++++++++++++++++++++++++++++++++ 5 files changed, 207 insertions(+) diff --git a/target/arm/tcg/helper-sve.h b/target/arm/tcg/helper-sve.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-sve.h +++ b/target/arm/tcg/helper-sve.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_4(sve2p1_andqv_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_4(sve2p1_andqv_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_4(sve2p1_andqv_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_4(sve2p1_andqv_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) + +DEF_HELPER_FLAGS_3(pmov_pv_h, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(pmov_pv_s, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(pmov_pv_d, TCG_CALL_NO_RWG, void, ptr, ptr, i32) + +DEF_HELPER_FLAGS_3(pmov_vp_h, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(pmov_vp_s, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(pmov_vp_d, TCG_CALL_NO_RWG, void, ptr, ptr, i32) diff --git a/target/arm/tcg/vec_internal.h b/target/arm/tcg/vec_internal.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/vec_internal.h +++ b/target/arm/tcg/vec_internal.h @@ -XXX,XX +XXX,XX @@ decode_counter(unsigned png, unsigned vl, unsigned v_esz) return ret; } +/* Extract @len bits from an array of uint64_t at offset @pos bits. */ +static inline uint64_t extractn(uint64_t *p, unsigned pos, unsigned len) +{ + uint64_t x; + + p += pos / 64; + pos = pos % 64; + + x = p[0]; + if (pos + len > 64) { + x = (x >> pos) | (p[1] << (-pos & 63)); + pos = 0; + } + return extract64(x, pos, len); +} + +/* Deposit @len bits into an array of uint64_t at offset @pos bits. */ +static inline void depositn(uint64_t *p, unsigned pos, + unsigned len, uint64_t val) +{ + p += pos / 64; + pos = pos % 64; + + if (pos + len <= 64) { + p[0] = deposit64(p[0], pos, len, val); + } else { + unsigned len0 = 64 - pos; + unsigned len1 = len - len0; + + p[0] = deposit64(p[0], pos, len0, val); + p[1] = deposit64(p[1], 0, len1, val >> len0); + } +} + #endif /* TARGET_ARM_VEC_INTERNAL_H */ diff --git a/target/arm/tcg/sve.decode b/target/arm/tcg/sve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve.decode +++ b/target/arm/tcg/sve.decode @@ -XXX,XX +XXX,XX @@ %size_23 23:2 %dtype_23_13 23:2 13:2 %index3_22_19 22:1 19:2 +%index3_22_17 22:1 17:2 %index3_19_11 19:2 11:1 %index2_20_11 20:1 11:1 @@ -XXX,XX +XXX,XX @@ INSR_r 00000101 .. 1 00100 001110 ..... ..... @rdn_rm # SVE reverse vector elements REV_v 00000101 .. 1 11000 001110 ..... ..... @rd_rn +# SVE move predicate to/from vector + +PMOV_pv 00000101 00 101 01 0001110 rn:5 0 rd:4 \ + &rri_esz esz=0 imm=0 +PMOV_pv 00000101 00 101 1 imm:1 0001110 rn:5 0 rd:4 &rri_esz esz=1 +PMOV_pv 00000101 01 101 imm:2 0001110 rn:5 0 rd:4 &rri_esz esz=2 +PMOV_pv 00000101 1. 101 .. 0001110 rn:5 0 rd:4 \ + &rri_esz esz=3 imm=%index3_22_17 + +PMOV_vp 00000101 00 101 01 1001110 0 rn:4 rd:5 \ + &rri_esz esz=0 imm=0 +PMOV_vp 00000101 00 101 1 imm:1 1001110 0 rn:4 rd:5 &rri_esz esz=1 +PMOV_vp 00000101 01 101 imm:2 1001110 0 rn:4 rd:5 &rri_esz esz=2 +PMOV_vp 00000101 1. 101 .. 1001110 0 rn:4 rd:5 \ + &rri_esz esz=3 imm=%index3_22_17 + # SVE vector table lookup TBL 00000101 .. 1 ..... 001100 ..... ..... @rd_rn_rm diff --git a/target/arm/tcg/sve_helper.c b/target/arm/tcg/sve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve_helper.c +++ b/target/arm/tcg/sve_helper.c @@ -XXX,XX +XXX,XX @@ void HELPER(sve_rev_d)(void *vd, void *vn, uint32_t desc) } } +/* + * TODO: This could use half_shuffle64 and similar bit tricks to + * expand blocks of bits at once. + */ +#define DO_PMOV_PV(NAME, ESIZE) \ +void HELPER(NAME)(void *vd, void *vs, uint32_t desc) \ +{ \ + unsigned vl = simd_oprsz(desc); \ + unsigned idx = simd_data(desc); \ + unsigned elements = vl / ESIZE; \ + ARMPredicateReg *d = vd; \ + ARMVectorReg *s = vs; \ + memset(d, 0, sizeof(*d)); \ + for (unsigned e = 0; e < elements; ++e) { \ + depositn(d->p, e * ESIZE, 1, extractn(s->d, elements * idx + e, 1)); \ + } \ +} + +DO_PMOV_PV(pmov_pv_h, 2) +DO_PMOV_PV(pmov_pv_s, 4) +DO_PMOV_PV(pmov_pv_d, 8) + +#undef DO_PMOV_PV + +/* + * TODO: This could use half_unshuffle64 and similar bit tricks to + * compress blocks of bits at once. + */ +#define DO_PMOV_VP(NAME, ESIZE) \ +void HELPER(NAME)(void *vd, void *vs, uint32_t desc) \ +{ \ + unsigned vl = simd_oprsz(desc); \ + unsigned idx = simd_data(desc); \ + unsigned elements = vl / ESIZE; \ + ARMVectorReg *d = vd; \ + ARMPredicateReg *s = vs; \ + if (idx == 0) { \ + memset(d, 0, vl); \ + } \ + for (unsigned e = 0; e < elements; ++e) { \ + depositn(d->d, elements * idx + e, 1, extractn(s->p, e * ESIZE, 1)); \ + } \ +} + +DO_PMOV_VP(pmov_vp_h, 2) +DO_PMOV_VP(pmov_vp_s, 4) +DO_PMOV_VP(pmov_vp_d, 8) + +#undef DO_PMOV_VP + typedef void tb_impl_fn(void *, void *, void *, void *, uintptr_t, bool); static inline void do_tbl1(void *vd, void *vn, void *vm, uint32_t desc, diff --git a/target/arm/tcg/translate-sve.c b/target/arm/tcg/translate-sve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sve.c +++ b/target/arm/tcg/translate-sve.c @@ -XXX,XX +XXX,XX @@ static gen_helper_gvec_3 * const tbx_fns[4] = { }; TRANS_FEAT(TBX, aa64_sve2, gen_gvec_ool_arg_zzz, tbx_fns[a->esz], a, 0) +static bool trans_PMOV_pv(DisasContext *s, arg_PMOV_pv *a) +{ + static gen_helper_gvec_2 * const fns[4] = { + NULL, gen_helper_pmov_pv_h, + gen_helper_pmov_pv_s, gen_helper_pmov_pv_d + }; + unsigned vl, pl, vofs, pofs; + TCGv_i64 tmp; + + if (!dc_isar_feature(aa64_sme2p1_or_sve2p1, s)) { + return false; + } + if (!sve_access_check(s)) { + return true; + } + + vl = vec_full_reg_size(s); + if (a->esz != MO_8) { + tcg_gen_gvec_2_ool(pred_full_reg_offset(s, a->rd), + vec_full_reg_offset(s, a->rn), + vl, vl, a->imm, fns[a->esz]); + return true; + } + + /* + * Copy the low PL bytes from vector Zn, zero-extending to a + * multiple of 8 bytes, so that Pd is properly cleared. + */ + + pl = vl / 8; + pofs = pred_full_reg_offset(s, a->rd); + vofs = vec_full_reg_offset(s, a->rn); + + QEMU_BUILD_BUG_ON(sizeof(ARMPredicateReg) != 32); + for (unsigned i = 32; i >= 8; i >>= 1) { + if (pl & i) { + tcg_gen_gvec_mov(MO_64, pofs, vofs, i, i); + pofs += i; + vofs += i; + } + } + switch (pl & 7) { + case 0: + return true; + case 2: + tmp = tcg_temp_new_i64(); + tcg_gen_ld16u_i64(tmp, tcg_env, vofs + (HOST_BIG_ENDIAN ? 6 : 0)); + break; + case 4: + tmp = tcg_temp_new_i64(); + tcg_gen_ld32u_i64(tmp, tcg_env, vofs + (HOST_BIG_ENDIAN ? 4 : 0)); + break; + case 6: + tmp = tcg_temp_new_i64(); + tcg_gen_ld_i64(tmp, tcg_env, vofs); + tcg_gen_extract_i64(tmp, tmp, 0, 48); + break; + default: + g_assert_not_reached(); + } + tcg_gen_st_i64(tmp, tcg_env, pofs); + return true; +} + +static bool trans_PMOV_vp(DisasContext *s, arg_PMOV_pv *a) +{ + static gen_helper_gvec_2 * const fns[4] = { + NULL, gen_helper_pmov_vp_h, + gen_helper_pmov_vp_s, gen_helper_pmov_vp_d + }; + unsigned vl; + + if (!dc_isar_feature(aa64_sme2p1_or_sve2p1, s)) { + return false; + } + if (!sve_access_check(s)) { + return true; + } + + vl = vec_full_reg_size(s); + + if (a->esz == MO_8) { + /* + * The low PL bytes are copied from Pn to Zd unchanged. + * We know that the unused portion of Pn is zero, and + * that imm == 0, so the balance of Zd must be zeroed. + */ + tcg_gen_gvec_mov(MO_64, vec_full_reg_offset(s, a->rd), + pred_full_reg_offset(s, a->rn), + size_for_gvec(vl / 8), vl); + } else { + tcg_gen_gvec_2_ool(vec_full_reg_offset(s, a->rd), + pred_full_reg_offset(s, a->rn), + vl, vl, a->imm, fns[a->esz]); + } + return true; +} + static bool trans_UNPK(DisasContext *s, arg_UNPK *a) { static gen_helper_gvec_2 * const fns[4][2] = { -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-92-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper-sve.h | 10 ++++++++++ target/arm/tcg/sve.decode | 6 ++++++ target/arm/tcg/sve_helper.c | 29 +++++++++++++++++++++++++++++ target/arm/tcg/translate-sve.c | 19 ++++++++++++++++++- 4 files changed, 63 insertions(+), 1 deletion(-) diff --git a/target/arm/tcg/helper-sve.h b/target/arm/tcg/helper-sve.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-sve.h +++ b/target/arm/tcg/helper-sve.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_4(sve_zip_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_4(sve_zip_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_4(sve2_zip_q, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_zipq_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_zipq_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_zipq_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_zipq_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) + DEF_HELPER_FLAGS_4(sve_uzp_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_4(sve_uzp_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_4(sve_uzp_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_4(sve_uzp_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_4(sve2_uzp_q, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_uzpq_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_uzpq_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_uzpq_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_uzpq_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) + DEF_HELPER_FLAGS_4(sve_trn_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_4(sve_trn_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_4(sve_trn_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) diff --git a/target/arm/tcg/sve.decode b/target/arm/tcg/sve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve.decode +++ b/target/arm/tcg/sve.decode @@ -XXX,XX +XXX,XX @@ UZP2_q 00000101 10 1 ..... 000 011 ..... ..... @rd_rn_rm_e0 TRN1_q 00000101 10 1 ..... 000 110 ..... ..... @rd_rn_rm_e0 TRN2_q 00000101 10 1 ..... 000 111 ..... ..... @rd_rn_rm_e0 +# SVE2.1 permute vector elements (quadwords) +ZIPQ1 01000100 .. 0 ..... 111 000 ..... ..... @rd_rn_rm +ZIPQ2 01000100 .. 0 ..... 111 001 ..... ..... @rd_rn_rm +UZPQ1 01000100 .. 0 ..... 111 010 ..... ..... @rd_rn_rm +UZPQ2 01000100 .. 0 ..... 111 011 ..... ..... @rd_rn_rm + ### SVE Permute - Predicated Group # SVE compress active elements diff --git a/target/arm/tcg/sve_helper.c b/target/arm/tcg/sve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve_helper.c +++ b/target/arm/tcg/sve_helper.c @@ -XXX,XX +XXX,XX @@ DO_UZP(sve_uzp_s, uint32_t, H1_4) DO_UZP(sve_uzp_d, uint64_t, H1_8) DO_UZP(sve2_uzp_q, Int128, ) +typedef void perseg_zzz_fn(void *vd, void *vn, void *vm, uint32_t desc); + +static void do_perseg_zzz(void *vd, void *vn, void *vm, + uint32_t desc, perseg_zzz_fn *fn) +{ + intptr_t oprsz = simd_oprsz(desc); + + desc = simd_desc(16, 16, simd_data(desc)); + for (intptr_t i = 0; i < oprsz; i += 16) { + fn(vd + i, vn + i, vm + i, desc); + } +} + +#define DO_PERSEG_ZZZ(NAME, FUNC) \ + void HELPER(NAME)(void *vd, void *vn, void *vm, uint32_t desc) \ + { do_perseg_zzz(vd, vn, vm, desc, FUNC); } + +DO_PERSEG_ZZZ(sve2p1_uzpq_b, helper_sve_uzp_b) +DO_PERSEG_ZZZ(sve2p1_uzpq_h, helper_sve_uzp_h) +DO_PERSEG_ZZZ(sve2p1_uzpq_s, helper_sve_uzp_s) +DO_PERSEG_ZZZ(sve2p1_uzpq_d, helper_sve_uzp_d) + +DO_PERSEG_ZZZ(sve2p1_zipq_b, helper_sve_zip_b) +DO_PERSEG_ZZZ(sve2p1_zipq_h, helper_sve_zip_h) +DO_PERSEG_ZZZ(sve2p1_zipq_s, helper_sve_zip_s) +DO_PERSEG_ZZZ(sve2p1_zipq_d, helper_sve_zip_d) + +#undef DO_PERSEG_ZZZ + #define DO_TRN(NAME, TYPE, H) \ void HELPER(NAME)(void *vd, void *vn, void *vm, uint32_t desc) \ { \ diff --git a/target/arm/tcg/translate-sve.c b/target/arm/tcg/translate-sve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sve.c +++ b/target/arm/tcg/translate-sve.c @@ -XXX,XX +XXX,XX @@ TRANS_FEAT_NONSTREAMING(ZIP2_q, aa64_sve_f64mm, do_interleave_q, gen_helper_sve2_zip_q, a, QEMU_ALIGN_DOWN(vec_full_reg_size(s), 32) / 2) +static gen_helper_gvec_3 * const zipq_fns[4] = { + gen_helper_sve2p1_zipq_b, gen_helper_sve2p1_zipq_h, + gen_helper_sve2p1_zipq_s, gen_helper_sve2p1_zipq_d, +}; +TRANS_FEAT(ZIPQ1, aa64_sme2p1_or_sve2p1, gen_gvec_ool_arg_zzz, + zipq_fns[a->esz], a, 0) +TRANS_FEAT(ZIPQ2, aa64_sme2p1_or_sve2p1, gen_gvec_ool_arg_zzz, + zipq_fns[a->esz], a, 16 / 2) + static gen_helper_gvec_3 * const uzp_fns[4] = { gen_helper_sve_uzp_b, gen_helper_sve_uzp_h, gen_helper_sve_uzp_s, gen_helper_sve_uzp_d, }; - TRANS_FEAT(UZP1_z, aa64_sve, gen_gvec_ool_arg_zzz, uzp_fns[a->esz], a, 0) TRANS_FEAT(UZP2_z, aa64_sve, gen_gvec_ool_arg_zzz, @@ -XXX,XX +XXX,XX @@ TRANS_FEAT_NONSTREAMING(UZP1_q, aa64_sve_f64mm, do_interleave_q, TRANS_FEAT_NONSTREAMING(UZP2_q, aa64_sve_f64mm, do_interleave_q, gen_helper_sve2_uzp_q, a, 16) +static gen_helper_gvec_3 * const uzpq_fns[4] = { + gen_helper_sve2p1_uzpq_b, gen_helper_sve2p1_uzpq_h, + gen_helper_sve2p1_uzpq_s, gen_helper_sve2p1_uzpq_d, +}; +TRANS_FEAT(UZPQ1, aa64_sme2p1_or_sve2p1, gen_gvec_ool_arg_zzz, + uzpq_fns[a->esz], a, 0) +TRANS_FEAT(UZPQ2, aa64_sme2p1_or_sve2p1, gen_gvec_ool_arg_zzz, + uzpq_fns[a->esz], a, 1 << a->esz) + static gen_helper_gvec_3 * const trn_fns[4] = { gen_helper_sve_trn_b, gen_helper_sve_trn_h, gen_helper_sve_trn_s, gen_helper_sve_trn_d, -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-93-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper-sve.h | 10 ++++++++++ target/arm/tcg/sve.decode | 3 +++ target/arm/tcg/sve_helper.c | 10 ++++++++++ target/arm/tcg/translate-sve.c | 14 ++++++++++++++ 4 files changed, 37 insertions(+) diff --git a/target/arm/tcg/helper-sve.h b/target/arm/tcg/helper-sve.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-sve.h +++ b/target/arm/tcg/helper-sve.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_5(sve2_tbl_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_5(sve2_tbl_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_5(sve2_tbl_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_tblq_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_tblq_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_tblq_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_tblq_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) + DEF_HELPER_FLAGS_4(sve2_tbx_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_4(sve2_tbx_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_4(sve2_tbx_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_4(sve2_tbx_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_tbxq_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_tbxq_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_tbxq_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(sve2p1_tbxq_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) + DEF_HELPER_FLAGS_3(sve_sunpk_h, TCG_CALL_NO_RWG, void, ptr, ptr, i32) DEF_HELPER_FLAGS_3(sve_sunpk_s, TCG_CALL_NO_RWG, void, ptr, ptr, i32) DEF_HELPER_FLAGS_3(sve_sunpk_d, TCG_CALL_NO_RWG, void, ptr, ptr, i32) diff --git a/target/arm/tcg/sve.decode b/target/arm/tcg/sve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve.decode +++ b/target/arm/tcg/sve.decode @@ -XXX,XX +XXX,XX @@ ZIPQ2 01000100 .. 0 ..... 111 001 ..... ..... @rd_rn_rm UZPQ1 01000100 .. 0 ..... 111 010 ..... ..... @rd_rn_rm UZPQ2 01000100 .. 0 ..... 111 011 ..... ..... @rd_rn_rm +TBLQ 01000100 .. 0 ..... 111 110 ..... ..... @rd_rn_rm +TBXQ 00000101 .. 1 ..... 001 101 ..... ..... @rd_rn_rm + ### SVE Permute - Predicated Group # SVE compress active elements diff --git a/target/arm/tcg/sve_helper.c b/target/arm/tcg/sve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve_helper.c +++ b/target/arm/tcg/sve_helper.c @@ -XXX,XX +XXX,XX @@ DO_PERSEG_ZZZ(sve2p1_zipq_h, helper_sve_zip_h) DO_PERSEG_ZZZ(sve2p1_zipq_s, helper_sve_zip_s) DO_PERSEG_ZZZ(sve2p1_zipq_d, helper_sve_zip_d) +DO_PERSEG_ZZZ(sve2p1_tblq_b, helper_sve_tbl_b) +DO_PERSEG_ZZZ(sve2p1_tblq_h, helper_sve_tbl_h) +DO_PERSEG_ZZZ(sve2p1_tblq_s, helper_sve_tbl_s) +DO_PERSEG_ZZZ(sve2p1_tblq_d, helper_sve_tbl_d) + +DO_PERSEG_ZZZ(sve2p1_tbxq_b, helper_sve2_tbx_b) +DO_PERSEG_ZZZ(sve2p1_tbxq_h, helper_sve2_tbx_h) +DO_PERSEG_ZZZ(sve2p1_tbxq_s, helper_sve2_tbx_s) +DO_PERSEG_ZZZ(sve2p1_tbxq_d, helper_sve2_tbx_d) + #undef DO_PERSEG_ZZZ #define DO_TRN(NAME, TYPE, H) \ diff --git a/target/arm/tcg/translate-sve.c b/target/arm/tcg/translate-sve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sve.c +++ b/target/arm/tcg/translate-sve.c @@ -XXX,XX +XXX,XX @@ static gen_helper_gvec_4 * const sve2_tbl_fns[4] = { TRANS_FEAT(TBL_sve2, aa64_sve2, gen_gvec_ool_zzzz, sve2_tbl_fns[a->esz], a->rd, a->rn, (a->rn + 1) % 32, a->rm, 0) +static gen_helper_gvec_3 * const tblq_fns[4] = { + gen_helper_sve2p1_tblq_b, gen_helper_sve2p1_tblq_h, + gen_helper_sve2p1_tblq_s, gen_helper_sve2p1_tblq_d +}; +TRANS_FEAT(TBLQ, aa64_sme2p1_or_sve2p1, gen_gvec_ool_arg_zzz, + tblq_fns[a->esz], a, 0) + static gen_helper_gvec_3 * const tbx_fns[4] = { gen_helper_sve2_tbx_b, gen_helper_sve2_tbx_h, gen_helper_sve2_tbx_s, gen_helper_sve2_tbx_d }; TRANS_FEAT(TBX, aa64_sve2, gen_gvec_ool_arg_zzz, tbx_fns[a->esz], a, 0) +static gen_helper_gvec_3 * const tbxq_fns[4] = { + gen_helper_sve2p1_tbxq_b, gen_helper_sve2p1_tbxq_h, + gen_helper_sve2p1_tbxq_s, gen_helper_sve2p1_tbxq_d +}; +TRANS_FEAT(TBXQ, aa64_sme2p1_or_sve2p1, gen_gvec_ool_arg_zzz, + tbxq_fns[a->esz], a, 0) + static bool trans_PMOV_pv(DisasContext *s, arg_PMOV_pv *a) { static gen_helper_gvec_2 * const fns[4] = { -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Implement the SVE2p1 consecutive register LD1/ST1, and the SME2 strided register LD1/ST1. Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-94-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper-sve.h | 16 ++ target/arm/tcg/sve.decode | 50 ++++ target/arm/tcg/sve_helper.c | 493 +++++++++++++++++++++++++++++++++ target/arm/tcg/translate-sve.c | 103 +++++++ 4 files changed, 662 insertions(+) diff --git a/target/arm/tcg/helper-sve.h b/target/arm/tcg/helper-sve.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-sve.h +++ b/target/arm/tcg/helper-sve.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_3(pmov_pv_d, TCG_CALL_NO_RWG, void, ptr, ptr, i32) DEF_HELPER_FLAGS_3(pmov_vp_h, TCG_CALL_NO_RWG, void, ptr, ptr, i32) DEF_HELPER_FLAGS_3(pmov_vp_s, TCG_CALL_NO_RWG, void, ptr, ptr, i32) DEF_HELPER_FLAGS_3(pmov_vp_d, TCG_CALL_NO_RWG, void, ptr, ptr, i32) + +DEF_HELPER_FLAGS_5(sve2p1_ld1bb_c, TCG_CALL_NO_WG, void, env, ptr, tl, i32, i32) +DEF_HELPER_FLAGS_5(sve2p1_ld1hh_le_c, TCG_CALL_NO_WG, void, env, ptr, tl, i32, i32) +DEF_HELPER_FLAGS_5(sve2p1_ld1hh_be_c, TCG_CALL_NO_WG, void, env, ptr, tl, i32, i32) +DEF_HELPER_FLAGS_5(sve2p1_ld1ss_le_c, TCG_CALL_NO_WG, void, env, ptr, tl, i32, i32) +DEF_HELPER_FLAGS_5(sve2p1_ld1ss_be_c, TCG_CALL_NO_WG, void, env, ptr, tl, i32, i32) +DEF_HELPER_FLAGS_5(sve2p1_ld1dd_le_c, TCG_CALL_NO_WG, void, env, ptr, tl, i32, i32) +DEF_HELPER_FLAGS_5(sve2p1_ld1dd_be_c, TCG_CALL_NO_WG, void, env, ptr, tl, i32, i32) + +DEF_HELPER_FLAGS_5(sve2p1_st1bb_c, TCG_CALL_NO_WG, void, env, ptr, tl, i32, i32) +DEF_HELPER_FLAGS_5(sve2p1_st1hh_le_c, TCG_CALL_NO_WG, void, env, ptr, tl, i32, i32) +DEF_HELPER_FLAGS_5(sve2p1_st1hh_be_c, TCG_CALL_NO_WG, void, env, ptr, tl, i32, i32) +DEF_HELPER_FLAGS_5(sve2p1_st1ss_le_c, TCG_CALL_NO_WG, void, env, ptr, tl, i32, i32) +DEF_HELPER_FLAGS_5(sve2p1_st1ss_be_c, TCG_CALL_NO_WG, void, env, ptr, tl, i32, i32) +DEF_HELPER_FLAGS_5(sve2p1_st1dd_le_c, TCG_CALL_NO_WG, void, env, ptr, tl, i32, i32) +DEF_HELPER_FLAGS_5(sve2p1_st1dd_be_c, TCG_CALL_NO_WG, void, env, ptr, tl, i32, i32) diff --git a/target/arm/tcg/sve.decode b/target/arm/tcg/sve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve.decode +++ b/target/arm/tcg/sve.decode @@ -XXX,XX +XXX,XX @@ SCLAMP 01000100 .. 0 ..... 110000 ..... ..... @rda_rn_rm UCLAMP 01000100 .. 0 ..... 110001 ..... ..... @rda_rn_rm FCLAMP 01100100 .. 1 ..... 001001 ..... ..... @rda_rn_rm + +### SVE2p1 multi-vec contiguous load + +&zcrr_ldst rd png rn rm esz nreg +&zcri_ldst rd png rn imm esz nreg +%png 10:3 !function=plus_8 +%zd_ax2 1:4 !function=times_2 +%zd_ax4 2:3 !function=times_4 + +LD1_zcrr 10100000000 rm:5 0 esz:2 ... rn:5 .... - \ + &zcrr_ldst %png rd=%zd_ax2 nreg=2 +LD1_zcrr 10100000000 rm:5 1 esz:2 ... rn:5 ... 0- \ + &zcrr_ldst %png rd=%zd_ax4 nreg=4 + +ST1_zcrr 10100000001 rm:5 0 esz:2 ... rn:5 .... - \ + &zcrr_ldst %png rd=%zd_ax2 nreg=2 +ST1_zcrr 10100000001 rm:5 1 esz:2 ... rn:5 ... 0- \ + &zcrr_ldst %png rd=%zd_ax4 nreg=4 + +LD1_zcri 101000000100 imm:s4 0 esz:2 ... rn:5 .... - \ + &zcri_ldst %png rd=%zd_ax2 nreg=2 +LD1_zcri 101000000100 imm:s4 1 esz:2 ... rn:5 ... 0- \ + &zcri_ldst %png rd=%zd_ax4 nreg=4 + +ST1_zcri 101000000110 imm:s4 0 esz:2 ... rn:5 .... - \ + &zcri_ldst %png rd=%zd_ax2 nreg=2 +ST1_zcri 101000000110 imm:s4 1 esz:2 ... rn:5 ... 0- \ + &zcri_ldst %png rd=%zd_ax4 nreg=4 + +# Note: N bit and 0 bit (for nreg4) still mashed in rd. +# This is handled within gen_ldst_c(). +LD1_zcrr_stride 10100001000 rm:5 0 esz:2 ... rn:5 rd:5 \ + &zcrr_ldst %png nreg=2 +LD1_zcrr_stride 10100001000 rm:5 1 esz:2 ... rn:5 rd:5 \ + &zcrr_ldst %png nreg=4 + +ST1_zcrr_stride 10100001001 rm:5 0 esz:2 ... rn:5 rd:5 \ + &zcrr_ldst %png nreg=2 +ST1_zcrr_stride 10100001001 rm:5 1 esz:2 ... rn:5 rd:5 \ + &zcrr_ldst %png nreg=4 + +LD1_zcri_stride 101000010100 imm:s4 0 esz:2 ... rn:5 rd:5 \ + &zcri_ldst %png nreg=2 +LD1_zcri_stride 101000010100 imm:s4 1 esz:2 ... rn:5 rd:5 \ + &zcri_ldst %png nreg=4 + +ST1_zcri_stride 101000010110 imm:s4 0 esz:2 ... rn:5 rd:5 \ + &zcri_ldst %png nreg=2 +ST1_zcri_stride 101000010110 imm:s4 1 esz:2 ... rn:5 rd:5 \ + &zcri_ldst %png nreg=4 diff --git a/target/arm/tcg/sve_helper.c b/target/arm/tcg/sve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve_helper.c +++ b/target/arm/tcg/sve_helper.c @@ -XXX,XX +XXX,XX @@ DO_ST1_ZPZ_D(dd_be, zd, MO_64) #undef DO_ST1_ZPZ_S #undef DO_ST1_ZPZ_D +/* + * SVE2.1 consecutive register load/store + */ + +static unsigned sve2p1_cont_ldst_elements(SVEContLdSt *info, vaddr addr, + uint32_t png, intptr_t reg_max, + int N, int v_esz) +{ + const int esize = 1 << v_esz; + intptr_t reg_off_first = -1, reg_off_last = -1, reg_off_split; + DecodeCounter p = decode_counter(png, reg_max, v_esz); + unsigned b_count = p.count << v_esz; + unsigned b_stride = 1 << (v_esz + p.lg2_stride); + intptr_t page_split; + + /* Set all of the element indices to -1, and the TLB data to 0. */ + memset(info, -1, offsetof(SVEContLdSt, page)); + memset(info->page, 0, sizeof(info->page)); + + if (p.invert) { + if (b_count >= reg_max * N) { + return 0; + } + reg_off_first = b_count; + reg_off_last = reg_max * N - b_stride; + } else { + if (b_count == 0) { + return 0; + } + reg_off_first = 0; + reg_off_last = MIN(b_count - esize, reg_max * N - b_stride); + } + + info->reg_off_first[0] = reg_off_first; + info->mem_off_first[0] = reg_off_first; + + page_split = -(addr | TARGET_PAGE_MASK); + if (reg_off_last + esize <= page_split || reg_off_first >= page_split) { + /* The entire operation fits within a single page. */ + info->reg_off_last[0] = reg_off_last; + return b_stride; + } + + info->page_split = page_split; + reg_off_split = ROUND_DOWN(page_split, esize); + + /* + * This is the last full element on the first page, but it is not + * necessarily active. If there is no full element, i.e. the first + * active element is the one that's split, this value remains -1. + * It is useful as iteration bounds. + */ + if (reg_off_split != 0) { + info->reg_off_last[0] = ROUND_DOWN(reg_off_split - esize, b_stride); + } + + /* Determine if an unaligned element spans the pages. */ + if (page_split & (esize - 1)) { + /* It is helpful to know if the split element is active. */ + if ((reg_off_split & (b_stride - 1)) == 0) { + info->reg_off_split = reg_off_split; + info->mem_off_split = reg_off_split; + } + reg_off_split += esize; + } + + /* + * We do want the first active element on the second page, because + * this may affect the address reported in an exception. + */ + reg_off_split = ROUND_UP(reg_off_split, b_stride); + if (reg_off_split <= reg_off_last) { + info->reg_off_first[1] = reg_off_split; + info->mem_off_first[1] = reg_off_split; + info->reg_off_last[1] = reg_off_last; + } + return b_stride; +} + +static void sve2p1_cont_ldst_watchpoints(SVEContLdSt *info, CPUARMState *env, + target_ulong addr, unsigned estride, + int esize, int wp_access, uintptr_t ra) +{ +#ifndef CONFIG_USER_ONLY + intptr_t count_off, count_last; + int flags0 = info->page[0].flags; + int flags1 = info->page[1].flags; + + if (likely(!((flags0 | flags1) & TLB_WATCHPOINT))) { + return; + } + + /* Indicate that watchpoints are handled. */ + info->page[0].flags = flags0 & ~TLB_WATCHPOINT; + info->page[1].flags = flags1 & ~TLB_WATCHPOINT; + + if (flags0 & TLB_WATCHPOINT) { + count_off = info->reg_off_first[0]; + count_last = info->reg_off_split; + if (count_last < 0) { + count_last = info->reg_off_last[0]; + } + do { + cpu_check_watchpoint(env_cpu(env), addr + count_off, + esize, info->page[0].attrs, wp_access, ra); + count_off += estride; + } while (count_off <= count_last); + } + + count_off = info->reg_off_first[1]; + if ((flags1 & TLB_WATCHPOINT) && count_off >= 0) { + count_last = info->reg_off_last[1]; + do { + cpu_check_watchpoint(env_cpu(env), addr + count_off, + esize, info->page[1].attrs, + wp_access, ra); + count_off += estride; + } while (count_off <= count_last); + } +#endif +} + +static void sve2p1_cont_ldst_mte_check(SVEContLdSt *info, CPUARMState *env, + target_ulong addr, unsigned estride, + int esize, uint32_t mtedesc, + uintptr_t ra) +{ + intptr_t count_off, count_last; + + /* + * TODO: estride is always a small power of two, <= 8. + * Manipulate the stride within the loops such that + * - first iteration hits addr + off, as required, + * - second iteration hits ALIGN_UP(addr, 16), + * - other iterations advance addr by 16. + * This will minimize the probing to once per MTE granule. + */ + + /* Process the page only if MemAttr == Tagged. */ + if (info->page[0].tagged) { + count_off = info->reg_off_first[0]; + count_last = info->reg_off_split; + if (count_last < 0) { + count_last = info->reg_off_last[0]; + } + + do { + mte_check(env, mtedesc, addr + count_off, ra); + count_off += estride; + } while (count_off <= count_last); + } + + count_off = info->reg_off_first[1]; + if (count_off >= 0 && info->page[1].tagged) { + count_last = info->reg_off_last[1]; + do { + mte_check(env, mtedesc, addr + count_off, ra); + count_off += estride; + } while (count_off <= count_last); + } +} + +static inline QEMU_ALWAYS_INLINE +void sve2p1_ld1_c(CPUARMState *env, ARMVectorReg *zd, const vaddr addr, + uint32_t png, uint32_t desc, + const uintptr_t ra, const MemOp esz, + sve_ldst1_host_fn *host_fn, + sve_ldst1_tlb_fn *tlb_fn) +{ + const unsigned N = (desc >> SIMD_DATA_SHIFT) & 1 ? 4 : 2; + const unsigned rstride = 1 << ((desc >> (SIMD_DATA_SHIFT + 1)) % 4); + uint32_t mtedesc = desc >> (SIMD_DATA_SHIFT + SVE_MTEDESC_SHIFT); + const intptr_t reg_max = simd_oprsz(desc); + const unsigned esize = 1 << esz; + intptr_t count_off, count_last; + intptr_t reg_off, reg_last, reg_n; + SVEContLdSt info; + unsigned estride, flags; + void *host; + + estride = sve2p1_cont_ldst_elements(&info, addr, png, reg_max, N, esz); + if (estride == 0) { + /* The entire predicate was false; no load occurs. */ + for (unsigned n = 0; n < N; n++) { + memset(zd + n * rstride, 0, reg_max); + } + return; + } + + /* Probe the page(s). Exit with exception for any invalid page. */ + sve_cont_ldst_pages(&info, FAULT_ALL, env, addr, MMU_DATA_LOAD, ra); + + /* Handle watchpoints for all active elements. */ + sve2p1_cont_ldst_watchpoints(&info, env, addr, estride, + esize, BP_MEM_READ, ra); + + /* + * Handle mte checks for all active elements. + * Since TBI must be set for MTE, !mtedesc => !mte_active. + */ + if (mtedesc) { + sve2p1_cont_ldst_mte_check(&info, env, estride, addr, + esize, mtedesc, ra); + } + + flags = info.page[0].flags | info.page[1].flags; + if (unlikely(flags != 0)) { + /* + * At least one page includes MMIO. + * Any bus operation can fail with cpu_transaction_failed, + * which for ARM will raise SyncExternal. Perform the load + * into scratch memory to preserve register state until the end. + */ + ARMVectorReg scratch[4] = { }; + + count_off = info.reg_off_first[0]; + count_last = info.reg_off_last[1]; + if (count_last < 0) { + count_last = info.reg_off_split; + if (count_last < 0) { + count_last = info.reg_off_last[0]; + } + } + reg_off = count_off % reg_max; + reg_n = count_off / reg_max; + + do { + reg_last = MIN(count_last - count_off, reg_max - esize); + do { + tlb_fn(env, &scratch[reg_n], reg_off, addr + count_off, ra); + reg_off += estride; + count_off += estride; + } while (reg_off <= reg_last); + reg_off = 0; + reg_n++; + } while (count_off <= count_last); + + for (unsigned n = 0; n < N; ++n) { + memcpy(&zd[n * rstride], &scratch[n], reg_max); + } + return; + } + + /* The entire operation is in RAM, on valid pages. */ + + for (unsigned n = 0; n < N; ++n) { + memset(&zd[n * rstride], 0, reg_max); + } + + count_off = info.reg_off_first[0]; + count_last = info.reg_off_last[0]; + reg_off = count_off % reg_max; + reg_n = count_off / reg_max; + host = info.page[0].host; + + set_helper_retaddr(ra); + + do { + reg_last = MIN(count_last - reg_n * reg_max, reg_max - esize); + do { + host_fn(&zd[reg_n * rstride], reg_off, host + count_off); + reg_off += estride; + count_off += estride; + } while (reg_off <= reg_last); + reg_off = 0; + reg_n++; + } while (count_off <= count_last); + + clear_helper_retaddr(); + + /* + * Use the slow path to manage the cross-page misalignment. + * But we know this is RAM and cannot trap. + */ + count_off = info.reg_off_split; + if (unlikely(count_off >= 0)) { + reg_off = count_off % reg_max; + reg_n = count_off / reg_max; + tlb_fn(env, &zd[reg_n * rstride], reg_off, addr + count_off, ra); + } + + count_off = info.reg_off_first[1]; + if (unlikely(count_off >= 0)) { + count_last = info.reg_off_last[1]; + reg_off = count_off % reg_max; + reg_n = count_off / reg_max; + host = info.page[1].host; + + set_helper_retaddr(ra); + + do { + reg_last = MIN(count_last - reg_n * reg_max, reg_max - esize); + do { + host_fn(&zd[reg_n * rstride], reg_off, host + count_off); + reg_off += estride; + count_off += estride; + } while (reg_off <= reg_last); + reg_off = 0; + reg_n++; + } while (count_off <= count_last); + + clear_helper_retaddr(); + } +} + +void HELPER(sve2p1_ld1bb_c)(CPUARMState *env, void *vd, target_ulong addr, + uint32_t png, uint32_t desc) +{ + sve2p1_ld1_c(env, vd, addr, png, desc, GETPC(), MO_8, + sve_ld1bb_host, sve_ld1bb_tlb); +} + +#define DO_LD1_2(NAME, ESZ) \ +void HELPER(sve2p1_##NAME##_le_c)(CPUARMState *env, void *vd, \ + target_ulong addr, uint32_t png, \ + uint32_t desc) \ +{ \ + sve2p1_ld1_c(env, vd, addr, png, desc, GETPC(), ESZ, \ + sve_##NAME##_le_host, sve_##NAME##_le_tlb); \ +} \ +void HELPER(sve2p1_##NAME##_be_c)(CPUARMState *env, void *vd, \ + target_ulong addr, uint32_t png, \ + uint32_t desc) \ +{ \ + sve2p1_ld1_c(env, vd, addr, png, desc, GETPC(), ESZ, \ + sve_##NAME##_be_host, sve_##NAME##_be_tlb); \ +} + +DO_LD1_2(ld1hh, MO_16) +DO_LD1_2(ld1ss, MO_32) +DO_LD1_2(ld1dd, MO_64) + +#undef DO_LD1_2 + +static inline QEMU_ALWAYS_INLINE +void sve2p1_st1_c(CPUARMState *env, ARMVectorReg *zd, const vaddr addr, + uint32_t png, uint32_t desc, + const uintptr_t ra, const int esz, + sve_ldst1_host_fn *host_fn, + sve_ldst1_tlb_fn *tlb_fn) +{ + const unsigned N = (desc >> SIMD_DATA_SHIFT) & 1 ? 4 : 2; + const unsigned rstride = 1 << ((desc >> (SIMD_DATA_SHIFT + 1)) % 4); + uint32_t mtedesc = desc >> (SIMD_DATA_SHIFT + SVE_MTEDESC_SHIFT); + const intptr_t reg_max = simd_oprsz(desc); + const unsigned esize = 1 << esz; + intptr_t count_off, count_last; + intptr_t reg_off, reg_last, reg_n; + SVEContLdSt info; + unsigned estride, flags; + void *host; + + estride = sve2p1_cont_ldst_elements(&info, addr, png, reg_max, N, esz); + if (estride == 0) { + /* The entire predicate was false; no store occurs. */ + return; + } + + /* Probe the page(s). Exit with exception for any invalid page. */ + sve_cont_ldst_pages(&info, FAULT_ALL, env, addr, MMU_DATA_STORE, ra); + + /* Handle watchpoints for all active elements. */ + sve2p1_cont_ldst_watchpoints(&info, env, addr, estride, + esize, BP_MEM_WRITE, ra); + + /* + * Handle mte checks for all active elements. + * Since TBI must be set for MTE, !mtedesc => !mte_active. + */ + if (mtedesc) { + sve2p1_cont_ldst_mte_check(&info, env, estride, addr, + esize, mtedesc, ra); + } + + flags = info.page[0].flags | info.page[1].flags; + if (unlikely(flags != 0)) { + /* + * At least one page includes MMIO. + * Any bus operation can fail with cpu_transaction_failed, + * which for ARM will raise SyncExternal. Perform the load + * into scratch memory to preserve register state until the end. + */ + count_off = info.reg_off_first[0]; + count_last = info.reg_off_last[1]; + if (count_last < 0) { + count_last = info.reg_off_split; + if (count_last < 0) { + count_last = info.reg_off_last[0]; + } + } + reg_off = count_off % reg_max; + reg_n = count_off / reg_max; + + do { + reg_last = MIN(count_last - count_off, reg_max - esize); + do { + tlb_fn(env, &zd[reg_n * rstride], reg_off, addr + count_off, ra); + reg_off += estride; + count_off += estride; + } while (reg_off <= reg_last); + reg_off = 0; + reg_n++; + } while (count_off <= count_last); + return; + } + + /* The entire operation is in RAM, on valid pages. */ + + count_off = info.reg_off_first[0]; + count_last = info.reg_off_last[0]; + reg_off = count_off % reg_max; + reg_n = count_off / reg_max; + host = info.page[0].host; + + set_helper_retaddr(ra); + + do { + reg_last = MIN(count_last - reg_n * reg_max, reg_max - esize); + do { + host_fn(&zd[reg_n * rstride], reg_off, host + count_off); + reg_off += estride; + count_off += estride; + } while (reg_off <= reg_last); + reg_off = 0; + reg_n++; + } while (count_off <= count_last); + + clear_helper_retaddr(); + + /* + * Use the slow path to manage the cross-page misalignment. + * But we know this is RAM and cannot trap. + */ + count_off = info.reg_off_split; + if (unlikely(count_off >= 0)) { + reg_off = count_off % reg_max; + reg_n = count_off / reg_max; + tlb_fn(env, &zd[reg_n * rstride], reg_off, addr + count_off, ra); + } + + count_off = info.reg_off_first[1]; + if (unlikely(count_off >= 0)) { + count_last = info.reg_off_last[1]; + reg_off = count_off % reg_max; + reg_n = count_off / reg_max; + host = info.page[1].host; + + set_helper_retaddr(ra); + + do { + reg_last = MIN(count_last - reg_n * reg_max, reg_max - esize); + do { + host_fn(&zd[reg_n * rstride], reg_off, host + count_off); + reg_off += estride; + count_off += estride; + } while (reg_off <= reg_last); + reg_off = 0; + reg_n++; + } while (count_off <= count_last); + + clear_helper_retaddr(); + } +} + +void HELPER(sve2p1_st1bb_c)(CPUARMState *env, void *vd, target_ulong addr, + uint32_t png, uint32_t desc) +{ + sve2p1_st1_c(env, vd, addr, png, desc, GETPC(), MO_8, + sve_st1bb_host, sve_st1bb_tlb); +} + +#define DO_ST1_2(NAME, ESZ) \ +void HELPER(sve2p1_##NAME##_le_c)(CPUARMState *env, void *vd, \ + target_ulong addr, uint32_t png, \ + uint32_t desc) \ +{ \ + sve2p1_st1_c(env, vd, addr, png, desc, GETPC(), ESZ, \ + sve_##NAME##_le_host, sve_##NAME##_le_tlb); \ +} \ +void HELPER(sve2p1_##NAME##_be_c)(CPUARMState *env, void *vd, \ + target_ulong addr, uint32_t png, \ + uint32_t desc) \ +{ \ + sve2p1_st1_c(env, vd, addr, png, desc, GETPC(), ESZ, \ + sve_##NAME##_be_host, sve_##NAME##_be_tlb); \ +} + +DO_ST1_2(st1hh, MO_16) +DO_ST1_2(st1ss, MO_32) +DO_ST1_2(st1dd, MO_64) + +#undef DO_ST1_2 + void HELPER(sve2_eor3)(void *vd, void *vn, void *vm, void *vk, uint32_t desc) { intptr_t i, opr_sz = simd_oprsz(desc) / 8; diff --git a/target/arm/tcg/translate-sve.c b/target/arm/tcg/translate-sve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sve.c +++ b/target/arm/tcg/translate-sve.c @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(UQCVTN_sh, aa64_sme2_or_sve2p1, gen_gvec_ool_zz, gen_helper_sme2_uqcvtn_sh, a->rd, a->rn, 0) TRANS_FEAT(SQCVTUN_sh, aa64_sme2_or_sve2p1, gen_gvec_ool_zz, gen_helper_sme2_sqcvtun_sh, a->rd, a->rn, 0) + +static bool gen_ldst_c(DisasContext *s, TCGv_i64 addr, int zd, int png, + MemOp esz, bool is_write, int n, bool strided) +{ + typedef void ldst_c_fn(TCGv_env, TCGv_ptr, TCGv_i64, + TCGv_i32, TCGv_i32); + static ldst_c_fn * const f_ldst[2][2][4] = { + { { gen_helper_sve2p1_ld1bb_c, + gen_helper_sve2p1_ld1hh_le_c, + gen_helper_sve2p1_ld1ss_le_c, + gen_helper_sve2p1_ld1dd_le_c, }, + { gen_helper_sve2p1_ld1bb_c, + gen_helper_sve2p1_ld1hh_be_c, + gen_helper_sve2p1_ld1ss_be_c, + gen_helper_sve2p1_ld1dd_be_c, } }, + + { { gen_helper_sve2p1_st1bb_c, + gen_helper_sve2p1_st1hh_le_c, + gen_helper_sve2p1_st1ss_le_c, + gen_helper_sve2p1_st1dd_le_c, }, + { gen_helper_sve2p1_st1bb_c, + gen_helper_sve2p1_st1hh_be_c, + gen_helper_sve2p1_st1ss_be_c, + gen_helper_sve2p1_st1dd_be_c, } } + }; + + TCGv_i32 t_png, t_desc; + TCGv_ptr t_zd; + uint32_t desc, lg2_rstride = 0; + bool be = s->be_data == MO_BE; + + assert(n == 2 || n == 4); + if (strided) { + lg2_rstride = 3; + if (n == 4) { + /* Validate ZD alignment. */ + if (zd & 4) { + return false; + } + lg2_rstride = 2; + } + /* Ignore non-temporal bit */ + zd &= ~8; + } + + if (strided || !dc_isar_feature(aa64_sve2p1, s) + ? !sme_sm_enabled_check(s) + : !sve_access_check(s)) { + return true; + } + + if (!s->mte_active[0]) { + addr = clean_data_tbi(s, addr); + } + + desc = n == 2 ? 0 : 1; + desc = desc | (lg2_rstride << 1); + desc = make_svemte_desc(s, vec_full_reg_size(s), 1, esz, is_write, desc); + t_desc = tcg_constant_i32(desc); + + t_png = tcg_temp_new_i32(); + tcg_gen_ld16u_i32(t_png, tcg_env, + pred_full_reg_offset(s, png) ^ + (HOST_BIG_ENDIAN ? 6 : 0)); + + t_zd = tcg_temp_new_ptr(); + tcg_gen_addi_ptr(t_zd, tcg_env, vec_full_reg_offset(s, zd)); + + f_ldst[is_write][be][esz](tcg_env, t_zd, addr, t_png, t_desc); + return true; +} + +static bool gen_ldst_zcrr_c(DisasContext *s, arg_zcrr_ldst *a, + bool is_write, bool strided) +{ + TCGv_i64 addr = tcg_temp_new_i64(); + + tcg_gen_shli_i64(addr, cpu_reg(s, a->rm), a->esz); + tcg_gen_add_i64(addr, addr, cpu_reg_sp(s, a->rn)); + return gen_ldst_c(s, addr, a->rd, a->png, a->esz, is_write, + a->nreg, strided); +} + +static bool gen_ldst_zcri_c(DisasContext *s, arg_zcri_ldst *a, + bool is_write, bool strided) +{ + TCGv_i64 addr = tcg_temp_new_i64(); + + tcg_gen_addi_i64(addr, cpu_reg_sp(s, a->rn), + a->imm * a->nreg * vec_full_reg_size(s)); + return gen_ldst_c(s, addr, a->rd, a->png, a->esz, is_write, + a->nreg, strided); +} + +TRANS_FEAT(LD1_zcrr, aa64_sme2_or_sve2p1, gen_ldst_zcrr_c, a, false, false) +TRANS_FEAT(LD1_zcri, aa64_sme2_or_sve2p1, gen_ldst_zcri_c, a, false, false) +TRANS_FEAT(ST1_zcrr, aa64_sme2_or_sve2p1, gen_ldst_zcrr_c, a, true, false) +TRANS_FEAT(ST1_zcri, aa64_sme2_or_sve2p1, gen_ldst_zcri_c, a, true, false) + +TRANS_FEAT(LD1_zcrr_stride, aa64_sme2, gen_ldst_zcrr_c, a, false, true) +TRANS_FEAT(LD1_zcri_stride, aa64_sme2, gen_ldst_zcri_c, a, false, true) +TRANS_FEAT(ST1_zcrr_stride, aa64_sme2, gen_ldst_zcrr_c, a, true, true) +TRANS_FEAT(ST1_zcri_stride, aa64_sme2, gen_ldst_zcri_c, a, true, true) -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> The msz > esz encodings are reserved, and some of them are about to be reused. Split these patterns so that the new insns do not overlap. Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-95-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/sve.decode | 26 ++++++++++++++++++-------- 1 file changed, 18 insertions(+), 8 deletions(-) diff --git a/target/arm/tcg/sve.decode b/target/arm/tcg/sve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve.decode +++ b/target/arm/tcg/sve.decode @@ -XXX,XX +XXX,XX @@ # Stores; user must fill in ESZ, MSZ, NREG as needed. @rprr_store ....... .. .. rm:5 ... pg:3 rn:5 rd:5 &rprr_store -@rpri_store_msz ....... msz:2 .. . imm:s4 ... pg:3 rn:5 rd:5 &rpri_store +@rpri_store ....... .. .. . imm:s4 ... pg:3 rn:5 rd:5 &rpri_store @rprr_store_esz_n0 ....... .. esz:2 rm:5 ... pg:3 rn:5 rd:5 \ &rprr_store nreg=0 @rprr_scatter_store ....... msz:2 .. rm:5 ... pg:3 rn:5 rd:5 \ @@ -XXX,XX +XXX,XX @@ STR_zri 1110010 11 0. ..... 010 ... ..... ..... @rd_rn_i9 # SVE contiguous store (scalar plus immediate) # ST1B, ST1H, ST1W, ST1D; require msz <= esz -ST_zpri 1110010 .. esz:2 0.... 111 ... ..... ..... \ - @rpri_store_msz nreg=0 +ST_zpri 1110010 00 esz:2 0.... 111 ... ..... ..... \ + @rpri_store msz=0 nreg=0 +ST_zpri 1110010 01 esz:2 0.... 111 ... ..... ..... \ + @rpri_store msz=1 nreg=0 +ST_zpri 1110010 10 10 0.... 111 ... ..... ..... \ + @rpri_store msz=2 esz=2 nreg=0 +ST_zpri 1110010 10 11 0.... 111 ... ..... ..... \ + @rpri_store msz=2 esz=3 nreg=0 +ST_zpri 1110010 11 11 0.... 111 ... ..... ..... \ + @rpri_store msz=3 esz=3 nreg=0 # SVE contiguous store (scalar plus scalar) # ST1B, ST1H, ST1W, ST1D; require msz <= esz @@ -XXX,XX +XXX,XX @@ ST_zprr 1110010 00 .. ..... 010 ... ..... ..... \ @rprr_store_esz_n0 msz=0 ST_zprr 1110010 01 .. ..... 010 ... ..... ..... \ @rprr_store_esz_n0 msz=1 -ST_zprr 1110010 10 .. ..... 010 ... ..... ..... \ - @rprr_store_esz_n0 msz=2 +ST_zprr 1110010 10 10 ..... 010 ... ..... ..... \ + @rprr_store msz=2 esz=2 nreg=0 +ST_zprr 1110010 10 11 ..... 010 ... ..... ..... \ + @rprr_store msz=2 esz=3 nreg=0 ST_zprr 1110010 11 11 ..... 010 ... ..... ..... \ @rprr_store msz=3 esz=3 nreg=0 # SVE contiguous non-temporal store (scalar plus immediate) (nreg == 0) # SVE store multiple structures (scalar plus immediate) (nreg != 0) ST_zpri 1110010 .. nreg:2 1.... 111 ... ..... ..... \ - @rpri_store_msz esz=%size_23 + @rpri_store msz=%size_23 esz=%size_23 # SVE contiguous non-temporal store (scalar plus scalar) (nreg == 0) # SVE store multiple structures (scalar plus scalar) (nreg != 0) -ST_zprr 1110010 msz:2 nreg:2 ..... 011 ... ..... ..... \ - @rprr_store esz=%size_23 +ST_zprr 1110010 .. nreg:2 ..... 011 ... ..... ..... \ + @rprr_store msz=%size_23 esz=%size_23 # SVE 32-bit scatter store (scalar plus 32-bit scaled offsets) # Require msz > 0 && msz <= esz. -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-96-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper-sve.h | 22 +++++ target/arm/tcg/sve_ldst_internal.h | 26 ++++++ target/arm/tcg/sve.decode | 20 +++++ target/arm/tcg/sve_helper.c | 6 ++ target/arm/tcg/translate-sve.c | 136 +++++++++++++++++++++++------ 5 files changed, 183 insertions(+), 27 deletions(-) diff --git a/target/arm/tcg/helper-sve.h b/target/arm/tcg/helper-sve.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-sve.h +++ b/target/arm/tcg/helper-sve.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_4(sve_ld1hds_be_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) DEF_HELPER_FLAGS_4(sve_ld1sdu_le_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) DEF_HELPER_FLAGS_4(sve_ld1sds_le_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) +DEF_HELPER_FLAGS_4(sve_ld1squ_le_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) +DEF_HELPER_FLAGS_4(sve_ld1dqu_le_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) + DEF_HELPER_FLAGS_4(sve_ld1sdu_be_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) DEF_HELPER_FLAGS_4(sve_ld1sds_be_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) +DEF_HELPER_FLAGS_4(sve_ld1squ_be_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) +DEF_HELPER_FLAGS_4(sve_ld1dqu_be_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) + DEF_HELPER_FLAGS_4(sve_ld1bb_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) DEF_HELPER_FLAGS_4(sve_ld2bb_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) DEF_HELPER_FLAGS_4(sve_ld3bb_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_4(sve_ld1hds_be_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) DEF_HELPER_FLAGS_4(sve_ld1sdu_le_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) DEF_HELPER_FLAGS_4(sve_ld1sds_le_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) +DEF_HELPER_FLAGS_4(sve_ld1squ_le_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) +DEF_HELPER_FLAGS_4(sve_ld1dqu_le_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) + DEF_HELPER_FLAGS_4(sve_ld1sdu_be_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) DEF_HELPER_FLAGS_4(sve_ld1sds_be_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) +DEF_HELPER_FLAGS_4(sve_ld1squ_be_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) +DEF_HELPER_FLAGS_4(sve_ld1dqu_be_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) + DEF_HELPER_FLAGS_4(sve_ldff1bb_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) DEF_HELPER_FLAGS_4(sve_ldff1bhu_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) DEF_HELPER_FLAGS_4(sve_ldff1bsu_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_4(sve_st1hd_be_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) DEF_HELPER_FLAGS_4(sve_st1sd_le_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) DEF_HELPER_FLAGS_4(sve_st1sd_be_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) +DEF_HELPER_FLAGS_4(sve_st1sq_le_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) +DEF_HELPER_FLAGS_4(sve_st1sq_be_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) +DEF_HELPER_FLAGS_4(sve_st1dq_le_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) +DEF_HELPER_FLAGS_4(sve_st1dq_be_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) + DEF_HELPER_FLAGS_4(sve_st1bb_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) DEF_HELPER_FLAGS_4(sve_st2bb_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) DEF_HELPER_FLAGS_4(sve_st3bb_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_4(sve_st1hd_be_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) DEF_HELPER_FLAGS_4(sve_st1sd_le_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) DEF_HELPER_FLAGS_4(sve_st1sd_be_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) +DEF_HELPER_FLAGS_4(sve_st1sq_le_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) +DEF_HELPER_FLAGS_4(sve_st1sq_be_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) +DEF_HELPER_FLAGS_4(sve_st1dq_le_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) +DEF_HELPER_FLAGS_4(sve_st1dq_be_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) + DEF_HELPER_FLAGS_6(sve_ldbsu_zsu, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr, tl, i32) DEF_HELPER_FLAGS_6(sve_ldhsu_le_zsu, TCG_CALL_NO_WG, diff --git a/target/arm/tcg/sve_ldst_internal.h b/target/arm/tcg/sve_ldst_internal.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve_ldst_internal.h +++ b/target/arm/tcg/sve_ldst_internal.h @@ -XXX,XX +XXX,XX @@ DO_ST_PRIM_2(sd, H1_8, uint64_t, uint32_t, stl) DO_LD_PRIM_2(dd, H1_8, uint64_t, uint64_t, ldq) DO_ST_PRIM_2(dd, H1_8, uint64_t, uint64_t, stq) +#define DO_LD_PRIM_3(NAME, FUNC) \ + static inline void sve_##NAME##_host(void *vd, \ + intptr_t reg_off, void *host) \ + { sve_##FUNC##_host(vd, reg_off, host); \ + *(uint64_t *)(vd + reg_off + 8) = 0; } \ + static inline void sve_##NAME##_tlb(CPUARMState *env, void *vd, \ + intptr_t reg_off, target_ulong addr, uintptr_t ra) \ + { sve_##FUNC##_tlb(env, vd, reg_off, addr, ra); \ + *(uint64_t *)(vd + reg_off + 8) = 0; } + +DO_LD_PRIM_3(ld1squ_be, ld1sdu_be) +DO_LD_PRIM_3(ld1squ_le, ld1sdu_le) +DO_LD_PRIM_3(ld1dqu_be, ld1dd_be) +DO_LD_PRIM_3(ld1dqu_le, ld1dd_le) + +#define sve_st1sq_be_host sve_st1sd_be_host +#define sve_st1sq_le_host sve_st1sd_le_host +#define sve_st1sq_be_tlb sve_st1sd_be_tlb +#define sve_st1sq_le_tlb sve_st1sd_le_tlb + +#define sve_st1dq_be_host sve_st1dd_be_host +#define sve_st1dq_le_host sve_st1dd_le_host +#define sve_st1dq_be_tlb sve_st1dd_be_tlb +#define sve_st1dq_le_tlb sve_st1dd_le_tlb + #undef DO_LD_TLB #undef DO_ST_TLB #undef DO_LD_HOST @@ -XXX,XX +XXX,XX @@ DO_ST_PRIM_2(dd, H1_8, uint64_t, uint64_t, stq) #undef DO_ST_PRIM_1 #undef DO_LD_PRIM_2 #undef DO_ST_PRIM_2 +#undef DO_LD_PRIM_3 /* * Resolve the guest virtual address to info->host and info->flags. diff --git a/target/arm/tcg/sve.decode b/target/arm/tcg/sve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve.decode +++ b/target/arm/tcg/sve.decode @@ -XXX,XX +XXX,XX @@ LD1_zpiz 1000010 .. 01 ..... 1.. ... ..... ..... \ # SVE contiguous load (scalar plus scalar) LD_zprr 1010010 .... ..... 010 ... ..... ..... @rprr_load_dt nreg=0 +# LD1W (128-bit element) +LD_zprr 1010010 1000 rm:5 100 pg:3 rn:5 rd:5 \ + &rprr_load dtype=16 nreg=0 +# LD1D (128-bit element) +LD_zprr 1010010 1100 rm:5 100 pg:3 rn:5 rd:5 \ + &rprr_load dtype=17 nreg=0 # SVE contiguous first-fault load (scalar plus scalar) LDFF1_zprr 1010010 .... ..... 011 ... ..... ..... @rprr_load_dt nreg=0 # SVE contiguous load (scalar plus immediate) LD_zpri 1010010 .... 0.... 101 ... ..... ..... @rpri_load_dt nreg=0 +# LD1W (128-bit element) +LD_zpri 1010010 1000 1 imm:s4 001 pg:3 rn:5 rd:5 \ + &rpri_load dtype=16 nreg=0 +# LD1D (128-bit element) +LD_zpri 1010010 1100 1 imm:s4 001 pg:3 rn:5 rd:5 \ + &rpri_load dtype=17 nreg=0 # SVE contiguous non-fault load (scalar plus immediate) LDNF1_zpri 1010010 .... 1.... 101 ... ..... ..... @rpri_load_dt nreg=0 @@ -XXX,XX +XXX,XX @@ ST_zpri 1110010 10 11 0.... 111 ... ..... ..... \ @rpri_store msz=2 esz=3 nreg=0 ST_zpri 1110010 11 11 0.... 111 ... ..... ..... \ @rpri_store msz=3 esz=3 nreg=0 +ST_zpri 1110010 10 00 0.... 111 ... ..... ..... \ + @rpri_store msz=2 esz=4 nreg=0 +ST_zpri 1110010 11 10 0.... 111 ... ..... ..... \ + @rpri_store msz=3 esz=4 nreg=0 # SVE contiguous store (scalar plus scalar) # ST1B, ST1H, ST1W, ST1D; require msz <= esz @@ -XXX,XX +XXX,XX @@ ST_zprr 1110010 10 11 ..... 010 ... ..... ..... \ @rprr_store msz=2 esz=3 nreg=0 ST_zprr 1110010 11 11 ..... 010 ... ..... ..... \ @rprr_store msz=3 esz=3 nreg=0 +ST_zprr 1110010 10 00 ..... 010 ... ..... ..... \ + @rprr_store msz=2 esz=4 nreg=0 +ST_zprr 1110010 11 10 ..... 010 ... ..... ..... \ + @rprr_store msz=3 esz=4 nreg=0 # SVE contiguous non-temporal store (scalar plus immediate) (nreg == 0) # SVE store multiple structures (scalar plus immediate) (nreg != 0) diff --git a/target/arm/tcg/sve_helper.c b/target/arm/tcg/sve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve_helper.c +++ b/target/arm/tcg/sve_helper.c @@ -XXX,XX +XXX,XX @@ DO_LD1_2(ld1sds, MO_64, MO_32) DO_LD1_2(ld1dd, MO_64, MO_64) +DO_LD1_2(ld1squ, MO_32, MO_128) +DO_LD1_2(ld1dqu, MO_64, MO_128) + #undef DO_LD1_1 #undef DO_LD1_2 @@ -XXX,XX +XXX,XX @@ DO_STN_2(2, dd, MO_64, MO_64) DO_STN_2(3, dd, MO_64, MO_64) DO_STN_2(4, dd, MO_64, MO_64) +DO_STN_2(1, sq, MO_128, MO_32) +DO_STN_2(1, dq, MO_128, MO_64) + #undef DO_STN_1 #undef DO_STN_2 diff --git a/target/arm/tcg/translate-sve.c b/target/arm/tcg/translate-sve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sve.c +++ b/target/arm/tcg/translate-sve.c @@ -XXX,XX +XXX,XX @@ static bool trans_STR_pri(DisasContext *s, arg_rri *a) */ /* The memory mode of the dtype. */ -static const MemOp dtype_mop[16] = { +static const MemOp dtype_mop[19] = { MO_UB, MO_UB, MO_UB, MO_UB, MO_SL, MO_UW, MO_UW, MO_UW, MO_SW, MO_SW, MO_UL, MO_UL, - MO_SB, MO_SB, MO_SB, MO_UQ + MO_SB, MO_SB, MO_SB, MO_UQ, + /* Artificial values used by decode */ + MO_UL, MO_UQ, MO_128, }; #define dtype_msz(x) (dtype_mop[x] & MO_SIZE) /* The vector element size of dtype. */ -static const uint8_t dtype_esz[16] = { +static const uint8_t dtype_esz[19] = { 0, 1, 2, 3, 3, 1, 2, 3, 3, 2, 2, 3, - 3, 2, 1, 3 + 3, 2, 1, 3, + /* Artificial values used by decode */ + 4, 4, 4, }; uint32_t make_svemte_desc(DisasContext *s, unsigned vsz, uint32_t nregs, @@ -XXX,XX +XXX,XX @@ static void do_mem_zpa(DisasContext *s, int zt, int pg, TCGv_i64 addr, } /* Indexed by [mte][be][dtype][nreg] */ -static gen_helper_gvec_mem * const ldr_fns[2][2][16][4] = { +static gen_helper_gvec_mem * const ldr_fns[2][2][18][4] = { { /* mte inactive, little-endian */ { { gen_helper_sve_ld1bb_r, gen_helper_sve_ld2bb_r, gen_helper_sve_ld3bb_r, gen_helper_sve_ld4bb_r }, @@ -XXX,XX +XXX,XX @@ static gen_helper_gvec_mem * const ldr_fns[2][2][16][4] = { { gen_helper_sve_ld1bss_r, NULL, NULL, NULL }, { gen_helper_sve_ld1bhs_r, NULL, NULL, NULL }, { gen_helper_sve_ld1dd_le_r, gen_helper_sve_ld2dd_le_r, - gen_helper_sve_ld3dd_le_r, gen_helper_sve_ld4dd_le_r } }, + gen_helper_sve_ld3dd_le_r, gen_helper_sve_ld4dd_le_r }, + + { gen_helper_sve_ld1squ_le_r, NULL, NULL, NULL }, + { gen_helper_sve_ld1dqu_le_r, NULL, NULL, NULL }, + }, /* mte inactive, big-endian */ { { gen_helper_sve_ld1bb_r, gen_helper_sve_ld2bb_r, @@ -XXX,XX +XXX,XX @@ static gen_helper_gvec_mem * const ldr_fns[2][2][16][4] = { { gen_helper_sve_ld1bss_r, NULL, NULL, NULL }, { gen_helper_sve_ld1bhs_r, NULL, NULL, NULL }, { gen_helper_sve_ld1dd_be_r, gen_helper_sve_ld2dd_be_r, - gen_helper_sve_ld3dd_be_r, gen_helper_sve_ld4dd_be_r } } }, + gen_helper_sve_ld3dd_be_r, gen_helper_sve_ld4dd_be_r }, + + { gen_helper_sve_ld1squ_be_r, NULL, NULL, NULL }, + { gen_helper_sve_ld1dqu_be_r, NULL, NULL, NULL }, + }, + }, { /* mte active, little-endian */ { { gen_helper_sve_ld1bb_r_mte, @@ -XXX,XX +XXX,XX @@ static gen_helper_gvec_mem * const ldr_fns[2][2][16][4] = { { gen_helper_sve_ld1dd_le_r_mte, gen_helper_sve_ld2dd_le_r_mte, gen_helper_sve_ld3dd_le_r_mte, - gen_helper_sve_ld4dd_le_r_mte } }, + gen_helper_sve_ld4dd_le_r_mte }, + + { gen_helper_sve_ld1squ_le_r_mte, NULL, NULL, NULL }, + { gen_helper_sve_ld1dqu_le_r_mte, NULL, NULL, NULL }, + }, /* mte active, big-endian */ { { gen_helper_sve_ld1bb_r_mte, @@ -XXX,XX +XXX,XX @@ static gen_helper_gvec_mem * const ldr_fns[2][2][16][4] = { { gen_helper_sve_ld1dd_be_r_mte, gen_helper_sve_ld2dd_be_r_mte, gen_helper_sve_ld3dd_be_r_mte, - gen_helper_sve_ld4dd_be_r_mte } } }, + gen_helper_sve_ld4dd_be_r_mte }, + + { gen_helper_sve_ld1squ_be_r_mte, NULL, NULL, NULL }, + { gen_helper_sve_ld1dqu_be_r_mte, NULL, NULL, NULL }, + }, + }, }; static void do_ld_zpa(DisasContext *s, int zt, int pg, @@ -XXX,XX +XXX,XX @@ static void do_ld_zpa(DisasContext *s, int zt, int pg, static bool trans_LD_zprr(DisasContext *s, arg_rprr_load *a) { - if (a->rm == 31 || !dc_isar_feature(aa64_sve, s)) { + if (a->rm == 31) { return false; } + + /* dtypes 16 and 17 are artificial, representing 128-bit element */ + if (a->dtype < 16) { + if (!dc_isar_feature(aa64_sve, s)) { + return false; + } + } else { + if (!dc_isar_feature(aa64_sve2p1, s)) { + return false; + } + s->is_nonstreaming = true; + } + if (sve_access_check(s)) { TCGv_i64 addr = tcg_temp_new_i64(); tcg_gen_shli_i64(addr, cpu_reg(s, a->rm), dtype_msz(a->dtype)); @@ -XXX,XX +XXX,XX @@ static bool trans_LD_zprr(DisasContext *s, arg_rprr_load *a) static bool trans_LD_zpri(DisasContext *s, arg_rpri_load *a) { - if (!dc_isar_feature(aa64_sve, s)) { - return false; + /* dtypes 16 and 17 are artificial, representing 128-bit element */ + if (a->dtype < 16) { + if (!dc_isar_feature(aa64_sve, s)) { + return false; + } + } else { + if (!dc_isar_feature(aa64_sve2p1, s)) { + return false; + } + s->is_nonstreaming = true; } + if (sve_access_check(s)) { int vsz = vec_full_reg_size(s); int elements = vsz >> dtype_esz[a->dtype]; @@ -XXX,XX +XXX,XX @@ static bool trans_LD1R_zpri(DisasContext *s, arg_rpri_load *a) static void do_st_zpa(DisasContext *s, int zt, int pg, TCGv_i64 addr, int msz, int esz, int nreg) { - static gen_helper_gvec_mem * const fn_single[2][2][4][4] = { + static gen_helper_gvec_mem * const fn_single[2][2][4][5] = { { { { gen_helper_sve_st1bb_r, gen_helper_sve_st1bh_r, gen_helper_sve_st1bs_r, @@ -XXX,XX +XXX,XX @@ static void do_st_zpa(DisasContext *s, int zt, int pg, TCGv_i64 addr, gen_helper_sve_st1hd_le_r }, { NULL, NULL, gen_helper_sve_st1ss_le_r, - gen_helper_sve_st1sd_le_r }, + gen_helper_sve_st1sd_le_r, + gen_helper_sve_st1sq_le_r, }, { NULL, NULL, NULL, - gen_helper_sve_st1dd_le_r } }, + gen_helper_sve_st1dd_le_r, + gen_helper_sve_st1dq_le_r, } }, { { gen_helper_sve_st1bb_r, gen_helper_sve_st1bh_r, gen_helper_sve_st1bs_r, @@ -XXX,XX +XXX,XX @@ static void do_st_zpa(DisasContext *s, int zt, int pg, TCGv_i64 addr, gen_helper_sve_st1hd_be_r }, { NULL, NULL, gen_helper_sve_st1ss_be_r, - gen_helper_sve_st1sd_be_r }, + gen_helper_sve_st1sd_be_r, + gen_helper_sve_st1sq_be_r }, { NULL, NULL, NULL, - gen_helper_sve_st1dd_be_r } } }, + gen_helper_sve_st1dd_be_r, + gen_helper_sve_st1dq_be_r } } }, { { { gen_helper_sve_st1bb_r_mte, gen_helper_sve_st1bh_r_mte, @@ -XXX,XX +XXX,XX @@ static void do_st_zpa(DisasContext *s, int zt, int pg, TCGv_i64 addr, gen_helper_sve_st1hd_le_r_mte }, { NULL, NULL, gen_helper_sve_st1ss_le_r_mte, - gen_helper_sve_st1sd_le_r_mte }, + gen_helper_sve_st1sd_le_r_mte, + gen_helper_sve_st1sq_le_r_mte }, { NULL, NULL, NULL, - gen_helper_sve_st1dd_le_r_mte } }, + gen_helper_sve_st1dd_le_r_mte, + gen_helper_sve_st1dq_le_r_mte } }, { { gen_helper_sve_st1bb_r_mte, gen_helper_sve_st1bh_r_mte, gen_helper_sve_st1bs_r_mte, @@ -XXX,XX +XXX,XX @@ static void do_st_zpa(DisasContext *s, int zt, int pg, TCGv_i64 addr, gen_helper_sve_st1hd_be_r_mte }, { NULL, NULL, gen_helper_sve_st1ss_be_r_mte, - gen_helper_sve_st1sd_be_r_mte }, + gen_helper_sve_st1sd_be_r_mte, + gen_helper_sve_st1sq_be_r_mte }, { NULL, NULL, NULL, - gen_helper_sve_st1dd_be_r_mte } } }, + gen_helper_sve_st1dd_be_r_mte, + gen_helper_sve_st1dq_be_r_mte } } }, }; static gen_helper_gvec_mem * const fn_multiple[2][2][3][4] = { { { { gen_helper_sve_st2bb_r, @@ -XXX,XX +XXX,XX @@ static void do_st_zpa(DisasContext *s, int zt, int pg, TCGv_i64 addr, static bool trans_ST_zprr(DisasContext *s, arg_rprr_store *a) { - if (!dc_isar_feature(aa64_sve, s)) { - return false; - } if (a->rm == 31 || a->msz > a->esz) { return false; } + switch (a->esz) { + case MO_8 ... MO_64: + if (!dc_isar_feature(aa64_sve, s)) { + return false; + } + break; + case MO_128: + assert(a->msz < a->esz); + assert(a->nreg == 0); + if (!dc_isar_feature(aa64_sve2p1, s)) { + return false; + } + s->is_nonstreaming = true; + break; + default: + g_assert_not_reached(); + } + if (sve_access_check(s)) { TCGv_i64 addr = tcg_temp_new_i64(); tcg_gen_shli_i64(addr, cpu_reg(s, a->rm), a->msz); @@ -XXX,XX +XXX,XX @@ static bool trans_ST_zprr(DisasContext *s, arg_rprr_store *a) static bool trans_ST_zpri(DisasContext *s, arg_rpri_store *a) { - if (!dc_isar_feature(aa64_sve, s)) { - return false; - } if (a->msz > a->esz) { return false; } + switch (a->esz) { + case MO_8 ... MO_64: + if (!dc_isar_feature(aa64_sve, s)) { + return false; + } + break; + case MO_128: + assert(a->msz < a->esz); + assert(a->nreg == 0); + if (!dc_isar_feature(aa64_sve2p1, s)) { + return false; + } + s->is_nonstreaming = true; + break; + default: + g_assert_not_reached(); + } + if (sve_access_check(s)) { int vsz = vec_full_reg_size(s); int elements = vsz >> a->esz; -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Move from sme_helper.c to the shared header. Add a comment noting the lack of atomicity. Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-97-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/sve_ldst_internal.h | 63 ++++++++++++++++++++++++++++++ target/arm/tcg/sme_helper.c | 44 +++------------------ 2 files changed, 69 insertions(+), 38 deletions(-) diff --git a/target/arm/tcg/sve_ldst_internal.h b/target/arm/tcg/sve_ldst_internal.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve_ldst_internal.h +++ b/target/arm/tcg/sve_ldst_internal.h @@ -XXX,XX +XXX,XX @@ DO_LD_PRIM_3(ld1dqu_le, ld1dd_le) #define sve_st1dq_be_tlb sve_st1dd_be_tlb #define sve_st1dq_le_tlb sve_st1dd_le_tlb +/* + * The ARMVectorReg elements are stored in host-endian 64-bit units. + * For 128-bit quantities, the sequence defined by the Elem[] pseudocode + * corresponds to storing the two 64-bit pieces in little-endian order. + */ +/* FIXME: Nothing in this file makes any effort at atomicity. */ + +static inline void sve_ld1qq_be_host(void *vd, intptr_t reg_off, void *host) +{ + sve_ld1dd_be_host(vd, reg_off + 8, host); + sve_ld1dd_be_host(vd, reg_off, host + 8); +} + +static inline void sve_ld1qq_le_host(void *vd, intptr_t reg_off, void *host) +{ + sve_ld1dd_le_host(vd, reg_off, host); + sve_ld1dd_le_host(vd, reg_off + 8, host + 8); +} + +static inline void +sve_ld1qq_be_tlb(CPUARMState *env, void *vd, intptr_t reg_off, + target_ulong addr, uintptr_t ra) +{ + sve_ld1dd_be_tlb(env, vd, reg_off + 8, addr, ra); + sve_ld1dd_be_tlb(env, vd, reg_off, addr + 8, ra); +} + +static inline void +sve_ld1qq_le_tlb(CPUARMState *env, void *vd, intptr_t reg_off, + target_ulong addr, uintptr_t ra) +{ + sve_ld1dd_le_tlb(env, vd, reg_off, addr, ra); + sve_ld1dd_le_tlb(env, vd, reg_off + 8, addr + 8, ra); +} + +static inline void sve_st1qq_be_host(void *vd, intptr_t reg_off, void *host) +{ + sve_st1dd_be_host(vd, reg_off + 8, host); + sve_st1dd_be_host(vd, reg_off, host + 8); +} + +static inline void sve_st1qq_le_host(void *vd, intptr_t reg_off, void *host) +{ + sve_st1dd_le_host(vd, reg_off, host); + sve_st1dd_le_host(vd, reg_off + 8, host + 8); +} + +static inline void +sve_st1qq_be_tlb(CPUARMState *env, void *vd, intptr_t reg_off, + target_ulong addr, uintptr_t ra) +{ + sve_st1dd_be_tlb(env, vd, reg_off + 8, addr, ra); + sve_st1dd_be_tlb(env, vd, reg_off, addr + 8, ra); +} + +static inline void +sve_st1qq_le_tlb(CPUARMState *env, void *vd, intptr_t reg_off, + target_ulong addr, uintptr_t ra) +{ + sve_st1dd_le_tlb(env, vd, reg_off, addr, ra); + sve_st1dd_le_tlb(env, vd, reg_off + 8, addr + 8, ra); +} + #undef DO_LD_TLB #undef DO_ST_TLB #undef DO_LD_HOST diff --git a/target/arm/tcg/sme_helper.c b/target/arm/tcg/sme_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme_helper.c +++ b/target/arm/tcg/sme_helper.c @@ -XXX,XX +XXX,XX @@ static inline void sme_##NAME##_v_tlb(CPUARMState *env, void *za, \ TLB(env, useronly_clean_ptr(addr), val, ra); \ } -/* - * The ARMVectorReg elements are stored in host-endian 64-bit units. - * For 128-bit quantities, the sequence defined by the Elem[] pseudocode - * corresponds to storing the two 64-bit pieces in little-endian order. - */ -#define DO_LDQ(HNAME, VNAME, BE, HOST, TLB) \ -static inline void HNAME##_host(void *za, intptr_t off, void *host) \ -{ \ - uint64_t val0 = HOST(host), val1 = HOST(host + 8); \ - uint64_t *ptr = za + off; \ - ptr[0] = BE ? val1 : val0, ptr[1] = BE ? val0 : val1; \ -} \ +#define DO_LDQ(HNAME, VNAME) \ static inline void VNAME##_v_host(void *za, intptr_t off, void *host) \ { \ HNAME##_host(za, tile_vslice_offset(off), host); \ } \ -static inline void HNAME##_tlb(CPUARMState *env, void *za, intptr_t off, \ - target_ulong addr, uintptr_t ra) \ -{ \ - uint64_t val0 = TLB(env, useronly_clean_ptr(addr), ra); \ - uint64_t val1 = TLB(env, useronly_clean_ptr(addr + 8), ra); \ - uint64_t *ptr = za + off; \ - ptr[0] = BE ? val1 : val0, ptr[1] = BE ? val0 : val1; \ -} \ static inline void VNAME##_v_tlb(CPUARMState *env, void *za, intptr_t off, \ target_ulong addr, uintptr_t ra) \ { \ HNAME##_tlb(env, za, tile_vslice_offset(off), addr, ra); \ } -#define DO_STQ(HNAME, VNAME, BE, HOST, TLB) \ -static inline void HNAME##_host(void *za, intptr_t off, void *host) \ -{ \ - uint64_t *ptr = za + off; \ - HOST(host, ptr[BE]); \ - HOST(host + 8, ptr[!BE]); \ -} \ +#define DO_STQ(HNAME, VNAME) \ static inline void VNAME##_v_host(void *za, intptr_t off, void *host) \ { \ HNAME##_host(za, tile_vslice_offset(off), host); \ } \ -static inline void HNAME##_tlb(CPUARMState *env, void *za, intptr_t off, \ - target_ulong addr, uintptr_t ra) \ -{ \ - uint64_t *ptr = za + off; \ - TLB(env, useronly_clean_ptr(addr), ptr[BE], ra); \ - TLB(env, useronly_clean_ptr(addr + 8), ptr[!BE], ra); \ -} \ static inline void VNAME##_v_tlb(CPUARMState *env, void *za, intptr_t off, \ target_ulong addr, uintptr_t ra) \ { \ @@ -XXX,XX +XXX,XX @@ DO_LD(ld1s_le, uint32_t, ldl_le_p, cpu_ldl_le_data_ra) DO_LD(ld1d_be, uint64_t, ldq_be_p, cpu_ldq_be_data_ra) DO_LD(ld1d_le, uint64_t, ldq_le_p, cpu_ldq_le_data_ra) -DO_LDQ(sve_ld1qq_be, sme_ld1q_be, 1, ldq_be_p, cpu_ldq_be_data_ra) -DO_LDQ(sve_ld1qq_le, sme_ld1q_le, 0, ldq_le_p, cpu_ldq_le_data_ra) +DO_LDQ(sve_ld1qq_be, sme_ld1q_be) +DO_LDQ(sve_ld1qq_le, sme_ld1q_le) DO_ST(st1b, uint8_t, stb_p, cpu_stb_data_ra) DO_ST(st1h_be, uint16_t, stw_be_p, cpu_stw_be_data_ra) @@ -XXX,XX +XXX,XX @@ DO_ST(st1s_le, uint32_t, stl_le_p, cpu_stl_le_data_ra) DO_ST(st1d_be, uint64_t, stq_be_p, cpu_stq_be_data_ra) DO_ST(st1d_le, uint64_t, stq_le_p, cpu_stq_le_data_ra) -DO_STQ(sve_st1qq_be, sme_st1q_be, 1, stq_be_p, cpu_stq_be_data_ra) -DO_STQ(sve_st1qq_le, sme_st1q_le, 0, stq_le_p, cpu_stq_le_data_ra) +DO_STQ(sve_st1qq_be, sme_st1q_be) +DO_STQ(sve_st1qq_le, sme_st1q_le) #undef DO_LD #undef DO_ST -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-98-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper-sve.h | 32 +++++++++ target/arm/tcg/sve.decode | 31 +++++++++ target/arm/tcg/sve_helper.c | 8 +++ target/arm/tcg/translate-sve.c | 116 ++++++++++++++++++++++++--------- 4 files changed, 156 insertions(+), 31 deletions(-) diff --git a/target/arm/tcg/helper-sve.h b/target/arm/tcg/helper-sve.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-sve.h +++ b/target/arm/tcg/helper-sve.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_4(sve_ld2dd_be_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) DEF_HELPER_FLAGS_4(sve_ld3dd_be_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) DEF_HELPER_FLAGS_4(sve_ld4dd_be_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) +DEF_HELPER_FLAGS_4(sve_ld2qq_le_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) +DEF_HELPER_FLAGS_4(sve_ld3qq_le_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) +DEF_HELPER_FLAGS_4(sve_ld4qq_le_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) + +DEF_HELPER_FLAGS_4(sve_ld2qq_be_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) +DEF_HELPER_FLAGS_4(sve_ld3qq_be_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) +DEF_HELPER_FLAGS_4(sve_ld4qq_be_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) + DEF_HELPER_FLAGS_4(sve_ld1bhu_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) DEF_HELPER_FLAGS_4(sve_ld1bsu_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) DEF_HELPER_FLAGS_4(sve_ld1bdu_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_4(sve_ld2dd_be_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) DEF_HELPER_FLAGS_4(sve_ld3dd_be_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) DEF_HELPER_FLAGS_4(sve_ld4dd_be_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) +DEF_HELPER_FLAGS_4(sve_ld2qq_le_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) +DEF_HELPER_FLAGS_4(sve_ld3qq_le_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) +DEF_HELPER_FLAGS_4(sve_ld4qq_le_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) + +DEF_HELPER_FLAGS_4(sve_ld2qq_be_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) +DEF_HELPER_FLAGS_4(sve_ld3qq_be_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) +DEF_HELPER_FLAGS_4(sve_ld4qq_be_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) + DEF_HELPER_FLAGS_4(sve_ld1bhu_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) DEF_HELPER_FLAGS_4(sve_ld1bsu_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) DEF_HELPER_FLAGS_4(sve_ld1bdu_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_4(sve_st2dd_be_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) DEF_HELPER_FLAGS_4(sve_st3dd_be_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) DEF_HELPER_FLAGS_4(sve_st4dd_be_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) +DEF_HELPER_FLAGS_4(sve_st2qq_le_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) +DEF_HELPER_FLAGS_4(sve_st3qq_le_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) +DEF_HELPER_FLAGS_4(sve_st4qq_le_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) + +DEF_HELPER_FLAGS_4(sve_st2qq_be_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) +DEF_HELPER_FLAGS_4(sve_st3qq_be_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) +DEF_HELPER_FLAGS_4(sve_st4qq_be_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) + DEF_HELPER_FLAGS_4(sve_st1bh_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) DEF_HELPER_FLAGS_4(sve_st1bs_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) DEF_HELPER_FLAGS_4(sve_st1bd_r, TCG_CALL_NO_WG, void, env, ptr, tl, i32) @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_4(sve_st2dd_be_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) DEF_HELPER_FLAGS_4(sve_st3dd_be_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) DEF_HELPER_FLAGS_4(sve_st4dd_be_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) +DEF_HELPER_FLAGS_4(sve_st2qq_le_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) +DEF_HELPER_FLAGS_4(sve_st3qq_le_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) +DEF_HELPER_FLAGS_4(sve_st4qq_le_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) + +DEF_HELPER_FLAGS_4(sve_st2qq_be_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) +DEF_HELPER_FLAGS_4(sve_st3qq_be_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) +DEF_HELPER_FLAGS_4(sve_st4qq_be_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) + DEF_HELPER_FLAGS_4(sve_st1bh_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) DEF_HELPER_FLAGS_4(sve_st1bs_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) DEF_HELPER_FLAGS_4(sve_st1bd_r_mte, TCG_CALL_NO_WG, void, env, ptr, tl, i32) diff --git a/target/arm/tcg/sve.decode b/target/arm/tcg/sve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve.decode +++ b/target/arm/tcg/sve.decode @@ -XXX,XX +XXX,XX @@ @rprr_load_dt ....... dtype:4 rm:5 ... pg:3 rn:5 rd:5 &rprr_load @rpri_load_dt ....... dtype:4 . imm:s4 ... pg:3 rn:5 rd:5 &rpri_load +@rprr_load ....... .... rm:5 ... pg:3 rn:5 rd:5 &rprr_load +@rpri_load ....... .... . imm:s4 ... pg:3 rn:5 rd:5 &rpri_load + @rprr_load_msz ....... .... rm:5 ... pg:3 rn:5 rd:5 \ &rprr_load dtype=%msz_dtype @rpri_load_msz ....... .... . imm:s4 ... pg:3 rn:5 rd:5 \ @@ -XXX,XX +XXX,XX @@ LDNF1_zpri 1010010 .... 1.... 101 ... ..... ..... @rpri_load_dt nreg=0 # SVE load multiple structures (scalar plus scalar) # LD2B, LD2H, LD2W, LD2D; etc. LD_zprr 1010010 .. nreg:2 ..... 110 ... ..... ..... @rprr_load_msz +# LD[234]Q +LD_zprr 1010010 01 01 ..... 100 ... ..... ..... \ + @rprr_load dtype=18 nreg=1 +LD_zprr 1010010 10 01 ..... 100 ... ..... ..... \ + @rprr_load dtype=18 nreg=2 +LD_zprr 1010010 11 01 ..... 100 ... ..... ..... \ + @rprr_load dtype=18 nreg=3 # SVE contiguous non-temporal load (scalar plus immediate) # LDNT1B, LDNT1H, LDNT1W, LDNT1D # SVE load multiple structures (scalar plus immediate) # LD2B, LD2H, LD2W, LD2D; etc. LD_zpri 1010010 .. nreg:2 0.... 111 ... ..... ..... @rpri_load_msz +# LD[234]Q +LD_zpri 1010010 01 001 .... 111 ... ..... ..... \ + @rpri_load dtype=18 nreg=1 +LD_zpri 1010010 10 001 .... 111 ... ..... ..... \ + @rpri_load dtype=18 nreg=2 +LD_zpri 1010010 11 001 .... 111 ... ..... ..... \ + @rpri_load dtype=18 nreg=3 # SVE load and broadcast quadword (scalar plus scalar) LD1RQ_zprr 1010010 .. 00 ..... 000 ... ..... ..... \ @@ -XXX,XX +XXX,XX @@ ST_zprr 1110010 11 10 ..... 010 ... ..... ..... \ # SVE store multiple structures (scalar plus immediate) (nreg != 0) ST_zpri 1110010 .. nreg:2 1.... 111 ... ..... ..... \ @rpri_store msz=%size_23 esz=%size_23 +# ST[234]Q +ST_zpri 11100100 01 00 .... 000 ... ..... ..... \ + @rpri_store msz=4 esz=4 nreg=1 +ST_zpri 11100100 10 00 .... 000 ... ..... ..... \ + @rpri_store msz=4 esz=4 nreg=2 +ST_zpri 11100100 11 00 .... 000 ... ..... ..... \ + @rpri_store msz=4 esz=4 nreg=3 # SVE contiguous non-temporal store (scalar plus scalar) (nreg == 0) # SVE store multiple structures (scalar plus scalar) (nreg != 0) ST_zprr 1110010 .. nreg:2 ..... 011 ... ..... ..... \ @rprr_store msz=%size_23 esz=%size_23 +# ST[234]Q +ST_zprr 11100100 01 1 ..... 000 ... ..... ..... \ + @rprr_store msz=4 esz=4 nreg=1 +ST_zprr 11100100 10 1 ..... 000 ... ..... ..... \ + @rprr_store msz=4 esz=4 nreg=2 +ST_zprr 11100100 11 1 ..... 000 ... ..... ..... \ + @rprr_store msz=4 esz=4 nreg=3 # SVE 32-bit scatter store (scalar plus 32-bit scaled offsets) # Require msz > 0 && msz <= esz. diff --git a/target/arm/tcg/sve_helper.c b/target/arm/tcg/sve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve_helper.c +++ b/target/arm/tcg/sve_helper.c @@ -XXX,XX +XXX,XX @@ DO_LDN_2(2, dd, MO_64) DO_LDN_2(3, dd, MO_64) DO_LDN_2(4, dd, MO_64) +DO_LDN_2(2, qq, MO_128) +DO_LDN_2(3, qq, MO_128) +DO_LDN_2(4, qq, MO_128) + #undef DO_LDN_1 #undef DO_LDN_2 @@ -XXX,XX +XXX,XX @@ DO_STN_2(4, dd, MO_64, MO_64) DO_STN_2(1, sq, MO_128, MO_32) DO_STN_2(1, dq, MO_128, MO_64) +DO_STN_2(2, qq, MO_128, MO_128) +DO_STN_2(3, qq, MO_128, MO_128) +DO_STN_2(4, qq, MO_128, MO_128) + #undef DO_STN_1 #undef DO_STN_2 diff --git a/target/arm/tcg/translate-sve.c b/target/arm/tcg/translate-sve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sve.c +++ b/target/arm/tcg/translate-sve.c @@ -XXX,XX +XXX,XX @@ static inline int expand_imm_sh8u(DisasContext *s, int x) */ static inline int msz_dtype(DisasContext *s, int msz) { - static const uint8_t dtype[4] = { 0, 5, 10, 15 }; + static const uint8_t dtype[5] = { 0, 5, 10, 15, 18 }; return dtype[msz]; } @@ -XXX,XX +XXX,XX @@ static void do_mem_zpa(DisasContext *s, int zt, int pg, TCGv_i64 addr, } /* Indexed by [mte][be][dtype][nreg] */ -static gen_helper_gvec_mem * const ldr_fns[2][2][18][4] = { +static gen_helper_gvec_mem * const ldr_fns[2][2][19][4] = { { /* mte inactive, little-endian */ { { gen_helper_sve_ld1bb_r, gen_helper_sve_ld2bb_r, gen_helper_sve_ld3bb_r, gen_helper_sve_ld4bb_r }, @@ -XXX,XX +XXX,XX @@ static gen_helper_gvec_mem * const ldr_fns[2][2][18][4] = { { gen_helper_sve_ld1squ_le_r, NULL, NULL, NULL }, { gen_helper_sve_ld1dqu_le_r, NULL, NULL, NULL }, + { NULL, gen_helper_sve_ld2qq_le_r, + gen_helper_sve_ld3qq_le_r, gen_helper_sve_ld4qq_le_r }, }, /* mte inactive, big-endian */ @@ -XXX,XX +XXX,XX @@ static gen_helper_gvec_mem * const ldr_fns[2][2][18][4] = { { gen_helper_sve_ld1squ_be_r, NULL, NULL, NULL }, { gen_helper_sve_ld1dqu_be_r, NULL, NULL, NULL }, + { NULL, gen_helper_sve_ld2qq_be_r, + gen_helper_sve_ld3qq_be_r, gen_helper_sve_ld4qq_be_r }, }, }, @@ -XXX,XX +XXX,XX @@ static gen_helper_gvec_mem * const ldr_fns[2][2][18][4] = { { gen_helper_sve_ld1squ_le_r_mte, NULL, NULL, NULL }, { gen_helper_sve_ld1dqu_le_r_mte, NULL, NULL, NULL }, + { NULL, + gen_helper_sve_ld2qq_le_r_mte, + gen_helper_sve_ld3qq_le_r_mte, + gen_helper_sve_ld4qq_le_r_mte }, }, /* mte active, big-endian */ @@ -XXX,XX +XXX,XX @@ static gen_helper_gvec_mem * const ldr_fns[2][2][18][4] = { { gen_helper_sve_ld1squ_be_r_mte, NULL, NULL, NULL }, { gen_helper_sve_ld1dqu_be_r_mte, NULL, NULL, NULL }, + { NULL, + gen_helper_sve_ld2qq_be_r_mte, + gen_helper_sve_ld3qq_be_r_mte, + gen_helper_sve_ld4qq_be_r_mte }, }, }, }; @@ -XXX,XX +XXX,XX @@ static bool trans_LD_zprr(DisasContext *s, arg_rprr_load *a) return false; } - /* dtypes 16 and 17 are artificial, representing 128-bit element */ - if (a->dtype < 16) { + /* dtypes 16-18 are artificial, representing 128-bit element */ + switch (a->dtype) { + case 0 ... 15: if (!dc_isar_feature(aa64_sve, s)) { return false; } - } else { + break; + case 16: case 17: if (!dc_isar_feature(aa64_sve2p1, s)) { return false; } s->is_nonstreaming = true; + break; + case 18: + if (!dc_isar_feature(aa64_sme2p1_or_sve2p1, s)) { + return false; + } + break; + default: + g_assert_not_reached(); } if (sve_access_check(s)) { @@ -XXX,XX +XXX,XX @@ static bool trans_LD_zprr(DisasContext *s, arg_rprr_load *a) static bool trans_LD_zpri(DisasContext *s, arg_rpri_load *a) { - /* dtypes 16 and 17 are artificial, representing 128-bit element */ - if (a->dtype < 16) { + /* dtypes 16-18 are artificial, representing 128-bit element */ + switch (a->dtype) { + case 0 ... 15: if (!dc_isar_feature(aa64_sve, s)) { return false; } - } else { + break; + case 16: case 17: if (!dc_isar_feature(aa64_sve2p1, s)) { return false; } s->is_nonstreaming = true; + break; + case 18: + if (!dc_isar_feature(aa64_sme2p1_or_sve2p1, s)) { + return false; + } + break; + default: + g_assert_not_reached(); } if (sve_access_check(s)) { @@ -XXX,XX +XXX,XX @@ static void do_st_zpa(DisasContext *s, int zt, int pg, TCGv_i64 addr, gen_helper_sve_st1dd_be_r_mte, gen_helper_sve_st1dq_be_r_mte } } }, }; - static gen_helper_gvec_mem * const fn_multiple[2][2][3][4] = { + static gen_helper_gvec_mem * const fn_multiple[2][2][3][5] = { { { { gen_helper_sve_st2bb_r, gen_helper_sve_st2hh_le_r, gen_helper_sve_st2ss_le_r, - gen_helper_sve_st2dd_le_r }, + gen_helper_sve_st2dd_le_r, + gen_helper_sve_st2qq_le_r }, { gen_helper_sve_st3bb_r, gen_helper_sve_st3hh_le_r, gen_helper_sve_st3ss_le_r, - gen_helper_sve_st3dd_le_r }, + gen_helper_sve_st3dd_le_r, + gen_helper_sve_st3qq_le_r }, { gen_helper_sve_st4bb_r, gen_helper_sve_st4hh_le_r, gen_helper_sve_st4ss_le_r, - gen_helper_sve_st4dd_le_r } }, + gen_helper_sve_st4dd_le_r, + gen_helper_sve_st4qq_le_r } }, { { gen_helper_sve_st2bb_r, gen_helper_sve_st2hh_be_r, gen_helper_sve_st2ss_be_r, - gen_helper_sve_st2dd_be_r }, + gen_helper_sve_st2dd_be_r, + gen_helper_sve_st2qq_be_r }, { gen_helper_sve_st3bb_r, gen_helper_sve_st3hh_be_r, gen_helper_sve_st3ss_be_r, - gen_helper_sve_st3dd_be_r }, + gen_helper_sve_st3dd_be_r, + gen_helper_sve_st3qq_be_r }, { gen_helper_sve_st4bb_r, gen_helper_sve_st4hh_be_r, gen_helper_sve_st4ss_be_r, - gen_helper_sve_st4dd_be_r } } }, + gen_helper_sve_st4dd_be_r, + gen_helper_sve_st4qq_be_r } } }, { { { gen_helper_sve_st2bb_r_mte, gen_helper_sve_st2hh_le_r_mte, gen_helper_sve_st2ss_le_r_mte, - gen_helper_sve_st2dd_le_r_mte }, + gen_helper_sve_st2dd_le_r_mte, + gen_helper_sve_st2qq_le_r_mte }, { gen_helper_sve_st3bb_r_mte, gen_helper_sve_st3hh_le_r_mte, gen_helper_sve_st3ss_le_r_mte, - gen_helper_sve_st3dd_le_r_mte }, + gen_helper_sve_st3dd_le_r_mte, + gen_helper_sve_st3qq_le_r_mte }, { gen_helper_sve_st4bb_r_mte, gen_helper_sve_st4hh_le_r_mte, gen_helper_sve_st4ss_le_r_mte, - gen_helper_sve_st4dd_le_r_mte } }, + gen_helper_sve_st4dd_le_r_mte, + gen_helper_sve_st4qq_le_r_mte } }, { { gen_helper_sve_st2bb_r_mte, gen_helper_sve_st2hh_be_r_mte, gen_helper_sve_st2ss_be_r_mte, - gen_helper_sve_st2dd_be_r_mte }, + gen_helper_sve_st2dd_be_r_mte, + gen_helper_sve_st2qq_be_r_mte }, { gen_helper_sve_st3bb_r_mte, gen_helper_sve_st3hh_be_r_mte, gen_helper_sve_st3ss_be_r_mte, - gen_helper_sve_st3dd_be_r_mte }, + gen_helper_sve_st3dd_be_r_mte, + gen_helper_sve_st3qq_be_r_mte }, { gen_helper_sve_st4bb_r_mte, gen_helper_sve_st4hh_be_r_mte, gen_helper_sve_st4ss_be_r_mte, - gen_helper_sve_st4dd_be_r_mte } } }, + gen_helper_sve_st4dd_be_r_mte, + gen_helper_sve_st4qq_be_r_mte } } }, }; gen_helper_gvec_mem *fn; int be = s->be_data == MO_BE; @@ -XXX,XX +XXX,XX @@ static bool trans_ST_zprr(DisasContext *s, arg_rprr_store *a) } break; case MO_128: - assert(a->msz < a->esz); - assert(a->nreg == 0); - if (!dc_isar_feature(aa64_sve2p1, s)) { - return false; + if (a->nreg == 0) { + assert(a->msz < a->esz); + if (!dc_isar_feature(aa64_sve2p1, s)) { + return false; + } + s->is_nonstreaming = true; + } else { + if (!dc_isar_feature(aa64_sme2p1_or_sve2p1, s)) { + return false; + } } - s->is_nonstreaming = true; break; default: g_assert_not_reached(); @@ -XXX,XX +XXX,XX @@ static bool trans_ST_zpri(DisasContext *s, arg_rpri_store *a) } break; case MO_128: - assert(a->msz < a->esz); - assert(a->nreg == 0); - if (!dc_isar_feature(aa64_sve2p1, s)) { - return false; + if (a->nreg == 0) { + assert(a->msz < a->esz); + if (!dc_isar_feature(aa64_sve2p1, s)) { + return false; + } + s->is_nonstreaming = true; + } else { + if (!dc_isar_feature(aa64_sme2p1_or_sve2p1, s)) { + return false; + } } - s->is_nonstreaming = true; break; default: g_assert_not_reached(); -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-99-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper-sve.h | 16 ++++++++++++++++ target/arm/tcg/sve.decode | 8 ++++++++ target/arm/tcg/sve_helper.c | 6 ++++++ target/arm/tcg/translate-sve.c | 34 ++++++++++++++++++++++++++++++++-- 4 files changed, 62 insertions(+), 2 deletions(-) diff --git a/target/arm/tcg/helper-sve.h b/target/arm/tcg/helper-sve.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-sve.h +++ b/target/arm/tcg/helper-sve.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_6(sve_ldsds_le_zd, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr, tl, i32) DEF_HELPER_FLAGS_6(sve_ldsds_be_zd, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr, tl, i32) +DEF_HELPER_FLAGS_6(sve_ldqq_le_zd, TCG_CALL_NO_WG, + void, env, ptr, ptr, ptr, tl, i32) +DEF_HELPER_FLAGS_6(sve_ldqq_be_zd, TCG_CALL_NO_WG, + void, env, ptr, ptr, ptr, tl, i32) DEF_HELPER_FLAGS_6(sve_ldbsu_zsu_mte, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr, tl, i32) @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_6(sve_ldsds_le_zd_mte, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr, tl, i32) DEF_HELPER_FLAGS_6(sve_ldsds_be_zd_mte, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr, tl, i32) +DEF_HELPER_FLAGS_6(sve_ldqq_le_zd_mte, TCG_CALL_NO_WG, + void, env, ptr, ptr, ptr, tl, i32) +DEF_HELPER_FLAGS_6(sve_ldqq_be_zd_mte, TCG_CALL_NO_WG, + void, env, ptr, ptr, ptr, tl, i32) DEF_HELPER_FLAGS_6(sve_ldffbsu_zsu, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr, tl, i32) @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_6(sve_stdd_le_zd, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr, tl, i32) DEF_HELPER_FLAGS_6(sve_stdd_be_zd, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr, tl, i32) +DEF_HELPER_FLAGS_6(sve_stqq_le_zd, TCG_CALL_NO_WG, + void, env, ptr, ptr, ptr, tl, i32) +DEF_HELPER_FLAGS_6(sve_stqq_be_zd, TCG_CALL_NO_WG, + void, env, ptr, ptr, ptr, tl, i32) DEF_HELPER_FLAGS_6(sve_stbs_zsu_mte, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr, tl, i32) @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_6(sve_stdd_le_zd_mte, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr, tl, i32) DEF_HELPER_FLAGS_6(sve_stdd_be_zd_mte, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr, tl, i32) +DEF_HELPER_FLAGS_6(sve_stqq_le_zd_mte, TCG_CALL_NO_WG, + void, env, ptr, ptr, ptr, tl, i32) +DEF_HELPER_FLAGS_6(sve_stqq_be_zd_mte, TCG_CALL_NO_WG, + void, env, ptr, ptr, ptr, tl, i32) DEF_HELPER_FLAGS_4(sve2_sqdmull_zzz_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) diff --git a/target/arm/tcg/sve.decode b/target/arm/tcg/sve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve.decode +++ b/target/arm/tcg/sve.decode @@ -XXX,XX +XXX,XX @@ LD1_zprz 1100010 10 1. ..... 1.. ... ..... ..... \ LD1_zprz 1100010 11 1. ..... 11. ... ..... ..... \ @rprr_g_load_sc esz=3 msz=3 u=1 +# LD1Q +LD1_zprz 1100 0100 000 rm:5 101 pg:3 rn:5 rd:5 \ + &rprr_gather_load u=0 ff=0 xs=2 esz=4 msz=4 scale=0 + # SVE 64-bit gather load (vector plus immediate) LD1_zpiz 1100010 .. 01 ..... 1.. ... ..... ..... \ @rpri_g_load esz=3 @@ -XXX,XX +XXX,XX @@ ST1_zprz 1110010 .. 01 ..... 101 ... ..... ..... \ ST1_zprz 1110010 .. 00 ..... 101 ... ..... ..... \ @rprr_scatter_store xs=2 esz=3 scale=0 +# ST1Q +ST1_zprz 1110 0100 001 rm:5 001 pg:3 rn:5 rd:5 \ + &rprr_scatter_store xs=2 msz=4 esz=4 scale=0 + # SVE 64-bit scatter store (vector plus immediate) ST1_zpiz 1110010 .. 10 ..... 101 ... ..... ..... \ @rpri_scatter_store esz=3 diff --git a/target/arm/tcg/sve_helper.c b/target/arm/tcg/sve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sve_helper.c +++ b/target/arm/tcg/sve_helper.c @@ -XXX,XX +XXX,XX @@ DO_LD1_ZPZ_D(dd_be, zsu, MO_64) DO_LD1_ZPZ_D(dd_be, zss, MO_64) DO_LD1_ZPZ_D(dd_be, zd, MO_64) +DO_LD1_ZPZ_D(qq_le, zd, MO_128) +DO_LD1_ZPZ_D(qq_be, zd, MO_128) + #undef DO_LD1_ZPZ_S #undef DO_LD1_ZPZ_D @@ -XXX,XX +XXX,XX @@ DO_ST1_ZPZ_D(sd_be, zd, MO_32) DO_ST1_ZPZ_D(dd_le, zd, MO_64) DO_ST1_ZPZ_D(dd_be, zd, MO_64) +DO_ST1_ZPZ_D(qq_le, zd, MO_128) +DO_ST1_ZPZ_D(qq_be, zd, MO_128) + #undef DO_ST1_ZPZ_S #undef DO_ST1_ZPZ_D diff --git a/target/arm/tcg/translate-sve.c b/target/arm/tcg/translate-sve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sve.c +++ b/target/arm/tcg/translate-sve.c @@ -XXX,XX +XXX,XX @@ gather_load_fn64[2][2][2][3][2][4] = { gen_helper_sve_ldffdd_be_zd_mte, } } } } }, }; +static gen_helper_gvec_mem_scatter * const +gather_load_fn128[2][2] = { + { gen_helper_sve_ldqq_le_zd, + gen_helper_sve_ldqq_be_zd }, + { gen_helper_sve_ldqq_le_zd_mte, + gen_helper_sve_ldqq_be_zd_mte } +}; + static bool trans_LD1_zprz(DisasContext *s, arg_LD1_zprz *a) { gen_helper_gvec_mem_scatter *fn = NULL; bool be = s->be_data == MO_BE; bool mte = s->mte_active[0]; - if (!dc_isar_feature(aa64_sve, s)) { + if (a->esz < MO_128 + ? !dc_isar_feature(aa64_sve, s) + : !dc_isar_feature(aa64_sve2p1, s)) { return false; } s->is_nonstreaming = true; @@ -XXX,XX +XXX,XX @@ static bool trans_LD1_zprz(DisasContext *s, arg_LD1_zprz *a) case MO_64: fn = gather_load_fn64[mte][be][a->ff][a->xs][a->u][a->msz]; break; + case MO_128: + assert(!a->ff && a->u && a->xs == 2 && a->msz == MO_128); + fn = gather_load_fn128[mte][be]; + break; + default: + g_assert_not_reached(); } assert(fn != NULL); @@ -XXX,XX +XXX,XX @@ static gen_helper_gvec_mem_scatter * const scatter_store_fn64[2][2][3][4] = { gen_helper_sve_stdd_be_zd_mte, } } }, }; +static gen_helper_gvec_mem_scatter * const +scatter_store_fn128[2][2] = { + { gen_helper_sve_stqq_le_zd, + gen_helper_sve_stqq_be_zd }, + { gen_helper_sve_stqq_le_zd_mte, + gen_helper_sve_stqq_be_zd_mte } +}; + static bool trans_ST1_zprz(DisasContext *s, arg_ST1_zprz *a) { gen_helper_gvec_mem_scatter *fn; @@ -XXX,XX +XXX,XX @@ static bool trans_ST1_zprz(DisasContext *s, arg_ST1_zprz *a) if (a->esz < a->msz || (a->msz == 0 && a->scale)) { return false; } - if (!dc_isar_feature(aa64_sve, s)) { + if (a->esz < MO_128 + ? !dc_isar_feature(aa64_sve, s) + : !dc_isar_feature(aa64_sve2p1, s)) { return false; } s->is_nonstreaming = true; @@ -XXX,XX +XXX,XX @@ static bool trans_ST1_zprz(DisasContext *s, arg_ST1_zprz *a) case MO_64: fn = scatter_store_fn64[mte][be][a->xs][a->msz]; break; + case MO_128: + assert(a->xs == 2 && a->msz == MO_128); + fn = scatter_store_fn128[mte][be]; + break; default: g_assert_not_reached(); } -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-100-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper-sme.h | 6 ++++ target/arm/tcg/sme.decode | 36 ++++++++++++++++++++ target/arm/tcg/sme_helper.c | 60 ++++++++++++++++++++++++++++++++++ target/arm/tcg/translate-sme.c | 46 +++++++++++++++++++------- 4 files changed, 137 insertions(+), 11 deletions(-) diff --git a/target/arm/tcg/helper-sme.h b/target/arm/tcg/helper-sme.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-sme.h +++ b/target/arm/tcg/helper-sme.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_3(sme2_mova_zc_s, TCG_CALL_NO_RWG, void, ptr, ptr, i32) DEF_HELPER_FLAGS_3(sme2_mova_cz_d, TCG_CALL_NO_RWG, void, ptr, ptr, i32) DEF_HELPER_FLAGS_3(sme2_mova_zc_d, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2p1_movaz_zc_b, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2p1_movaz_zc_h, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2p1_movaz_zc_s, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2p1_movaz_zc_d, TCG_CALL_NO_RWG, void, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(sme2p1_movaz_zc_q, TCG_CALL_NO_RWG, void, ptr, ptr, i32) + DEF_HELPER_FLAGS_5(sme_ld1b_h, TCG_CALL_NO_WG, void, env, ptr, ptr, tl, i32) DEF_HELPER_FLAGS_5(sme_ld1b_v, TCG_CALL_NO_WG, void, env, ptr, ptr, tl, i32) DEF_HELPER_FLAGS_5(sme_ld1b_h_mte, TCG_CALL_NO_WG, void, env, ptr, ptr, tl, i32) diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ MOVA_za2 11000000 00 00011 00 .. 010 00 off:3 zr:4 0 \ MOVA_za4 11000000 00 00011 00 .. 011 00 off:3 zr:3 00 \ &mova_a rv=%mova_rv +### SME Move and Zero + +MOVAZ_za2 11000000 00000110 0 .. 01010 off:3 zr:4 0 \ + &mova_a rv=%mova_rv +MOVAZ_za4 11000000 00000110 0 .. 01110 off:3 zr:3 00 \ + &mova_a rv=%mova_rv + +MOVAZ_zt 11000000 00 00001 0 v:1 .. 0001 off:4 zr:5 \ + &mova_t rs=%mova_rs esz=0 za=0 +MOVAZ_zt 11000000 01 00001 0 v:1 .. 0001 za:1 off:3 zr:5 \ + &mova_t rs=%mova_rs esz=1 +MOVAZ_zt 11000000 10 00001 0 v:1 .. 0001 za:2 off:2 zr:5 \ + &mova_t rs=%mova_rs esz=2 +MOVAZ_zt 11000000 11 00001 0 v:1 .. 0001 za:3 off:1 zr:5 \ + &mova_t rs=%mova_rs esz=3 +MOVAZ_zt 11000000 11 00001 1 v:1 .. 0001 za:4 zr:5 \ + &mova_t rs=%mova_rs esz=4 off=0 + +MOVAZ_zt2 11000000 00 00011 0 v:1 .. 00010 off:3 zr:4 0 \ + &mova_t rs=%mova_rs esz=0 za=0 +MOVAZ_zt2 11000000 01 00011 0 v:1 .. 00010 za:1 off:2 zr:4 0 \ + &mova_t rs=%mova_rs esz=1 +MOVAZ_zt2 11000000 10 00011 0 v:1 .. 00010 za:2 off:1 zr:4 0 \ + &mova_t rs=%mova_rs esz=2 +MOVAZ_zt2 11000000 11 00011 0 v:1 .. 00010 za:3 zr:4 0 \ + &mova_t rs=%mova_rs esz=3 off=0 + +MOVAZ_zt4 11000000 00 00011 0 v:1 .. 001100 off:2 zr:3 00 \ + &mova_t rs=%mova_rs esz=0 za=0 +MOVAZ_zt4 11000000 01 00011 0 v:1 .. 001100 za:1 off:1 zr:3 00 \ + &mova_t rs=%mova_rs esz=1 +MOVAZ_zt4 11000000 10 00011 0 v:1 .. 001100 za:2 zr:3 00 \ + &mova_t rs=%mova_rs esz=2 off=0 +MOVAZ_zt4 11000000 11 00011 0 v:1 .. 00110 za:3 zr:3 00 \ + &mova_t rs=%mova_rs esz=3 off=0 + ### SME Move into/from ZT0 MOVT_rzt 1100 0000 0100 1100 0 off:3 00 11111 rt:5 diff --git a/target/arm/tcg/sme_helper.c b/target/arm/tcg/sme_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme_helper.c +++ b/target/arm/tcg/sme_helper.c @@ -XXX,XX +XXX,XX @@ void HELPER(sme2_mova_zc_d)(void *vdst, void *vsrc, uint32_t desc) } } +void HELPER(sme2p1_movaz_zc_b)(void *vdst, void *vsrc, uint32_t desc) +{ + uint8_t *src = vsrc; + uint8_t *dst = vdst; + size_t i, n = simd_oprsz(desc); + + for (i = 0; i < n; ++i) { + dst[i] = src[tile_vslice_index(i)]; + src[tile_vslice_index(i)] = 0; + } +} + +void HELPER(sme2p1_movaz_zc_h)(void *vdst, void *vsrc, uint32_t desc) +{ + uint16_t *src = vsrc; + uint16_t *dst = vdst; + size_t i, n = simd_oprsz(desc) / 2; + + for (i = 0; i < n; ++i) { + dst[i] = src[tile_vslice_index(i)]; + src[tile_vslice_index(i)] = 0; + } +} + +void HELPER(sme2p1_movaz_zc_s)(void *vdst, void *vsrc, uint32_t desc) +{ + uint32_t *src = vsrc; + uint32_t *dst = vdst; + size_t i, n = simd_oprsz(desc) / 4; + + for (i = 0; i < n; ++i) { + dst[i] = src[tile_vslice_index(i)]; + src[tile_vslice_index(i)] = 0; + } +} + +void HELPER(sme2p1_movaz_zc_d)(void *vdst, void *vsrc, uint32_t desc) +{ + uint64_t *src = vsrc; + uint64_t *dst = vdst; + size_t i, n = simd_oprsz(desc) / 8; + + for (i = 0; i < n; ++i) { + dst[i] = src[tile_vslice_index(i)]; + src[tile_vslice_index(i)] = 0; + } +} + +void HELPER(sme2p1_movaz_zc_q)(void *vdst, void *vsrc, uint32_t desc) +{ + Int128 *src = vsrc; + Int128 *dst = vdst; + size_t i, n = simd_oprsz(desc) / 16; + + for (i = 0; i < n; ++i) { + dst[i] = src[tile_vslice_index(i)]; + memset(&src[tile_vslice_index(i)], 0, 16); + } +} + /* * Clear elements in a tile slice comprising len bytes. */ diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ static bool do_mova_tile(DisasContext *s, arg_mova_p *a, bool to_vec) TRANS_FEAT(MOVA_tz, aa64_sme, do_mova_tile, a, false) TRANS_FEAT(MOVA_zt, aa64_sme, do_mova_tile, a, true) -static bool do_mova_tile_n(DisasContext *s, arg_mova_t *a, int n, bool to_vec) +static bool do_mova_tile_n(DisasContext *s, arg_mova_t *a, int n, + bool to_vec, bool zero) { static gen_helper_gvec_2 * const cz_fns[] = { gen_helper_sme2_mova_cz_b, gen_helper_sme2_mova_cz_h, @@ -XXX,XX +XXX,XX @@ static bool do_mova_tile_n(DisasContext *s, arg_mova_t *a, int n, bool to_vec) gen_helper_sme2_mova_zc_b, gen_helper_sme2_mova_zc_h, gen_helper_sme2_mova_zc_s, gen_helper_sme2_mova_zc_d, }; + static gen_helper_gvec_2 * const zc_z_fns[] = { + gen_helper_sme2p1_movaz_zc_b, gen_helper_sme2p1_movaz_zc_h, + gen_helper_sme2p1_movaz_zc_s, gen_helper_sme2p1_movaz_zc_d, + gen_helper_sme2p1_movaz_zc_q, + }; TCGv_ptr t_za; int svl, bytes_per_op = n << a->esz; @@ -XXX,XX +XXX,XX @@ static bool do_mova_tile_n(DisasContext *s, arg_mova_t *a, int n, bool to_vec) return true; } + assert(a->esz <= MO_64 + zero); + if (!sme_smza_enabled_check(s)) { return true; } @@ -XXX,XX +XXX,XX @@ static bool do_mova_tile_n(DisasContext *s, arg_mova_t *a, int n, bool to_vec) TCGv_ptr t_zr = vec_full_reg_ptr(s, a->zr * n + i); t_za = get_tile_rowcol(s, a->esz, a->rs, a->za, a->off * n + i, 1, n, a->v); - if (to_vec) { + if (zero) { + zc_z_fns[a->esz](t_zr, t_za, t_desc); + } else if (to_vec) { zc_fns[a->esz](t_zr, t_za, t_desc); } else { cz_fns[a->esz](t_za, t_zr, t_desc); @@ -XXX,XX +XXX,XX @@ static bool do_mova_tile_n(DisasContext *s, arg_mova_t *a, int n, bool to_vec) a->off * n + i, 1, n, a->v); if (to_vec) { tcg_gen_gvec_mov_var(MO_8, tcg_env, o_zr, t_za, 0, svl, svl); + if (zero) { + tcg_gen_gvec_dup_imm_var(MO_8, t_za, 0, svl, svl, 0); + } } else { tcg_gen_gvec_mov_var(MO_8, t_za, 0, tcg_env, o_zr, svl, svl); } @@ -XXX,XX +XXX,XX @@ static bool do_mova_tile_n(DisasContext *s, arg_mova_t *a, int n, bool to_vec) return true; } -TRANS_FEAT(MOVA_tz2, aa64_sme2, do_mova_tile_n, a, 2, false) -TRANS_FEAT(MOVA_tz4, aa64_sme2, do_mova_tile_n, a, 4, false) -TRANS_FEAT(MOVA_zt2, aa64_sme2, do_mova_tile_n, a, 2, true) -TRANS_FEAT(MOVA_zt4, aa64_sme2, do_mova_tile_n, a, 4, true) +TRANS_FEAT(MOVA_tz2, aa64_sme2, do_mova_tile_n, a, 2, false, false) +TRANS_FEAT(MOVA_tz4, aa64_sme2, do_mova_tile_n, a, 4, false, false) +TRANS_FEAT(MOVA_zt2, aa64_sme2, do_mova_tile_n, a, 2, true, false) +TRANS_FEAT(MOVA_zt4, aa64_sme2, do_mova_tile_n, a, 4, true, false) -static bool do_mova_array_n(DisasContext *s, arg_mova_a *a, int n, bool to_vec) +TRANS_FEAT(MOVAZ_zt, aa64_sme2p1, do_mova_tile_n, a, 1, true, true) +TRANS_FEAT(MOVAZ_zt2, aa64_sme2p1, do_mova_tile_n, a, 2, true, true) +TRANS_FEAT(MOVAZ_zt4, aa64_sme2p1, do_mova_tile_n, a, 4, true, true) + +static bool do_mova_array_n(DisasContext *s, arg_mova_a *a, int n, + bool to_vec, bool zero) { TCGv_ptr t_za; int svl; @@ -XXX,XX +XXX,XX @@ static bool do_mova_array_n(DisasContext *s, arg_mova_a *a, int n, bool to_vec) if (to_vec) { tcg_gen_gvec_mov_var(MO_8, tcg_env, o_zr, t_za, o_za, svl, svl); + if (zero) { + tcg_gen_gvec_dup_imm_var(MO_8, t_za, o_za, svl, svl, 0); + } } else { tcg_gen_gvec_mov_var(MO_8, t_za, o_za, tcg_env, o_zr, svl, svl); } @@ -XXX,XX +XXX,XX @@ static bool do_mova_array_n(DisasContext *s, arg_mova_a *a, int n, bool to_vec) return true; } -TRANS_FEAT(MOVA_az2, aa64_sme2, do_mova_array_n, a, 2, false) -TRANS_FEAT(MOVA_az4, aa64_sme2, do_mova_array_n, a, 4, false) -TRANS_FEAT(MOVA_za2, aa64_sme2, do_mova_array_n, a, 2, true) -TRANS_FEAT(MOVA_za4, aa64_sme2, do_mova_array_n, a, 4, true) +TRANS_FEAT(MOVA_az2, aa64_sme2, do_mova_array_n, a, 2, false, false) +TRANS_FEAT(MOVA_az4, aa64_sme2, do_mova_array_n, a, 4, false, false) +TRANS_FEAT(MOVA_za2, aa64_sme2, do_mova_array_n, a, 2, true, false) +TRANS_FEAT(MOVA_za4, aa64_sme2, do_mova_array_n, a, 4, true, false) + +TRANS_FEAT(MOVAZ_za2, aa64_sme2p1, do_mova_array_n, a, 2, true, true) +TRANS_FEAT(MOVAZ_za4, aa64_sme2p1, do_mova_array_n, a, 4, true, true) static bool do_movt(DisasContext *s, arg_MOVT_rzt *a, void (*func)(TCGv_i64, TCGv_ptr, tcg_target_long)) -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-101-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper.h | 24 ++++++++++ target/arm/tcg/sme.decode | 42 ++++++++++++++++ target/arm/tcg/translate-sme.c | 56 ++++++++++++++++++++++ target/arm/tcg/vec_helper.c | 88 ++++++++++++++++++++++++++++++++++ 4 files changed, 210 insertions(+) diff --git a/target/arm/tcg/helper.h b/target/arm/tcg/helper.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper.h +++ b/target/arm/tcg/helper.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_4(gvec_uminp_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_3(gvec_urecpe_s, TCG_CALL_NO_RWG, void, ptr, ptr, i32) DEF_HELPER_FLAGS_3(gvec_ursqrte_s, TCG_CALL_NO_RWG, void, ptr, ptr, i32) + +DEF_HELPER_FLAGS_4(sme2_luti2_1b, TCG_CALL_NO_RWG, void, ptr, ptr, env, i32) +DEF_HELPER_FLAGS_4(sme2_luti2_1h, TCG_CALL_NO_RWG, void, ptr, ptr, env, i32) +DEF_HELPER_FLAGS_4(sme2_luti2_1s, TCG_CALL_NO_RWG, void, ptr, ptr, env, i32) + +DEF_HELPER_FLAGS_4(sme2_luti2_2b, TCG_CALL_NO_RWG, void, ptr, ptr, env, i32) +DEF_HELPER_FLAGS_4(sme2_luti2_2h, TCG_CALL_NO_RWG, void, ptr, ptr, env, i32) +DEF_HELPER_FLAGS_4(sme2_luti2_2s, TCG_CALL_NO_RWG, void, ptr, ptr, env, i32) + +DEF_HELPER_FLAGS_4(sme2_luti2_4b, TCG_CALL_NO_RWG, void, ptr, ptr, env, i32) +DEF_HELPER_FLAGS_4(sme2_luti2_4h, TCG_CALL_NO_RWG, void, ptr, ptr, env, i32) +DEF_HELPER_FLAGS_4(sme2_luti2_4s, TCG_CALL_NO_RWG, void, ptr, ptr, env, i32) + +DEF_HELPER_FLAGS_4(sme2_luti4_1b, TCG_CALL_NO_RWG, void, ptr, ptr, env, i32) +DEF_HELPER_FLAGS_4(sme2_luti4_1h, TCG_CALL_NO_RWG, void, ptr, ptr, env, i32) +DEF_HELPER_FLAGS_4(sme2_luti4_1s, TCG_CALL_NO_RWG, void, ptr, ptr, env, i32) + +DEF_HELPER_FLAGS_4(sme2_luti4_2b, TCG_CALL_NO_RWG, void, ptr, ptr, env, i32) +DEF_HELPER_FLAGS_4(sme2_luti4_2h, TCG_CALL_NO_RWG, void, ptr, ptr, env, i32) +DEF_HELPER_FLAGS_4(sme2_luti4_2s, TCG_CALL_NO_RWG, void, ptr, ptr, env, i32) + +DEF_HELPER_FLAGS_4(sme2_luti4_4b, TCG_CALL_NO_RWG, void, ptr, ptr, env, i32) +DEF_HELPER_FLAGS_4(sme2_luti4_4h, TCG_CALL_NO_RWG, void, ptr, ptr, env, i32) +DEF_HELPER_FLAGS_4(sme2_luti4_4s, TCG_CALL_NO_RWG, void, ptr, ptr, env, i32) diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ ZERO_za 11000000 000011 110 .. 0000000000 00. \ &zero_za ngrp=2 nvec=4 rv=%mova_rv off=%off1_x4 ZERO_za 11000000 000011 111 .. 0000000000 00. \ &zero_za ngrp=4 nvec=4 rv=%mova_rv off=%off1_x4 + +### SME Lookup Table Read + +&lut zd zn idx + +# LUTI2, consecutive +LUTI2_c_1b 1100 0000 1100 11 idx:4 00 00 zn:5 zd:5 &lut +LUTI2_c_1h 1100 0000 1100 11 idx:4 01 00 zn:5 zd:5 &lut +LUTI2_c_1s 1100 0000 1100 11 idx:4 10 00 zn:5 zd:5 &lut + +LUTI2_c_2b 1100 0000 1000 11 idx:3 1 00 00 zn:5 .... 0 &lut zd=%zd_ax2 +LUTI2_c_2h 1100 0000 1000 11 idx:3 1 01 00 zn:5 .... 0 &lut zd=%zd_ax2 +LUTI2_c_2s 1100 0000 1000 11 idx:3 1 10 00 zn:5 .... 0 &lut zd=%zd_ax2 + +LUTI2_c_4b 1100 0000 1000 11 idx:2 10 00 00 zn:5 ... 00 &lut zd=%zd_ax4 +LUTI2_c_4h 1100 0000 1000 11 idx:2 10 01 00 zn:5 ... 00 &lut zd=%zd_ax4 +LUTI2_c_4s 1100 0000 1000 11 idx:2 10 10 00 zn:5 ... 00 &lut zd=%zd_ax4 + +# LUTI2, strided (must check zd alignment) +LUTI2_s_2b 1100 0000 1001 11 idx:3 1 00 00 zn:5 zd:5 &lut +LUTI2_s_2h 1100 0000 1001 11 idx:3 1 01 00 zn:5 zd:5 &lut + +LUTI2_s_4b 1100 0000 1001 11 idx:2 10 00 00 zn:5 zd:5 &lut +LUTI2_s_4h 1100 0000 1001 11 idx:2 10 01 00 zn:5 zd:5 &lut + +# LUTI4, consecutive +LUTI4_c_1b 1100 0000 1100 101 idx:3 00 00 zn:5 zd:5 &lut +LUTI4_c_1h 1100 0000 1100 101 idx:3 01 00 zn:5 zd:5 &lut +LUTI4_c_1s 1100 0000 1100 101 idx:3 10 00 zn:5 zd:5 &lut + +LUTI4_c_2b 1100 0000 1000 101 idx:2 1 00 00 zn:5 .... 0 &lut zd=%zd_ax2 +LUTI4_c_2h 1100 0000 1000 101 idx:2 1 01 00 zn:5 .... 0 &lut zd=%zd_ax2 +LUTI4_c_2s 1100 0000 1000 101 idx:2 1 10 00 zn:5 .... 0 &lut zd=%zd_ax2 + +LUTI4_c_4h 1100 0000 1000 101 idx:1 10 01 00 zn:5 ... 00 &lut zd=%zd_ax4 +LUTI4_c_4s 1100 0000 1000 101 idx:1 10 10 00 zn:5 ... 00 &lut zd=%zd_ax4 + +# LUTI4, strided (must check zd alignment) +LUTI4_s_2b 1100 0000 1001 101 idx:2 1 00 00 zn:5 zd:5 &lut +LUTI4_s_2h 1100 0000 1001 101 idx:2 1 01 00 zn:5 zd:5 &lut + +LUTI4_s_4h 1100 0000 1001 101 idx:1 10 01 00 zn:5 zd:5 &lut diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ static bool trans_SEL(DisasContext *s, arg_SEL *a) } return true; } + +static bool do_lut(DisasContext *s, arg_lut *a, + gen_helper_gvec_2_ptr *fn, bool strided) +{ + if (sme_sm_enabled_check(s) && sme2_zt0_enabled_check(s)) { + int svl = streaming_vec_reg_size(s); + tcg_gen_gvec_2_ptr(vec_full_reg_offset(s, a->zd), + vec_full_reg_offset(s, a->zn), + tcg_env, svl, svl, strided | (a->idx << 1), fn); + } + return true; +} + +TRANS_FEAT(LUTI2_c_1b, aa64_sme2, do_lut, a, gen_helper_sme2_luti2_1b, false) +TRANS_FEAT(LUTI2_c_1h, aa64_sme2, do_lut, a, gen_helper_sme2_luti2_1h, false) +TRANS_FEAT(LUTI2_c_1s, aa64_sme2, do_lut, a, gen_helper_sme2_luti2_1s, false) + +TRANS_FEAT(LUTI2_c_2b, aa64_sme2, do_lut, a, gen_helper_sme2_luti2_2b, false) +TRANS_FEAT(LUTI2_c_2h, aa64_sme2, do_lut, a, gen_helper_sme2_luti2_2h, false) +TRANS_FEAT(LUTI2_c_2s, aa64_sme2, do_lut, a, gen_helper_sme2_luti2_2s, false) + +TRANS_FEAT(LUTI2_c_4b, aa64_sme2, do_lut, a, gen_helper_sme2_luti2_4b, false) +TRANS_FEAT(LUTI2_c_4h, aa64_sme2, do_lut, a, gen_helper_sme2_luti2_4h, false) +TRANS_FEAT(LUTI2_c_4s, aa64_sme2, do_lut, a, gen_helper_sme2_luti2_4s, false) + +TRANS_FEAT(LUTI4_c_1b, aa64_sme2, do_lut, a, gen_helper_sme2_luti4_1b, false) +TRANS_FEAT(LUTI4_c_1h, aa64_sme2, do_lut, a, gen_helper_sme2_luti4_1h, false) +TRANS_FEAT(LUTI4_c_1s, aa64_sme2, do_lut, a, gen_helper_sme2_luti4_1s, false) + +TRANS_FEAT(LUTI4_c_2b, aa64_sme2, do_lut, a, gen_helper_sme2_luti4_2b, false) +TRANS_FEAT(LUTI4_c_2h, aa64_sme2, do_lut, a, gen_helper_sme2_luti4_2h, false) +TRANS_FEAT(LUTI4_c_2s, aa64_sme2, do_lut, a, gen_helper_sme2_luti4_2s, false) + +TRANS_FEAT(LUTI4_c_4h, aa64_sme2, do_lut, a, gen_helper_sme2_luti4_4h, false) +TRANS_FEAT(LUTI4_c_4s, aa64_sme2, do_lut, a, gen_helper_sme2_luti4_4s, false) + +static bool do_lut_s4(DisasContext *s, arg_lut *a, gen_helper_gvec_2_ptr *fn) +{ + return !(a->zd & 0b01100) && do_lut(s, a, fn, true); +} + +static bool do_lut_s8(DisasContext *s, arg_lut *a, gen_helper_gvec_2_ptr *fn) +{ + return !(a->zd & 0b01000) && do_lut(s, a, fn, true); +} + +TRANS_FEAT(LUTI2_s_2b, aa64_sme2p1, do_lut_s8, a, gen_helper_sme2_luti2_2b) +TRANS_FEAT(LUTI2_s_2h, aa64_sme2p1, do_lut_s8, a, gen_helper_sme2_luti2_2h) + +TRANS_FEAT(LUTI2_s_4b, aa64_sme2p1, do_lut_s4, a, gen_helper_sme2_luti2_4b) +TRANS_FEAT(LUTI2_s_4h, aa64_sme2p1, do_lut_s4, a, gen_helper_sme2_luti2_4h) + +TRANS_FEAT(LUTI4_s_2b, aa64_sme2p1, do_lut_s8, a, gen_helper_sme2_luti4_2b) +TRANS_FEAT(LUTI4_s_2h, aa64_sme2p1, do_lut_s8, a, gen_helper_sme2_luti4_2h) + +TRANS_FEAT(LUTI4_s_4h, aa64_sme2p1, do_lut_s4, a, gen_helper_sme2_luti4_4h) diff --git a/target/arm/tcg/vec_helper.c b/target/arm/tcg/vec_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/vec_helper.c +++ b/target/arm/tcg/vec_helper.c @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_ursqrte_s)(void *vd, void *vn, uint32_t desc) } clear_tail(d, opr_sz, simd_maxsz(desc)); } + +static inline void do_lut_b(void *zd, uint64_t *indexes, uint64_t *table, + unsigned elements, unsigned segbase, + unsigned dstride, unsigned isize, + unsigned tsize, unsigned nreg) +{ + for (unsigned r = 0; r < nreg; ++r) { + uint8_t *dst = zd + dstride * r; + unsigned base = segbase + r * elements; + + for (unsigned e = 0; e < elements; ++e) { + unsigned index = extractn(indexes, (base + e) * isize, isize); + dst[H1(e)] = extractn(table, index * tsize, 8); + } + } +} + +static inline void do_lut_h(void *zd, uint64_t *indexes, uint64_t *table, + unsigned elements, unsigned segbase, + unsigned dstride, unsigned isize, + unsigned tsize, unsigned nreg) +{ + for (unsigned r = 0; r < nreg; ++r) { + uint16_t *dst = zd + dstride * r; + unsigned base = segbase + r * elements; + + for (unsigned e = 0; e < elements; ++e) { + unsigned index = extractn(indexes, (base + e) * isize, isize); + dst[H2(e)] = extractn(table, index * tsize, 16); + } + } +} + +static inline void do_lut_s(void *zd, uint64_t *indexes, uint32_t *table, + unsigned elements, unsigned segbase, + unsigned dstride, unsigned isize, + unsigned tsize, unsigned nreg) +{ + for (unsigned r = 0; r < nreg; ++r) { + uint32_t *dst = zd + dstride * r; + unsigned base = segbase + r * elements; + + for (unsigned e = 0; e < elements; ++e) { + unsigned index = extractn(indexes, (base + e) * isize, isize); + dst[H4(e)] = table[H4(index)]; + } + } +} + +#define DO_SME2_LUT(ISIZE, NREG, SUFF, ESIZE) \ +void helper_sme2_luti##ISIZE##_##NREG##SUFF \ + (void *zd, void *zn, CPUARMState *env, uint32_t desc) \ +{ \ + unsigned vl = simd_oprsz(desc); \ + unsigned strided = extract32(desc, SIMD_DATA_SHIFT, 1); \ + unsigned idx = extract32(desc, SIMD_DATA_SHIFT + 1, 4); \ + unsigned elements = vl / ESIZE; \ + unsigned dstride = (!strided ? 1 : NREG == 4 ? 4 : 8); \ + unsigned segments = (ESIZE * 8) / (ISIZE * NREG); \ + unsigned segment = idx & (segments - 1); \ + ARMVectorReg indexes; \ + memcpy(&indexes, zn, vl); \ + do_lut_##SUFF(zd, indexes.d, (void *)env->za_state.zt0, elements, \ + segment * NREG * elements, \ + dstride * sizeof(ARMVectorReg), ISIZE, 32, NREG); \ +} + +DO_SME2_LUT(2,1,b, 1) +DO_SME2_LUT(2,1,h, 2) +DO_SME2_LUT(2,1,s, 4) +DO_SME2_LUT(2,2,b, 1) +DO_SME2_LUT(2,2,h, 2) +DO_SME2_LUT(2,2,s, 4) +DO_SME2_LUT(2,4,b, 1) +DO_SME2_LUT(2,4,h, 2) +DO_SME2_LUT(2,4,s, 4) + +DO_SME2_LUT(4,1,b, 1) +DO_SME2_LUT(4,1,h, 2) +DO_SME2_LUT(4,1,s, 4) +DO_SME2_LUT(4,2,b, 1) +DO_SME2_LUT(4,2,h, 2) +DO_SME2_LUT(4,2,s, 4) +DO_SME2_LUT(4,4,b, 1) +DO_SME2_LUT(4,4,h, 2) +DO_SME2_LUT(4,4,s, 4) + +#undef DO_SME2_LUT -- 2.43.0 The pattern we currently have as FMOPA_h is the "widening" insn that takes fp16 inputs and produces single-precision outputs. This is unlike FMOPA_s and FMOPA_d, which are non-widening produce outputs the same size as their inputs. SME2 introduces a non-widening fp16 FMOPA operation; rename FMOPA_h to FMOPA_w_h (for 'widening'), so we can use FMOPA_h for the non-widening version, giving it a name in line with the other non-widening ops FMOPA_s and FMOPA_d. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-102-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper-sme.h | 2 +- target/arm/tcg/sme.decode | 2 +- target/arm/tcg/sme_helper.c | 4 ++-- target/arm/tcg/translate-sme.c | 4 ++-- 4 files changed, 6 insertions(+), 6 deletions(-) diff --git a/target/arm/tcg/helper-sme.h b/target/arm/tcg/helper-sme.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-sme.h +++ b/target/arm/tcg/helper-sme.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_5(sme_addva_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_5(sme_addha_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_5(sme_addva_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) -DEF_HELPER_FLAGS_7(sme_fmopa_h, TCG_CALL_NO_RWG, +DEF_HELPER_FLAGS_7(sme_fmopa_w_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, ptr, env, i32) DEF_HELPER_FLAGS_7(sme_fmopa_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, ptr, fpst, i32) diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ FMOPA_s 10000000 100 ..... ... ... ..... . 00 .. @op_32 FMOPA_d 10000000 110 ..... ... ... ..... . 0 ... @op_64 BFMOPA 10000001 100 ..... ... ... ..... . 00 .. @op_32 -FMOPA_h 10000001 101 ..... ... ... ..... . 00 .. @op_32 +FMOPA_w_h 10000001 101 ..... ... ... ..... . 00 .. @op_32 SMOPA_s 1010000 0 10 0 ..... ... ... ..... . 00 .. @op_32 SUMOPA_s 1010000 0 10 1 ..... ... ... ..... . 00 .. @op_32 diff --git a/target/arm/tcg/sme_helper.c b/target/arm/tcg/sme_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme_helper.c +++ b/target/arm/tcg/sme_helper.c @@ -XXX,XX +XXX,XX @@ static float32 f16_dotadd(float32 sum, uint32_t e1, uint32_t e2, return float32_add(sum, t32, s_std); } -void HELPER(sme_fmopa_h)(void *vza, void *vzn, void *vzm, void *vpn, - void *vpm, CPUARMState *env, uint32_t desc) +void HELPER(sme_fmopa_w_h)(void *vza, void *vzn, void *vzm, void *vpn, + void *vpm, CPUARMState *env, uint32_t desc) { intptr_t row, col, oprsz = simd_maxsz(desc); uint32_t neg = simd_data(desc) * 0x80008000u; diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ static bool do_outprod_env(DisasContext *s, arg_op *a, MemOp esz, return true; } -TRANS_FEAT(FMOPA_h, aa64_sme, do_outprod_env, a, - MO_32, gen_helper_sme_fmopa_h) +TRANS_FEAT(FMOPA_w_h, aa64_sme, do_outprod_env, a, + MO_32, gen_helper_sme_fmopa_w_h) TRANS_FEAT(FMOPA_s, aa64_sme, do_outprod_fpst, a, MO_32, FPST_ZA, gen_helper_sme_fmopa_s) TRANS_FEAT(FMOPA_d, aa64_sme_f64f64, do_outprod_fpst, a, -- 2.43.0 Our current BFMOPA opcode pattern is the widening version of the insn. Rename it to BFMOPA_w, to make way for the non-widening version added in SME2. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-103-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper-sme.h | 2 +- target/arm/tcg/sme.decode | 2 +- target/arm/tcg/sme_helper.c | 4 ++-- target/arm/tcg/translate-sme.c | 2 +- 4 files changed, 5 insertions(+), 5 deletions(-) diff --git a/target/arm/tcg/helper-sme.h b/target/arm/tcg/helper-sme.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-sme.h +++ b/target/arm/tcg/helper-sme.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_7(sme_fmopa_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_7(sme_fmopa_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, ptr, fpst, i32) -DEF_HELPER_FLAGS_7(sme_bfmopa, TCG_CALL_NO_RWG, +DEF_HELPER_FLAGS_7(sme_bfmopa_w, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, ptr, env, i32) DEF_HELPER_FLAGS_6(sme_smopa_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, ptr, i32) diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ ADDVA_d 11000000 11 01000 1 ... ... ..... 00 ... @adda_64 FMOPA_s 10000000 100 ..... ... ... ..... . 00 .. @op_32 FMOPA_d 10000000 110 ..... ... ... ..... . 0 ... @op_64 -BFMOPA 10000001 100 ..... ... ... ..... . 00 .. @op_32 +BFMOPA_w 10000001 100 ..... ... ... ..... . 00 .. @op_32 FMOPA_w_h 10000001 101 ..... ... ... ..... . 00 .. @op_32 SMOPA_s 1010000 0 10 0 ..... ... ... ..... . 00 .. @op_32 diff --git a/target/arm/tcg/sme_helper.c b/target/arm/tcg/sme_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme_helper.c +++ b/target/arm/tcg/sme_helper.c @@ -XXX,XX +XXX,XX @@ void HELPER(sme2_fvdot_idx_h)(void *vd, void *vn, void *vm, void *va, } } -void HELPER(sme_bfmopa)(void *vza, void *vzn, void *vzm, - void *vpn, void *vpm, CPUARMState *env, uint32_t desc) +void HELPER(sme_bfmopa_w)(void *vza, void *vzn, void *vzm, + void *vpn, void *vpm, CPUARMState *env, uint32_t desc) { intptr_t row, col, oprsz = simd_maxsz(desc); uint32_t neg = simd_data(desc) * 0x80008000u; diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(FMOPA_s, aa64_sme, do_outprod_fpst, a, TRANS_FEAT(FMOPA_d, aa64_sme_f64f64, do_outprod_fpst, a, MO_64, FPST_ZA, gen_helper_sme_fmopa_d) -TRANS_FEAT(BFMOPA, aa64_sme, do_outprod_env, a, MO_32, gen_helper_sme_bfmopa) +TRANS_FEAT(BFMOPA_w, aa64_sme, do_outprod_env, a, MO_32, gen_helper_sme_bfmopa_w) TRANS_FEAT(SMOPA_s, aa64_sme, do_outprod, a, MO_32, gen_helper_sme_smopa_s) TRANS_FEAT(UMOPA_s, aa64_sme, do_outprod, a, MO_32, gen_helper_sme_umopa_s) -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> For non-widening, we can use float_muladd_negate_product, For widening, which uses dot-product, we need to handle the negation explicitly. Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-104-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper-sme.h | 19 +++++ target/arm/tcg/vec_internal.h | 5 ++ target/arm/tcg/sme_helper.c | 141 +++++++++++++++++++++++++++------ target/arm/tcg/translate-sme.c | 27 ++++--- 4 files changed, 160 insertions(+), 32 deletions(-) diff --git a/target/arm/tcg/helper-sme.h b/target/arm/tcg/helper-sme.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-sme.h +++ b/target/arm/tcg/helper-sme.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_7(sme_fmopa_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_7(sme_bfmopa_w, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, ptr, env, i32) + +DEF_HELPER_FLAGS_7(sme_fmops_w_h, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, ptr, env, i32) +DEF_HELPER_FLAGS_7(sme_fmops_s, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_7(sme_fmops_d, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_7(sme_bfmops_w, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, ptr, env, i32) + +DEF_HELPER_FLAGS_7(sme_ah_fmops_w_h, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, ptr, env, i32) +DEF_HELPER_FLAGS_7(sme_ah_fmops_s, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_7(sme_ah_fmops_d, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, ptr, fpst, i32) +DEF_HELPER_FLAGS_7(sme_ah_bfmops_w, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, ptr, env, i32) + DEF_HELPER_FLAGS_6(sme_smopa_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_6(sme_umopa_s, TCG_CALL_NO_RWG, diff --git a/target/arm/tcg/vec_internal.h b/target/arm/tcg/vec_internal.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/vec_internal.h +++ b/target/arm/tcg/vec_internal.h @@ -XXX,XX +XXX,XX @@ bool is_ebf(CPUARMState *env, float_status *statusp, float_status *oddstatusp); /* * Negate as for FPCR.AH=1 -- do not negate NaNs. */ +static inline float16 bfloat16_ah_chs(float16 a) +{ + return bfloat16_is_any_nan(a) ? a : bfloat16_chs(a); +} + static inline float16 float16_ah_chs(float16 a) { return float16_is_any_nan(a) ? a : float16_chs(a); diff --git a/target/arm/tcg/sme_helper.c b/target/arm/tcg/sme_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme_helper.c +++ b/target/arm/tcg/sme_helper.c @@ -XXX,XX +XXX,XX @@ void HELPER(sme_addva_d)(void *vzda, void *vzn, void *vpn, } } -void HELPER(sme_fmopa_s)(void *vza, void *vzn, void *vzm, void *vpn, - void *vpm, float_status *fpst, uint32_t desc) +static void do_fmopa_s(void *vza, void *vzn, void *vzm, uint16_t *pn, + uint16_t *pm, float_status *fpst, uint32_t desc, + uint32_t negx, int negf) { intptr_t row, col, oprsz = simd_maxsz(desc); - uint32_t neg = simd_data(desc) << 31; - uint16_t *pn = vpn, *pm = vpm; for (row = 0; row < oprsz; ) { uint16_t pa = pn[H2(row >> 4)]; do { if (pa & 1) { void *vza_row = vza + tile_vslice_offset(row); - uint32_t n = *(uint32_t *)(vzn + H1_4(row)) ^ neg; + uint32_t n = *(uint32_t *)(vzn + H1_4(row)) ^ negx; for (col = 0; col < oprsz; ) { uint16_t pb = pm[H2(col >> 4)]; @@ -XXX,XX +XXX,XX @@ void HELPER(sme_fmopa_s)(void *vza, void *vzn, void *vzm, void *vpn, if (pb & 1) { uint32_t *a = vza_row + H1_4(col); uint32_t *m = vzm + H1_4(col); - *a = float32_muladd(n, *m, *a, 0, fpst); + *a = float32_muladd(n, *m, *a, negf, fpst); } col += 4; pb >>= 4; @@ -XXX,XX +XXX,XX @@ void HELPER(sme_fmopa_s)(void *vza, void *vzn, void *vzm, void *vpn, } } -void HELPER(sme_fmopa_d)(void *vza, void *vzn, void *vzm, void *vpn, +void HELPER(sme_fmopa_s)(void *vza, void *vzn, void *vzm, void *vpn, void *vpm, float_status *fpst, uint32_t desc) +{ + do_fmopa_s(vza, vzn, vzm, vpn, vpm, fpst, desc, 0, 0); +} + +void HELPER(sme_fmops_s)(void *vza, void *vzn, void *vzm, void *vpn, + void *vpm, float_status *fpst, uint32_t desc) +{ + do_fmopa_s(vza, vzn, vzm, vpn, vpm, fpst, desc, 1u << 31, 0); +} + +void HELPER(sme_ah_fmops_s)(void *vza, void *vzn, void *vzm, void *vpn, + void *vpm, float_status *fpst, uint32_t desc) +{ + do_fmopa_s(vza, vzn, vzm, vpn, vpm, fpst, desc, 0, + float_muladd_negate_product); +} + +static void do_fmopa_d(uint64_t *za, uint64_t *zn, uint64_t *zm, uint8_t *pn, + uint8_t *pm, float_status *fpst, uint32_t desc, + uint64_t negx, int negf) { intptr_t row, col, oprsz = simd_oprsz(desc) / 8; - uint64_t neg = (uint64_t)simd_data(desc) << 63; - uint64_t *za = vza, *zn = vzn, *zm = vzm; - uint8_t *pn = vpn, *pm = vpm; for (row = 0; row < oprsz; ++row) { if (pn[H1(row)] & 1) { uint64_t *za_row = &za[tile_vslice_index(row)]; - uint64_t n = zn[row] ^ neg; + uint64_t n = zn[row] ^ negx; for (col = 0; col < oprsz; ++col) { if (pm[H1(col)] & 1) { uint64_t *a = &za_row[col]; - *a = float64_muladd(n, zm[col], *a, 0, fpst); + *a = float64_muladd(n, zm[col], *a, negf, fpst); } } } } } +void HELPER(sme_fmopa_d)(void *vza, void *vzn, void *vzm, void *vpn, + void *vpm, float_status *fpst, uint32_t desc) +{ + do_fmopa_d(vza, vzn, vzm, vpn, vpm, fpst, desc, 0, 0); +} + +void HELPER(sme_fmops_d)(void *vza, void *vzn, void *vzm, void *vpn, + void *vpm, float_status *fpst, uint32_t desc) +{ + do_fmopa_d(vza, vzn, vzm, vpn, vpm, fpst, desc, 1ull << 63, 0); +} + +void HELPER(sme_ah_fmops_d)(void *vza, void *vzn, void *vzm, void *vpn, + void *vpm, float_status *fpst, uint32_t desc) +{ + do_fmopa_d(vza, vzn, vzm, vpn, vpm, fpst, desc, 0, + float_muladd_negate_product); +} + /* * Alter PAIR as needed for controlling predicates being false, * and for NEG on an enabled row element. @@ -XXX,XX +XXX,XX @@ static inline uint32_t f16mop_adj_pair(uint32_t pair, uint32_t pg, uint32_t neg) return pair; } +static inline uint32_t f16mop_ah_neg_adj_pair(uint32_t pair, uint32_t pg) +{ + uint32_t l = pg & 1 ? float16_ah_chs(pair) : 0; + uint32_t h = pg & 4 ? float16_ah_chs(pair >> 16) : 0; + return l | (h << 16); +} + +static inline uint32_t bf16mop_ah_neg_adj_pair(uint32_t pair, uint32_t pg) +{ + uint32_t l = pg & 1 ? bfloat16_ah_chs(pair) : 0; + uint32_t h = pg & 4 ? bfloat16_ah_chs(pair >> 16) : 0; + return l | (h << 16); +} + static float32 f16_dotadd(float32 sum, uint32_t e1, uint32_t e2, float_status *s_f16, float_status *s_std, float_status *s_odd) @@ -XXX,XX +XXX,XX @@ static float32 f16_dotadd(float32 sum, uint32_t e1, uint32_t e2, return float32_add(sum, t32, s_std); } -void HELPER(sme_fmopa_w_h)(void *vza, void *vzn, void *vzm, void *vpn, - void *vpm, CPUARMState *env, uint32_t desc) +static void do_fmopa_w_h(void *vza, void *vzn, void *vzm, uint16_t *pn, + uint16_t *pm, CPUARMState *env, uint32_t desc, + uint32_t negx, bool ah_neg) { intptr_t row, col, oprsz = simd_maxsz(desc); - uint32_t neg = simd_data(desc) * 0x80008000u; - uint16_t *pn = vpn, *pm = vpm; float_status fpst_odd = env->vfp.fp_status[FPST_ZA]; set_float_rounding_mode(float_round_to_odd, &fpst_odd); @@ -XXX,XX +XXX,XX @@ void HELPER(sme_fmopa_w_h)(void *vza, void *vzn, void *vzm, void *vpn, void *vza_row = vza + tile_vslice_offset(row); uint32_t n = *(uint32_t *)(vzn + H1_4(row)); - n = f16mop_adj_pair(n, prow, neg); + if (ah_neg) { + n = f16mop_ah_neg_adj_pair(n, prow); + } else { + n = f16mop_adj_pair(n, prow, negx); + } for (col = 0; col < oprsz; ) { uint16_t pcol = pm[H2(col >> 4)]; @@ -XXX,XX +XXX,XX @@ void HELPER(sme_fmopa_w_h)(void *vza, void *vzn, void *vzm, void *vpn, } } +void HELPER(sme_fmopa_w_h)(void *vza, void *vzn, void *vzm, void *vpn, + void *vpm, CPUARMState *env, uint32_t desc) +{ + do_fmopa_w_h(vza, vzn, vzm, vpn, vpm, env, desc, 0, false); +} + +void HELPER(sme_fmops_w_h)(void *vza, void *vzn, void *vzm, void *vpn, + void *vpm, CPUARMState *env, uint32_t desc) +{ + do_fmopa_w_h(vza, vzn, vzm, vpn, vpm, env, desc, 0x80008000u, false); +} + +void HELPER(sme_ah_fmops_w_h)(void *vza, void *vzn, void *vzm, void *vpn, + void *vpm, CPUARMState *env, uint32_t desc) +{ + do_fmopa_w_h(vza, vzn, vzm, vpn, vpm, env, desc, 0, true); +} + void HELPER(sme2_fdot_h)(void *vd, void *vn, void *vm, void *va, CPUARMState *env, uint32_t desc) { @@ -XXX,XX +XXX,XX @@ void HELPER(sme2_fvdot_idx_h)(void *vd, void *vn, void *vm, void *va, } } -void HELPER(sme_bfmopa_w)(void *vza, void *vzn, void *vzm, - void *vpn, void *vpm, CPUARMState *env, uint32_t desc) +static void do_bfmopa_w(void *vza, void *vzn, void *vzm, + uint16_t *pn, uint16_t *pm, CPUARMState *env, + uint32_t desc, uint32_t negx, bool ah_neg) { intptr_t row, col, oprsz = simd_maxsz(desc); - uint32_t neg = simd_data(desc) * 0x80008000u; - uint16_t *pn = vpn, *pm = vpm; float_status fpst, fpst_odd; if (is_ebf(env, &fpst, &fpst_odd)) { @@ -XXX,XX +XXX,XX @@ void HELPER(sme_bfmopa_w)(void *vza, void *vzn, void *vzm, void *vza_row = vza + tile_vslice_offset(row); uint32_t n = *(uint32_t *)(vzn + H1_4(row)); - n = f16mop_adj_pair(n, prow, neg); + if (ah_neg) { + n = bf16mop_ah_neg_adj_pair(n, prow); + } else { + n = f16mop_adj_pair(n, prow, negx); + } for (col = 0; col < oprsz; ) { uint16_t pcol = pm[H2(col >> 4)]; @@ -XXX,XX +XXX,XX @@ void HELPER(sme_bfmopa_w)(void *vza, void *vzn, void *vzm, void *vza_row = vza + tile_vslice_offset(row); uint32_t n = *(uint32_t *)(vzn + H1_4(row)); - n = f16mop_adj_pair(n, prow, neg); + if (ah_neg) { + n = bf16mop_ah_neg_adj_pair(n, prow); + } else { + n = f16mop_adj_pair(n, prow, negx); + } for (col = 0; col < oprsz; ) { uint16_t pcol = pm[H2(col >> 4)]; @@ -XXX,XX +XXX,XX @@ void HELPER(sme_bfmopa_w)(void *vza, void *vzn, void *vzm, } } +void HELPER(sme_bfmopa_w)(void *vza, void *vzn, void *vzm, void *vpn, + void *vpm, CPUARMState *env, uint32_t desc) +{ + do_bfmopa_w(vza, vzn, vzm, vpn, vpm, env, desc, 0, false); +} + +void HELPER(sme_bfmops_w)(void *vza, void *vzn, void *vzm, void *vpn, + void *vpm, CPUARMState *env, uint32_t desc) +{ + do_bfmopa_w(vza, vzn, vzm, vpn, vpm, env, desc, 0x80008000u, false); +} + +void HELPER(sme_ah_bfmops_w)(void *vza, void *vzn, void *vzm, void *vpn, + void *vpm, CPUARMState *env, uint32_t desc) +{ + do_bfmopa_w(vza, vzn, vzm, vpn, vpm, env, desc, 0, true); +} + typedef uint32_t IMOPFn32(uint32_t, uint32_t, uint32_t, uint8_t, bool); static inline void do_imopa_s(uint32_t *za, uint32_t *zn, uint32_t *zm, uint8_t *pn, uint8_t *pm, diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ static bool do_outprod_fpst(DisasContext *s, arg_op *a, MemOp esz, gen_helper_gvec_5_ptr *fn) { int svl = streaming_vec_reg_size(s); - uint32_t desc = simd_desc(svl, svl, a->sub); + uint32_t desc = simd_desc(svl, svl, 0); TCGv_ptr za, zn, zm, pn, pm, fpst; if (!sme_smza_enabled_check(s)) { @@ -XXX,XX +XXX,XX @@ static bool do_outprod_env(DisasContext *s, arg_op *a, MemOp esz, gen_helper_gvec_5_ptr *fn) { int svl = streaming_vec_reg_size(s); - uint32_t desc = simd_desc(svl, svl, a->sub); + uint32_t desc = simd_desc(svl, svl, 0); TCGv_ptr za, zn, zm, pn, pm; if (!sme_smza_enabled_check(s)) { @@ -XXX,XX +XXX,XX @@ static bool do_outprod_env(DisasContext *s, arg_op *a, MemOp esz, return true; } -TRANS_FEAT(FMOPA_w_h, aa64_sme, do_outprod_env, a, - MO_32, gen_helper_sme_fmopa_w_h) -TRANS_FEAT(FMOPA_s, aa64_sme, do_outprod_fpst, a, - MO_32, FPST_ZA, gen_helper_sme_fmopa_s) -TRANS_FEAT(FMOPA_d, aa64_sme_f64f64, do_outprod_fpst, a, - MO_64, FPST_ZA, gen_helper_sme_fmopa_d) +TRANS_FEAT(FMOPA_w_h, aa64_sme, do_outprod_env, a, MO_32, + !a->sub ? gen_helper_sme_fmopa_w_h + : !s->fpcr_ah ? gen_helper_sme_fmops_w_h + : gen_helper_sme_ah_fmops_w_h) +TRANS_FEAT(FMOPA_s, aa64_sme, do_outprod_fpst, a, MO_32, FPST_ZA, + !a->sub ? gen_helper_sme_fmopa_s + : !s->fpcr_ah ? gen_helper_sme_fmops_s + : gen_helper_sme_ah_fmops_s) +TRANS_FEAT(FMOPA_d, aa64_sme_f64f64, do_outprod_fpst, a, MO_64, FPST_ZA, + !a->sub ? gen_helper_sme_fmopa_d + : !s->fpcr_ah ? gen_helper_sme_fmops_d + : gen_helper_sme_ah_fmops_d) -TRANS_FEAT(BFMOPA_w, aa64_sme, do_outprod_env, a, MO_32, gen_helper_sme_bfmopa_w) +TRANS_FEAT(BFMOPA_w, aa64_sme, do_outprod_env, a, MO_32, + !a->sub ? gen_helper_sme_bfmopa_w + : !s->fpcr_ah ? gen_helper_sme_bfmops_w + : gen_helper_sme_ah_bfmops_w) TRANS_FEAT(SMOPA_s, aa64_sme, do_outprod, a, MO_32, gen_helper_sme_smopa_s) TRANS_FEAT(UMOPA_s, aa64_sme, do_outprod, a, MO_32, gen_helper_sme_umopa_s) -- 2.43.0 Reviewed-by: Alex Bennée <alex.bennee@linaro.org> Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-105-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper-sme.h | 6 ++++ target/arm/tcg/sme.decode | 2 ++ target/arm/tcg/sme_helper.c | 51 ++++++++++++++++++++++++++++++++++ target/arm/tcg/translate-sme.c | 4 +++ 4 files changed, 63 insertions(+) diff --git a/target/arm/tcg/helper-sme.h b/target/arm/tcg/helper-sme.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-sme.h +++ b/target/arm/tcg/helper-sme.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_5(sme_addva_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_7(sme_fmopa_w_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, ptr, env, i32) +DEF_HELPER_FLAGS_7(sme_fmopa_h, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_7(sme_fmopa_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_7(sme_fmopa_d, TCG_CALL_NO_RWG, @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_7(sme_bfmopa_w, TCG_CALL_NO_RWG, DEF_HELPER_FLAGS_7(sme_fmops_w_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, ptr, env, i32) +DEF_HELPER_FLAGS_7(sme_fmops_h, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_7(sme_fmops_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_7(sme_fmops_d, TCG_CALL_NO_RWG, @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_7(sme_bfmops_w, TCG_CALL_NO_RWG, DEF_HELPER_FLAGS_7(sme_ah_fmops_w_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, ptr, env, i32) +DEF_HELPER_FLAGS_7(sme_ah_fmops_h, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_7(sme_ah_fmops_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_7(sme_ah_fmops_d, TCG_CALL_NO_RWG, diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ ADDVA_d 11000000 11 01000 1 ... ... ..... 00 ... @adda_64 ### SME Outer Product &op zad zn zm pm pn sub:bool +@op_16 ........ ... zm:5 pm:3 pn:3 zn:5 sub:1 ... zad:1 &op @op_32 ........ ... zm:5 pm:3 pn:3 zn:5 sub:1 .. zad:2 &op @op_64 ........ ... zm:5 pm:3 pn:3 zn:5 sub:1 . zad:3 &op +FMOPA_h 10000001 100 ..... ... ... ..... . 100 . @op_16 FMOPA_s 10000000 100 ..... ... ... ..... . 00 .. @op_32 FMOPA_d 10000000 110 ..... ... ... ..... . 0 ... @op_64 diff --git a/target/arm/tcg/sme_helper.c b/target/arm/tcg/sme_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme_helper.c +++ b/target/arm/tcg/sme_helper.c @@ -XXX,XX +XXX,XX @@ void HELPER(sme_addva_d)(void *vzda, void *vzn, void *vpn, } } +static void do_fmopa_h(void *vza, void *vzn, void *vzm, uint16_t *pn, + uint16_t *pm, float_status *fpst, uint32_t desc, + uint16_t negx, int negf) +{ + intptr_t row, col, oprsz = simd_maxsz(desc); + + for (row = 0; row < oprsz; ) { + uint16_t pa = pn[H2(row >> 4)]; + do { + if (pa & 1) { + void *vza_row = vza + tile_vslice_offset(row); + uint16_t n = *(uint32_t *)(vzn + H1_2(row)) ^ negx; + + for (col = 0; col < oprsz; ) { + uint16_t pb = pm[H2(col >> 4)]; + do { + if (pb & 1) { + uint16_t *a = vza_row + H1_2(col); + uint16_t *m = vzm + H1_2(col); + *a = float16_muladd(n, *m, *a, negf, fpst); + } + col += 2; + pb >>= 2; + } while (col & 15); + } + } + row += 2; + pa >>= 2; + } while (row & 15); + } +} + +void HELPER(sme_fmopa_h)(void *vza, void *vzn, void *vzm, void *vpn, + void *vpm, float_status *fpst, uint32_t desc) +{ + do_fmopa_h(vza, vzn, vzm, vpn, vpm, fpst, desc, 0, 0); +} + +void HELPER(sme_fmops_h)(void *vza, void *vzn, void *vzm, void *vpn, + void *vpm, float_status *fpst, uint32_t desc) +{ + do_fmopa_h(vza, vzn, vzm, vpn, vpm, fpst, desc, 1u << 15, 0); +} + +void HELPER(sme_ah_fmops_h)(void *vza, void *vzn, void *vzm, void *vpn, + void *vpm, float_status *fpst, uint32_t desc) +{ + do_fmopa_h(vza, vzn, vzm, vpn, vpm, fpst, desc, 0, + float_muladd_negate_product); +} + static void do_fmopa_s(void *vza, void *vzn, void *vzm, uint16_t *pn, uint16_t *pm, float_status *fpst, uint32_t desc, uint32_t negx, int negf) diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(FMOPA_w_h, aa64_sme, do_outprod_env, a, MO_32, !a->sub ? gen_helper_sme_fmopa_w_h : !s->fpcr_ah ? gen_helper_sme_fmops_w_h : gen_helper_sme_ah_fmops_w_h) +TRANS_FEAT(FMOPA_h, aa64_sme_f16f16, do_outprod_fpst, a, MO_16, FPST_ZA_F16, + !a->sub ? gen_helper_sme_fmopa_h + : !s->fpcr_ah ? gen_helper_sme_fmops_h + : gen_helper_sme_ah_fmops_h) TRANS_FEAT(FMOPA_s, aa64_sme, do_outprod_fpst, a, MO_32, FPST_ZA, !a->sub ? gen_helper_sme_fmopa_s : !s->fpcr_ah ? gen_helper_sme_fmops_s -- 2.43.0 Reviewed-by: Alex Bennée <alex.bennee@linaro.org> Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-106-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/helper-sme.h | 6 ++++ target/arm/tcg/sme.decode | 2 ++ target/arm/tcg/sme_helper.c | 51 ++++++++++++++++++++++++++++++++++ target/arm/tcg/translate-sme.c | 5 ++++ 4 files changed, 64 insertions(+) diff --git a/target/arm/tcg/helper-sme.h b/target/arm/tcg/helper-sme.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-sme.h +++ b/target/arm/tcg/helper-sme.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_7(sme_fmopa_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_7(sme_bfmopa_w, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, ptr, env, i32) +DEF_HELPER_FLAGS_7(sme_bfmopa, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_7(sme_fmops_w_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, ptr, env, i32) @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_7(sme_fmops_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_7(sme_bfmops_w, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, ptr, env, i32) +DEF_HELPER_FLAGS_7(sme_bfmops, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_7(sme_ah_fmops_w_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, ptr, env, i32) @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_7(sme_ah_fmops_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_7(sme_ah_bfmops_w, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, ptr, env, i32) +DEF_HELPER_FLAGS_7(sme_ah_bfmops, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, ptr, fpst, i32) DEF_HELPER_FLAGS_6(sme_smopa_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, ptr, i32) diff --git a/target/arm/tcg/sme.decode b/target/arm/tcg/sme.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme.decode +++ b/target/arm/tcg/sme.decode @@ -XXX,XX +XXX,XX @@ FMOPA_h 10000001 100 ..... ... ... ..... . 100 . @op_16 FMOPA_s 10000000 100 ..... ... ... ..... . 00 .. @op_32 FMOPA_d 10000000 110 ..... ... ... ..... . 0 ... @op_64 +BFMOPA 10000001 101 ..... ... ... ..... . 100 . @op_16 + BFMOPA_w 10000001 100 ..... ... ... ..... . 00 .. @op_32 FMOPA_w_h 10000001 101 ..... ... ... ..... . 00 .. @op_32 diff --git a/target/arm/tcg/sme_helper.c b/target/arm/tcg/sme_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/sme_helper.c +++ b/target/arm/tcg/sme_helper.c @@ -XXX,XX +XXX,XX @@ void HELPER(sme_ah_fmops_d)(void *vza, void *vzn, void *vzm, void *vpn, float_muladd_negate_product); } +static void do_bfmopa(void *vza, void *vzn, void *vzm, uint16_t *pn, + uint16_t *pm, float_status *fpst, uint32_t desc, + uint16_t negx, int negf) +{ + intptr_t row, col, oprsz = simd_maxsz(desc); + + for (row = 0; row < oprsz; ) { + uint16_t pa = pn[H2(row >> 4)]; + do { + if (pa & 1) { + void *vza_row = vza + tile_vslice_offset(row); + uint16_t n = *(uint32_t *)(vzn + H1_2(row)) ^ negx; + + for (col = 0; col < oprsz; ) { + uint16_t pb = pm[H2(col >> 4)]; + do { + if (pb & 1) { + uint16_t *a = vza_row + H1_2(col); + uint16_t *m = vzm + H1_2(col); + *a = bfloat16_muladd(n, *m, *a, negf, fpst); + } + col += 2; + pb >>= 2; + } while (col & 15); + } + } + row += 2; + pa >>= 2; + } while (row & 15); + } +} + +void HELPER(sme_bfmopa)(void *vza, void *vzn, void *vzm, void *vpn, + void *vpm, float_status *fpst, uint32_t desc) +{ + do_bfmopa(vza, vzn, vzm, vpn, vpm, fpst, desc, 0, 0); +} + +void HELPER(sme_bfmops)(void *vza, void *vzn, void *vzm, void *vpn, + void *vpm, float_status *fpst, uint32_t desc) +{ + do_bfmopa(vza, vzn, vzm, vpn, vpm, fpst, desc, 1u << 15, 0); +} + +void HELPER(sme_ah_bfmops)(void *vza, void *vzn, void *vzm, void *vpn, + void *vpm, float_status *fpst, uint32_t desc) +{ + do_bfmopa(vza, vzn, vzm, vpn, vpm, fpst, desc, 0, + float_muladd_negate_product); +} + /* * Alter PAIR as needed for controlling predicates being false, * and for NEG on an enabled row element. diff --git a/target/arm/tcg/translate-sme.c b/target/arm/tcg/translate-sme.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-sme.c +++ b/target/arm/tcg/translate-sme.c @@ -XXX,XX +XXX,XX @@ TRANS_FEAT(FMOPA_d, aa64_sme_f64f64, do_outprod_fpst, a, MO_64, FPST_ZA, : !s->fpcr_ah ? gen_helper_sme_fmops_d : gen_helper_sme_ah_fmops_d) +TRANS_FEAT(BFMOPA, aa64_sme_b16b16, do_outprod_fpst, a, MO_16, FPST_ZA, + !a->sub ? gen_helper_sme_bfmopa + : !s->fpcr_ah ? gen_helper_sme_bfmops + : gen_helper_sme_ah_bfmops) + TRANS_FEAT(BFMOPA_w, aa64_sme, do_outprod_env, a, MO_32, !a->sub ? gen_helper_sme_bfmopa_w : !s->fpcr_ah ? gen_helper_sme_bfmops_w -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-107-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- docs/system/arm/emulation.rst | 6 ++++++ target/arm/tcg/cpu64.c | 10 ++++++++-- 2 files changed, 14 insertions(+), 2 deletions(-) diff --git a/docs/system/arm/emulation.rst b/docs/system/arm/emulation.rst index XXXXXXX..XXXXXXX 100644 --- a/docs/system/arm/emulation.rst +++ b/docs/system/arm/emulation.rst @@ -XXX,XX +XXX,XX @@ the following architecture extensions: - FEAT_SM3 (Advanced SIMD SM3 instructions) - FEAT_SM4 (Advanced SIMD SM4 instructions) - FEAT_SME (Scalable Matrix Extension) +- FEAT_SME2 (Scalable Matrix Extension version 2) +- FEAT_SME2p1 (Scalable Matrix Extension version 2.1) +- FEAT_SME_B16B16 (Non-widening BFloat16 arithmetic for SME2) - FEAT_SME_FA64 (Full A64 instruction set in Streaming SVE mode) +- FEAT_SME_F16F16 (Non-widening half-precision FP16 arithmetic for SME2) - FEAT_SME_F64F64 (Double-precision floating-point outer product instructions) - FEAT_SME_I16I64 (16-bit to 64-bit integer widening outer product instructions) - FEAT_SVE (Scalable Vector Extension) - FEAT_SVE_AES (Scalable Vector AES instructions) +- FEAT_SVE_B16B16 (Non-widening BFloat16 arithmetic for SVE2) - FEAT_SVE_BitPerm (Scalable Vector Bit Permutes instructions) - FEAT_SVE_PMULL128 (Scalable Vector PMULL instructions) - FEAT_SVE_SHA3 (Scalable Vector SHA3 instructions) - FEAT_SVE_SM4 (Scalable Vector SM4 instructions) - FEAT_SVE2 (Scalable Vector Extension version 2) +- FEAT_SVE2p1 (Scalable Vector Extension version 2.1) - FEAT_SPECRES (Speculation restriction instructions) - FEAT_SSBS (Speculative Store Bypass Safe) - FEAT_SSBS2 (MRS and MSR instructions for SSBS version 2) diff --git a/target/arm/tcg/cpu64.c b/target/arm/tcg/cpu64.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/cpu64.c +++ b/target/arm/tcg/cpu64.c @@ -XXX,XX +XXX,XX @@ void aarch64_max_tcg_initfn(Object *obj) */ t = FIELD_DP64(t, ID_AA64PFR1, MTE, 3); /* FEAT_MTE3 */ t = FIELD_DP64(t, ID_AA64PFR1, RAS_FRAC, 0); /* FEAT_RASv1p1 + FEAT_DoubleFault */ - t = FIELD_DP64(t, ID_AA64PFR1, SME, 1); /* FEAT_SME */ + t = FIELD_DP64(t, ID_AA64PFR1, SME, 2); /* FEAT_SME2 */ t = FIELD_DP64(t, ID_AA64PFR1, CSV2_FRAC, 0); /* FEAT_CSV2_3 */ t = FIELD_DP64(t, ID_AA64PFR1, NMI, 1); /* FEAT_NMI */ SET_IDREG(isar, ID_AA64PFR1, t); @@ -XXX,XX +XXX,XX @@ void aarch64_max_tcg_initfn(Object *obj) FIELD_DP64_IDREG(isar, ID_AA64MMFR3, SPEC_FPACC, 1); /* FEAT_FPACC_SPEC */ t = GET_IDREG(isar, ID_AA64ZFR0); - t = FIELD_DP64(t, ID_AA64ZFR0, SVEVER, 1); + t = FIELD_DP64(t, ID_AA64ZFR0, SVEVER, 2); /* FEAT_SVE2p1 */ t = FIELD_DP64(t, ID_AA64ZFR0, AES, 2); /* FEAT_SVE_PMULL128 */ t = FIELD_DP64(t, ID_AA64ZFR0, BITPERM, 1); /* FEAT_SVE_BitPerm */ t = FIELD_DP64(t, ID_AA64ZFR0, BFLOAT16, 2); /* FEAT_BF16, FEAT_EBF16 */ + t = FIELD_DP64(t, ID_AA64ZFR0, B16B16, 1); /* FEAT_SVE_B16B16 */ t = FIELD_DP64(t, ID_AA64ZFR0, SHA3, 1); /* FEAT_SVE_SHA3 */ t = FIELD_DP64(t, ID_AA64ZFR0, SM4, 1); /* FEAT_SVE_SM4 */ t = FIELD_DP64(t, ID_AA64ZFR0, I8MM, 1); /* FEAT_I8MM */ @@ -XXX,XX +XXX,XX @@ void aarch64_max_tcg_initfn(Object *obj) t = GET_IDREG(isar, ID_AA64SMFR0); t = FIELD_DP64(t, ID_AA64SMFR0, F32F32, 1); /* FEAT_SME */ + t = FIELD_DP64(t, ID_AA64SMFR0, BI32I32, 1); /* FEAT_SME2 */ t = FIELD_DP64(t, ID_AA64SMFR0, B16F32, 1); /* FEAT_SME */ t = FIELD_DP64(t, ID_AA64SMFR0, F16F32, 1); /* FEAT_SME */ t = FIELD_DP64(t, ID_AA64SMFR0, I8I32, 0xf); /* FEAT_SME */ + t = FIELD_DP64(t, ID_AA64SMFR0, F16F16, 1); /* FEAT_SME_F16F16 */ + t = FIELD_DP64(t, ID_AA64SMFR0, B16B16, 1); /* FEAT_SME_B16B16 */ + t = FIELD_DP64(t, ID_AA64SMFR0, I16I32, 5); /* FEAT_SME2 */ t = FIELD_DP64(t, ID_AA64SMFR0, F64F64, 1); /* FEAT_SME_F64F64 */ t = FIELD_DP64(t, ID_AA64SMFR0, I16I64, 0xf); /* FEAT_SME_I16I64 */ + t = FIELD_DP64(t, ID_AA64SMFR0, SMEVER, 2); /* FEAT_SME2p1 */ t = FIELD_DP64(t, ID_AA64SMFR0, FA64, 1); /* FEAT_SME_FA64 */ SET_IDREG(isar, ID_AA64SMFR0, t); -- 2.43.0 From: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20250704142112.1018902-108-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- linux-user/elfload.c | 8 ++++++++ 1 file changed, 8 insertions(+) diff --git a/linux-user/elfload.c b/linux-user/elfload.c index XXXXXXX..XXXXXXX 100644 --- a/linux-user/elfload.c +++ b/linux-user/elfload.c @@ -XXX,XX +XXX,XX @@ uint64_t get_elf_hwcap2(void) GET_FEATURE_ID(aa64_sme_fa64, ARM_HWCAP2_A64_SME_FA64); GET_FEATURE_ID(aa64_hbc, ARM_HWCAP2_A64_HBC); GET_FEATURE_ID(aa64_mops, ARM_HWCAP2_A64_MOPS); + GET_FEATURE_ID(aa64_sve2p1, ARM_HWCAP2_A64_SVE2P1); + GET_FEATURE_ID(aa64_sme2, (ARM_HWCAP2_A64_SME2 | + ARM_HWCAP2_A64_SME_I16I32 | + ARM_HWCAP2_A64_SME_BI32I32)); + GET_FEATURE_ID(aa64_sme2p1, ARM_HWCAP2_A64_SME2P1); + GET_FEATURE_ID(aa64_sme_b16b16, ARM_HWCAP2_A64_SME_B16B16); + GET_FEATURE_ID(aa64_sme_f16f16, ARM_HWCAP2_A64_SME_F16F16); + GET_FEATURE_ID(aa64_sve_b16b16, ARM_HWCAP2_A64_SVE_B16B16); return hwcaps; } -- 2.43.0 Hi; the bulk of this arm pullreq is Shameer's SMMUv3 acceleration series; there are a few other minor items in here also. thanks -- PMM The following changes since commit 65fb9de4888f9bf32bc4f60c880da3dff04c4fd9: Merge tag 'pr-build_fix-20260128' of https://gitlab.com/pbo-linaro/qemu into staging (2026-01-29 13:29:33 +1100) are available in the Git repository at: https://gitlab.com/pm215/qemu.git tags/pull-target-arm-20260129 for you to fetch changes up to 5252c077467a5570bf3b6b6278d206c46f51f9b8: arm: add DCZID_EL0 to idregs array (2026-01-29 13:45:39 +0000) ---------------------------------------------------------------- target-arm queue: * Support SMMUv3 acceleration * A few other minor cleanups and fixes ---------------------------------------------------------------- Alex Bennée (2): MAINTAINERS: add emulation.rst to ARM TCG CPUs docs/system: update FEAT_BBML[12] references Cornelia Huck (2): arm: add {get,set}_dczid_bs helpers arm: add DCZID_EL0 to idregs array Eric Auger (2): hw/pci-host/gpex: Allow to generate preserve boot config DSM #5 hw/arm/virt-acpi-build: Add IORT RMR regions to handle MSI nested binding Nicolin Chen (4): backends/iommufd: Introduce iommufd_backend_alloc_viommu backends/iommufd: Introduce iommufd_backend_alloc_vdev hw/arm/smmuv3-accel: Add set/unset_iommu_device callback hw/arm/smmuv3-accel: Add nested vSTE install/uninstall support Philippe Mathieu-Daudé (2): target/arm/hvf: Move hvf_sysreg_[read, write]_cp() functions around target/arm/hvf: Sync CNTV_CTL_EL0 & CNTV_CVAL_EL0 Shameer Kolothum (31): hw/arm/smmu-common: Factor out common helper functions and export hw/arm/smmu-common: Make iommu ops part of SMMUState hw/arm/smmuv3-accel: Introduce smmuv3 accel device hw/arm/smmuv3-accel: Initialize shared system address space hw/pci/pci: Move pci_init_bus_master() after adding device to bus hw/pci/pci: Add optional supports_address_space() callback hw/pci-bridge/pci_expander_bridge: Move TYPE_PXB_PCIE_DEV to header hw/arm/smmuv3-accel: Restrict accelerated SMMUv3 to vfio-pci endpoints with iommufd hw/arm/smmuv3: Implement get_viommu_cap() callback hw/arm/smmuv3: propagate smmuv3_cmdq_consume() errors to caller hw/arm/smmuv3-accel: Install SMMUv3 GBPA based hwpt hw/pci/pci: Introduce a callback to retrieve the MSI doorbell GPA directly hw/arm/smmuv3-accel: Implement get_msi_direct_gpa callback hw/arm/virt: Set msi-gpa property hw/arm/smmuv3-accel: Add support to issue invalidation cmd to host hw/arm/smmuv3: Initialize ID registers early during realize() hw/arm/smmuv3-accel: Get host SMMUv3 hw info and validate hw/arm/virt: Set PCI preserve_config for accel SMMUv3 tests/qtest/bios-tables-test: Prepare for IORT revison upgrade tests/qtest/bios-tables-test: Update IORT blobs after revision upgrade hw/arm/smmuv3: Block migration when accel is enabled hw/arm/smmuv3: Add accel property for SMMUv3 device hw/arm/smmuv3-accel: Add a property to specify RIL support hw/arm/smmuv3-accel: Add support for ATS hw/arm/smmuv3-accel: Add property to specify OAS bits backends/iommufd: Retrieve PASID width from iommufd_backend_get_device_info() backends/iommufd: Add get_pasid_info() callback hw/pci: Add helper to insert PCIe extended capability at a fixed offset hw/pci: Factor out common PASID capability initialization hw/vfio/pci: Synthesize PASID capability for vfio-pci devices hw/arm/smmuv3-accel: Make SubstreamID support configurable MAINTAINERS | 1 + backends/iommufd.c | 76 ++- backends/trace-events | 2 + docs/system/arm/emulation.rst | 3 +- hw/arm/Kconfig | 5 + hw/arm/meson.build | 3 +- hw/arm/smmu-common.c | 51 +- hw/arm/smmuv3-accel.c | 768 ++++++++++++++++++++++++ hw/arm/smmuv3-accel.h | 88 +++ hw/arm/smmuv3-internal.h | 1 + hw/arm/smmuv3.c | 227 +++++-- hw/arm/trace-events | 6 + hw/arm/virt-acpi-build.c | 131 +++- hw/arm/virt.c | 39 +- hw/pci-bridge/pci_expander_bridge.c | 1 - hw/pci-host/gpex-acpi.c | 29 +- hw/pci/pci.c | 43 +- hw/pci/pcie.c | 88 ++- hw/vfio/iommufd.c | 6 +- hw/vfio/pci.c | 75 +++ hw/vfio/pci.h | 1 + hw/vfio/trace-events | 1 + include/hw/arm/smmu-common.h | 7 + include/hw/arm/smmuv3-common.h | 27 +- include/hw/arm/smmuv3.h | 10 + include/hw/arm/virt.h | 1 + include/hw/core/iommu.h | 1 + include/hw/pci-host/gpex.h | 1 + include/hw/pci/pci.h | 36 ++ include/hw/pci/pci_bridge.h | 1 + include/hw/pci/pcie.h | 4 + include/system/host_iommu_device.h | 20 + include/system/iommufd.h | 29 +- target/arm/cpu-sysregs.h.inc | 1 + target/arm/cpu.c | 2 +- target/arm/cpu.h | 14 +- target/arm/cpu64.c | 4 +- target/arm/helper.c | 5 +- target/arm/hvf/hvf.c | 167 +++--- target/arm/kvm.c | 18 +- target/arm/tcg/cpu64.c | 22 +- target/arm/tcg/helper-a64.c | 2 +- target/arm/tcg/mte_helper.c | 4 +- target/arm/tcg/translate-a64.c | 2 +- target/arm/tcg/translate.h | 2 +- tests/data/acpi/aarch64/virt/IORT | Bin 128 -> 128 bytes tests/data/acpi/aarch64/virt/IORT.its_off | Bin 172 -> 172 bytes tests/data/acpi/aarch64/virt/IORT.smmuv3-dev | Bin 364 -> 364 bytes tests/data/acpi/aarch64/virt/IORT.smmuv3-legacy | Bin 276 -> 276 bytes 49 files changed, 1829 insertions(+), 196 deletions(-) create mode 100644 hw/arm/smmuv3-accel.c create mode 100644 hw/arm/smmuv3-accel.h From: Nicolin Chen <nicolinc@nvidia.com> Add a helper to allocate a viommu object. Also introduce a struct IOMMUFDViommu that can be used later by vendor IOMMU implementations. Signed-off-by: Nicolin Chen <nicolinc@nvidia.com> Reviewed-by: Eric Auger <eric.auger@redhat.com> Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com> Signed-off-by: Shameer Kolothum <shameerali.kolothum.thodi@huawei.com> Tested-by: Zhangfei Gao <zhangfei.gao@linaro.org> Tested-by: Eric Auger <eric.auger@redhat.com> Signed-off-by: Shameer Kolothum <skolothumtho@nvidia.com> Reviewed-by: Yi Liu <yi.l.liu@intel.com> Message-id: 20260126104342.253965-2-skolothumtho@nvidia.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- backends/iommufd.c | 26 ++++++++++++++++++++++++++ backends/trace-events | 1 + include/system/iommufd.h | 14 ++++++++++++++ 3 files changed, 41 insertions(+) diff --git a/backends/iommufd.c b/backends/iommufd.c index XXXXXXX..XXXXXXX 100644 --- a/backends/iommufd.c +++ b/backends/iommufd.c @@ -XXX,XX +XXX,XX @@ bool iommufd_backend_invalidate_cache(IOMMUFDBackend *be, uint32_t id, return !ret; } +bool iommufd_backend_alloc_viommu(IOMMUFDBackend *be, uint32_t dev_id, + uint32_t viommu_type, uint32_t hwpt_id, + uint32_t *out_viommu_id, Error **errp) +{ + int ret; + struct iommu_viommu_alloc alloc_viommu = { + .size = sizeof(alloc_viommu), + .type = viommu_type, + .dev_id = dev_id, + .hwpt_id = hwpt_id, + }; + + ret = ioctl(be->fd, IOMMU_VIOMMU_ALLOC, &alloc_viommu); + + trace_iommufd_backend_alloc_viommu(be->fd, dev_id, viommu_type, hwpt_id, + alloc_viommu.out_viommu_id, ret); + if (ret) { + error_setg_errno(errp, errno, "IOMMU_VIOMMU_ALLOC failed"); + return false; + } + + g_assert(out_viommu_id); + *out_viommu_id = alloc_viommu.out_viommu_id; + return true; +} + bool host_iommu_device_iommufd_attach_hwpt(HostIOMMUDeviceIOMMUFD *idev, uint32_t hwpt_id, Error **errp) { diff --git a/backends/trace-events b/backends/trace-events index XXXXXXX..XXXXXXX 100644 --- a/backends/trace-events +++ b/backends/trace-events @@ -XXX,XX +XXX,XX @@ iommufd_backend_free_id(int iommufd, uint32_t id, int ret) " iommufd=%d id=%d (% iommufd_backend_set_dirty(int iommufd, uint32_t hwpt_id, bool start, int ret) " iommufd=%d hwpt=%u enable=%d (%d)" iommufd_backend_get_dirty_bitmap(int iommufd, uint32_t hwpt_id, uint64_t iova, uint64_t size, uint64_t flags, uint64_t page_size, int ret) " iommufd=%d hwpt=%u iova=0x%"PRIx64" size=0x%"PRIx64" flags=0x%"PRIx64" page_size=0x%"PRIx64" (%d)" iommufd_backend_invalidate_cache(int iommufd, uint32_t id, uint32_t data_type, uint32_t entry_len, uint32_t entry_num, uint32_t done_num, uint64_t data_ptr, int ret) " iommufd=%d id=%u data_type=%u entry_len=%u entry_num=%u done_num=%u data_ptr=0x%"PRIx64" (%d)" +iommufd_backend_alloc_viommu(int iommufd, uint32_t dev_id, uint32_t type, uint32_t hwpt_id, uint32_t viommu_id, int ret) " iommufd=%d type=%u dev_id=%u hwpt_id=%u viommu_id=%u (%d)" diff --git a/include/system/iommufd.h b/include/system/iommufd.h index XXXXXXX..XXXXXXX 100644 --- a/include/system/iommufd.h +++ b/include/system/iommufd.h @@ -XXX,XX +XXX,XX @@ struct IOMMUFDBackend { /*< public >*/ }; +/* + * Virtual IOMMU object that represents physical IOMMU's virtualization + * support + */ +typedef struct IOMMUFDViommu { + IOMMUFDBackend *iommufd; + uint32_t s2_hwpt_id; /* ID of stage 2 HWPT */ + uint32_t viommu_id; /* virtual IOMMU ID of allocated object */ +} IOMMUFDViommu; + bool iommufd_backend_connect(IOMMUFDBackend *be, Error **errp); void iommufd_backend_disconnect(IOMMUFDBackend *be); @@ -XXX,XX +XXX,XX @@ bool iommufd_backend_alloc_hwpt(IOMMUFDBackend *be, uint32_t dev_id, uint32_t data_type, uint32_t data_len, void *data_ptr, uint32_t *out_hwpt, Error **errp); +bool iommufd_backend_alloc_viommu(IOMMUFDBackend *be, uint32_t dev_id, + uint32_t viommu_type, uint32_t hwpt_id, + uint32_t *out_hwpt, Error **errp); + bool iommufd_backend_set_dirty_tracking(IOMMUFDBackend *be, uint32_t hwpt_id, bool start, Error **errp); bool iommufd_backend_get_dirty_bitmap(IOMMUFDBackend *be, uint32_t hwpt_id, -- 2.43.0 From: Nicolin Chen <nicolinc@nvidia.com> Add a helper to allocate an iommufd device's virtual device (in the user space) per a viommu instance. While at it, introduce a struct IOMMUFDVdev for later use by vendor IOMMU implementations. Signed-off-by: Nicolin Chen <nicolinc@nvidia.com> Reviewed-by: Eric Auger <eric.auger@redhat.com> Signed-off-by: Shameer Kolothum <shameerali.kolothum.thodi@huawei.com> Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com> Tested-by: Zhangfei Gao <zhangfei.gao@linaro.org> Tested-by: Eric Auger <eric.auger@redhat.com> Signed-off-by: Shameer Kolothum <skolothumtho@nvidia.com> Reviewed-by: Yi Liu <yi.l.liu@intel.com> Message-id: 20260126104342.253965-3-skolothumtho@nvidia.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- backends/iommufd.c | 27 +++++++++++++++++++++++++++ backends/trace-events | 1 + include/system/iommufd.h | 12 ++++++++++++ 3 files changed, 40 insertions(+) diff --git a/backends/iommufd.c b/backends/iommufd.c index XXXXXXX..XXXXXXX 100644 --- a/backends/iommufd.c +++ b/backends/iommufd.c @@ -XXX,XX +XXX,XX @@ bool iommufd_backend_alloc_viommu(IOMMUFDBackend *be, uint32_t dev_id, return true; } +bool iommufd_backend_alloc_vdev(IOMMUFDBackend *be, uint32_t dev_id, + uint32_t viommu_id, uint64_t virt_id, + uint32_t *out_vdev_id, Error **errp) +{ + int ret; + struct iommu_vdevice_alloc alloc_vdev = { + .size = sizeof(alloc_vdev), + .viommu_id = viommu_id, + .dev_id = dev_id, + .virt_id = virt_id, + }; + + ret = ioctl(be->fd, IOMMU_VDEVICE_ALLOC, &alloc_vdev); + + trace_iommufd_backend_alloc_vdev(be->fd, dev_id, viommu_id, virt_id, + alloc_vdev.out_vdevice_id, ret); + + if (ret) { + error_setg_errno(errp, errno, "IOMMU_VDEVICE_ALLOC failed"); + return false; + } + + g_assert(out_vdev_id); + *out_vdev_id = alloc_vdev.out_vdevice_id; + return true; +} + bool host_iommu_device_iommufd_attach_hwpt(HostIOMMUDeviceIOMMUFD *idev, uint32_t hwpt_id, Error **errp) { diff --git a/backends/trace-events b/backends/trace-events index XXXXXXX..XXXXXXX 100644 --- a/backends/trace-events +++ b/backends/trace-events @@ -XXX,XX +XXX,XX @@ iommufd_backend_set_dirty(int iommufd, uint32_t hwpt_id, bool start, int ret) " iommufd_backend_get_dirty_bitmap(int iommufd, uint32_t hwpt_id, uint64_t iova, uint64_t size, uint64_t flags, uint64_t page_size, int ret) " iommufd=%d hwpt=%u iova=0x%"PRIx64" size=0x%"PRIx64" flags=0x%"PRIx64" page_size=0x%"PRIx64" (%d)" iommufd_backend_invalidate_cache(int iommufd, uint32_t id, uint32_t data_type, uint32_t entry_len, uint32_t entry_num, uint32_t done_num, uint64_t data_ptr, int ret) " iommufd=%d id=%u data_type=%u entry_len=%u entry_num=%u done_num=%u data_ptr=0x%"PRIx64" (%d)" iommufd_backend_alloc_viommu(int iommufd, uint32_t dev_id, uint32_t type, uint32_t hwpt_id, uint32_t viommu_id, int ret) " iommufd=%d type=%u dev_id=%u hwpt_id=%u viommu_id=%u (%d)" +iommufd_backend_alloc_vdev(int iommufd, uint32_t dev_id, uint32_t viommu_id, uint64_t virt_id, uint32_t vdev_id, int ret) " iommufd=%d dev_id=%u viommu_id=%u virt_id=0x%"PRIx64" vdev_id=%u (%d)" diff --git a/include/system/iommufd.h b/include/system/iommufd.h index XXXXXXX..XXXXXXX 100644 --- a/include/system/iommufd.h +++ b/include/system/iommufd.h @@ -XXX,XX +XXX,XX @@ typedef struct IOMMUFDViommu { uint32_t viommu_id; /* virtual IOMMU ID of allocated object */ } IOMMUFDViommu; +/* + * Virtual device object for a physical device bind to a vIOMMU. + */ +typedef struct IOMMUFDVdev { + uint32_t vdevice_id; /* object handle for vDevice */ + uint32_t virt_id; /* virtual device ID */ +} IOMMUFDVdev; + bool iommufd_backend_connect(IOMMUFDBackend *be, Error **errp); void iommufd_backend_disconnect(IOMMUFDBackend *be); @@ -XXX,XX +XXX,XX @@ bool iommufd_backend_alloc_viommu(IOMMUFDBackend *be, uint32_t dev_id, uint32_t viommu_type, uint32_t hwpt_id, uint32_t *out_hwpt, Error **errp); +bool iommufd_backend_alloc_vdev(IOMMUFDBackend *be, uint32_t dev_id, + uint32_t viommu_id, uint64_t virt_id, + uint32_t *out_vdev_id, Error **errp); + bool iommufd_backend_set_dirty_tracking(IOMMUFDBackend *be, uint32_t hwpt_id, bool start, Error **errp); bool iommufd_backend_get_dirty_bitmap(IOMMUFDBackend *be, uint32_t hwpt_id, -- 2.43.0 From: Shameer Kolothum <skolothumtho@nvidia.com> Factor out common helper functions and export. Subsequent patches for smmuv3 accel support will make use of this. Signed-off-by: Nicolin Chen <nicolinc@nvidia.com> Reviewed-by: Eric Auger <eric.auger@redhat.com> Signed-off-by: Shameer Kolothum <shameerali.kolothum.thodi@huawei.com> Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com> Tested-by: Zhangfei Gao <zhangfei.gao@linaro.org> Tested-by: Eric Auger <eric.auger@redhat.com> Signed-off-by: Shameer Kolothum <skolothumtho@nvidia.com> Message-id: 20260126104342.253965-4-skolothumtho@nvidia.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/arm/smmu-common.c | 44 +++++++++++++++++++++--------------- include/hw/arm/smmu-common.h | 6 +++++ 2 files changed, 32 insertions(+), 18 deletions(-) diff --git a/hw/arm/smmu-common.c b/hw/arm/smmu-common.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/smmu-common.c +++ b/hw/arm/smmu-common.c @@ -XXX,XX +XXX,XX @@ SMMUPciBus *smmu_find_smmu_pcibus(SMMUState *s, uint8_t bus_num) return NULL; } -static AddressSpace *smmu_find_add_as(PCIBus *bus, void *opaque, int devfn) +void smmu_init_sdev(SMMUState *s, SMMUDevice *sdev, PCIBus *bus, int devfn) { - SMMUState *s = opaque; - SMMUPciBus *sbus = g_hash_table_lookup(s->smmu_pcibus_by_busptr, bus); - SMMUDevice *sdev; static unsigned int index; + g_autofree char *name = g_strdup_printf("%s-%d-%d", s->mrtypename, devfn, + index++); + sdev->smmu = s; + sdev->bus = bus; + sdev->devfn = devfn; + + memory_region_init_iommu(&sdev->iommu, sizeof(sdev->iommu), + s->mrtypename, OBJECT(s), name, UINT64_MAX); + address_space_init(&sdev->as, MEMORY_REGION(&sdev->iommu), name); + trace_smmu_add_mr(name); +} + +SMMUPciBus *smmu_get_sbus(SMMUState *s, PCIBus *bus) +{ + SMMUPciBus *sbus = g_hash_table_lookup(s->smmu_pcibus_by_busptr, bus); if (!sbus) { sbus = g_malloc0(sizeof(SMMUPciBus) + @@ -XXX,XX +XXX,XX @@ static AddressSpace *smmu_find_add_as(PCIBus *bus, void *opaque, int devfn) g_hash_table_insert(s->smmu_pcibus_by_busptr, bus, sbus); } + return sbus; +} + +static AddressSpace *smmu_find_add_as(PCIBus *bus, void *opaque, int devfn) +{ + SMMUState *s = opaque; + SMMUPciBus *sbus = smmu_get_sbus(s, bus); + SMMUDevice *sdev; + sdev = sbus->pbdev[devfn]; if (!sdev) { - char *name = g_strdup_printf("%s-%d-%d", s->mrtypename, devfn, index++); - sdev = sbus->pbdev[devfn] = g_new0(SMMUDevice, 1); - - sdev->smmu = s; - sdev->bus = bus; - sdev->devfn = devfn; - - memory_region_init_iommu(&sdev->iommu, sizeof(sdev->iommu), - s->mrtypename, - OBJECT(s), name, UINT64_MAX); - address_space_init(&sdev->as, - MEMORY_REGION(&sdev->iommu), name); - trace_smmu_add_mr(name); - g_free(name); + smmu_init_sdev(s, sdev, bus, devfn); } return &sdev->as; diff --git a/include/hw/arm/smmu-common.h b/include/hw/arm/smmu-common.h index XXXXXXX..XXXXXXX 100644 --- a/include/hw/arm/smmu-common.h +++ b/include/hw/arm/smmu-common.h @@ -XXX,XX +XXX,XX @@ OBJECT_DECLARE_TYPE(SMMUState, SMMUBaseClass, ARM_SMMU) /* Return the SMMUPciBus handle associated to a PCI bus number */ SMMUPciBus *smmu_find_smmu_pcibus(SMMUState *s, uint8_t bus_num); +/* Return the SMMUPciBus handle associated to a PCI bus */ +SMMUPciBus *smmu_get_sbus(SMMUState *s, PCIBus *bus); + +/* Initialize SMMUDevice handle associated to a SMMUPciBus */ +void smmu_init_sdev(SMMUState *s, SMMUDevice *sdev, PCIBus *bus, int devfn); + /* Return the stream ID of an SMMU device */ static inline uint16_t smmu_get_sid(SMMUDevice *sdev) { -- 2.43.0 From: Shameer Kolothum <skolothumtho@nvidia.com> Make iommu ops part of SMMUState and set to the current default smmu_ops. No functional change intended. This will allow SMMUv3 accel implementation to set a different iommu ops later. Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com> Reviewed-by: Nicolin Chen <nicolinc@nvidia.com> Reviewed-by: Eric Auger <eric.auger@redhat.com> Tested-by: Zhangfei Gao <zhangfei.gao@linaro.org> Tested-by: Eric Auger <eric.auger@redhat.com> Signed-off-by: Shameer Kolothum <skolothumtho@nvidia.com> Message-id: 20260126104342.253965-5-skolothumtho@nvidia.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/arm/smmu-common.c | 7 +++++-- include/hw/arm/smmu-common.h | 1 + 2 files changed, 6 insertions(+), 2 deletions(-) diff --git a/hw/arm/smmu-common.c b/hw/arm/smmu-common.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/smmu-common.c +++ b/hw/arm/smmu-common.c @@ -XXX,XX +XXX,XX @@ static void smmu_base_realize(DeviceState *dev, Error **errp) "smmu-secure-memory-view"); } + if (!s->iommu_ops) { + s->iommu_ops = &smmu_ops; + } /* * We only allow default PCIe Root Complex(pcie.0) or pxb-pcie based extra * root complexes to be associated with SMMU. @@ -XXX,XX +XXX,XX @@ static void smmu_base_realize(DeviceState *dev, Error **errp) } if (s->smmu_per_bus) { - pci_setup_iommu_per_bus(pci_bus, &smmu_ops, s); + pci_setup_iommu_per_bus(pci_bus, s->iommu_ops, s); } else { - pci_setup_iommu(pci_bus, &smmu_ops, s); + pci_setup_iommu(pci_bus, s->iommu_ops, s); } return; } diff --git a/include/hw/arm/smmu-common.h b/include/hw/arm/smmu-common.h index XXXXXXX..XXXXXXX 100644 --- a/include/hw/arm/smmu-common.h +++ b/include/hw/arm/smmu-common.h @@ -XXX,XX +XXX,XX @@ struct SMMUState { AddressSpace memory_as; MemoryRegion *secure_memory; AddressSpace secure_memory_as; + const PCIIOMMUOps *iommu_ops; }; struct SMMUBaseClass { -- 2.43.0 From: Shameer Kolothum <skolothumtho@nvidia.com> Set up dedicated PCIIOMMUOps for the accel SMMUv3, since it will need different callback handling in upcoming patches. This also adds a CONFIG_ARM_SMMUV3_ACCEL build option so the feature can be disabled at compile time. Because we now include CONFIG_DEVICES in the header to check for ARM_SMMUV3_ACCEL, the meson file entry for smmuv3.c needs to be changed to arm_ss.add. The “accel” property isn’t user visible yet and it will be introduced in a later patch once all the supporting pieces are ready. Signed-off-by: Shameer Kolothum <shameerali.kolothum.thodi@huawei.com> Reviewed-by: Nicolin Chen <nicolinc@nvidia.com> Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com> Reviewed-by: Eric Auger <eric.auger@redhat.com> Tested-by: Zhangfei Gao <zhangfei.gao@linaro.org> Tested-by: Eric Auger <eric.auger@redhat.com> Signed-off-by: Shameer Kolothum <skolothumtho@nvidia.com> Message-id: 20260126104342.253965-6-skolothumtho@nvidia.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/arm/Kconfig | 5 ++++ hw/arm/meson.build | 3 ++- hw/arm/smmuv3-accel.c | 59 +++++++++++++++++++++++++++++++++++++++++ hw/arm/smmuv3-accel.h | 27 +++++++++++++++++++ hw/arm/smmuv3.c | 5 ++++ include/hw/arm/smmuv3.h | 3 +++ 6 files changed, 101 insertions(+), 1 deletion(-) create mode 100644 hw/arm/smmuv3-accel.c create mode 100644 hw/arm/smmuv3-accel.h diff --git a/hw/arm/Kconfig b/hw/arm/Kconfig index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/Kconfig +++ b/hw/arm/Kconfig @@ -XXX,XX +XXX,XX @@ config FSL_IMX8MP_EVK depends on TCG select FSL_IMX8MP +config ARM_SMMUV3_ACCEL + bool + depends on ARM_SMMUV3 + config ARM_SMMUV3 bool + select ARM_SMMUV3_ACCEL if IOMMUFD config FSL_IMX6UL bool diff --git a/hw/arm/meson.build b/hw/arm/meson.build index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/meson.build +++ b/hw/arm/meson.build @@ -XXX,XX +XXX,XX @@ arm_common_ss.add(when: 'CONFIG_ARMSSE', if_true: files('armsse.c')) arm_common_ss.add(when: 'CONFIG_FSL_IMX7', if_true: files('fsl-imx7.c', 'mcimx7d-sabre.c')) arm_common_ss.add(when: 'CONFIG_FSL_IMX8MP', if_true: files('fsl-imx8mp.c')) arm_common_ss.add(when: 'CONFIG_FSL_IMX8MP_EVK', if_true: files('imx8mp-evk.c')) -arm_common_ss.add(when: 'CONFIG_ARM_SMMUV3', if_true: files('smmuv3.c')) +arm_ss.add(when: 'CONFIG_ARM_SMMUV3', if_true: files('smmuv3.c')) +arm_ss.add(when: 'CONFIG_ARM_SMMUV3_ACCEL', if_true: files('smmuv3-accel.c')) arm_common_ss.add(when: 'CONFIG_FSL_IMX6UL', if_true: files('fsl-imx6ul.c', 'mcimx6ul-evk.c')) arm_common_ss.add(when: 'CONFIG_NRF51_SOC', if_true: files('nrf51_soc.c')) arm_common_ss.add(when: 'CONFIG_XEN', if_true: files( diff --git a/hw/arm/smmuv3-accel.c b/hw/arm/smmuv3-accel.c new file mode 100644 index XXXXXXX..XXXXXXX --- /dev/null +++ b/hw/arm/smmuv3-accel.c @@ -XXX,XX +XXX,XX @@ +/* + * Copyright (c) 2025 Huawei Technologies R & D (UK) Ltd + * Copyright (C) 2025 NVIDIA + * Written by Nicolin Chen, Shameer Kolothum + * + * SPDX-License-Identifier: GPL-2.0-or-later + */ + +#include "qemu/osdep.h" + +#include "hw/arm/smmuv3.h" +#include "smmuv3-accel.h" + +static SMMUv3AccelDevice *smmuv3_accel_get_dev(SMMUState *bs, SMMUPciBus *sbus, + PCIBus *bus, int devfn) +{ + SMMUDevice *sdev = sbus->pbdev[devfn]; + SMMUv3AccelDevice *accel_dev; + + if (sdev) { + return container_of(sdev, SMMUv3AccelDevice, sdev); + } + + accel_dev = g_new0(SMMUv3AccelDevice, 1); + sdev = &accel_dev->sdev; + + sbus->pbdev[devfn] = sdev; + smmu_init_sdev(bs, sdev, bus, devfn); + return accel_dev; +} + +/* + * Find or add an address space for the given PCI device. + * + * If a device matching @bus and @devfn already exists, return its + * corresponding address space. Otherwise, create a new device entry + * and initialize address space for it. + */ +static AddressSpace *smmuv3_accel_find_add_as(PCIBus *bus, void *opaque, + int devfn) +{ + SMMUState *bs = opaque; + SMMUPciBus *sbus = smmu_get_sbus(bs, bus); + SMMUv3AccelDevice *accel_dev = smmuv3_accel_get_dev(bs, sbus, bus, devfn); + SMMUDevice *sdev = &accel_dev->sdev; + + return &sdev->as; +} + +static const PCIIOMMUOps smmuv3_accel_ops = { + .get_address_space = smmuv3_accel_find_add_as, +}; + +void smmuv3_accel_init(SMMUv3State *s) +{ + SMMUState *bs = ARM_SMMU(s); + + bs->iommu_ops = &smmuv3_accel_ops; +} diff --git a/hw/arm/smmuv3-accel.h b/hw/arm/smmuv3-accel.h new file mode 100644 index XXXXXXX..XXXXXXX --- /dev/null +++ b/hw/arm/smmuv3-accel.h @@ -XXX,XX +XXX,XX @@ +/* + * Copyright (c) 2025 Huawei Technologies R & D (UK) Ltd + * Copyright (C) 2025 NVIDIA + * Written by Nicolin Chen, Shameer Kolothum + * + * SPDX-License-Identifier: GPL-2.0-or-later + */ + +#ifndef HW_ARM_SMMUV3_ACCEL_H +#define HW_ARM_SMMUV3_ACCEL_H + +#include "hw/arm/smmu-common.h" +#include CONFIG_DEVICES + +typedef struct SMMUv3AccelDevice { + SMMUDevice sdev; +} SMMUv3AccelDevice; + +#ifdef CONFIG_ARM_SMMUV3_ACCEL +void smmuv3_accel_init(SMMUv3State *s); +#else +static inline void smmuv3_accel_init(SMMUv3State *s) +{ +} +#endif + +#endif /* HW_ARM_SMMUV3_ACCEL_H */ diff --git a/hw/arm/smmuv3.c b/hw/arm/smmuv3.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/smmuv3.c +++ b/hw/arm/smmuv3.c @@ -XXX,XX +XXX,XX @@ #include "qapi/error.h" #include "hw/arm/smmuv3.h" +#include "smmuv3-accel.h" #include "smmuv3-internal.h" #include "smmu-internal.h" @@ -XXX,XX +XXX,XX @@ static void smmu_realize(DeviceState *d, Error **errp) SysBusDevice *dev = SYS_BUS_DEVICE(d); Error *local_err = NULL; + if (s->accel) { + smmuv3_accel_init(s); + } + c->parent_realize(d, &local_err); if (local_err) { error_propagate(errp, local_err); diff --git a/include/hw/arm/smmuv3.h b/include/hw/arm/smmuv3.h index XXXXXXX..XXXXXXX 100644 --- a/include/hw/arm/smmuv3.h +++ b/include/hw/arm/smmuv3.h @@ -XXX,XX +XXX,XX @@ struct SMMUv3State { qemu_irq irq[4]; QemuMutex mutex; char *stage; + + /* SMMU has HW accelerator support for nested S1 + s2 */ + bool accel; }; typedef enum { -- 2.43.0 From: Shameer Kolothum <skolothumtho@nvidia.com> To support accelerated SMMUv3 instances, introduce a shared system-wide AddressSpace (shared_as_sysmem) that aliases the global system memory. This shared AddressSpace will be used in a subsequent patch for all vfio-pci devices behind all accelerated SMMUv3 instances within a VM. Sharing a single system AddressSpace ensures that all devices behind accelerated SMMUv3s use the same system address space pointer. This allows VFIO/iommufd to reuse a single IOAS ID in iommufd_cdev_attach(), enabling the Stage-2 page tables to be shared within the VM rather than duplicated for each SMMUv3 instance. Reviewed-by: Nicolin Chen <nicolinc@nvidia.com> Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com> Reviewed-by: Eric Auger <eric.auger@redhat.com> Tested-by: Zhangfei Gao <zhangfei.gao@linaro.org> Tested-by: Eric Auger <eric.auger@redhat.com> Signed-off-by: Shameer Kolothum <skolothumtho@nvidia.com> Message-id: 20260126104342.253965-7-skolothumtho@nvidia.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/arm/smmuv3-accel.c | 26 ++++++++++++++++++++++++++ 1 file changed, 26 insertions(+) diff --git a/hw/arm/smmuv3-accel.c b/hw/arm/smmuv3-accel.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/smmuv3-accel.c +++ b/hw/arm/smmuv3-accel.c @@ -XXX,XX +XXX,XX @@ #include "hw/arm/smmuv3.h" #include "smmuv3-accel.h" +/* + * The root region aliases the global system memory, and shared_as_sysmem + * provides a shared Address Space referencing it. This Address Space is used + * by all vfio-pci devices behind all accelerated SMMUv3 instances within a VM. + */ +static MemoryRegion root, sysmem; +static AddressSpace *shared_as_sysmem; + static SMMUv3AccelDevice *smmuv3_accel_get_dev(SMMUState *bs, SMMUPciBus *sbus, PCIBus *bus, int devfn) { @@ -XXX,XX +XXX,XX @@ static const PCIIOMMUOps smmuv3_accel_ops = { .get_address_space = smmuv3_accel_find_add_as, }; +static void smmuv3_accel_as_init(SMMUv3State *s) +{ + + if (shared_as_sysmem) { + return; + } + + memory_region_init(&root, OBJECT(s), "root", UINT64_MAX); + memory_region_init_alias(&sysmem, OBJECT(s), "smmuv3-accel-sysmem", + get_system_memory(), 0, + memory_region_size(get_system_memory())); + memory_region_add_subregion(&root, 0, &sysmem); + + shared_as_sysmem = g_new0(AddressSpace, 1); + address_space_init(shared_as_sysmem, &root, "smmuv3-accel-as-sysmem"); +} + void smmuv3_accel_init(SMMUv3State *s) { SMMUState *bs = ARM_SMMU(s); bs->iommu_ops = &smmuv3_accel_ops; + smmuv3_accel_as_init(s); } -- 2.43.0 From: Shameer Kolothum <skolothumtho@nvidia.com> During PCI hotplug, in do_pci_register_device(), pci_init_bus_master() is called before storing the pci_dev pointer in bus->devices[devfn]. This causes a problem if pci_init_bus_master() (via its get_address_space() callback) attempts to retrieve the device using pci_find_device(), since the PCI device is not yet visible on the bus. Fix this by moving the pci_init_bus_master() call to after the device has been added to bus->devices[devfn]. This prepares for a subsequent patch where the accel SMMUv3 get_address_space() callback retrieves the pci_dev to identify the attached device type. No functional change intended. Cc: Michael S. Tsirkin <mst@redhat.com> Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com> Reviewed-by: Eric Auger <eric.auger@redhat.com> Reviewed-by: Nicolin Chen <nicolinc@nvidia.com> Reviewed-by: Michael S. Tsirkin <mst@redhat.com> Tested-by: Eric Auger <eric.auger@redhat.com> Tested-by: Zhangfei Gao <zhangfei.gao@linaro.org> Signed-off-by: Shameer Kolothum <skolothumtho@nvidia.com> Reviewed-by: Yi Liu <yi.l.liu@intel.com> Message-id: 20260126104342.253965-8-skolothumtho@nvidia.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/pci/pci.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/hw/pci/pci.c b/hw/pci/pci.c index XXXXXXX..XXXXXXX 100644 --- a/hw/pci/pci.c +++ b/hw/pci/pci.c @@ -XXX,XX +XXX,XX @@ static PCIDevice *do_pci_register_device(PCIDevice *pci_dev, pci_dev->bus_master_as.max_bounce_buffer_size = pci_dev->max_bounce_buffer_size; - if (phase_check(PHASE_MACHINE_READY)) { - pci_init_bus_master(pci_dev); - } pci_dev->irq_state = 0; pci_config_alloc(pci_dev); @@ -XXX,XX +XXX,XX @@ static PCIDevice *do_pci_register_device(PCIDevice *pci_dev, pci_dev->config_write = config_write; bus->devices[devfn] = pci_dev; pci_dev->version_id = 2; /* Current pci device vmstate version */ + if (phase_check(PHASE_MACHINE_READY)) { + pci_init_bus_master(pci_dev); + } return pci_dev; } -- 2.43.0 From: Shameer Kolothum <skolothumtho@nvidia.com> Introduce an optional supports_address_space() callback in PCIIOMMUOps to allow a vIOMMU implementation to reject devices that should not be attached to it. Currently, get_address_space() is the first and mandatory callback into the vIOMMU layer, which always returns an address space. For certain setups, such as hardware accelerated vIOMMUs (e.g. ARM SMMUv3 with accel=on), attaching emulated endpoint devices is undesirable as it may impact the behavior or performance of VFIO passthrough devices, for example, by triggering unnecessary invalidations on the host IOMMU. The new callback allows a vIOMMU to check and reject unsupported devices early during PCI device registration. Cc: Michael S. Tsirkin <mst@redhat.com> Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com> Reviewed-by: Eric Auger <eric.auger@redhat.com> Reviewed-by: Nicolin Chen <nicolinc@nvidia.com> Reviewed-by: Michael S. Tsirkin <mst@redhat.com> Tested-by: Eric Auger <eric.auger@redhat.com> Tested-by: Zhangfei Gao <zhangfei.gao@linaro.org> Signed-off-by: Shameer Kolothum <skolothumtho@nvidia.com> Reviewed-by: Yi Liu <yi.l.liu@intel.com> Message-id: 20260126104342.253965-9-skolothumtho@nvidia.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/pci/pci.c | 20 ++++++++++++++++++++ include/hw/pci/pci.h | 19 +++++++++++++++++++ 2 files changed, 39 insertions(+) diff --git a/hw/pci/pci.c b/hw/pci/pci.c index XXXXXXX..XXXXXXX 100644 --- a/hw/pci/pci.c +++ b/hw/pci/pci.c @@ -XXX,XX +XXX,XX @@ static void pci_set_master(PCIDevice *d, bool enable) d->is_master = enable; /* cache the status */ } +static bool +pci_device_supports_iommu_address_space(PCIDevice *dev, Error **errp) +{ + PCIBus *bus; + PCIBus *iommu_bus; + int devfn; + + pci_device_get_iommu_bus_devfn(dev, &iommu_bus, &bus, &devfn); + if (iommu_bus && iommu_bus->iommu_ops->supports_address_space) { + return iommu_bus->iommu_ops->supports_address_space(bus, + iommu_bus->iommu_opaque, devfn, errp); + } + return true; +} + static void pci_init_bus_master(PCIDevice *pci_dev) { AddressSpace *dma_as = pci_device_iommu_address_space(pci_dev); @@ -XXX,XX +XXX,XX @@ static PCIDevice *do_pci_register_device(PCIDevice *pci_dev, pci_dev->config_write = config_write; bus->devices[devfn] = pci_dev; pci_dev->version_id = 2; /* Current pci device vmstate version */ + if (!pci_device_supports_iommu_address_space(pci_dev, errp)) { + do_pci_unregister_device(pci_dev); + bus->devices[devfn] = NULL; + return NULL; + } if (phase_check(PHASE_MACHINE_READY)) { pci_init_bus_master(pci_dev); } diff --git a/include/hw/pci/pci.h b/include/hw/pci/pci.h index XXXXXXX..XXXXXXX 100644 --- a/include/hw/pci/pci.h +++ b/include/hw/pci/pci.h @@ -XXX,XX +XXX,XX @@ typedef struct IOMMUPRINotifier { * framework for a set of devices on a PCI bus. */ typedef struct PCIIOMMUOps { + /** + * @supports_address_space: Optional pre-check to determine whether a PCI + * device can be associated with an IOMMU. If this callback returns true, + * the IOMMU accepts the device association and get_address_space() can be + * called to obtain the address_space to be used. + * + * @bus: the #PCIBus being accessed. + * + * @opaque: the data passed to pci_setup_iommu(). + * + * @devfn: device and function number. + * + * @errp: pass an Error out only when return false + * + * Returns: true if the device can be associated with an IOMMU, false + * otherwise with errp set. + */ + bool (*supports_address_space)(PCIBus *bus, void *opaque, int devfn, + Error **errp); /** * @get_address_space: get the address space for a set of devices * on a PCI bus. -- 2.43.0 From: Shameer Kolothum <skolothumtho@nvidia.com> Move the TYPE_PXB_PCIE_DEV definition to header so that it can be referenced by other code in subsequent patch. Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com> Reviewed-by: Eric Auger <eric.auger@redhat.com> Reviewed-by: Nicolin Chen <nicolinc@nvidia.com> Tested-by: Eric Auger <eric.auger@redhat.com> Tested-by: Zhangfei Gao <zhangfei.gao@linaro.org> Signed-off-by: Shameer Kolothum <skolothumtho@nvidia.com> Reviewed-by: Yi Liu <yi.l.liu@intel.com> Message-id: 20260126104342.253965-10-skolothumtho@nvidia.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/pci-bridge/pci_expander_bridge.c | 1 - include/hw/pci/pci_bridge.h | 1 + 2 files changed, 1 insertion(+), 1 deletion(-) diff --git a/hw/pci-bridge/pci_expander_bridge.c b/hw/pci-bridge/pci_expander_bridge.c index XXXXXXX..XXXXXXX 100644 --- a/hw/pci-bridge/pci_expander_bridge.c +++ b/hw/pci-bridge/pci_expander_bridge.c @@ -XXX,XX +XXX,XX @@ struct PXBBus { char bus_path[8]; }; -#define TYPE_PXB_PCIE_DEV "pxb-pcie" OBJECT_DECLARE_SIMPLE_TYPE(PXBPCIEDev, PXB_PCIE_DEV) static GList *pxb_dev_list; diff --git a/include/hw/pci/pci_bridge.h b/include/hw/pci/pci_bridge.h index XXXXXXX..XXXXXXX 100644 --- a/include/hw/pci/pci_bridge.h +++ b/include/hw/pci/pci_bridge.h @@ -XXX,XX +XXX,XX @@ typedef struct PXBPCIEDev { #define TYPE_PXB_PCIE_BUS "pxb-pcie-bus" #define TYPE_PXB_CXL_BUS "pxb-cxl-bus" +#define TYPE_PXB_PCIE_DEV "pxb-pcie" #define TYPE_PXB_DEV "pxb" OBJECT_DECLARE_SIMPLE_TYPE(PXBDev, PXB_DEV) -- 2.43.0 From: Shameer Kolothum <skolothumtho@nvidia.com> Accelerated SMMUv3 is only meaningful when a device can leverage the host SMMUv3 in nested mode (S1+S2 translation). To keep the model consistent and correct, this mode is restricted to vfio-pci endpoint devices using the iommufd backend. Non-endpoint emulated devices such as PCIe root ports and bridges are also permitted so that vfio-pci devices can be attached downstream. All other device types are unsupported in accelerated mode. Implement supports_address_space() callback to reject all such unsupported devices. This restriction also avoids complications with IOTLB invalidations. Some TLBI commands (e.g. CMD_TLBI_NH_ASID) lack an associated SID, making it difficult to trace the originating device. Allowing emulated endpoints would require invalidating both QEMU’s software IOTLB and the host’s hardware IOTLB, which can significantly degrade performance. A key design choice is the address space returned for accelerated vfio-pci endpoints. VFIO core has a container that manages an HWPT. By default, it allocates a stage-1 normal HWPT, unless vIOMMU requests for a nesting parent HWPT for accelerated cases. VFIO core adds a listener for that HWPT and sets up a handler vfio_container_region_add() where it checks the memory region. -If the region is a non-IOMMU translated one (system address space), VFIO treats it as RAM and handles all stage-2 mappings for the core allocated nesting parent HWPT. -If the region is an IOMMU address space, VFIO instead enables IOTLB notifier handling and translation replay, skipping the RAM listener and therefore not installing stage-2 mappings. For accelerated SMMUv3, correct operation requires the S1+S2 nesting model, and therefore VFIO must take the "system address space" path so that stage-2 mappings are properly built. Returning an alias of the system address space ensures this happens. Returning the IOMMU address space would omit stage-2 mapping and break nested translation. Another option considered was forcing a pre-registration path using vfio_prereg_listener() to set up stage-2 mappings, but this requires changes in VFIO core and was not adopted. Returning an alias of the system address space keeps the design aligned with existing VFIO/iommufd nesting flows and avoids the need for cross-subsystem changes. In summary: - vfio-pci devices(with iommufd as backend) return an address space aliased to system address space. - bridges and root ports return the IOMMU address space. Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com> Reviewed-by: Nicolin Chen <nicolinc@nvidia.com> Reviewed-by: Eric Auger <eric.auger@redhat.com> Tested-by: Eric Auger <eric.auger@redhat.com> Tested-by: Zhangfei Gao <zhangfei.gao@linaro.org> Signed-off-by: Shameer Kolothum <skolothumtho@nvidia.com> Message-id: 20260126104342.253965-11-skolothumtho@nvidia.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/arm/smmuv3-accel.c | 77 ++++++++++++++++++++++++++++++++++++++++++- 1 file changed, 76 insertions(+), 1 deletion(-) diff --git a/hw/arm/smmuv3-accel.c b/hw/arm/smmuv3-accel.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/smmuv3-accel.c +++ b/hw/arm/smmuv3-accel.c @@ -XXX,XX +XXX,XX @@ */ #include "qemu/osdep.h" +#include "qemu/error-report.h" #include "hw/arm/smmuv3.h" +#include "hw/pci/pci_bridge.h" +#include "hw/pci-host/gpex.h" +#include "hw/vfio/pci.h" + #include "smmuv3-accel.h" /* @@ -XXX,XX +XXX,XX @@ static SMMUv3AccelDevice *smmuv3_accel_get_dev(SMMUState *bs, SMMUPciBus *sbus, return accel_dev; } +/* + * Only allow PCIe bridges, pxb-pcie roots, and GPEX roots so vfio-pci + * endpoints can sit downstream. Accelerated SMMUv3 requires a vfio-pci + * endpoint using the iommufd backend; all other device types are rejected. + * This avoids supporting emulated endpoints, which would complicate IOTLB + * invalidation and hurt performance. + */ +static bool smmuv3_accel_pdev_allowed(PCIDevice *pdev, bool *vfio_pci) +{ + + if (object_dynamic_cast(OBJECT(pdev), TYPE_PCI_BRIDGE) || + object_dynamic_cast(OBJECT(pdev), TYPE_PXB_PCIE_DEV) || + object_dynamic_cast(OBJECT(pdev), TYPE_GPEX_ROOT_DEVICE)) { + return true; + } else if ((object_dynamic_cast(OBJECT(pdev), TYPE_VFIO_PCI))) { + *vfio_pci = true; + if (object_property_get_link(OBJECT(pdev), "iommufd", NULL)) { + return true; + } + } + return false; +} + +static bool smmuv3_accel_supports_as(PCIBus *bus, void *opaque, int devfn, + Error **errp) +{ + PCIDevice *pdev = pci_find_device(bus, pci_bus_num(bus), devfn); + bool vfio_pci = false; + + if (pdev && !smmuv3_accel_pdev_allowed(pdev, &vfio_pci)) { + if (vfio_pci) { + error_setg(errp, "vfio-pci endpoint devices without an iommufd " + "backend not allowed when using arm-smmuv3,accel=on"); + + } else { + error_setg(errp, "Emulated endpoint devices are not allowed when " + "using arm-smmuv3,accel=on"); + } + return false; + } + return true; +} /* * Find or add an address space for the given PCI device. * @@ -XXX,XX +XXX,XX @@ static SMMUv3AccelDevice *smmuv3_accel_get_dev(SMMUState *bs, SMMUPciBus *sbus, static AddressSpace *smmuv3_accel_find_add_as(PCIBus *bus, void *opaque, int devfn) { + PCIDevice *pdev = pci_find_device(bus, pci_bus_num(bus), devfn); SMMUState *bs = opaque; SMMUPciBus *sbus = smmu_get_sbus(bs, bus); SMMUv3AccelDevice *accel_dev = smmuv3_accel_get_dev(bs, sbus, bus, devfn); SMMUDevice *sdev = &accel_dev->sdev; + bool vfio_pci = false; - return &sdev->as; + if (pdev && !smmuv3_accel_pdev_allowed(pdev, &vfio_pci)) { + /* Should never be here: supports_address_space() filters these out */ + g_assert_not_reached(); + } + + /* + * In the accelerated mode, a vfio-pci device attached via the iommufd + * backend must remain in the system address space. Such a device is + * always translated by its physical SMMU (using either a stage-2-only + * STE or a nested STE), where the parent stage-2 page table is allocated + * by the VFIO core to back the system address space. + * + * Return the shared_as_sysmem aliased to the global system memory in this + * case. Sharing address_space_memory also allows devices under different + * vSMMU instances in the same VM to reuse a single nesting parent HWPT in + * the VFIO core. + * + * For non-endpoint emulated devices such as PCIe root ports and bridges, + * which may use the normal emulated translation path and software IOTLBs, + * return the SMMU's IOMMU address space. + */ + if (vfio_pci) { + return shared_as_sysmem; + } else { + return &sdev->as; + } } static const PCIIOMMUOps smmuv3_accel_ops = { + .supports_address_space = smmuv3_accel_supports_as, .get_address_space = smmuv3_accel_find_add_as, }; -- 2.43.0 From: Shameer Kolothum <skolothumtho@nvidia.com> For accelerated SMMUv3, we need nested parent domain creation. Add the callback support so that VFIO can create a nested parent. Reviewed-by: Nicolin Chen <nicolinc@nvidia.com> Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com> Signed-off-by: Shameer Kolothum <shameerali.kolothum.thodi@huawei.com> Tested-by: Zhangfei Gao <zhangfei.gao@linaro.org> Reviewed-by: Eric Auger <eric.auger@redhat.com> Tested-by: Eric Auger <eric.auger@redhat.com> Signed-off-by: Shameer Kolothum <skolothumtho@nvidia.com> Message-id: 20260126104342.253965-12-skolothumtho@nvidia.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/arm/smmuv3-accel.c | 13 +++++++++++++ 1 file changed, 13 insertions(+) diff --git a/hw/arm/smmuv3-accel.c b/hw/arm/smmuv3-accel.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/smmuv3-accel.c +++ b/hw/arm/smmuv3-accel.c @@ -XXX,XX +XXX,XX @@ #include "qemu/error-report.h" #include "hw/arm/smmuv3.h" +#include "hw/core/iommu.h" #include "hw/pci/pci_bridge.h" #include "hw/pci-host/gpex.h" #include "hw/vfio/pci.h" @@ -XXX,XX +XXX,XX @@ static AddressSpace *smmuv3_accel_find_add_as(PCIBus *bus, void *opaque, } } +static uint64_t smmuv3_accel_get_viommu_flags(void *opaque) +{ + /* + * We return VIOMMU_FLAG_WANT_NESTING_PARENT to inform VFIO core to create a + * nesting parent which is required for accelerated SMMUv3 support. + * The real HW nested support should be reported from host SMMUv3 and if + * it doesn't, the nesting parent allocation will fail anyway in VFIO core. + */ + return VIOMMU_FLAG_WANT_NESTING_PARENT; +} + static const PCIIOMMUOps smmuv3_accel_ops = { .supports_address_space = smmuv3_accel_supports_as, .get_address_space = smmuv3_accel_find_add_as, + .get_viommu_flags = smmuv3_accel_get_viommu_flags, }; static void smmuv3_accel_as_init(SMMUv3State *s) -- 2.43.0 From: Nicolin Chen <nicolinc@nvidia.com> Implement the VFIO/PCI callbacks to attach and detach a HostIOMMUDevice to a vSMMUv3 when accel=on, - set_iommu_device(): attach a HostIOMMUDevice to a vIOMMU - unset_iommu_device(): detach and release associated resources In SMMUv3 accel=on mode, the guest SMMUv3 is backed by the host SMMUv3 via IOMMUFD. A vIOMMU object (created via IOMMU_VIOMMU_ALLOC) provides a per-VM, security-isolated handle to the physical SMMUv3. Without a vIOMMU, the vSMMUv3 cannot relay guest operations to the host hardware nor maintain isolation across VMs or devices. Therefore, set_iommu_device() allocates a vIOMMU object if one does not already exist. There are two main points to consider in this implementation: 1) VFIO core allocates and attaches a S2 HWPT that acts as the nesting parent for nested HWPTs(IOMMU_DOMAIN_NESTED). This parent HWPT will be shared across multiple vSMMU instances within a VM. 2) A device cannot attach directly to a vIOMMU. Instead, it attaches through a proxy nested HWPT (IOMMU_DOMAIN_NESTED). Based on the STE configuration,there are three types of nested HWPTs: bypass, abort, and translate. -The bypass and abort proxy HWPTs are pre-allocated. When SMMUv3 operates in global abort or bypass modes, as controlled by the GBPA register, or issues a vSTE for bypass or abort we attach these pre-allocated nested HWPTs. -The translate HWPT requires a vDEVICE to be allocated first, since invalidations and events depend on a valid vSID. -The vDEVICE allocation and attach operations for vSTE based HWPTs are implemented in subsequent patches. In summary, a device placed behind a vSMMU instance must have a vSID for translate vSTE. The bypass and abort vSTEs are pre-allocated as proxy nested HWPTs and is attached based on GBPA register. The core-managed nesting parent S2 HWPT is used as parent S2 HWPT for all the nested HWPTs and is intended to be shared across vSMMU instances within the same VM. set_iommu_device(): - Reuse an existing vIOMMU for the same physical SMMU if available. If not, allocate a new one using the nesting parent S2 HWPT. - Pre-allocate two proxy nested HWPTs (bypass and abort) under the vIOMMU and install one based on GBPA.ABORT value. - Add the device to the vIOMMU’s device list. unset_iommu_device(): - Re-attach device to the nesting parent S2 HWPT. - Remove the device from the vIOMMU’s device list. - If the list is empty, free the proxy HWPTs (bypass and abort) and release the vIOMMU object. Introduce struct SMMUv3AccelState, representing an accelerated SMMUv3 instance backed by an iommufd vIOMMU object, and storing the bypass and abort proxy HWPT IDs. Signed-off-by: Nicolin Chen <nicolinc@nvidia.com> Signed-off-by: Shameer Kolothum <shameerali.kolothum.thodi@huawei.com Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com> Tested-by: Zhangfei Gao <zhangfei.gao@linaro.org> Reviewed-by: Eric Auger <eric.auger@redhat.com> Tested-by: Eric Auger <eric.auger@redhat.com> Signed-off-by: Shameer Kolothum <skolothumtho@nvidia.com> Message-id: 20260126104342.253965-13-skolothumtho@nvidia.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/arm/smmuv3-accel.c | 156 +++++++++++++++++++++++++++++++++ hw/arm/smmuv3-accel.h | 18 ++++ hw/arm/trace-events | 4 + include/hw/arm/smmuv3-common.h | 3 + include/hw/arm/smmuv3.h | 1 + 5 files changed, 182 insertions(+) diff --git a/hw/arm/smmuv3-accel.c b/hw/arm/smmuv3-accel.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/smmuv3-accel.c +++ b/hw/arm/smmuv3-accel.c @@ -XXX,XX +XXX,XX @@ #include "qemu/osdep.h" #include "qemu/error-report.h" +#include "trace.h" #include "hw/arm/smmuv3.h" #include "hw/core/iommu.h" @@ -XXX,XX +XXX,XX @@ #include "hw/pci-host/gpex.h" #include "hw/vfio/pci.h" +#include "smmuv3-internal.h" #include "smmuv3-accel.h" /* @@ -XXX,XX +XXX,XX @@ static SMMUv3AccelDevice *smmuv3_accel_get_dev(SMMUState *bs, SMMUPciBus *sbus, return accel_dev; } +static uint32_t smmuv3_accel_gbpa_hwpt(SMMUv3State *s, SMMUv3AccelState *accel) +{ + return FIELD_EX32(s->gbpa, GBPA, ABORT) ? + accel->abort_hwpt_id : accel->bypass_hwpt_id; +} + +static bool +smmuv3_accel_alloc_viommu(SMMUv3State *s, HostIOMMUDeviceIOMMUFD *idev, + Error **errp) +{ + SMMUv3AccelState *accel = s->s_accel; + struct iommu_hwpt_arm_smmuv3 bypass_data = { + .ste = { SMMU_STE_CFG_BYPASS | SMMU_STE_VALID, 0x0ULL }, + }; + struct iommu_hwpt_arm_smmuv3 abort_data = { + .ste = { SMMU_STE_VALID, 0x0ULL }, + }; + uint32_t s2_hwpt_id = idev->hwpt_id; + uint32_t viommu_id, hwpt_id; + IOMMUFDViommu *viommu; + + if (!iommufd_backend_alloc_viommu(idev->iommufd, idev->devid, + IOMMU_VIOMMU_TYPE_ARM_SMMUV3, + s2_hwpt_id, &viommu_id, errp)) { + return false; + } + + viommu = g_new0(IOMMUFDViommu, 1); + viommu->viommu_id = viommu_id; + viommu->s2_hwpt_id = s2_hwpt_id; + viommu->iommufd = idev->iommufd; + + /* + * Pre-allocate HWPTs for S1 bypass and abort cases. These will be attached + * later for guest STEs or GBPAs that require bypass or abort configuration. + */ + if (!iommufd_backend_alloc_hwpt(idev->iommufd, idev->devid, viommu_id, + 0, IOMMU_HWPT_DATA_ARM_SMMUV3, + sizeof(abort_data), &abort_data, + &accel->abort_hwpt_id, errp)) { + goto free_viommu; + } + + if (!iommufd_backend_alloc_hwpt(idev->iommufd, idev->devid, viommu_id, + 0, IOMMU_HWPT_DATA_ARM_SMMUV3, + sizeof(bypass_data), &bypass_data, + &accel->bypass_hwpt_id, errp)) { + goto free_abort_hwpt; + } + + /* Attach a HWPT based on SMMUv3 GBPA.ABORT value */ + hwpt_id = smmuv3_accel_gbpa_hwpt(s, accel); + if (!host_iommu_device_iommufd_attach_hwpt(idev, hwpt_id, errp)) { + goto free_bypass_hwpt; + } + accel->viommu = viommu; + return true; + +free_bypass_hwpt: + iommufd_backend_free_id(idev->iommufd, accel->bypass_hwpt_id); +free_abort_hwpt: + iommufd_backend_free_id(idev->iommufd, accel->abort_hwpt_id); +free_viommu: + iommufd_backend_free_id(idev->iommufd, viommu->viommu_id); + g_free(viommu); + return false; +} + +static bool smmuv3_accel_set_iommu_device(PCIBus *bus, void *opaque, int devfn, + HostIOMMUDevice *hiod, Error **errp) +{ + HostIOMMUDeviceIOMMUFD *idev = HOST_IOMMU_DEVICE_IOMMUFD(hiod); + SMMUState *bs = opaque; + SMMUv3State *s = ARM_SMMUV3(bs); + SMMUPciBus *sbus = smmu_get_sbus(bs, bus); + SMMUv3AccelDevice *accel_dev = smmuv3_accel_get_dev(bs, sbus, bus, devfn); + + if (!idev) { + return true; + } + + if (accel_dev->idev) { + if (accel_dev->idev != idev) { + error_setg(errp, "Device already has an associated idev 0x%x", + idev->devid); + return false; + } + return true; + } + + if (s->s_accel->viommu) { + goto done; + } + + if (!smmuv3_accel_alloc_viommu(s, idev, errp)) { + error_append_hint(errp, "Unable to alloc vIOMMU: idev devid 0x%x: ", + idev->devid); + return false; + } + +done: + accel_dev->idev = idev; + accel_dev->s_accel = s->s_accel; + QLIST_INSERT_HEAD(&s->s_accel->device_list, accel_dev, next); + trace_smmuv3_accel_set_iommu_device(devfn, idev->devid); + return true; +} + +static void smmuv3_accel_unset_iommu_device(PCIBus *bus, void *opaque, + int devfn) +{ + SMMUState *bs = opaque; + SMMUPciBus *sbus = g_hash_table_lookup(bs->smmu_pcibus_by_busptr, bus); + HostIOMMUDeviceIOMMUFD *idev; + SMMUv3AccelDevice *accel_dev; + SMMUv3AccelState *accel; + SMMUDevice *sdev; + + if (!sbus) { + return; + } + + sdev = sbus->pbdev[devfn]; + if (!sdev) { + return; + } + + accel_dev = container_of(sdev, SMMUv3AccelDevice, sdev); + idev = accel_dev->idev; + accel = accel_dev->s_accel; + /* Re-attach the default s2 hwpt id */ + if (!host_iommu_device_iommufd_attach_hwpt(idev, idev->hwpt_id, NULL)) { + error_report("Unable to attach the default HW pagetable: idev devid " + "0x%x", idev->devid); + } + + accel_dev->idev = NULL; + accel_dev->s_accel = NULL; + QLIST_REMOVE(accel_dev, next); + trace_smmuv3_accel_unset_iommu_device(devfn, idev->devid); + + if (QLIST_EMPTY(&accel->device_list)) { + iommufd_backend_free_id(accel->viommu->iommufd, accel->bypass_hwpt_id); + iommufd_backend_free_id(accel->viommu->iommufd, accel->abort_hwpt_id); + iommufd_backend_free_id(accel->viommu->iommufd, + accel->viommu->viommu_id); + g_free(accel->viommu); + accel->viommu = NULL; + } +} + /* * Only allow PCIe bridges, pxb-pcie roots, and GPEX roots so vfio-pci * endpoints can sit downstream. Accelerated SMMUv3 requires a vfio-pci @@ -XXX,XX +XXX,XX @@ static const PCIIOMMUOps smmuv3_accel_ops = { .supports_address_space = smmuv3_accel_supports_as, .get_address_space = smmuv3_accel_find_add_as, .get_viommu_flags = smmuv3_accel_get_viommu_flags, + .set_iommu_device = smmuv3_accel_set_iommu_device, + .unset_iommu_device = smmuv3_accel_unset_iommu_device, }; static void smmuv3_accel_as_init(SMMUv3State *s) @@ -XXX,XX +XXX,XX @@ void smmuv3_accel_init(SMMUv3State *s) { SMMUState *bs = ARM_SMMU(s); + s->s_accel = g_new0(SMMUv3AccelState, 1); bs->iommu_ops = &smmuv3_accel_ops; smmuv3_accel_as_init(s); } diff --git a/hw/arm/smmuv3-accel.h b/hw/arm/smmuv3-accel.h index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/smmuv3-accel.h +++ b/hw/arm/smmuv3-accel.h @@ -XXX,XX +XXX,XX @@ #define HW_ARM_SMMUV3_ACCEL_H #include "hw/arm/smmu-common.h" +#include "system/iommufd.h" +#ifdef CONFIG_LINUX +#include <linux/iommufd.h> +#endif #include CONFIG_DEVICES +/* + * Represents an accelerated SMMU instance backed by an iommufd vIOMMU object. + * Holds bypass and abort proxy HWPT IDs used for device attachment. + */ +typedef struct SMMUv3AccelState { + IOMMUFDViommu *viommu; + uint32_t bypass_hwpt_id; + uint32_t abort_hwpt_id; + QLIST_HEAD(, SMMUv3AccelDevice) device_list; +} SMMUv3AccelState; + typedef struct SMMUv3AccelDevice { SMMUDevice sdev; + HostIOMMUDeviceIOMMUFD *idev; + QLIST_ENTRY(SMMUv3AccelDevice) next; + SMMUv3AccelState *s_accel; } SMMUv3AccelDevice; #ifdef CONFIG_ARM_SMMUV3_ACCEL diff --git a/hw/arm/trace-events b/hw/arm/trace-events index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/trace-events +++ b/hw/arm/trace-events @@ -XXX,XX +XXX,XX @@ smmuv3_notify_flag_del(const char *iommu) "DEL SMMUNotifier node for iommu mr=%s smmuv3_inv_notifiers_iova(const char *name, int asid, int vmid, uint64_t iova, uint8_t tg, uint64_t num_pages, int stage) "iommu mr=%s asid=%d vmid=%d iova=0x%"PRIx64" tg=%d num_pages=0x%"PRIx64" stage=%d" smmu_reset_exit(void) "" +#smmuv3-accel.c +smmuv3_accel_set_iommu_device(int devfn, uint32_t devid) "devfn=0x%x (idev devid=0x%x)" +smmuv3_accel_unset_iommu_device(int devfn, uint32_t devid) "devfn=0x%x (idev devid=0x%x)" + # strongarm.c strongarm_uart_update_parameters(const char *label, int speed, char parity, int data_bits, int stop_bits) "%s speed=%d parity=%c data=%d stop=%d" strongarm_ssp_read_underrun(void) "SSP rx underrun" diff --git a/include/hw/arm/smmuv3-common.h b/include/hw/arm/smmuv3-common.h index XXXXXXX..XXXXXXX 100644 --- a/include/hw/arm/smmuv3-common.h +++ b/include/hw/arm/smmuv3-common.h @@ -XXX,XX +XXX,XX @@ REG32(STE_7, 28) #define STE_CFG_ABORT(config) (!(config & 0x4)) #define STE_CFG_BYPASS(config) (config == 0x4) +#define SMMU_STE_VALID (1ULL << 0) +#define SMMU_STE_CFG_BYPASS (1ULL << 3) + /* Update STE fields */ #define STE_SET_VALID(ste, v) \ ((ste)->word[0] = FIELD_DP32((ste)->word[0], STE_0, VALID, (v))) diff --git a/include/hw/arm/smmuv3.h b/include/hw/arm/smmuv3.h index XXXXXXX..XXXXXXX 100644 --- a/include/hw/arm/smmuv3.h +++ b/include/hw/arm/smmuv3.h @@ -XXX,XX +XXX,XX @@ struct SMMUv3State { /* SMMU has HW accelerator support for nested S1 + s2 */ bool accel; + struct SMMUv3AccelState *s_accel; }; typedef enum { -- 2.43.0 From: Shameer Kolothum <skolothumtho@nvidia.com> smmuv3_cmdq_consume() is updated to return detailed errors via errp. Although this is currently a no-op, it prepares the ground for accel SMMUv3 specific command handling where proper error reporting will be useful. Reviewed-by: Nicolin Chen <nicolinc@nvidia.com> Reviewed-by: Eric Auger <eric.auger@redhat.com> Tested-by: Eric Auger <eric.auger@redhat.com> Tested-by: Zhangfei Gao <zhangfei.gao@linaro.org> Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com> Signed-off-by: Shameer Kolothum <skolothumtho@nvidia.com> Message-id: 20260126104342.253965-14-skolothumtho@nvidia.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/arm/smmuv3.c | 67 +++++++++++++++++++++++++++---------------------- 1 file changed, 37 insertions(+), 30 deletions(-) diff --git a/hw/arm/smmuv3.c b/hw/arm/smmuv3.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/smmuv3.c +++ b/hw/arm/smmuv3.c @@ -XXX,XX +XXX,XX @@ static void smmuv3_range_inval(SMMUState *s, Cmd *cmd, SMMUStage stage) } } -static int smmuv3_cmdq_consume(SMMUv3State *s) +static int smmuv3_cmdq_consume(SMMUv3State *s, Error **errp) { SMMUState *bs = ARM_SMMU(s); SMMUCmdError cmd_error = SMMU_CERROR_NONE; @@ -XXX,XX +XXX,XX @@ static MemTxResult smmu_writell(SMMUv3State *s, hwaddr offset, static MemTxResult smmu_writel(SMMUv3State *s, hwaddr offset, uint64_t data, MemTxAttrs attrs) { + Error *local_err = NULL; + switch (offset) { case A_CR0: s->cr[0] = data; s->cr0ack = data & ~SMMU_CR0_RESERVED; /* in case the command queue has been enabled */ - smmuv3_cmdq_consume(s); - return MEMTX_OK; + smmuv3_cmdq_consume(s, &local_err); + break; case A_CR1: s->cr[1] = data; - return MEMTX_OK; + break; case A_CR2: s->cr[2] = data; - return MEMTX_OK; + break; case A_IRQ_CTRL: s->irq_ctrl = data; - return MEMTX_OK; + break; case A_GERRORN: smmuv3_write_gerrorn(s, data); /* * By acknowledging the CMDQ_ERR, SW may notify cmds can * be processed again */ - smmuv3_cmdq_consume(s); - return MEMTX_OK; + smmuv3_cmdq_consume(s, &local_err); + break; case A_GERROR_IRQ_CFG0: /* 64b */ s->gerror_irq_cfg0 = deposit64(s->gerror_irq_cfg0, 0, 32, data); - return MEMTX_OK; + break; case A_GERROR_IRQ_CFG0 + 4: s->gerror_irq_cfg0 = deposit64(s->gerror_irq_cfg0, 32, 32, data); - return MEMTX_OK; + break; case A_GERROR_IRQ_CFG1: s->gerror_irq_cfg1 = data; - return MEMTX_OK; + break; case A_GERROR_IRQ_CFG2: s->gerror_irq_cfg2 = data; - return MEMTX_OK; + break; case A_GBPA: /* * If UPDATE is not set, the write is ignored. This is the only @@ -XXX,XX +XXX,XX @@ static MemTxResult smmu_writel(SMMUv3State *s, hwaddr offset, /* Ignore update bit as write is synchronous. */ s->gbpa = data & ~R_GBPA_UPDATE_MASK; } - return MEMTX_OK; + break; case A_STRTAB_BASE: /* 64b */ s->strtab_base = deposit64(s->strtab_base, 0, 32, data); - return MEMTX_OK; + break; case A_STRTAB_BASE + 4: s->strtab_base = deposit64(s->strtab_base, 32, 32, data); - return MEMTX_OK; + break; case A_STRTAB_BASE_CFG: s->strtab_base_cfg = data; if (FIELD_EX32(data, STRTAB_BASE_CFG, FMT) == 1) { s->sid_split = FIELD_EX32(data, STRTAB_BASE_CFG, SPLIT); s->features |= SMMU_FEATURE_2LVL_STE; } - return MEMTX_OK; + break; case A_CMDQ_BASE: /* 64b */ s->cmdq.base = deposit64(s->cmdq.base, 0, 32, data); s->cmdq.log2size = extract64(s->cmdq.base, 0, 5); if (s->cmdq.log2size > SMMU_CMDQS) { s->cmdq.log2size = SMMU_CMDQS; } - return MEMTX_OK; + break; case A_CMDQ_BASE + 4: /* 64b */ s->cmdq.base = deposit64(s->cmdq.base, 32, 32, data); - return MEMTX_OK; + break; case A_CMDQ_PROD: s->cmdq.prod = data; - smmuv3_cmdq_consume(s); - return MEMTX_OK; + smmuv3_cmdq_consume(s, &local_err); + break; case A_CMDQ_CONS: s->cmdq.cons = data; - return MEMTX_OK; + break; case A_EVENTQ_BASE: /* 64b */ s->eventq.base = deposit64(s->eventq.base, 0, 32, data); s->eventq.log2size = extract64(s->eventq.base, 0, 5); if (s->eventq.log2size > SMMU_EVENTQS) { s->eventq.log2size = SMMU_EVENTQS; } - return MEMTX_OK; + break; case A_EVENTQ_BASE + 4: s->eventq.base = deposit64(s->eventq.base, 32, 32, data); - return MEMTX_OK; + break; case A_EVENTQ_PROD: s->eventq.prod = data; - return MEMTX_OK; + break; case A_EVENTQ_CONS: s->eventq.cons = data; - return MEMTX_OK; + break; case A_EVENTQ_IRQ_CFG0: /* 64b */ s->eventq_irq_cfg0 = deposit64(s->eventq_irq_cfg0, 0, 32, data); - return MEMTX_OK; + break; case A_EVENTQ_IRQ_CFG0 + 4: s->eventq_irq_cfg0 = deposit64(s->eventq_irq_cfg0, 32, 32, data); - return MEMTX_OK; + break; case A_EVENTQ_IRQ_CFG1: s->eventq_irq_cfg1 = data; - return MEMTX_OK; + break; case A_EVENTQ_IRQ_CFG2: s->eventq_irq_cfg2 = data; - return MEMTX_OK; + break; default: qemu_log_mask(LOG_UNIMP, "%s Unexpected 32-bit access to 0x%"PRIx64" (WI)\n", __func__, offset); - return MEMTX_OK; + break; } + + if (local_err) { + error_report_err(local_err); + } + return MEMTX_OK; } static MemTxResult smmu_write_mmio(void *opaque, hwaddr offset, uint64_t data, -- 2.43.0 From: Nicolin Chen <nicolinc@nvidia.com> A device placed behind a vSMMU instance must have corresponding vSTEs (bypass, abort, or translate) installed. The bypass and abort proxy nested HWPTs are pre-allocated. For translat HWPT, a vDEVICE object is allocated and associated with the vIOMMU for each guest device. This allows the host kernel to establish a virtual SID to physical SID mapping, which is required for handling invalidations and event reporting. An translate HWPT is allocated based on the guest STE configuration and attached to the device when the guest issues SMMU_CMD_CFGI_STE or SMMU_CMD_CFGI_STE_RANGE, provided the STE enables S1 translation. If the guest STE is invalid or S1 translation is disabled, the device is attached to one of the pre-allocated ABORT or BYPASS HWPTs instead. While at it, export smmu_find_ste() for use here. Signed-off-by: Nicolin Chen <nicolinc@nvidia.com> Signed-off-by: Shameer Kolothum <shameerali.kolothum.thodi@huawei.com> Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com> Reviewed-by: Eric Auger <eric.auger@redhat.com> Tested-by: Eric Auger <eric.auger@redhat.com> Tested-by: Zhangfei Gao <zhangfei.gao@linaro.org> Signed-off-by: Shameer Kolothum <skolothumtho@nvidia.com> Message-id: 20260126104342.253965-15-skolothumtho@nvidia.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/arm/smmuv3-accel.c | 197 +++++++++++++++++++++++++++++++++ hw/arm/smmuv3-accel.h | 22 ++++ hw/arm/smmuv3-internal.h | 1 + hw/arm/smmuv3.c | 11 +- hw/arm/trace-events | 2 + include/hw/arm/smmuv3-common.h | 18 +++ 6 files changed, 249 insertions(+), 2 deletions(-) diff --git a/hw/arm/smmuv3-accel.c b/hw/arm/smmuv3-accel.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/smmuv3-accel.c +++ b/hw/arm/smmuv3-accel.c @@ -XXX,XX +XXX,XX @@ static uint32_t smmuv3_accel_gbpa_hwpt(SMMUv3State *s, SMMUv3AccelState *accel) accel->abort_hwpt_id : accel->bypass_hwpt_id; } +static bool +smmuv3_accel_alloc_vdev(SMMUv3AccelDevice *accel_dev, int sid, Error **errp) +{ + SMMUv3AccelState *accel = accel_dev->s_accel; + HostIOMMUDeviceIOMMUFD *idev = accel_dev->idev; + IOMMUFDVdev *vdev = accel_dev->vdev; + uint32_t vdevice_id; + + if (!idev || vdev) { + return true; + } + + if (!iommufd_backend_alloc_vdev(idev->iommufd, idev->devid, + accel->viommu->viommu_id, sid, + &vdevice_id, errp)) { + return false; + } + + vdev = g_new(IOMMUFDVdev, 1); + vdev->vdevice_id = vdevice_id; + vdev->virt_id = sid; + accel_dev->vdev = vdev; + return true; +} + +static SMMUS1Hwpt * +smmuv3_accel_dev_alloc_translate(SMMUv3AccelDevice *accel_dev, STE *ste, + Error **errp) +{ + uint64_t ste_0 = (uint64_t)ste->word[0] | (uint64_t)ste->word[1] << 32; + uint64_t ste_1 = (uint64_t)ste->word[2] | (uint64_t)ste->word[3] << 32; + HostIOMMUDeviceIOMMUFD *idev = accel_dev->idev; + SMMUv3AccelState *accel = accel_dev->s_accel; + struct iommu_hwpt_arm_smmuv3 nested_data = { + .ste = { + cpu_to_le64(ste_0 & STE0_MASK), + cpu_to_le64(ste_1 & STE1_MASK), + }, + }; + uint32_t hwpt_id = 0, flags = 0; + SMMUS1Hwpt *s1_hwpt; + + if (!iommufd_backend_alloc_hwpt(idev->iommufd, idev->devid, + accel->viommu->viommu_id, flags, + IOMMU_HWPT_DATA_ARM_SMMUV3, + sizeof(nested_data), &nested_data, + &hwpt_id, errp)) { + return NULL; + } + + s1_hwpt = g_new0(SMMUS1Hwpt, 1); + s1_hwpt->hwpt_id = hwpt_id; + trace_smmuv3_accel_translate_ste(accel_dev->vdev->virt_id, hwpt_id, + nested_data.ste[1], nested_data.ste[0]); + return s1_hwpt; +} + +bool smmuv3_accel_install_ste(SMMUv3State *s, SMMUDevice *sdev, int sid, + Error **errp) +{ + SMMUEventInfo event = {.type = SMMU_EVT_NONE, .sid = sid, + .inval_ste_allowed = true}; + SMMUv3AccelState *accel = s->s_accel; + SMMUv3AccelDevice *accel_dev; + HostIOMMUDeviceIOMMUFD *idev; + uint32_t config, hwpt_id = 0; + SMMUS1Hwpt *s1_hwpt = NULL; + const char *type; + STE ste; + + if (!accel || !accel->viommu) { + return true; + } + + accel_dev = container_of(sdev, SMMUv3AccelDevice, sdev); + if (!accel_dev->s_accel) { + return true; + } + + idev = accel_dev->idev; + if (!smmuv3_accel_alloc_vdev(accel_dev, sid, errp)) { + return false; + } + + if (smmu_find_ste(sdev->smmu, sid, &ste, &event)) { + /* No STE found, nothing to install */ + return true; + } + + /* + * Install the STE based on SMMU enabled/config: + * - attach a pre-allocated HWPT for abort/bypass + * - or a new HWPT for translate STE + * + * Note: The vdev remains associated with accel_dev even if HWPT + * attach/alloc fails, since the Guest–Host SID mapping stays + * valid as long as the device is behind the accelerated SMMUv3. + */ + if (!smmu_enabled(s)) { + hwpt_id = smmuv3_accel_gbpa_hwpt(s, accel); + } else { + config = STE_CONFIG(&ste); + + if (!STE_VALID(&ste) || STE_CFG_ABORT(config)) { + hwpt_id = accel->abort_hwpt_id; + } else if (STE_CFG_BYPASS(config)) { + hwpt_id = accel->bypass_hwpt_id; + } else if (STE_CFG_S1_TRANSLATE(config)) { + s1_hwpt = smmuv3_accel_dev_alloc_translate(accel_dev, &ste, errp); + if (!s1_hwpt) { + return false; + } + hwpt_id = s1_hwpt->hwpt_id; + } + } + + if (!hwpt_id) { + error_setg(errp, "Invalid STE config for sid 0x%x", + smmu_get_sid(&accel_dev->sdev)); + return false; + } + + if (!host_iommu_device_iommufd_attach_hwpt(idev, hwpt_id, errp)) { + if (s1_hwpt) { + iommufd_backend_free_id(idev->iommufd, s1_hwpt->hwpt_id); + g_free(s1_hwpt); + } + return false; + } + + /* Free the previous s1_hwpt */ + if (accel_dev->s1_hwpt) { + iommufd_backend_free_id(idev->iommufd, accel_dev->s1_hwpt->hwpt_id); + g_free(accel_dev->s1_hwpt); + } + + accel_dev->s1_hwpt = s1_hwpt; + if (hwpt_id == accel->abort_hwpt_id) { + type = "abort"; + } else if (hwpt_id == accel->bypass_hwpt_id) { + type = "bypass"; + } else { + type = "translate"; + } + + trace_smmuv3_accel_install_ste(sid, type, hwpt_id); + return true; +} + +bool smmuv3_accel_install_ste_range(SMMUv3State *s, SMMUSIDRange *range, + Error **errp) +{ + SMMUv3AccelState *accel = s->s_accel; + SMMUv3AccelDevice *accel_dev; + Error *local_err = NULL; + bool all_ok = true; + + if (!accel || !accel->viommu) { + return true; + } + + QLIST_FOREACH(accel_dev, &accel->device_list, next) { + uint32_t sid = smmu_get_sid(&accel_dev->sdev); + + if (sid >= range->start && sid <= range->end) { + if (!smmuv3_accel_install_ste(s, &accel_dev->sdev, + sid, &local_err)) { + error_append_hint(&local_err, "Device 0x%x: Failed to install " + "STE\n", sid); + error_report_err(local_err); + local_err = NULL; + all_ok = false; + } + } + } + + if (!all_ok) { + error_setg(errp, "Failed to install all STEs properly"); + } + return all_ok; +} + static bool smmuv3_accel_alloc_viommu(SMMUv3State *s, HostIOMMUDeviceIOMMUFD *idev, Error **errp) @@ -XXX,XX +XXX,XX @@ static void smmuv3_accel_unset_iommu_device(PCIBus *bus, void *opaque, HostIOMMUDeviceIOMMUFD *idev; SMMUv3AccelDevice *accel_dev; SMMUv3AccelState *accel; + IOMMUFDVdev *vdev; SMMUDevice *sdev; if (!sbus) { @@ -XXX,XX +XXX,XX @@ static void smmuv3_accel_unset_iommu_device(PCIBus *bus, void *opaque, "0x%x", idev->devid); } + if (accel_dev->s1_hwpt) { + iommufd_backend_free_id(accel_dev->idev->iommufd, + accel_dev->s1_hwpt->hwpt_id); + g_free(accel_dev->s1_hwpt); + accel_dev->s1_hwpt = NULL; + } + + vdev = accel_dev->vdev; + if (vdev) { + iommufd_backend_free_id(accel->viommu->iommufd, vdev->vdevice_id); + g_free(vdev); + accel_dev->vdev = NULL; + } + accel_dev->idev = NULL; accel_dev->s_accel = NULL; QLIST_REMOVE(accel_dev, next); diff --git a/hw/arm/smmuv3-accel.h b/hw/arm/smmuv3-accel.h index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/smmuv3-accel.h +++ b/hw/arm/smmuv3-accel.h @@ -XXX,XX +XXX,XX @@ typedef struct SMMUv3AccelState { QLIST_HEAD(, SMMUv3AccelDevice) device_list; } SMMUv3AccelState; +typedef struct SMMUS1Hwpt { + uint32_t hwpt_id; +} SMMUS1Hwpt; + typedef struct SMMUv3AccelDevice { SMMUDevice sdev; HostIOMMUDeviceIOMMUFD *idev; + SMMUS1Hwpt *s1_hwpt; + IOMMUFDVdev *vdev; QLIST_ENTRY(SMMUv3AccelDevice) next; SMMUv3AccelState *s_accel; } SMMUv3AccelDevice; #ifdef CONFIG_ARM_SMMUV3_ACCEL void smmuv3_accel_init(SMMUv3State *s); +bool smmuv3_accel_install_ste(SMMUv3State *s, SMMUDevice *sdev, int sid, + Error **errp); +bool smmuv3_accel_install_ste_range(SMMUv3State *s, SMMUSIDRange *range, + Error **errp); #else static inline void smmuv3_accel_init(SMMUv3State *s) { } +static inline bool +smmuv3_accel_install_ste(SMMUv3State *s, SMMUDevice *sdev, int sid, + Error **errp) +{ + return true; +} +static inline bool +smmuv3_accel_install_ste_range(SMMUv3State *s, SMMUSIDRange *range, + Error **errp) +{ + return true; +} #endif #endif /* HW_ARM_SMMUV3_ACCEL_H */ diff --git a/hw/arm/smmuv3-internal.h b/hw/arm/smmuv3-internal.h index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/smmuv3-internal.h +++ b/hw/arm/smmuv3-internal.h @@ -XXX,XX +XXX,XX @@ typedef struct SMMUEventInfo { } while (0) void smmuv3_record_event(SMMUv3State *s, SMMUEventInfo *event); +int smmu_find_ste(SMMUv3State *s, uint32_t sid, STE *ste, SMMUEventInfo *event); static inline int oas2bits(int oas_field) { diff --git a/hw/arm/smmuv3.c b/hw/arm/smmuv3.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/smmuv3.c +++ b/hw/arm/smmuv3.c @@ -XXX,XX +XXX,XX @@ bad_ste: * Supports linear and 2-level stream table * Return 0 on success, -EINVAL otherwise */ -static int smmu_find_ste(SMMUv3State *s, uint32_t sid, STE *ste, - SMMUEventInfo *event) +int smmu_find_ste(SMMUv3State *s, uint32_t sid, STE *ste, SMMUEventInfo *event) { dma_addr_t addr, strtab_base; uint32_t log2size; @@ -XXX,XX +XXX,XX @@ static int smmuv3_cmdq_consume(SMMUv3State *s, Error **errp) } trace_smmuv3_cmdq_cfgi_ste(sid); + if (!smmuv3_accel_install_ste(s, sdev, sid, errp)) { + cmd_error = SMMU_CERROR_ILL; + break; + } smmuv3_flush_config(sdev); break; @@ -XXX,XX +XXX,XX @@ static int smmuv3_cmdq_consume(SMMUv3State *s, Error **errp) sid_range.end = sid_range.start + mask; trace_smmuv3_cmdq_cfgi_ste_range(sid_range.start, sid_range.end); + if (!smmuv3_accel_install_ste_range(s, &sid_range, errp)) { + cmd_error = SMMU_CERROR_ILL; + break; + } smmu_configs_inv_sid_range(bs, sid_range); break; } diff --git a/hw/arm/trace-events b/hw/arm/trace-events index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/trace-events +++ b/hw/arm/trace-events @@ -XXX,XX +XXX,XX @@ smmu_reset_exit(void) "" #smmuv3-accel.c smmuv3_accel_set_iommu_device(int devfn, uint32_t devid) "devfn=0x%x (idev devid=0x%x)" smmuv3_accel_unset_iommu_device(int devfn, uint32_t devid) "devfn=0x%x (idev devid=0x%x)" +smmuv3_accel_translate_ste(uint32_t vsid, uint32_t hwpt_id, uint64_t ste_1, uint64_t ste_0) "vSID=0x%x hwpt_id=0x%x ste=%"PRIx64":%"PRIx64 +smmuv3_accel_install_ste(uint32_t vsid, const char * type, uint32_t hwpt_id) "vSID=0x%x ste type=%s hwpt_id=0x%x" # strongarm.c strongarm_uart_update_parameters(const char *label, int speed, char parity, int data_bits, int stop_bits) "%s speed=%d parity=%c data=%d stop=%d" diff --git a/include/hw/arm/smmuv3-common.h b/include/hw/arm/smmuv3-common.h index XXXXXXX..XXXXXXX 100644 --- a/include/hw/arm/smmuv3-common.h +++ b/include/hw/arm/smmuv3-common.h @@ -XXX,XX +XXX,XX @@ REG32(STE_7, 28) #define STE_CFG_S2_ENABLED(config) (config & 0x2) #define STE_CFG_ABORT(config) (!(config & 0x4)) #define STE_CFG_BYPASS(config) (config == 0x4) +#define STE_CFG_S1_TRANSLATE(config) (config == 0x5) #define SMMU_STE_VALID (1ULL << 0) #define SMMU_STE_CFG_BYPASS (1ULL << 3) +#define STE0_V MAKE_64BIT_MASK(0, 1) +#define STE0_CONFIG MAKE_64BIT_MASK(1, 3) +#define STE0_S1FMT MAKE_64BIT_MASK(4, 2) +#define STE0_CTXPTR MAKE_64BIT_MASK(6, 50) +#define STE0_S1CDMAX MAKE_64BIT_MASK(59, 5) +#define STE0_MASK (STE0_S1CDMAX | STE0_CTXPTR | STE0_S1FMT | STE0_CONFIG | \ + STE0_V) + +#define STE1_S1DSS MAKE_64BIT_MASK(0, 2) +#define STE1_S1CIR MAKE_64BIT_MASK(2, 2) +#define STE1_S1COR MAKE_64BIT_MASK(4, 2) +#define STE1_S1CSH MAKE_64BIT_MASK(6, 2) +#define STE1_S1STALLD MAKE_64BIT_MASK(27, 1) +#define STE1_EATS MAKE_64BIT_MASK(28, 2) +#define STE1_MASK (STE1_EATS | STE1_S1STALLD | STE1_S1CSH | STE1_S1COR | \ + STE1_S1CIR | STE1_S1DSS) + /* Update STE fields */ #define STE_SET_VALID(ste, v) \ ((ste)->word[0] = FIELD_DP32((ste)->word[0], STE_0, VALID, (v))) -- 2.43.0 From: Shameer Kolothum <skolothumtho@nvidia.com> On guest reboot or on GBPA update, attach a nested HWPT based on the GPBA.ABORT bit which either aborts all incoming transactions or bypasses them. Reviewed-by: Eric Auger <eric.auger@redhat.com> Reviewed-by: Nicolin Chen <nicolinc@nvidia.com> Tested-by: Eric Auger <eric.auger@redhat.com> Tested-by: Zhangfei Gao <zhangfei.gao@linaro.org> Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com> Signed-off-by: Shameer Kolothum <skolothumtho@nvidia.com> Message-id: 20260126104342.253965-16-skolothumtho@nvidia.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/arm/smmuv3-accel.c | 36 ++++++++++++++++++++++++++++++++++++ hw/arm/smmuv3-accel.h | 9 +++++++++ hw/arm/smmuv3.c | 2 ++ 3 files changed, 47 insertions(+) diff --git a/hw/arm/smmuv3-accel.c b/hw/arm/smmuv3-accel.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/smmuv3-accel.c +++ b/hw/arm/smmuv3-accel.c @@ -XXX,XX +XXX,XX @@ static const PCIIOMMUOps smmuv3_accel_ops = { .unset_iommu_device = smmuv3_accel_unset_iommu_device, }; +/* Based on SMUUv3 GPBA.ABORT configuration, attach a corresponding HWPT */ +bool smmuv3_accel_attach_gbpa_hwpt(SMMUv3State *s, Error **errp) +{ + SMMUv3AccelState *accel = s->s_accel; + SMMUv3AccelDevice *accel_dev; + Error *local_err = NULL; + bool all_ok = true; + uint32_t hwpt_id; + + if (!accel || !accel->viommu) { + return true; + } + + hwpt_id = smmuv3_accel_gbpa_hwpt(s, accel); + QLIST_FOREACH(accel_dev, &accel->device_list, next) { + if (!host_iommu_device_iommufd_attach_hwpt(accel_dev->idev, hwpt_id, + &local_err)) { + error_append_hint(&local_err, "Failed to attach GBPA hwpt %u for " + "idev devid %u", hwpt_id, accel_dev->idev->devid); + error_report_err(local_err); + local_err = NULL; + all_ok = false; + } + } + if (!all_ok) { + error_setg(errp, "Failed to attach all GBPA based HWPTs properly"); + } + return all_ok; +} + +void smmuv3_accel_reset(SMMUv3State *s) +{ + /* Attach a HWPT based on GBPA reset value */ + smmuv3_accel_attach_gbpa_hwpt(s, NULL); +} + static void smmuv3_accel_as_init(SMMUv3State *s) { diff --git a/hw/arm/smmuv3-accel.h b/hw/arm/smmuv3-accel.h index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/smmuv3-accel.h +++ b/hw/arm/smmuv3-accel.h @@ -XXX,XX +XXX,XX @@ bool smmuv3_accel_install_ste(SMMUv3State *s, SMMUDevice *sdev, int sid, Error **errp); bool smmuv3_accel_install_ste_range(SMMUv3State *s, SMMUSIDRange *range, Error **errp); +bool smmuv3_accel_attach_gbpa_hwpt(SMMUv3State *s, Error **errp); +void smmuv3_accel_reset(SMMUv3State *s); #else static inline void smmuv3_accel_init(SMMUv3State *s) { @@ -XXX,XX +XXX,XX @@ smmuv3_accel_install_ste_range(SMMUv3State *s, SMMUSIDRange *range, { return true; } +static inline bool smmuv3_accel_attach_gbpa_hwpt(SMMUv3State *s, Error **errp) +{ + return true; +} +static inline void smmuv3_accel_reset(SMMUv3State *s) +{ +} #endif #endif /* HW_ARM_SMMUV3_ACCEL_H */ diff --git a/hw/arm/smmuv3.c b/hw/arm/smmuv3.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/smmuv3.c +++ b/hw/arm/smmuv3.c @@ -XXX,XX +XXX,XX @@ static MemTxResult smmu_writel(SMMUv3State *s, hwaddr offset, if (data & R_GBPA_UPDATE_MASK) { /* Ignore update bit as write is synchronous. */ s->gbpa = data & ~R_GBPA_UPDATE_MASK; + smmuv3_accel_attach_gbpa_hwpt(s, &local_err); } break; case A_STRTAB_BASE: /* 64b */ @@ -XXX,XX +XXX,XX @@ static void smmu_reset_exit(Object *obj, ResetType type) } smmuv3_init_regs(s); + smmuv3_accel_reset(s); } static void smmu_realize(DeviceState *d, Error **errp) -- 2.43.0 From: Shameer Kolothum <skolothumtho@nvidia.com> For certain vIOMMU implementations, such as SMMUv3 in accelerated mode, the translation tables are programmed directly into the physical SMMUv3 in a nested configuration. While QEMU knows where the guest tables live, safely walking them in software would require trapping and ordering all guest invalidations on every command queue. Without this, QEMU could race with guest updates and walk stale or freed page tables. This constraint is fundamental to the design of HW-accelerated vSMMU when used with downstream vfio-pci endpoint devices, where QEMU must never walk guest translation tables and must rely on the physical SMMU for translation. Future accelerated vSMMU features, such as virtual CMDQ, will also prevent trapping invalidations, reinforcing this restriction. For vfio-pci endpoints behind such a vSMMU, the only translation QEMU needs is for the MSI doorbell used when setting up KVM MSI route tables. Instead of attempting a software walk, introduce an optional vIOMMU callback that returns the MSI doorbell GPA directly. kvm_arch_fixup_msi_route() uses this callback when available and ignores the guest provided IOVA in that case. If the vIOMMU does not implement the callback, we fall back to the existing IOMMU based address space translation path. This ensures correct MSI routing for accelerated SMMUv3 + VFIO passthrough while avoiding unsafe software walks of guest translation tables. As a related change, replace RCU_READ_LOCK_GUARD() with explicit rcu_read_lock()/rcu_read_unlock(). The introduction of an early goto (set_doorbell) path means the RCU read side critical section can no longer be safely scoped using RCU_READ_LOCK_GUARD(). Cc: Michael S. Tsirkin <mst@redhat.com> Reviewed-by: Nicolin Chen <nicolinc@nvidia.com> Reviewed-by: Eric Auger <eric.auger@redhat.com> Reviewed-by: Michael S. Tsirkin <mst@redhat.com> Tested-by: Eric Auger <eric.auger@redhat.com> Tested-by: Zhangfei Gao <zhangfei.gao@linaro.org> Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com> Signed-off-by: Shameer Kolothum <skolothumtho@nvidia.com> Message-id: 20260126104342.253965-17-skolothumtho@nvidia.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/pci/pci.c | 17 +++++++++++++++++ include/hw/pci/pci.h | 17 +++++++++++++++++ target/arm/kvm.c | 18 +++++++++++++++++- 3 files changed, 51 insertions(+), 1 deletion(-) diff --git a/hw/pci/pci.c b/hw/pci/pci.c index XXXXXXX..XXXXXXX 100644 --- a/hw/pci/pci.c +++ b/hw/pci/pci.c @@ -XXX,XX +XXX,XX @@ bool pci_device_get_iommu_bus_devfn(PCIDevice *dev, PCIBus **piommu_bus, return aliased; } +bool pci_device_iommu_msi_direct_gpa(PCIDevice *dev, hwaddr *out_doorbell) +{ + PCIBus *bus; + PCIBus *iommu_bus; + int devfn; + + pci_device_get_iommu_bus_devfn(dev, &iommu_bus, &bus, &devfn); + if (iommu_bus) { + if (iommu_bus->iommu_ops->get_msi_direct_gpa) { + *out_doorbell = iommu_bus->iommu_ops->get_msi_direct_gpa(bus, + iommu_bus->iommu_opaque, devfn); + return true; + } + } + return false; +} + AddressSpace *pci_device_iommu_address_space(PCIDevice *dev) { PCIBus *bus; diff --git a/include/hw/pci/pci.h b/include/hw/pci/pci.h index XXXXXXX..XXXXXXX 100644 --- a/include/hw/pci/pci.h +++ b/include/hw/pci/pci.h @@ -XXX,XX +XXX,XX @@ typedef struct PCIIOMMUOps { uint32_t pasid, bool priv_req, bool exec_req, hwaddr addr, bool lpig, uint16_t prgi, bool is_read, bool is_write); + /** + * @get_msi_direct_gpa: get the guest physical address of MSI doorbell + * for the device on a PCI bus. + * + * Optional callback. If implemented, it must return a valid guest + * physical address for the MSI doorbell + * + * @bus: the #PCIBus being accessed. + * + * @opaque: the data passed to pci_setup_iommu(). + * + * @devfn: device and function number + * + * Returns: the guest physical address of the MSI doorbell. + */ + uint64_t (*get_msi_direct_gpa)(PCIBus *bus, void *opaque, int devfn); } PCIIOMMUOps; bool pci_device_get_iommu_bus_devfn(PCIDevice *dev, PCIBus **piommu_bus, @@ -XXX,XX +XXX,XX @@ AddressSpace *pci_device_iommu_address_space(PCIDevice *dev); bool pci_device_set_iommu_device(PCIDevice *dev, HostIOMMUDevice *hiod, Error **errp); void pci_device_unset_iommu_device(PCIDevice *dev); +bool pci_device_iommu_msi_direct_gpa(PCIDevice *dev, hwaddr *out_doorbell); /** * pci_device_get_viommu_flags: get vIOMMU flags. diff --git a/target/arm/kvm.c b/target/arm/kvm.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/kvm.c +++ b/target/arm/kvm.c @@ -XXX,XX +XXX,XX @@ int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route, return 0; } + /* + * We do have an IOMMU address space, but for some vIOMMU implementations + * (e.g. accelerated SMMUv3) the translation tables are programmed into + * the physical SMMUv3 in the host (nested S1=guest, S2=host). QEMU cannot + * walk these tables in a safe way, so in that case we obtain the MSI + * doorbell GPA directly from the vIOMMU backend and ignore the gIOVA + * @address. + */ + if (pci_device_iommu_msi_direct_gpa(dev, &doorbell_gpa)) { + goto set_doorbell; + } + /* MSI doorbell address is translated by an IOMMU */ - RCU_READ_LOCK_GUARD(); + rcu_read_lock(); mr = address_space_translate(as, address, &xlat, &len, true, MEMTXATTRS_UNSPECIFIED); if (!mr) { + rcu_read_unlock(); return 1; } mrs = memory_region_find(mr, xlat, 1); if (!mrs.mr) { + rcu_read_unlock(); return 1; } doorbell_gpa = mrs.offset_within_address_space; memory_region_unref(mrs.mr); + rcu_read_unlock(); +set_doorbell: route->u.msi.address_lo = doorbell_gpa; route->u.msi.address_hi = doorbell_gpa >> 32; -- 2.43.0 From: Shameer Kolothum <skolothumtho@nvidia.com> Accelerated SMMUv3 instances rely on the physical SMMUv3 for nested translation (guest Stage-1, host Stage-2). In this mode, the guest Stage-1 tables are programmed directly into hardware, and QEMU must not attempt to walk them for translation, as doing so is not reliably safe. For vfio-pci endpoints behind such a vSMMU, the only translation QEMU needs to perform is for the MSI doorbell used during KVM MSI setup. Implement the callback so that kvm_arch_fixup_msi_route() can retrieve the MSI doorbell GPA directly, instead of attempting a software walk of the guest translation tables. Also introduce an SMMUv3 device property to carry the MSI doorbell GPA. This property will be set by the virt machine in a subsequent patch. Reviewed-by: Nicolin Chen <nicolinc@nvidia.com> Reviewed-by: Eric Auger <eric.auger@redhat.com> Tested-by: Eric Auger <eric.auger@redhat.com> Tested-by: Zhangfei Gao <zhangfei.gao@linaro.org> Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com> Signed-off-by: Shameer Kolothum <skolothumtho@nvidia.com> Message-id: 20260126104342.253965-18-skolothumtho@nvidia.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/arm/smmuv3-accel.c | 10 ++++++++++ hw/arm/smmuv3.c | 2 ++ include/hw/arm/smmuv3.h | 1 + 3 files changed, 13 insertions(+) diff --git a/hw/arm/smmuv3-accel.c b/hw/arm/smmuv3-accel.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/smmuv3-accel.c +++ b/hw/arm/smmuv3-accel.c @@ -XXX,XX +XXX,XX @@ static void smmuv3_accel_unset_iommu_device(PCIBus *bus, void *opaque, } } +static uint64_t smmuv3_accel_get_msi_gpa(PCIBus *bus, void *opaque, int devfn) +{ + SMMUState *bs = opaque; + SMMUv3State *s = ARM_SMMUV3(bs); + + g_assert(s->msi_gpa); + return s->msi_gpa; +} + /* * Only allow PCIe bridges, pxb-pcie roots, and GPEX roots so vfio-pci * endpoints can sit downstream. Accelerated SMMUv3 requires a vfio-pci @@ -XXX,XX +XXX,XX @@ static const PCIIOMMUOps smmuv3_accel_ops = { .get_viommu_flags = smmuv3_accel_get_viommu_flags, .set_iommu_device = smmuv3_accel_set_iommu_device, .unset_iommu_device = smmuv3_accel_unset_iommu_device, + .get_msi_direct_gpa = smmuv3_accel_get_msi_gpa, }; /* Based on SMUUv3 GPBA.ABORT configuration, attach a corresponding HWPT */ diff --git a/hw/arm/smmuv3.c b/hw/arm/smmuv3.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/smmuv3.c +++ b/hw/arm/smmuv3.c @@ -XXX,XX +XXX,XX @@ static const Property smmuv3_properties[] = { * Defaults to stage 1 */ DEFINE_PROP_STRING("stage", SMMUv3State, stage), + /* GPA of MSI doorbell, for SMMUv3 accel use. */ + DEFINE_PROP_UINT64("msi-gpa", SMMUv3State, msi_gpa, 0), }; static void smmuv3_instance_init(Object *obj) diff --git a/include/hw/arm/smmuv3.h b/include/hw/arm/smmuv3.h index XXXXXXX..XXXXXXX 100644 --- a/include/hw/arm/smmuv3.h +++ b/include/hw/arm/smmuv3.h @@ -XXX,XX +XXX,XX @@ struct SMMUv3State { /* SMMU has HW accelerator support for nested S1 + s2 */ bool accel; struct SMMUv3AccelState *s_accel; + uint64_t msi_gpa; }; typedef enum { -- 2.43.0 From: Shameer Kolothum <skolothumtho@nvidia.com> Set the MSI doorbell GPA property for accelerated SMMUv3 devices for use by KVM MSI setup. Also, since any meaningful use of vfio-pci devices with an accelerated SMMUv3 requires both KVM and a kernel irqchip, ensure those are specified when accel=on is selected. Reviewed-by: Nicolin Chen <nicolinc@nvidia.com> Tested-by: Eric Auger <eric.auger@redhat.com> Tested-by: Zhangfei Gao <zhangfei.gao@linaro.org> Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com> Signed-off-by: Shameer Kolothum <skolothumtho@nvidia.com> Message-id: 20260126104342.253965-19-skolothumtho@nvidia.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/arm/virt.c | 20 ++++++++++++++++++++ 1 file changed, 20 insertions(+) diff --git a/hw/arm/virt.c b/hw/arm/virt.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/virt.c +++ b/hw/arm/virt.c @@ -XXX,XX +XXX,XX @@ static void virt_machine_device_pre_plug_cb(HotplugHandler *hotplug_dev, object_property_set_link(OBJECT(dev), "secure-memory", OBJECT(vms->secure_sysmem), NULL); } + if (object_property_find(OBJECT(dev), "accel") && + object_property_get_bool(OBJECT(dev), "accel", &error_abort)) { + hwaddr db_start = 0; + + if (!kvm_enabled() || !kvm_irqchip_in_kernel()) { + error_setg(errp, "SMMUv3 accel=on requires KVM with " + "kernel-irqchip=on support"); + return; + } + + if (vms->msi_controller == VIRT_MSI_CTRL_ITS) { + /* GITS_TRANSLATER page + offset */ + db_start = base_memmap[VIRT_GIC_ITS].base + 0x10000 + 0x40; + } else if (vms->msi_controller == VIRT_MSI_CTRL_GICV2M) { + /* MSI_SETSPI_NS page + offset */ + db_start = base_memmap[VIRT_GIC_V2M].base + 0x40; + } + object_property_set_uint(OBJECT(dev), "msi-gpa", db_start, + &error_abort); + } } } -- 2.43.0 From: Shameer Kolothum <skolothumtho@nvidia.com> Provide a helper and use that to issue the invalidation cmd to host SMMUv3. We only issue one cmd at a time for now. Support for batching of commands will be added later after analysing the impact. Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com> Tested-by: Zhangfei Gao <zhangfei.gao@linaro.org> Reviewed-by: Nicolin Chen <nicolinc@nvidia.com> Reviewed-by: Eric Auger <eric.auger@redhat.com> Tested-by: Eric Auger <eric.auger@redhat.com> Signed-off-by: Shameer Kolothum <skolothumtho@nvidia.com> Message-id: 20260126104342.253965-20-skolothumtho@nvidia.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/arm/smmuv3-accel.c | 36 ++++++++++++++++++++++++++++++++++++ hw/arm/smmuv3-accel.h | 8 ++++++++ hw/arm/smmuv3.c | 16 ++++++++++++++++ 3 files changed, 60 insertions(+) diff --git a/hw/arm/smmuv3-accel.c b/hw/arm/smmuv3-accel.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/smmuv3-accel.c +++ b/hw/arm/smmuv3-accel.c @@ -XXX,XX +XXX,XX @@ bool smmuv3_accel_install_ste_range(SMMUv3State *s, SMMUSIDRange *range, return all_ok; } +/* + * This issues the invalidation cmd to the host SMMUv3. + * + * sdev is non-NULL for SID based invalidations (e.g. CFGI_CD), and NULL for + * non SID invalidations such as SMMU_CMD_TLBI_NH_ASID and SMMU_CMD_TLBI_NH_VA. + */ +bool smmuv3_accel_issue_inv_cmd(SMMUv3State *bs, void *cmd, SMMUDevice *sdev, + Error **errp) +{ + SMMUv3State *s = ARM_SMMUV3(bs); + SMMUv3AccelState *accel = s->s_accel; + uint32_t entry_num = 1; + + /* + * No accel or viommu means no VFIO/IOMMUFD devices, nothing to + * invalidate. + */ + if (!accel || !accel->viommu) { + return true; + } + + /* + * SID based invalidations (e.g. CFGI_CD) apply only to vfio-pci endpoints + * with a valid vIOMMU vdev. + */ + if (sdev && !container_of(sdev, SMMUv3AccelDevice, sdev)->vdev) { + return true; + } + + /* Single command (entry_num = 1); no need to check returned entry_num */ + return iommufd_backend_invalidate_cache( + accel->viommu->iommufd, accel->viommu->viommu_id, + IOMMU_VIOMMU_INVALIDATE_DATA_ARM_SMMUV3, + sizeof(Cmd), &entry_num, cmd, errp); +} + static bool smmuv3_accel_alloc_viommu(SMMUv3State *s, HostIOMMUDeviceIOMMUFD *idev, Error **errp) diff --git a/hw/arm/smmuv3-accel.h b/hw/arm/smmuv3-accel.h index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/smmuv3-accel.h +++ b/hw/arm/smmuv3-accel.h @@ -XXX,XX +XXX,XX @@ bool smmuv3_accel_install_ste(SMMUv3State *s, SMMUDevice *sdev, int sid, bool smmuv3_accel_install_ste_range(SMMUv3State *s, SMMUSIDRange *range, Error **errp); bool smmuv3_accel_attach_gbpa_hwpt(SMMUv3State *s, Error **errp); +bool smmuv3_accel_issue_inv_cmd(SMMUv3State *s, void *cmd, SMMUDevice *sdev, + Error **errp); void smmuv3_accel_reset(SMMUv3State *s); #else static inline void smmuv3_accel_init(SMMUv3State *s) @@ -XXX,XX +XXX,XX @@ static inline bool smmuv3_accel_attach_gbpa_hwpt(SMMUv3State *s, Error **errp) { return true; } +static inline bool +smmuv3_accel_issue_inv_cmd(SMMUv3State *s, void *cmd, SMMUDevice *sdev, + Error **errp) +{ + return true; +} static inline void smmuv3_accel_reset(SMMUv3State *s) { } diff --git a/hw/arm/smmuv3.c b/hw/arm/smmuv3.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/smmuv3.c +++ b/hw/arm/smmuv3.c @@ -XXX,XX +XXX,XX @@ static int smmuv3_cmdq_consume(SMMUv3State *s, Error **errp) trace_smmuv3_cmdq_cfgi_cd(sid); smmuv3_flush_config(sdev); + if (!smmuv3_accel_issue_inv_cmd(s, &cmd, sdev, errp)) { + cmd_error = SMMU_CERROR_ILL; + break; + } break; } case SMMU_CMD_TLBI_NH_ASID: @@ -XXX,XX +XXX,XX @@ static int smmuv3_cmdq_consume(SMMUv3State *s, Error **errp) trace_smmuv3_cmdq_tlbi_nh_asid(asid); smmu_inv_notifiers_all(&s->smmu_state); smmu_iotlb_inv_asid_vmid(bs, asid, vmid); + if (!smmuv3_accel_issue_inv_cmd(s, &cmd, NULL, errp)) { + cmd_error = SMMU_CERROR_ILL; + break; + } break; } case SMMU_CMD_TLBI_NH_ALL: @@ -XXX,XX +XXX,XX @@ static int smmuv3_cmdq_consume(SMMUv3State *s, Error **errp) trace_smmuv3_cmdq_tlbi_nsnh(); smmu_inv_notifiers_all(&s->smmu_state); smmu_iotlb_inv_all(bs); + if (!smmuv3_accel_issue_inv_cmd(s, &cmd, NULL, errp)) { + cmd_error = SMMU_CERROR_ILL; + break; + } break; case SMMU_CMD_TLBI_NH_VAA: case SMMU_CMD_TLBI_NH_VA: @@ -XXX,XX +XXX,XX @@ static int smmuv3_cmdq_consume(SMMUv3State *s, Error **errp) break; } smmuv3_range_inval(bs, &cmd, SMMU_STAGE_1); + if (!smmuv3_accel_issue_inv_cmd(s, &cmd, NULL, errp)) { + cmd_error = SMMU_CERROR_ILL; + break; + } break; case SMMU_CMD_TLBI_S12_VMALL: { -- 2.43.0 From: Shameer Kolothum <skolothumtho@nvidia.com> Factor out ID register init into smmuv3_init_id_regs() and call it from realize(). This ensures ID registers are initialized early for use in the accelerated SMMUv3 path and will be utilized in subsequent patch. Other registers remain initialized in smmuv3_reset(). Reviewed-by: Nicolin Chen <nicolinc@nvidia.com> Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com> Reviewed-by: Eric Auger <eric.auger@redhat.com> Tested-by: Eric Auger <eric.auger@redhat.com> Tested-by: Zhangfei Gao <zhangfei.gao@linaro.org> Signed-off-by: Shameer Kolothum <skolothumtho@nvidia.com> Message-id: 20260126104342.253965-21-skolothumtho@nvidia.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/arm/smmuv3.c | 15 ++++++++++++--- 1 file changed, 12 insertions(+), 3 deletions(-) diff --git a/hw/arm/smmuv3.c b/hw/arm/smmuv3.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/smmuv3.c +++ b/hw/arm/smmuv3.c @@ -XXX,XX +XXX,XX @@ void smmuv3_record_event(SMMUv3State *s, SMMUEventInfo *info) info->recorded = true; } -static void smmuv3_init_regs(SMMUv3State *s) +/* + * Called during realize(), as the ID registers will be accessed early in the + * SMMUv3 accel path for feature compatibility checks. The remaining registers + * are initialized later in smmuv3_reset(). + */ +static void smmuv3_init_id_regs(SMMUv3State *s) { /* Based on sys property, the stages supported in smmu will be advertised.*/ if (s->stage && !strcmp("2", s->stage)) { @@ -XXX,XX +XXX,XX @@ static void smmuv3_init_regs(SMMUv3State *s) s->idr[5] = FIELD_DP32(s->idr[5], IDR5, GRAN4K, 1); s->idr[5] = FIELD_DP32(s->idr[5], IDR5, GRAN16K, 1); s->idr[5] = FIELD_DP32(s->idr[5], IDR5, GRAN64K, 1); + s->aidr = 0x1; +} +static void smmuv3_reset(SMMUv3State *s) +{ s->cmdq.base = deposit64(s->cmdq.base, 0, 5, SMMU_CMDQS); s->cmdq.prod = 0; s->cmdq.cons = 0; @@ -XXX,XX +XXX,XX @@ static void smmuv3_init_regs(SMMUv3State *s) s->features = 0; s->sid_split = 0; - s->aidr = 0x1; s->cr[0] = 0; s->cr0ack = 0; s->irq_ctrl = 0; @@ -XXX,XX +XXX,XX @@ static void smmu_reset_exit(Object *obj, ResetType type) c->parent_phases.exit(obj, type); } - smmuv3_init_regs(s); + smmuv3_reset(s); smmuv3_accel_reset(s); } @@ -XXX,XX +XXX,XX @@ static void smmu_realize(DeviceState *d, Error **errp) sysbus_init_mmio(dev, &sys->iomem); smmu_init_irq(s, dev); + smmuv3_init_id_regs(s); } static const VMStateDescription vmstate_smmuv3_queue = { -- 2.43.0 From: Shameer Kolothum <skolothumtho@nvidia.com> Just before the device gets attached to the SMMUv3, make sure QEMU SMMUv3 features are compatible with the host SMMUv3. Not all fields in the host SMMUv3 IDR registers are meaningful for userspace. Only the following fields can be used: - IDR0: ST_LEVEL, TERM_MODEL, STALL_MODEL, TTENDIAN, CD2L, ASID16, TTF - IDR1: SIDSIZE, SSIDSIZE - IDR3: BBML, RIL - IDR5: VAX, GRAN64K, GRAN16K, GRAN4K For now, the check is to make sure the features are in sync to enable basic accelerated SMMUv3 support. AIDR is not checked, as hardware implementations often provide a mix of architecture features regardless of the revision reported in AIDR. Note that SSIDSIZE check will be added later when support for PASID is introduced. Reviewed-by: Nicolin Chen <nicolinc@nvidia.com> Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com> Reviewed-by: Eric Auger <eric.auger@redhat.com> Tested-by: Eric Auger <eric.auger@redhat.com> Tested-by: Zhangfei Gao <zhangfei.gao@linaro.org> Signed-off-by: Shameer Kolothum <skolothumtho@nvidia.com> Message-id: 20260126104342.253965-22-skolothumtho@nvidia.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/arm/smmuv3-accel.c | 101 ++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 101 insertions(+) diff --git a/hw/arm/smmuv3-accel.c b/hw/arm/smmuv3-accel.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/smmuv3-accel.c +++ b/hw/arm/smmuv3-accel.c @@ -XXX,XX +XXX,XX @@ static MemoryRegion root, sysmem; static AddressSpace *shared_as_sysmem; +static bool +smmuv3_accel_check_hw_compatible(SMMUv3State *s, + struct iommu_hw_info_arm_smmuv3 *info, + Error **errp) +{ + /* QEMU SMMUv3 supports both linear and 2-level stream tables */ + if (FIELD_EX32(info->idr[0], IDR0, STLEVEL) != + FIELD_EX32(s->idr[0], IDR0, STLEVEL)) { + error_setg(errp, "Host SMMUv3 Stream Table format mismatch " + "(host STLEVEL=%u, QEMU STLEVEL=%u)", + FIELD_EX32(info->idr[0], IDR0, STLEVEL), + FIELD_EX32(s->idr[0], IDR0, STLEVEL)); + return false; + } + + /* QEMU SMMUv3 supports only little-endian translation table walks */ + if (FIELD_EX32(info->idr[0], IDR0, TTENDIAN) > + FIELD_EX32(s->idr[0], IDR0, TTENDIAN)) { + error_setg(errp, "Host SMMUv3 doesn't support Little-endian " + "translation table"); + return false; + } + + /* QEMU SMMUv3 supports only AArch64 translation table format */ + if (FIELD_EX32(info->idr[0], IDR0, TTF) < + FIELD_EX32(s->idr[0], IDR0, TTF)) { + error_setg(errp, "Host SMMUv3 doesn't support AArch64 translation " + "table format"); + return false; + } + + /* QEMU SMMUv3 supports SIDSIZE 16 */ + if (FIELD_EX32(info->idr[1], IDR1, SIDSIZE) < + FIELD_EX32(s->idr[1], IDR1, SIDSIZE)) { + error_setg(errp, "Host SMMUv3 SIDSIZE not compatible " + "(host=%u, QEMU=%u)", + FIELD_EX32(info->idr[1], IDR1, SIDSIZE), + FIELD_EX32(s->idr[1], IDR1, SIDSIZE)); + return false; + } + + /* QEMU SMMUv3 supports Range Invalidation by default */ + if (FIELD_EX32(info->idr[3], IDR3, RIL) != + FIELD_EX32(s->idr[3], IDR3, RIL)) { + error_setg(errp, "Host SMMUv3 doesn't support Range Invalidation"); + return false; + } + + /* QEMU SMMUv3 supports GRAN4K/GRAN16K/GRAN64K translation granules */ + if (FIELD_EX32(info->idr[5], IDR5, GRAN4K) != + FIELD_EX32(s->idr[5], IDR5, GRAN4K)) { + error_setg(errp, "Host SMMUv3 doesn't support 4K translation granule"); + return false; + } + if (FIELD_EX32(info->idr[5], IDR5, GRAN16K) != + FIELD_EX32(s->idr[5], IDR5, GRAN16K)) { + error_setg(errp, "Host SMMUv3 doesn't support 16K translation granule"); + return false; + } + if (FIELD_EX32(info->idr[5], IDR5, GRAN64K) != + FIELD_EX32(s->idr[5], IDR5, GRAN64K)) { + error_setg(errp, "Host SMMUv3 doesn't support 64K translation granule"); + return false; + } + + return true; +} + +static bool +smmuv3_accel_hw_compatible(SMMUv3State *s, HostIOMMUDeviceIOMMUFD *idev, + Error **errp) +{ + struct iommu_hw_info_arm_smmuv3 info; + uint32_t data_type; + uint64_t caps; + + if (!iommufd_backend_get_device_info(idev->iommufd, idev->devid, &data_type, + &info, sizeof(info), &caps, errp)) { + return false; + } + + if (data_type != IOMMU_HW_INFO_TYPE_ARM_SMMUV3) { + error_setg(errp, "Wrong data type (%d) for Host SMMUv3 device info", + data_type); + return false; + } + + if (!smmuv3_accel_check_hw_compatible(s, &info, errp)) { + return false; + } + return true; +} + static SMMUv3AccelDevice *smmuv3_accel_get_dev(SMMUState *bs, SMMUPciBus *sbus, PCIBus *bus, int devfn) { @@ -XXX,XX +XXX,XX @@ static bool smmuv3_accel_set_iommu_device(PCIBus *bus, void *opaque, int devfn, return true; } + /* + * Check the host SMMUv3 associated with the dev is compatible with the + * QEMU SMMUv3 accel. + */ + if (!smmuv3_accel_hw_compatible(s, idev, errp)) { + return false; + } + if (s->s_accel->viommu) { goto done; } -- 2.43.0 From: Eric Auger <eric.auger@redhat.com> Add a 'preserve_config' field in struct GPEXConfig and, if set, generate the _DSM function #5 for preserving PCI boot configurations. This will be used for SMMUv3 accel=on support in subsequent patch. When SMMUv3 acceleration (accel=on) is enabled, QEMU exposes IORT Reserved Memory Region (RMR) nodes to support MSI doorbell translations. As per the Arm IORT specification, using IORT RMRs mandates the presence of _DSM function #5 so that the OS retains the firmware-assigned PCI configuration. Hence, this patch adds conditional support for generating _DSM #5. According to the ACPI Specification, Revision 6.6, Section 9.1.1 - “_DSM (Device Specific Method)”, " If Function Index is zero, the return is a buffer containing one bit for each function index, starting with zero. Bit 0 indicates whether there is support for any functions other than function 0 for the specified UUID and Revision ID. If set to zero, no functions are supported (other than function zero) for the specified UUID and Revision ID. If set to one, at least one additional function is supported. For all other bits in the buffer, a bit is set to zero to indicate if that function index is not supported for the specific UUID and Revision ID. (For example, bit 1 set to 0 indicates that function index 1 is not supported for the specific UUID and Revision ID.) " Please refer PCI Firmware Specification, Revision 3.3, Section 4.6.5 — "_DSM for Preserving PCI Boot Configurations" for Function 5 of _DSM method. Also, while at it, move the byte_list declaration to the top of the function for clarity. At the moment, DSM generation is not yet enabled. The resulting AML when preserve_config=true is: Method (_DSM, 4, NotSerialized) { If ((Arg0 == ToUUID ("e5c937d0-3553-4d7a-9117-ea4d19c3434d"))) { If ((Arg2 == Zero)) { Return (Buffer (One) { 0x21 }) } If ((Arg2 == 0x05)) { Return (Zero) } } ... } Cc: Michael S. Tsirkin <mst@redhat.com> Signed-off-by: Eric Auger <eric.auger@redhat.com> Signed-off-by: Shameer Kolothum <shameerali.kolothum.thodi@huawei.com> Signed-off-by: Shameer Kolothum <skolothumtho@nvidia.com> Tested-by: Zhangfei Gao <zhangfei.gao@linaro.org> Tested-by: Eric Auger <eric.auger@redhat.com> Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com> Reviewed-by: Michael S. Tsirkin <mst@redhat.com> Message-id: 20260126104342.253965-23-skolothumtho@nvidia.com [Shameer: Removed possible duplicate _DSM creations] Signed-off-by: Shameer Kolothum <shameerali.kolothum.thodi@huawei.com> Tested-by: Zhangfei Gao <zhangfei.gao@linaro.org> Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com> Reviewed-by: Michael S. Tsirkin <mst@redhat.com> Tested-by: Eric Auger <eric.auger@redhat.com> Signed-off-by: Shameer Kolothum <skolothumtho@nvidia.com> Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/pci-host/gpex-acpi.c | 29 +++++++++++++++++++++++------ include/hw/pci-host/gpex.h | 1 + 2 files changed, 24 insertions(+), 6 deletions(-) diff --git a/hw/pci-host/gpex-acpi.c b/hw/pci-host/gpex-acpi.c index XXXXXXX..XXXXXXX 100644 --- a/hw/pci-host/gpex-acpi.c +++ b/hw/pci-host/gpex-acpi.c @@ -XXX,XX +XXX,XX @@ static void acpi_dsdt_add_pci_route_table(Aml *dev, uint32_t irq, } } -static Aml *build_pci_host_bridge_dsm_method(void) +static Aml *build_pci_host_bridge_dsm_method(bool preserve_config) { Aml *method = aml_method("_DSM", 4, AML_NOTSERIALIZED); Aml *UUID, *ifctx, *ifctx1, *buf; + uint8_t byte_list[1] = {0}; /* PCI Firmware Specification 3.0 * 4.6.1. _DSM for PCI Express Slot Information @@ -XXX,XX +XXX,XX @@ static Aml *build_pci_host_bridge_dsm_method(void) UUID = aml_touuid("E5C937D0-3553-4D7A-9117-EA4D19C3434D"); ifctx = aml_if(aml_equal(aml_arg(0), UUID)); ifctx1 = aml_if(aml_equal(aml_arg(2), aml_int(0))); - uint8_t byte_list[1] = {0}; + if (preserve_config) { + /* support functions other than 0, specifically function 5 */ + byte_list[0] = 0x21; + } buf = aml_buffer(1, byte_list); aml_append(ifctx1, aml_return(buf)); aml_append(ifctx, ifctx1); + if (preserve_config) { + Aml *ifctx2 = aml_if(aml_equal(aml_arg(2), aml_int(5))); + /* + * 0 - The operating system must not ignore the PCI configuration that + * firmware has done at boot time. + */ + aml_append(ifctx2, aml_return(aml_int(0))); + aml_append(ifctx, ifctx2); + } + aml_append(method, ifctx); byte_list[0] = 0; @@ -XXX,XX +XXX,XX @@ static Aml *build_pci_host_bridge_dsm_method(void) } static void acpi_dsdt_add_host_bridge_methods(Aml *dev, - bool enable_native_pcie_hotplug) + bool enable_native_pcie_hotplug, + bool preserve_config) { /* Declare an _OSC (OS Control Handoff) method */ aml_append(dev, build_pci_host_bridge_osc_method(enable_native_pcie_hotplug)); - aml_append(dev, build_pci_host_bridge_dsm_method()); + aml_append(dev, build_pci_host_bridge_dsm_method(preserve_config)); } void acpi_dsdt_add_gpex(Aml *scope, struct GPEXConfig *cfg) @@ -XXX,XX +XXX,XX @@ void acpi_dsdt_add_gpex(Aml *scope, struct GPEXConfig *cfg) build_cxl_osc_method(dev); } else { /* pxb bridges do not have ACPI PCI Hot-plug enabled */ - acpi_dsdt_add_host_bridge_methods(dev, true); + acpi_dsdt_add_host_bridge_methods(dev, true, + cfg->preserve_config); } aml_append(scope, dev); @@ -XXX,XX +XXX,XX @@ void acpi_dsdt_add_gpex(Aml *scope, struct GPEXConfig *cfg) } aml_append(dev, aml_name_decl("_CRS", rbuf)); - acpi_dsdt_add_host_bridge_methods(dev, cfg->pci_native_hotplug); + acpi_dsdt_add_host_bridge_methods(dev, cfg->pci_native_hotplug, + cfg->preserve_config); Aml *dev_res0 = aml_device("%s", "RES0"); aml_append(dev_res0, aml_name_decl("_HID", aml_string("PNP0C02"))); diff --git a/include/hw/pci-host/gpex.h b/include/hw/pci-host/gpex.h index XXXXXXX..XXXXXXX 100644 --- a/include/hw/pci-host/gpex.h +++ b/include/hw/pci-host/gpex.h @@ -XXX,XX +XXX,XX @@ struct GPEXConfig { int irq; PCIBus *bus; bool pci_native_hotplug; + bool preserve_config; }; typedef struct GPEXIrq GPEXIrq; -- 2.43.0 From: Shameer Kolothum <skolothumtho@nvidia.com> Introduce a new pci_preserve_config field in virt machine state which allows the generation of DSM #5. This field is only set if accel SMMU is instantiated. In a subsequent patch, SMMUv3 accel mode will make use of IORT RMR nodes to enable nested translation of MSI doorbell addresses. IORT RMR requires _DSM #5 to be set for the PCI host bridge so that the Guest kernel preserves the PCI boot configuration. Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com> Tested-by: Zhangfei Gao <zhangfei.gao@linaro.org> Reviewed-by: Eric Auger <eric.auger@redhat.com> Tested-by: Eric Auger <eric.auger@redhat.com> Signed-off-by: Shameer Kolothum <skolothumtho@nvidia.com> Message-id: 20260126104342.253965-24-skolothumtho@nvidia.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/arm/virt-acpi-build.c | 8 ++++++++ hw/arm/virt.c | 1 + include/hw/arm/virt.h | 1 + 3 files changed, 10 insertions(+) diff --git a/hw/arm/virt-acpi-build.c b/hw/arm/virt-acpi-build.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/virt-acpi-build.c +++ b/hw/arm/virt-acpi-build.c @@ -XXX,XX +XXX,XX @@ static void acpi_dsdt_add_pci(Aml *scope, const MemMapEntry *memmap, .pci_native_hotplug = !acpi_pcihp, }; + /* + * Accel SMMU requires RMRs for MSI 1-1 mapping, which require _DSM + * function 5 (_DSM for Preserving PCI Boot Configurations). + */ + if (vms->pci_preserve_config) { + cfg.preserve_config = true; + } + if (vms->highmem_mmio) { cfg.mmio64 = memmap[VIRT_HIGH_PCIE_MMIO]; } diff --git a/hw/arm/virt.c b/hw/arm/virt.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/virt.c +++ b/hw/arm/virt.c @@ -XXX,XX +XXX,XX @@ static void virt_machine_device_pre_plug_cb(HotplugHandler *hotplug_dev, } object_property_set_uint(OBJECT(dev), "msi-gpa", db_start, &error_abort); + vms->pci_preserve_config = true; } } } diff --git a/include/hw/arm/virt.h b/include/hw/arm/virt.h index XXXXXXX..XXXXXXX 100644 --- a/include/hw/arm/virt.h +++ b/include/hw/arm/virt.h @@ -XXX,XX +XXX,XX @@ struct VirtMachineState { bool legacy_smmuv3_present; MemoryRegion *sysmem; MemoryRegion *secure_sysmem; + bool pci_preserve_config; }; #define VIRT_ECAM_ID(high) (high ? VIRT_HIGH_PCIE_ECAM : VIRT_PCIE_ECAM) -- 2.43.0 From: Shameer Kolothum <skolothumtho@nvidia.com> Subsequent patch will upgrade IORT revision to 5 to add support for IORT RMR nodes. Add the affected IORT blobs to allowed-diff list for bios-table tests. Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com> Reviewed-by: Eric Auger <eric.auger@redhat.com> Tested-by: Eric Auger <eric.auger@redhat.com> Tested-by: Zhangfei Gao <zhangfei.gao@linaro.org> Signed-off-by: Shameer Kolothum <skolothumtho@nvidia.com> Message-id: 20260126104342.253965-25-skolothumtho@nvidia.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- tests/qtest/bios-tables-test-allowed-diff.h | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/tests/qtest/bios-tables-test-allowed-diff.h b/tests/qtest/bios-tables-test-allowed-diff.h index XXXXXXX..XXXXXXX 100644 --- a/tests/qtest/bios-tables-test-allowed-diff.h +++ b/tests/qtest/bios-tables-test-allowed-diff.h @@ -1 +1,5 @@ /* List of comma-separated changed AML files to ignore */ +"tests/data/acpi/aarch64/virt/IORT", +"tests/data/acpi/aarch64/virt/IORT.its_off", +"tests/data/acpi/aarch64/virt/IORT.smmuv3-legacy", +"tests/data/acpi/aarch64/virt/IORT.smmuv3-dev", -- 2.43.0 From: Eric Auger <eric.auger@redhat.com> To handle SMMUv3 accel=on mode(which configures the host SMMUv3 in nested mode), it is practical to expose the guest with reserved memory regions (RMRs) covering the IOVAs used by the host kernel to map physical MSI doorbells. Those IOVAs belong to [0x8000000, 0x8100000] matching MSI_IOVA_BASE and MSI_IOVA_LENGTH definitions in kernel arm-smmu-v3 driver. This is the window used to allocate IOVAs matching physical MSI doorbells. With those RMRs, the guest is forced to use a flat mapping for this range. Hence the assigned device is programmed with one IOVA from this range. Stage 1, owned by the guest has a flat mapping for this IOVA. Stage2, owned by the VMM then enforces a mapping from this IOVA to the physical MSI doorbell. The creation of those RMR nodes is only relevant if nested stage SMMU is in use, along with VFIO. As VFIO devices can be hotplugged, all RMRs need to be created in advance. Initialise AcpiIortSMMUv3Dev structures to avoid using uninitialised state when building the IORT, as legacy and device SMMUv3 paths populate different fields now(e.g. accel). Signed-off-by: Eric Auger <eric.auger@redhat.com> Signed-off-by: Shameer Kolothum <shameerali.kolothum.thodi@huawei.com> Signed-off-by: Nicolin Chen <nicolinc@nvidia.com> Signed-off-by: Shameer Kolothum <skolothumtho@nvidia.com> Tested-by: Zhangfei Gao <zhangfei.gao@linaro.org> Tested-by: Eric Auger <eric.auger@redhat.com> Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com> Message-id: 20260126104342.253965-26-skolothumtho@nvidia.com Suggested-by: Jean-Philippe Brucker <jean-philippe@linaro.org> Signed-off-by: Nicolin Chen <nicolinc@nvidia.com> Signed-off-by: Shameer Kolothum <shameerali.kolothum.thodi@huawei.com> Tested-by: Zhangfei Gao <zhangfei.gao@linaro.org> Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com> Tested-by: Eric Auger <eric.auger@redhat.com> Signed-off-by: Shameer Kolothum <skolothumtho@nvidia.com> Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/arm/virt-acpi-build.c | 115 +++++++++++++++++++++++++++++++++++---- 1 file changed, 105 insertions(+), 10 deletions(-) diff --git a/hw/arm/virt-acpi-build.c b/hw/arm/virt-acpi-build.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/virt-acpi-build.c +++ b/hw/arm/virt-acpi-build.c @@ -XXX,XX +XXX,XX @@ static void acpi_dsdt_add_tpm(Aml *scope, VirtMachineState *vms) #define ROOT_COMPLEX_ENTRY_SIZE 36 #define IORT_NODE_OFFSET 48 +#define IORT_RMR_NUM_ID_MAPPINGS 1 +#define IORT_RMR_NUM_MEM_RANGE_DESC 1 +#define IORT_RMR_COMMON_HEADER_SIZE 28 +#define IORT_RMR_MEM_RANGE_DESC_SIZE 20 + +/* + * IORT RMR flags: + * Bit[0] = 0 Disallow remapping of reserved ranges + * Bit[1] = 0 Unprivileged access + * Bits[9:2] = 0x00 Device nGnRnE memory + */ +#define IORT_RMR_FLAGS 0 + +/* + * MSI doorbell IOVA window used by the host kernel SMMUv3 driver. + * Described in IORT RMR nodes to reserve the IOVA range where the host + * kernel maps physical MSI doorbells for devices. This ensures guests + * preserve a flat mapping for MSI doorbell in nested SMMUv3(accel=on) + * configurations. + */ +#define MSI_IOVA_BASE 0x8000000 +#define MSI_IOVA_LENGTH 0x100000 + /* * Append an ID mapping entry as described by "Table 4 ID mapping format" in * "IO Remapping Table System Software on ARM Platforms", Chapter 3. @@ -XXX,XX +XXX,XX @@ static void acpi_dsdt_add_tpm(Aml *scope, VirtMachineState *vms) * Note that @id_count gets internally subtracted by one, following the spec. */ static void build_iort_id_mapping(GArray *table_data, uint32_t input_base, - uint32_t id_count, uint32_t out_ref) + uint32_t id_count, uint32_t out_ref, + uint32_t flags) { build_append_int_noprefix(table_data, input_base, 4); /* Input base */ /* Number of IDs - The number of IDs in the range minus one */ @@ -XXX,XX +XXX,XX @@ static void build_iort_id_mapping(GArray *table_data, uint32_t input_base, build_append_int_noprefix(table_data, input_base, 4); /* Output base */ build_append_int_noprefix(table_data, out_ref, 4); /* Output Reference */ /* Flags */ - build_append_int_noprefix(table_data, 0 /* Single mapping (disabled) */, 4); + build_append_int_noprefix(table_data, flags, 4); } struct AcpiIortIdMapping { @@ -XXX,XX +XXX,XX @@ typedef struct AcpiIortSMMUv3Dev { GArray *rc_smmu_idmaps; /* Offset of the SMMUv3 IORT Node relative to the start of the IORT */ size_t offset; + bool accel; } AcpiIortSMMUv3Dev; /* @@ -XXX,XX +XXX,XX @@ typedef struct AcpiIortSMMUv3Dev { static int populate_smmuv3_legacy_dev(GArray *sdev_blob) { VirtMachineState *vms = VIRT_MACHINE(qdev_get_machine()); - AcpiIortSMMUv3Dev sdev; + AcpiIortSMMUv3Dev sdev = {0}; sdev.rc_smmu_idmaps = g_array_new(false, true, sizeof(AcpiIortIdMapping)); object_child_foreach_recursive(object_get_root(), iort_host_bridges, @@ -XXX,XX +XXX,XX @@ static int smmuv3_dev_idmap_compare(gconstpointer a, gconstpointer b) static int iort_smmuv3_devices(Object *obj, void *opaque) { VirtMachineState *vms = VIRT_MACHINE(qdev_get_machine()); + AcpiIortSMMUv3Dev sdev = {0}; GArray *sdev_blob = opaque; AcpiIortIdMapping idmap; PlatformBusDevice *pbus; - AcpiIortSMMUv3Dev sdev; int min_bus, max_bus; SysBusDevice *sbdev; PCIBus *bus; @@ -XXX,XX +XXX,XX @@ static int iort_smmuv3_devices(Object *obj, void *opaque) } bus = PCI_BUS(object_property_get_link(obj, "primary-bus", &error_abort)); + if (object_property_find(obj, "accel")) { + sdev.accel = object_property_get_bool(obj, "accel", &error_abort); + } pbus = PLATFORM_BUS_DEVICE(vms->platform_bus_dev); sbdev = SYS_BUS_DEVICE(obj); sdev.base = platform_bus_get_mmio_addr(pbus, sbdev, 0); @@ -XXX,XX +XXX,XX @@ static void create_rc_its_idmaps(GArray *its_idmaps, GArray *smmuv3_devs) } } +static void +build_iort_rmr_nodes(GArray *table_data, GArray *smmuv3_devices, uint32_t *id) +{ + AcpiIortSMMUv3Dev *sdev; + AcpiIortIdMapping *idmap; + int i; + + for (i = 0; i < smmuv3_devices->len; i++) { + uint16_t rmr_len; + int bdf; + + sdev = &g_array_index(smmuv3_devices, AcpiIortSMMUv3Dev, i); + if (!sdev->accel) { + continue; + } + + /* + * Spec reference:Arm IO Remapping Table(IORT), ARM DEN 0049E.d, + * Section 3.1.1.5 "Reserved Memory Range node" + */ + idmap = &g_array_index(sdev->rc_smmu_idmaps, AcpiIortIdMapping, 0); + bdf = idmap->input_base; + rmr_len = IORT_RMR_COMMON_HEADER_SIZE + + (IORT_RMR_NUM_ID_MAPPINGS * ID_MAPPING_ENTRY_SIZE) + + (IORT_RMR_NUM_MEM_RANGE_DESC * IORT_RMR_MEM_RANGE_DESC_SIZE); + + /* Table 18 Reserved Memory Range Node */ + build_append_int_noprefix(table_data, 6 /* RMR */, 1); /* Type */ + /* Length */ + build_append_int_noprefix(table_data, rmr_len, 2); + build_append_int_noprefix(table_data, 3, 1); /* Revision */ + build_append_int_noprefix(table_data, (*id)++, 4); /* Identifier */ + /* Number of ID mappings */ + build_append_int_noprefix(table_data, IORT_RMR_NUM_ID_MAPPINGS, 4); + /* Reference to ID Array */ + build_append_int_noprefix(table_data, IORT_RMR_COMMON_HEADER_SIZE, 4); + + /* RMR specific data */ + + /* Flags */ + build_append_int_noprefix(table_data, IORT_RMR_FLAGS, 4); + /* Number of Memory Range Descriptors */ + build_append_int_noprefix(table_data, IORT_RMR_NUM_MEM_RANGE_DESC, 4); + /* Reference to Memory Range Descriptors */ + build_append_int_noprefix(table_data, IORT_RMR_COMMON_HEADER_SIZE + + (IORT_RMR_NUM_ID_MAPPINGS * ID_MAPPING_ENTRY_SIZE), 4); + build_iort_id_mapping(table_data, bdf, idmap->id_count, sdev->offset, + 1); + + /* Table 19 Memory Range Descriptor */ + + /* Physical Range offset */ + build_append_int_noprefix(table_data, MSI_IOVA_BASE, 8); + /* Physical Range length */ + build_append_int_noprefix(table_data, MSI_IOVA_LENGTH, 8); + build_append_int_noprefix(table_data, 0, 4); /* Reserved */ + } +} + /* * Input Output Remapping Table (IORT) * Conforms to "IO Remapping Table System Software on ARM Platforms", - * Document number: ARM DEN 0049E.b, Feb 2021 + * Document number: ARM DEN 0049E.d, Feb 2022 */ static void build_iort(GArray *table_data, BIOSLinker *linker, VirtMachineState *vms) @@ -XXX,XX +XXX,XX @@ build_iort(GArray *table_data, BIOSLinker *linker, VirtMachineState *vms) GArray *smmuv3_devs = g_array_new(false, true, sizeof(AcpiIortSMMUv3Dev)); GArray *rc_its_idmaps = g_array_new(false, true, sizeof(AcpiIortIdMapping)); - AcpiTable table = { .sig = "IORT", .rev = 3, .oem_id = vms->oem_id, + AcpiTable table = { .sig = "IORT", .rev = 5, .oem_id = vms->oem_id, .oem_table_id = vms->oem_table_id }; /* Table 2 The IORT */ acpi_table_begin(&table, table_data); @@ -XXX,XX +XXX,XX @@ build_iort(GArray *table_data, BIOSLinker *linker, VirtMachineState *vms) nb_nodes++; /* ITS */ rc_mapping_count += rc_its_idmaps->len; } + /* Calculate RMR nodes required. One per SMMUv3 with accelerated mode */ + for (i = 0; i < num_smmus; i++) { + sdev = &g_array_index(smmuv3_devs, AcpiIortSMMUv3Dev, i); + if (sdev->accel) { + nb_nodes++; + } + } } else { if (vms->its) { nb_nodes = 2; /* RC and ITS */ @@ -XXX,XX +XXX,XX @@ build_iort(GArray *table_data, BIOSLinker *linker, VirtMachineState *vms) /* Array of ID mappings */ if (smmu_mapping_count) { /* Output IORT node is the ITS Group node (the first node). */ - build_iort_id_mapping(table_data, 0, 0x10000, IORT_NODE_OFFSET); + build_iort_id_mapping(table_data, 0, 0x10000, IORT_NODE_OFFSET, 0); } } @@ -XXX,XX +XXX,XX @@ build_iort(GArray *table_data, BIOSLinker *linker, VirtMachineState *vms) AcpiIortIdMapping, j); /* Output IORT node is the SMMUv3 node. */ build_iort_id_mapping(table_data, range->input_base, - range->id_count, sdev->offset); + range->id_count, sdev->offset, 0); } } @@ -XXX,XX +XXX,XX @@ build_iort(GArray *table_data, BIOSLinker *linker, VirtMachineState *vms) range = &g_array_index(rc_its_idmaps, AcpiIortIdMapping, i); /* Output IORT node is the ITS Group node (the first node). */ build_iort_id_mapping(table_data, range->input_base, - range->id_count, IORT_NODE_OFFSET); + range->id_count, IORT_NODE_OFFSET, 0); } } } else { @@ -XXX,XX +XXX,XX @@ build_iort(GArray *table_data, BIOSLinker *linker, VirtMachineState *vms) * SMMU: RC -> ITS. * Output IORT node is the ITS Group node (the first node). */ - build_iort_id_mapping(table_data, 0, 0x10000, IORT_NODE_OFFSET); + build_iort_id_mapping(table_data, 0, 0x10000, IORT_NODE_OFFSET, 0); } + build_iort_rmr_nodes(table_data, smmuv3_devs, &id); acpi_table_end(linker, &table); g_array_free(rc_its_idmaps, true); for (i = 0; i < num_smmus; i++) { -- 2.43.0 From: Shameer Kolothum <skolothumtho@nvidia.com> Update the reference IORT blobs after revision upgrade for RMR node support. This affects the aarch64 'virt' IORT tests. IORT diff is the same for all the tests: /* * Intel ACPI Component Architecture * AML/ASL+ Disassembler version 20230628 (64-bit version) * Copyright (c) 2000 - 2023 Intel Corporation * - * Disassembly of tests/data/acpi/aarch64/virt/IORT, Mon Oct 20 14:42:41 2025 + * Disassembly of /tmp/aml-B4ZRE3, Mon Oct 20 14:42:41 2025 * * ACPI Data Table [IORT] * * Format: [HexOffset DecimalOffset ByteLength] FieldName : FieldValue (in hex) */ [000h 0000 004h] Signature : "IORT" [IO Remapping Table] [004h 0004 004h] Table Length : 00000080 -[008h 0008 001h] Revision : 03 -[009h 0009 001h] Checksum : B3 +[008h 0008 001h] Revision : 05 +[009h 0009 001h] Checksum : B1 [00Ah 0010 006h] Oem ID : "BOCHS " [010h 0016 008h] Oem Table ID : "BXPC " [018h 0024 004h] Oem Revision : 00000001 [01Ch 0028 004h] Asl Compiler ID : "BXPC" [020h 0032 004h] Asl Compiler Revision : 00000001 ... Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com> Reviewed-by: Eric Auger <eric.auger@redhat.com> Tested-by: Eric Auger <eric.auger@redhat.com> Tested-by: Zhangfei Gao <zhangfei.gao@linaro.org> Signed-off-by: Shameer Kolothum <skolothumtho@nvidia.com> Message-id: 20260126104342.253965-27-skolothumtho@nvidia.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- tests/data/acpi/aarch64/virt/IORT | Bin 128 -> 128 bytes tests/data/acpi/aarch64/virt/IORT.its_off | Bin 172 -> 172 bytes tests/data/acpi/aarch64/virt/IORT.smmuv3-dev | Bin 364 -> 364 bytes tests/data/acpi/aarch64/virt/IORT.smmuv3-legacy | Bin 276 -> 276 bytes tests/qtest/bios-tables-test-allowed-diff.h | 4 ---- 5 files changed, 4 deletions(-) diff --git a/tests/data/acpi/aarch64/virt/IORT b/tests/data/acpi/aarch64/virt/IORT index XXXXXXX..XXXXXXX 100644 GIT binary patch delta 18 ZcmZo*Y+&T_^bZPYU|?Wi-8hk}3;-#Q1d;#% delta 18 ZcmZo*Y+&T_^bZPYU|?Wi-aL`33;-#O1d;#% diff --git a/tests/data/acpi/aarch64/virt/IORT.its_off b/tests/data/acpi/aarch64/virt/IORT.its_off index XXXXXXX..XXXXXXX 100644 GIT binary patch delta 18 ZcmZ3(xQ3C-(?2L=4FdxM>(q%{ivTdM1ttIh delta 18 ZcmZ3(xQ3C-(?2L=4FdxM^Yn>aivTdK1ttIh diff --git a/tests/data/acpi/aarch64/virt/IORT.smmuv3-dev b/tests/data/acpi/aarch64/virt/IORT.smmuv3-dev index XXXXXXX..XXXXXXX 100644 GIT binary patch delta 19 acmaFE^oEJc(?2LAhmnDS^~6T5Bt`%|fCYU3 delta 19 acmaFE^oEJc(?2LAhmnDS`P4?PBt`%|eg%C1 diff --git a/tests/data/acpi/aarch64/virt/IORT.smmuv3-legacy b/tests/data/acpi/aarch64/virt/IORT.smmuv3-legacy index XXXXXXX..XXXXXXX 100644 GIT binary patch delta 19 acmbQjG=+)F(?2Lggpq-P)oUXc7b5^FiUXej delta 19 acmbQjG=+)F(?2Lggpq-P*=Hjc7b5^Fhy$Mh diff --git a/tests/qtest/bios-tables-test-allowed-diff.h b/tests/qtest/bios-tables-test-allowed-diff.h index XXXXXXX..XXXXXXX 100644 --- a/tests/qtest/bios-tables-test-allowed-diff.h +++ b/tests/qtest/bios-tables-test-allowed-diff.h @@ -1,5 +1 @@ /* List of comma-separated changed AML files to ignore */ -"tests/data/acpi/aarch64/virt/IORT", -"tests/data/acpi/aarch64/virt/IORT.its_off", -"tests/data/acpi/aarch64/virt/IORT.smmuv3-legacy", -"tests/data/acpi/aarch64/virt/IORT.smmuv3-dev", -- 2.43.0 From: Shameer Kolothum <skolothumtho@nvidia.com> Live migration is not supported when the SMMUv3 accelerator mode is enabled. Add a migration blocker to prevent migration in this configuration. Reviewed-by: Nicolin Chen <nicolinc@nvidia.com> Reviewed-by: Eric Auger <eric.auger@redhat.com> Tested-by: Zhangfei Gao <zhangfei.gao@linaro.org> Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com> Tested-by: Eric Auger <eric.auger@redhat.com> Signed-off-by: Shameer Kolothum <skolothumtho@nvidia.com> Message-id: 20260126104342.253965-28-skolothumtho@nvidia.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/arm/smmuv3.c | 6 ++++++ include/hw/arm/smmuv3.h | 1 + 2 files changed, 7 insertions(+) diff --git a/hw/arm/smmuv3.c b/hw/arm/smmuv3.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/smmuv3.c +++ b/hw/arm/smmuv3.c @@ -XXX,XX +XXX,XX @@ #include "qemu/bitops.h" #include "hw/core/irq.h" #include "hw/core/sysbus.h" +#include "migration/blocker.h" #include "migration/vmstate.h" #include "hw/core/qdev-properties.h" #include "hw/core/qdev.h" @@ -XXX,XX +XXX,XX @@ static void smmu_realize(DeviceState *d, Error **errp) if (s->accel) { smmuv3_accel_init(s); + error_setg(&s->migration_blocker, "Migration not supported with SMMUv3 " + "accelerator mode enabled"); + if (migrate_add_blocker(&s->migration_blocker, errp) < 0) { + return; + } } c->parent_realize(d, &local_err); diff --git a/include/hw/arm/smmuv3.h b/include/hw/arm/smmuv3.h index XXXXXXX..XXXXXXX 100644 --- a/include/hw/arm/smmuv3.h +++ b/include/hw/arm/smmuv3.h @@ -XXX,XX +XXX,XX @@ struct SMMUv3State { bool accel; struct SMMUv3AccelState *s_accel; uint64_t msi_gpa; + Error *migration_blocker; }; typedef enum { -- 2.43.0 From: Shameer Kolothum <skolothumtho@nvidia.com> Add an "accel" property to enable SMMUv3 accelerator mode. Accelerator mode relies on IORT RMR entries for MSI support and is therefore not supported when booting with a device tree. In this mode, the host SMMUv3 operates in nested translation (Stage-1 + Stage-2), with the guest owning the Stage-1 page tables. Expose only Stage-1 to the guest to ensure it uses the correct page table format Reviewed-by: Nicolin Chen <nicolinc@nvidia.com> Reviewed-by: Eric Auger <eric.auger@redhat.com> Tested-by: Zhangfei Gao <zhangfei.gao@linaro.org> Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com> Tested-by: Eric Auger <eric.auger@redhat.com> Signed-off-by: Shameer Kolothum <skolothumtho@nvidia.com> Message-id: 20260126104342.253965-29-skolothumtho@nvidia.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/arm/smmuv3.c | 32 ++++++++++++++++++++++++++++++++ hw/arm/virt-acpi-build.c | 4 +--- hw/arm/virt.c | 22 +++++++++++++--------- 3 files changed, 46 insertions(+), 12 deletions(-) diff --git a/hw/arm/smmuv3.c b/hw/arm/smmuv3.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/smmuv3.c +++ b/hw/arm/smmuv3.c @@ -XXX,XX +XXX,XX @@ static void smmu_reset_exit(Object *obj, ResetType type) smmuv3_accel_reset(s); } +static bool smmu_validate_property(SMMUv3State *s, Error **errp) +{ +#ifndef CONFIG_ARM_SMMUV3_ACCEL + if (s->accel) { + error_setg(errp, "accel=on support not compiled in"); + return false; + } +#endif + + if (!s->accel) { + return true; + } + + /* If no stage specified, SMMUv3 defaults to stage 1 */ + if (s->stage && strcmp(s->stage, "1")) { + error_setg(errp, + "Only stage1 is supported for SMMUv3 with accel=on"); + return false; + } + + return true; +} + static void smmu_realize(DeviceState *d, Error **errp) { SMMUState *sys = ARM_SMMU(d); @@ -XXX,XX +XXX,XX @@ static void smmu_realize(DeviceState *d, Error **errp) SysBusDevice *dev = SYS_BUS_DEVICE(d); Error *local_err = NULL; + if (!smmu_validate_property(s, errp)) { + return; + } + if (s->accel) { smmuv3_accel_init(s); error_setg(&s->migration_blocker, "Migration not supported with SMMUv3 " @@ -XXX,XX +XXX,XX @@ static const Property smmuv3_properties[] = { * Defaults to stage 1 */ DEFINE_PROP_STRING("stage", SMMUv3State, stage), + DEFINE_PROP_BOOL("accel", SMMUv3State, accel, false), /* GPA of MSI doorbell, for SMMUv3 accel use. */ DEFINE_PROP_UINT64("msi-gpa", SMMUv3State, msi_gpa, 0), }; @@ -XXX,XX +XXX,XX @@ static void smmuv3_class_init(ObjectClass *klass, const void *data) device_class_set_props(dc, smmuv3_properties); dc->hotpluggable = false; dc->user_creatable = true; + + object_class_property_set_description(klass, "accel", + "Enable SMMUv3 accelerator support. Allows host SMMUv3 to be " + "configured in nested mode for vfio-pci dev assignment"); } static int smmuv3_notify_flag_changed(IOMMUMemoryRegion *iommu, diff --git a/hw/arm/virt-acpi-build.c b/hw/arm/virt-acpi-build.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/virt-acpi-build.c +++ b/hw/arm/virt-acpi-build.c @@ -XXX,XX +XXX,XX @@ static int iort_smmuv3_devices(Object *obj, void *opaque) } bus = PCI_BUS(object_property_get_link(obj, "primary-bus", &error_abort)); - if (object_property_find(obj, "accel")) { - sdev.accel = object_property_get_bool(obj, "accel", &error_abort); - } + sdev.accel = object_property_get_bool(obj, "accel", &error_abort); pbus = PLATFORM_BUS_DEVICE(vms->platform_bus_dev); sbdev = SYS_BUS_DEVICE(obj); sdev.base = platform_bus_get_mmio_addr(pbus, sbdev, 0); diff --git a/hw/arm/virt.c b/hw/arm/virt.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/virt.c +++ b/hw/arm/virt.c @@ -XXX,XX +XXX,XX @@ static void create_smmuv3_dt_bindings(const VirtMachineState *vms, hwaddr base, g_free(node); } -static void create_smmuv3_dev_dtb(VirtMachineState *vms, - DeviceState *dev, PCIBus *bus) +static void create_smmuv3_dev_dtb(VirtMachineState *vms, DeviceState *dev, + PCIBus *bus, Error **errp) { PlatformBusDevice *pbus = PLATFORM_BUS_DEVICE(vms->platform_bus_dev); SysBusDevice *sbdev = SYS_BUS_DEVICE(dev); @@ -XXX,XX +XXX,XX @@ static void create_smmuv3_dev_dtb(VirtMachineState *vms, hwaddr base = platform_bus_get_mmio_addr(pbus, sbdev, 0); MachineState *ms = MACHINE(vms); - if (!(vms->bootinfo.firmware_loaded && virt_is_acpi_enabled(vms)) && - strcmp("pcie.0", bus->qbus.name)) { - warn_report("SMMUv3 device only supported with pcie.0 for DT"); - return; + if (!(vms->bootinfo.firmware_loaded && virt_is_acpi_enabled(vms))) { + if (object_property_get_bool(OBJECT(dev), "accel", &error_abort)) { + error_setg(errp, "SMMUv3 with accel=on not supported for DT"); + return; + } + if (strcmp("pcie.0", bus->qbus.name)) { + warn_report("SMMUv3 device only supported with pcie.0 for DT"); + return; + } } base += vms->memmap[VIRT_PLATFORM_BUS].base; irq += vms->irqmap[VIRT_PLATFORM_BUS]; @@ -XXX,XX +XXX,XX @@ static void virt_machine_device_pre_plug_cb(HotplugHandler *hotplug_dev, object_property_set_link(OBJECT(dev), "secure-memory", OBJECT(vms->secure_sysmem), NULL); } - if (object_property_find(OBJECT(dev), "accel") && - object_property_get_bool(OBJECT(dev), "accel", &error_abort)) { + if (object_property_get_bool(OBJECT(dev), "accel", &error_abort)) { hwaddr db_start = 0; if (!kvm_enabled() || !kvm_irqchip_in_kernel()) { @@ -XXX,XX +XXX,XX @@ static void virt_machine_device_plug_cb(HotplugHandler *hotplug_dev, return; } - create_smmuv3_dev_dtb(vms, dev, bus); + create_smmuv3_dev_dtb(vms, dev, bus, errp); } } -- 2.43.0 From: Shameer Kolothum <skolothumtho@nvidia.com> Currently QEMU SMMUv3 has RIL support by default. But if accelerated mode is enabled, RIL has to be compatible with host SMMUv3 support. Add a property so that the user can specify this. Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com> Tested-by: Zhangfei Gao <zhangfei.gao@linaro.org> Reviewed-by: Eric Auger <eric.auger@redhat.com> Reviewed-by: Nicolin Chen <nicolinc@nvidia.com> Tested-by: Eric Auger <eric.auger@redhat.com> Signed-off-by: Shameer Kolothum <skolothumtho@nvidia.com> Message-id: 20260126104342.253965-30-skolothumtho@nvidia.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/arm/smmuv3-accel.c | 14 ++++++++++++-- hw/arm/smmuv3-accel.h | 4 ++++ hw/arm/smmuv3.c | 9 +++++++++ include/hw/arm/smmuv3.h | 1 + 4 files changed, 26 insertions(+), 2 deletions(-) diff --git a/hw/arm/smmuv3-accel.c b/hw/arm/smmuv3-accel.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/smmuv3-accel.c +++ b/hw/arm/smmuv3-accel.c @@ -XXX,XX +XXX,XX @@ smmuv3_accel_check_hw_compatible(SMMUv3State *s, return false; } - /* QEMU SMMUv3 supports Range Invalidation by default */ - if (FIELD_EX32(info->idr[3], IDR3, RIL) != + /* User can disable QEMU SMMUv3 Range Invalidation support */ + if (FIELD_EX32(info->idr[3], IDR3, RIL) < FIELD_EX32(s->idr[3], IDR3, RIL)) { error_setg(errp, "Host SMMUv3 doesn't support Range Invalidation"); return false; @@ -XXX,XX +XXX,XX @@ static const PCIIOMMUOps smmuv3_accel_ops = { .get_msi_direct_gpa = smmuv3_accel_get_msi_gpa, }; +void smmuv3_accel_idr_override(SMMUv3State *s) +{ + if (!s->accel) { + return; + } + + /* By default QEMU SMMUv3 has RIL. Update IDR3 if user has disabled it */ + s->idr[3] = FIELD_DP32(s->idr[3], IDR3, RIL, s->ril); +} + /* Based on SMUUv3 GPBA.ABORT configuration, attach a corresponding HWPT */ bool smmuv3_accel_attach_gbpa_hwpt(SMMUv3State *s, Error **errp) { diff --git a/hw/arm/smmuv3-accel.h b/hw/arm/smmuv3-accel.h index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/smmuv3-accel.h +++ b/hw/arm/smmuv3-accel.h @@ -XXX,XX +XXX,XX @@ bool smmuv3_accel_install_ste_range(SMMUv3State *s, SMMUSIDRange *range, bool smmuv3_accel_attach_gbpa_hwpt(SMMUv3State *s, Error **errp); bool smmuv3_accel_issue_inv_cmd(SMMUv3State *s, void *cmd, SMMUDevice *sdev, Error **errp); +void smmuv3_accel_idr_override(SMMUv3State *s); void smmuv3_accel_reset(SMMUv3State *s); #else static inline void smmuv3_accel_init(SMMUv3State *s) @@ -XXX,XX +XXX,XX @@ smmuv3_accel_issue_inv_cmd(SMMUv3State *s, void *cmd, SMMUDevice *sdev, { return true; } +static inline void smmuv3_accel_idr_override(SMMUv3State *s) +{ +} static inline void smmuv3_accel_reset(SMMUv3State *s) { } diff --git a/hw/arm/smmuv3.c b/hw/arm/smmuv3.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/smmuv3.c +++ b/hw/arm/smmuv3.c @@ -XXX,XX +XXX,XX @@ static void smmuv3_init_id_regs(SMMUv3State *s) s->idr[5] = FIELD_DP32(s->idr[5], IDR5, GRAN16K, 1); s->idr[5] = FIELD_DP32(s->idr[5], IDR5, GRAN64K, 1); s->aidr = 0x1; + smmuv3_accel_idr_override(s); } static void smmuv3_reset(SMMUv3State *s) @@ -XXX,XX +XXX,XX @@ static bool smmu_validate_property(SMMUv3State *s, Error **errp) #endif if (!s->accel) { + if (!s->ril) { + error_setg(errp, "ril can only be disabled if accel=on"); + return false; + } return true; } @@ -XXX,XX +XXX,XX @@ static const Property smmuv3_properties[] = { DEFINE_PROP_BOOL("accel", SMMUv3State, accel, false), /* GPA of MSI doorbell, for SMMUv3 accel use. */ DEFINE_PROP_UINT64("msi-gpa", SMMUv3State, msi_gpa, 0), + /* RIL can be turned off for accel cases */ + DEFINE_PROP_BOOL("ril", SMMUv3State, ril, true), }; static void smmuv3_instance_init(Object *obj) @@ -XXX,XX +XXX,XX @@ static void smmuv3_class_init(ObjectClass *klass, const void *data) object_class_property_set_description(klass, "accel", "Enable SMMUv3 accelerator support. Allows host SMMUv3 to be " "configured in nested mode for vfio-pci dev assignment"); + object_class_property_set_description(klass, "ril", + "Disable range invalidation support (for accel=on)"); } static int smmuv3_notify_flag_changed(IOMMUMemoryRegion *iommu, diff --git a/include/hw/arm/smmuv3.h b/include/hw/arm/smmuv3.h index XXXXXXX..XXXXXXX 100644 --- a/include/hw/arm/smmuv3.h +++ b/include/hw/arm/smmuv3.h @@ -XXX,XX +XXX,XX @@ struct SMMUv3State { struct SMMUv3AccelState *s_accel; uint64_t msi_gpa; Error *migration_blocker; + bool ril; }; typedef enum { -- 2.43.0 From: Shameer Kolothum <skolothumtho@nvidia.com> QEMU SMMUv3 does not enable ATS (Address Translation Services) by default. When accelerated mode is enabled and the host SMMUv3 supports ATS, it can be useful to report ATS capability to the guest so it can take advantage of it if the device also supports ATS. Note: ATS support cannot be reliably detected from the host SMMUv3 IDR registers alone, as firmware ACPI IORT tables may override them. The user must therefore ensure the support before enabling it. The ATS support enabled here is only relevant for vfio-pci endpoints, as SMMUv3 accelerated mode does not support emulated endpoint devices. QEMU’s SMMUv3 implementation still lacks support for handling ATS translation requests, which would be required for emulated endpoints. Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com> Tested-by: Zhangfei Gao <zhangfei.gao@linaro.org> Reviewed-by: Nicolin Chen <nicolinc@nvidia.com> Reviewed-by: Eric Auger <eric.auger@redhat.com> Tested-by: Eric Auger <eric.auger@redhat.com> Signed-off-by: Shameer Kolothum <skolothumtho@nvidia.com> Message-id: 20260126104342.253965-31-skolothumtho@nvidia.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/arm/smmuv3-accel.c | 3 +++ hw/arm/smmuv3.c | 24 +++++++++++++++++++++++- hw/arm/virt-acpi-build.c | 10 ++++++++-- include/hw/arm/smmuv3.h | 1 + 4 files changed, 35 insertions(+), 3 deletions(-) diff --git a/hw/arm/smmuv3-accel.c b/hw/arm/smmuv3-accel.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/smmuv3-accel.c +++ b/hw/arm/smmuv3-accel.c @@ -XXX,XX +XXX,XX @@ void smmuv3_accel_idr_override(SMMUv3State *s) /* By default QEMU SMMUv3 has RIL. Update IDR3 if user has disabled it */ s->idr[3] = FIELD_DP32(s->idr[3], IDR3, RIL, s->ril); + + /* QEMU SMMUv3 has no ATS. Advertise ATS if opt-in by property */ + s->idr[0] = FIELD_DP32(s->idr[0], IDR0, ATS, s->ats); } /* Based on SMUUv3 GPBA.ABORT configuration, attach a corresponding HWPT */ diff --git a/hw/arm/smmuv3.c b/hw/arm/smmuv3.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/smmuv3.c +++ b/hw/arm/smmuv3.c @@ -XXX,XX +XXX,XX @@ static int smmuv3_cmdq_consume(SMMUv3State *s, Error **errp) */ smmuv3_range_inval(bs, &cmd, SMMU_STAGE_2); break; + case SMMU_CMD_ATC_INV: + { + SMMUDevice *sdev = smmu_find_sdev(bs, CMD_SID(&cmd)); + + if (!sdev || !s->ats) { + trace_smmuv3_unhandled_cmd(type); + break; + } + + if (!smmuv3_accel_issue_inv_cmd(s, &cmd, sdev, errp)) { + cmd_error = SMMU_CERROR_ILL; + break; + } + break; + } case SMMU_CMD_TLBI_EL3_ALL: case SMMU_CMD_TLBI_EL3_VA: case SMMU_CMD_TLBI_EL2_ALL: case SMMU_CMD_TLBI_EL2_ASID: case SMMU_CMD_TLBI_EL2_VA: case SMMU_CMD_TLBI_EL2_VAA: - case SMMU_CMD_ATC_INV: case SMMU_CMD_PRI_RESP: case SMMU_CMD_RESUME: case SMMU_CMD_STALL_TERM: @@ -XXX,XX +XXX,XX @@ static bool smmu_validate_property(SMMUv3State *s, Error **errp) error_setg(errp, "ril can only be disabled if accel=on"); return false; } + if (s->ats) { + error_setg(errp, "ats can only be enabled if accel=on"); + return false; + } return true; } @@ -XXX,XX +XXX,XX @@ static const Property smmuv3_properties[] = { DEFINE_PROP_UINT64("msi-gpa", SMMUv3State, msi_gpa, 0), /* RIL can be turned off for accel cases */ DEFINE_PROP_BOOL("ril", SMMUv3State, ril, true), + DEFINE_PROP_BOOL("ats", SMMUv3State, ats, false), }; static void smmuv3_instance_init(Object *obj) @@ -XXX,XX +XXX,XX @@ static void smmuv3_class_init(ObjectClass *klass, const void *data) "configured in nested mode for vfio-pci dev assignment"); object_class_property_set_description(klass, "ril", "Disable range invalidation support (for accel=on)"); + object_class_property_set_description(klass, "ats", + "Enable/disable ATS support (for accel=on). Please ensure host " + "platform has ATS support before enabling this"); } static int smmuv3_notify_flag_changed(IOMMUMemoryRegion *iommu, diff --git a/hw/arm/virt-acpi-build.c b/hw/arm/virt-acpi-build.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/virt-acpi-build.c +++ b/hw/arm/virt-acpi-build.c @@ -XXX,XX +XXX,XX @@ typedef struct AcpiIortSMMUv3Dev { /* Offset of the SMMUv3 IORT Node relative to the start of the IORT */ size_t offset; bool accel; + bool ats; } AcpiIortSMMUv3Dev; /* @@ -XXX,XX +XXX,XX @@ static int iort_smmuv3_devices(Object *obj, void *opaque) bus = PCI_BUS(object_property_get_link(obj, "primary-bus", &error_abort)); sdev.accel = object_property_get_bool(obj, "accel", &error_abort); + sdev.ats = object_property_get_bool(obj, "ats", &error_abort); pbus = PLATFORM_BUS_DEVICE(vms->platform_bus_dev); sbdev = SYS_BUS_DEVICE(obj); sdev.base = platform_bus_get_mmio_addr(pbus, sbdev, 0); @@ -XXX,XX +XXX,XX @@ build_iort(GArray *table_data, BIOSLinker *linker, VirtMachineState *vms) int i, nb_nodes, rc_mapping_count; AcpiIortSMMUv3Dev *sdev; size_t node_size; + bool ats_needed = false; int num_smmus = 0; uint32_t id = 0; int rc_smmu_idmaps_len = 0; @@ -XXX,XX +XXX,XX @@ build_iort(GArray *table_data, BIOSLinker *linker, VirtMachineState *vms) /* Calculate RMR nodes required. One per SMMUv3 with accelerated mode */ for (i = 0; i < num_smmus; i++) { sdev = &g_array_index(smmuv3_devs, AcpiIortSMMUv3Dev, i); + if (sdev->ats) { + ats_needed = true; + } if (sdev->accel) { nb_nodes++; } @@ -XXX,XX +XXX,XX @@ build_iort(GArray *table_data, BIOSLinker *linker, VirtMachineState *vms) build_append_int_noprefix(table_data, 0, 2); /* Reserved */ /* Table 15 Memory Access Flags */ build_append_int_noprefix(table_data, 0x3 /* CCA = CPM = DACS = 1 */, 1); - - build_append_int_noprefix(table_data, 0, 4); /* ATS Attribute */ + /* ATS Attribute */ + build_append_int_noprefix(table_data, ats_needed, 4); /* MCFG pci_segment */ build_append_int_noprefix(table_data, 0, 4); /* PCI Segment number */ diff --git a/include/hw/arm/smmuv3.h b/include/hw/arm/smmuv3.h index XXXXXXX..XXXXXXX 100644 --- a/include/hw/arm/smmuv3.h +++ b/include/hw/arm/smmuv3.h @@ -XXX,XX +XXX,XX @@ struct SMMUv3State { uint64_t msi_gpa; Error *migration_blocker; bool ril; + bool ats; }; typedef enum { -- 2.43.0 From: Shameer Kolothum <skolothumtho@nvidia.com> QEMU SMMUv3 currently sets the output address size (OAS) to 44 bits. With accelerator mode enabled, a device may use SVA, where CPU page tables are shared with the SMMU, requiring an OAS at least as large as the CPU’s output address size. A user option is added to configure this. However, the OAS value advertised by the virtual SMMU must remain compatible with the capabilities of the host SMMUv3. In accelerated mode, the host SMMU performs stage-2 translation and must be able to consume the intermediate physical addresses (IPA) produced by stage-1. The OAS exposed by the virtual SMMU defines the maximum IPA width that stage-1 translations may generate. For AArch64 implementations, the maximum usable IPA size on the host SMMU is determined by its own OAS. Check that the configured OAS does not exceed what the host SMMU can safely support. Tested-by: Zhangfei Gao <zhangfei.gao@linaro.org> Reviewed-by: Nicolin Chen <nicolinc@nvidia.com> Reviewed-by: Eric Auger <eric.auger@redhat.com> Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com> Tested-by: Eric Auger <eric.auger@redhat.com> Signed-off-by: Shameer Kolothum <skolothumtho@nvidia.com> Message-id: 20260126104342.253965-32-skolothumtho@nvidia.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/arm/smmuv3-accel.c | 22 ++++++++++++++++++++++ hw/arm/smmuv3.c | 16 +++++++++++++++- include/hw/arm/smmuv3-common.h | 5 ++++- include/hw/arm/smmuv3.h | 1 + 4 files changed, 42 insertions(+), 2 deletions(-) diff --git a/hw/arm/smmuv3-accel.c b/hw/arm/smmuv3-accel.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/smmuv3-accel.c +++ b/hw/arm/smmuv3-accel.c @@ -XXX,XX +XXX,XX @@ static MemoryRegion root, sysmem; static AddressSpace *shared_as_sysmem; +static int smmuv3_oas_bits(uint32_t oas) +{ + static const int map[] = { 32, 36, 40, 42, 44, 48, 52, 56 }; + + g_assert(oas < ARRAY_SIZE(map)); + return map[oas]; +} + static bool smmuv3_accel_check_hw_compatible(SMMUv3State *s, struct iommu_hw_info_arm_smmuv3 *info, @@ -XXX,XX +XXX,XX @@ smmuv3_accel_check_hw_compatible(SMMUv3State *s, error_setg(errp, "Host SMMUv3 doesn't support Range Invalidation"); return false; } + /* Check OAS value opted is compatible with Host SMMUv3 IPA */ + if (FIELD_EX32(info->idr[5], IDR5, OAS) < + FIELD_EX32(s->idr[5], IDR5, OAS)) { + error_setg(errp, "Host SMMUv3 supports only %d-bit IPA, but the vSMMU " + "OAS implies %d-bit IPA", + smmuv3_oas_bits(FIELD_EX32(info->idr[5], IDR5, OAS)), + smmuv3_oas_bits(FIELD_EX32(s->idr[5], IDR5, OAS))); + return false; + } /* QEMU SMMUv3 supports GRAN4K/GRAN16K/GRAN64K translation granules */ if (FIELD_EX32(info->idr[5], IDR5, GRAN4K) != @@ -XXX,XX +XXX,XX @@ void smmuv3_accel_idr_override(SMMUv3State *s) /* QEMU SMMUv3 has no ATS. Advertise ATS if opt-in by property */ s->idr[0] = FIELD_DP32(s->idr[0], IDR0, ATS, s->ats); + + /* Advertise 48-bit OAS in IDR5 when requested (default is 44 bits). */ + if (s->oas == SMMU_OAS_48BIT) { + s->idr[5] = FIELD_DP32(s->idr[5], IDR5, OAS, SMMU_IDR5_OAS_48); + } } /* Based on SMUUv3 GPBA.ABORT configuration, attach a corresponding HWPT */ diff --git a/hw/arm/smmuv3.c b/hw/arm/smmuv3.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/smmuv3.c +++ b/hw/arm/smmuv3.c @@ -XXX,XX +XXX,XX @@ static void smmuv3_init_id_regs(SMMUv3State *s) s->idr[3] = FIELD_DP32(s->idr[3], IDR3, RIL, 1); s->idr[3] = FIELD_DP32(s->idr[3], IDR3, BBML, 2); - s->idr[5] = FIELD_DP32(s->idr[5], IDR5, OAS, SMMU_IDR5_OAS); /* 44 bits */ + /* OAS: 44 bits */ + s->idr[5] = FIELD_DP32(s->idr[5], IDR5, OAS, SMMU_IDR5_OAS_44); /* 4K, 16K and 64K granule support */ s->idr[5] = FIELD_DP32(s->idr[5], IDR5, GRAN4K, 1); s->idr[5] = FIELD_DP32(s->idr[5], IDR5, GRAN16K, 1); @@ -XXX,XX +XXX,XX @@ static bool smmu_validate_property(SMMUv3State *s, Error **errp) error_setg(errp, "ats can only be enabled if accel=on"); return false; } + if (s->oas != SMMU_OAS_44BIT) { + error_setg(errp, "OAS must be 44 bits when accel=off"); + return false; + } return true; } @@ -XXX,XX +XXX,XX @@ static bool smmu_validate_property(SMMUv3State *s, Error **errp) return false; } + if (s->oas != SMMU_OAS_44BIT && s->oas != SMMU_OAS_48BIT) { + error_setg(errp, "OAS can only be set to 44 or 48 bits"); + return false; + } + return true; } @@ -XXX,XX +XXX,XX @@ static const Property smmuv3_properties[] = { /* RIL can be turned off for accel cases */ DEFINE_PROP_BOOL("ril", SMMUv3State, ril, true), DEFINE_PROP_BOOL("ats", SMMUv3State, ats, false), + DEFINE_PROP_UINT8("oas", SMMUv3State, oas, 44), }; static void smmuv3_instance_init(Object *obj) @@ -XXX,XX +XXX,XX @@ static void smmuv3_class_init(ObjectClass *klass, const void *data) object_class_property_set_description(klass, "ats", "Enable/disable ATS support (for accel=on). Please ensure host " "platform has ATS support before enabling this"); + object_class_property_set_description(klass, "oas", + "Specify Output Address Size (for accel=on). Supported values " + "are 44 or 48 bits. Defaults to 44 bits"); } static int smmuv3_notify_flag_changed(IOMMUMemoryRegion *iommu, diff --git a/include/hw/arm/smmuv3-common.h b/include/hw/arm/smmuv3-common.h index XXXXXXX..XXXXXXX 100644 --- a/include/hw/arm/smmuv3-common.h +++ b/include/hw/arm/smmuv3-common.h @@ -XXX,XX +XXX,XX @@ REG32(IDR5, 0x14) FIELD(IDR5, VAX, 10, 2); FIELD(IDR5, STALL_MAX, 16, 16); -#define SMMU_IDR5_OAS 4 +#define SMMU_OAS_44BIT 44 +#define SMMU_OAS_48BIT 48 +#define SMMU_IDR5_OAS_44 4 +#define SMMU_IDR5_OAS_48 5 REG32(IIDR, 0x18) REG32(AIDR, 0x1c) diff --git a/include/hw/arm/smmuv3.h b/include/hw/arm/smmuv3.h index XXXXXXX..XXXXXXX 100644 --- a/include/hw/arm/smmuv3.h +++ b/include/hw/arm/smmuv3.h @@ -XXX,XX +XXX,XX @@ struct SMMUv3State { Error *migration_blocker; bool ril; bool ats; + uint8_t oas; }; typedef enum { -- 2.43.0 From: Shameer Kolothum <skolothumtho@nvidia.com> Retrieve PASID width from iommufd_backend_get_device_info() and store it in HostIOMMUDeviceCaps for later use. Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com> Reviewed-by: Eric Auger <eric.auger@redhat.com> Tested-by: Zhangfei Gao <zhangfei.gao@linaro.org> Reviewed-by: Nicolin Chen <nicolinc@nvidia.com> Reviewed-by: Cédric Le Goater <clg@redhat.com> Tested-by: Eric Auger <eric.auger@redhat.com> Signed-off-by: Shameer Kolothum <skolothumtho@nvidia.com> Reviewed-by: Yi Liu <yi.l.liu@intel.com> Message-id: 20260126104342.253965-33-skolothumtho@nvidia.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- backends/iommufd.c | 6 +++++- hw/arm/smmuv3-accel.c | 3 ++- hw/vfio/iommufd.c | 6 ++++-- include/system/host_iommu_device.h | 3 +++ include/system/iommufd.h | 3 ++- 5 files changed, 16 insertions(+), 5 deletions(-) diff --git a/backends/iommufd.c b/backends/iommufd.c index XXXXXXX..XXXXXXX 100644 --- a/backends/iommufd.c +++ b/backends/iommufd.c @@ -XXX,XX +XXX,XX @@ bool iommufd_backend_get_dirty_bitmap(IOMMUFDBackend *be, bool iommufd_backend_get_device_info(IOMMUFDBackend *be, uint32_t devid, uint32_t *type, void *data, uint32_t len, - uint64_t *caps, Error **errp) + uint64_t *caps, uint8_t *max_pasid_log2, + Error **errp) { struct iommu_hw_info info = { .size = sizeof(info), @@ -XXX,XX +XXX,XX @@ bool iommufd_backend_get_device_info(IOMMUFDBackend *be, uint32_t devid, g_assert(caps); *caps = info.out_capabilities; + if (max_pasid_log2) { + *max_pasid_log2 = info.out_max_pasid_log2; + } return true; } diff --git a/hw/arm/smmuv3-accel.c b/hw/arm/smmuv3-accel.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/smmuv3-accel.c +++ b/hw/arm/smmuv3-accel.c @@ -XXX,XX +XXX,XX @@ smmuv3_accel_hw_compatible(SMMUv3State *s, HostIOMMUDeviceIOMMUFD *idev, uint64_t caps; if (!iommufd_backend_get_device_info(idev->iommufd, idev->devid, &data_type, - &info, sizeof(info), &caps, errp)) { + &info, sizeof(info), &caps, NULL, + errp)) { return false; } diff --git a/hw/vfio/iommufd.c b/hw/vfio/iommufd.c index XXXXXXX..XXXXXXX 100644 --- a/hw/vfio/iommufd.c +++ b/hw/vfio/iommufd.c @@ -XXX,XX +XXX,XX @@ static bool iommufd_cdev_autodomains_get(VFIODevice *vbasedev, */ if (!iommufd_backend_get_device_info(vbasedev->iommufd, vbasedev->devid, &type, &caps, sizeof(caps), &hw_caps, - errp)) { + NULL, errp)) { return false; } @@ -XXX,XX +XXX,XX @@ static bool hiod_iommufd_vfio_realize(HostIOMMUDevice *hiod, void *opaque, HostIOMMUDeviceCaps *caps = &hiod->caps; VendorCaps *vendor_caps = &caps->vendor_caps; enum iommu_hw_info_type type; + uint8_t max_pasid_log2; uint64_t hw_caps; hiod->agent = opaque; if (!iommufd_backend_get_device_info(vdev->iommufd, vdev->devid, &type, vendor_caps, sizeof(*vendor_caps), - &hw_caps, errp)) { + &hw_caps, &max_pasid_log2, errp)) { return false; } hiod->name = g_strdup(vdev->name); caps->type = type; caps->hw_caps = hw_caps; + caps->max_pasid_log2 = max_pasid_log2; idev = HOST_IOMMU_DEVICE_IOMMUFD(hiod); idev->iommufd = vdev->iommufd; diff --git a/include/system/host_iommu_device.h b/include/system/host_iommu_device.h index XXXXXXX..XXXXXXX 100644 --- a/include/system/host_iommu_device.h +++ b/include/system/host_iommu_device.h @@ -XXX,XX +XXX,XX @@ typedef union VendorCaps { * @hw_caps: host platform IOMMU capabilities (e.g. on IOMMUFD this represents * the @out_capabilities value returned from IOMMU_GET_HW_INFO ioctl) * + * @max_pasid_log2: width of PASIDs supported by host IOMMU device + * * @vendor_caps: host platform IOMMU vendor specific capabilities (e.g. on * IOMMUFD this represents a user-space buffer filled by kernel * with host IOMMU @type specific hardware information data) @@ -XXX,XX +XXX,XX @@ typedef union VendorCaps { typedef struct HostIOMMUDeviceCaps { uint32_t type; uint64_t hw_caps; + uint8_t max_pasid_log2; VendorCaps vendor_caps; } HostIOMMUDeviceCaps; #endif diff --git a/include/system/iommufd.h b/include/system/iommufd.h index XXXXXXX..XXXXXXX 100644 --- a/include/system/iommufd.h +++ b/include/system/iommufd.h @@ -XXX,XX +XXX,XX @@ int iommufd_backend_unmap_dma(IOMMUFDBackend *be, uint32_t ioas_id, hwaddr iova, uint64_t size); bool iommufd_backend_get_device_info(IOMMUFDBackend *be, uint32_t devid, uint32_t *type, void *data, uint32_t len, - uint64_t *caps, Error **errp); + uint64_t *caps, uint8_t *max_pasid_log2, + Error **errp); bool iommufd_backend_alloc_hwpt(IOMMUFDBackend *be, uint32_t dev_id, uint32_t pt_id, uint32_t flags, uint32_t data_type, uint32_t data_len, -- 2.43.0 From: Shameer Kolothum <skolothumtho@nvidia.com> The get_pasid_info callback retrieves PASID capability information when the HostIOMMUDevice backend supports it. Currently, only the Linux IOMMUFD backend provides this information. This will be used by a subsequent patch to synthesize a PASID capability for vfio-pci devices behind a vIOMMU that supports PASID. Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com> Reviewed-by: Eric Auger <eric.auger@redhat.com> Tested-by: Eric Auger <eric.auger@redhat.com> Tested-by: Zhangfei Gao <zhangfei.gao@linaro.org> Reviewed-by: Nicolin Chen <nicolinc@nvidia.com> Signed-off-by: Shameer Kolothum <skolothumtho@nvidia.com> Reviewed-by: Yi Liu <yi.l.liu@intel.com> Message-id: 20260126104342.253965-34-skolothumtho@nvidia.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- backends/iommufd.c | 17 +++++++++++++++++ include/system/host_iommu_device.h | 17 +++++++++++++++++ 2 files changed, 34 insertions(+) diff --git a/backends/iommufd.c b/backends/iommufd.c index XXXXXXX..XXXXXXX 100644 --- a/backends/iommufd.c +++ b/backends/iommufd.c @@ -XXX,XX +XXX,XX @@ static int hiod_iommufd_get_cap(HostIOMMUDevice *hiod, int cap, Error **errp) } } +static bool hiod_iommufd_get_pasid_info(HostIOMMUDevice *hiod, + PasidInfo *pasid_info) +{ + HostIOMMUDeviceCaps *caps = &hiod->caps; + + if (!caps->max_pasid_log2) { + return false; + } + + g_assert(pasid_info); + pasid_info->exec_perm = (caps->hw_caps & IOMMU_HW_CAP_PCI_PASID_EXEC); + pasid_info->priv_mod = (caps->hw_caps & IOMMU_HW_CAP_PCI_PASID_PRIV); + pasid_info->max_pasid_log2 = caps->max_pasid_log2; + return true; +} + static void hiod_iommufd_class_init(ObjectClass *oc, const void *data) { HostIOMMUDeviceClass *hioc = HOST_IOMMU_DEVICE_CLASS(oc); hioc->get_cap = hiod_iommufd_get_cap; + hioc->get_pasid_info = hiod_iommufd_get_pasid_info; }; static const TypeInfo types[] = { diff --git a/include/system/host_iommu_device.h b/include/system/host_iommu_device.h index XXXXXXX..XXXXXXX 100644 --- a/include/system/host_iommu_device.h +++ b/include/system/host_iommu_device.h @@ -XXX,XX +XXX,XX @@ struct HostIOMMUDevice { #endif }; +typedef struct PasidInfo { + bool exec_perm; + bool priv_mod; + uint8_t max_pasid_log2; +} PasidInfo; + /** * struct HostIOMMUDeviceClass - The base class for all host IOMMU devices. * @@ -XXX,XX +XXX,XX @@ struct HostIOMMUDeviceClass { * @hiod: handle to the host IOMMU device */ uint64_t (*get_page_size_mask)(HostIOMMUDevice *hiod); + /** + * @get_pasid_info: Return the PASID information associated with the + * @hiod Host IOMMU device. + * + * @hiod: handle to the host IOMMU device + * + * @pasid_info: If success, returns the PASID related information. + * + * Returns: true on success, false on failure. + */ + bool (*get_pasid_info)(HostIOMMUDevice *hiod, PasidInfo *pasid_info); }; /* -- 2.43.0 From: Shameer Kolothum <skolothumtho@nvidia.com> Add pcie_insert_capability(), a helper to insert a PCIe extended capability into an existing extended capability list at a caller specified offset. Unlike pcie_add_capability(), which always appends a capability to the end of the list, this helper preserves the existing list ordering while allowing insertion at an arbitrary offset. The helper only validates that the insertion does not overwrite an existing PCIe extended capability header, since corrupting a header would break the extended capability linked list. Validation of overlaps with other configuration space registers or capability-specific register blocks is left to the caller. Reviewed-by: Michael S. Tsirkin <mst@redhat.com> Reviewed-by: Eric Auger <eric.auger@redhat.com> Tested-by: Eric Auger <eric.auger@redhat.com> Tested-by: Zhangfei Gao <zhangfei.gao@linaro.org> Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com> Signed-off-by: Shameer Kolothum <skolothumtho@nvidia.com> Reviewed-by: Yi Liu <yi.l.liu@intel.com> Message-id: 20260126104342.253965-35-skolothumtho@nvidia.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/pci/pcie.c | 69 +++++++++++++++++++++++++++++++++++++++++++ include/hw/pci/pcie.h | 2 ++ 2 files changed, 71 insertions(+) diff --git a/hw/pci/pcie.c b/hw/pci/pcie.c index XXXXXXX..XXXXXXX 100644 --- a/hw/pci/pcie.c +++ b/hw/pci/pcie.c @@ -XXX,XX +XXX,XX @@ static void pcie_ext_cap_set_next(PCIDevice *dev, uint16_t pos, uint16_t next) pci_set_long(dev->config + pos, header); } +/* + * Insert a PCIe extended capability at a given offset. + * + * This helper only validates that the insertion does not overwrite an + * existing PCIe extended capability header, as corrupting a header would + * break the extended capability linked list. + * + * The caller must ensure that (offset, size) does not overlap with other + * registers or capability-specific register blocks. Overlaps with + * capability-specific registers are not checked and are considered a + * user-controlled override. + * + * Note: Best effort helper. The PCIe spec does not require extended + * capabilities to be ordered, but most devices use a forward-linked list. + * Devices that do not consistently use a forward-linked list may cause + * insertion to fail. + */ +bool pcie_insert_capability(PCIDevice *dev, uint16_t cap_id, uint8_t cap_ver, + uint16_t offset, uint16_t size) +{ + uint16_t pos = PCI_CONFIG_SPACE_SIZE, prev = 0; + uint32_t header; + + assert(pci_is_express(dev)); + + if (!QEMU_IS_ALIGNED(offset, PCI_EXT_CAP_ALIGN) || + size < 8 || + offset < PCI_CONFIG_SPACE_SIZE || + offset >= PCIE_CONFIG_SPACE_SIZE || + offset + size > PCIE_CONFIG_SPACE_SIZE) { + return false; + } + + header = pci_get_long(dev->config + pos); + if (!header) { + /* No extended capability present, insertion must be at the ECAP head */ + if (offset != pos) { + return false; + } + pci_set_long(dev->config + pos, PCI_EXT_CAP(cap_id, cap_ver, 0)); + goto out; + } + + while (header && pos && offset >= pos) { + uint16_t next = PCI_EXT_CAP_NEXT(header); + + /* Reject insertion inside an existing ECAP header (4 bytes) */ + if (offset < pos + PCI_EXT_CAP_ALIGN) { + return false; + } + + prev = pos; + pos = next; + header = pos ? pci_get_long(dev->config + pos) : 0; + } + + pci_set_long(dev->config + offset, PCI_EXT_CAP(cap_id, cap_ver, pos)); + if (prev) { + pcie_ext_cap_set_next(dev, prev, offset); + } + +out: + /* Make capability read-only by default */ + memset(dev->wmask + offset, 0, size); + memset(dev->w1cmask + offset, 0, size); + /* Check capability by default */ + memset(dev->cmask + offset, 0xFF, size); + return true; +} /* * Caller must supply valid (offset, size) such that the range wouldn't * overlap with other capability or other registers. diff --git a/include/hw/pci/pcie.h b/include/hw/pci/pcie.h index XXXXXXX..XXXXXXX 100644 --- a/include/hw/pci/pcie.h +++ b/include/hw/pci/pcie.h @@ -XXX,XX +XXX,XX @@ uint16_t pcie_find_capability(PCIDevice *dev, uint16_t cap_id); void pcie_add_capability(PCIDevice *dev, uint16_t cap_id, uint8_t cap_ver, uint16_t offset, uint16_t size); +bool pcie_insert_capability(PCIDevice *dev, uint16_t cap_id, uint8_t cap_ver, + uint16_t offset, uint16_t size); void pcie_sync_bridge_lnk(PCIDevice *dev); void pcie_acs_init(PCIDevice *dev, uint16_t offset); -- 2.43.0 From: Shameer Kolothum <skolothumtho@nvidia.com> Refactor PCIe PASID capability initialization by moving the common register init into a new helper, pcie_pasid_common_init(). Subsequent patch to synthesize a vPASID will make use of this helper. No functional change intended. Cc: Michael S. Tsirkin <mst@redhat.com> Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com> Reviewed-by: Eric Auger <eric.auger@redhat.com> Reviewed-by: Michael S. Tsirkin <mst@redhat.com> Tested-by: Eric Auger <eric.auger@redhat.com> Tested-by: Zhangfei Gao <zhangfei.gao@linaro.org> Signed-off-by: Shameer Kolothum <skolothumtho@nvidia.com> Reviewed-by: Yi Liu <yi.l.liu@intel.com> Message-id: 20260126104342.253965-36-skolothumtho@nvidia.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/pci/pcie.c | 19 ++++++++++++------- include/hw/pci/pcie.h | 2 ++ 2 files changed, 14 insertions(+), 7 deletions(-) diff --git a/hw/pci/pcie.c b/hw/pci/pcie.c index XXXXXXX..XXXXXXX 100644 --- a/hw/pci/pcie.c +++ b/hw/pci/pcie.c @@ -XXX,XX +XXX,XX @@ void pcie_acs_reset(PCIDevice *dev) } } -/* PASID */ -void pcie_pasid_init(PCIDevice *dev, uint16_t offset, uint8_t pasid_width, - bool exec_perm, bool priv_mod) +void pcie_pasid_common_init(PCIDevice *dev, uint16_t offset, + uint8_t pasid_width, bool exec_perm, bool priv_mod) { static const uint16_t control_reg_rw_mask = 0x07; uint16_t capability_reg; assert(pasid_width <= PCI_EXT_CAP_PASID_MAX_WIDTH); - - pcie_add_capability(dev, PCI_EXT_CAP_ID_PASID, PCI_PASID_VER, offset, - PCI_EXT_CAP_PASID_SIZEOF); - capability_reg = ((uint16_t)pasid_width) << PCI_PASID_CAP_WIDTH_SHIFT; capability_reg |= exec_perm ? PCI_PASID_CAP_EXEC : 0; capability_reg |= priv_mod ? PCI_PASID_CAP_PRIV : 0; @@ -XXX,XX +XXX,XX @@ void pcie_pasid_init(PCIDevice *dev, uint16_t offset, uint8_t pasid_width, pci_set_word(dev->wmask + offset + PCI_PASID_CTRL, control_reg_rw_mask); dev->exp.pasid_cap = offset; + +} + +/* PASID */ +void pcie_pasid_init(PCIDevice *dev, uint16_t offset, uint8_t pasid_width, + bool exec_perm, bool priv_mod) +{ + pcie_add_capability(dev, PCI_EXT_CAP_ID_PASID, PCI_PASID_VER, offset, + PCI_EXT_CAP_PASID_SIZEOF); + pcie_pasid_common_init(dev, offset, pasid_width, exec_perm, priv_mod); } /* PRI */ diff --git a/include/hw/pci/pcie.h b/include/hw/pci/pcie.h index XXXXXXX..XXXXXXX 100644 --- a/include/hw/pci/pcie.h +++ b/include/hw/pci/pcie.h @@ -XXX,XX +XXX,XX @@ void pcie_cap_slot_unplug_cb(HotplugHandler *hotplug_dev, DeviceState *dev, void pcie_cap_slot_unplug_request_cb(HotplugHandler *hotplug_dev, DeviceState *dev, Error **errp); +void pcie_pasid_common_init(PCIDevice *dev, uint16_t offset, + uint8_t pasid_width, bool exec_perm, bool priv_mod); void pcie_pasid_init(PCIDevice *dev, uint16_t offset, uint8_t pasid_width, bool exec_perm, bool priv_mod); void pcie_pri_init(PCIDevice *dev, uint16_t offset, uint32_t outstanding_pr_cap, -- 2.43.0 From: Shameer Kolothum <skolothumtho@nvidia.com> Add support for synthesizing a PCIe PASID extended capability for vfio-pci devices when PASID is enabled via a vIOMMU and supported by the host IOMMU backend. PASID capability parameters are retrieved via IOMMUFD APIs and the capability is inserted into the PCIe extended capability list using the insertion helper. A new x-vpasid-cap-offset property allows explicit control over the placement; by default the capability is placed at the end of the PCIe extended configuration space. If the kernel does not expose PASID information or insertion fails, the device continues without PASID support. Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com> Tested-by: Eric Auger <eric.auger@redhat.com> Tested-by: Zhangfei Gao <zhangfei.gao@linaro.org> Reviewed-by: Cédric Le Goater <clg@redhat.com> Signed-off-by: Shameer Kolothum <skolothumtho@nvidia.com> Reviewed-by: Yi Liu <yi.l.liu@intel.com> Message-id: 20260126104342.253965-37-skolothumtho@nvidia.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/vfio/pci.c | 75 +++++++++++++++++++++++++++++++++++++++++ hw/vfio/pci.h | 1 + hw/vfio/trace-events | 1 + include/hw/core/iommu.h | 1 + 4 files changed, 78 insertions(+) diff --git a/hw/vfio/pci.c b/hw/vfio/pci.c index XXXXXXX..XXXXXXX 100644 --- a/hw/vfio/pci.c +++ b/hw/vfio/pci.c @@ -XXX,XX +XXX,XX @@ #include <sys/ioctl.h> #include "hw/core/hw-error.h" +#include "hw/core/iommu.h" #include "hw/pci/msi.h" #include "hw/pci/msix.h" #include "hw/pci/pci_bridge.h" @@ -XXX,XX +XXX,XX @@ static int vfio_setup_rebar_ecap(VFIOPCIDevice *vdev, uint16_t pos) return 0; } +/* + * Try to retrieve PASID capability information via IOMMUFD APIs and, + * if supported, synthesize a PASID PCIe extended capability for the + * VFIO device. + * + * Use user-specified PASID capability offset if provided, otherwise + * place it at the end of the PCIe extended configuration space. + */ +static bool vfio_pci_synthesize_pasid_cap(VFIOPCIDevice *vdev, Error **errp) +{ + HostIOMMUDevice *hiod = vdev->vbasedev.hiod; + HostIOMMUDeviceClass *hiodc; + PasidInfo pasid_info; + PCIDevice *pdev = PCI_DEVICE(vdev); + uint16_t pasid_offset; + + if (!hiod) { + return true; + } + + hiodc = HOST_IOMMU_DEVICE_GET_CLASS(hiod); + if (!hiodc || !hiodc->get_pasid_info || + !hiodc->get_pasid_info(hiod, &pasid_info) || + !(pci_device_get_viommu_flags(pdev) & VIOMMU_FLAG_PASID_SUPPORTED)) { + return true; + } + + /* Use user-specified offset if set, otherwise place PASID at the end. */ + if (vdev->vpasid_cap_offset) { + pasid_offset = vdev->vpasid_cap_offset; + } else { + pasid_offset = PCIE_CONFIG_SPACE_SIZE - PCI_EXT_CAP_PASID_SIZEOF; + } + + if (!pcie_insert_capability(pdev, PCI_EXT_CAP_ID_PASID, PCI_PASID_VER, + pasid_offset, PCI_EXT_CAP_PASID_SIZEOF)) { + error_setg(errp, "vfio: Placing PASID capability at offset 0x%x failed", + pasid_offset); + return false; + } + trace_vfio_pci_synthesize_pasid_cap(vdev->vbasedev.name, pasid_offset); + + pcie_pasid_common_init(pdev, pasid_offset, pasid_info.max_pasid_log2, + pasid_info.exec_perm, pasid_info.priv_mod); + + /* PASID capability is fully emulated by QEMU */ + memset(vdev->emulated_config_bits + pdev->exp.pasid_cap, 0xff, + PCI_EXT_CAP_PASID_SIZEOF); + return true; +} + static void vfio_add_ext_cap(VFIOPCIDevice *vdev) { PCIDevice *pdev = PCI_DEVICE(vdev); + bool pasid_cap_added = false; + Error *err = NULL; uint32_t header; uint16_t cap_id, next, size; uint8_t cap_ver; @@ -XXX,XX +XXX,XX @@ static void vfio_add_ext_cap(VFIOPCIDevice *vdev) pcie_add_capability(pdev, cap_id, cap_ver, next, size); } break; + /* + * VFIO kernel does not expose the PASID CAP today. We may synthesize + * one later through IOMMUFD APIs. If VFIO ever starts exposing it, + * record its presence here so we do not create a duplicate CAP. + */ + case PCI_EXT_CAP_ID_PASID: + pasid_cap_added = true; + /* fallthrough */ default: pcie_add_capability(pdev, cap_id, cap_ver, next, size); } } + if (!pasid_cap_added && !vfio_pci_synthesize_pasid_cap(vdev, &err)) { + error_report_err(err); + } + /* Cleanup chain head ID if necessary */ if (pci_get_word(pdev->config + PCI_CONFIG_SPACE_SIZE) == 0xFFFF) { pci_set_word(pdev->config + PCI_CONFIG_SPACE_SIZE, 0); @@ -XXX,XX +XXX,XX @@ static const Property vfio_pci_properties[] = { TYPE_IOMMUFD_BACKEND, IOMMUFDBackend *), #endif DEFINE_PROP_BOOL("skip-vsc-check", VFIOPCIDevice, skip_vsc_check, true), + DEFINE_PROP_UINT16("x-vpasid-cap-offset", VFIOPCIDevice, + vpasid_cap_offset, 0), }; #ifdef CONFIG_IOMMUFD @@ -XXX,XX +XXX,XX @@ static void vfio_pci_class_init(ObjectClass *klass, const void *data) "destination when doing live " "migration of device state via " "multifd channels"); + object_class_property_set_description(klass, /* 11.0 */ + "x-vpasid-cap-offset", + "PCIe extended configuration space offset at which to place a " + "synthetic PASID extended capability when PASID is enabled via " + "a vIOMMU. A value of 0 (default) places the capability at the " + "end of the extended configuration space. The offset must be " + "4-byte aligned and within the PCIe extended configuration space"); } static const TypeInfo vfio_pci_info = { diff --git a/hw/vfio/pci.h b/hw/vfio/pci.h index XXXXXXX..XXXXXXX 100644 --- a/hw/vfio/pci.h +++ b/hw/vfio/pci.h @@ -XXX,XX +XXX,XX @@ struct VFIOPCIDevice { bool defer_kvm_irq_routing; bool clear_parent_atomics_on_exit; bool skip_vsc_check; + uint16_t vpasid_cap_offset; VFIODisplay *dpy; Notifier irqchip_change_notifier; VFIOPCICPR cpr; diff --git a/hw/vfio/trace-events b/hw/vfio/trace-events index XXXXXXX..XXXXXXX 100644 --- a/hw/vfio/trace-events +++ b/hw/vfio/trace-events @@ -XXX,XX +XXX,XX @@ vfio_pci_hot_reset_result(const char *name, const char *result) "%s hot reset: % vfio_pci_populate_device_config(const char *name, unsigned long size, unsigned long offset, unsigned long flags) "Device '%s' config: size: 0x%lx, offset: 0x%lx, flags: 0x%lx" vfio_pci_populate_device_get_irq_info_failure(const char *errstr) "VFIO_DEVICE_GET_IRQ_INFO failure: %s" vfio_mdev(const char *name, bool is_mdev) " (%s) is_mdev %d" +vfio_pci_synthesize_pasid_cap(const char *name, uint16_t offset) "%s offset: 0x%x" vfio_add_ext_cap_dropped(const char *name, uint16_t cap, uint16_t offset) "%s 0x%x@0x%x" vfio_pci_reset(const char *name) " (%s)" vfio_pci_reset_flr(const char *name) "%s FLR/VFIO_DEVICE_RESET" diff --git a/include/hw/core/iommu.h b/include/hw/core/iommu.h index XXXXXXX..XXXXXXX 100644 --- a/include/hw/core/iommu.h +++ b/include/hw/core/iommu.h @@ -XXX,XX +XXX,XX @@ enum viommu_flags { /* vIOMMU needs nesting parent HWPT to create nested HWPT */ VIOMMU_FLAG_WANT_NESTING_PARENT = BIT_ULL(0), + VIOMMU_FLAG_PASID_SUPPORTED = BIT_ULL(1), }; /* Host IOMMU quirks. Extracted from host IOMMU capabilities */ -- 2.43.0 From: Shameer Kolothum <skolothumtho@nvidia.com> QEMU SMMUv3 currently reports no SubstreamID support, forcing SSID to zero. This prevents accelerated use cases such as Shared Virtual Addressing (SVA), which require multiple Stage-1 context descriptors indexed by SubstreamID. Add a new "ssidsize" property to explicitly configure the number of bits used for SubstreamIDs. A value greater than zero enables SubstreamID support and advertises PASID capability to the vIOMMU. The requested SSIDSIZE is validated against host SMMUv3 capabilities and is only supported when accel=on. Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com> Reviewed-by: Eric Auger <eric.auger@redhat.com> Reviewed-by: Nicolin Chen <nicolinc@nvidia.com> Tested-by: Eric Auger <eric.auger@redhat.com> Tested-by: Zhangfei Gao <zhangfei.gao@linaro.org> Signed-off-by: Shameer Kolothum <skolothumtho@nvidia.com> Message-id: 20260126104342.253965-38-skolothumtho@nvidia.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/arm/smmuv3-accel.c | 25 ++++++++++++++++++++++++- hw/arm/smmuv3.c | 22 ++++++++++++++++++++-- include/hw/arm/smmuv3-common.h | 1 + include/hw/arm/smmuv3.h | 1 + 4 files changed, 46 insertions(+), 3 deletions(-) diff --git a/hw/arm/smmuv3-accel.c b/hw/arm/smmuv3-accel.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/smmuv3-accel.c +++ b/hw/arm/smmuv3-accel.c @@ -XXX,XX +XXX,XX @@ smmuv3_accel_check_hw_compatible(SMMUv3State *s, return false; } + /* Check SSIDSIZE value opted-in is compatible with Host SMMUv3 SSIDSIZE */ + if (FIELD_EX32(info->idr[1], IDR1, SSIDSIZE) < + FIELD_EX32(s->idr[1], IDR1, SSIDSIZE)) { + error_setg(errp, "Host SMMUv3 SSIDSIZE not compatible " + "(host=%u, QEMU=%u)", + FIELD_EX32(info->idr[1], IDR1, SSIDSIZE), + FIELD_EX32(s->idr[1], IDR1, SSIDSIZE)); + return false; + } + /* User can disable QEMU SMMUv3 Range Invalidation support */ if (FIELD_EX32(info->idr[3], IDR3, RIL) < FIELD_EX32(s->idr[3], IDR3, RIL)) { @@ -XXX,XX +XXX,XX @@ static uint64_t smmuv3_accel_get_viommu_flags(void *opaque) * The real HW nested support should be reported from host SMMUv3 and if * it doesn't, the nesting parent allocation will fail anyway in VFIO core. */ - return VIOMMU_FLAG_WANT_NESTING_PARENT; + uint64_t flags = VIOMMU_FLAG_WANT_NESTING_PARENT; + SMMUState *bs = opaque; + SMMUv3State *s = ARM_SMMUV3(bs); + + if (s->ssidsize) { + flags |= VIOMMU_FLAG_PASID_SUPPORTED; + } + return flags; } static const PCIIOMMUOps smmuv3_accel_ops = { @@ -XXX,XX +XXX,XX @@ void smmuv3_accel_idr_override(SMMUv3State *s) if (s->oas == SMMU_OAS_48BIT) { s->idr[5] = FIELD_DP32(s->idr[5], IDR5, OAS, SMMU_IDR5_OAS_48); } + + /* + * By default QEMU SMMUv3 has no SubstreamID support. Update IDR1 if user + * has enabled it. + */ + s->idr[1] = FIELD_DP32(s->idr[1], IDR1, SSIDSIZE, s->ssidsize); } /* Based on SMUUv3 GPBA.ABORT configuration, attach a corresponding HWPT */ diff --git a/hw/arm/smmuv3.c b/hw/arm/smmuv3.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/smmuv3.c +++ b/hw/arm/smmuv3.c @@ -XXX,XX +XXX,XX @@ static int decode_ste(SMMUv3State *s, SMMUTransCfg *cfg, } } - if (STE_S1CDMAX(ste) != 0) { + /* Multiple context descriptors require SubstreamID support */ + if (!s->ssidsize && STE_S1CDMAX(ste) != 0) { qemu_log_mask(LOG_UNIMP, - "SMMUv3 does not support multiple context descriptors yet\n"); + "SMMUv3: multiple S1 context descriptors require SubstreamID support. " + "Configure ssidsize > 0 (requires accel=on)\n"); goto bad_ste; } @@ -XXX,XX +XXX,XX @@ static bool smmu_validate_property(SMMUv3State *s, Error **errp) error_setg(errp, "OAS must be 44 bits when accel=off"); return false; } + if (s->ssidsize) { + error_setg(errp, "ssidsize can only be set if accel=on"); + return false; + } return true; } @@ -XXX,XX +XXX,XX @@ static bool smmu_validate_property(SMMUv3State *s, Error **errp) error_setg(errp, "OAS can only be set to 44 or 48 bits"); return false; } + if (s->ssidsize > SMMU_SSID_MAX_BITS) { + error_setg(errp, "ssidsize must be in the range 0 to %d", + SMMU_SSID_MAX_BITS); + return false; + } return true; } @@ -XXX,XX +XXX,XX @@ static const Property smmuv3_properties[] = { DEFINE_PROP_BOOL("ril", SMMUv3State, ril, true), DEFINE_PROP_BOOL("ats", SMMUv3State, ats, false), DEFINE_PROP_UINT8("oas", SMMUv3State, oas, 44), + DEFINE_PROP_UINT8("ssidsize", SMMUv3State, ssidsize, 0), }; static void smmuv3_instance_init(Object *obj) @@ -XXX,XX +XXX,XX @@ static void smmuv3_class_init(ObjectClass *klass, const void *data) object_class_property_set_description(klass, "oas", "Specify Output Address Size (for accel=on). Supported values " "are 44 or 48 bits. Defaults to 44 bits"); + object_class_property_set_description(klass, "ssidsize", + "Number of bits used to represent SubstreamIDs (SSIDs). " + "A value of N allows SSIDs in the range [0 .. 2^N - 1]. " + "Valid range is 0-20, where 0 disables SubstreamID support. " + "Defaults to 0. A value greater than 0 is required to enable " + "PASID support."); } static int smmuv3_notify_flag_changed(IOMMUMemoryRegion *iommu, diff --git a/include/hw/arm/smmuv3-common.h b/include/hw/arm/smmuv3-common.h index XXXXXXX..XXXXXXX 100644 --- a/include/hw/arm/smmuv3-common.h +++ b/include/hw/arm/smmuv3-common.h @@ -XXX,XX +XXX,XX @@ REG32(IDR1, 0x4) FIELD(IDR1, TABLES_PRESET, 30, 1) FIELD(IDR1, ECMDQ, 31, 1) +#define SMMU_SSID_MAX_BITS 20 #define SMMU_IDR1_SIDSIZE 16 #define SMMU_CMDQS 19 #define SMMU_EVENTQS 19 diff --git a/include/hw/arm/smmuv3.h b/include/hw/arm/smmuv3.h index XXXXXXX..XXXXXXX 100644 --- a/include/hw/arm/smmuv3.h +++ b/include/hw/arm/smmuv3.h @@ -XXX,XX +XXX,XX @@ struct SMMUv3State { bool ril; bool ats; uint8_t oas; + uint8_t ssidsize; }; typedef enum { -- 2.43.0 From: Philippe Mathieu-Daudé <philmd@linaro.org> Next commit will use these functions prototype earlier. Rather than forward-declaring them, move them around. Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Akihiko Odaki <odaki@rsg.ci.i.u-tokyo.ac.jp> Message-id: 20260118215945.46693-2-philmd@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/hvf/hvf.c | 142 +++++++++++++++++++++---------------------- 1 file changed, 71 insertions(+), 71 deletions(-) diff --git a/target/arm/hvf/hvf.c b/target/arm/hvf/hvf.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/hvf/hvf.c +++ b/target/arm/hvf/hvf.c @@ -XXX,XX +XXX,XX @@ static const hv_sys_reg_t hvf_sreg_list[] = { #undef DEF_SYSREG +static uint32_t hvf_reg2cp_reg(uint32_t reg) +{ + return ENCODE_AA64_CP_REG((reg >> SYSREG_OP0_SHIFT) & SYSREG_OP0_MASK, + (reg >> SYSREG_OP1_SHIFT) & SYSREG_OP1_MASK, + (reg >> SYSREG_CRN_SHIFT) & SYSREG_CRN_MASK, + (reg >> SYSREG_CRM_SHIFT) & SYSREG_CRM_MASK, + (reg >> SYSREG_OP2_SHIFT) & SYSREG_OP2_MASK); +} + +static bool hvf_sysreg_read_cp(CPUState *cpu, const char *cpname, + uint32_t reg, uint64_t *val) +{ + ARMCPU *arm_cpu = ARM_CPU(cpu); + CPUARMState *env = &arm_cpu->env; + const ARMCPRegInfo *ri; + + ri = get_arm_cp_reginfo(arm_cpu->cp_regs, hvf_reg2cp_reg(reg)); + if (ri) { + if (!cp_access_ok(1, ri, true)) { + return false; + } + if (ri->accessfn) { + if (ri->accessfn(env, ri, true) != CP_ACCESS_OK) { + return false; + } + } + if (ri->type & ARM_CP_CONST) { + *val = ri->resetvalue; + } else if (ri->readfn) { + *val = ri->readfn(env, ri); + } else { + *val = raw_read(env, ri); + } + trace_hvf_emu_reginfo_read(cpname, ri->name, *val); + return true; + } + + return false; +} + +static bool hvf_sysreg_write_cp(CPUState *cpu, const char *cpname, + uint32_t reg, uint64_t val) +{ + ARMCPU *arm_cpu = ARM_CPU(cpu); + CPUARMState *env = &arm_cpu->env; + const ARMCPRegInfo *ri; + + ri = get_arm_cp_reginfo(arm_cpu->cp_regs, hvf_reg2cp_reg(reg)); + + if (ri) { + if (!cp_access_ok(1, ri, false)) { + return false; + } + if (ri->accessfn) { + if (ri->accessfn(env, ri, false) != CP_ACCESS_OK) { + return false; + } + } + if (ri->writefn) { + ri->writefn(env, ri, val); + } else { + raw_write(env, ri, val); + } + + trace_hvf_emu_reginfo_write(cpname, ri->name, val); + return true; + } + + return false; +} + int hvf_arch_get_registers(CPUState *cpu) { ARMCPU *arm_cpu = ARM_CPU(cpu); @@ -XXX,XX +XXX,XX @@ static bool is_id_sysreg(uint32_t reg) SYSREG_CRM(reg) < 8; } -static uint32_t hvf_reg2cp_reg(uint32_t reg) -{ - return ENCODE_AA64_CP_REG((reg >> SYSREG_OP0_SHIFT) & SYSREG_OP0_MASK, - (reg >> SYSREG_OP1_SHIFT) & SYSREG_OP1_MASK, - (reg >> SYSREG_CRN_SHIFT) & SYSREG_CRN_MASK, - (reg >> SYSREG_CRM_SHIFT) & SYSREG_CRM_MASK, - (reg >> SYSREG_OP2_SHIFT) & SYSREG_OP2_MASK); -} - -static bool hvf_sysreg_read_cp(CPUState *cpu, const char *cpname, - uint32_t reg, uint64_t *val) -{ - ARMCPU *arm_cpu = ARM_CPU(cpu); - CPUARMState *env = &arm_cpu->env; - const ARMCPRegInfo *ri; - - ri = get_arm_cp_reginfo(arm_cpu->cp_regs, hvf_reg2cp_reg(reg)); - if (ri) { - if (!cp_access_ok(1, ri, true)) { - return false; - } - if (ri->accessfn) { - if (ri->accessfn(env, ri, true) != CP_ACCESS_OK) { - return false; - } - } - if (ri->type & ARM_CP_CONST) { - *val = ri->resetvalue; - } else if (ri->readfn) { - *val = ri->readfn(env, ri); - } else { - *val = raw_read(env, ri); - } - trace_hvf_emu_reginfo_read(cpname, ri->name, *val); - return true; - } - - return false; -} - static int hvf_sysreg_read(CPUState *cpu, uint32_t reg, uint64_t *val) { ARMCPU *arm_cpu = ARM_CPU(cpu); @@ -XXX,XX +XXX,XX @@ static void pmswinc_write(CPUARMState *env, uint64_t value) } } -static bool hvf_sysreg_write_cp(CPUState *cpu, const char *cpname, - uint32_t reg, uint64_t val) -{ - ARMCPU *arm_cpu = ARM_CPU(cpu); - CPUARMState *env = &arm_cpu->env; - const ARMCPRegInfo *ri; - - ri = get_arm_cp_reginfo(arm_cpu->cp_regs, hvf_reg2cp_reg(reg)); - - if (ri) { - if (!cp_access_ok(1, ri, false)) { - return false; - } - if (ri->accessfn) { - if (ri->accessfn(env, ri, false) != CP_ACCESS_OK) { - return false; - } - } - if (ri->writefn) { - ri->writefn(env, ri, val); - } else { - raw_write(env, ri, val); - } - - trace_hvf_emu_reginfo_write(cpname, ri->name, val); - return true; - } - - return false; -} - static int hvf_sysreg_write(CPUState *cpu, uint32_t reg, uint64_t val) { ARMCPU *arm_cpu = ARM_CPU(cpu); -- 2.43.0 From: Philippe Mathieu-Daudé <philmd@linaro.org> Keep CNTV_CTL_EL0 and CNTV_CVAL_EL0 synchronized with the host hardware accelerator. Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Akihiko Odaki <odaki@rsg.ci.i.u-tokyo.ac.jp> Message-id: 20260118215945.46693-3-philmd@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/hvf/hvf.c | 25 +++++++++++++++++++++++++ 1 file changed, 25 insertions(+) diff --git a/target/arm/hvf/hvf.c b/target/arm/hvf/hvf.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/hvf/hvf.c +++ b/target/arm/hvf/hvf.c @@ -XXX,XX +XXX,XX @@ void hvf_arm_init_debug(void) #define SYSREG_PMCEID0_EL0 SYSREG(3, 3, 9, 12, 6) #define SYSREG_PMCEID1_EL0 SYSREG(3, 3, 9, 12, 7) #define SYSREG_PMCCNTR_EL0 SYSREG(3, 3, 9, 13, 0) + +#define SYSREG_CNTV_CTL_EL0 SYSREG(3, 3, 14, 3, 1) +#define SYSREG_CNTV_CVAL_EL0 SYSREG(3, 3, 14, 3, 2) #define SYSREG_PMCCFILTR_EL0 SYSREG(3, 3, 14, 15, 7) #define SYSREG_ICC_AP0R0_EL1 SYSREG(3, 0, 12, 8, 4) @@ -XXX,XX +XXX,XX @@ int hvf_arch_get_registers(CPUState *cpu) uint64_t val; hv_simd_fp_uchar16_t fpval; int i, n; + bool b; for (i = 0; i < ARRAY_SIZE(hvf_reg_match); i++) { ret = hv_vcpu_get_reg(cpu->accel->fd, hvf_reg_match[i].reg, &val); @@ -XXX,XX +XXX,XX @@ int hvf_arch_get_registers(CPUState *cpu) aarch64_restore_sp(env, arm_current_el(env)); + ret = hv_vcpu_get_sys_reg(cpu->accel->fd, HV_SYS_REG_CNTV_CVAL_EL0, &val); + assert_hvf_ok(ret); + b = hvf_sysreg_write_cp(cpu, "VTimer", SYSREG_CNTV_CVAL_EL0, val); + assert(b); + + ret = hv_vcpu_get_sys_reg(cpu->accel->fd, HV_SYS_REG_CNTV_CTL_EL0, &val); + assert_hvf_ok(ret); + b = hvf_sysreg_write_cp(cpu, "VTimer", SYSREG_CNTV_CTL_EL0, val); + assert(b); + return 0; } @@ -XXX,XX +XXX,XX @@ int hvf_arch_put_registers(CPUState *cpu) uint64_t val; hv_simd_fp_uchar16_t fpval; int i, n; + bool b; for (i = 0; i < ARRAY_SIZE(hvf_reg_match); i++) { val = *(uint64_t *)((void *)env + hvf_reg_match[i].offset); @@ -XXX,XX +XXX,XX @@ int hvf_arch_put_registers(CPUState *cpu) ret = hv_vcpu_set_vtimer_offset(cpu->accel->fd, hvf_state->vtimer_offset); assert_hvf_ok(ret); + b = hvf_sysreg_read_cp(cpu, "VTimer", SYSREG_CNTV_CVAL_EL0, &val); + assert(b); + ret = hv_vcpu_set_sys_reg(cpu->accel->fd, HV_SYS_REG_CNTV_CVAL_EL0, val); + assert_hvf_ok(ret); + + b = hvf_sysreg_read_cp(cpu, "VTimer", SYSREG_CNTV_CTL_EL0, &val); + assert(b); + ret = hv_vcpu_set_sys_reg(cpu->accel->fd, HV_SYS_REG_CNTV_CTL_EL0, val); + assert_hvf_ok(ret); + return 0; } -- 2.43.0 From: Alex Bennée <alex.bennee@linaro.org> This is updated as Arm architectural features are added so we should catch changes to the docs as well. Signed-off-by: Alex Bennée <alex.bennee@linaro.org> Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Message-id: 20260127145928.3073826-1-alex.bennee@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- MAINTAINERS | 1 + 1 file changed, 1 insertion(+) diff --git a/MAINTAINERS b/MAINTAINERS index XXXXXXX..XXXXXXX 100644 --- a/MAINTAINERS +++ b/MAINTAINERS @@ -XXX,XX +XXX,XX @@ F: hw/cpu/a*mpcore.c F: include/hw/cpu/a*mpcore.h F: docs/system/target-arm.rst F: docs/system/arm/cpu-features.rst +F: docs/system/arm/emulation.rst F: gdb-xml/arm*.xml F: gdb-xml/aarch64*.xml -- 2.43.0 From: Alex Bennée <alex.bennee@linaro.org> It looks like the features were renamed to include the levels at some point. To make it easier to match features up to the Arm ARM update to use the full name. Signed-off-by: Alex Bennée <alex.bennee@linaro.org> Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org Message-id: 20260127145555.3070590-1-alex.bennee@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- docs/system/arm/emulation.rst | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/docs/system/arm/emulation.rst b/docs/system/arm/emulation.rst index XXXXXXX..XXXXXXX 100644 --- a/docs/system/arm/emulation.rst +++ b/docs/system/arm/emulation.rst @@ -XXX,XX +XXX,XX @@ the following architecture extensions: - FEAT_ASID16 (16 bit ASID) - FEAT_ASID2 (Concurrent use of two ASIDs) - FEAT_ATS1A (Address Translation operations that ignore stage 1 permissions) -- FEAT_BBM at level 2 (Translation table break-before-make levels) +- FEAT_BBML1 (Translation table break-before-make level 1) +- FEAT_BBML2 (Translation table break-before-make level 2) - FEAT_BF16 (AArch64 BFloat16 instructions) - FEAT_BTI (Branch Target Identification) - FEAT_CCIDX (Extended cache index) -- 2.43.0 From: Cornelia Huck <cohuck@redhat.com> Most accesses to cpu->dcz_blocksize really care about DCZID_EL0.BS (i.e. the part of the register that does not change at different EL.) Wean them off directly dealing with cpu->dcz_blocksize so that we can switch to handling DCZID_EL0 differently in a followup patch. Signed-off-by: Cornelia Huck <cohuck@redhat.com> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Sebastian Ott <sebott@redhat.com> Message-id: 20260105154119.59853-2-cohuck@redhat.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/cpu.c | 2 +- target/arm/cpu.h | 10 ++++++++++ target/arm/cpu64.c | 4 ++-- target/arm/helper.c | 1 + target/arm/tcg/cpu64.c | 22 +++++++++++----------- target/arm/tcg/helper-a64.c | 2 +- target/arm/tcg/mte_helper.c | 4 ++-- target/arm/tcg/translate-a64.c | 2 +- 8 files changed, 29 insertions(+), 18 deletions(-) diff --git a/target/arm/cpu.c b/target/arm/cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/cpu.c +++ b/target/arm/cpu.c @@ -XXX,XX +XXX,XX @@ static void arm_cpu_realizefn(DeviceState *dev, Error **errp) #endif if (tcg_enabled()) { - int dcz_blocklen = 4 << cpu->dcz_blocksize; + int dcz_blocklen = 4 << get_dczid_bs(cpu); /* * We only support DCZ blocklen that fits on one page. diff --git a/target/arm/cpu.h b/target/arm/cpu.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/cpu.h +++ b/target/arm/cpu.h @@ -XXX,XX +XXX,XX @@ struct ARMCPUClass { ResettablePhases parent_phases; }; +static inline uint8_t get_dczid_bs(ARMCPU *cpu) +{ + return cpu->dcz_blocksize; +} + +static inline void set_dczid_bs(ARMCPU *cpu, uint8_t bs) +{ + cpu->dcz_blocksize = bs; +} + /* Callback functions for the generic timer's timers. */ void arm_gt_ptimer_cb(void *opaque); void arm_gt_vtimer_cb(void *opaque); diff --git a/target/arm/cpu64.c b/target/arm/cpu64.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/cpu64.c +++ b/target/arm/cpu64.c @@ -XXX,XX +XXX,XX @@ static void aarch64_a57_initfn(Object *obj) cpu->ccsidr[1] = make_ccsidr(CCSIDR_FORMAT_LEGACY, 3, 64, 48 * KiB, 2); /* 2048KB L2 cache */ cpu->ccsidr[2] = make_ccsidr(CCSIDR_FORMAT_LEGACY, 16, 64, 2 * MiB, 7); - cpu->dcz_blocksize = 4; /* 64 bytes */ + set_dczid_bs(cpu, 4); /* 64 bytes */ cpu->gic_num_lrs = 4; cpu->gic_vpribits = 5; cpu->gic_vprebits = 5; @@ -XXX,XX +XXX,XX @@ static void aarch64_a53_initfn(Object *obj) cpu->ccsidr[1] = make_ccsidr(CCSIDR_FORMAT_LEGACY, 1, 64, 32 * KiB, 2); /* 1024KB L2 cache */ cpu->ccsidr[2] = make_ccsidr(CCSIDR_FORMAT_LEGACY, 16, 64, 1 * MiB, 7); - cpu->dcz_blocksize = 4; /* 64 bytes */ + set_dczid_bs(cpu, 4); /* 64 bytes */ cpu->gic_num_lrs = 4; cpu->gic_vpribits = 5; cpu->gic_vprebits = 5; diff --git a/target/arm/helper.c b/target/arm/helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/helper.c +++ b/target/arm/helper.c @@ -XXX,XX +XXX,XX @@ static uint64_t aa64_dczid_read(CPUARMState *env, const ARMCPRegInfo *ri) if (aa64_zva_access(env, NULL, false) == CP_ACCESS_OK) { dzp_bit = 0; } + return cpu->dcz_blocksize | dzp_bit; } diff --git a/target/arm/tcg/cpu64.c b/target/arm/tcg/cpu64.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/cpu64.c +++ b/target/arm/tcg/cpu64.c @@ -XXX,XX +XXX,XX @@ static void aarch64_a35_initfn(Object *obj) SET_IDREG(isar, ID_AA64MMFR0, 0x00101122); SET_IDREG(isar, ID_AA64MMFR1, 0); SET_IDREG(isar, CLIDR, 0x0a200023); - cpu->dcz_blocksize = 4; + set_dczid_bs(cpu, 4); /* From B2.4 AArch64 Virtual Memory control registers */ cpu->reset_sctlr = 0x00c50838; @@ -XXX,XX +XXX,XX @@ static void aarch64_a55_initfn(Object *obj) /* Ordered by B2.4 AArch64 registers by functional group */ SET_IDREG(isar, CLIDR, 0x82000023); cpu->ctr = 0x84448004; /* L1Ip = VIPT */ - cpu->dcz_blocksize = 4; /* 64 bytes */ + set_dczid_bs(cpu, 4); /* 64 bytes */ SET_IDREG(isar, ID_AA64DFR0, 0x0000000010305408ull); SET_IDREG(isar, ID_AA64ISAR0, 0x0000100010211120ull); SET_IDREG(isar, ID_AA64ISAR1, 0x0000000000100001ull); @@ -XXX,XX +XXX,XX @@ static void aarch64_a72_initfn(Object *obj) cpu->ccsidr[1] = make_ccsidr(CCSIDR_FORMAT_LEGACY, 3, 64, 48 * KiB, 2); /* 1MB L2 cache */ cpu->ccsidr[2] = make_ccsidr(CCSIDR_FORMAT_LEGACY, 16, 64, 1 * MiB, 7); - cpu->dcz_blocksize = 4; /* 64 bytes */ + set_dczid_bs(cpu, 4); /* 64 bytes */ cpu->gic_num_lrs = 4; cpu->gic_vpribits = 5; cpu->gic_vprebits = 5; @@ -XXX,XX +XXX,XX @@ static void aarch64_a76_initfn(Object *obj) /* Ordered by B2.4 AArch64 registers by functional group */ SET_IDREG(isar, CLIDR, 0x82000023); cpu->ctr = 0x8444C004; - cpu->dcz_blocksize = 4; + set_dczid_bs(cpu, 4); SET_IDREG(isar, ID_AA64DFR0, 0x0000000010305408ull); SET_IDREG(isar, ID_AA64ISAR0, 0x0000100010211120ull); SET_IDREG(isar, ID_AA64ISAR1, 0x0000000000100001ull); @@ -XXX,XX +XXX,XX @@ static void aarch64_a78ae_initfn(Object *obj) /* Ordered by 3.2.4 AArch64 registers by functional group */ SET_IDREG(isar, CLIDR, 0x82000023); cpu->ctr = 0x9444c004; - cpu->dcz_blocksize = 4; + set_dczid_bs(cpu, 4); SET_IDREG(isar, ID_AA64DFR0, 0x0000000110305408ull); SET_IDREG(isar, ID_AA64ISAR0, 0x0010100010211120ull); SET_IDREG(isar, ID_AA64ISAR1, 0x0000000001200031ull); @@ -XXX,XX +XXX,XX @@ static void aarch64_a64fx_initfn(Object *obj) cpu->ccsidr[1] = make_ccsidr(CCSIDR_FORMAT_LEGACY, 4, 256, 64 * KiB, 2); /* 8MB L2 cache */ cpu->ccsidr[2] = make_ccsidr(CCSIDR_FORMAT_LEGACY, 16, 256, 8 * MiB, 7); - cpu->dcz_blocksize = 6; /* 256 bytes */ + set_dczid_bs(cpu, 6); /* 256 bytes */ cpu->gic_num_lrs = 4; cpu->gic_vpribits = 5; cpu->gic_vprebits = 5; @@ -XXX,XX +XXX,XX @@ static void aarch64_neoverse_n1_initfn(Object *obj) /* Ordered by B2.4 AArch64 registers by functional group */ SET_IDREG(isar, CLIDR, 0x82000023); cpu->ctr = 0x8444c004; - cpu->dcz_blocksize = 4; + set_dczid_bs(cpu, 4); SET_IDREG(isar, ID_AA64DFR0, 0x0000000110305408ull); SET_IDREG(isar, ID_AA64ISAR0, 0x0000100010211120ull); SET_IDREG(isar, ID_AA64ISAR1, 0x0000000000100001ull); @@ -XXX,XX +XXX,XX @@ static void aarch64_neoverse_v1_initfn(Object *obj) /* Ordered by 3.2.4 AArch64 registers by functional group */ SET_IDREG(isar, CLIDR, 0x82000023); cpu->ctr = 0xb444c004; /* With DIC and IDC set */ - cpu->dcz_blocksize = 4; + set_dczid_bs(cpu, 4); SET_IDREG(isar, ID_AA64AFR0, 0x00000000); SET_IDREG(isar, ID_AA64AFR1, 0x00000000); SET_IDREG(isar, ID_AA64DFR0, 0x000001f210305519ull); @@ -XXX,XX +XXX,XX @@ static void aarch64_a710_initfn(Object *obj) SET_IDREG(isar, CLIDR, 0x0000001482000023ull); cpu->gm_blocksize = 4; cpu->ctr = 0x000000049444c004ull; - cpu->dcz_blocksize = 4; + set_dczid_bs(cpu, 4); /* TODO FEAT_MPAM: mpamidr_el1 = 0x0000_0001_0006_003f */ /* Section B.5.2: PMCR_EL0 */ @@ -XXX,XX +XXX,XX @@ static void aarch64_neoverse_n2_initfn(Object *obj) SET_IDREG(isar, CLIDR, 0x0000001482000023ull); cpu->gm_blocksize = 4; cpu->ctr = 0x00000004b444c004ull; - cpu->dcz_blocksize = 4; + set_dczid_bs(cpu, 4); /* TODO FEAT_MPAM: mpamidr_el1 = 0x0000_0001_001e_01ff */ /* Section B.7.2: PMCR_EL0 */ @@ -XXX,XX +XXX,XX @@ void aarch64_max_tcg_initfn(Object *obj) * blocksize since we don't have to follow what the hardware does. */ cpu->ctr = 0x80038003; /* 32 byte I and D cacheline size, VIPT icache */ - cpu->dcz_blocksize = 7; /* 512 bytes */ + set_dczid_bs(cpu, 7); /* 512 bytes */ #endif cpu->gm_blocksize = 6; /* 256 bytes */ diff --git a/target/arm/tcg/helper-a64.c b/target/arm/tcg/helper-a64.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-a64.c +++ b/target/arm/tcg/helper-a64.c @@ -XXX,XX +XXX,XX @@ void HELPER(dc_zva)(CPUARMState *env, uint64_t vaddr_in) * (which matches the usual QEMU behaviour of not implementing either * alignment faults or any memory attribute handling). */ - int blocklen = 4 << env_archcpu(env)->dcz_blocksize; + int blocklen = 4 << get_dczid_bs(env_archcpu(env)); uint64_t vaddr = vaddr_in & ~(blocklen - 1); int mmu_idx = arm_env_mmu_index(env); void *mem; diff --git a/target/arm/tcg/mte_helper.c b/target/arm/tcg/mte_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/mte_helper.c +++ b/target/arm/tcg/mte_helper.c @@ -XXX,XX +XXX,XX @@ void HELPER(stzgm_tags)(CPUARMState *env, uint64_t ptr, uint64_t val) * i.e. 32 bytes, which is an unreasonably small dcz anyway, * to make sure that we can access one complete tag byte here. */ - log2_dcz_bytes = env_archcpu(env)->dcz_blocksize + 2; + log2_dcz_bytes = get_dczid_bs(env_archcpu(env)) + 2; log2_tag_bytes = log2_dcz_bytes - (LOG2_TAG_GRANULE + 1); dcz_bytes = (intptr_t)1 << log2_dcz_bytes; tag_bytes = (intptr_t)1 << log2_tag_bytes; @@ -XXX,XX +XXX,XX @@ uint64_t HELPER(mte_check_zva)(CPUARMState *env, uint32_t desc, uint64_t ptr) * i.e. 32 bytes, which is an unreasonably small dcz anyway, to make * sure that we can access one complete tag byte here. */ - log2_dcz_bytes = env_archcpu(env)->dcz_blocksize + 2; + log2_dcz_bytes = get_dczid_bs(env_archcpu(env)) + 2; log2_tag_bytes = log2_dcz_bytes - (LOG2_TAG_GRANULE + 1); dcz_bytes = (intptr_t)1 << log2_dcz_bytes; tag_bytes = (intptr_t)1 << log2_tag_bytes; diff --git a/target/arm/tcg/translate-a64.c b/target/arm/tcg/translate-a64.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate-a64.c +++ b/target/arm/tcg/translate-a64.c @@ -XXX,XX +XXX,XX @@ static void aarch64_tr_init_disas_context(DisasContextBase *dcbase, dc->vec_stride = 0; dc->cp_regs = arm_cpu->cp_regs; dc->features = env->features; - dc->dcz_blocksize = arm_cpu->dcz_blocksize; + dc->dcz_blocksize = get_dczid_bs(arm_cpu); dc->gm_blocksize = arm_cpu->gm_blocksize; #ifdef CONFIG_USER_ONLY -- 2.43.0 From: Cornelia Huck <cohuck@redhat.com> Continue moving ID registers to the idregs array, so that we eventually can switch to an autogenerated cpu-sysregs.h.inc. This requires a bit of care, since we still have to handle the EL specific part (DCZID_EL0.DZP). The value previously saved in cpu->dcz_blocksize is now kept in DCZID_EL.BS (transparent to callers using the wrappers.) KVM currently does not support DCZID_EL0 via ONE_REG, assert that we're not trying to do anything with it until it does. Signed-off-by: Cornelia Huck <cohuck@redhat.com> Reviewed-by: Sebastian Ott <sebott@redhat.com> Message-id: 20260105154119.59853-3-cohuck@redhat.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/cpu-sysregs.h.inc | 1 + target/arm/cpu.h | 8 ++++---- target/arm/helper.c | 4 +++- target/arm/tcg/translate.h | 2 +- 4 files changed, 9 insertions(+), 6 deletions(-) diff --git a/target/arm/cpu-sysregs.h.inc b/target/arm/cpu-sysregs.h.inc index XXXXXXX..XXXXXXX 100644 --- a/target/arm/cpu-sysregs.h.inc +++ b/target/arm/cpu-sysregs.h.inc @@ -XXX,XX +XXX,XX @@ DEF(ID_MMFR5_EL1, 3, 0, 0, 3, 6) DEF(CLIDR_EL1, 3, 1, 0, 0, 1) DEF(ID_AA64ZFR0_EL1, 3, 0, 0, 4, 4) DEF(CTR_EL0, 3, 3, 0, 0, 1) +DEF(DCZID_EL0, 3, 3, 0, 0, 7) diff --git a/target/arm/cpu.h b/target/arm/cpu.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/cpu.h +++ b/target/arm/cpu.h @@ -XXX,XX +XXX,XX @@ struct ArchCPU { bool prop_pauth_qarma5; bool prop_lpa2; - /* DCZ blocksize, in log_2(words), ie low 4 bits of DCZID_EL0 */ - uint8_t dcz_blocksize; /* GM blocksize, in log_2(words), ie low 4 bits of GMID_EL0 */ uint8_t gm_blocksize; @@ -XXX,XX +XXX,XX @@ struct ARMCPUClass { static inline uint8_t get_dczid_bs(ARMCPU *cpu) { - return cpu->dcz_blocksize; + return extract64(cpu->isar.idregs[DCZID_EL0_IDX], 0, 4); } static inline void set_dczid_bs(ARMCPU *cpu, uint8_t bs) { - cpu->dcz_blocksize = bs; + /* keep dzp unchanged */ + cpu->isar.idregs[DCZID_EL0_IDX] = + deposit64(cpu->isar.idregs[DCZID_EL0_IDX], 0, 4, bs); } /* Callback functions for the generic timer's timers. */ diff --git a/target/arm/helper.c b/target/arm/helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/helper.c +++ b/target/arm/helper.c @@ -XXX,XX +XXX,XX @@ static uint64_t aa64_dczid_read(CPUARMState *env, const ARMCPRegInfo *ri) ARMCPU *cpu = env_archcpu(env); int dzp_bit = 1 << 4; + assert(!kvm_enabled()); + /* DZP indicates whether DC ZVA access is allowed */ if (aa64_zva_access(env, NULL, false) == CP_ACCESS_OK) { dzp_bit = 0; } - return cpu->dcz_blocksize | dzp_bit; + return cpu->isar.idregs[DCZID_EL0_IDX] | dzp_bit; } static CPAccessResult sp_el0_access(CPUARMState *env, const ARMCPRegInfo *ri, diff --git a/target/arm/tcg/translate.h b/target/arm/tcg/translate.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate.h +++ b/target/arm/tcg/translate.h @@ -XXX,XX +XXX,XX @@ typedef struct DisasContext { * < 0, set by the current instruction. */ int8_t btype; - /* A copy of cpu->dcz_blocksize. */ + /* A copy of DCZID_EL0.BS. */ uint8_t dcz_blocksize; /* A copy of cpu->gm_blocksize. */ uint8_t gm_blocksize; -- 2.43.0