Mostly my decodetree stuff, but also some patches for various smaller bugs/features from others. thanks -- PMM The following changes since commit 53550e81e2cafe7c03a39526b95cd21b5194d9b1: Merge remote-tracking branch 'remotes/berrange/tags/qcrypto-next-pull-request' into staging (2020-06-15 16:36:34 +0100) are available in the Git repository at: https://git.linaro.org/people/pmaydell/qemu-arm.git tags/pull-target-arm-20200616 for you to fetch changes up to 64b397417a26509bcdff44ab94356a35c7901c79: hw: arm: Set vendor property for IMX SDHCI emulations (2020-06-16 10:32:29 +0100) ---------------------------------------------------------------- * hw: arm: Set vendor property for IMX SDHCI emulations * sd: sdhci: Implement basic vendor specific register support * hw/net/imx_fec: Convert debug fprintf() to trace events * target/arm/cpu: adjust virtual time for all KVM arm cpus * Implement configurable descriptor size in ftgmac100 * hw/misc/imx6ul_ccm: Implement non writable bits in CCM registers * target/arm: More Neon decodetree conversion work ---------------------------------------------------------------- Erik Smit (1): Implement configurable descriptor size in ftgmac100 Guenter Roeck (2): sd: sdhci: Implement basic vendor specific register support hw: arm: Set vendor property for IMX SDHCI emulations Jean-Christophe Dubois (2): hw/misc/imx6ul_ccm: Implement non writable bits in CCM registers hw/net/imx_fec: Convert debug fprintf() to trace events Peter Maydell (17): target/arm: Fix missing temp frees in do_vshll_2sh target/arm: Convert Neon 3-reg-diff prewidening ops to decodetree target/arm: Convert Neon 3-reg-diff narrowing ops to decodetree target/arm: Convert Neon 3-reg-diff VABAL, VABDL to decodetree target/arm: Convert Neon 3-reg-diff long multiplies target/arm: Convert Neon 3-reg-diff saturating doubling multiplies target/arm: Convert Neon 3-reg-diff polynomial VMULL target/arm: Add 'static' and 'const' annotations to VSHLL function arrays target/arm: Add missing TCG temp free in do_2shift_env_64() target/arm: Convert Neon 2-reg-scalar integer multiplies to decodetree target/arm: Convert Neon 2-reg-scalar float multiplies to decodetree target/arm: Convert Neon 2-reg-scalar VQDMULH, VQRDMULH to decodetree target/arm: Convert Neon 2-reg-scalar VQRDMLAH, VQRDMLSH to decodetree target/arm: Convert Neon 2-reg-scalar long multiplies to decodetree target/arm: Convert Neon VEXT to decodetree target/arm: Convert Neon VTBL, VTBX to decodetree target/arm: Convert Neon VDUP (scalar) to decodetree fangying (1): target/arm/cpu: adjust virtual time for all KVM arm cpus hw/sd/sdhci-internal.h | 5 + include/hw/sd/sdhci.h | 5 + target/arm/translate.h | 1 + target/arm/neon-dp.decode | 130 +++++ hw/arm/fsl-imx25.c | 6 + hw/arm/fsl-imx6.c | 6 + hw/arm/fsl-imx6ul.c | 2 + hw/arm/fsl-imx7.c | 2 + hw/misc/imx6ul_ccm.c | 76 ++- hw/net/ftgmac100.c | 26 +- hw/net/imx_fec.c | 106 ++-- hw/sd/sdhci.c | 18 +- target/arm/cpu.c | 6 +- target/arm/cpu64.c | 1 - target/arm/kvm.c | 21 +- target/arm/translate-neon.inc.c | 1148 ++++++++++++++++++++++++++++++++++++++- target/arm/translate.c | 684 +---------------------- hw/net/trace-events | 18 + 18 files changed, 1495 insertions(+), 766 deletions(-) The widenfn() in do_vshll_2sh() does not free the input 32-bit TCGv, so we need to do this in the calling code. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org> --- target/arm/translate-neon.inc.c | 2 ++ 1 file changed, 2 insertions(+) diff --git a/target/arm/translate-neon.inc.c b/target/arm/translate-neon.inc.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-neon.inc.c +++ b/target/arm/translate-neon.inc.c @@ -XXX,XX +XXX,XX @@ static bool do_vshll_2sh(DisasContext *s, arg_2reg_shift *a, tmp = tcg_temp_new_i64(); widenfn(tmp, rm0); + tcg_temp_free_i32(rm0); if (a->shift != 0) { tcg_gen_shli_i64(tmp, tmp, a->shift); tcg_gen_andi_i64(tmp, tmp, ~widen_mask); @@ -XXX,XX +XXX,XX @@ static bool do_vshll_2sh(DisasContext *s, arg_2reg_shift *a, neon_store_reg64(tmp, a->vd); widenfn(tmp, rm1); + tcg_temp_free_i32(rm1); if (a->shift != 0) { tcg_gen_shli_i64(tmp, tmp, a->shift); tcg_gen_andi_i64(tmp, tmp, ~widen_mask); -- 2.20.1 Convert the "pre-widening" insns VADDL, VSUBL, VADDW and VSUBW in the Neon 3-registers-different-lengths group to decodetree. These insns work by widening one or both inputs to double their size, performing an add or subtract at the doubled size and then storing the double-size result. As usual, rather than copying the loop of the original decoder (which needs awkward code to avoid problems when source and destination registers overlap) we just unroll the two passes. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/neon-dp.decode | 43 +++++++++++++ target/arm/translate-neon.inc.c | 104 ++++++++++++++++++++++++++++++++ target/arm/translate.c | 16 ++--- 3 files changed, 151 insertions(+), 12 deletions(-) diff --git a/target/arm/neon-dp.decode b/target/arm/neon-dp.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/neon-dp.decode +++ b/target/arm/neon-dp.decode @@ -XXX,XX +XXX,XX @@ VCVT_FU_2sh 1111 001 1 1 . ...... .... 1111 0 . . 1 .... @2reg_vcvt # So we have a single decode line and check the cmode/op in the # trans function. Vimm_1r 1111 001 . 1 . 000 ... .... cmode:4 0 . op:1 1 .... @1reg_imm + +###################################################################### +# Within the "two registers, or three registers of different lengths" +# grouping ([23,4]=0b10), bits [21:20] are either part of the opcode +# decode: 0b11 for VEXT, two-reg-misc, VTBL, and duplicate-scalar; +# or they are a size field for the three-reg-different-lengths and +# two-reg-and-scalar insn groups (where size cannot be 0b11). This +# is slightly awkward for decodetree: we handle it with this +# non-exclusive group which contains within it two exclusive groups: +# one for the size=0b11 patterns, and one for the size-not-0b11 +# patterns. This allows us to check that none of the insns within +# each subgroup accidentally overlap each other. Note that all the +# trans functions for the size-not-0b11 patterns must check and +# return false for size==3. +###################################################################### +{ + # 0b11 subgroup will go here + + # Subgroup for size != 0b11 + [ + ################################################################## + # 3-reg-different-length grouping: + # 1111 001 U 1 D sz!=11 Vn:4 Vd:4 opc:4 N 0 M 0 Vm:4 + ################################################################## + + &3diff vm vn vd size + + @3diff .... ... . . . size:2 .... .... .... . . . . .... \ + &3diff vm=%vm_dp vn=%vn_dp vd=%vd_dp + + VADDL_S_3d 1111 001 0 1 . .. .... .... 0000 . 0 . 0 .... @3diff + VADDL_U_3d 1111 001 1 1 . .. .... .... 0000 . 0 . 0 .... @3diff + + VADDW_S_3d 1111 001 0 1 . .. .... .... 0001 . 0 . 0 .... @3diff + VADDW_U_3d 1111 001 1 1 . .. .... .... 0001 . 0 . 0 .... @3diff + + VSUBL_S_3d 1111 001 0 1 . .. .... .... 0010 . 0 . 0 .... @3diff + VSUBL_U_3d 1111 001 1 1 . .. .... .... 0010 . 0 . 0 .... @3diff + + VSUBW_S_3d 1111 001 0 1 . .. .... .... 0011 . 0 . 0 .... @3diff + VSUBW_U_3d 1111 001 1 1 . .. .... .... 0011 . 0 . 0 .... @3diff + ] +} diff --git a/target/arm/translate-neon.inc.c b/target/arm/translate-neon.inc.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-neon.inc.c +++ b/target/arm/translate-neon.inc.c @@ -XXX,XX +XXX,XX @@ static bool trans_Vimm_1r(DisasContext *s, arg_1reg_imm *a) } return do_1reg_imm(s, a, fn); } + +static bool do_prewiden_3d(DisasContext *s, arg_3diff *a, + NeonGenWidenFn *widenfn, + NeonGenTwo64OpFn *opfn, + bool src1_wide) +{ + /* 3-regs different lengths, prewidening case (VADDL/VSUBL/VAADW/VSUBW) */ + TCGv_i64 rn0_64, rn1_64, rm_64; + TCGv_i32 rm; + + if (!arm_dc_feature(s, ARM_FEATURE_NEON)) { + return false; + } + + /* UNDEF accesses to D16-D31 if they don't exist. */ + if (!dc_isar_feature(aa32_simd_r32, s) && + ((a->vd | a->vn | a->vm) & 0x10)) { + return false; + } + + if (!widenfn || !opfn) { + /* size == 3 case, which is an entirely different insn group */ + return false; + } + + if ((a->vd & 1) || (src1_wide && (a->vn & 1))) { + return false; + } + + if (!vfp_access_check(s)) { + return true; + } + + rn0_64 = tcg_temp_new_i64(); + rn1_64 = tcg_temp_new_i64(); + rm_64 = tcg_temp_new_i64(); + + if (src1_wide) { + neon_load_reg64(rn0_64, a->vn); + } else { + TCGv_i32 tmp = neon_load_reg(a->vn, 0); + widenfn(rn0_64, tmp); + tcg_temp_free_i32(tmp); + } + rm = neon_load_reg(a->vm, 0); + + widenfn(rm_64, rm); + tcg_temp_free_i32(rm); + opfn(rn0_64, rn0_64, rm_64); + + /* + * Load second pass inputs before storing the first pass result, to + * avoid incorrect results if a narrow input overlaps with the result. + */ + if (src1_wide) { + neon_load_reg64(rn1_64, a->vn + 1); + } else { + TCGv_i32 tmp = neon_load_reg(a->vn, 1); + widenfn(rn1_64, tmp); + tcg_temp_free_i32(tmp); + } + rm = neon_load_reg(a->vm, 1); + + neon_store_reg64(rn0_64, a->vd); + + widenfn(rm_64, rm); + tcg_temp_free_i32(rm); + opfn(rn1_64, rn1_64, rm_64); + neon_store_reg64(rn1_64, a->vd + 1); + + tcg_temp_free_i64(rn0_64); + tcg_temp_free_i64(rn1_64); + tcg_temp_free_i64(rm_64); + + return true; +} + +#define DO_PREWIDEN(INSN, S, EXT, OP, SRC1WIDE) \ + static bool trans_##INSN##_3d(DisasContext *s, arg_3diff *a) \ + { \ + static NeonGenWidenFn * const widenfn[] = { \ + gen_helper_neon_widen_##S##8, \ + gen_helper_neon_widen_##S##16, \ + tcg_gen_##EXT##_i32_i64, \ + NULL, \ + }; \ + static NeonGenTwo64OpFn * const addfn[] = { \ + gen_helper_neon_##OP##l_u16, \ + gen_helper_neon_##OP##l_u32, \ + tcg_gen_##OP##_i64, \ + NULL, \ + }; \ + return do_prewiden_3d(s, a, widenfn[a->size], \ + addfn[a->size], SRC1WIDE); \ + } + +DO_PREWIDEN(VADDL_S, s, ext, add, false) +DO_PREWIDEN(VADDL_U, u, extu, add, false) +DO_PREWIDEN(VSUBL_S, s, ext, sub, false) +DO_PREWIDEN(VSUBL_U, u, extu, sub, false) +DO_PREWIDEN(VADDW_S, s, ext, add, true) +DO_PREWIDEN(VADDW_U, u, extu, add, true) +DO_PREWIDEN(VSUBW_S, s, ext, sub, true) +DO_PREWIDEN(VSUBW_U, u, extu, sub, true) diff --git a/target/arm/translate.c b/target/arm/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate.c +++ b/target/arm/translate.c @@ -XXX,XX +XXX,XX @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn) /* Three registers of different lengths. */ int src1_wide; int src2_wide; - int prewiden; /* undefreq: bit 0 : UNDEF if size == 0 * bit 1 : UNDEF if size == 1 * bit 2 : UNDEF if size == 2 @@ -XXX,XX +XXX,XX @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn) int undefreq; /* prewiden, src1_wide, src2_wide, undefreq */ static const int neon_3reg_wide[16][4] = { - {1, 0, 0, 0}, /* VADDL */ - {1, 1, 0, 0}, /* VADDW */ - {1, 0, 0, 0}, /* VSUBL */ - {1, 1, 0, 0}, /* VSUBW */ + {0, 0, 0, 7}, /* VADDL: handled by decodetree */ + {0, 0, 0, 7}, /* VADDW: handled by decodetree */ + {0, 0, 0, 7}, /* VSUBL: handled by decodetree */ + {0, 0, 0, 7}, /* VSUBW: handled by decodetree */ {0, 1, 1, 0}, /* VADDHN */ {0, 0, 0, 0}, /* VABAL */ {0, 1, 1, 0}, /* VSUBHN */ @@ -XXX,XX +XXX,XX @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn) {0, 0, 0, 7}, /* Reserved: always UNDEF */ }; - prewiden = neon_3reg_wide[op][0]; src1_wide = neon_3reg_wide[op][1]; src2_wide = neon_3reg_wide[op][2]; undefreq = neon_3reg_wide[op][3]; @@ -XXX,XX +XXX,XX @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn) } else { tmp = neon_load_reg(rn, pass); } - if (prewiden) { - gen_neon_widen(cpu_V0, tmp, size, u); - } } if (src2_wide) { neon_load_reg64(cpu_V1, rm + pass); @@ -XXX,XX +XXX,XX @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn) } else { tmp2 = neon_load_reg(rm, pass); } - if (prewiden) { - gen_neon_widen(cpu_V1, tmp2, size, u); - } } switch (op) { case 0: case 1: case 4: /* VADDL, VADDW, VADDHN, VRADDHN */ -- 2.20.1 Convert the narrow-to-high-half insns VADDHN, VSUBHN, VRADDHN, VRSUBHN in the Neon 3-registers-different-lengths group to decodetree. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/neon-dp.decode | 6 +++ target/arm/translate-neon.inc.c | 87 +++++++++++++++++++++++++++++++ target/arm/translate.c | 91 ++++----------------------------- 3 files changed, 104 insertions(+), 80 deletions(-) diff --git a/target/arm/neon-dp.decode b/target/arm/neon-dp.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/neon-dp.decode +++ b/target/arm/neon-dp.decode @@ -XXX,XX +XXX,XX @@ Vimm_1r 1111 001 . 1 . 000 ... .... cmode:4 0 . op:1 1 .... @1reg_imm VSUBW_S_3d 1111 001 0 1 . .. .... .... 0011 . 0 . 0 .... @3diff VSUBW_U_3d 1111 001 1 1 . .. .... .... 0011 . 0 . 0 .... @3diff + + VADDHN_3d 1111 001 0 1 . .. .... .... 0100 . 0 . 0 .... @3diff + VRADDHN_3d 1111 001 1 1 . .. .... .... 0100 . 0 . 0 .... @3diff + + VSUBHN_3d 1111 001 0 1 . .. .... .... 0110 . 0 . 0 .... @3diff + VRSUBHN_3d 1111 001 1 1 . .. .... .... 0110 . 0 . 0 .... @3diff ] } diff --git a/target/arm/translate-neon.inc.c b/target/arm/translate-neon.inc.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-neon.inc.c +++ b/target/arm/translate-neon.inc.c @@ -XXX,XX +XXX,XX @@ DO_PREWIDEN(VADDW_S, s, ext, add, true) DO_PREWIDEN(VADDW_U, u, extu, add, true) DO_PREWIDEN(VSUBW_S, s, ext, sub, true) DO_PREWIDEN(VSUBW_U, u, extu, sub, true) + +static bool do_narrow_3d(DisasContext *s, arg_3diff *a, + NeonGenTwo64OpFn *opfn, NeonGenNarrowFn *narrowfn) +{ + /* 3-regs different lengths, narrowing (VADDHN/VSUBHN/VRADDHN/VRSUBHN) */ + TCGv_i64 rn_64, rm_64; + TCGv_i32 rd0, rd1; + + if (!arm_dc_feature(s, ARM_FEATURE_NEON)) { + return false; + } + + /* UNDEF accesses to D16-D31 if they don't exist. */ + if (!dc_isar_feature(aa32_simd_r32, s) && + ((a->vd | a->vn | a->vm) & 0x10)) { + return false; + } + + if (!opfn || !narrowfn) { + /* size == 3 case, which is an entirely different insn group */ + return false; + } + + if ((a->vn | a->vm) & 1) { + return false; + } + + if (!vfp_access_check(s)) { + return true; + } + + rn_64 = tcg_temp_new_i64(); + rm_64 = tcg_temp_new_i64(); + rd0 = tcg_temp_new_i32(); + rd1 = tcg_temp_new_i32(); + + neon_load_reg64(rn_64, a->vn); + neon_load_reg64(rm_64, a->vm); + + opfn(rn_64, rn_64, rm_64); + + narrowfn(rd0, rn_64); + + neon_load_reg64(rn_64, a->vn + 1); + neon_load_reg64(rm_64, a->vm + 1); + + opfn(rn_64, rn_64, rm_64); + + narrowfn(rd1, rn_64); + + neon_store_reg(a->vd, 0, rd0); + neon_store_reg(a->vd, 1, rd1); + + tcg_temp_free_i64(rn_64); + tcg_temp_free_i64(rm_64); + + return true; +} + +#define DO_NARROW_3D(INSN, OP, NARROWTYPE, EXTOP) \ + static bool trans_##INSN##_3d(DisasContext *s, arg_3diff *a) \ + { \ + static NeonGenTwo64OpFn * const addfn[] = { \ + gen_helper_neon_##OP##l_u16, \ + gen_helper_neon_##OP##l_u32, \ + tcg_gen_##OP##_i64, \ + NULL, \ + }; \ + static NeonGenNarrowFn * const narrowfn[] = { \ + gen_helper_neon_##NARROWTYPE##_high_u8, \ + gen_helper_neon_##NARROWTYPE##_high_u16, \ + EXTOP, \ + NULL, \ + }; \ + return do_narrow_3d(s, a, addfn[a->size], narrowfn[a->size]); \ + } + +static void gen_narrow_round_high_u32(TCGv_i32 rd, TCGv_i64 rn) +{ + tcg_gen_addi_i64(rn, rn, 1u << 31); + tcg_gen_extrh_i64_i32(rd, rn); +} + +DO_NARROW_3D(VADDHN, add, narrow, tcg_gen_extrh_i64_i32) +DO_NARROW_3D(VSUBHN, sub, narrow, tcg_gen_extrh_i64_i32) +DO_NARROW_3D(VRADDHN, add, narrow_round, gen_narrow_round_high_u32) +DO_NARROW_3D(VRSUBHN, sub, narrow_round, gen_narrow_round_high_u32) diff --git a/target/arm/translate.c b/target/arm/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate.c +++ b/target/arm/translate.c @@ -XXX,XX +XXX,XX @@ static inline void gen_neon_addl(int size) } } -static inline void gen_neon_subl(int size) -{ - switch (size) { - case 0: gen_helper_neon_subl_u16(CPU_V001); break; - case 1: gen_helper_neon_subl_u32(CPU_V001); break; - case 2: tcg_gen_sub_i64(CPU_V001); break; - default: abort(); - } -} - static inline void gen_neon_negl(TCGv_i64 var, int size) { switch (size) { @@ -XXX,XX +XXX,XX @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn) op = (insn >> 8) & 0xf; if ((insn & (1 << 6)) == 0) { /* Three registers of different lengths. */ - int src1_wide; - int src2_wide; /* undefreq: bit 0 : UNDEF if size == 0 * bit 1 : UNDEF if size == 1 * bit 2 : UNDEF if size == 2 @@ -XXX,XX +XXX,XX @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn) {0, 0, 0, 7}, /* VADDW: handled by decodetree */ {0, 0, 0, 7}, /* VSUBL: handled by decodetree */ {0, 0, 0, 7}, /* VSUBW: handled by decodetree */ - {0, 1, 1, 0}, /* VADDHN */ + {0, 0, 0, 7}, /* VADDHN: handled by decodetree */ {0, 0, 0, 0}, /* VABAL */ - {0, 1, 1, 0}, /* VSUBHN */ + {0, 0, 0, 7}, /* VSUBHN: handled by decodetree */ {0, 0, 0, 0}, /* VABDL */ {0, 0, 0, 0}, /* VMLAL */ {0, 0, 0, 9}, /* VQDMLAL */ @@ -XXX,XX +XXX,XX @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn) {0, 0, 0, 7}, /* Reserved: always UNDEF */ }; - src1_wide = neon_3reg_wide[op][1]; - src2_wide = neon_3reg_wide[op][2]; undefreq = neon_3reg_wide[op][3]; if ((undefreq & (1 << size)) || ((undefreq & 8) && u)) { return 1; } - if ((src1_wide && (rn & 1)) || - (src2_wide && (rm & 1)) || - (!src2_wide && (rd & 1))) { + if (rd & 1) { return 1; } @@ -XXX,XX +XXX,XX @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn) /* Avoid overlapping operands. Wide source operands are always aligned so will never overlap with wide destinations in problematic ways. */ - if (rd == rm && !src2_wide) { + if (rd == rm) { tmp = neon_load_reg(rm, 1); neon_store_scratch(2, tmp); - } else if (rd == rn && !src1_wide) { + } else if (rd == rn) { tmp = neon_load_reg(rn, 1); neon_store_scratch(2, tmp); } tmp3 = NULL; for (pass = 0; pass < 2; pass++) { - if (src1_wide) { - neon_load_reg64(cpu_V0, rn + pass); - tmp = NULL; + if (pass == 1 && rd == rn) { + tmp = neon_load_scratch(2); } else { - if (pass == 1 && rd == rn) { - tmp = neon_load_scratch(2); - } else { - tmp = neon_load_reg(rn, pass); - } + tmp = neon_load_reg(rn, pass); } - if (src2_wide) { - neon_load_reg64(cpu_V1, rm + pass); - tmp2 = NULL; + if (pass == 1 && rd == rm) { + tmp2 = neon_load_scratch(2); } else { - if (pass == 1 && rd == rm) { - tmp2 = neon_load_scratch(2); - } else { - tmp2 = neon_load_reg(rm, pass); - } + tmp2 = neon_load_reg(rm, pass); } switch (op) { - case 0: case 1: case 4: /* VADDL, VADDW, VADDHN, VRADDHN */ - gen_neon_addl(size); - break; - case 2: case 3: case 6: /* VSUBL, VSUBW, VSUBHN, VRSUBHN */ - gen_neon_subl(size); - break; case 5: case 7: /* VABAL, VABDL */ switch ((size << 1) | u) { case 0: @@ -XXX,XX +XXX,XX @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn) abort(); } neon_store_reg64(cpu_V0, rd + pass); - } else if (op == 4 || op == 6) { - /* Narrowing operation. */ - tmp = tcg_temp_new_i32(); - if (!u) { - switch (size) { - case 0: - gen_helper_neon_narrow_high_u8(tmp, cpu_V0); - break; - case 1: - gen_helper_neon_narrow_high_u16(tmp, cpu_V0); - break; - case 2: - tcg_gen_extrh_i64_i32(tmp, cpu_V0); - break; - default: abort(); - } - } else { - switch (size) { - case 0: - gen_helper_neon_narrow_round_high_u8(tmp, cpu_V0); - break; - case 1: - gen_helper_neon_narrow_round_high_u16(tmp, cpu_V0); - break; - case 2: - tcg_gen_addi_i64(cpu_V0, cpu_V0, 1u << 31); - tcg_gen_extrh_i64_i32(tmp, cpu_V0); - break; - default: abort(); - } - } - if (pass == 0) { - tmp3 = tmp; - } else { - neon_store_reg(rd, 0, tmp3); - neon_store_reg(rd, 1, tmp); - } } else { /* Write back the result. */ neon_store_reg64(cpu_V0, rd + pass); -- 2.20.1 Convert the Neon 3-reg-diff insns VABAL and VABDL to decodetree. Like almost all the remaining insns in this group, these are a combination of a two-input operation which returns a double width result and then a possible accumulation of that double width result into the destination. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/translate.h | 1 + target/arm/neon-dp.decode | 6 ++ target/arm/translate-neon.inc.c | 132 ++++++++++++++++++++++++++++++++ target/arm/translate.c | 31 +------- 4 files changed, 142 insertions(+), 28 deletions(-) diff --git a/target/arm/translate.h b/target/arm/translate.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate.h +++ b/target/arm/translate.h @@ -XXX,XX +XXX,XX @@ typedef void NeonGenTwo64OpEnvFn(TCGv_i64, TCGv_ptr, TCGv_i64, TCGv_i64); typedef void NeonGenNarrowFn(TCGv_i32, TCGv_i64); typedef void NeonGenNarrowEnvFn(TCGv_i32, TCGv_ptr, TCGv_i64); typedef void NeonGenWidenFn(TCGv_i64, TCGv_i32); +typedef void NeonGenTwoOpWidenFn(TCGv_i64, TCGv_i32, TCGv_i32); typedef void NeonGenTwoSingleOPFn(TCGv_i32, TCGv_i32, TCGv_i32, TCGv_ptr); typedef void NeonGenTwoDoubleOPFn(TCGv_i64, TCGv_i64, TCGv_i64, TCGv_ptr); typedef void NeonGenOneOpFn(TCGv_i64, TCGv_i64); diff --git a/target/arm/neon-dp.decode b/target/arm/neon-dp.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/neon-dp.decode +++ b/target/arm/neon-dp.decode @@ -XXX,XX +XXX,XX @@ Vimm_1r 1111 001 . 1 . 000 ... .... cmode:4 0 . op:1 1 .... @1reg_imm VADDHN_3d 1111 001 0 1 . .. .... .... 0100 . 0 . 0 .... @3diff VRADDHN_3d 1111 001 1 1 . .. .... .... 0100 . 0 . 0 .... @3diff + VABAL_S_3d 1111 001 0 1 . .. .... .... 0101 . 0 . 0 .... @3diff + VABAL_U_3d 1111 001 1 1 . .. .... .... 0101 . 0 . 0 .... @3diff + VSUBHN_3d 1111 001 0 1 . .. .... .... 0110 . 0 . 0 .... @3diff VRSUBHN_3d 1111 001 1 1 . .. .... .... 0110 . 0 . 0 .... @3diff + + VABDL_S_3d 1111 001 0 1 . .. .... .... 0111 . 0 . 0 .... @3diff + VABDL_U_3d 1111 001 1 1 . .. .... .... 0111 . 0 . 0 .... @3diff ] } diff --git a/target/arm/translate-neon.inc.c b/target/arm/translate-neon.inc.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-neon.inc.c +++ b/target/arm/translate-neon.inc.c @@ -XXX,XX +XXX,XX @@ DO_NARROW_3D(VADDHN, add, narrow, tcg_gen_extrh_i64_i32) DO_NARROW_3D(VSUBHN, sub, narrow, tcg_gen_extrh_i64_i32) DO_NARROW_3D(VRADDHN, add, narrow_round, gen_narrow_round_high_u32) DO_NARROW_3D(VRSUBHN, sub, narrow_round, gen_narrow_round_high_u32) + +static bool do_long_3d(DisasContext *s, arg_3diff *a, + NeonGenTwoOpWidenFn *opfn, + NeonGenTwo64OpFn *accfn) +{ + /* + * 3-regs different lengths, long operations. + * These perform an operation on two inputs that returns a double-width + * result, and then possibly perform an accumulation operation of + * that result into the double-width destination. + */ + TCGv_i64 rd0, rd1, tmp; + TCGv_i32 rn, rm; + + if (!arm_dc_feature(s, ARM_FEATURE_NEON)) { + return false; + } + + /* UNDEF accesses to D16-D31 if they don't exist. */ + if (!dc_isar_feature(aa32_simd_r32, s) && + ((a->vd | a->vn | a->vm) & 0x10)) { + return false; + } + + if (!opfn) { + /* size == 3 case, which is an entirely different insn group */ + return false; + } + + if (a->vd & 1) { + return false; + } + + if (!vfp_access_check(s)) { + return true; + } + + rd0 = tcg_temp_new_i64(); + rd1 = tcg_temp_new_i64(); + + rn = neon_load_reg(a->vn, 0); + rm = neon_load_reg(a->vm, 0); + opfn(rd0, rn, rm); + tcg_temp_free_i32(rn); + tcg_temp_free_i32(rm); + + rn = neon_load_reg(a->vn, 1); + rm = neon_load_reg(a->vm, 1); + opfn(rd1, rn, rm); + tcg_temp_free_i32(rn); + tcg_temp_free_i32(rm); + + /* Don't store results until after all loads: they might overlap */ + if (accfn) { + tmp = tcg_temp_new_i64(); + neon_load_reg64(tmp, a->vd); + accfn(tmp, tmp, rd0); + neon_store_reg64(tmp, a->vd); + neon_load_reg64(tmp, a->vd + 1); + accfn(tmp, tmp, rd1); + neon_store_reg64(tmp, a->vd + 1); + tcg_temp_free_i64(tmp); + } else { + neon_store_reg64(rd0, a->vd); + neon_store_reg64(rd1, a->vd + 1); + } + + tcg_temp_free_i64(rd0); + tcg_temp_free_i64(rd1); + + return true; +} + +static bool trans_VABDL_S_3d(DisasContext *s, arg_3diff *a) +{ + static NeonGenTwoOpWidenFn * const opfn[] = { + gen_helper_neon_abdl_s16, + gen_helper_neon_abdl_s32, + gen_helper_neon_abdl_s64, + NULL, + }; + + return do_long_3d(s, a, opfn[a->size], NULL); +} + +static bool trans_VABDL_U_3d(DisasContext *s, arg_3diff *a) +{ + static NeonGenTwoOpWidenFn * const opfn[] = { + gen_helper_neon_abdl_u16, + gen_helper_neon_abdl_u32, + gen_helper_neon_abdl_u64, + NULL, + }; + + return do_long_3d(s, a, opfn[a->size], NULL); +} + +static bool trans_VABAL_S_3d(DisasContext *s, arg_3diff *a) +{ + static NeonGenTwoOpWidenFn * const opfn[] = { + gen_helper_neon_abdl_s16, + gen_helper_neon_abdl_s32, + gen_helper_neon_abdl_s64, + NULL, + }; + static NeonGenTwo64OpFn * const addfn[] = { + gen_helper_neon_addl_u16, + gen_helper_neon_addl_u32, + tcg_gen_add_i64, + NULL, + }; + + return do_long_3d(s, a, opfn[a->size], addfn[a->size]); +} + +static bool trans_VABAL_U_3d(DisasContext *s, arg_3diff *a) +{ + static NeonGenTwoOpWidenFn * const opfn[] = { + gen_helper_neon_abdl_u16, + gen_helper_neon_abdl_u32, + gen_helper_neon_abdl_u64, + NULL, + }; + static NeonGenTwo64OpFn * const addfn[] = { + gen_helper_neon_addl_u16, + gen_helper_neon_addl_u32, + tcg_gen_add_i64, + NULL, + }; + + return do_long_3d(s, a, opfn[a->size], addfn[a->size]); +} diff --git a/target/arm/translate.c b/target/arm/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate.c +++ b/target/arm/translate.c @@ -XXX,XX +XXX,XX @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn) {0, 0, 0, 7}, /* VSUBL: handled by decodetree */ {0, 0, 0, 7}, /* VSUBW: handled by decodetree */ {0, 0, 0, 7}, /* VADDHN: handled by decodetree */ - {0, 0, 0, 0}, /* VABAL */ + {0, 0, 0, 7}, /* VABAL */ {0, 0, 0, 7}, /* VSUBHN: handled by decodetree */ - {0, 0, 0, 0}, /* VABDL */ + {0, 0, 0, 7}, /* VABDL */ {0, 0, 0, 0}, /* VMLAL */ {0, 0, 0, 9}, /* VQDMLAL */ {0, 0, 0, 0}, /* VMLSL */ @@ -XXX,XX +XXX,XX @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn) tmp2 = neon_load_reg(rm, pass); } switch (op) { - case 5: case 7: /* VABAL, VABDL */ - switch ((size << 1) | u) { - case 0: - gen_helper_neon_abdl_s16(cpu_V0, tmp, tmp2); - break; - case 1: - gen_helper_neon_abdl_u16(cpu_V0, tmp, tmp2); - break; - case 2: - gen_helper_neon_abdl_s32(cpu_V0, tmp, tmp2); - break; - case 3: - gen_helper_neon_abdl_u32(cpu_V0, tmp, tmp2); - break; - case 4: - gen_helper_neon_abdl_s64(cpu_V0, tmp, tmp2); - break; - case 5: - gen_helper_neon_abdl_u64(cpu_V0, tmp, tmp2); - break; - default: abort(); - } - tcg_temp_free_i32(tmp2); - tcg_temp_free_i32(tmp); - break; case 8: case 9: case 10: case 11: case 12: case 13: /* VMLAL, VQDMLAL, VMLSL, VQDMLSL, VMULL, VQDMULL */ gen_neon_mull(cpu_V0, tmp, tmp2, size, u); @@ -XXX,XX +XXX,XX @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn) case 10: /* VMLSL */ gen_neon_negl(cpu_V0, size); /* Fall through */ - case 5: case 8: /* VABAL, VMLAL */ + case 8: /* VABAL, VMLAL */ gen_neon_addl(size); break; case 9: case 11: /* VQDMLAL, VQDMLSL */ -- 2.20.1 Convert the Neon 3-reg-diff insns VMULL, VMLAL and VMLSL; these perform a 32x32->64 multiply with possible accumulate. Note that for VMLSL we do the accumulate directly with a subtraction rather than doing a negate-then-add as the old code did. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/neon-dp.decode | 9 +++++ target/arm/translate-neon.inc.c | 71 +++++++++++++++++++++++++++++++++ target/arm/translate.c | 21 +++------- 3 files changed, 86 insertions(+), 15 deletions(-) diff --git a/target/arm/neon-dp.decode b/target/arm/neon-dp.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/neon-dp.decode +++ b/target/arm/neon-dp.decode @@ -XXX,XX +XXX,XX @@ Vimm_1r 1111 001 . 1 . 000 ... .... cmode:4 0 . op:1 1 .... @1reg_imm VABDL_S_3d 1111 001 0 1 . .. .... .... 0111 . 0 . 0 .... @3diff VABDL_U_3d 1111 001 1 1 . .. .... .... 0111 . 0 . 0 .... @3diff + + VMLAL_S_3d 1111 001 0 1 . .. .... .... 1000 . 0 . 0 .... @3diff + VMLAL_U_3d 1111 001 1 1 . .. .... .... 1000 . 0 . 0 .... @3diff + + VMLSL_S_3d 1111 001 0 1 . .. .... .... 1010 . 0 . 0 .... @3diff + VMLSL_U_3d 1111 001 1 1 . .. .... .... 1010 . 0 . 0 .... @3diff + + VMULL_S_3d 1111 001 0 1 . .. .... .... 1100 . 0 . 0 .... @3diff + VMULL_U_3d 1111 001 1 1 . .. .... .... 1100 . 0 . 0 .... @3diff ] } diff --git a/target/arm/translate-neon.inc.c b/target/arm/translate-neon.inc.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-neon.inc.c +++ b/target/arm/translate-neon.inc.c @@ -XXX,XX +XXX,XX @@ static bool trans_VABAL_U_3d(DisasContext *s, arg_3diff *a) return do_long_3d(s, a, opfn[a->size], addfn[a->size]); } + +static void gen_mull_s32(TCGv_i64 rd, TCGv_i32 rn, TCGv_i32 rm) +{ + TCGv_i32 lo = tcg_temp_new_i32(); + TCGv_i32 hi = tcg_temp_new_i32(); + + tcg_gen_muls2_i32(lo, hi, rn, rm); + tcg_gen_concat_i32_i64(rd, lo, hi); + + tcg_temp_free_i32(lo); + tcg_temp_free_i32(hi); +} + +static void gen_mull_u32(TCGv_i64 rd, TCGv_i32 rn, TCGv_i32 rm) +{ + TCGv_i32 lo = tcg_temp_new_i32(); + TCGv_i32 hi = tcg_temp_new_i32(); + + tcg_gen_mulu2_i32(lo, hi, rn, rm); + tcg_gen_concat_i32_i64(rd, lo, hi); + + tcg_temp_free_i32(lo); + tcg_temp_free_i32(hi); +} + +static bool trans_VMULL_S_3d(DisasContext *s, arg_3diff *a) +{ + static NeonGenTwoOpWidenFn * const opfn[] = { + gen_helper_neon_mull_s8, + gen_helper_neon_mull_s16, + gen_mull_s32, + NULL, + }; + + return do_long_3d(s, a, opfn[a->size], NULL); +} + +static bool trans_VMULL_U_3d(DisasContext *s, arg_3diff *a) +{ + static NeonGenTwoOpWidenFn * const opfn[] = { + gen_helper_neon_mull_u8, + gen_helper_neon_mull_u16, + gen_mull_u32, + NULL, + }; + + return do_long_3d(s, a, opfn[a->size], NULL); +} + +#define DO_VMLAL(INSN,MULL,ACC) \ + static bool trans_##INSN##_3d(DisasContext *s, arg_3diff *a) \ + { \ + static NeonGenTwoOpWidenFn * const opfn[] = { \ + gen_helper_neon_##MULL##8, \ + gen_helper_neon_##MULL##16, \ + gen_##MULL##32, \ + NULL, \ + }; \ + static NeonGenTwo64OpFn * const accfn[] = { \ + gen_helper_neon_##ACC##l_u16, \ + gen_helper_neon_##ACC##l_u32, \ + tcg_gen_##ACC##_i64, \ + NULL, \ + }; \ + return do_long_3d(s, a, opfn[a->size], accfn[a->size]); \ + } + +DO_VMLAL(VMLAL_S,mull_s,add) +DO_VMLAL(VMLAL_U,mull_u,add) +DO_VMLAL(VMLSL_S,mull_s,sub) +DO_VMLAL(VMLSL_U,mull_u,sub) diff --git a/target/arm/translate.c b/target/arm/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate.c +++ b/target/arm/translate.c @@ -XXX,XX +XXX,XX @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn) {0, 0, 0, 7}, /* VABAL */ {0, 0, 0, 7}, /* VSUBHN: handled by decodetree */ {0, 0, 0, 7}, /* VABDL */ - {0, 0, 0, 0}, /* VMLAL */ + {0, 0, 0, 7}, /* VMLAL */ {0, 0, 0, 9}, /* VQDMLAL */ - {0, 0, 0, 0}, /* VMLSL */ + {0, 0, 0, 7}, /* VMLSL */ {0, 0, 0, 9}, /* VQDMLSL */ - {0, 0, 0, 0}, /* Integer VMULL */ + {0, 0, 0, 7}, /* Integer VMULL */ {0, 0, 0, 9}, /* VQDMULL */ {0, 0, 0, 0xa}, /* Polynomial VMULL */ {0, 0, 0, 7}, /* Reserved: always UNDEF */ @@ -XXX,XX +XXX,XX @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn) tmp2 = neon_load_reg(rm, pass); } switch (op) { - case 8: case 9: case 10: case 11: case 12: case 13: - /* VMLAL, VQDMLAL, VMLSL, VQDMLSL, VMULL, VQDMULL */ + case 9: case 11: case 13: + /* VQDMLAL, VQDMLSL, VQDMULL */ gen_neon_mull(cpu_V0, tmp, tmp2, size, u); break; default: /* 15 is RESERVED: caught earlier */ @@ -XXX,XX +XXX,XX @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn) /* VQDMULL */ gen_neon_addl_saturate(cpu_V0, cpu_V0, size); neon_store_reg64(cpu_V0, rd + pass); - } else if (op == 5 || (op >= 8 && op <= 11)) { + } else { /* Accumulate. */ neon_load_reg64(cpu_V1, rd + pass); switch (op) { - case 10: /* VMLSL */ - gen_neon_negl(cpu_V0, size); - /* Fall through */ - case 8: /* VABAL, VMLAL */ - gen_neon_addl(size); - break; case 9: case 11: /* VQDMLAL, VQDMLSL */ gen_neon_addl_saturate(cpu_V0, cpu_V0, size); if (op == 11) { @@ -XXX,XX +XXX,XX @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn) abort(); } neon_store_reg64(cpu_V0, rd + pass); - } else { - /* Write back the result. */ - neon_store_reg64(cpu_V0, rd + pass); } } } else { -- 2.20.1 Convert the Neon 3-reg-diff insns VQDMULL, VQDMLAL and VQDMLSL: these are all saturating doubling long multiplies with a possible accumulate step. These are the last insns in the group which use the pass-over-each elements loop, so we can delete that code. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/neon-dp.decode | 6 +++ target/arm/translate-neon.inc.c | 82 +++++++++++++++++++++++++++++++++ target/arm/translate.c | 59 ++---------------------- 3 files changed, 92 insertions(+), 55 deletions(-) diff --git a/target/arm/neon-dp.decode b/target/arm/neon-dp.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/neon-dp.decode +++ b/target/arm/neon-dp.decode @@ -XXX,XX +XXX,XX @@ Vimm_1r 1111 001 . 1 . 000 ... .... cmode:4 0 . op:1 1 .... @1reg_imm VMLAL_S_3d 1111 001 0 1 . .. .... .... 1000 . 0 . 0 .... @3diff VMLAL_U_3d 1111 001 1 1 . .. .... .... 1000 . 0 . 0 .... @3diff + VQDMLAL_3d 1111 001 0 1 . .. .... .... 1001 . 0 . 0 .... @3diff + VMLSL_S_3d 1111 001 0 1 . .. .... .... 1010 . 0 . 0 .... @3diff VMLSL_U_3d 1111 001 1 1 . .. .... .... 1010 . 0 . 0 .... @3diff + VQDMLSL_3d 1111 001 0 1 . .. .... .... 1011 . 0 . 0 .... @3diff + VMULL_S_3d 1111 001 0 1 . .. .... .... 1100 . 0 . 0 .... @3diff VMULL_U_3d 1111 001 1 1 . .. .... .... 1100 . 0 . 0 .... @3diff + + VQDMULL_3d 1111 001 0 1 . .. .... .... 1101 . 0 . 0 .... @3diff ] } diff --git a/target/arm/translate-neon.inc.c b/target/arm/translate-neon.inc.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-neon.inc.c +++ b/target/arm/translate-neon.inc.c @@ -XXX,XX +XXX,XX @@ DO_VMLAL(VMLAL_S,mull_s,add) DO_VMLAL(VMLAL_U,mull_u,add) DO_VMLAL(VMLSL_S,mull_s,sub) DO_VMLAL(VMLSL_U,mull_u,sub) + +static void gen_VQDMULL_16(TCGv_i64 rd, TCGv_i32 rn, TCGv_i32 rm) +{ + gen_helper_neon_mull_s16(rd, rn, rm); + gen_helper_neon_addl_saturate_s32(rd, cpu_env, rd, rd); +} + +static void gen_VQDMULL_32(TCGv_i64 rd, TCGv_i32 rn, TCGv_i32 rm) +{ + gen_mull_s32(rd, rn, rm); + gen_helper_neon_addl_saturate_s64(rd, cpu_env, rd, rd); +} + +static bool trans_VQDMULL_3d(DisasContext *s, arg_3diff *a) +{ + static NeonGenTwoOpWidenFn * const opfn[] = { + NULL, + gen_VQDMULL_16, + gen_VQDMULL_32, + NULL, + }; + + return do_long_3d(s, a, opfn[a->size], NULL); +} + +static void gen_VQDMLAL_acc_16(TCGv_i64 rd, TCGv_i64 rn, TCGv_i64 rm) +{ + gen_helper_neon_addl_saturate_s32(rd, cpu_env, rn, rm); +} + +static void gen_VQDMLAL_acc_32(TCGv_i64 rd, TCGv_i64 rn, TCGv_i64 rm) +{ + gen_helper_neon_addl_saturate_s64(rd, cpu_env, rn, rm); +} + +static bool trans_VQDMLAL_3d(DisasContext *s, arg_3diff *a) +{ + static NeonGenTwoOpWidenFn * const opfn[] = { + NULL, + gen_VQDMULL_16, + gen_VQDMULL_32, + NULL, + }; + static NeonGenTwo64OpFn * const accfn[] = { + NULL, + gen_VQDMLAL_acc_16, + gen_VQDMLAL_acc_32, + NULL, + }; + + return do_long_3d(s, a, opfn[a->size], accfn[a->size]); +} + +static void gen_VQDMLSL_acc_16(TCGv_i64 rd, TCGv_i64 rn, TCGv_i64 rm) +{ + gen_helper_neon_negl_u32(rm, rm); + gen_helper_neon_addl_saturate_s32(rd, cpu_env, rn, rm); +} + +static void gen_VQDMLSL_acc_32(TCGv_i64 rd, TCGv_i64 rn, TCGv_i64 rm) +{ + tcg_gen_neg_i64(rm, rm); + gen_helper_neon_addl_saturate_s64(rd, cpu_env, rn, rm); +} + +static bool trans_VQDMLSL_3d(DisasContext *s, arg_3diff *a) +{ + static NeonGenTwoOpWidenFn * const opfn[] = { + NULL, + gen_VQDMULL_16, + gen_VQDMULL_32, + NULL, + }; + static NeonGenTwo64OpFn * const accfn[] = { + NULL, + gen_VQDMLSL_acc_16, + gen_VQDMLSL_acc_32, + NULL, + }; + + return do_long_3d(s, a, opfn[a->size], accfn[a->size]); +} diff --git a/target/arm/translate.c b/target/arm/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate.c +++ b/target/arm/translate.c @@ -XXX,XX +XXX,XX @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn) {0, 0, 0, 7}, /* VSUBHN: handled by decodetree */ {0, 0, 0, 7}, /* VABDL */ {0, 0, 0, 7}, /* VMLAL */ - {0, 0, 0, 9}, /* VQDMLAL */ + {0, 0, 0, 7}, /* VQDMLAL */ {0, 0, 0, 7}, /* VMLSL */ - {0, 0, 0, 9}, /* VQDMLSL */ + {0, 0, 0, 7}, /* VQDMLSL */ {0, 0, 0, 7}, /* Integer VMULL */ - {0, 0, 0, 9}, /* VQDMULL */ + {0, 0, 0, 7}, /* VQDMULL */ {0, 0, 0, 0xa}, /* Polynomial VMULL */ {0, 0, 0, 7}, /* Reserved: always UNDEF */ }; @@ -XXX,XX +XXX,XX @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn) } return 0; } - - /* Avoid overlapping operands. Wide source operands are - always aligned so will never overlap with wide - destinations in problematic ways. */ - if (rd == rm) { - tmp = neon_load_reg(rm, 1); - neon_store_scratch(2, tmp); - } else if (rd == rn) { - tmp = neon_load_reg(rn, 1); - neon_store_scratch(2, tmp); - } - tmp3 = NULL; - for (pass = 0; pass < 2; pass++) { - if (pass == 1 && rd == rn) { - tmp = neon_load_scratch(2); - } else { - tmp = neon_load_reg(rn, pass); - } - if (pass == 1 && rd == rm) { - tmp2 = neon_load_scratch(2); - } else { - tmp2 = neon_load_reg(rm, pass); - } - switch (op) { - case 9: case 11: case 13: - /* VQDMLAL, VQDMLSL, VQDMULL */ - gen_neon_mull(cpu_V0, tmp, tmp2, size, u); - break; - default: /* 15 is RESERVED: caught earlier */ - abort(); - } - if (op == 13) { - /* VQDMULL */ - gen_neon_addl_saturate(cpu_V0, cpu_V0, size); - neon_store_reg64(cpu_V0, rd + pass); - } else { - /* Accumulate. */ - neon_load_reg64(cpu_V1, rd + pass); - switch (op) { - case 9: case 11: /* VQDMLAL, VQDMLSL */ - gen_neon_addl_saturate(cpu_V0, cpu_V0, size); - if (op == 11) { - gen_neon_negl(cpu_V0, size); - } - gen_neon_addl_saturate(cpu_V0, cpu_V1, size); - break; - default: - abort(); - } - neon_store_reg64(cpu_V0, rd + pass); - } - } + abort(); /* all others handled by decodetree */ } else { /* Two registers and a scalar. NB that for ops of this form * the ARM ARM labels bit 24 as Q, but it is in our variable -- 2.20.1 Convert the Neon 3-reg-diff insn polynomial VMULL. This is the last insn in this group to be converted. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/neon-dp.decode | 2 ++ target/arm/translate-neon.inc.c | 43 +++++++++++++++++++++++ target/arm/translate.c | 60 ++------------------------------- 3 files changed, 48 insertions(+), 57 deletions(-) diff --git a/target/arm/neon-dp.decode b/target/arm/neon-dp.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/neon-dp.decode +++ b/target/arm/neon-dp.decode @@ -XXX,XX +XXX,XX @@ Vimm_1r 1111 001 . 1 . 000 ... .... cmode:4 0 . op:1 1 .... @1reg_imm VMULL_U_3d 1111 001 1 1 . .. .... .... 1100 . 0 . 0 .... @3diff VQDMULL_3d 1111 001 0 1 . .. .... .... 1101 . 0 . 0 .... @3diff + + VMULL_P_3d 1111 001 0 1 . .. .... .... 1110 . 0 . 0 .... @3diff ] } diff --git a/target/arm/translate-neon.inc.c b/target/arm/translate-neon.inc.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-neon.inc.c +++ b/target/arm/translate-neon.inc.c @@ -XXX,XX +XXX,XX @@ static bool trans_VQDMLSL_3d(DisasContext *s, arg_3diff *a) return do_long_3d(s, a, opfn[a->size], accfn[a->size]); } + +static bool trans_VMULL_P_3d(DisasContext *s, arg_3diff *a) +{ + gen_helper_gvec_3 *fn_gvec; + + if (!arm_dc_feature(s, ARM_FEATURE_NEON)) { + return false; + } + + /* UNDEF accesses to D16-D31 if they don't exist. */ + if (!dc_isar_feature(aa32_simd_r32, s) && + ((a->vd | a->vn | a->vm) & 0x10)) { + return false; + } + + if (a->vd & 1) { + return false; + } + + switch (a->size) { + case 0: + fn_gvec = gen_helper_neon_pmull_h; + break; + case 2: + if (!dc_isar_feature(aa32_pmull, s)) { + return false; + } + fn_gvec = gen_helper_gvec_pmull_q; + break; + default: + return false; + } + + if (!vfp_access_check(s)) { + return true; + } + + tcg_gen_gvec_3_ool(neon_reg_offset(a->vd, 0), + neon_reg_offset(a->vn, 0), + neon_reg_offset(a->vm, 0), + 16, 16, 0, fn_gvec); + return true; +} diff --git a/target/arm/translate.c b/target/arm/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate.c +++ b/target/arm/translate.c @@ -XXX,XX +XXX,XX @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn) { int op; int q; - int rd, rn, rm, rd_ofs, rn_ofs, rm_ofs; + int rd, rn, rm, rd_ofs, rm_ofs; int size; int pass; int u; @@ -XXX,XX +XXX,XX @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn) size = (insn >> 20) & 3; vec_size = q ? 16 : 8; rd_ofs = neon_reg_offset(rd, 0); - rn_ofs = neon_reg_offset(rn, 0); rm_ofs = neon_reg_offset(rm, 0); if ((insn & (1 << 23)) == 0) { @@ -XXX,XX +XXX,XX @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn) if (size != 3) { op = (insn >> 8) & 0xf; if ((insn & (1 << 6)) == 0) { - /* Three registers of different lengths. */ - /* undefreq: bit 0 : UNDEF if size == 0 - * bit 1 : UNDEF if size == 1 - * bit 2 : UNDEF if size == 2 - * bit 3 : UNDEF if U == 1 - * Note that [2:0] set implies 'always UNDEF' - */ - int undefreq; - /* prewiden, src1_wide, src2_wide, undefreq */ - static const int neon_3reg_wide[16][4] = { - {0, 0, 0, 7}, /* VADDL: handled by decodetree */ - {0, 0, 0, 7}, /* VADDW: handled by decodetree */ - {0, 0, 0, 7}, /* VSUBL: handled by decodetree */ - {0, 0, 0, 7}, /* VSUBW: handled by decodetree */ - {0, 0, 0, 7}, /* VADDHN: handled by decodetree */ - {0, 0, 0, 7}, /* VABAL */ - {0, 0, 0, 7}, /* VSUBHN: handled by decodetree */ - {0, 0, 0, 7}, /* VABDL */ - {0, 0, 0, 7}, /* VMLAL */ - {0, 0, 0, 7}, /* VQDMLAL */ - {0, 0, 0, 7}, /* VMLSL */ - {0, 0, 0, 7}, /* VQDMLSL */ - {0, 0, 0, 7}, /* Integer VMULL */ - {0, 0, 0, 7}, /* VQDMULL */ - {0, 0, 0, 0xa}, /* Polynomial VMULL */ - {0, 0, 0, 7}, /* Reserved: always UNDEF */ - }; - - undefreq = neon_3reg_wide[op][3]; - - if ((undefreq & (1 << size)) || - ((undefreq & 8) && u)) { - return 1; - } - if (rd & 1) { - return 1; - } - - /* Handle polynomial VMULL in a single pass. */ - if (op == 14) { - if (size == 0) { - /* VMULL.P8 */ - tcg_gen_gvec_3_ool(rd_ofs, rn_ofs, rm_ofs, 16, 16, - 0, gen_helper_neon_pmull_h); - } else { - /* VMULL.P64 */ - if (!dc_isar_feature(aa32_pmull, s)) { - return 1; - } - tcg_gen_gvec_3_ool(rd_ofs, rn_ofs, rm_ofs, 16, 16, - 0, gen_helper_gvec_pmull_q); - } - return 0; - } - abort(); /* all others handled by decodetree */ + /* Three registers of different lengths: handled by decodetree */ + return 1; } else { /* Two registers and a scalar. NB that for ops of this form * the ARM ARM labels bit 24 as Q, but it is in our variable -- 2.20.1 Mark the arrays of function pointers in trans_VSHLL_S_2sh() and trans_VSHLL_U_2sh() as both 'static' and 'const'. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/translate-neon.inc.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/target/arm/translate-neon.inc.c b/target/arm/translate-neon.inc.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-neon.inc.c +++ b/target/arm/translate-neon.inc.c @@ -XXX,XX +XXX,XX @@ static bool do_vshll_2sh(DisasContext *s, arg_2reg_shift *a, static bool trans_VSHLL_S_2sh(DisasContext *s, arg_2reg_shift *a) { - NeonGenWidenFn *widenfn[] = { + static NeonGenWidenFn * const widenfn[] = { gen_helper_neon_widen_s8, gen_helper_neon_widen_s16, tcg_gen_ext_i32_i64, @@ -XXX,XX +XXX,XX @@ static bool trans_VSHLL_S_2sh(DisasContext *s, arg_2reg_shift *a) static bool trans_VSHLL_U_2sh(DisasContext *s, arg_2reg_shift *a) { - NeonGenWidenFn *widenfn[] = { + static NeonGenWidenFn * const widenfn[] = { gen_helper_neon_widen_u8, gen_helper_neon_widen_u16, tcg_gen_extu_i32_i64, -- 2.20.1 In commit 37bfce81b10450071 we accidentally introduced a leak of a TCG temporary in do_2shift_env_64(); free it. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/translate-neon.inc.c | 1 + 1 file changed, 1 insertion(+) diff --git a/target/arm/translate-neon.inc.c b/target/arm/translate-neon.inc.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-neon.inc.c +++ b/target/arm/translate-neon.inc.c @@ -XXX,XX +XXX,XX @@ static bool do_2shift_env_64(DisasContext *s, arg_2reg_shift *a, neon_load_reg64(tmp, a->vm + pass); fn(tmp, cpu_env, tmp, constimm); neon_store_reg64(tmp, a->vd + pass); + tcg_temp_free_i64(tmp); } tcg_temp_free_i64(constimm); return true; -- 2.20.1 Convert the VMLA, VMLS and VMUL insns in the Neon "2 registers and a scalar" group to decodetree. These are 32x32->32 operations where one of the inputs is the scalar, followed by a possible accumulate operation of the 32-bit result. The refactoring removes some of the oddities of the old decoder: * operands to the operation and accumulation were often reversed (taking advantage of the fact that most of these ops are commutative); the new code follows the pseudocode order * the Q bit in the insn was in a local variable 'u'; in the new code it is decoded into a->q Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/neon-dp.decode | 15 ++++ target/arm/translate-neon.inc.c | 133 ++++++++++++++++++++++++++++++++ target/arm/translate.c | 77 ++---------------- 3 files changed, 154 insertions(+), 71 deletions(-) diff --git a/target/arm/neon-dp.decode b/target/arm/neon-dp.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/neon-dp.decode +++ b/target/arm/neon-dp.decode @@ -XXX,XX +XXX,XX @@ Vimm_1r 1111 001 . 1 . 000 ... .... cmode:4 0 . op:1 1 .... @1reg_imm VQDMULL_3d 1111 001 0 1 . .. .... .... 1101 . 0 . 0 .... @3diff VMULL_P_3d 1111 001 0 1 . .. .... .... 1110 . 0 . 0 .... @3diff + + ################################################################## + # 2-regs-plus-scalar grouping: + # 1111 001 Q 1 D sz!=11 Vn:4 Vd:4 opc:4 N 1 M 0 Vm:4 + ################################################################## + &2scalar vm vn vd size q + + @2scalar .... ... q:1 . . size:2 .... .... .... . . . . .... \ + &2scalar vm=%vm_dp vn=%vn_dp vd=%vd_dp + + VMLA_2sc 1111 001 . 1 . .. .... .... 0000 . 1 . 0 .... @2scalar + + VMLS_2sc 1111 001 . 1 . .. .... .... 0100 . 1 . 0 .... @2scalar + + VMUL_2sc 1111 001 . 1 . .. .... .... 1000 . 1 . 0 .... @2scalar ] } diff --git a/target/arm/translate-neon.inc.c b/target/arm/translate-neon.inc.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-neon.inc.c +++ b/target/arm/translate-neon.inc.c @@ -XXX,XX +XXX,XX @@ static bool trans_VMULL_P_3d(DisasContext *s, arg_3diff *a) 16, 16, 0, fn_gvec); return true; } + +static void gen_neon_dup_low16(TCGv_i32 var) +{ + TCGv_i32 tmp = tcg_temp_new_i32(); + tcg_gen_ext16u_i32(var, var); + tcg_gen_shli_i32(tmp, var, 16); + tcg_gen_or_i32(var, var, tmp); + tcg_temp_free_i32(tmp); +} + +static void gen_neon_dup_high16(TCGv_i32 var) +{ + TCGv_i32 tmp = tcg_temp_new_i32(); + tcg_gen_andi_i32(var, var, 0xffff0000); + tcg_gen_shri_i32(tmp, var, 16); + tcg_gen_or_i32(var, var, tmp); + tcg_temp_free_i32(tmp); +} + +static inline TCGv_i32 neon_get_scalar(int size, int reg) +{ + TCGv_i32 tmp; + if (size == 1) { + tmp = neon_load_reg(reg & 7, reg >> 4); + if (reg & 8) { + gen_neon_dup_high16(tmp); + } else { + gen_neon_dup_low16(tmp); + } + } else { + tmp = neon_load_reg(reg & 15, reg >> 4); + } + return tmp; +} + +static bool do_2scalar(DisasContext *s, arg_2scalar *a, + NeonGenTwoOpFn *opfn, NeonGenTwoOpFn *accfn) +{ + /* + * Two registers and a scalar: perform an operation between + * the input elements and the scalar, and then possibly + * perform an accumulation operation of that result into the + * destination. + */ + TCGv_i32 scalar; + int pass; + + if (!arm_dc_feature(s, ARM_FEATURE_NEON)) { + return false; + } + + /* UNDEF accesses to D16-D31 if they don't exist. */ + if (!dc_isar_feature(aa32_simd_r32, s) && + ((a->vd | a->vn | a->vm) & 0x10)) { + return false; + } + + if (!opfn) { + /* Bad size (including size == 3, which is a different insn group) */ + return false; + } + + if (a->q && ((a->vd | a->vn) & 1)) { + return false; + } + + if (!vfp_access_check(s)) { + return true; + } + + scalar = neon_get_scalar(a->size, a->vm); + + for (pass = 0; pass < (a->q ? 4 : 2); pass++) { + TCGv_i32 tmp = neon_load_reg(a->vn, pass); + opfn(tmp, tmp, scalar); + if (accfn) { + TCGv_i32 rd = neon_load_reg(a->vd, pass); + accfn(tmp, rd, tmp); + tcg_temp_free_i32(rd); + } + neon_store_reg(a->vd, pass, tmp); + } + tcg_temp_free_i32(scalar); + return true; +} + +static bool trans_VMUL_2sc(DisasContext *s, arg_2scalar *a) +{ + static NeonGenTwoOpFn * const opfn[] = { + NULL, + gen_helper_neon_mul_u16, + tcg_gen_mul_i32, + NULL, + }; + + return do_2scalar(s, a, opfn[a->size], NULL); +} + +static bool trans_VMLA_2sc(DisasContext *s, arg_2scalar *a) +{ + static NeonGenTwoOpFn * const opfn[] = { + NULL, + gen_helper_neon_mul_u16, + tcg_gen_mul_i32, + NULL, + }; + static NeonGenTwoOpFn * const accfn[] = { + NULL, + gen_helper_neon_add_u16, + tcg_gen_add_i32, + NULL, + }; + + return do_2scalar(s, a, opfn[a->size], accfn[a->size]); +} + +static bool trans_VMLS_2sc(DisasContext *s, arg_2scalar *a) +{ + static NeonGenTwoOpFn * const opfn[] = { + NULL, + gen_helper_neon_mul_u16, + tcg_gen_mul_i32, + NULL, + }; + static NeonGenTwoOpFn * const accfn[] = { + NULL, + gen_helper_neon_sub_u16, + tcg_gen_sub_i32, + NULL, + }; + + return do_2scalar(s, a, opfn[a->size], accfn[a->size]); +} diff --git a/target/arm/translate.c b/target/arm/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate.c +++ b/target/arm/translate.c @@ -XXX,XX +XXX,XX @@ static int disas_dsp_insn(DisasContext *s, uint32_t insn) #define VFP_DREG_N(reg, insn) VFP_DREG(reg, insn, 16, 7) #define VFP_DREG_M(reg, insn) VFP_DREG(reg, insn, 0, 5) -static void gen_neon_dup_low16(TCGv_i32 var) -{ - TCGv_i32 tmp = tcg_temp_new_i32(); - tcg_gen_ext16u_i32(var, var); - tcg_gen_shli_i32(tmp, var, 16); - tcg_gen_or_i32(var, var, tmp); - tcg_temp_free_i32(tmp); -} - -static void gen_neon_dup_high16(TCGv_i32 var) -{ - TCGv_i32 tmp = tcg_temp_new_i32(); - tcg_gen_andi_i32(var, var, 0xffff0000); - tcg_gen_shri_i32(tmp, var, 16); - tcg_gen_or_i32(var, var, tmp); - tcg_temp_free_i32(tmp); -} - static inline bool use_goto_tb(DisasContext *s, target_ulong dest) { #ifndef CONFIG_USER_ONLY @@ -XXX,XX +XXX,XX @@ static void gen_exception_return(DisasContext *s, TCGv_i32 pc) #define CPU_V001 cpu_V0, cpu_V0, cpu_V1 -static inline void gen_neon_add(int size, TCGv_i32 t0, TCGv_i32 t1) -{ - switch (size) { - case 0: gen_helper_neon_add_u8(t0, t0, t1); break; - case 1: gen_helper_neon_add_u16(t0, t0, t1); break; - case 2: tcg_gen_add_i32(t0, t0, t1); break; - default: abort(); - } -} - -static inline void gen_neon_rsb(int size, TCGv_i32 t0, TCGv_i32 t1) -{ - switch (size) { - case 0: gen_helper_neon_sub_u8(t0, t1, t0); break; - case 1: gen_helper_neon_sub_u16(t0, t1, t0); break; - case 2: tcg_gen_sub_i32(t0, t1, t0); break; - default: return; - } -} - static TCGv_i32 neon_load_scratch(int scratch) { TCGv_i32 tmp = tcg_temp_new_i32(); @@ -XXX,XX +XXX,XX @@ static void neon_store_scratch(int scratch, TCGv_i32 var) tcg_temp_free_i32(var); } -static inline TCGv_i32 neon_get_scalar(int size, int reg) -{ - TCGv_i32 tmp; - if (size == 1) { - tmp = neon_load_reg(reg & 7, reg >> 4); - if (reg & 8) { - gen_neon_dup_high16(tmp); - } else { - gen_neon_dup_low16(tmp); - } - } else { - tmp = neon_load_reg(reg & 15, reg >> 4); - } - return tmp; -} - static int gen_neon_unzip(int rd, int rm, int size, int q) { TCGv_ptr pd, pm; @@ -XXX,XX +XXX,XX @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn) return 1; } switch (op) { + case 0: /* Integer VMLA scalar */ + case 4: /* Integer VMLS scalar */ + case 8: /* Integer VMUL scalar */ + return 1; /* handled by decodetree */ + case 1: /* Float VMLA scalar */ case 5: /* Floating point VMLS scalar */ case 9: /* Floating point VMUL scalar */ @@ -XXX,XX +XXX,XX @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn) return 1; } /* fall through */ - case 0: /* Integer VMLA scalar */ - case 4: /* Integer VMLS scalar */ - case 8: /* Integer VMUL scalar */ case 12: /* VQDMULH scalar */ case 13: /* VQRDMULH scalar */ if (u && ((rd | rn) & 1)) { @@ -XXX,XX +XXX,XX @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn) } else { gen_helper_neon_qrdmulh_s32(tmp, cpu_env, tmp, tmp2); } - } else if (op & 1) { + } else { TCGv_ptr fpstatus = get_fpstatus_ptr(1); gen_helper_vfp_muls(tmp, tmp, tmp2, fpstatus); tcg_temp_free_ptr(fpstatus); - } else { - switch (size) { - case 0: gen_helper_neon_mul_u8(tmp, tmp, tmp2); break; - case 1: gen_helper_neon_mul_u16(tmp, tmp, tmp2); break; - case 2: tcg_gen_mul_i32(tmp, tmp, tmp2); break; - default: abort(); - } } tcg_temp_free_i32(tmp2); if (op < 8) { /* Accumulate. */ tmp2 = neon_load_reg(rd, pass); switch (op) { - case 0: - gen_neon_add(size, tmp, tmp2); - break; case 1: { TCGv_ptr fpstatus = get_fpstatus_ptr(1); @@ -XXX,XX +XXX,XX @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn) tcg_temp_free_ptr(fpstatus); break; } - case 4: - gen_neon_rsb(size, tmp, tmp2); - break; case 5: { TCGv_ptr fpstatus = get_fpstatus_ptr(1); -- 2.20.1 Convert the float versions of VMLA, VMLS and VMUL in the Neon 2-reg-scalar group to decodetree. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- As noted in the comment on the WRAP_FP_FN macro, we could have had a do_2scalar_fp() function, but for 3 insns it seemed simpler to just do the wrapping to get hold of the fpstatus ptr. (These are the only fp insns in the group.) Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/neon-dp.decode | 3 ++ target/arm/translate-neon.inc.c | 65 +++++++++++++++++++++++++++++++++ target/arm/translate.c | 37 ++----------------- 3 files changed, 71 insertions(+), 34 deletions(-) diff --git a/target/arm/neon-dp.decode b/target/arm/neon-dp.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/neon-dp.decode +++ b/target/arm/neon-dp.decode @@ -XXX,XX +XXX,XX @@ Vimm_1r 1111 001 . 1 . 000 ... .... cmode:4 0 . op:1 1 .... @1reg_imm &2scalar vm=%vm_dp vn=%vn_dp vd=%vd_dp VMLA_2sc 1111 001 . 1 . .. .... .... 0000 . 1 . 0 .... @2scalar + VMLA_F_2sc 1111 001 . 1 . .. .... .... 0001 . 1 . 0 .... @2scalar VMLS_2sc 1111 001 . 1 . .. .... .... 0100 . 1 . 0 .... @2scalar + VMLS_F_2sc 1111 001 . 1 . .. .... .... 0101 . 1 . 0 .... @2scalar VMUL_2sc 1111 001 . 1 . .. .... .... 1000 . 1 . 0 .... @2scalar + VMUL_F_2sc 1111 001 . 1 . .. .... .... 1001 . 1 . 0 .... @2scalar ] } diff --git a/target/arm/translate-neon.inc.c b/target/arm/translate-neon.inc.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-neon.inc.c +++ b/target/arm/translate-neon.inc.c @@ -XXX,XX +XXX,XX @@ static bool trans_VMLS_2sc(DisasContext *s, arg_2scalar *a) return do_2scalar(s, a, opfn[a->size], accfn[a->size]); } + +/* + * Rather than have a float-specific version of do_2scalar just for + * three insns, we wrap a NeonGenTwoSingleOpFn to turn it into + * a NeonGenTwoOpFn. + */ +#define WRAP_FP_FN(WRAPNAME, FUNC) \ + static void WRAPNAME(TCGv_i32 rd, TCGv_i32 rn, TCGv_i32 rm) \ + { \ + TCGv_ptr fpstatus = get_fpstatus_ptr(1); \ + FUNC(rd, rn, rm, fpstatus); \ + tcg_temp_free_ptr(fpstatus); \ + } + +WRAP_FP_FN(gen_VMUL_F_mul, gen_helper_vfp_muls) +WRAP_FP_FN(gen_VMUL_F_add, gen_helper_vfp_adds) +WRAP_FP_FN(gen_VMUL_F_sub, gen_helper_vfp_subs) + +static bool trans_VMUL_F_2sc(DisasContext *s, arg_2scalar *a) +{ + static NeonGenTwoOpFn * const opfn[] = { + NULL, + NULL, /* TODO: fp16 support */ + gen_VMUL_F_mul, + NULL, + }; + + return do_2scalar(s, a, opfn[a->size], NULL); +} + +static bool trans_VMLA_F_2sc(DisasContext *s, arg_2scalar *a) +{ + static NeonGenTwoOpFn * const opfn[] = { + NULL, + NULL, /* TODO: fp16 support */ + gen_VMUL_F_mul, + NULL, + }; + static NeonGenTwoOpFn * const accfn[] = { + NULL, + NULL, /* TODO: fp16 support */ + gen_VMUL_F_add, + NULL, + }; + + return do_2scalar(s, a, opfn[a->size], accfn[a->size]); +} + +static bool trans_VMLS_F_2sc(DisasContext *s, arg_2scalar *a) +{ + static NeonGenTwoOpFn * const opfn[] = { + NULL, + NULL, /* TODO: fp16 support */ + gen_VMUL_F_mul, + NULL, + }; + static NeonGenTwoOpFn * const accfn[] = { + NULL, + NULL, /* TODO: fp16 support */ + gen_VMUL_F_sub, + NULL, + }; + + return do_2scalar(s, a, opfn[a->size], accfn[a->size]); +} diff --git a/target/arm/translate.c b/target/arm/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate.c +++ b/target/arm/translate.c @@ -XXX,XX +XXX,XX @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn) case 0: /* Integer VMLA scalar */ case 4: /* Integer VMLS scalar */ case 8: /* Integer VMUL scalar */ - return 1; /* handled by decodetree */ - case 1: /* Float VMLA scalar */ case 5: /* Floating point VMLS scalar */ case 9: /* Floating point VMUL scalar */ - if (size == 1) { - return 1; - } - /* fall through */ + return 1; /* handled by decodetree */ + case 12: /* VQDMULH scalar */ case 13: /* VQRDMULH scalar */ if (u && ((rd | rn) & 1)) { @@ -XXX,XX +XXX,XX @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn) } else { gen_helper_neon_qdmulh_s32(tmp, cpu_env, tmp, tmp2); } - } else if (op == 13) { + } else { if (size == 1) { gen_helper_neon_qrdmulh_s16(tmp, cpu_env, tmp, tmp2); } else { gen_helper_neon_qrdmulh_s32(tmp, cpu_env, tmp, tmp2); } - } else { - TCGv_ptr fpstatus = get_fpstatus_ptr(1); - gen_helper_vfp_muls(tmp, tmp, tmp2, fpstatus); - tcg_temp_free_ptr(fpstatus); } tcg_temp_free_i32(tmp2); - if (op < 8) { - /* Accumulate. */ - tmp2 = neon_load_reg(rd, pass); - switch (op) { - case 1: - { - TCGv_ptr fpstatus = get_fpstatus_ptr(1); - gen_helper_vfp_adds(tmp, tmp, tmp2, fpstatus); - tcg_temp_free_ptr(fpstatus); - break; - } - case 5: - { - TCGv_ptr fpstatus = get_fpstatus_ptr(1); - gen_helper_vfp_subs(tmp, tmp2, tmp, fpstatus); - tcg_temp_free_ptr(fpstatus); - break; - } - default: - abort(); - } - tcg_temp_free_i32(tmp2); - } neon_store_reg(rd, pass, tmp); } break; -- 2.20.1 Convert the VQDMULH and VQRDMULH insns in the 2-reg-scalar group to decodetree. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/neon-dp.decode | 3 +++ target/arm/translate-neon.inc.c | 29 +++++++++++++++++++++++ target/arm/translate.c | 42 ++------------------------------- 3 files changed, 34 insertions(+), 40 deletions(-) diff --git a/target/arm/neon-dp.decode b/target/arm/neon-dp.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/neon-dp.decode +++ b/target/arm/neon-dp.decode @@ -XXX,XX +XXX,XX @@ Vimm_1r 1111 001 . 1 . 000 ... .... cmode:4 0 . op:1 1 .... @1reg_imm VMUL_2sc 1111 001 . 1 . .. .... .... 1000 . 1 . 0 .... @2scalar VMUL_F_2sc 1111 001 . 1 . .. .... .... 1001 . 1 . 0 .... @2scalar + + VQDMULH_2sc 1111 001 . 1 . .. .... .... 1100 . 1 . 0 .... @2scalar + VQRDMULH_2sc 1111 001 . 1 . .. .... .... 1101 . 1 . 0 .... @2scalar ] } diff --git a/target/arm/translate-neon.inc.c b/target/arm/translate-neon.inc.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-neon.inc.c +++ b/target/arm/translate-neon.inc.c @@ -XXX,XX +XXX,XX @@ static bool trans_VMLS_F_2sc(DisasContext *s, arg_2scalar *a) return do_2scalar(s, a, opfn[a->size], accfn[a->size]); } + +WRAP_ENV_FN(gen_VQDMULH_16, gen_helper_neon_qdmulh_s16) +WRAP_ENV_FN(gen_VQDMULH_32, gen_helper_neon_qdmulh_s32) +WRAP_ENV_FN(gen_VQRDMULH_16, gen_helper_neon_qrdmulh_s16) +WRAP_ENV_FN(gen_VQRDMULH_32, gen_helper_neon_qrdmulh_s32) + +static bool trans_VQDMULH_2sc(DisasContext *s, arg_2scalar *a) +{ + static NeonGenTwoOpFn * const opfn[] = { + NULL, + gen_VQDMULH_16, + gen_VQDMULH_32, + NULL, + }; + + return do_2scalar(s, a, opfn[a->size], NULL); +} + +static bool trans_VQRDMULH_2sc(DisasContext *s, arg_2scalar *a) +{ + static NeonGenTwoOpFn * const opfn[] = { + NULL, + gen_VQRDMULH_16, + gen_VQRDMULH_32, + NULL, + }; + + return do_2scalar(s, a, opfn[a->size], NULL); +} diff --git a/target/arm/translate.c b/target/arm/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate.c +++ b/target/arm/translate.c @@ -XXX,XX +XXX,XX @@ static void gen_exception_return(DisasContext *s, TCGv_i32 pc) #define CPU_V001 cpu_V0, cpu_V0, cpu_V1 -static TCGv_i32 neon_load_scratch(int scratch) -{ - TCGv_i32 tmp = tcg_temp_new_i32(); - tcg_gen_ld_i32(tmp, cpu_env, offsetof(CPUARMState, vfp.scratch[scratch])); - return tmp; -} - -static void neon_store_scratch(int scratch, TCGv_i32 var) -{ - tcg_gen_st_i32(var, cpu_env, offsetof(CPUARMState, vfp.scratch[scratch])); - tcg_temp_free_i32(var); -} - static int gen_neon_unzip(int rd, int rm, int size, int q) { TCGv_ptr pd, pm; @@ -XXX,XX +XXX,XX @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn) case 1: /* Float VMLA scalar */ case 5: /* Floating point VMLS scalar */ case 9: /* Floating point VMUL scalar */ - return 1; /* handled by decodetree */ - case 12: /* VQDMULH scalar */ case 13: /* VQRDMULH scalar */ - if (u && ((rd | rn) & 1)) { - return 1; - } - tmp = neon_get_scalar(size, rm); - neon_store_scratch(0, tmp); - for (pass = 0; pass < (u ? 4 : 2); pass++) { - tmp = neon_load_scratch(0); - tmp2 = neon_load_reg(rn, pass); - if (op == 12) { - if (size == 1) { - gen_helper_neon_qdmulh_s16(tmp, cpu_env, tmp, tmp2); - } else { - gen_helper_neon_qdmulh_s32(tmp, cpu_env, tmp, tmp2); - } - } else { - if (size == 1) { - gen_helper_neon_qrdmulh_s16(tmp, cpu_env, tmp, tmp2); - } else { - gen_helper_neon_qrdmulh_s32(tmp, cpu_env, tmp, tmp2); - } - } - tcg_temp_free_i32(tmp2); - neon_store_reg(rd, pass, tmp); - } - break; + return 1; /* handled by decodetree */ + case 3: /* VQDMLAL scalar */ case 7: /* VQDMLSL scalar */ case 11: /* VQDMULL scalar */ -- 2.20.1 Convert the VQRDMLAH and VQRDMLSH insns in the 2-reg-scalar group to decodetree. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/neon-dp.decode | 3 ++ target/arm/translate-neon.inc.c | 74 +++++++++++++++++++++++++++++++++ target/arm/translate.c | 38 +---------------- 3 files changed, 79 insertions(+), 36 deletions(-) diff --git a/target/arm/neon-dp.decode b/target/arm/neon-dp.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/neon-dp.decode +++ b/target/arm/neon-dp.decode @@ -XXX,XX +XXX,XX @@ Vimm_1r 1111 001 . 1 . 000 ... .... cmode:4 0 . op:1 1 .... @1reg_imm VQDMULH_2sc 1111 001 . 1 . .. .... .... 1100 . 1 . 0 .... @2scalar VQRDMULH_2sc 1111 001 . 1 . .. .... .... 1101 . 1 . 0 .... @2scalar + + VQRDMLAH_2sc 1111 001 . 1 . .. .... .... 1110 . 1 . 0 .... @2scalar + VQRDMLSH_2sc 1111 001 . 1 . .. .... .... 1111 . 1 . 0 .... @2scalar ] } diff --git a/target/arm/translate-neon.inc.c b/target/arm/translate-neon.inc.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-neon.inc.c +++ b/target/arm/translate-neon.inc.c @@ -XXX,XX +XXX,XX @@ static bool trans_VQRDMULH_2sc(DisasContext *s, arg_2scalar *a) return do_2scalar(s, a, opfn[a->size], NULL); } + +static bool do_vqrdmlah_2sc(DisasContext *s, arg_2scalar *a, + NeonGenThreeOpEnvFn *opfn) +{ + /* + * VQRDMLAH/VQRDMLSH: this is like do_2scalar, but the opfn + * performs a kind of fused op-then-accumulate using a helper + * function that takes all of rd, rn and the scalar at once. + */ + TCGv_i32 scalar; + int pass; + + if (!arm_dc_feature(s, ARM_FEATURE_NEON)) { + return false; + } + + if (!dc_isar_feature(aa32_rdm, s)) { + return false; + } + + /* UNDEF accesses to D16-D31 if they don't exist. */ + if (!dc_isar_feature(aa32_simd_r32, s) && + ((a->vd | a->vn | a->vm) & 0x10)) { + return false; + } + + if (!opfn) { + /* Bad size (including size == 3, which is a different insn group) */ + return false; + } + + if (a->q && ((a->vd | a->vn) & 1)) { + return false; + } + + if (!vfp_access_check(s)) { + return true; + } + + scalar = neon_get_scalar(a->size, a->vm); + + for (pass = 0; pass < (a->q ? 4 : 2); pass++) { + TCGv_i32 rn = neon_load_reg(a->vn, pass); + TCGv_i32 rd = neon_load_reg(a->vd, pass); + opfn(rd, cpu_env, rn, scalar, rd); + tcg_temp_free_i32(rn); + neon_store_reg(a->vd, pass, rd); + } + tcg_temp_free_i32(scalar); + + return true; +} + +static bool trans_VQRDMLAH_2sc(DisasContext *s, arg_2scalar *a) +{ + static NeonGenThreeOpEnvFn *opfn[] = { + NULL, + gen_helper_neon_qrdmlah_s16, + gen_helper_neon_qrdmlah_s32, + NULL, + }; + return do_vqrdmlah_2sc(s, a, opfn[a->size]); +} + +static bool trans_VQRDMLSH_2sc(DisasContext *s, arg_2scalar *a) +{ + static NeonGenThreeOpEnvFn *opfn[] = { + NULL, + gen_helper_neon_qrdmlsh_s16, + gen_helper_neon_qrdmlsh_s32, + NULL, + }; + return do_vqrdmlah_2sc(s, a, opfn[a->size]); +} diff --git a/target/arm/translate.c b/target/arm/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate.c +++ b/target/arm/translate.c @@ -XXX,XX +XXX,XX @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn) case 9: /* Floating point VMUL scalar */ case 12: /* VQDMULH scalar */ case 13: /* VQRDMULH scalar */ + case 14: /* VQRDMLAH scalar */ + case 15: /* VQRDMLSH scalar */ return 1; /* handled by decodetree */ case 3: /* VQDMLAL scalar */ @@ -XXX,XX +XXX,XX @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn) neon_store_reg64(cpu_V0, rd + pass); } break; - case 14: /* VQRDMLAH scalar */ - case 15: /* VQRDMLSH scalar */ - { - NeonGenThreeOpEnvFn *fn; - - if (!dc_isar_feature(aa32_rdm, s)) { - return 1; - } - if (u && ((rd | rn) & 1)) { - return 1; - } - if (op == 14) { - if (size == 1) { - fn = gen_helper_neon_qrdmlah_s16; - } else { - fn = gen_helper_neon_qrdmlah_s32; - } - } else { - if (size == 1) { - fn = gen_helper_neon_qrdmlsh_s16; - } else { - fn = gen_helper_neon_qrdmlsh_s32; - } - } - - tmp2 = neon_get_scalar(size, rm); - for (pass = 0; pass < (u ? 4 : 2); pass++) { - tmp = neon_load_reg(rn, pass); - tmp3 = neon_load_reg(rd, pass); - fn(tmp, cpu_env, tmp, tmp2, tmp3); - tcg_temp_free_i32(tmp3); - neon_store_reg(rd, pass, tmp); - } - tcg_temp_free_i32(tmp2); - } - break; default: g_assert_not_reached(); } -- 2.20.1 Convert the Neon 2-reg-scalar long multiplies to decodetree. These are the last instructions in the group. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/neon-dp.decode | 18 ++++ target/arm/translate-neon.inc.c | 163 ++++++++++++++++++++++++++++ target/arm/translate.c | 182 ++------------------------------ 3 files changed, 187 insertions(+), 176 deletions(-) diff --git a/target/arm/neon-dp.decode b/target/arm/neon-dp.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/neon-dp.decode +++ b/target/arm/neon-dp.decode @@ -XXX,XX +XXX,XX @@ Vimm_1r 1111 001 . 1 . 000 ... .... cmode:4 0 . op:1 1 .... @1reg_imm @2scalar .... ... q:1 . . size:2 .... .... .... . . . . .... \ &2scalar vm=%vm_dp vn=%vn_dp vd=%vd_dp + # For the 'long' ops the Q bit is part of insn decode + @2scalar_q0 .... ... . . . size:2 .... .... .... . . . . .... \ + &2scalar vm=%vm_dp vn=%vn_dp vd=%vd_dp q=0 VMLA_2sc 1111 001 . 1 . .. .... .... 0000 . 1 . 0 .... @2scalar VMLA_F_2sc 1111 001 . 1 . .. .... .... 0001 . 1 . 0 .... @2scalar + VMLAL_S_2sc 1111 001 0 1 . .. .... .... 0010 . 1 . 0 .... @2scalar_q0 + VMLAL_U_2sc 1111 001 1 1 . .. .... .... 0010 . 1 . 0 .... @2scalar_q0 + + VQDMLAL_2sc 1111 001 0 1 . .. .... .... 0011 . 1 . 0 .... @2scalar_q0 + VMLS_2sc 1111 001 . 1 . .. .... .... 0100 . 1 . 0 .... @2scalar VMLS_F_2sc 1111 001 . 1 . .. .... .... 0101 . 1 . 0 .... @2scalar + VMLSL_S_2sc 1111 001 0 1 . .. .... .... 0110 . 1 . 0 .... @2scalar_q0 + VMLSL_U_2sc 1111 001 1 1 . .. .... .... 0110 . 1 . 0 .... @2scalar_q0 + + VQDMLSL_2sc 1111 001 0 1 . .. .... .... 0111 . 1 . 0 .... @2scalar_q0 + VMUL_2sc 1111 001 . 1 . .. .... .... 1000 . 1 . 0 .... @2scalar VMUL_F_2sc 1111 001 . 1 . .. .... .... 1001 . 1 . 0 .... @2scalar + VMULL_S_2sc 1111 001 0 1 . .. .... .... 1010 . 1 . 0 .... @2scalar_q0 + VMULL_U_2sc 1111 001 1 1 . .. .... .... 1010 . 1 . 0 .... @2scalar_q0 + + VQDMULL_2sc 1111 001 0 1 . .. .... .... 1011 . 1 . 0 .... @2scalar_q0 + VQDMULH_2sc 1111 001 . 1 . .. .... .... 1100 . 1 . 0 .... @2scalar VQRDMULH_2sc 1111 001 . 1 . .. .... .... 1101 . 1 . 0 .... @2scalar diff --git a/target/arm/translate-neon.inc.c b/target/arm/translate-neon.inc.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-neon.inc.c +++ b/target/arm/translate-neon.inc.c @@ -XXX,XX +XXX,XX @@ static bool trans_VQRDMLSH_2sc(DisasContext *s, arg_2scalar *a) }; return do_vqrdmlah_2sc(s, a, opfn[a->size]); } + +static bool do_2scalar_long(DisasContext *s, arg_2scalar *a, + NeonGenTwoOpWidenFn *opfn, + NeonGenTwo64OpFn *accfn) +{ + /* + * Two registers and a scalar, long operations: perform an + * operation on the input elements and the scalar which produces + * a double-width result, and then possibly perform an accumulation + * operation of that result into the destination. + */ + TCGv_i32 scalar, rn; + TCGv_i64 rn0_64, rn1_64; + + if (!arm_dc_feature(s, ARM_FEATURE_NEON)) { + return false; + } + + /* UNDEF accesses to D16-D31 if they don't exist. */ + if (!dc_isar_feature(aa32_simd_r32, s) && + ((a->vd | a->vn | a->vm) & 0x10)) { + return false; + } + + if (!opfn) { + /* Bad size (including size == 3, which is a different insn group) */ + return false; + } + + if (a->vd & 1) { + return false; + } + + if (!vfp_access_check(s)) { + return true; + } + + scalar = neon_get_scalar(a->size, a->vm); + + /* Load all inputs before writing any outputs, in case of overlap */ + rn = neon_load_reg(a->vn, 0); + rn0_64 = tcg_temp_new_i64(); + opfn(rn0_64, rn, scalar); + tcg_temp_free_i32(rn); + + rn = neon_load_reg(a->vn, 1); + rn1_64 = tcg_temp_new_i64(); + opfn(rn1_64, rn, scalar); + tcg_temp_free_i32(rn); + tcg_temp_free_i32(scalar); + + if (accfn) { + TCGv_i64 t64 = tcg_temp_new_i64(); + neon_load_reg64(t64, a->vd); + accfn(t64, t64, rn0_64); + neon_store_reg64(t64, a->vd); + neon_load_reg64(t64, a->vd + 1); + accfn(t64, t64, rn1_64); + neon_store_reg64(t64, a->vd + 1); + tcg_temp_free_i64(t64); + } else { + neon_store_reg64(rn0_64, a->vd); + neon_store_reg64(rn1_64, a->vd + 1); + } + tcg_temp_free_i64(rn0_64); + tcg_temp_free_i64(rn1_64); + return true; +} + +static bool trans_VMULL_S_2sc(DisasContext *s, arg_2scalar *a) +{ + static NeonGenTwoOpWidenFn * const opfn[] = { + NULL, + gen_helper_neon_mull_s16, + gen_mull_s32, + NULL, + }; + + return do_2scalar_long(s, a, opfn[a->size], NULL); +} + +static bool trans_VMULL_U_2sc(DisasContext *s, arg_2scalar *a) +{ + static NeonGenTwoOpWidenFn * const opfn[] = { + NULL, + gen_helper_neon_mull_u16, + gen_mull_u32, + NULL, + }; + + return do_2scalar_long(s, a, opfn[a->size], NULL); +} + +#define DO_VMLAL_2SC(INSN, MULL, ACC) \ + static bool trans_##INSN##_2sc(DisasContext *s, arg_2scalar *a) \ + { \ + static NeonGenTwoOpWidenFn * const opfn[] = { \ + NULL, \ + gen_helper_neon_##MULL##16, \ + gen_##MULL##32, \ + NULL, \ + }; \ + static NeonGenTwo64OpFn * const accfn[] = { \ + NULL, \ + gen_helper_neon_##ACC##l_u32, \ + tcg_gen_##ACC##_i64, \ + NULL, \ + }; \ + return do_2scalar_long(s, a, opfn[a->size], accfn[a->size]); \ + } + +DO_VMLAL_2SC(VMLAL_S, mull_s, add) +DO_VMLAL_2SC(VMLAL_U, mull_u, add) +DO_VMLAL_2SC(VMLSL_S, mull_s, sub) +DO_VMLAL_2SC(VMLSL_U, mull_u, sub) + +static bool trans_VQDMULL_2sc(DisasContext *s, arg_2scalar *a) +{ + static NeonGenTwoOpWidenFn * const opfn[] = { + NULL, + gen_VQDMULL_16, + gen_VQDMULL_32, + NULL, + }; + + return do_2scalar_long(s, a, opfn[a->size], NULL); +} + +static bool trans_VQDMLAL_2sc(DisasContext *s, arg_2scalar *a) +{ + static NeonGenTwoOpWidenFn * const opfn[] = { + NULL, + gen_VQDMULL_16, + gen_VQDMULL_32, + NULL, + }; + static NeonGenTwo64OpFn * const accfn[] = { + NULL, + gen_VQDMLAL_acc_16, + gen_VQDMLAL_acc_32, + NULL, + }; + + return do_2scalar_long(s, a, opfn[a->size], accfn[a->size]); +} + +static bool trans_VQDMLSL_2sc(DisasContext *s, arg_2scalar *a) +{ + static NeonGenTwoOpWidenFn * const opfn[] = { + NULL, + gen_VQDMULL_16, + gen_VQDMULL_32, + NULL, + }; + static NeonGenTwo64OpFn * const accfn[] = { + NULL, + gen_VQDMLSL_acc_16, + gen_VQDMLSL_acc_32, + NULL, + }; + + return do_2scalar_long(s, a, opfn[a->size], accfn[a->size]); +} diff --git a/target/arm/translate.c b/target/arm/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate.c +++ b/target/arm/translate.c @@ -XXX,XX +XXX,XX @@ static void gen_revsh(TCGv_i32 dest, TCGv_i32 var) tcg_gen_ext16s_i32(dest, var); } -/* 32x32->64 multiply. Marks inputs as dead. */ -static TCGv_i64 gen_mulu_i64_i32(TCGv_i32 a, TCGv_i32 b) -{ - TCGv_i32 lo = tcg_temp_new_i32(); - TCGv_i32 hi = tcg_temp_new_i32(); - TCGv_i64 ret; - - tcg_gen_mulu2_i32(lo, hi, a, b); - tcg_temp_free_i32(a); - tcg_temp_free_i32(b); - - ret = tcg_temp_new_i64(); - tcg_gen_concat_i32_i64(ret, lo, hi); - tcg_temp_free_i32(lo); - tcg_temp_free_i32(hi); - - return ret; -} - -static TCGv_i64 gen_muls_i64_i32(TCGv_i32 a, TCGv_i32 b) -{ - TCGv_i32 lo = tcg_temp_new_i32(); - TCGv_i32 hi = tcg_temp_new_i32(); - TCGv_i64 ret; - - tcg_gen_muls2_i32(lo, hi, a, b); - tcg_temp_free_i32(a); - tcg_temp_free_i32(b); - - ret = tcg_temp_new_i64(); - tcg_gen_concat_i32_i64(ret, lo, hi); - tcg_temp_free_i32(lo); - tcg_temp_free_i32(hi); - - return ret; -} - /* Swap low and high halfwords. */ static void gen_swap_half(TCGv_i32 var) { @@ -XXX,XX +XXX,XX @@ static inline void gen_neon_addl(int size) } } -static inline void gen_neon_negl(TCGv_i64 var, int size) -{ - switch (size) { - case 0: gen_helper_neon_negl_u16(var, var); break; - case 1: gen_helper_neon_negl_u32(var, var); break; - case 2: - tcg_gen_neg_i64(var, var); - break; - default: abort(); - } -} - -static inline void gen_neon_addl_saturate(TCGv_i64 op0, TCGv_i64 op1, int size) -{ - switch (size) { - case 1: gen_helper_neon_addl_saturate_s32(op0, cpu_env, op0, op1); break; - case 2: gen_helper_neon_addl_saturate_s64(op0, cpu_env, op0, op1); break; - default: abort(); - } -} - -static inline void gen_neon_mull(TCGv_i64 dest, TCGv_i32 a, TCGv_i32 b, - int size, int u) -{ - TCGv_i64 tmp; - - switch ((size << 1) | u) { - case 0: gen_helper_neon_mull_s8(dest, a, b); break; - case 1: gen_helper_neon_mull_u8(dest, a, b); break; - case 2: gen_helper_neon_mull_s16(dest, a, b); break; - case 3: gen_helper_neon_mull_u16(dest, a, b); break; - case 4: - tmp = gen_muls_i64_i32(a, b); - tcg_gen_mov_i64(dest, tmp); - tcg_temp_free_i64(tmp); - break; - case 5: - tmp = gen_mulu_i64_i32(a, b); - tcg_gen_mov_i64(dest, tmp); - tcg_temp_free_i64(tmp); - break; - default: abort(); - } - - /* gen_helper_neon_mull_[su]{8|16} do not free their parameters. - Don't forget to clean them now. */ - if (size < 2) { - tcg_temp_free_i32(a); - tcg_temp_free_i32(b); - } -} - static void gen_neon_narrow_op(int op, int u, int size, TCGv_i32 dest, TCGv_i64 src) { @@ -XXX,XX +XXX,XX @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn) int u; int vec_size; uint32_t imm; - TCGv_i32 tmp, tmp2, tmp3, tmp4, tmp5; + TCGv_i32 tmp, tmp2, tmp3, tmp5; TCGv_ptr ptr1; TCGv_i64 tmp64; @@ -XXX,XX +XXX,XX @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn) return 1; } else { /* (insn & 0x00800010 == 0x00800000) */ if (size != 3) { - op = (insn >> 8) & 0xf; - if ((insn & (1 << 6)) == 0) { - /* Three registers of different lengths: handled by decodetree */ - return 1; - } else { - /* Two registers and a scalar. NB that for ops of this form - * the ARM ARM labels bit 24 as Q, but it is in our variable - * 'u', not 'q'. - */ - if (size == 0) { - return 1; - } - switch (op) { - case 0: /* Integer VMLA scalar */ - case 4: /* Integer VMLS scalar */ - case 8: /* Integer VMUL scalar */ - case 1: /* Float VMLA scalar */ - case 5: /* Floating point VMLS scalar */ - case 9: /* Floating point VMUL scalar */ - case 12: /* VQDMULH scalar */ - case 13: /* VQRDMULH scalar */ - case 14: /* VQRDMLAH scalar */ - case 15: /* VQRDMLSH scalar */ - return 1; /* handled by decodetree */ - - case 3: /* VQDMLAL scalar */ - case 7: /* VQDMLSL scalar */ - case 11: /* VQDMULL scalar */ - if (u == 1) { - return 1; - } - /* fall through */ - case 2: /* VMLAL sclar */ - case 6: /* VMLSL scalar */ - case 10: /* VMULL scalar */ - if (rd & 1) { - return 1; - } - tmp2 = neon_get_scalar(size, rm); - /* We need a copy of tmp2 because gen_neon_mull - * deletes it during pass 0. */ - tmp4 = tcg_temp_new_i32(); - tcg_gen_mov_i32(tmp4, tmp2); - tmp3 = neon_load_reg(rn, 1); - - for (pass = 0; pass < 2; pass++) { - if (pass == 0) { - tmp = neon_load_reg(rn, 0); - } else { - tmp = tmp3; - tmp2 = tmp4; - } - gen_neon_mull(cpu_V0, tmp, tmp2, size, u); - if (op != 11) { - neon_load_reg64(cpu_V1, rd + pass); - } - switch (op) { - case 6: - gen_neon_negl(cpu_V0, size); - /* Fall through */ - case 2: - gen_neon_addl(size); - break; - case 3: case 7: - gen_neon_addl_saturate(cpu_V0, cpu_V0, size); - if (op == 7) { - gen_neon_negl(cpu_V0, size); - } - gen_neon_addl_saturate(cpu_V0, cpu_V1, size); - break; - case 10: - /* no-op */ - break; - case 11: - gen_neon_addl_saturate(cpu_V0, cpu_V0, size); - break; - default: - abort(); - } - neon_store_reg64(cpu_V0, rd + pass); - } - break; - default: - g_assert_not_reached(); - } - } + /* + * Three registers of different lengths, or two registers and + * a scalar: handled by decodetree + */ + return 1; } else { /* size == 3 */ if (!u) { /* Extract. */ -- 2.20.1 Convert the Neon VEXT insn to decodetree. Rather than keeping the old implementation which used fixed temporaries cpu_V0 and cpu_V1 and did the extraction with by-hand shift and logic ops, we use the TCG extract2 insn. We don't need to special case 0 or 8 immediates any more as the optimizer is smart enough to throw away the dead code. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/neon-dp.decode | 8 +++- target/arm/translate-neon.inc.c | 76 +++++++++++++++++++++++++++++++++ target/arm/translate.c | 58 +------------------------ 3 files changed, 85 insertions(+), 57 deletions(-) diff --git a/target/arm/neon-dp.decode b/target/arm/neon-dp.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/neon-dp.decode +++ b/target/arm/neon-dp.decode @@ -XXX,XX +XXX,XX @@ Vimm_1r 1111 001 . 1 . 000 ... .... cmode:4 0 . op:1 1 .... @1reg_imm # return false for size==3. ###################################################################### { - # 0b11 subgroup will go here + [ + ################################################################## + # Miscellaneous size=0b11 insns + ################################################################## + VEXT 1111 001 0 1 . 11 .... .... imm:4 . q:1 . 0 .... \ + vm=%vm_dp vn=%vn_dp vd=%vd_dp + ] # Subgroup for size != 0b11 [ diff --git a/target/arm/translate-neon.inc.c b/target/arm/translate-neon.inc.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-neon.inc.c +++ b/target/arm/translate-neon.inc.c @@ -XXX,XX +XXX,XX @@ static bool trans_VQDMLSL_2sc(DisasContext *s, arg_2scalar *a) return do_2scalar_long(s, a, opfn[a->size], accfn[a->size]); } + +static bool trans_VEXT(DisasContext *s, arg_VEXT *a) +{ + if (!arm_dc_feature(s, ARM_FEATURE_NEON)) { + return false; + } + + /* UNDEF accesses to D16-D31 if they don't exist. */ + if (!dc_isar_feature(aa32_simd_r32, s) && + ((a->vd | a->vn | a->vm) & 0x10)) { + return false; + } + + if ((a->vn | a->vm | a->vd) & a->q) { + return false; + } + + if (a->imm > 7 && !a->q) { + return false; + } + + if (!vfp_access_check(s)) { + return true; + } + + if (!a->q) { + /* Extract 64 bits from <Vm:Vn> */ + TCGv_i64 left, right, dest; + + left = tcg_temp_new_i64(); + right = tcg_temp_new_i64(); + dest = tcg_temp_new_i64(); + + neon_load_reg64(right, a->vn); + neon_load_reg64(left, a->vm); + tcg_gen_extract2_i64(dest, right, left, a->imm * 8); + neon_store_reg64(dest, a->vd); + + tcg_temp_free_i64(left); + tcg_temp_free_i64(right); + tcg_temp_free_i64(dest); + } else { + /* Extract 128 bits from <Vm+1:Vm:Vn+1:Vn> */ + TCGv_i64 left, middle, right, destleft, destright; + + left = tcg_temp_new_i64(); + middle = tcg_temp_new_i64(); + right = tcg_temp_new_i64(); + destleft = tcg_temp_new_i64(); + destright = tcg_temp_new_i64(); + + if (a->imm < 8) { + neon_load_reg64(right, a->vn); + neon_load_reg64(middle, a->vn + 1); + tcg_gen_extract2_i64(destright, right, middle, a->imm * 8); + neon_load_reg64(left, a->vm); + tcg_gen_extract2_i64(destleft, middle, left, a->imm * 8); + } else { + neon_load_reg64(right, a->vn + 1); + neon_load_reg64(middle, a->vm); + tcg_gen_extract2_i64(destright, right, middle, (a->imm - 8) * 8); + neon_load_reg64(left, a->vm + 1); + tcg_gen_extract2_i64(destleft, middle, left, (a->imm - 8) * 8); + } + + neon_store_reg64(destright, a->vd); + neon_store_reg64(destleft, a->vd + 1); + + tcg_temp_free_i64(destright); + tcg_temp_free_i64(destleft); + tcg_temp_free_i64(right); + tcg_temp_free_i64(middle); + tcg_temp_free_i64(left); + } + return true; +} diff --git a/target/arm/translate.c b/target/arm/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate.c +++ b/target/arm/translate.c @@ -XXX,XX +XXX,XX @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn) int pass; int u; int vec_size; - uint32_t imm; TCGv_i32 tmp, tmp2, tmp3, tmp5; TCGv_ptr ptr1; - TCGv_i64 tmp64; if (!arm_dc_feature(s, ARM_FEATURE_NEON)) { return 1; @@ -XXX,XX +XXX,XX @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn) return 1; } else { /* size == 3 */ if (!u) { - /* Extract. */ - imm = (insn >> 8) & 0xf; - - if (imm > 7 && !q) - return 1; - - if (q && ((rd | rn | rm) & 1)) { - return 1; - } - - if (imm == 0) { - neon_load_reg64(cpu_V0, rn); - if (q) { - neon_load_reg64(cpu_V1, rn + 1); - } - } else if (imm == 8) { - neon_load_reg64(cpu_V0, rn + 1); - if (q) { - neon_load_reg64(cpu_V1, rm); - } - } else if (q) { - tmp64 = tcg_temp_new_i64(); - if (imm < 8) { - neon_load_reg64(cpu_V0, rn); - neon_load_reg64(tmp64, rn + 1); - } else { - neon_load_reg64(cpu_V0, rn + 1); - neon_load_reg64(tmp64, rm); - } - tcg_gen_shri_i64(cpu_V0, cpu_V0, (imm & 7) * 8); - tcg_gen_shli_i64(cpu_V1, tmp64, 64 - ((imm & 7) * 8)); - tcg_gen_or_i64(cpu_V0, cpu_V0, cpu_V1); - if (imm < 8) { - neon_load_reg64(cpu_V1, rm); - } else { - neon_load_reg64(cpu_V1, rm + 1); - imm -= 8; - } - tcg_gen_shli_i64(cpu_V1, cpu_V1, 64 - (imm * 8)); - tcg_gen_shri_i64(tmp64, tmp64, imm * 8); - tcg_gen_or_i64(cpu_V1, cpu_V1, tmp64); - tcg_temp_free_i64(tmp64); - } else { - /* BUGFIX */ - neon_load_reg64(cpu_V0, rn); - tcg_gen_shri_i64(cpu_V0, cpu_V0, imm * 8); - neon_load_reg64(cpu_V1, rm); - tcg_gen_shli_i64(cpu_V1, cpu_V1, 64 - (imm * 8)); - tcg_gen_or_i64(cpu_V0, cpu_V0, cpu_V1); - } - neon_store_reg64(cpu_V0, rd); - if (q) { - neon_store_reg64(cpu_V1, rd + 1); - } + /* Extract: handled by decodetree */ + return 1; } else if ((insn & (1 << 11)) == 0) { /* Two register misc. */ op = ((insn >> 12) & 0x30) | ((insn >> 7) & 0xf); -- 2.20.1 Convert the Neon VTBL, VTBX instructions to decodetree. The actual implementation of the insn is copied across to the new trans function unchanged except for renaming 'tmp5' to 'tmp4'. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/neon-dp.decode | 3 ++ target/arm/translate-neon.inc.c | 56 +++++++++++++++++++++++++++++++++ target/arm/translate.c | 41 +++--------------------- 3 files changed, 63 insertions(+), 37 deletions(-) diff --git a/target/arm/neon-dp.decode b/target/arm/neon-dp.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/neon-dp.decode +++ b/target/arm/neon-dp.decode @@ -XXX,XX +XXX,XX @@ Vimm_1r 1111 001 . 1 . 000 ... .... cmode:4 0 . op:1 1 .... @1reg_imm ################################################################## VEXT 1111 001 0 1 . 11 .... .... imm:4 . q:1 . 0 .... \ vm=%vm_dp vn=%vn_dp vd=%vd_dp + + VTBL 1111 001 1 1 . 11 .... .... 10 len:2 . op:1 . 0 .... \ + vm=%vm_dp vn=%vn_dp vd=%vd_dp ] # Subgroup for size != 0b11 diff --git a/target/arm/translate-neon.inc.c b/target/arm/translate-neon.inc.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-neon.inc.c +++ b/target/arm/translate-neon.inc.c @@ -XXX,XX +XXX,XX @@ static bool trans_VEXT(DisasContext *s, arg_VEXT *a) } return true; } + +static bool trans_VTBL(DisasContext *s, arg_VTBL *a) +{ + int n; + TCGv_i32 tmp, tmp2, tmp3, tmp4; + TCGv_ptr ptr1; + + if (!arm_dc_feature(s, ARM_FEATURE_NEON)) { + return false; + } + + /* UNDEF accesses to D16-D31 if they don't exist. */ + if (!dc_isar_feature(aa32_simd_r32, s) && + ((a->vd | a->vn | a->vm) & 0x10)) { + return false; + } + + if (!vfp_access_check(s)) { + return true; + } + + n = a->len + 1; + if ((a->vn + n) > 32) { + /* + * This is UNPREDICTABLE; we choose to UNDEF to avoid the + * helper function running off the end of the register file. + */ + return false; + } + n <<= 3; + if (a->op) { + tmp = neon_load_reg(a->vd, 0); + } else { + tmp = tcg_temp_new_i32(); + tcg_gen_movi_i32(tmp, 0); + } + tmp2 = neon_load_reg(a->vm, 0); + ptr1 = vfp_reg_ptr(true, a->vn); + tmp4 = tcg_const_i32(n); + gen_helper_neon_tbl(tmp2, tmp2, tmp, ptr1, tmp4); + tcg_temp_free_i32(tmp); + if (a->op) { + tmp = neon_load_reg(a->vd, 1); + } else { + tmp = tcg_temp_new_i32(); + tcg_gen_movi_i32(tmp, 0); + } + tmp3 = neon_load_reg(a->vm, 1); + gen_helper_neon_tbl(tmp3, tmp3, tmp, ptr1, tmp4); + tcg_temp_free_i32(tmp4); + tcg_temp_free_ptr(ptr1); + neon_store_reg(a->vd, 0, tmp2); + neon_store_reg(a->vd, 1, tmp3); + tcg_temp_free_i32(tmp); + return true; +} diff --git a/target/arm/translate.c b/target/arm/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate.c +++ b/target/arm/translate.c @@ -XXX,XX +XXX,XX @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn) { int op; int q; - int rd, rn, rm, rd_ofs, rm_ofs; + int rd, rm, rd_ofs, rm_ofs; int size; int pass; int u; int vec_size; - TCGv_i32 tmp, tmp2, tmp3, tmp5; - TCGv_ptr ptr1; + TCGv_i32 tmp, tmp2, tmp3; if (!arm_dc_feature(s, ARM_FEATURE_NEON)) { return 1; @@ -XXX,XX +XXX,XX @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn) q = (insn & (1 << 6)) != 0; u = (insn >> 24) & 1; VFP_DREG_D(rd, insn); - VFP_DREG_N(rn, insn); VFP_DREG_M(rm, insn); size = (insn >> 20) & 3; vec_size = q ? 16 : 8; @@ -XXX,XX +XXX,XX @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn) break; } } else if ((insn & (1 << 10)) == 0) { - /* VTBL, VTBX. */ - int n = ((insn >> 8) & 3) + 1; - if ((rn + n) > 32) { - /* This is UNPREDICTABLE; we choose to UNDEF to avoid the - * helper function running off the end of the register file. - */ - return 1; - } - n <<= 3; - if (insn & (1 << 6)) { - tmp = neon_load_reg(rd, 0); - } else { - tmp = tcg_temp_new_i32(); - tcg_gen_movi_i32(tmp, 0); - } - tmp2 = neon_load_reg(rm, 0); - ptr1 = vfp_reg_ptr(true, rn); - tmp5 = tcg_const_i32(n); - gen_helper_neon_tbl(tmp2, tmp2, tmp, ptr1, tmp5); - tcg_temp_free_i32(tmp); - if (insn & (1 << 6)) { - tmp = neon_load_reg(rd, 1); - } else { - tmp = tcg_temp_new_i32(); - tcg_gen_movi_i32(tmp, 0); - } - tmp3 = neon_load_reg(rm, 1); - gen_helper_neon_tbl(tmp3, tmp3, tmp, ptr1, tmp5); - tcg_temp_free_i32(tmp5); - tcg_temp_free_ptr(ptr1); - neon_store_reg(rd, 0, tmp2); - neon_store_reg(rd, 1, tmp3); - tcg_temp_free_i32(tmp); + /* VTBL, VTBX: handled by decodetree */ + return 1; } else if ((insn & 0x380) == 0) { /* VDUP */ int element; -- 2.20.1 Convert the Neon VDUP (scalar) insn to decodetree. (Note that we can't call this just "VDUP" as we used that already in vfp.decode for the "VDUP (general purpose register" insn.) Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/neon-dp.decode | 7 +++++++ target/arm/translate-neon.inc.c | 26 ++++++++++++++++++++++++++ target/arm/translate.c | 25 +------------------------ 3 files changed, 34 insertions(+), 24 deletions(-) diff --git a/target/arm/neon-dp.decode b/target/arm/neon-dp.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/neon-dp.decode +++ b/target/arm/neon-dp.decode @@ -XXX,XX +XXX,XX @@ Vimm_1r 1111 001 . 1 . 000 ... .... cmode:4 0 . op:1 1 .... @1reg_imm VTBL 1111 001 1 1 . 11 .... .... 10 len:2 . op:1 . 0 .... \ vm=%vm_dp vn=%vn_dp vd=%vd_dp + + VDUP_scalar 1111 001 1 1 . 11 index:3 1 .... 11 000 q:1 . 0 .... \ + vm=%vm_dp vd=%vd_dp size=0 + VDUP_scalar 1111 001 1 1 . 11 index:2 10 .... 11 000 q:1 . 0 .... \ + vm=%vm_dp vd=%vd_dp size=1 + VDUP_scalar 1111 001 1 1 . 11 index:1 100 .... 11 000 q:1 . 0 .... \ + vm=%vm_dp vd=%vd_dp size=2 ] # Subgroup for size != 0b11 diff --git a/target/arm/translate-neon.inc.c b/target/arm/translate-neon.inc.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-neon.inc.c +++ b/target/arm/translate-neon.inc.c @@ -XXX,XX +XXX,XX @@ static bool trans_VTBL(DisasContext *s, arg_VTBL *a) tcg_temp_free_i32(tmp); return true; } + +static bool trans_VDUP_scalar(DisasContext *s, arg_VDUP_scalar *a) +{ + if (!arm_dc_feature(s, ARM_FEATURE_NEON)) { + return false; + } + + /* UNDEF accesses to D16-D31 if they don't exist. */ + if (!dc_isar_feature(aa32_simd_r32, s) && + ((a->vd | a->vm) & 0x10)) { + return false; + } + + if (a->vd & a->q) { + return false; + } + + if (!vfp_access_check(s)) { + return true; + } + + tcg_gen_gvec_dup_mem(a->size, neon_reg_offset(a->vd, 0), + neon_element_offset(a->vm, a->index, a->size), + a->q ? 16 : 8, a->q ? 16 : 8); + return true; +} diff --git a/target/arm/translate.c b/target/arm/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate.c +++ b/target/arm/translate.c @@ -XXX,XX +XXX,XX @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn) } break; } - } else if ((insn & (1 << 10)) == 0) { - /* VTBL, VTBX: handled by decodetree */ - return 1; - } else if ((insn & 0x380) == 0) { - /* VDUP */ - int element; - MemOp size; - - if ((insn & (7 << 16)) == 0 || (q && (rd & 1))) { - return 1; - } - if (insn & (1 << 16)) { - size = MO_8; - element = (insn >> 17) & 7; - } else if (insn & (1 << 17)) { - size = MO_16; - element = (insn >> 18) & 3; - } else { - size = MO_32; - element = (insn >> 19) & 1; - } - tcg_gen_gvec_dup_mem(size, neon_reg_offset(rd, 0), - neon_element_offset(rm, element, size), - q ? 16 : 8, q ? 16 : 8); } else { + /* VTBL, VTBX, VDUP: handled by decodetree */ return 1; } } -- 2.20.1 From: Jean-Christophe Dubois <jcd@tribudubois.net> Some bits of the CCM registers are non writable. This was left undone in the initial commit (all bits of registers were writable). This patch adds the required code to protect the non writable bits. Signed-off-by: Jean-Christophe Dubois <jcd@tribudubois.net> Message-id: 20200608133508.550046-1-jcd@tribudubois.net Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/misc/imx6ul_ccm.c | 76 ++++++++++++++++++++++++++++++++++++-------- 1 file changed, 63 insertions(+), 13 deletions(-) diff --git a/hw/misc/imx6ul_ccm.c b/hw/misc/imx6ul_ccm.c index XXXXXXX..XXXXXXX 100644 --- a/hw/misc/imx6ul_ccm.c +++ b/hw/misc/imx6ul_ccm.c @@ -XXX,XX +XXX,XX @@ #include "trace.h" +static const uint32_t ccm_mask[CCM_MAX] = { + [CCM_CCR] = 0xf01fef80, + [CCM_CCDR] = 0xfffeffff, + [CCM_CSR] = 0xffffffff, + [CCM_CCSR] = 0xfffffef2, + [CCM_CACRR] = 0xfffffff8, + [CCM_CBCDR] = 0xc1f8e000, + [CCM_CBCMR] = 0xfc03cfff, + [CCM_CSCMR1] = 0x80700000, + [CCM_CSCMR2] = 0xe01ff003, + [CCM_CSCDR1] = 0xfe00c780, + [CCM_CS1CDR] = 0xfe00fe00, + [CCM_CS2CDR] = 0xf8007000, + [CCM_CDCDR] = 0xf00fffff, + [CCM_CHSCCDR] = 0xfffc01ff, + [CCM_CSCDR2] = 0xfe0001ff, + [CCM_CSCDR3] = 0xffffc1ff, + [CCM_CDHIPR] = 0xffffffff, + [CCM_CTOR] = 0x00000000, + [CCM_CLPCR] = 0xf39ff01c, + [CCM_CISR] = 0xfb85ffbe, + [CCM_CIMR] = 0xfb85ffbf, + [CCM_CCOSR] = 0xfe00fe00, + [CCM_CGPR] = 0xfffc3fea, + [CCM_CCGR0] = 0x00000000, + [CCM_CCGR1] = 0x00000000, + [CCM_CCGR2] = 0x00000000, + [CCM_CCGR3] = 0x00000000, + [CCM_CCGR4] = 0x00000000, + [CCM_CCGR5] = 0x00000000, + [CCM_CCGR6] = 0x00000000, + [CCM_CMEOR] = 0xafffff1f, +}; + +static const uint32_t analog_mask[CCM_ANALOG_MAX] = { + [CCM_ANALOG_PLL_ARM] = 0xfff60f80, + [CCM_ANALOG_PLL_USB1] = 0xfffe0fbc, + [CCM_ANALOG_PLL_USB2] = 0xfffe0fbc, + [CCM_ANALOG_PLL_SYS] = 0xfffa0ffe, + [CCM_ANALOG_PLL_SYS_SS] = 0x00000000, + [CCM_ANALOG_PLL_SYS_NUM] = 0xc0000000, + [CCM_ANALOG_PLL_SYS_DENOM] = 0xc0000000, + [CCM_ANALOG_PLL_AUDIO] = 0xffe20f80, + [CCM_ANALOG_PLL_AUDIO_NUM] = 0xc0000000, + [CCM_ANALOG_PLL_AUDIO_DENOM] = 0xc0000000, + [CCM_ANALOG_PLL_VIDEO] = 0xffe20f80, + [CCM_ANALOG_PLL_VIDEO_NUM] = 0xc0000000, + [CCM_ANALOG_PLL_VIDEO_DENOM] = 0xc0000000, + [CCM_ANALOG_PLL_ENET] = 0xffc20ff0, + [CCM_ANALOG_PFD_480] = 0x40404040, + [CCM_ANALOG_PFD_528] = 0x40404040, + [PMU_MISC0] = 0x01fe8306, + [PMU_MISC1] = 0x07fcede0, + [PMU_MISC2] = 0x005f5f5f, +}; + static const char *imx6ul_ccm_reg_name(uint32_t reg) { static char unknown[20]; @@ -XXX,XX +XXX,XX @@ static void imx6ul_ccm_write(void *opaque, hwaddr offset, uint64_t value, trace_ccm_write_reg(imx6ul_ccm_reg_name(index), (uint32_t)value); - /* - * We will do a better implementation later. In particular some bits - * cannot be written to. - */ - s->ccm[index] = (uint32_t)value; + s->ccm[index] = (s->ccm[index] & ccm_mask[index]) | + ((uint32_t)value & ~ccm_mask[index]); } static uint64_t imx6ul_analog_read(void *opaque, hwaddr offset, unsigned size) @@ -XXX,XX +XXX,XX @@ static void imx6ul_analog_write(void *opaque, hwaddr offset, uint64_t value, * the REG_NAME register. So we change the value of the * REG_NAME register, setting bits passed in the value. */ - s->analog[index - 1] |= value; + s->analog[index - 1] |= (value & ~analog_mask[index - 1]); break; case CCM_ANALOG_PLL_ARM_CLR: case CCM_ANALOG_PLL_USB1_CLR: @@ -XXX,XX +XXX,XX @@ static void imx6ul_analog_write(void *opaque, hwaddr offset, uint64_t value, * the REG_NAME register. So we change the value of the * REG_NAME register, unsetting bits passed in the value. */ - s->analog[index - 2] &= ~value; + s->analog[index - 2] &= ~(value & ~analog_mask[index - 2]); break; case CCM_ANALOG_PLL_ARM_TOG: case CCM_ANALOG_PLL_USB1_TOG: @@ -XXX,XX +XXX,XX @@ static void imx6ul_analog_write(void *opaque, hwaddr offset, uint64_t value, * the REG_NAME register. So we change the value of the * REG_NAME register, toggling bits passed in the value. */ - s->analog[index - 3] ^= value; + s->analog[index - 3] ^= (value & ~analog_mask[index - 3]); break; default: - /* - * We will do a better implementation later. In particular some bits - * cannot be written to. - */ - s->analog[index] = value; + s->analog[index] = (s->analog[index] & analog_mask[index]) | + (value & ~analog_mask[index]); break; } } -- 2.20.1 From: Erik Smit <erik.lucas.smit@gmail.com> The hardware supports configurable descriptor sizes, configured in the DBLAC register. Most drivers use the default 4 word descriptor, which is currently hardcoded, but Aspeed SDK configures 8 words to store extra data. Signed-off-by: Erik Smit <erik.lucas.smit@gmail.com> Reviewed-by: Cédric Le Goater <clg@kaod.org> [PMM: removed unnecessary parens] Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/net/ftgmac100.c | 26 ++++++++++++++++++++++++-- 1 file changed, 24 insertions(+), 2 deletions(-) diff --git a/hw/net/ftgmac100.c b/hw/net/ftgmac100.c index XXXXXXX..XXXXXXX 100644 --- a/hw/net/ftgmac100.c +++ b/hw/net/ftgmac100.c @@ -XXX,XX +XXX,XX @@ #define FTGMAC100_APTC_TXPOLL_CNT(x) (((x) >> 8) & 0xf) #define FTGMAC100_APTC_TXPOLL_TIME_SEL (1 << 12) +/* + * DMA burst length and arbitration control register + */ +#define FTGMAC100_DBLAC_RXBURST_SIZE(x) (((x) >> 8) & 0x3) +#define FTGMAC100_DBLAC_TXBURST_SIZE(x) (((x) >> 10) & 0x3) +#define FTGMAC100_DBLAC_RXDES_SIZE(x) ((((x) >> 12) & 0xf) * 8) +#define FTGMAC100_DBLAC_TXDES_SIZE(x) ((((x) >> 16) & 0xf) * 8) +#define FTGMAC100_DBLAC_IFG_CNT(x) (((x) >> 20) & 0x7) +#define FTGMAC100_DBLAC_IFG_INC (1 << 23) + /* * PHY control register */ @@ -XXX,XX +XXX,XX @@ static void ftgmac100_do_tx(FTGMAC100State *s, uint32_t tx_ring, if (bd.des0 & s->txdes0_edotr) { addr = tx_ring; } else { - addr += sizeof(FTGMAC100Desc); + addr += FTGMAC100_DBLAC_TXDES_SIZE(s->dblac); } } @@ -XXX,XX +XXX,XX @@ static void ftgmac100_write(void *opaque, hwaddr addr, s->phydata = value & 0xffff; break; case FTGMAC100_DBLAC: /* DMA Burst Length and Arbitration Control */ + if (FTGMAC100_DBLAC_TXDES_SIZE(s->dblac) < sizeof(FTGMAC100Desc)) { + qemu_log_mask(LOG_GUEST_ERROR, + "%s: transmit descriptor too small : %d bytes\n", + __func__, FTGMAC100_DBLAC_TXDES_SIZE(s->dblac)); + break; + } + if (FTGMAC100_DBLAC_RXDES_SIZE(s->dblac) < sizeof(FTGMAC100Desc)) { + qemu_log_mask(LOG_GUEST_ERROR, + "%s: receive descriptor too small : %d bytes\n", + __func__, FTGMAC100_DBLAC_RXDES_SIZE(s->dblac)); + break; + } s->dblac = value; break; case FTGMAC100_REVR: /* Feature Register */ @@ -XXX,XX +XXX,XX @@ static ssize_t ftgmac100_receive(NetClientState *nc, const uint8_t *buf, if (bd.des0 & s->rxdes0_edorr) { addr = s->rx_ring; } else { - addr += sizeof(FTGMAC100Desc); + addr += FTGMAC100_DBLAC_RXDES_SIZE(s->dblac); } } s->rx_descriptor = addr; -- 2.20.1 From: fangying <fangying1@huawei.com> Virtual time adjustment was implemented for virt-5.0 machine type, but the cpu property was enabled only for host-passthrough and max cpu model. Let's add it for any KVM arm cpu which has the generic timer feature enabled. Signed-off-by: Ying Fang <fangying1@huawei.com> Reviewed-by: Andrew Jones <drjones@redhat.com> Message-id: 20200608121243.2076-1-fangying1@huawei.com [PMM: minor commit message tweak, removed inaccurate suggested-by tag] Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/cpu.c | 6 ++++-- target/arm/cpu64.c | 1 - target/arm/kvm.c | 21 +++++++++++---------- 3 files changed, 15 insertions(+), 13 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 @@ void arm_cpu_post_init(Object *obj) if (arm_feature(&cpu->env, ARM_FEATURE_GENERIC_TIMER)) { qdev_property_add_static(DEVICE(cpu), &arm_cpu_gt_cntfrq_property); } + + if (kvm_enabled()) { + kvm_arm_add_vcpu_properties(obj); + } } static void arm_cpu_finalizefn(Object *obj) @@ -XXX,XX +XXX,XX @@ static void arm_max_initfn(Object *obj) if (kvm_enabled()) { kvm_arm_set_cpu_features_from_host(cpu); - kvm_arm_add_vcpu_properties(obj); } else { cortex_a15_initfn(obj); @@ -XXX,XX +XXX,XX @@ static void arm_host_initfn(Object *obj) if (arm_feature(&cpu->env, ARM_FEATURE_AARCH64)) { aarch64_add_sve_properties(obj); } - kvm_arm_add_vcpu_properties(obj); arm_cpu_post_init(obj); } 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_max_initfn(Object *obj) if (kvm_enabled()) { kvm_arm_set_cpu_features_from_host(cpu); - kvm_arm_add_vcpu_properties(obj); } else { uint64_t t; uint32_t u; 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 @@ static void kvm_no_adjvtime_set(Object *obj, bool value, Error **errp) /* KVM VCPU properties should be prefixed with "kvm-". */ void kvm_arm_add_vcpu_properties(Object *obj) { - if (!kvm_enabled()) { - return; - } + ARMCPU *cpu = ARM_CPU(obj); + CPUARMState *env = &cpu->env; - ARM_CPU(obj)->kvm_adjvtime = true; - object_property_add_bool(obj, "kvm-no-adjvtime", kvm_no_adjvtime_get, - kvm_no_adjvtime_set); - object_property_set_description(obj, "kvm-no-adjvtime", - "Set on to disable the adjustment of " - "the virtual counter. VM stopped time " - "will be counted."); + if (arm_feature(env, ARM_FEATURE_GENERIC_TIMER)) { + cpu->kvm_adjvtime = true; + object_property_add_bool(obj, "kvm-no-adjvtime", kvm_no_adjvtime_get, + kvm_no_adjvtime_set); + object_property_set_description(obj, "kvm-no-adjvtime", + "Set on to disable the adjustment of " + "the virtual counter. VM stopped time " + "will be counted."); + } } bool kvm_arm_pmu_supported(CPUState *cpu) -- 2.20.1 From: Jean-Christophe Dubois <jcd@tribudubois.net> Signed-off-by: Jean-Christophe Dubois <jcd@tribudubois.net> Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org> Tested-by: Philippe Mathieu-Daudé <f4bug@amsat.org> [PMD: Fixed 32-bit format string using PRIx32/PRIx64] Signed-off-by: Philippe Mathieu-Daudé <f4bug@amsat.org> Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/net/imx_fec.c | 106 +++++++++++++++++++------------------------- hw/net/trace-events | 18 ++++++++ 2 files changed, 63 insertions(+), 61 deletions(-) diff --git a/hw/net/imx_fec.c b/hw/net/imx_fec.c index XXXXXXX..XXXXXXX 100644 --- a/hw/net/imx_fec.c +++ b/hw/net/imx_fec.c @@ -XXX,XX +XXX,XX @@ #include "qemu/module.h" #include "net/checksum.h" #include "net/eth.h" +#include "trace.h" /* For crc32 */ #include <zlib.h> -#ifndef DEBUG_IMX_FEC -#define DEBUG_IMX_FEC 0 -#endif - -#define FEC_PRINTF(fmt, args...) \ - do { \ - if (DEBUG_IMX_FEC) { \ - fprintf(stderr, "[%s]%s: " fmt , TYPE_IMX_FEC, \ - __func__, ##args); \ - } \ - } while (0) - -#ifndef DEBUG_IMX_PHY -#define DEBUG_IMX_PHY 0 -#endif - -#define PHY_PRINTF(fmt, args...) \ - do { \ - if (DEBUG_IMX_PHY) { \ - fprintf(stderr, "[%s.phy]%s: " fmt , TYPE_IMX_FEC, \ - __func__, ##args); \ - } \ - } while (0) - #define IMX_MAX_DESC 1024 static const char *imx_default_reg_name(IMXFECState *s, uint32_t index) @@ -XXX,XX +XXX,XX @@ static void imx_eth_update(IMXFECState *s); * For now we don't handle any GPIO/interrupt line, so the OS will * have to poll for the PHY status. */ -static void phy_update_irq(IMXFECState *s) +static void imx_phy_update_irq(IMXFECState *s) { imx_eth_update(s); } -static void phy_update_link(IMXFECState *s) +static void imx_phy_update_link(IMXFECState *s) { /* Autonegotiation status mirrors link status. */ if (qemu_get_queue(s->nic)->link_down) { - PHY_PRINTF("link is down\n"); + trace_imx_phy_update_link("down"); s->phy_status &= ~0x0024; s->phy_int |= PHY_INT_DOWN; } else { - PHY_PRINTF("link is up\n"); + trace_imx_phy_update_link("up"); s->phy_status |= 0x0024; s->phy_int |= PHY_INT_ENERGYON; s->phy_int |= PHY_INT_AUTONEG_COMPLETE; } - phy_update_irq(s); + imx_phy_update_irq(s); } static void imx_eth_set_link(NetClientState *nc) { - phy_update_link(IMX_FEC(qemu_get_nic_opaque(nc))); + imx_phy_update_link(IMX_FEC(qemu_get_nic_opaque(nc))); } -static void phy_reset(IMXFECState *s) +static void imx_phy_reset(IMXFECState *s) { + trace_imx_phy_reset(); + s->phy_status = 0x7809; s->phy_control = 0x3000; s->phy_advertise = 0x01e1; s->phy_int_mask = 0; s->phy_int = 0; - phy_update_link(s); + imx_phy_update_link(s); } -static uint32_t do_phy_read(IMXFECState *s, int reg) +static uint32_t imx_phy_read(IMXFECState *s, int reg) { uint32_t val; @@ -XXX,XX +XXX,XX @@ static uint32_t do_phy_read(IMXFECState *s, int reg) case 29: /* Interrupt source. */ val = s->phy_int; s->phy_int = 0; - phy_update_irq(s); + imx_phy_update_irq(s); break; case 30: /* Interrupt mask */ val = s->phy_int_mask; @@ -XXX,XX +XXX,XX @@ static uint32_t do_phy_read(IMXFECState *s, int reg) break; } - PHY_PRINTF("read 0x%04x @ %d\n", val, reg); + trace_imx_phy_read(val, reg); return val; } -static void do_phy_write(IMXFECState *s, int reg, uint32_t val) +static void imx_phy_write(IMXFECState *s, int reg, uint32_t val) { - PHY_PRINTF("write 0x%04x @ %d\n", val, reg); + trace_imx_phy_write(val, reg); if (reg > 31) { /* we only advertise one phy */ @@ -XXX,XX +XXX,XX @@ static void do_phy_write(IMXFECState *s, int reg, uint32_t val) switch (reg) { case 0: /* Basic Control */ if (val & 0x8000) { - phy_reset(s); + imx_phy_reset(s); } else { s->phy_control = val & 0x7980; /* Complete autonegotiation immediately. */ @@ -XXX,XX +XXX,XX @@ static void do_phy_write(IMXFECState *s, int reg, uint32_t val) break; case 30: /* Interrupt mask */ s->phy_int_mask = val & 0xff; - phy_update_irq(s); + imx_phy_update_irq(s); break; case 17: case 18: @@ -XXX,XX +XXX,XX @@ static void do_phy_write(IMXFECState *s, int reg, uint32_t val) static void imx_fec_read_bd(IMXFECBufDesc *bd, dma_addr_t addr) { dma_memory_read(&address_space_memory, addr, bd, sizeof(*bd)); + + trace_imx_fec_read_bd(addr, bd->flags, bd->length, bd->data); } static void imx_fec_write_bd(IMXFECBufDesc *bd, dma_addr_t addr) @@ -XXX,XX +XXX,XX @@ static void imx_fec_write_bd(IMXFECBufDesc *bd, dma_addr_t addr) static void imx_enet_read_bd(IMXENETBufDesc *bd, dma_addr_t addr) { dma_memory_read(&address_space_memory, addr, bd, sizeof(*bd)); + + trace_imx_enet_read_bd(addr, bd->flags, bd->length, bd->data, + bd->option, bd->status); } static void imx_enet_write_bd(IMXENETBufDesc *bd, dma_addr_t addr) @@ -XXX,XX +XXX,XX @@ static void imx_fec_do_tx(IMXFECState *s) int len; imx_fec_read_bd(&bd, addr); - FEC_PRINTF("tx_bd %x flags %04x len %d data %08x\n", - addr, bd.flags, bd.length, bd.data); if ((bd.flags & ENET_BD_R) == 0) { + /* Run out of descriptors to transmit. */ - FEC_PRINTF("tx_bd ran out of descriptors to transmit\n"); + trace_imx_eth_tx_bd_busy(); + break; } len = bd.length; @@ -XXX,XX +XXX,XX @@ static void imx_enet_do_tx(IMXFECState *s, uint32_t index) int len; imx_enet_read_bd(&bd, addr); - FEC_PRINTF("tx_bd %x flags %04x len %d data %08x option %04x " - "status %04x\n", addr, bd.flags, bd.length, bd.data, - bd.option, bd.status); if ((bd.flags & ENET_BD_R) == 0) { /* Run out of descriptors to transmit. */ + + trace_imx_eth_tx_bd_busy(); + break; } len = bd.length; @@ -XXX,XX +XXX,XX @@ static void imx_eth_enable_rx(IMXFECState *s, bool flush) s->regs[ENET_RDAR] = (bd.flags & ENET_BD_E) ? ENET_RDAR_RDAR : 0; if (!s->regs[ENET_RDAR]) { - FEC_PRINTF("RX buffer full\n"); + trace_imx_eth_rx_bd_full(); } else if (flush) { qemu_flush_queued_packets(qemu_get_queue(s->nic)); } @@ -XXX,XX +XXX,XX @@ static void imx_eth_reset(DeviceState *d) memset(s->tx_descriptor, 0, sizeof(s->tx_descriptor)); /* We also reset the PHY */ - phy_reset(s); + imx_phy_reset(s); } static uint32_t imx_default_read(IMXFECState *s, uint32_t index) @@ -XXX,XX +XXX,XX @@ static uint64_t imx_eth_read(void *opaque, hwaddr offset, unsigned size) break; } - FEC_PRINTF("reg[%s] => 0x%" PRIx32 "\n", imx_eth_reg_name(s, index), - value); + trace_imx_eth_read(index, imx_eth_reg_name(s, index), value); return value; } @@ -XXX,XX +XXX,XX @@ static void imx_eth_write(void *opaque, hwaddr offset, uint64_t value, const bool single_tx_ring = !imx_eth_is_multi_tx_ring(s); uint32_t index = offset >> 2; - FEC_PRINTF("reg[%s] <= 0x%" PRIx32 "\n", imx_eth_reg_name(s, index), - (uint32_t)value); + trace_imx_eth_write(index, imx_eth_reg_name(s, index), value); switch (index) { case ENET_EIR: @@ -XXX,XX +XXX,XX @@ static void imx_eth_write(void *opaque, hwaddr offset, uint64_t value, if (extract32(value, 29, 1)) { /* This is a read operation */ s->regs[ENET_MMFR] = deposit32(s->regs[ENET_MMFR], 0, 16, - do_phy_read(s, + imx_phy_read(s, extract32(value, 18, 10))); } else { /* This a write operation */ - do_phy_write(s, extract32(value, 18, 10), extract32(value, 0, 16)); + imx_phy_write(s, extract32(value, 18, 10), extract32(value, 0, 16)); } /* raise the interrupt as the PHY operation is done */ s->regs[ENET_EIR] |= ENET_INT_MII; @@ -XXX,XX +XXX,XX @@ static bool imx_eth_can_receive(NetClientState *nc) { IMXFECState *s = IMX_FEC(qemu_get_nic_opaque(nc)); - FEC_PRINTF("\n"); - return !!s->regs[ENET_RDAR]; } @@ -XXX,XX +XXX,XX @@ static ssize_t imx_fec_receive(NetClientState *nc, const uint8_t *buf, unsigned int buf_len; size_t size = len; - FEC_PRINTF("len %d\n", (int)size); + trace_imx_fec_receive(size); if (!s->regs[ENET_RDAR]) { qemu_log_mask(LOG_GUEST_ERROR, "[%s]%s: Unexpected packet\n", @@ -XXX,XX +XXX,XX @@ static ssize_t imx_fec_receive(NetClientState *nc, const uint8_t *buf, bd.length = buf_len; size -= buf_len; - FEC_PRINTF("rx_bd 0x%x length %d\n", addr, bd.length); + trace_imx_fec_receive_len(addr, bd.length); /* The last 4 bytes are the CRC. */ if (size < 4) { @@ -XXX,XX +XXX,XX @@ static ssize_t imx_fec_receive(NetClientState *nc, const uint8_t *buf, if (size == 0) { /* Last buffer in frame. */ bd.flags |= flags | ENET_BD_L; - FEC_PRINTF("rx frame flags %04x\n", bd.flags); + + trace_imx_fec_receive_last(bd.flags); + s->regs[ENET_EIR] |= ENET_INT_RXF; } else { s->regs[ENET_EIR] |= ENET_INT_RXB; @@ -XXX,XX +XXX,XX @@ static ssize_t imx_enet_receive(NetClientState *nc, const uint8_t *buf, size_t size = len; bool shift16 = s->regs[ENET_RACC] & ENET_RACC_SHIFT16; - FEC_PRINTF("len %d\n", (int)size); + trace_imx_enet_receive(size); if (!s->regs[ENET_RDAR]) { qemu_log_mask(LOG_GUEST_ERROR, "[%s]%s: Unexpected packet\n", @@ -XXX,XX +XXX,XX @@ static ssize_t imx_enet_receive(NetClientState *nc, const uint8_t *buf, bd.length = buf_len; size -= buf_len; - FEC_PRINTF("rx_bd 0x%x length %d\n", addr, bd.length); + trace_imx_enet_receive_len(addr, bd.length); /* The last 4 bytes are the CRC. */ if (size < 4) { @@ -XXX,XX +XXX,XX @@ static ssize_t imx_enet_receive(NetClientState *nc, const uint8_t *buf, if (size == 0) { /* Last buffer in frame. */ bd.flags |= flags | ENET_BD_L; - FEC_PRINTF("rx frame flags %04x\n", bd.flags); + + trace_imx_enet_receive_last(bd.flags); + /* Indicate that we've updated the last buffer descriptor. */ bd.last_buffer = ENET_BD_BDU; if (bd.option & ENET_BD_RX_INT) { diff --git a/hw/net/trace-events b/hw/net/trace-events index XXXXXXX..XXXXXXX 100644 --- a/hw/net/trace-events +++ b/hw/net/trace-events @@ -XXX,XX +XXX,XX @@ i82596_receive_packet(size_t sz) "len=%zu" i82596_new_mac(const char *id_with_mac) "New MAC for: %s" i82596_set_multicast(uint16_t count) "Added %d multicast entries" i82596_channel_attention(void *s) "%p: Received CHANNEL ATTENTION" + +# imx_fec.c +imx_phy_read(uint32_t val, int reg) "0x%04"PRIx32" <= reg[%d]" +imx_phy_write(uint32_t val, int reg) "0x%04"PRIx32" => reg[%d]" +imx_phy_update_link(const char *s) "%s" +imx_phy_reset(void) "" +imx_fec_read_bd(uint64_t addr, int flags, int len, int data) "tx_bd 0x%"PRIx64" flags 0x%04x len %d data 0x%08x" +imx_enet_read_bd(uint64_t addr, int flags, int len, int data, int options, int status) "tx_bd 0x%"PRIx64" flags 0x%04x len %d data 0x%08x option 0x%04x status 0x%04x" +imx_eth_tx_bd_busy(void) "tx_bd ran out of descriptors to transmit" +imx_eth_rx_bd_full(void) "RX buffer is full" +imx_eth_read(int reg, const char *reg_name, uint32_t value) "reg[%d:%s] => 0x%08"PRIx32 +imx_eth_write(int reg, const char *reg_name, uint64_t value) "reg[%d:%s] <= 0x%08"PRIx64 +imx_fec_receive(size_t size) "len %zu" +imx_fec_receive_len(uint64_t addr, int len) "rx_bd 0x%"PRIx64" length %d" +imx_fec_receive_last(int last) "rx frame flags 0x%04x" +imx_enet_receive(size_t size) "len %zu" +imx_enet_receive_len(uint64_t addr, int len) "rx_bd 0x%"PRIx64" length %d" +imx_enet_receive_last(int last) "rx frame flags 0x%04x" -- 2.20.1 From: Guenter Roeck <linux@roeck-us.net> The Linux kernel's IMX code now uses vendor specific commands. This results in endless warnings when booting the Linux kernel. sdhci-esdhc-imx 2194000.usdhc: esdhc_wait_for_card_clock_gate_off: card clock still not gate off in 100us!. Implement support for the vendor specific command implemented in IMX hardware to be able to avoid this warning. Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org> Tested-by: Philippe Mathieu-Daudé <f4bug@amsat.org> Signed-off-by: Guenter Roeck <linux@roeck-us.net> Message-id: 20200603145258.195920-2-linux@roeck-us.net Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/sd/sdhci-internal.h | 5 +++++ include/hw/sd/sdhci.h | 5 +++++ hw/sd/sdhci.c | 18 +++++++++++++++++- 3 files changed, 27 insertions(+), 1 deletion(-) diff --git a/hw/sd/sdhci-internal.h b/hw/sd/sdhci-internal.h index XXXXXXX..XXXXXXX 100644 --- a/hw/sd/sdhci-internal.h +++ b/hw/sd/sdhci-internal.h @@ -XXX,XX +XXX,XX @@ #define SDHC_CMD_INHIBIT 0x00000001 #define SDHC_DATA_INHIBIT 0x00000002 #define SDHC_DAT_LINE_ACTIVE 0x00000004 +#define SDHC_IMX_CLOCK_GATE_OFF 0x00000080 #define SDHC_DOING_WRITE 0x00000100 #define SDHC_DOING_READ 0x00000200 #define SDHC_SPACE_AVAILABLE 0x00000400 @@ -XXX,XX +XXX,XX @@ extern const VMStateDescription sdhci_vmstate; #define ESDHC_MIX_CTRL 0x48 + #define ESDHC_VENDOR_SPEC 0xc0 +#define ESDHC_IMX_FRC_SDCLK_ON (1 << 8) + #define ESDHC_DLL_CTRL 0x60 #define ESDHC_TUNING_CTRL 0xcc @@ -XXX,XX +XXX,XX @@ extern const VMStateDescription sdhci_vmstate; #define DEFINE_SDHCI_COMMON_PROPERTIES(_state) \ DEFINE_PROP_UINT8("sd-spec-version", _state, sd_spec_version, 2), \ DEFINE_PROP_UINT8("uhs", _state, uhs_mode, UHS_NOT_SUPPORTED), \ + DEFINE_PROP_UINT8("vendor", _state, vendor, SDHCI_VENDOR_NONE), \ \ /* Capabilities registers provide information on supported * features of this specific host controller implementation */ \ diff --git a/include/hw/sd/sdhci.h b/include/hw/sd/sdhci.h index XXXXXXX..XXXXXXX 100644 --- a/include/hw/sd/sdhci.h +++ b/include/hw/sd/sdhci.h @@ -XXX,XX +XXX,XX @@ typedef struct SDHCIState { uint16_t acmd12errsts; /* Auto CMD12 error status register */ uint16_t hostctl2; /* Host Control 2 */ uint64_t admasysaddr; /* ADMA System Address Register */ + uint16_t vendor_spec; /* Vendor specific register */ /* Read-only registers */ uint64_t capareg; /* Capabilities Register */ @@ -XXX,XX +XXX,XX @@ typedef struct SDHCIState { uint32_t quirks; uint8_t sd_spec_version; uint8_t uhs_mode; + uint8_t vendor; /* For vendor specific functionality */ } SDHCIState; +#define SDHCI_VENDOR_NONE 0 +#define SDHCI_VENDOR_IMX 1 + /* * Controller does not provide transfer-complete interrupt when not * busy. diff --git a/hw/sd/sdhci.c b/hw/sd/sdhci.c index XXXXXXX..XXXXXXX 100644 --- a/hw/sd/sdhci.c +++ b/hw/sd/sdhci.c @@ -XXX,XX +XXX,XX @@ static uint64_t usdhc_read(void *opaque, hwaddr offset, unsigned size) } break; + case ESDHC_VENDOR_SPEC: + ret = s->vendor_spec; + break; case ESDHC_DLL_CTRL: case ESDHC_TUNE_CTRL_STATUS: case ESDHC_UNDOCUMENTED_REG27: case ESDHC_TUNING_CTRL: - case ESDHC_VENDOR_SPEC: case ESDHC_MIX_CTRL: case ESDHC_WTMK_LVL: ret = 0; @@ -XXX,XX +XXX,XX @@ usdhc_write(void *opaque, hwaddr offset, uint64_t val, unsigned size) case ESDHC_UNDOCUMENTED_REG27: case ESDHC_TUNING_CTRL: case ESDHC_WTMK_LVL: + break; + case ESDHC_VENDOR_SPEC: + s->vendor_spec = value; + switch (s->vendor) { + case SDHCI_VENDOR_IMX: + if (value & ESDHC_IMX_FRC_SDCLK_ON) { + s->prnsts &= ~SDHC_IMX_CLOCK_GATE_OFF; + } else { + s->prnsts |= SDHC_IMX_CLOCK_GATE_OFF; + } + break; + default: + break; + } break; case SDHC_HOSTCTL: -- 2.20.1 From: Guenter Roeck <linux@roeck-us.net> Set vendor property to IMX to enable IMX specific functionality in sdhci code. Tested-by: Philippe Mathieu-Daudé <f4bug@amsat.org> Signed-off-by: Guenter Roeck <linux@roeck-us.net> Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org> Message-id: 20200603145258.195920-3-linux@roeck-us.net Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/arm/fsl-imx25.c | 6 ++++++ hw/arm/fsl-imx6.c | 6 ++++++ hw/arm/fsl-imx6ul.c | 2 ++ hw/arm/fsl-imx7.c | 2 ++ 4 files changed, 16 insertions(+) diff --git a/hw/arm/fsl-imx25.c b/hw/arm/fsl-imx25.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/fsl-imx25.c +++ b/hw/arm/fsl-imx25.c @@ -XXX,XX +XXX,XX @@ static void fsl_imx25_realize(DeviceState *dev, Error **errp) &err); object_property_set_uint(OBJECT(&s->esdhc[i]), IMX25_ESDHC_CAPABILITIES, "capareg", &err); + object_property_set_uint(OBJECT(&s->esdhc[i]), SDHCI_VENDOR_IMX, + "vendor", &err); + if (err) { + error_propagate(errp, err); + return; + } object_property_set_bool(OBJECT(&s->esdhc[i]), true, "realized", &err); if (err) { error_propagate(errp, err); diff --git a/hw/arm/fsl-imx6.c b/hw/arm/fsl-imx6.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/fsl-imx6.c +++ b/hw/arm/fsl-imx6.c @@ -XXX,XX +XXX,XX @@ static void fsl_imx6_realize(DeviceState *dev, Error **errp) &err); object_property_set_uint(OBJECT(&s->esdhc[i]), IMX6_ESDHC_CAPABILITIES, "capareg", &err); + object_property_set_uint(OBJECT(&s->esdhc[i]), SDHCI_VENDOR_IMX, + "vendor", &err); + if (err) { + error_propagate(errp, err); + return; + } object_property_set_bool(OBJECT(&s->esdhc[i]), true, "realized", &err); if (err) { error_propagate(errp, err); diff --git a/hw/arm/fsl-imx6ul.c b/hw/arm/fsl-imx6ul.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/fsl-imx6ul.c +++ b/hw/arm/fsl-imx6ul.c @@ -XXX,XX +XXX,XX @@ static void fsl_imx6ul_realize(DeviceState *dev, Error **errp) FSL_IMX6UL_USDHC2_IRQ, }; + object_property_set_uint(OBJECT(&s->usdhc[i]), SDHCI_VENDOR_IMX, + "vendor", &error_abort); object_property_set_bool(OBJECT(&s->usdhc[i]), true, "realized", &error_abort); diff --git a/hw/arm/fsl-imx7.c b/hw/arm/fsl-imx7.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/fsl-imx7.c +++ b/hw/arm/fsl-imx7.c @@ -XXX,XX +XXX,XX @@ static void fsl_imx7_realize(DeviceState *dev, Error **errp) FSL_IMX7_USDHC3_IRQ, }; + object_property_set_uint(OBJECT(&s->usdhc[i]), SDHCI_VENDOR_IMX, + "vendor", &error_abort); object_property_set_bool(OBJECT(&s->usdhc[i]), true, "realized", &error_abort); -- 2.20.1 Small pile of bug fixes for rc1. I've included my patches to get our docs building with Sphinx 3, just for convenience... -- PMM The following changes since commit b149dea55cce97cb226683d06af61984a1c11e96: Merge remote-tracking branch 'remotes/cschoenebeck/tags/pull-9p-20201102' into staging (2020-11-02 10:57:48 +0000) are available in the Git repository at: https://git.linaro.org/people/pmaydell/qemu-arm.git tags/pull-target-arm-20201102 for you to fetch changes up to ffb4fbf90a2f63c9cb33e4bb9f854c79bf04ca4a: tests/qtest/npcm7xx_rng-test: Disable randomness tests (2020-11-02 16:52:18 +0000) ---------------------------------------------------------------- target-arm queue: * target/arm: Fix Neon emulation bugs on big-endian hosts * target/arm: fix handling of HCR.FB * target/arm: fix LORID_EL1 access check * disas/capstone: Fix monitor disassembly of >32 bytes * hw/arm/smmuv3: Fix potential integer overflow (CID 1432363) * hw/arm/boot: fix SVE for EL3 direct kernel boot * hw/display/omap_lcdc: Fix potential NULL pointer dereference * hw/display/exynos4210_fimd: Fix potential NULL pointer dereference * target/arm: Get correct MMU index for other-security-state * configure: Test that gio libs from pkg-config work * hw/intc/arm_gicv3_cpuif: Make GIC maintenance interrupts work * docs: Fix building with Sphinx 3 * tests/qtest/npcm7xx_rng-test: Disable randomness tests ---------------------------------------------------------------- AlexChen (2): hw/display/omap_lcdc: Fix potential NULL pointer dereference hw/display/exynos4210_fimd: Fix potential NULL pointer dereference Peter Maydell (9): target/arm: Fix float16 pairwise Neon ops on big-endian hosts target/arm: Fix VUDOT/VSDOT (scalar) on big-endian hosts disas/capstone: Fix monitor disassembly of >32 bytes target/arm: Get correct MMU index for other-security-state configure: Test that gio libs from pkg-config work hw/intc/arm_gicv3_cpuif: Make GIC maintenance interrupts work scripts/kerneldoc: For Sphinx 3 use c:macro for macros with arguments qemu-option-trace.rst.inc: Don't use option:: markup tests/qtest/npcm7xx_rng-test: Disable randomness tests Philippe Mathieu-Daudé (1): hw/arm/smmuv3: Fix potential integer overflow (CID 1432363) Richard Henderson (11): target/arm: Introduce neon_full_reg_offset target/arm: Move neon_element_offset to translate.c target/arm: Use neon_element_offset in neon_load/store_reg target/arm: Use neon_element_offset in vfp_reg_offset target/arm: Add read/write_neon_element32 target/arm: Expand read/write_neon_element32 to all MemOp target/arm: Rename neon_load_reg32 to vfp_load_reg32 target/arm: Add read/write_neon_element64 target/arm: Rename neon_load_reg64 to vfp_load_reg64 target/arm: Simplify do_long_3d and do_2scalar_long target/arm: Improve do_prewiden_3d Rémi Denis-Courmont (3): target/arm: fix handling of HCR.FB target/arm: fix LORID_EL1 access check hw/arm/boot: fix SVE for EL3 direct kernel boot docs/qemu-option-trace.rst.inc | 6 +- configure | 10 +- include/hw/intc/arm_gicv3_common.h | 1 - disas/capstone.c | 2 +- hw/arm/boot.c | 3 + hw/arm/smmuv3.c | 3 +- hw/display/exynos4210_fimd.c | 4 +- hw/display/omap_lcdc.c | 10 +- hw/intc/arm_gicv3_cpuif.c | 5 +- target/arm/helper.c | 24 +- target/arm/m_helper.c | 3 +- target/arm/translate.c | 153 +++++++++--- target/arm/vec_helper.c | 12 +- tests/qtest/npcm7xx_rng-test.c | 14 +- scripts/kernel-doc | 18 +- target/arm/translate-neon.c.inc | 472 ++++++++++++++++++++----------------- target/arm/translate-vfp.c.inc | 341 +++++++++++---------------- 17 files changed, 588 insertions(+), 493 deletions(-) From: Richard Henderson <richard.henderson@linaro.org> This function makes it clear that we're talking about the whole register, and not the 32-bit piece at index 0. This fixes a bug when running on a big-endian host. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20201030022618.785675-2-richard.henderson@linaro.org Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/translate.c | 8 ++++++ target/arm/translate-neon.c.inc | 44 ++++++++++++++++----------------- target/arm/translate-vfp.c.inc | 2 +- 3 files changed, 31 insertions(+), 23 deletions(-) diff --git a/target/arm/translate.c b/target/arm/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate.c +++ b/target/arm/translate.c @@ -XXX,XX +XXX,XX @@ static inline void gen_hlt(DisasContext *s, int imm) unallocated_encoding(s); } +/* + * Return the offset of a "full" NEON Dreg. + */ +static long neon_full_reg_offset(unsigned reg) +{ + return offsetof(CPUARMState, vfp.zregs[reg >> 1].d[reg & 1]); +} + static inline long vfp_reg_offset(bool dp, unsigned reg) { if (dp) { diff --git a/target/arm/translate-neon.c.inc b/target/arm/translate-neon.c.inc index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-neon.c.inc +++ b/target/arm/translate-neon.c.inc @@ -XXX,XX +XXX,XX @@ neon_element_offset(int reg, int element, MemOp size) ofs ^= 8 - element_size; } #endif - return neon_reg_offset(reg, 0) + ofs; + return neon_full_reg_offset(reg) + ofs; } static void neon_load_element(TCGv_i32 var, int reg, int ele, MemOp mop) @@ -XXX,XX +XXX,XX @@ static bool trans_VLD_all_lanes(DisasContext *s, arg_VLD_all_lanes *a) * We cannot write 16 bytes at once because the * destination is unaligned. */ - tcg_gen_gvec_dup_i32(size, neon_reg_offset(vd, 0), + tcg_gen_gvec_dup_i32(size, neon_full_reg_offset(vd), 8, 8, tmp); - tcg_gen_gvec_mov(0, neon_reg_offset(vd + 1, 0), - neon_reg_offset(vd, 0), 8, 8); + tcg_gen_gvec_mov(0, neon_full_reg_offset(vd + 1), + neon_full_reg_offset(vd), 8, 8); } else { - tcg_gen_gvec_dup_i32(size, neon_reg_offset(vd, 0), + tcg_gen_gvec_dup_i32(size, neon_full_reg_offset(vd), vec_size, vec_size, tmp); } tcg_gen_addi_i32(addr, addr, 1 << size); @@ -XXX,XX +XXX,XX @@ static bool trans_VLDST_single(DisasContext *s, arg_VLDST_single *a) static bool do_3same(DisasContext *s, arg_3same *a, GVecGen3Fn fn) { int vec_size = a->q ? 16 : 8; - int rd_ofs = neon_reg_offset(a->vd, 0); - int rn_ofs = neon_reg_offset(a->vn, 0); - int rm_ofs = neon_reg_offset(a->vm, 0); + int rd_ofs = neon_full_reg_offset(a->vd); + int rn_ofs = neon_full_reg_offset(a->vn); + int rm_ofs = neon_full_reg_offset(a->vm); if (!arm_dc_feature(s, ARM_FEATURE_NEON)) { return false; @@ -XXX,XX +XXX,XX @@ static bool do_vector_2sh(DisasContext *s, arg_2reg_shift *a, GVecGen2iFn *fn) { /* Handle a 2-reg-shift insn which can be vectorized. */ int vec_size = a->q ? 16 : 8; - int rd_ofs = neon_reg_offset(a->vd, 0); - int rm_ofs = neon_reg_offset(a->vm, 0); + int rd_ofs = neon_full_reg_offset(a->vd); + int rm_ofs = neon_full_reg_offset(a->vm); if (!arm_dc_feature(s, ARM_FEATURE_NEON)) { return false; @@ -XXX,XX +XXX,XX @@ static bool do_fp_2sh(DisasContext *s, arg_2reg_shift *a, { /* FP operations in 2-reg-and-shift group */ int vec_size = a->q ? 16 : 8; - int rd_ofs = neon_reg_offset(a->vd, 0); - int rm_ofs = neon_reg_offset(a->vm, 0); + int rd_ofs = neon_full_reg_offset(a->vd); + int rm_ofs = neon_full_reg_offset(a->vm); TCGv_ptr fpst; if (!arm_dc_feature(s, ARM_FEATURE_NEON)) { @@ -XXX,XX +XXX,XX @@ static bool do_1reg_imm(DisasContext *s, arg_1reg_imm *a, return true; } - reg_ofs = neon_reg_offset(a->vd, 0); + reg_ofs = neon_full_reg_offset(a->vd); vec_size = a->q ? 16 : 8; imm = asimd_imm_const(a->imm, a->cmode, a->op); @@ -XXX,XX +XXX,XX @@ static bool trans_VMULL_P_3d(DisasContext *s, arg_3diff *a) return true; } - tcg_gen_gvec_3_ool(neon_reg_offset(a->vd, 0), - neon_reg_offset(a->vn, 0), - neon_reg_offset(a->vm, 0), + tcg_gen_gvec_3_ool(neon_full_reg_offset(a->vd), + neon_full_reg_offset(a->vn), + neon_full_reg_offset(a->vm), 16, 16, 0, fn_gvec); return true; } @@ -XXX,XX +XXX,XX @@ static bool do_2scalar_fp_vec(DisasContext *s, arg_2scalar *a, { /* Two registers and a scalar, using gvec */ int vec_size = a->q ? 16 : 8; - int rd_ofs = neon_reg_offset(a->vd, 0); - int rn_ofs = neon_reg_offset(a->vn, 0); + int rd_ofs = neon_full_reg_offset(a->vd); + int rn_ofs = neon_full_reg_offset(a->vn); int rm_ofs; int idx; TCGv_ptr fpstatus; @@ -XXX,XX +XXX,XX @@ static bool do_2scalar_fp_vec(DisasContext *s, arg_2scalar *a, /* a->vm is M:Vm, which encodes both register and index */ idx = extract32(a->vm, a->size + 2, 2); a->vm = extract32(a->vm, 0, a->size + 2); - rm_ofs = neon_reg_offset(a->vm, 0); + rm_ofs = neon_full_reg_offset(a->vm); fpstatus = fpstatus_ptr(a->size == 1 ? FPST_STD_F16 : FPST_STD); tcg_gen_gvec_3_ptr(rd_ofs, rn_ofs, rm_ofs, fpstatus, @@ -XXX,XX +XXX,XX @@ static bool trans_VDUP_scalar(DisasContext *s, arg_VDUP_scalar *a) return true; } - tcg_gen_gvec_dup_mem(a->size, neon_reg_offset(a->vd, 0), + tcg_gen_gvec_dup_mem(a->size, neon_full_reg_offset(a->vd), neon_element_offset(a->vm, a->index, a->size), a->q ? 16 : 8, a->q ? 16 : 8); return true; @@ -XXX,XX +XXX,XX @@ static bool trans_VCVT_F32_F16(DisasContext *s, arg_2misc *a) static bool do_2misc_vec(DisasContext *s, arg_2misc *a, GVecGen2Fn *fn) { int vec_size = a->q ? 16 : 8; - int rd_ofs = neon_reg_offset(a->vd, 0); - int rm_ofs = neon_reg_offset(a->vm, 0); + int rd_ofs = neon_full_reg_offset(a->vd); + int rm_ofs = neon_full_reg_offset(a->vm); if (!arm_dc_feature(s, ARM_FEATURE_NEON)) { return false; diff --git a/target/arm/translate-vfp.c.inc b/target/arm/translate-vfp.c.inc index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-vfp.c.inc +++ b/target/arm/translate-vfp.c.inc @@ -XXX,XX +XXX,XX @@ static bool trans_VDUP(DisasContext *s, arg_VDUP *a) } tmp = load_reg(s, a->rt); - tcg_gen_gvec_dup_i32(size, neon_reg_offset(a->vn, 0), + tcg_gen_gvec_dup_i32(size, neon_full_reg_offset(a->vn), vec_size, vec_size, tmp); tcg_temp_free_i32(tmp); -- 2.20.1 From: Richard Henderson <richard.henderson@linaro.org> This will shortly have users outside of translate-neon.c.inc. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20201030022618.785675-3-richard.henderson@linaro.org Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/translate.c | 20 ++++++++++++++++++++ target/arm/translate-neon.c.inc | 19 ------------------- 2 files changed, 20 insertions(+), 19 deletions(-) diff --git a/target/arm/translate.c b/target/arm/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate.c +++ b/target/arm/translate.c @@ -XXX,XX +XXX,XX @@ static long neon_full_reg_offset(unsigned reg) return offsetof(CPUARMState, vfp.zregs[reg >> 1].d[reg & 1]); } +/* + * Return the offset of a 2**SIZE piece of a NEON register, at index ELE, + * where 0 is the least significant end of the register. + */ +static long neon_element_offset(int reg, int element, MemOp size) +{ + int element_size = 1 << size; + int ofs = element * element_size; +#ifdef HOST_WORDS_BIGENDIAN + /* + * Calculate the offset assuming fully little-endian, + * then XOR to account for the order of the 8-byte units. + */ + if (element_size < 8) { + ofs ^= 8 - element_size; + } +#endif + return neon_full_reg_offset(reg) + ofs; +} + static inline long vfp_reg_offset(bool dp, unsigned reg) { if (dp) { diff --git a/target/arm/translate-neon.c.inc b/target/arm/translate-neon.c.inc index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-neon.c.inc +++ b/target/arm/translate-neon.c.inc @@ -XXX,XX +XXX,XX @@ static inline int neon_3same_fp_size(DisasContext *s, int x) #include "decode-neon-ls.c.inc" #include "decode-neon-shared.c.inc" -/* Return the offset of a 2**SIZE piece of a NEON register, at index ELE, - * where 0 is the least significant end of the register. - */ -static inline long -neon_element_offset(int reg, int element, MemOp size) -{ - int element_size = 1 << size; - int ofs = element * element_size; -#ifdef HOST_WORDS_BIGENDIAN - /* Calculate the offset assuming fully little-endian, - * then XOR to account for the order of the 8-byte units. - */ - if (element_size < 8) { - ofs ^= 8 - element_size; - } -#endif - return neon_full_reg_offset(reg) + ofs; -} - static void neon_load_element(TCGv_i32 var, int reg, int ele, MemOp mop) { long offset = neon_element_offset(reg, ele, mop & MO_SIZE); -- 2.20.1 From: Richard Henderson <richard.henderson@linaro.org> These are the only users of neon_reg_offset, so remove that. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20201030022618.785675-4-richard.henderson@linaro.org Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/translate.c | 14 ++------------ 1 file changed, 2 insertions(+), 12 deletions(-) diff --git a/target/arm/translate.c b/target/arm/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate.c +++ b/target/arm/translate.c @@ -XXX,XX +XXX,XX @@ static inline long vfp_reg_offset(bool dp, unsigned reg) } } -/* Return the offset of a 32-bit piece of a NEON register. - zero is the least significant end of the register. */ -static inline long -neon_reg_offset (int reg, int n) -{ - int sreg; - sreg = reg * 2 + n; - return vfp_reg_offset(0, sreg); -} - static TCGv_i32 neon_load_reg(int reg, int pass) { TCGv_i32 tmp = tcg_temp_new_i32(); - tcg_gen_ld_i32(tmp, cpu_env, neon_reg_offset(reg, pass)); + tcg_gen_ld_i32(tmp, cpu_env, neon_element_offset(reg, pass, MO_32)); return tmp; } static void neon_store_reg(int reg, int pass, TCGv_i32 var) { - tcg_gen_st_i32(var, cpu_env, neon_reg_offset(reg, pass)); + tcg_gen_st_i32(var, cpu_env, neon_element_offset(reg, pass, MO_32)); tcg_temp_free_i32(var); } -- 2.20.1 From: Richard Henderson <richard.henderson@linaro.org> This seems a bit more readable than using offsetof CPU_DoubleU. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20201030022618.785675-5-richard.henderson@linaro.org Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/translate.c | 13 ++++--------- 1 file changed, 4 insertions(+), 9 deletions(-) diff --git a/target/arm/translate.c b/target/arm/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate.c +++ b/target/arm/translate.c @@ -XXX,XX +XXX,XX @@ static long neon_element_offset(int reg, int element, MemOp size) return neon_full_reg_offset(reg) + ofs; } -static inline long vfp_reg_offset(bool dp, unsigned reg) +/* Return the offset of a VFP Dreg (dp = true) or VFP Sreg (dp = false). */ +static long vfp_reg_offset(bool dp, unsigned reg) { if (dp) { - return offsetof(CPUARMState, vfp.zregs[reg >> 1].d[reg & 1]); + return neon_element_offset(reg, 0, MO_64); } else { - long ofs = offsetof(CPUARMState, vfp.zregs[reg >> 2].d[(reg >> 1) & 1]); - if (reg & 1) { - ofs += offsetof(CPU_DoubleU, l.upper); - } else { - ofs += offsetof(CPU_DoubleU, l.lower); - } - return ofs; + return neon_element_offset(reg >> 1, reg & 1, MO_32); } } -- 2.20.1 From: Richard Henderson <richard.henderson@linaro.org> Model these off the aa64 read/write_vec_element functions. Use it within translate-neon.c.inc. The new functions do not allocate or free temps, so this rearranges the calling code a bit. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20201030022618.785675-6-richard.henderson@linaro.org Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/translate.c | 26 ++++ target/arm/translate-neon.c.inc | 256 ++++++++++++++++++++------------ 2 files changed, 183 insertions(+), 99 deletions(-) diff --git a/target/arm/translate.c b/target/arm/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate.c +++ b/target/arm/translate.c @@ -XXX,XX +XXX,XX @@ static inline void neon_store_reg32(TCGv_i32 var, int reg) tcg_gen_st_i32(var, cpu_env, vfp_reg_offset(false, reg)); } +static void read_neon_element32(TCGv_i32 dest, int reg, int ele, MemOp size) +{ + long off = neon_element_offset(reg, ele, size); + + switch (size) { + case MO_32: + tcg_gen_ld_i32(dest, cpu_env, off); + break; + default: + g_assert_not_reached(); + } +} + +static void write_neon_element32(TCGv_i32 src, int reg, int ele, MemOp size) +{ + long off = neon_element_offset(reg, ele, size); + + switch (size) { + case MO_32: + tcg_gen_st_i32(src, cpu_env, off); + break; + default: + g_assert_not_reached(); + } +} + static TCGv_ptr vfp_reg_ptr(bool dp, int reg) { TCGv_ptr ret = tcg_temp_new_ptr(); diff --git a/target/arm/translate-neon.c.inc b/target/arm/translate-neon.c.inc index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-neon.c.inc +++ b/target/arm/translate-neon.c.inc @@ -XXX,XX +XXX,XX @@ static bool do_3same_pair(DisasContext *s, arg_3same *a, NeonGenTwoOpFn *fn) * early. Since Q is 0 there are always just two passes, so instead * of a complicated loop over each pass we just unroll. */ - tmp = neon_load_reg(a->vn, 0); - tmp2 = neon_load_reg(a->vn, 1); + tmp = tcg_temp_new_i32(); + tmp2 = tcg_temp_new_i32(); + tmp3 = tcg_temp_new_i32(); + + read_neon_element32(tmp, a->vn, 0, MO_32); + read_neon_element32(tmp2, a->vn, 1, MO_32); fn(tmp, tmp, tmp2); - tcg_temp_free_i32(tmp2); - tmp3 = neon_load_reg(a->vm, 0); - tmp2 = neon_load_reg(a->vm, 1); + read_neon_element32(tmp3, a->vm, 0, MO_32); + read_neon_element32(tmp2, a->vm, 1, MO_32); fn(tmp3, tmp3, tmp2); - tcg_temp_free_i32(tmp2); - neon_store_reg(a->vd, 0, tmp); - neon_store_reg(a->vd, 1, tmp3); + write_neon_element32(tmp, a->vd, 0, MO_32); + write_neon_element32(tmp3, a->vd, 1, MO_32); + + tcg_temp_free_i32(tmp); + tcg_temp_free_i32(tmp2); + tcg_temp_free_i32(tmp3); return true; } @@ -XXX,XX +XXX,XX @@ static bool do_2shift_env_32(DisasContext *s, arg_2reg_shift *a, * 2-reg-and-shift operations, size < 3 case, where the * helper needs to be passed cpu_env. */ - TCGv_i32 constimm; + TCGv_i32 constimm, tmp; int pass; if (!arm_dc_feature(s, ARM_FEATURE_NEON)) { @@ -XXX,XX +XXX,XX @@ static bool do_2shift_env_32(DisasContext *s, arg_2reg_shift *a, * by immediate using the variable shift operations. */ constimm = tcg_const_i32(dup_const(a->size, a->shift)); + tmp = tcg_temp_new_i32(); for (pass = 0; pass < (a->q ? 4 : 2); pass++) { - TCGv_i32 tmp = neon_load_reg(a->vm, pass); + read_neon_element32(tmp, a->vm, pass, MO_32); fn(tmp, cpu_env, tmp, constimm); - neon_store_reg(a->vd, pass, tmp); + write_neon_element32(tmp, a->vd, pass, MO_32); } + tcg_temp_free_i32(tmp); tcg_temp_free_i32(constimm); return true; } @@ -XXX,XX +XXX,XX @@ static bool do_2shift_narrow_64(DisasContext *s, arg_2reg_shift *a, constimm = tcg_const_i64(-a->shift); rm1 = tcg_temp_new_i64(); rm2 = tcg_temp_new_i64(); + rd = tcg_temp_new_i32(); /* Load both inputs first to avoid potential overwrite if rm == rd */ neon_load_reg64(rm1, a->vm); neon_load_reg64(rm2, a->vm + 1); shiftfn(rm1, rm1, constimm); - rd = tcg_temp_new_i32(); narrowfn(rd, cpu_env, rm1); - neon_store_reg(a->vd, 0, rd); + write_neon_element32(rd, a->vd, 0, MO_32); shiftfn(rm2, rm2, constimm); - rd = tcg_temp_new_i32(); narrowfn(rd, cpu_env, rm2); - neon_store_reg(a->vd, 1, rd); + write_neon_element32(rd, a->vd, 1, MO_32); + tcg_temp_free_i32(rd); tcg_temp_free_i64(rm1); tcg_temp_free_i64(rm2); tcg_temp_free_i64(constimm); @@ -XXX,XX +XXX,XX @@ static bool do_2shift_narrow_32(DisasContext *s, arg_2reg_shift *a, constimm = tcg_const_i32(imm); /* Load all inputs first to avoid potential overwrite */ - rm1 = neon_load_reg(a->vm, 0); - rm2 = neon_load_reg(a->vm, 1); - rm3 = neon_load_reg(a->vm + 1, 0); - rm4 = neon_load_reg(a->vm + 1, 1); + rm1 = tcg_temp_new_i32(); + rm2 = tcg_temp_new_i32(); + rm3 = tcg_temp_new_i32(); + rm4 = tcg_temp_new_i32(); + read_neon_element32(rm1, a->vm, 0, MO_32); + read_neon_element32(rm2, a->vm, 1, MO_32); + read_neon_element32(rm3, a->vm, 2, MO_32); + read_neon_element32(rm4, a->vm, 3, MO_32); rtmp = tcg_temp_new_i64(); shiftfn(rm1, rm1, constimm); @@ -XXX,XX +XXX,XX @@ static bool do_2shift_narrow_32(DisasContext *s, arg_2reg_shift *a, tcg_temp_free_i32(rm2); narrowfn(rm1, cpu_env, rtmp); - neon_store_reg(a->vd, 0, rm1); + write_neon_element32(rm1, a->vd, 0, MO_32); + tcg_temp_free_i32(rm1); shiftfn(rm3, rm3, constimm); shiftfn(rm4, rm4, constimm); @@ -XXX,XX +XXX,XX @@ static bool do_2shift_narrow_32(DisasContext *s, arg_2reg_shift *a, narrowfn(rm3, cpu_env, rtmp); tcg_temp_free_i64(rtmp); - neon_store_reg(a->vd, 1, rm3); + write_neon_element32(rm3, a->vd, 1, MO_32); + tcg_temp_free_i32(rm3); return true; } @@ -XXX,XX +XXX,XX @@ static bool do_vshll_2sh(DisasContext *s, arg_2reg_shift *a, widen_mask = dup_const(a->size + 1, widen_mask); } - rm0 = neon_load_reg(a->vm, 0); - rm1 = neon_load_reg(a->vm, 1); + rm0 = tcg_temp_new_i32(); + rm1 = tcg_temp_new_i32(); + read_neon_element32(rm0, a->vm, 0, MO_32); + read_neon_element32(rm1, a->vm, 1, MO_32); tmp = tcg_temp_new_i64(); widenfn(tmp, rm0); @@ -XXX,XX +XXX,XX @@ static bool do_prewiden_3d(DisasContext *s, arg_3diff *a, if (src1_wide) { neon_load_reg64(rn0_64, a->vn); } else { - TCGv_i32 tmp = neon_load_reg(a->vn, 0); + TCGv_i32 tmp = tcg_temp_new_i32(); + read_neon_element32(tmp, a->vn, 0, MO_32); widenfn(rn0_64, tmp); tcg_temp_free_i32(tmp); } - rm = neon_load_reg(a->vm, 0); + rm = tcg_temp_new_i32(); + read_neon_element32(rm, a->vm, 0, MO_32); widenfn(rm_64, rm); tcg_temp_free_i32(rm); @@ -XXX,XX +XXX,XX @@ static bool do_prewiden_3d(DisasContext *s, arg_3diff *a, if (src1_wide) { neon_load_reg64(rn1_64, a->vn + 1); } else { - TCGv_i32 tmp = neon_load_reg(a->vn, 1); + TCGv_i32 tmp = tcg_temp_new_i32(); + read_neon_element32(tmp, a->vn, 1, MO_32); widenfn(rn1_64, tmp); tcg_temp_free_i32(tmp); } - rm = neon_load_reg(a->vm, 1); + rm = tcg_temp_new_i32(); + read_neon_element32(rm, a->vm, 1, MO_32); neon_store_reg64(rn0_64, a->vd); @@ -XXX,XX +XXX,XX @@ static bool do_narrow_3d(DisasContext *s, arg_3diff *a, narrowfn(rd1, rn_64); - neon_store_reg(a->vd, 0, rd0); - neon_store_reg(a->vd, 1, rd1); + write_neon_element32(rd0, a->vd, 0, MO_32); + write_neon_element32(rd1, a->vd, 1, MO_32); + tcg_temp_free_i32(rd0); + tcg_temp_free_i32(rd1); tcg_temp_free_i64(rn_64); tcg_temp_free_i64(rm_64); @@ -XXX,XX +XXX,XX @@ static bool do_long_3d(DisasContext *s, arg_3diff *a, rd0 = tcg_temp_new_i64(); rd1 = tcg_temp_new_i64(); - rn = neon_load_reg(a->vn, 0); - rm = neon_load_reg(a->vm, 0); + rn = tcg_temp_new_i32(); + rm = tcg_temp_new_i32(); + read_neon_element32(rn, a->vn, 0, MO_32); + read_neon_element32(rm, a->vm, 0, MO_32); opfn(rd0, rn, rm); - tcg_temp_free_i32(rn); - tcg_temp_free_i32(rm); - rn = neon_load_reg(a->vn, 1); - rm = neon_load_reg(a->vm, 1); + read_neon_element32(rn, a->vn, 1, MO_32); + read_neon_element32(rm, a->vm, 1, MO_32); opfn(rd1, rn, rm); tcg_temp_free_i32(rn); tcg_temp_free_i32(rm); @@ -XXX,XX +XXX,XX @@ static void gen_neon_dup_high16(TCGv_i32 var) static inline TCGv_i32 neon_get_scalar(int size, int reg) { - TCGv_i32 tmp; - if (size == 1) { - tmp = neon_load_reg(reg & 7, reg >> 4); + TCGv_i32 tmp = tcg_temp_new_i32(); + if (size == MO_16) { + read_neon_element32(tmp, reg & 7, reg >> 4, MO_32); if (reg & 8) { gen_neon_dup_high16(tmp); } else { gen_neon_dup_low16(tmp); } } else { - tmp = neon_load_reg(reg & 15, reg >> 4); + read_neon_element32(tmp, reg & 15, reg >> 4, MO_32); } return tmp; } @@ -XXX,XX +XXX,XX @@ static bool do_2scalar(DisasContext *s, arg_2scalar *a, * perform an accumulation operation of that result into the * destination. */ - TCGv_i32 scalar; + TCGv_i32 scalar, tmp; int pass; if (!arm_dc_feature(s, ARM_FEATURE_NEON)) { @@ -XXX,XX +XXX,XX @@ static bool do_2scalar(DisasContext *s, arg_2scalar *a, } scalar = neon_get_scalar(a->size, a->vm); + tmp = tcg_temp_new_i32(); for (pass = 0; pass < (a->q ? 4 : 2); pass++) { - TCGv_i32 tmp = neon_load_reg(a->vn, pass); + read_neon_element32(tmp, a->vn, pass, MO_32); opfn(tmp, tmp, scalar); if (accfn) { - TCGv_i32 rd = neon_load_reg(a->vd, pass); + TCGv_i32 rd = tcg_temp_new_i32(); + read_neon_element32(rd, a->vd, pass, MO_32); accfn(tmp, rd, tmp); tcg_temp_free_i32(rd); } - neon_store_reg(a->vd, pass, tmp); + write_neon_element32(tmp, a->vd, pass, MO_32); } + tcg_temp_free_i32(tmp); tcg_temp_free_i32(scalar); return true; } @@ -XXX,XX +XXX,XX @@ static bool do_vqrdmlah_2sc(DisasContext *s, arg_2scalar *a, * performs a kind of fused op-then-accumulate using a helper * function that takes all of rd, rn and the scalar at once. */ - TCGv_i32 scalar; + TCGv_i32 scalar, rn, rd; int pass; if (!arm_dc_feature(s, ARM_FEATURE_NEON)) { @@ -XXX,XX +XXX,XX @@ static bool do_vqrdmlah_2sc(DisasContext *s, arg_2scalar *a, } scalar = neon_get_scalar(a->size, a->vm); + rn = tcg_temp_new_i32(); + rd = tcg_temp_new_i32(); for (pass = 0; pass < (a->q ? 4 : 2); pass++) { - TCGv_i32 rn = neon_load_reg(a->vn, pass); - TCGv_i32 rd = neon_load_reg(a->vd, pass); + read_neon_element32(rn, a->vn, pass, MO_32); + read_neon_element32(rd, a->vd, pass, MO_32); opfn(rd, cpu_env, rn, scalar, rd); - tcg_temp_free_i32(rn); - neon_store_reg(a->vd, pass, rd); + write_neon_element32(rd, a->vd, pass, MO_32); } + tcg_temp_free_i32(rn); + tcg_temp_free_i32(rd); tcg_temp_free_i32(scalar); return true; @@ -XXX,XX +XXX,XX @@ static bool do_2scalar_long(DisasContext *s, arg_2scalar *a, scalar = neon_get_scalar(a->size, a->vm); /* Load all inputs before writing any outputs, in case of overlap */ - rn = neon_load_reg(a->vn, 0); + rn = tcg_temp_new_i32(); + read_neon_element32(rn, a->vn, 0, MO_32); rn0_64 = tcg_temp_new_i64(); opfn(rn0_64, rn, scalar); - tcg_temp_free_i32(rn); - rn = neon_load_reg(a->vn, 1); + read_neon_element32(rn, a->vn, 1, MO_32); rn1_64 = tcg_temp_new_i64(); opfn(rn1_64, rn, scalar); tcg_temp_free_i32(rn); @@ -XXX,XX +XXX,XX @@ static bool trans_VTBL(DisasContext *s, arg_VTBL *a) return false; } n <<= 3; + tmp = tcg_temp_new_i32(); if (a->op) { - tmp = neon_load_reg(a->vd, 0); + read_neon_element32(tmp, a->vd, 0, MO_32); } else { - tmp = tcg_temp_new_i32(); tcg_gen_movi_i32(tmp, 0); } - tmp2 = neon_load_reg(a->vm, 0); + tmp2 = tcg_temp_new_i32(); + read_neon_element32(tmp2, a->vm, 0, MO_32); ptr1 = vfp_reg_ptr(true, a->vn); tmp4 = tcg_const_i32(n); gen_helper_neon_tbl(tmp2, tmp2, tmp, ptr1, tmp4); - tcg_temp_free_i32(tmp); + if (a->op) { - tmp = neon_load_reg(a->vd, 1); + read_neon_element32(tmp, a->vd, 1, MO_32); } else { - tmp = tcg_temp_new_i32(); tcg_gen_movi_i32(tmp, 0); } - tmp3 = neon_load_reg(a->vm, 1); + tmp3 = tcg_temp_new_i32(); + read_neon_element32(tmp3, a->vm, 1, MO_32); gen_helper_neon_tbl(tmp3, tmp3, tmp, ptr1, tmp4); + tcg_temp_free_i32(tmp); tcg_temp_free_i32(tmp4); tcg_temp_free_ptr(ptr1); - neon_store_reg(a->vd, 0, tmp2); - neon_store_reg(a->vd, 1, tmp3); - tcg_temp_free_i32(tmp); + + write_neon_element32(tmp2, a->vd, 0, MO_32); + write_neon_element32(tmp3, a->vd, 1, MO_32); + tcg_temp_free_i32(tmp2); + tcg_temp_free_i32(tmp3); return true; } @@ -XXX,XX +XXX,XX @@ static bool trans_VDUP_scalar(DisasContext *s, arg_VDUP_scalar *a) static bool trans_VREV64(DisasContext *s, arg_VREV64 *a) { int pass, half; + TCGv_i32 tmp[2]; if (!arm_dc_feature(s, ARM_FEATURE_NEON)) { return false; @@ -XXX,XX +XXX,XX @@ static bool trans_VREV64(DisasContext *s, arg_VREV64 *a) return true; } - for (pass = 0; pass < (a->q ? 2 : 1); pass++) { - TCGv_i32 tmp[2]; + tmp[0] = tcg_temp_new_i32(); + tmp[1] = tcg_temp_new_i32(); + for (pass = 0; pass < (a->q ? 2 : 1); pass++) { for (half = 0; half < 2; half++) { - tmp[half] = neon_load_reg(a->vm, pass * 2 + half); + read_neon_element32(tmp[half], a->vm, pass * 2 + half, MO_32); switch (a->size) { case 0: tcg_gen_bswap32_i32(tmp[half], tmp[half]); @@ -XXX,XX +XXX,XX @@ static bool trans_VREV64(DisasContext *s, arg_VREV64 *a) g_assert_not_reached(); } } - neon_store_reg(a->vd, pass * 2, tmp[1]); - neon_store_reg(a->vd, pass * 2 + 1, tmp[0]); + write_neon_element32(tmp[1], a->vd, pass * 2, MO_32); + write_neon_element32(tmp[0], a->vd, pass * 2 + 1, MO_32); } + + tcg_temp_free_i32(tmp[0]); + tcg_temp_free_i32(tmp[1]); return true; } @@ -XXX,XX +XXX,XX @@ static bool do_2misc_pairwise(DisasContext *s, arg_2misc *a, rm0_64 = tcg_temp_new_i64(); rm1_64 = tcg_temp_new_i64(); rd_64 = tcg_temp_new_i64(); - tmp = neon_load_reg(a->vm, pass * 2); + + tmp = tcg_temp_new_i32(); + read_neon_element32(tmp, a->vm, pass * 2, MO_32); widenfn(rm0_64, tmp); - tcg_temp_free_i32(tmp); - tmp = neon_load_reg(a->vm, pass * 2 + 1); + read_neon_element32(tmp, a->vm, pass * 2 + 1, MO_32); widenfn(rm1_64, tmp); tcg_temp_free_i32(tmp); + opfn(rd_64, rm0_64, rm1_64); tcg_temp_free_i64(rm0_64); tcg_temp_free_i64(rm1_64); @@ -XXX,XX +XXX,XX @@ static bool do_vmovn(DisasContext *s, arg_2misc *a, narrowfn(rd0, cpu_env, rm); neon_load_reg64(rm, a->vm + 1); narrowfn(rd1, cpu_env, rm); - neon_store_reg(a->vd, 0, rd0); - neon_store_reg(a->vd, 1, rd1); + write_neon_element32(rd0, a->vd, 0, MO_32); + write_neon_element32(rd1, a->vd, 1, MO_32); + tcg_temp_free_i32(rd0); + tcg_temp_free_i32(rd1); tcg_temp_free_i64(rm); return true; } @@ -XXX,XX +XXX,XX @@ static bool trans_VSHLL(DisasContext *s, arg_2misc *a) } rd = tcg_temp_new_i64(); + rm0 = tcg_temp_new_i32(); + rm1 = tcg_temp_new_i32(); - rm0 = neon_load_reg(a->vm, 0); - rm1 = neon_load_reg(a->vm, 1); + read_neon_element32(rm0, a->vm, 0, MO_32); + read_neon_element32(rm1, a->vm, 1, MO_32); widenfn(rd, rm0); tcg_gen_shli_i64(rd, rd, 8 << a->size); @@ -XXX,XX +XXX,XX @@ static bool trans_VCVT_F16_F32(DisasContext *s, arg_2misc *a) fpst = fpstatus_ptr(FPST_STD); ahp = get_ahp_flag(); - tmp = neon_load_reg(a->vm, 0); + tmp = tcg_temp_new_i32(); + read_neon_element32(tmp, a->vm, 0, MO_32); gen_helper_vfp_fcvt_f32_to_f16(tmp, tmp, fpst, ahp); - tmp2 = neon_load_reg(a->vm, 1); + tmp2 = tcg_temp_new_i32(); + read_neon_element32(tmp2, a->vm, 1, MO_32); gen_helper_vfp_fcvt_f32_to_f16(tmp2, tmp2, fpst, ahp); tcg_gen_shli_i32(tmp2, tmp2, 16); tcg_gen_or_i32(tmp2, tmp2, tmp); - tcg_temp_free_i32(tmp); - tmp = neon_load_reg(a->vm, 2); + read_neon_element32(tmp, a->vm, 2, MO_32); gen_helper_vfp_fcvt_f32_to_f16(tmp, tmp, fpst, ahp); - tmp3 = neon_load_reg(a->vm, 3); - neon_store_reg(a->vd, 0, tmp2); + tmp3 = tcg_temp_new_i32(); + read_neon_element32(tmp3, a->vm, 3, MO_32); + write_neon_element32(tmp2, a->vd, 0, MO_32); + tcg_temp_free_i32(tmp2); gen_helper_vfp_fcvt_f32_to_f16(tmp3, tmp3, fpst, ahp); tcg_gen_shli_i32(tmp3, tmp3, 16); tcg_gen_or_i32(tmp3, tmp3, tmp); - neon_store_reg(a->vd, 1, tmp3); + write_neon_element32(tmp3, a->vd, 1, MO_32); + tcg_temp_free_i32(tmp3); tcg_temp_free_i32(tmp); tcg_temp_free_i32(ahp); tcg_temp_free_ptr(fpst); @@ -XXX,XX +XXX,XX @@ static bool trans_VCVT_F32_F16(DisasContext *s, arg_2misc *a) fpst = fpstatus_ptr(FPST_STD); ahp = get_ahp_flag(); tmp3 = tcg_temp_new_i32(); - tmp = neon_load_reg(a->vm, 0); - tmp2 = neon_load_reg(a->vm, 1); + tmp2 = tcg_temp_new_i32(); + tmp = tcg_temp_new_i32(); + read_neon_element32(tmp, a->vm, 0, MO_32); + read_neon_element32(tmp2, a->vm, 1, MO_32); tcg_gen_ext16u_i32(tmp3, tmp); gen_helper_vfp_fcvt_f16_to_f32(tmp3, tmp3, fpst, ahp); - neon_store_reg(a->vd, 0, tmp3); + write_neon_element32(tmp3, a->vd, 0, MO_32); tcg_gen_shri_i32(tmp, tmp, 16); gen_helper_vfp_fcvt_f16_to_f32(tmp, tmp, fpst, ahp); - neon_store_reg(a->vd, 1, tmp); - tmp3 = tcg_temp_new_i32(); + write_neon_element32(tmp, a->vd, 1, MO_32); + tcg_temp_free_i32(tmp); tcg_gen_ext16u_i32(tmp3, tmp2); gen_helper_vfp_fcvt_f16_to_f32(tmp3, tmp3, fpst, ahp); - neon_store_reg(a->vd, 2, tmp3); + write_neon_element32(tmp3, a->vd, 2, MO_32); + tcg_temp_free_i32(tmp3); tcg_gen_shri_i32(tmp2, tmp2, 16); gen_helper_vfp_fcvt_f16_to_f32(tmp2, tmp2, fpst, ahp); - neon_store_reg(a->vd, 3, tmp2); + write_neon_element32(tmp2, a->vd, 3, MO_32); + tcg_temp_free_i32(tmp2); tcg_temp_free_i32(ahp); tcg_temp_free_ptr(fpst); @@ -XXX,XX +XXX,XX @@ DO_2M_CRYPTO(SHA256SU0, aa32_sha2, 2) static bool do_2misc(DisasContext *s, arg_2misc *a, NeonGenOneOpFn *fn) { + TCGv_i32 tmp; int pass; /* Handle a 2-reg-misc operation by iterating 32 bits at a time */ @@ -XXX,XX +XXX,XX @@ static bool do_2misc(DisasContext *s, arg_2misc *a, NeonGenOneOpFn *fn) return true; } + tmp = tcg_temp_new_i32(); for (pass = 0; pass < (a->q ? 4 : 2); pass++) { - TCGv_i32 tmp = neon_load_reg(a->vm, pass); + read_neon_element32(tmp, a->vm, pass, MO_32); fn(tmp, tmp); - neon_store_reg(a->vd, pass, tmp); + write_neon_element32(tmp, a->vd, pass, MO_32); } + tcg_temp_free_i32(tmp); return true; } @@ -XXX,XX +XXX,XX @@ static bool trans_VTRN(DisasContext *s, arg_2misc *a) return true; } - if (a->size == 2) { + tmp = tcg_temp_new_i32(); + tmp2 = tcg_temp_new_i32(); + if (a->size == MO_32) { for (pass = 0; pass < (a->q ? 4 : 2); pass += 2) { - tmp = neon_load_reg(a->vm, pass); - tmp2 = neon_load_reg(a->vd, pass + 1); - neon_store_reg(a->vm, pass, tmp2); - neon_store_reg(a->vd, pass + 1, tmp); + read_neon_element32(tmp, a->vm, pass, MO_32); + read_neon_element32(tmp2, a->vd, pass + 1, MO_32); + write_neon_element32(tmp2, a->vm, pass, MO_32); + write_neon_element32(tmp, a->vd, pass + 1, MO_32); } } else { for (pass = 0; pass < (a->q ? 4 : 2); pass++) { - tmp = neon_load_reg(a->vm, pass); - tmp2 = neon_load_reg(a->vd, pass); - if (a->size == 0) { + read_neon_element32(tmp, a->vm, pass, MO_32); + read_neon_element32(tmp2, a->vd, pass, MO_32); + if (a->size == MO_8) { gen_neon_trn_u8(tmp, tmp2); } else { gen_neon_trn_u16(tmp, tmp2); } - neon_store_reg(a->vm, pass, tmp2); - neon_store_reg(a->vd, pass, tmp); + write_neon_element32(tmp2, a->vm, pass, MO_32); + write_neon_element32(tmp, a->vd, pass, MO_32); } } + tcg_temp_free_i32(tmp); + tcg_temp_free_i32(tmp2); return true; } -- 2.20.1 From: Richard Henderson <richard.henderson@linaro.org> We can then use this to improve VMOV (scalar to gp) and VMOV (gp to scalar) so that we simply perform the memory operation that we wanted, rather than inserting or extracting from a 32-bit quantity. These were the last uses of neon_load/store_reg, so remove them. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20201030022618.785675-7-richard.henderson@linaro.org Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/translate.c | 50 +++++++++++++----------- target/arm/translate-vfp.c.inc | 71 +++++----------------------------- 2 files changed, 37 insertions(+), 84 deletions(-) diff --git a/target/arm/translate.c b/target/arm/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate.c +++ b/target/arm/translate.c @@ -XXX,XX +XXX,XX @@ static long neon_full_reg_offset(unsigned reg) * Return the offset of a 2**SIZE piece of a NEON register, at index ELE, * where 0 is the least significant end of the register. */ -static long neon_element_offset(int reg, int element, MemOp size) +static long neon_element_offset(int reg, int element, MemOp memop) { - int element_size = 1 << size; + int element_size = 1 << (memop & MO_SIZE); int ofs = element * element_size; #ifdef HOST_WORDS_BIGENDIAN /* @@ -XXX,XX +XXX,XX @@ static long vfp_reg_offset(bool dp, unsigned reg) } } -static TCGv_i32 neon_load_reg(int reg, int pass) -{ - TCGv_i32 tmp = tcg_temp_new_i32(); - tcg_gen_ld_i32(tmp, cpu_env, neon_element_offset(reg, pass, MO_32)); - return tmp; -} - -static void neon_store_reg(int reg, int pass, TCGv_i32 var) -{ - tcg_gen_st_i32(var, cpu_env, neon_element_offset(reg, pass, MO_32)); - tcg_temp_free_i32(var); -} - static inline void neon_load_reg64(TCGv_i64 var, int reg) { tcg_gen_ld_i64(var, cpu_env, vfp_reg_offset(1, reg)); @@ -XXX,XX +XXX,XX @@ static inline void neon_store_reg32(TCGv_i32 var, int reg) tcg_gen_st_i32(var, cpu_env, vfp_reg_offset(false, reg)); } -static void read_neon_element32(TCGv_i32 dest, int reg, int ele, MemOp size) +static void read_neon_element32(TCGv_i32 dest, int reg, int ele, MemOp memop) { - long off = neon_element_offset(reg, ele, size); + long off = neon_element_offset(reg, ele, memop); - switch (size) { - case MO_32: + switch (memop) { + case MO_SB: + tcg_gen_ld8s_i32(dest, cpu_env, off); + break; + case MO_UB: + tcg_gen_ld8u_i32(dest, cpu_env, off); + break; + case MO_SW: + tcg_gen_ld16s_i32(dest, cpu_env, off); + break; + case MO_UW: + tcg_gen_ld16u_i32(dest, cpu_env, off); + break; + case MO_UL: + case MO_SL: tcg_gen_ld_i32(dest, cpu_env, off); break; default: @@ -XXX,XX +XXX,XX @@ static void read_neon_element32(TCGv_i32 dest, int reg, int ele, MemOp size) } } -static void write_neon_element32(TCGv_i32 src, int reg, int ele, MemOp size) +static void write_neon_element32(TCGv_i32 src, int reg, int ele, MemOp memop) { - long off = neon_element_offset(reg, ele, size); + long off = neon_element_offset(reg, ele, memop); - switch (size) { + switch (memop) { + case MO_8: + tcg_gen_st8_i32(src, cpu_env, off); + break; + case MO_16: + tcg_gen_st16_i32(src, cpu_env, off); + break; case MO_32: tcg_gen_st_i32(src, cpu_env, off); break; diff --git a/target/arm/translate-vfp.c.inc b/target/arm/translate-vfp.c.inc index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-vfp.c.inc +++ b/target/arm/translate-vfp.c.inc @@ -XXX,XX +XXX,XX @@ static bool trans_VMOV_to_gp(DisasContext *s, arg_VMOV_to_gp *a) { /* VMOV scalar to general purpose register */ TCGv_i32 tmp; - int pass; - uint32_t offset; - /* SIZE == 2 is a VFP instruction; otherwise NEON. */ - if (a->size == 2 + /* SIZE == MO_32 is a VFP instruction; otherwise NEON. */ + if (a->size == MO_32 ? !dc_isar_feature(aa32_fpsp_v2, s) : !arm_dc_feature(s, ARM_FEATURE_NEON)) { return false; @@ -XXX,XX +XXX,XX @@ static bool trans_VMOV_to_gp(DisasContext *s, arg_VMOV_to_gp *a) return false; } - offset = a->index << a->size; - pass = extract32(offset, 2, 1); - offset = extract32(offset, 0, 2) * 8; - if (!vfp_access_check(s)) { return true; } - tmp = neon_load_reg(a->vn, pass); - switch (a->size) { - case 0: - if (offset) { - tcg_gen_shri_i32(tmp, tmp, offset); - } - if (a->u) { - gen_uxtb(tmp); - } else { - gen_sxtb(tmp); - } - break; - case 1: - if (a->u) { - if (offset) { - tcg_gen_shri_i32(tmp, tmp, 16); - } else { - gen_uxth(tmp); - } - } else { - if (offset) { - tcg_gen_sari_i32(tmp, tmp, 16); - } else { - gen_sxth(tmp); - } - } - break; - case 2: - break; - } + tmp = tcg_temp_new_i32(); + read_neon_element32(tmp, a->vn, a->index, a->size | (a->u ? 0 : MO_SIGN)); store_reg(s, a->rt, tmp); return true; @@ -XXX,XX +XXX,XX @@ static bool trans_VMOV_to_gp(DisasContext *s, arg_VMOV_to_gp *a) static bool trans_VMOV_from_gp(DisasContext *s, arg_VMOV_from_gp *a) { /* VMOV general purpose register to scalar */ - TCGv_i32 tmp, tmp2; - int pass; - uint32_t offset; + TCGv_i32 tmp; - /* SIZE == 2 is a VFP instruction; otherwise NEON. */ - if (a->size == 2 + /* SIZE == MO_32 is a VFP instruction; otherwise NEON. */ + if (a->size == MO_32 ? !dc_isar_feature(aa32_fpsp_v2, s) : !arm_dc_feature(s, ARM_FEATURE_NEON)) { return false; @@ -XXX,XX +XXX,XX @@ static bool trans_VMOV_from_gp(DisasContext *s, arg_VMOV_from_gp *a) return false; } - offset = a->index << a->size; - pass = extract32(offset, 2, 1); - offset = extract32(offset, 0, 2) * 8; - if (!vfp_access_check(s)) { return true; } tmp = load_reg(s, a->rt); - switch (a->size) { - case 0: - tmp2 = neon_load_reg(a->vn, pass); - tcg_gen_deposit_i32(tmp, tmp2, tmp, offset, 8); - tcg_temp_free_i32(tmp2); - break; - case 1: - tmp2 = neon_load_reg(a->vn, pass); - tcg_gen_deposit_i32(tmp, tmp2, tmp, offset, 16); - tcg_temp_free_i32(tmp2); - break; - case 2: - break; - } - neon_store_reg(a->vn, pass, tmp); + write_neon_element32(tmp, a->vn, a->index, a->size); + tcg_temp_free_i32(tmp); return true; } -- 2.20.1 From: Richard Henderson <richard.henderson@linaro.org> The only uses of this function are for loading VFP single-precision values, and nothing to do with NEON. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20201030022618.785675-8-richard.henderson@linaro.org Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/translate.c | 4 +- target/arm/translate-vfp.c.inc | 184 ++++++++++++++++----------------- 2 files changed, 94 insertions(+), 94 deletions(-) diff --git a/target/arm/translate.c b/target/arm/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate.c +++ b/target/arm/translate.c @@ -XXX,XX +XXX,XX @@ static inline void neon_store_reg64(TCGv_i64 var, int reg) tcg_gen_st_i64(var, cpu_env, vfp_reg_offset(1, reg)); } -static inline void neon_load_reg32(TCGv_i32 var, int reg) +static inline void vfp_load_reg32(TCGv_i32 var, int reg) { tcg_gen_ld_i32(var, cpu_env, vfp_reg_offset(false, reg)); } -static inline void neon_store_reg32(TCGv_i32 var, int reg) +static inline void vfp_store_reg32(TCGv_i32 var, int reg) { tcg_gen_st_i32(var, cpu_env, vfp_reg_offset(false, reg)); } diff --git a/target/arm/translate-vfp.c.inc b/target/arm/translate-vfp.c.inc index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-vfp.c.inc +++ b/target/arm/translate-vfp.c.inc @@ -XXX,XX +XXX,XX @@ static bool trans_VSEL(DisasContext *s, arg_VSEL *a) frn = tcg_temp_new_i32(); frm = tcg_temp_new_i32(); dest = tcg_temp_new_i32(); - neon_load_reg32(frn, rn); - neon_load_reg32(frm, rm); + vfp_load_reg32(frn, rn); + vfp_load_reg32(frm, rm); switch (a->cc) { case 0: /* eq: Z */ tcg_gen_movcond_i32(TCG_COND_EQ, dest, cpu_ZF, zero, @@ -XXX,XX +XXX,XX @@ static bool trans_VSEL(DisasContext *s, arg_VSEL *a) if (sz == 1) { tcg_gen_andi_i32(dest, dest, 0xffff); } - neon_store_reg32(dest, rd); + vfp_store_reg32(dest, rd); tcg_temp_free_i32(frn); tcg_temp_free_i32(frm); tcg_temp_free_i32(dest); @@ -XXX,XX +XXX,XX @@ static bool trans_VRINT(DisasContext *s, arg_VRINT *a) TCGv_i32 tcg_res; tcg_op = tcg_temp_new_i32(); tcg_res = tcg_temp_new_i32(); - neon_load_reg32(tcg_op, rm); + vfp_load_reg32(tcg_op, rm); if (sz == 1) { gen_helper_rinth(tcg_res, tcg_op, fpst); } else { gen_helper_rints(tcg_res, tcg_op, fpst); } - neon_store_reg32(tcg_res, rd); + vfp_store_reg32(tcg_res, rd); tcg_temp_free_i32(tcg_op); tcg_temp_free_i32(tcg_res); } @@ -XXX,XX +XXX,XX @@ static bool trans_VCVT(DisasContext *s, arg_VCVT *a) gen_helper_vfp_tould(tcg_res, tcg_double, tcg_shift, fpst); } tcg_gen_extrl_i64_i32(tcg_tmp, tcg_res); - neon_store_reg32(tcg_tmp, rd); + vfp_store_reg32(tcg_tmp, rd); tcg_temp_free_i32(tcg_tmp); tcg_temp_free_i64(tcg_res); tcg_temp_free_i64(tcg_double); @@ -XXX,XX +XXX,XX @@ static bool trans_VCVT(DisasContext *s, arg_VCVT *a) TCGv_i32 tcg_single, tcg_res; tcg_single = tcg_temp_new_i32(); tcg_res = tcg_temp_new_i32(); - neon_load_reg32(tcg_single, rm); + vfp_load_reg32(tcg_single, rm); if (sz == 1) { if (is_signed) { gen_helper_vfp_toslh(tcg_res, tcg_single, tcg_shift, fpst); @@ -XXX,XX +XXX,XX @@ static bool trans_VCVT(DisasContext *s, arg_VCVT *a) gen_helper_vfp_touls(tcg_res, tcg_single, tcg_shift, fpst); } } - neon_store_reg32(tcg_res, rd); + vfp_store_reg32(tcg_res, rd); tcg_temp_free_i32(tcg_res); tcg_temp_free_i32(tcg_single); } @@ -XXX,XX +XXX,XX @@ static bool trans_VMOV_half(DisasContext *s, arg_VMOV_single *a) if (a->l) { /* VFP to general purpose register */ tmp = tcg_temp_new_i32(); - neon_load_reg32(tmp, a->vn); + vfp_load_reg32(tmp, a->vn); tcg_gen_andi_i32(tmp, tmp, 0xffff); store_reg(s, a->rt, tmp); } else { /* general purpose register to VFP */ tmp = load_reg(s, a->rt); tcg_gen_andi_i32(tmp, tmp, 0xffff); - neon_store_reg32(tmp, a->vn); + vfp_store_reg32(tmp, a->vn); tcg_temp_free_i32(tmp); } @@ -XXX,XX +XXX,XX @@ static bool trans_VMOV_single(DisasContext *s, arg_VMOV_single *a) if (a->l) { /* VFP to general purpose register */ tmp = tcg_temp_new_i32(); - neon_load_reg32(tmp, a->vn); + vfp_load_reg32(tmp, a->vn); if (a->rt == 15) { /* Set the 4 flag bits in the CPSR. */ gen_set_nzcv(tmp); @@ -XXX,XX +XXX,XX @@ static bool trans_VMOV_single(DisasContext *s, arg_VMOV_single *a) } else { /* general purpose register to VFP */ tmp = load_reg(s, a->rt); - neon_store_reg32(tmp, a->vn); + vfp_store_reg32(tmp, a->vn); tcg_temp_free_i32(tmp); } @@ -XXX,XX +XXX,XX @@ static bool trans_VMOV_64_sp(DisasContext *s, arg_VMOV_64_sp *a) if (a->op) { /* fpreg to gpreg */ tmp = tcg_temp_new_i32(); - neon_load_reg32(tmp, a->vm); + vfp_load_reg32(tmp, a->vm); store_reg(s, a->rt, tmp); tmp = tcg_temp_new_i32(); - neon_load_reg32(tmp, a->vm + 1); + vfp_load_reg32(tmp, a->vm + 1); store_reg(s, a->rt2, tmp); } else { /* gpreg to fpreg */ tmp = load_reg(s, a->rt); - neon_store_reg32(tmp, a->vm); + vfp_store_reg32(tmp, a->vm); tcg_temp_free_i32(tmp); tmp = load_reg(s, a->rt2); - neon_store_reg32(tmp, a->vm + 1); + vfp_store_reg32(tmp, a->vm + 1); tcg_temp_free_i32(tmp); } @@ -XXX,XX +XXX,XX @@ static bool trans_VMOV_64_dp(DisasContext *s, arg_VMOV_64_dp *a) if (a->op) { /* fpreg to gpreg */ tmp = tcg_temp_new_i32(); - neon_load_reg32(tmp, a->vm * 2); + vfp_load_reg32(tmp, a->vm * 2); store_reg(s, a->rt, tmp); tmp = tcg_temp_new_i32(); - neon_load_reg32(tmp, a->vm * 2 + 1); + vfp_load_reg32(tmp, a->vm * 2 + 1); store_reg(s, a->rt2, tmp); } else { /* gpreg to fpreg */ tmp = load_reg(s, a->rt); - neon_store_reg32(tmp, a->vm * 2); + vfp_store_reg32(tmp, a->vm * 2); tcg_temp_free_i32(tmp); tmp = load_reg(s, a->rt2); - neon_store_reg32(tmp, a->vm * 2 + 1); + vfp_store_reg32(tmp, a->vm * 2 + 1); tcg_temp_free_i32(tmp); } @@ -XXX,XX +XXX,XX @@ static bool trans_VLDR_VSTR_hp(DisasContext *s, arg_VLDR_VSTR_sp *a) tmp = tcg_temp_new_i32(); if (a->l) { gen_aa32_ld16u(s, tmp, addr, get_mem_index(s)); - neon_store_reg32(tmp, a->vd); + vfp_store_reg32(tmp, a->vd); } else { - neon_load_reg32(tmp, a->vd); + vfp_load_reg32(tmp, a->vd); gen_aa32_st16(s, tmp, addr, get_mem_index(s)); } tcg_temp_free_i32(tmp); @@ -XXX,XX +XXX,XX @@ static bool trans_VLDR_VSTR_sp(DisasContext *s, arg_VLDR_VSTR_sp *a) tmp = tcg_temp_new_i32(); if (a->l) { gen_aa32_ld32u(s, tmp, addr, get_mem_index(s)); - neon_store_reg32(tmp, a->vd); + vfp_store_reg32(tmp, a->vd); } else { - neon_load_reg32(tmp, a->vd); + vfp_load_reg32(tmp, a->vd); gen_aa32_st32(s, tmp, addr, get_mem_index(s)); } tcg_temp_free_i32(tmp); @@ -XXX,XX +XXX,XX @@ static bool trans_VLDM_VSTM_sp(DisasContext *s, arg_VLDM_VSTM_sp *a) if (a->l) { /* load */ gen_aa32_ld32u(s, tmp, addr, get_mem_index(s)); - neon_store_reg32(tmp, a->vd + i); + vfp_store_reg32(tmp, a->vd + i); } else { /* store */ - neon_load_reg32(tmp, a->vd + i); + vfp_load_reg32(tmp, a->vd + i); gen_aa32_st32(s, tmp, addr, get_mem_index(s)); } tcg_gen_addi_i32(addr, addr, offset); @@ -XXX,XX +XXX,XX @@ static bool do_vfp_3op_sp(DisasContext *s, VFPGen3OpSPFn *fn, fd = tcg_temp_new_i32(); fpst = fpstatus_ptr(FPST_FPCR); - neon_load_reg32(f0, vn); - neon_load_reg32(f1, vm); + vfp_load_reg32(f0, vn); + vfp_load_reg32(f1, vm); for (;;) { if (reads_vd) { - neon_load_reg32(fd, vd); + vfp_load_reg32(fd, vd); } fn(fd, f0, f1, fpst); - neon_store_reg32(fd, vd); + vfp_store_reg32(fd, vd); if (veclen == 0) { break; @@ -XXX,XX +XXX,XX @@ static bool do_vfp_3op_sp(DisasContext *s, VFPGen3OpSPFn *fn, veclen--; vd = vfp_advance_sreg(vd, delta_d); vn = vfp_advance_sreg(vn, delta_d); - neon_load_reg32(f0, vn); + vfp_load_reg32(f0, vn); if (delta_m) { vm = vfp_advance_sreg(vm, delta_m); - neon_load_reg32(f1, vm); + vfp_load_reg32(f1, vm); } } @@ -XXX,XX +XXX,XX @@ static bool do_vfp_3op_hp(DisasContext *s, VFPGen3OpSPFn *fn, fd = tcg_temp_new_i32(); fpst = fpstatus_ptr(FPST_FPCR_F16); - neon_load_reg32(f0, vn); - neon_load_reg32(f1, vm); + vfp_load_reg32(f0, vn); + vfp_load_reg32(f1, vm); if (reads_vd) { - neon_load_reg32(fd, vd); + vfp_load_reg32(fd, vd); } fn(fd, f0, f1, fpst); - neon_store_reg32(fd, vd); + vfp_store_reg32(fd, vd); tcg_temp_free_i32(f0); tcg_temp_free_i32(f1); @@ -XXX,XX +XXX,XX @@ static bool do_vfp_2op_sp(DisasContext *s, VFPGen2OpSPFn *fn, int vd, int vm) f0 = tcg_temp_new_i32(); fd = tcg_temp_new_i32(); - neon_load_reg32(f0, vm); + vfp_load_reg32(f0, vm); for (;;) { fn(fd, f0); - neon_store_reg32(fd, vd); + vfp_store_reg32(fd, vd); if (veclen == 0) { break; @@ -XXX,XX +XXX,XX @@ static bool do_vfp_2op_sp(DisasContext *s, VFPGen2OpSPFn *fn, int vd, int vm) /* single source one-many */ while (veclen--) { vd = vfp_advance_sreg(vd, delta_d); - neon_store_reg32(fd, vd); + vfp_store_reg32(fd, vd); } break; } @@ -XXX,XX +XXX,XX @@ static bool do_vfp_2op_sp(DisasContext *s, VFPGen2OpSPFn *fn, int vd, int vm) veclen--; vd = vfp_advance_sreg(vd, delta_d); vm = vfp_advance_sreg(vm, delta_m); - neon_load_reg32(f0, vm); + vfp_load_reg32(f0, vm); } tcg_temp_free_i32(f0); @@ -XXX,XX +XXX,XX @@ static bool do_vfp_2op_hp(DisasContext *s, VFPGen2OpSPFn *fn, int vd, int vm) } f0 = tcg_temp_new_i32(); - neon_load_reg32(f0, vm); + vfp_load_reg32(f0, vm); fn(f0, f0); - neon_store_reg32(f0, vd); + vfp_store_reg32(f0, vd); tcg_temp_free_i32(f0); return true; @@ -XXX,XX +XXX,XX @@ static bool do_vfm_hp(DisasContext *s, arg_VFMA_sp *a, bool neg_n, bool neg_d) vm = tcg_temp_new_i32(); vd = tcg_temp_new_i32(); - neon_load_reg32(vn, a->vn); - neon_load_reg32(vm, a->vm); + vfp_load_reg32(vn, a->vn); + vfp_load_reg32(vm, a->vm); if (neg_n) { /* VFNMS, VFMS */ gen_helper_vfp_negh(vn, vn); } - neon_load_reg32(vd, a->vd); + vfp_load_reg32(vd, a->vd); if (neg_d) { /* VFNMA, VFNMS */ gen_helper_vfp_negh(vd, vd); } fpst = fpstatus_ptr(FPST_FPCR_F16); gen_helper_vfp_muladdh(vd, vn, vm, vd, fpst); - neon_store_reg32(vd, a->vd); + vfp_store_reg32(vd, a->vd); tcg_temp_free_ptr(fpst); tcg_temp_free_i32(vn); @@ -XXX,XX +XXX,XX @@ static bool do_vfm_sp(DisasContext *s, arg_VFMA_sp *a, bool neg_n, bool neg_d) vm = tcg_temp_new_i32(); vd = tcg_temp_new_i32(); - neon_load_reg32(vn, a->vn); - neon_load_reg32(vm, a->vm); + vfp_load_reg32(vn, a->vn); + vfp_load_reg32(vm, a->vm); if (neg_n) { /* VFNMS, VFMS */ gen_helper_vfp_negs(vn, vn); } - neon_load_reg32(vd, a->vd); + vfp_load_reg32(vd, a->vd); if (neg_d) { /* VFNMA, VFNMS */ gen_helper_vfp_negs(vd, vd); } fpst = fpstatus_ptr(FPST_FPCR); gen_helper_vfp_muladds(vd, vn, vm, vd, fpst); - neon_store_reg32(vd, a->vd); + vfp_store_reg32(vd, a->vd); tcg_temp_free_ptr(fpst); tcg_temp_free_i32(vn); @@ -XXX,XX +XXX,XX @@ static bool trans_VMOV_imm_hp(DisasContext *s, arg_VMOV_imm_sp *a) } fd = tcg_const_i32(vfp_expand_imm(MO_16, a->imm)); - neon_store_reg32(fd, a->vd); + vfp_store_reg32(fd, a->vd); tcg_temp_free_i32(fd); return true; } @@ -XXX,XX +XXX,XX @@ static bool trans_VMOV_imm_sp(DisasContext *s, arg_VMOV_imm_sp *a) fd = tcg_const_i32(vfp_expand_imm(MO_32, a->imm)); for (;;) { - neon_store_reg32(fd, vd); + vfp_store_reg32(fd, vd); if (veclen == 0) { break; @@ -XXX,XX +XXX,XX @@ static bool trans_VCMP_hp(DisasContext *s, arg_VCMP_sp *a) vd = tcg_temp_new_i32(); vm = tcg_temp_new_i32(); - neon_load_reg32(vd, a->vd); + vfp_load_reg32(vd, a->vd); if (a->z) { tcg_gen_movi_i32(vm, 0); } else { - neon_load_reg32(vm, a->vm); + vfp_load_reg32(vm, a->vm); } if (a->e) { @@ -XXX,XX +XXX,XX @@ static bool trans_VCMP_sp(DisasContext *s, arg_VCMP_sp *a) vd = tcg_temp_new_i32(); vm = tcg_temp_new_i32(); - neon_load_reg32(vd, a->vd); + vfp_load_reg32(vd, a->vd); if (a->z) { tcg_gen_movi_i32(vm, 0); } else { - neon_load_reg32(vm, a->vm); + vfp_load_reg32(vm, a->vm); } if (a->e) { @@ -XXX,XX +XXX,XX @@ static bool trans_VCVT_f32_f16(DisasContext *s, arg_VCVT_f32_f16 *a) /* The T bit tells us if we want the low or high 16 bits of Vm */ tcg_gen_ld16u_i32(tmp, cpu_env, vfp_f16_offset(a->vm, a->t)); gen_helper_vfp_fcvt_f16_to_f32(tmp, tmp, fpst, ahp_mode); - neon_store_reg32(tmp, a->vd); + vfp_store_reg32(tmp, a->vd); tcg_temp_free_i32(ahp_mode); tcg_temp_free_ptr(fpst); tcg_temp_free_i32(tmp); @@ -XXX,XX +XXX,XX @@ static bool trans_VCVT_f16_f32(DisasContext *s, arg_VCVT_f16_f32 *a) ahp_mode = get_ahp_flag(); tmp = tcg_temp_new_i32(); - neon_load_reg32(tmp, a->vm); + vfp_load_reg32(tmp, a->vm); gen_helper_vfp_fcvt_f32_to_f16(tmp, tmp, fpst, ahp_mode); tcg_gen_st16_i32(tmp, cpu_env, vfp_f16_offset(a->vd, a->t)); tcg_temp_free_i32(ahp_mode); @@ -XXX,XX +XXX,XX @@ static bool trans_VRINTR_hp(DisasContext *s, arg_VRINTR_sp *a) } tmp = tcg_temp_new_i32(); - neon_load_reg32(tmp, a->vm); + vfp_load_reg32(tmp, a->vm); fpst = fpstatus_ptr(FPST_FPCR_F16); gen_helper_rinth(tmp, tmp, fpst); - neon_store_reg32(tmp, a->vd); + vfp_store_reg32(tmp, a->vd); tcg_temp_free_ptr(fpst); tcg_temp_free_i32(tmp); return true; @@ -XXX,XX +XXX,XX @@ static bool trans_VRINTR_sp(DisasContext *s, arg_VRINTR_sp *a) } tmp = tcg_temp_new_i32(); - neon_load_reg32(tmp, a->vm); + vfp_load_reg32(tmp, a->vm); fpst = fpstatus_ptr(FPST_FPCR); gen_helper_rints(tmp, tmp, fpst); - neon_store_reg32(tmp, a->vd); + vfp_store_reg32(tmp, a->vd); tcg_temp_free_ptr(fpst); tcg_temp_free_i32(tmp); return true; @@ -XXX,XX +XXX,XX @@ static bool trans_VRINTZ_hp(DisasContext *s, arg_VRINTZ_sp *a) } tmp = tcg_temp_new_i32(); - neon_load_reg32(tmp, a->vm); + vfp_load_reg32(tmp, a->vm); fpst = fpstatus_ptr(FPST_FPCR_F16); tcg_rmode = tcg_const_i32(float_round_to_zero); gen_helper_set_rmode(tcg_rmode, tcg_rmode, fpst); gen_helper_rinth(tmp, tmp, fpst); gen_helper_set_rmode(tcg_rmode, tcg_rmode, fpst); - neon_store_reg32(tmp, a->vd); + vfp_store_reg32(tmp, a->vd); tcg_temp_free_ptr(fpst); tcg_temp_free_i32(tcg_rmode); tcg_temp_free_i32(tmp); @@ -XXX,XX +XXX,XX @@ static bool trans_VRINTZ_sp(DisasContext *s, arg_VRINTZ_sp *a) } tmp = tcg_temp_new_i32(); - neon_load_reg32(tmp, a->vm); + vfp_load_reg32(tmp, a->vm); fpst = fpstatus_ptr(FPST_FPCR); tcg_rmode = tcg_const_i32(float_round_to_zero); gen_helper_set_rmode(tcg_rmode, tcg_rmode, fpst); gen_helper_rints(tmp, tmp, fpst); gen_helper_set_rmode(tcg_rmode, tcg_rmode, fpst); - neon_store_reg32(tmp, a->vd); + vfp_store_reg32(tmp, a->vd); tcg_temp_free_ptr(fpst); tcg_temp_free_i32(tcg_rmode); tcg_temp_free_i32(tmp); @@ -XXX,XX +XXX,XX @@ static bool trans_VRINTX_hp(DisasContext *s, arg_VRINTX_sp *a) } tmp = tcg_temp_new_i32(); - neon_load_reg32(tmp, a->vm); + vfp_load_reg32(tmp, a->vm); fpst = fpstatus_ptr(FPST_FPCR_F16); gen_helper_rinth_exact(tmp, tmp, fpst); - neon_store_reg32(tmp, a->vd); + vfp_store_reg32(tmp, a->vd); tcg_temp_free_ptr(fpst); tcg_temp_free_i32(tmp); return true; @@ -XXX,XX +XXX,XX @@ static bool trans_VRINTX_sp(DisasContext *s, arg_VRINTX_sp *a) } tmp = tcg_temp_new_i32(); - neon_load_reg32(tmp, a->vm); + vfp_load_reg32(tmp, a->vm); fpst = fpstatus_ptr(FPST_FPCR); gen_helper_rints_exact(tmp, tmp, fpst); - neon_store_reg32(tmp, a->vd); + vfp_store_reg32(tmp, a->vd); tcg_temp_free_ptr(fpst); tcg_temp_free_i32(tmp); return true; @@ -XXX,XX +XXX,XX @@ static bool trans_VCVT_sp(DisasContext *s, arg_VCVT_sp *a) vm = tcg_temp_new_i32(); vd = tcg_temp_new_i64(); - neon_load_reg32(vm, a->vm); + vfp_load_reg32(vm, a->vm); gen_helper_vfp_fcvtds(vd, vm, cpu_env); neon_store_reg64(vd, a->vd); tcg_temp_free_i32(vm); @@ -XXX,XX +XXX,XX @@ static bool trans_VCVT_dp(DisasContext *s, arg_VCVT_dp *a) vm = tcg_temp_new_i64(); neon_load_reg64(vm, a->vm); gen_helper_vfp_fcvtsd(vd, vm, cpu_env); - neon_store_reg32(vd, a->vd); + vfp_store_reg32(vd, a->vd); tcg_temp_free_i32(vd); tcg_temp_free_i64(vm); return true; @@ -XXX,XX +XXX,XX @@ static bool trans_VCVT_int_hp(DisasContext *s, arg_VCVT_int_sp *a) } vm = tcg_temp_new_i32(); - neon_load_reg32(vm, a->vm); + vfp_load_reg32(vm, a->vm); fpst = fpstatus_ptr(FPST_FPCR_F16); if (a->s) { /* i32 -> f16 */ @@ -XXX,XX +XXX,XX @@ static bool trans_VCVT_int_hp(DisasContext *s, arg_VCVT_int_sp *a) /* u32 -> f16 */ gen_helper_vfp_uitoh(vm, vm, fpst); } - neon_store_reg32(vm, a->vd); + vfp_store_reg32(vm, a->vd); tcg_temp_free_i32(vm); tcg_temp_free_ptr(fpst); return true; @@ -XXX,XX +XXX,XX @@ static bool trans_VCVT_int_sp(DisasContext *s, arg_VCVT_int_sp *a) } vm = tcg_temp_new_i32(); - neon_load_reg32(vm, a->vm); + vfp_load_reg32(vm, a->vm); fpst = fpstatus_ptr(FPST_FPCR); if (a->s) { /* i32 -> f32 */ @@ -XXX,XX +XXX,XX @@ static bool trans_VCVT_int_sp(DisasContext *s, arg_VCVT_int_sp *a) /* u32 -> f32 */ gen_helper_vfp_uitos(vm, vm, fpst); } - neon_store_reg32(vm, a->vd); + vfp_store_reg32(vm, a->vd); tcg_temp_free_i32(vm); tcg_temp_free_ptr(fpst); return true; @@ -XXX,XX +XXX,XX @@ static bool trans_VCVT_int_dp(DisasContext *s, arg_VCVT_int_dp *a) vm = tcg_temp_new_i32(); vd = tcg_temp_new_i64(); - neon_load_reg32(vm, a->vm); + vfp_load_reg32(vm, a->vm); fpst = fpstatus_ptr(FPST_FPCR); if (a->s) { /* i32 -> f64 */ @@ -XXX,XX +XXX,XX @@ static bool trans_VJCVT(DisasContext *s, arg_VJCVT *a) vd = tcg_temp_new_i32(); neon_load_reg64(vm, a->vm); gen_helper_vjcvt(vd, vm, cpu_env); - neon_store_reg32(vd, a->vd); + vfp_store_reg32(vd, a->vd); tcg_temp_free_i64(vm); tcg_temp_free_i32(vd); return true; @@ -XXX,XX +XXX,XX @@ static bool trans_VCVT_fix_hp(DisasContext *s, arg_VCVT_fix_sp *a) frac_bits = (a->opc & 1) ? (32 - a->imm) : (16 - a->imm); vd = tcg_temp_new_i32(); - neon_load_reg32(vd, a->vd); + vfp_load_reg32(vd, a->vd); fpst = fpstatus_ptr(FPST_FPCR_F16); shift = tcg_const_i32(frac_bits); @@ -XXX,XX +XXX,XX @@ static bool trans_VCVT_fix_hp(DisasContext *s, arg_VCVT_fix_sp *a) g_assert_not_reached(); } - neon_store_reg32(vd, a->vd); + vfp_store_reg32(vd, a->vd); tcg_temp_free_i32(vd); tcg_temp_free_i32(shift); tcg_temp_free_ptr(fpst); @@ -XXX,XX +XXX,XX @@ static bool trans_VCVT_fix_sp(DisasContext *s, arg_VCVT_fix_sp *a) frac_bits = (a->opc & 1) ? (32 - a->imm) : (16 - a->imm); vd = tcg_temp_new_i32(); - neon_load_reg32(vd, a->vd); + vfp_load_reg32(vd, a->vd); fpst = fpstatus_ptr(FPST_FPCR); shift = tcg_const_i32(frac_bits); @@ -XXX,XX +XXX,XX @@ static bool trans_VCVT_fix_sp(DisasContext *s, arg_VCVT_fix_sp *a) g_assert_not_reached(); } - neon_store_reg32(vd, a->vd); + vfp_store_reg32(vd, a->vd); tcg_temp_free_i32(vd); tcg_temp_free_i32(shift); tcg_temp_free_ptr(fpst); @@ -XXX,XX +XXX,XX @@ static bool trans_VCVT_hp_int(DisasContext *s, arg_VCVT_sp_int *a) fpst = fpstatus_ptr(FPST_FPCR_F16); vm = tcg_temp_new_i32(); - neon_load_reg32(vm, a->vm); + vfp_load_reg32(vm, a->vm); if (a->s) { if (a->rz) { @@ -XXX,XX +XXX,XX @@ static bool trans_VCVT_hp_int(DisasContext *s, arg_VCVT_sp_int *a) gen_helper_vfp_touih(vm, vm, fpst); } } - neon_store_reg32(vm, a->vd); + vfp_store_reg32(vm, a->vd); tcg_temp_free_i32(vm); tcg_temp_free_ptr(fpst); return true; @@ -XXX,XX +XXX,XX @@ static bool trans_VCVT_sp_int(DisasContext *s, arg_VCVT_sp_int *a) fpst = fpstatus_ptr(FPST_FPCR); vm = tcg_temp_new_i32(); - neon_load_reg32(vm, a->vm); + vfp_load_reg32(vm, a->vm); if (a->s) { if (a->rz) { @@ -XXX,XX +XXX,XX @@ static bool trans_VCVT_sp_int(DisasContext *s, arg_VCVT_sp_int *a) gen_helper_vfp_touis(vm, vm, fpst); } } - neon_store_reg32(vm, a->vd); + vfp_store_reg32(vm, a->vd); tcg_temp_free_i32(vm); tcg_temp_free_ptr(fpst); return true; @@ -XXX,XX +XXX,XX @@ static bool trans_VCVT_dp_int(DisasContext *s, arg_VCVT_dp_int *a) gen_helper_vfp_touid(vd, vm, fpst); } } - neon_store_reg32(vd, a->vd); + vfp_store_reg32(vd, a->vd); tcg_temp_free_i32(vd); tcg_temp_free_i64(vm); tcg_temp_free_ptr(fpst); @@ -XXX,XX +XXX,XX @@ static bool trans_VINS(DisasContext *s, arg_VINS *a) /* Insert low half of Vm into high half of Vd */ rm = tcg_temp_new_i32(); rd = tcg_temp_new_i32(); - neon_load_reg32(rm, a->vm); - neon_load_reg32(rd, a->vd); + vfp_load_reg32(rm, a->vm); + vfp_load_reg32(rd, a->vd); tcg_gen_deposit_i32(rd, rd, rm, 16, 16); - neon_store_reg32(rd, a->vd); + vfp_store_reg32(rd, a->vd); tcg_temp_free_i32(rm); tcg_temp_free_i32(rd); return true; @@ -XXX,XX +XXX,XX @@ static bool trans_VMOVX(DisasContext *s, arg_VINS *a) /* Set Vd to high half of Vm */ rm = tcg_temp_new_i32(); - neon_load_reg32(rm, a->vm); + vfp_load_reg32(rm, a->vm); tcg_gen_shri_i32(rm, rm, 16); - neon_store_reg32(rm, a->vd); + vfp_store_reg32(rm, a->vd); tcg_temp_free_i32(rm); return true; } -- 2.20.1 From: Richard Henderson <richard.henderson@linaro.org> Replace all uses of neon_load/store_reg64 within translate-neon.c.inc. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20201030022618.785675-9-richard.henderson@linaro.org Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/translate.c | 26 +++++++++ target/arm/translate-neon.c.inc | 94 ++++++++++++++++----------------- 2 files changed, 73 insertions(+), 47 deletions(-) diff --git a/target/arm/translate.c b/target/arm/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate.c +++ b/target/arm/translate.c @@ -XXX,XX +XXX,XX @@ static void read_neon_element32(TCGv_i32 dest, int reg, int ele, MemOp memop) } } +static void read_neon_element64(TCGv_i64 dest, int reg, int ele, MemOp memop) +{ + long off = neon_element_offset(reg, ele, memop); + + switch (memop) { + case MO_Q: + tcg_gen_ld_i64(dest, cpu_env, off); + break; + default: + g_assert_not_reached(); + } +} + static void write_neon_element32(TCGv_i32 src, int reg, int ele, MemOp memop) { long off = neon_element_offset(reg, ele, memop); @@ -XXX,XX +XXX,XX @@ static void write_neon_element32(TCGv_i32 src, int reg, int ele, MemOp memop) } } +static void write_neon_element64(TCGv_i64 src, int reg, int ele, MemOp memop) +{ + long off = neon_element_offset(reg, ele, memop); + + switch (memop) { + case MO_64: + tcg_gen_st_i64(src, cpu_env, off); + break; + default: + g_assert_not_reached(); + } +} + static TCGv_ptr vfp_reg_ptr(bool dp, int reg) { TCGv_ptr ret = tcg_temp_new_ptr(); diff --git a/target/arm/translate-neon.c.inc b/target/arm/translate-neon.c.inc index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-neon.c.inc +++ b/target/arm/translate-neon.c.inc @@ -XXX,XX +XXX,XX @@ static bool do_2shift_env_64(DisasContext *s, arg_2reg_shift *a, for (pass = 0; pass < a->q + 1; pass++) { TCGv_i64 tmp = tcg_temp_new_i64(); - neon_load_reg64(tmp, a->vm + pass); + read_neon_element64(tmp, a->vm, pass, MO_64); fn(tmp, cpu_env, tmp, constimm); - neon_store_reg64(tmp, a->vd + pass); + write_neon_element64(tmp, a->vd, pass, MO_64); tcg_temp_free_i64(tmp); } tcg_temp_free_i64(constimm); @@ -XXX,XX +XXX,XX @@ static bool do_2shift_narrow_64(DisasContext *s, arg_2reg_shift *a, rd = tcg_temp_new_i32(); /* Load both inputs first to avoid potential overwrite if rm == rd */ - neon_load_reg64(rm1, a->vm); - neon_load_reg64(rm2, a->vm + 1); + read_neon_element64(rm1, a->vm, 0, MO_64); + read_neon_element64(rm2, a->vm, 1, MO_64); shiftfn(rm1, rm1, constimm); narrowfn(rd, cpu_env, rm1); @@ -XXX,XX +XXX,XX @@ static bool do_vshll_2sh(DisasContext *s, arg_2reg_shift *a, tcg_gen_shli_i64(tmp, tmp, a->shift); tcg_gen_andi_i64(tmp, tmp, ~widen_mask); } - neon_store_reg64(tmp, a->vd); + write_neon_element64(tmp, a->vd, 0, MO_64); widenfn(tmp, rm1); tcg_temp_free_i32(rm1); @@ -XXX,XX +XXX,XX @@ static bool do_vshll_2sh(DisasContext *s, arg_2reg_shift *a, tcg_gen_shli_i64(tmp, tmp, a->shift); tcg_gen_andi_i64(tmp, tmp, ~widen_mask); } - neon_store_reg64(tmp, a->vd + 1); + write_neon_element64(tmp, a->vd, 1, MO_64); tcg_temp_free_i64(tmp); return true; } @@ -XXX,XX +XXX,XX @@ static bool do_prewiden_3d(DisasContext *s, arg_3diff *a, rm_64 = tcg_temp_new_i64(); if (src1_wide) { - neon_load_reg64(rn0_64, a->vn); + read_neon_element64(rn0_64, a->vn, 0, MO_64); } else { TCGv_i32 tmp = tcg_temp_new_i32(); read_neon_element32(tmp, a->vn, 0, MO_32); @@ -XXX,XX +XXX,XX @@ static bool do_prewiden_3d(DisasContext *s, arg_3diff *a, * avoid incorrect results if a narrow input overlaps with the result. */ if (src1_wide) { - neon_load_reg64(rn1_64, a->vn + 1); + read_neon_element64(rn1_64, a->vn, 1, MO_64); } else { TCGv_i32 tmp = tcg_temp_new_i32(); read_neon_element32(tmp, a->vn, 1, MO_32); @@ -XXX,XX +XXX,XX @@ static bool do_prewiden_3d(DisasContext *s, arg_3diff *a, rm = tcg_temp_new_i32(); read_neon_element32(rm, a->vm, 1, MO_32); - neon_store_reg64(rn0_64, a->vd); + write_neon_element64(rn0_64, a->vd, 0, MO_64); widenfn(rm_64, rm); tcg_temp_free_i32(rm); opfn(rn1_64, rn1_64, rm_64); - neon_store_reg64(rn1_64, a->vd + 1); + write_neon_element64(rn1_64, a->vd, 1, MO_64); tcg_temp_free_i64(rn0_64); tcg_temp_free_i64(rn1_64); @@ -XXX,XX +XXX,XX @@ static bool do_narrow_3d(DisasContext *s, arg_3diff *a, rd0 = tcg_temp_new_i32(); rd1 = tcg_temp_new_i32(); - neon_load_reg64(rn_64, a->vn); - neon_load_reg64(rm_64, a->vm); + read_neon_element64(rn_64, a->vn, 0, MO_64); + read_neon_element64(rm_64, a->vm, 0, MO_64); opfn(rn_64, rn_64, rm_64); narrowfn(rd0, rn_64); - neon_load_reg64(rn_64, a->vn + 1); - neon_load_reg64(rm_64, a->vm + 1); + read_neon_element64(rn_64, a->vn, 1, MO_64); + read_neon_element64(rm_64, a->vm, 1, MO_64); opfn(rn_64, rn_64, rm_64); @@ -XXX,XX +XXX,XX @@ static bool do_long_3d(DisasContext *s, arg_3diff *a, /* Don't store results until after all loads: they might overlap */ if (accfn) { tmp = tcg_temp_new_i64(); - neon_load_reg64(tmp, a->vd); + read_neon_element64(tmp, a->vd, 0, MO_64); accfn(tmp, tmp, rd0); - neon_store_reg64(tmp, a->vd); - neon_load_reg64(tmp, a->vd + 1); + write_neon_element64(tmp, a->vd, 0, MO_64); + read_neon_element64(tmp, a->vd, 1, MO_64); accfn(tmp, tmp, rd1); - neon_store_reg64(tmp, a->vd + 1); + write_neon_element64(tmp, a->vd, 1, MO_64); tcg_temp_free_i64(tmp); } else { - neon_store_reg64(rd0, a->vd); - neon_store_reg64(rd1, a->vd + 1); + write_neon_element64(rd0, a->vd, 0, MO_64); + write_neon_element64(rd1, a->vd, 1, MO_64); } tcg_temp_free_i64(rd0); @@ -XXX,XX +XXX,XX @@ static bool do_2scalar_long(DisasContext *s, arg_2scalar *a, if (accfn) { TCGv_i64 t64 = tcg_temp_new_i64(); - neon_load_reg64(t64, a->vd); + read_neon_element64(t64, a->vd, 0, MO_64); accfn(t64, t64, rn0_64); - neon_store_reg64(t64, a->vd); - neon_load_reg64(t64, a->vd + 1); + write_neon_element64(t64, a->vd, 0, MO_64); + read_neon_element64(t64, a->vd, 1, MO_64); accfn(t64, t64, rn1_64); - neon_store_reg64(t64, a->vd + 1); + write_neon_element64(t64, a->vd, 1, MO_64); tcg_temp_free_i64(t64); } else { - neon_store_reg64(rn0_64, a->vd); - neon_store_reg64(rn1_64, a->vd + 1); + write_neon_element64(rn0_64, a->vd, 0, MO_64); + write_neon_element64(rn1_64, a->vd, 1, MO_64); } tcg_temp_free_i64(rn0_64); tcg_temp_free_i64(rn1_64); @@ -XXX,XX +XXX,XX @@ static bool trans_VEXT(DisasContext *s, arg_VEXT *a) right = tcg_temp_new_i64(); dest = tcg_temp_new_i64(); - neon_load_reg64(right, a->vn); - neon_load_reg64(left, a->vm); + read_neon_element64(right, a->vn, 0, MO_64); + read_neon_element64(left, a->vm, 0, MO_64); tcg_gen_extract2_i64(dest, right, left, a->imm * 8); - neon_store_reg64(dest, a->vd); + write_neon_element64(dest, a->vd, 0, MO_64); tcg_temp_free_i64(left); tcg_temp_free_i64(right); @@ -XXX,XX +XXX,XX @@ static bool trans_VEXT(DisasContext *s, arg_VEXT *a) destright = tcg_temp_new_i64(); if (a->imm < 8) { - neon_load_reg64(right, a->vn); - neon_load_reg64(middle, a->vn + 1); + read_neon_element64(right, a->vn, 0, MO_64); + read_neon_element64(middle, a->vn, 1, MO_64); tcg_gen_extract2_i64(destright, right, middle, a->imm * 8); - neon_load_reg64(left, a->vm); + read_neon_element64(left, a->vm, 0, MO_64); tcg_gen_extract2_i64(destleft, middle, left, a->imm * 8); } else { - neon_load_reg64(right, a->vn + 1); - neon_load_reg64(middle, a->vm); + read_neon_element64(right, a->vn, 1, MO_64); + read_neon_element64(middle, a->vm, 0, MO_64); tcg_gen_extract2_i64(destright, right, middle, (a->imm - 8) * 8); - neon_load_reg64(left, a->vm + 1); + read_neon_element64(left, a->vm, 1, MO_64); tcg_gen_extract2_i64(destleft, middle, left, (a->imm - 8) * 8); } - neon_store_reg64(destright, a->vd); - neon_store_reg64(destleft, a->vd + 1); + write_neon_element64(destright, a->vd, 0, MO_64); + write_neon_element64(destleft, a->vd, 1, MO_64); tcg_temp_free_i64(destright); tcg_temp_free_i64(destleft); @@ -XXX,XX +XXX,XX @@ static bool do_2misc_pairwise(DisasContext *s, arg_2misc *a, if (accfn) { TCGv_i64 tmp64 = tcg_temp_new_i64(); - neon_load_reg64(tmp64, a->vd + pass); + read_neon_element64(tmp64, a->vd, pass, MO_64); accfn(rd_64, tmp64, rd_64); tcg_temp_free_i64(tmp64); } - neon_store_reg64(rd_64, a->vd + pass); + write_neon_element64(rd_64, a->vd, pass, MO_64); tcg_temp_free_i64(rd_64); } return true; @@ -XXX,XX +XXX,XX @@ static bool do_vmovn(DisasContext *s, arg_2misc *a, rd0 = tcg_temp_new_i32(); rd1 = tcg_temp_new_i32(); - neon_load_reg64(rm, a->vm); + read_neon_element64(rm, a->vm, 0, MO_64); narrowfn(rd0, cpu_env, rm); - neon_load_reg64(rm, a->vm + 1); + read_neon_element64(rm, a->vm, 1, MO_64); narrowfn(rd1, cpu_env, rm); write_neon_element32(rd0, a->vd, 0, MO_32); write_neon_element32(rd1, a->vd, 1, MO_32); @@ -XXX,XX +XXX,XX @@ static bool trans_VSHLL(DisasContext *s, arg_2misc *a) widenfn(rd, rm0); tcg_gen_shli_i64(rd, rd, 8 << a->size); - neon_store_reg64(rd, a->vd); + write_neon_element64(rd, a->vd, 0, MO_64); widenfn(rd, rm1); tcg_gen_shli_i64(rd, rd, 8 << a->size); - neon_store_reg64(rd, a->vd + 1); + write_neon_element64(rd, a->vd, 1, MO_64); tcg_temp_free_i64(rd); tcg_temp_free_i32(rm0); @@ -XXX,XX +XXX,XX @@ static bool trans_VSWP(DisasContext *s, arg_2misc *a) rm = tcg_temp_new_i64(); rd = tcg_temp_new_i64(); for (pass = 0; pass < (a->q ? 2 : 1); pass++) { - neon_load_reg64(rm, a->vm + pass); - neon_load_reg64(rd, a->vd + pass); - neon_store_reg64(rm, a->vd + pass); - neon_store_reg64(rd, a->vm + pass); + read_neon_element64(rm, a->vm, pass, MO_64); + read_neon_element64(rd, a->vd, pass, MO_64); + write_neon_element64(rm, a->vd, pass, MO_64); + write_neon_element64(rd, a->vm, pass, MO_64); } tcg_temp_free_i64(rm); tcg_temp_free_i64(rd); -- 2.20.1 From: Richard Henderson <richard.henderson@linaro.org> The only uses of this function are for loading VFP double-precision values, and nothing to do with NEON. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20201030022618.785675-10-richard.henderson@linaro.org Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/translate.c | 8 ++-- target/arm/translate-vfp.c.inc | 84 +++++++++++++++++----------------- 2 files changed, 46 insertions(+), 46 deletions(-) diff --git a/target/arm/translate.c b/target/arm/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate.c +++ b/target/arm/translate.c @@ -XXX,XX +XXX,XX @@ static long vfp_reg_offset(bool dp, unsigned reg) } } -static inline void neon_load_reg64(TCGv_i64 var, int reg) +static inline void vfp_load_reg64(TCGv_i64 var, int reg) { - tcg_gen_ld_i64(var, cpu_env, vfp_reg_offset(1, reg)); + tcg_gen_ld_i64(var, cpu_env, vfp_reg_offset(true, reg)); } -static inline void neon_store_reg64(TCGv_i64 var, int reg) +static inline void vfp_store_reg64(TCGv_i64 var, int reg) { - tcg_gen_st_i64(var, cpu_env, vfp_reg_offset(1, reg)); + tcg_gen_st_i64(var, cpu_env, vfp_reg_offset(true, reg)); } static inline void vfp_load_reg32(TCGv_i32 var, int reg) diff --git a/target/arm/translate-vfp.c.inc b/target/arm/translate-vfp.c.inc index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-vfp.c.inc +++ b/target/arm/translate-vfp.c.inc @@ -XXX,XX +XXX,XX @@ static bool trans_VSEL(DisasContext *s, arg_VSEL *a) tcg_gen_ext_i32_i64(nf, cpu_NF); tcg_gen_ext_i32_i64(vf, cpu_VF); - neon_load_reg64(frn, rn); - neon_load_reg64(frm, rm); + vfp_load_reg64(frn, rn); + vfp_load_reg64(frm, rm); switch (a->cc) { case 0: /* eq: Z */ tcg_gen_movcond_i64(TCG_COND_EQ, dest, zf, zero, @@ -XXX,XX +XXX,XX @@ static bool trans_VSEL(DisasContext *s, arg_VSEL *a) tcg_temp_free_i64(tmp); break; } - neon_store_reg64(dest, rd); + vfp_store_reg64(dest, rd); tcg_temp_free_i64(frn); tcg_temp_free_i64(frm); tcg_temp_free_i64(dest); @@ -XXX,XX +XXX,XX @@ static bool trans_VRINT(DisasContext *s, arg_VRINT *a) TCGv_i64 tcg_res; tcg_op = tcg_temp_new_i64(); tcg_res = tcg_temp_new_i64(); - neon_load_reg64(tcg_op, rm); + vfp_load_reg64(tcg_op, rm); gen_helper_rintd(tcg_res, tcg_op, fpst); - neon_store_reg64(tcg_res, rd); + vfp_store_reg64(tcg_res, rd); tcg_temp_free_i64(tcg_op); tcg_temp_free_i64(tcg_res); } else { @@ -XXX,XX +XXX,XX @@ static bool trans_VCVT(DisasContext *s, arg_VCVT *a) tcg_double = tcg_temp_new_i64(); tcg_res = tcg_temp_new_i64(); tcg_tmp = tcg_temp_new_i32(); - neon_load_reg64(tcg_double, rm); + vfp_load_reg64(tcg_double, rm); if (is_signed) { gen_helper_vfp_tosld(tcg_res, tcg_double, tcg_shift, fpst); } else { @@ -XXX,XX +XXX,XX @@ static bool trans_VLDR_VSTR_dp(DisasContext *s, arg_VLDR_VSTR_dp *a) tmp = tcg_temp_new_i64(); if (a->l) { gen_aa32_ld64(s, tmp, addr, get_mem_index(s)); - neon_store_reg64(tmp, a->vd); + vfp_store_reg64(tmp, a->vd); } else { - neon_load_reg64(tmp, a->vd); + vfp_load_reg64(tmp, a->vd); gen_aa32_st64(s, tmp, addr, get_mem_index(s)); } tcg_temp_free_i64(tmp); @@ -XXX,XX +XXX,XX @@ static bool trans_VLDM_VSTM_dp(DisasContext *s, arg_VLDM_VSTM_dp *a) if (a->l) { /* load */ gen_aa32_ld64(s, tmp, addr, get_mem_index(s)); - neon_store_reg64(tmp, a->vd + i); + vfp_store_reg64(tmp, a->vd + i); } else { /* store */ - neon_load_reg64(tmp, a->vd + i); + vfp_load_reg64(tmp, a->vd + i); gen_aa32_st64(s, tmp, addr, get_mem_index(s)); } tcg_gen_addi_i32(addr, addr, offset); @@ -XXX,XX +XXX,XX @@ static bool do_vfp_3op_dp(DisasContext *s, VFPGen3OpDPFn *fn, fd = tcg_temp_new_i64(); fpst = fpstatus_ptr(FPST_FPCR); - neon_load_reg64(f0, vn); - neon_load_reg64(f1, vm); + vfp_load_reg64(f0, vn); + vfp_load_reg64(f1, vm); for (;;) { if (reads_vd) { - neon_load_reg64(fd, vd); + vfp_load_reg64(fd, vd); } fn(fd, f0, f1, fpst); - neon_store_reg64(fd, vd); + vfp_store_reg64(fd, vd); if (veclen == 0) { break; @@ -XXX,XX +XXX,XX @@ static bool do_vfp_3op_dp(DisasContext *s, VFPGen3OpDPFn *fn, veclen--; vd = vfp_advance_dreg(vd, delta_d); vn = vfp_advance_dreg(vn, delta_d); - neon_load_reg64(f0, vn); + vfp_load_reg64(f0, vn); if (delta_m) { vm = vfp_advance_dreg(vm, delta_m); - neon_load_reg64(f1, vm); + vfp_load_reg64(f1, vm); } } @@ -XXX,XX +XXX,XX @@ static bool do_vfp_2op_dp(DisasContext *s, VFPGen2OpDPFn *fn, int vd, int vm) f0 = tcg_temp_new_i64(); fd = tcg_temp_new_i64(); - neon_load_reg64(f0, vm); + vfp_load_reg64(f0, vm); for (;;) { fn(fd, f0); - neon_store_reg64(fd, vd); + vfp_store_reg64(fd, vd); if (veclen == 0) { break; @@ -XXX,XX +XXX,XX @@ static bool do_vfp_2op_dp(DisasContext *s, VFPGen2OpDPFn *fn, int vd, int vm) /* single source one-many */ while (veclen--) { vd = vfp_advance_dreg(vd, delta_d); - neon_store_reg64(fd, vd); + vfp_store_reg64(fd, vd); } break; } @@ -XXX,XX +XXX,XX @@ static bool do_vfp_2op_dp(DisasContext *s, VFPGen2OpDPFn *fn, int vd, int vm) veclen--; vd = vfp_advance_dreg(vd, delta_d); vd = vfp_advance_dreg(vm, delta_m); - neon_load_reg64(f0, vm); + vfp_load_reg64(f0, vm); } tcg_temp_free_i64(f0); @@ -XXX,XX +XXX,XX @@ static bool do_vfm_dp(DisasContext *s, arg_VFMA_dp *a, bool neg_n, bool neg_d) vm = tcg_temp_new_i64(); vd = tcg_temp_new_i64(); - neon_load_reg64(vn, a->vn); - neon_load_reg64(vm, a->vm); + vfp_load_reg64(vn, a->vn); + vfp_load_reg64(vm, a->vm); if (neg_n) { /* VFNMS, VFMS */ gen_helper_vfp_negd(vn, vn); } - neon_load_reg64(vd, a->vd); + vfp_load_reg64(vd, a->vd); if (neg_d) { /* VFNMA, VFNMS */ gen_helper_vfp_negd(vd, vd); } fpst = fpstatus_ptr(FPST_FPCR); gen_helper_vfp_muladdd(vd, vn, vm, vd, fpst); - neon_store_reg64(vd, a->vd); + vfp_store_reg64(vd, a->vd); tcg_temp_free_ptr(fpst); tcg_temp_free_i64(vn); @@ -XXX,XX +XXX,XX @@ static bool trans_VMOV_imm_dp(DisasContext *s, arg_VMOV_imm_dp *a) fd = tcg_const_i64(vfp_expand_imm(MO_64, a->imm)); for (;;) { - neon_store_reg64(fd, vd); + vfp_store_reg64(fd, vd); if (veclen == 0) { break; @@ -XXX,XX +XXX,XX @@ static bool trans_VCMP_dp(DisasContext *s, arg_VCMP_dp *a) vd = tcg_temp_new_i64(); vm = tcg_temp_new_i64(); - neon_load_reg64(vd, a->vd); + vfp_load_reg64(vd, a->vd); if (a->z) { tcg_gen_movi_i64(vm, 0); } else { - neon_load_reg64(vm, a->vm); + vfp_load_reg64(vm, a->vm); } if (a->e) { @@ -XXX,XX +XXX,XX @@ static bool trans_VCVT_f64_f16(DisasContext *s, arg_VCVT_f64_f16 *a) tcg_gen_ld16u_i32(tmp, cpu_env, vfp_f16_offset(a->vm, a->t)); vd = tcg_temp_new_i64(); gen_helper_vfp_fcvt_f16_to_f64(vd, tmp, fpst, ahp_mode); - neon_store_reg64(vd, a->vd); + vfp_store_reg64(vd, a->vd); tcg_temp_free_i32(ahp_mode); tcg_temp_free_ptr(fpst); tcg_temp_free_i32(tmp); @@ -XXX,XX +XXX,XX @@ static bool trans_VCVT_f16_f64(DisasContext *s, arg_VCVT_f16_f64 *a) tmp = tcg_temp_new_i32(); vm = tcg_temp_new_i64(); - neon_load_reg64(vm, a->vm); + vfp_load_reg64(vm, a->vm); gen_helper_vfp_fcvt_f64_to_f16(tmp, vm, fpst, ahp_mode); tcg_temp_free_i64(vm); tcg_gen_st16_i32(tmp, cpu_env, vfp_f16_offset(a->vd, a->t)); @@ -XXX,XX +XXX,XX @@ static bool trans_VRINTR_dp(DisasContext *s, arg_VRINTR_dp *a) } tmp = tcg_temp_new_i64(); - neon_load_reg64(tmp, a->vm); + vfp_load_reg64(tmp, a->vm); fpst = fpstatus_ptr(FPST_FPCR); gen_helper_rintd(tmp, tmp, fpst); - neon_store_reg64(tmp, a->vd); + vfp_store_reg64(tmp, a->vd); tcg_temp_free_ptr(fpst); tcg_temp_free_i64(tmp); return true; @@ -XXX,XX +XXX,XX @@ static bool trans_VRINTZ_dp(DisasContext *s, arg_VRINTZ_dp *a) } tmp = tcg_temp_new_i64(); - neon_load_reg64(tmp, a->vm); + vfp_load_reg64(tmp, a->vm); fpst = fpstatus_ptr(FPST_FPCR); tcg_rmode = tcg_const_i32(float_round_to_zero); gen_helper_set_rmode(tcg_rmode, tcg_rmode, fpst); gen_helper_rintd(tmp, tmp, fpst); gen_helper_set_rmode(tcg_rmode, tcg_rmode, fpst); - neon_store_reg64(tmp, a->vd); + vfp_store_reg64(tmp, a->vd); tcg_temp_free_ptr(fpst); tcg_temp_free_i64(tmp); tcg_temp_free_i32(tcg_rmode); @@ -XXX,XX +XXX,XX @@ static bool trans_VRINTX_dp(DisasContext *s, arg_VRINTX_dp *a) } tmp = tcg_temp_new_i64(); - neon_load_reg64(tmp, a->vm); + vfp_load_reg64(tmp, a->vm); fpst = fpstatus_ptr(FPST_FPCR); gen_helper_rintd_exact(tmp, tmp, fpst); - neon_store_reg64(tmp, a->vd); + vfp_store_reg64(tmp, a->vd); tcg_temp_free_ptr(fpst); tcg_temp_free_i64(tmp); return true; @@ -XXX,XX +XXX,XX @@ static bool trans_VCVT_sp(DisasContext *s, arg_VCVT_sp *a) vd = tcg_temp_new_i64(); vfp_load_reg32(vm, a->vm); gen_helper_vfp_fcvtds(vd, vm, cpu_env); - neon_store_reg64(vd, a->vd); + vfp_store_reg64(vd, a->vd); tcg_temp_free_i32(vm); tcg_temp_free_i64(vd); return true; @@ -XXX,XX +XXX,XX @@ static bool trans_VCVT_dp(DisasContext *s, arg_VCVT_dp *a) vd = tcg_temp_new_i32(); vm = tcg_temp_new_i64(); - neon_load_reg64(vm, a->vm); + vfp_load_reg64(vm, a->vm); gen_helper_vfp_fcvtsd(vd, vm, cpu_env); vfp_store_reg32(vd, a->vd); tcg_temp_free_i32(vd); @@ -XXX,XX +XXX,XX @@ static bool trans_VCVT_int_dp(DisasContext *s, arg_VCVT_int_dp *a) /* u32 -> f64 */ gen_helper_vfp_uitod(vd, vm, fpst); } - neon_store_reg64(vd, a->vd); + vfp_store_reg64(vd, a->vd); tcg_temp_free_i32(vm); tcg_temp_free_i64(vd); tcg_temp_free_ptr(fpst); @@ -XXX,XX +XXX,XX @@ static bool trans_VJCVT(DisasContext *s, arg_VJCVT *a) vm = tcg_temp_new_i64(); vd = tcg_temp_new_i32(); - neon_load_reg64(vm, a->vm); + vfp_load_reg64(vm, a->vm); gen_helper_vjcvt(vd, vm, cpu_env); vfp_store_reg32(vd, a->vd); tcg_temp_free_i64(vm); @@ -XXX,XX +XXX,XX @@ static bool trans_VCVT_fix_dp(DisasContext *s, arg_VCVT_fix_dp *a) frac_bits = (a->opc & 1) ? (32 - a->imm) : (16 - a->imm); vd = tcg_temp_new_i64(); - neon_load_reg64(vd, a->vd); + vfp_load_reg64(vd, a->vd); fpst = fpstatus_ptr(FPST_FPCR); shift = tcg_const_i32(frac_bits); @@ -XXX,XX +XXX,XX @@ static bool trans_VCVT_fix_dp(DisasContext *s, arg_VCVT_fix_dp *a) g_assert_not_reached(); } - neon_store_reg64(vd, a->vd); + vfp_store_reg64(vd, a->vd); tcg_temp_free_i64(vd); tcg_temp_free_i32(shift); tcg_temp_free_ptr(fpst); @@ -XXX,XX +XXX,XX @@ static bool trans_VCVT_dp_int(DisasContext *s, arg_VCVT_dp_int *a) fpst = fpstatus_ptr(FPST_FPCR); vm = tcg_temp_new_i64(); vd = tcg_temp_new_i32(); - neon_load_reg64(vm, a->vm); + vfp_load_reg64(vm, a->vm); if (a->s) { if (a->rz) { -- 2.20.1 From: Richard Henderson <richard.henderson@linaro.org> In both cases, we can sink the write-back and perform the accumulate into the normal destination temps. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20201030022618.785675-11-richard.henderson@linaro.org Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/translate-neon.c.inc | 23 +++++++++-------------- 1 file changed, 9 insertions(+), 14 deletions(-) diff --git a/target/arm/translate-neon.c.inc b/target/arm/translate-neon.c.inc index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-neon.c.inc +++ b/target/arm/translate-neon.c.inc @@ -XXX,XX +XXX,XX @@ static bool do_long_3d(DisasContext *s, arg_3diff *a, if (accfn) { tmp = tcg_temp_new_i64(); read_neon_element64(tmp, a->vd, 0, MO_64); - accfn(tmp, tmp, rd0); - write_neon_element64(tmp, a->vd, 0, MO_64); + accfn(rd0, tmp, rd0); read_neon_element64(tmp, a->vd, 1, MO_64); - accfn(tmp, tmp, rd1); - write_neon_element64(tmp, a->vd, 1, MO_64); + accfn(rd1, tmp, rd1); tcg_temp_free_i64(tmp); - } else { - write_neon_element64(rd0, a->vd, 0, MO_64); - write_neon_element64(rd1, a->vd, 1, MO_64); } + write_neon_element64(rd0, a->vd, 0, MO_64); + write_neon_element64(rd1, a->vd, 1, MO_64); tcg_temp_free_i64(rd0); tcg_temp_free_i64(rd1); @@ -XXX,XX +XXX,XX @@ static bool do_2scalar_long(DisasContext *s, arg_2scalar *a, if (accfn) { TCGv_i64 t64 = tcg_temp_new_i64(); read_neon_element64(t64, a->vd, 0, MO_64); - accfn(t64, t64, rn0_64); - write_neon_element64(t64, a->vd, 0, MO_64); + accfn(rn0_64, t64, rn0_64); read_neon_element64(t64, a->vd, 1, MO_64); - accfn(t64, t64, rn1_64); - write_neon_element64(t64, a->vd, 1, MO_64); + accfn(rn1_64, t64, rn1_64); tcg_temp_free_i64(t64); - } else { - write_neon_element64(rn0_64, a->vd, 0, MO_64); - write_neon_element64(rn1_64, a->vd, 1, MO_64); } + + write_neon_element64(rn0_64, a->vd, 0, MO_64); + write_neon_element64(rn1_64, a->vd, 1, MO_64); tcg_temp_free_i64(rn0_64); tcg_temp_free_i64(rn1_64); return true; -- 2.20.1 From: Richard Henderson <richard.henderson@linaro.org> We can use proper widening loads to extend 32-bit inputs, and skip the "widenfn" step. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20201030022618.785675-12-richard.henderson@linaro.org Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/translate.c | 6 +++ target/arm/translate-neon.c.inc | 66 ++++++++++++++++++--------------- 2 files changed, 43 insertions(+), 29 deletions(-) diff --git a/target/arm/translate.c b/target/arm/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate.c +++ b/target/arm/translate.c @@ -XXX,XX +XXX,XX @@ static void read_neon_element64(TCGv_i64 dest, int reg, int ele, MemOp memop) long off = neon_element_offset(reg, ele, memop); switch (memop) { + case MO_SL: + tcg_gen_ld32s_i64(dest, cpu_env, off); + break; + case MO_UL: + tcg_gen_ld32u_i64(dest, cpu_env, off); + break; case MO_Q: tcg_gen_ld_i64(dest, cpu_env, off); break; diff --git a/target/arm/translate-neon.c.inc b/target/arm/translate-neon.c.inc index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-neon.c.inc +++ b/target/arm/translate-neon.c.inc @@ -XXX,XX +XXX,XX @@ static bool trans_Vimm_1r(DisasContext *s, arg_1reg_imm *a) static bool do_prewiden_3d(DisasContext *s, arg_3diff *a, NeonGenWidenFn *widenfn, NeonGenTwo64OpFn *opfn, - bool src1_wide) + int src1_mop, int src2_mop) { /* 3-regs different lengths, prewidening case (VADDL/VSUBL/VAADW/VSUBW) */ TCGv_i64 rn0_64, rn1_64, rm_64; - TCGv_i32 rm; if (!arm_dc_feature(s, ARM_FEATURE_NEON)) { return false; @@ -XXX,XX +XXX,XX @@ static bool do_prewiden_3d(DisasContext *s, arg_3diff *a, return false; } - if (!widenfn || !opfn) { + if (!opfn) { /* size == 3 case, which is an entirely different insn group */ return false; } - if ((a->vd & 1) || (src1_wide && (a->vn & 1))) { + if ((a->vd & 1) || (src1_mop == MO_Q && (a->vn & 1))) { return false; } @@ -XXX,XX +XXX,XX @@ static bool do_prewiden_3d(DisasContext *s, arg_3diff *a, rn1_64 = tcg_temp_new_i64(); rm_64 = tcg_temp_new_i64(); - if (src1_wide) { - read_neon_element64(rn0_64, a->vn, 0, MO_64); + if (src1_mop >= 0) { + read_neon_element64(rn0_64, a->vn, 0, src1_mop); } else { TCGv_i32 tmp = tcg_temp_new_i32(); read_neon_element32(tmp, a->vn, 0, MO_32); widenfn(rn0_64, tmp); tcg_temp_free_i32(tmp); } - rm = tcg_temp_new_i32(); - read_neon_element32(rm, a->vm, 0, MO_32); + if (src2_mop >= 0) { + read_neon_element64(rm_64, a->vm, 0, src2_mop); + } else { + TCGv_i32 tmp = tcg_temp_new_i32(); + read_neon_element32(tmp, a->vm, 0, MO_32); + widenfn(rm_64, tmp); + tcg_temp_free_i32(tmp); + } - widenfn(rm_64, rm); - tcg_temp_free_i32(rm); opfn(rn0_64, rn0_64, rm_64); /* * Load second pass inputs before storing the first pass result, to * avoid incorrect results if a narrow input overlaps with the result. */ - if (src1_wide) { - read_neon_element64(rn1_64, a->vn, 1, MO_64); + if (src1_mop >= 0) { + read_neon_element64(rn1_64, a->vn, 1, src1_mop); } else { TCGv_i32 tmp = tcg_temp_new_i32(); read_neon_element32(tmp, a->vn, 1, MO_32); widenfn(rn1_64, tmp); tcg_temp_free_i32(tmp); } - rm = tcg_temp_new_i32(); - read_neon_element32(rm, a->vm, 1, MO_32); + if (src2_mop >= 0) { + read_neon_element64(rm_64, a->vm, 1, src2_mop); + } else { + TCGv_i32 tmp = tcg_temp_new_i32(); + read_neon_element32(tmp, a->vm, 1, MO_32); + widenfn(rm_64, tmp); + tcg_temp_free_i32(tmp); + } write_neon_element64(rn0_64, a->vd, 0, MO_64); - widenfn(rm_64, rm); - tcg_temp_free_i32(rm); opfn(rn1_64, rn1_64, rm_64); write_neon_element64(rn1_64, a->vd, 1, MO_64); @@ -XXX,XX +XXX,XX @@ static bool do_prewiden_3d(DisasContext *s, arg_3diff *a, return true; } -#define DO_PREWIDEN(INSN, S, EXT, OP, SRC1WIDE) \ +#define DO_PREWIDEN(INSN, S, OP, SRC1WIDE, SIGN) \ static bool trans_##INSN##_3d(DisasContext *s, arg_3diff *a) \ { \ static NeonGenWidenFn * const widenfn[] = { \ gen_helper_neon_widen_##S##8, \ gen_helper_neon_widen_##S##16, \ - tcg_gen_##EXT##_i32_i64, \ - NULL, \ + NULL, NULL, \ }; \ static NeonGenTwo64OpFn * const addfn[] = { \ gen_helper_neon_##OP##l_u16, \ @@ -XXX,XX +XXX,XX @@ static bool do_prewiden_3d(DisasContext *s, arg_3diff *a, tcg_gen_##OP##_i64, \ NULL, \ }; \ - return do_prewiden_3d(s, a, widenfn[a->size], \ - addfn[a->size], SRC1WIDE); \ + int narrow_mop = a->size == MO_32 ? MO_32 | SIGN : -1; \ + return do_prewiden_3d(s, a, widenfn[a->size], addfn[a->size], \ + SRC1WIDE ? MO_Q : narrow_mop, \ + narrow_mop); \ } -DO_PREWIDEN(VADDL_S, s, ext, add, false) -DO_PREWIDEN(VADDL_U, u, extu, add, false) -DO_PREWIDEN(VSUBL_S, s, ext, sub, false) -DO_PREWIDEN(VSUBL_U, u, extu, sub, false) -DO_PREWIDEN(VADDW_S, s, ext, add, true) -DO_PREWIDEN(VADDW_U, u, extu, add, true) -DO_PREWIDEN(VSUBW_S, s, ext, sub, true) -DO_PREWIDEN(VSUBW_U, u, extu, sub, true) +DO_PREWIDEN(VADDL_S, s, add, false, MO_SIGN) +DO_PREWIDEN(VADDL_U, u, add, false, 0) +DO_PREWIDEN(VSUBL_S, s, sub, false, MO_SIGN) +DO_PREWIDEN(VSUBL_U, u, sub, false, 0) +DO_PREWIDEN(VADDW_S, s, add, true, MO_SIGN) +DO_PREWIDEN(VADDW_U, u, add, true, 0) +DO_PREWIDEN(VSUBW_S, s, sub, true, MO_SIGN) +DO_PREWIDEN(VSUBW_U, u, sub, true, 0) static bool do_narrow_3d(DisasContext *s, arg_3diff *a, NeonGenTwo64OpFn *opfn, NeonGenNarrowFn *narrowfn) -- 2.20.1 In the neon_padd/pmax/pmin helpers for float16, a cut-and-paste error meant we were using the H4() address swizzler macro rather than the H2() which is required for 2-byte data. This had no effect on little-endian hosts but meant we put the result data into the destination Dreg in the wrong order on big-endian hosts. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org> Message-id: 20201028191712.4910-2-peter.maydell@linaro.org --- target/arm/vec_helper.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/target/arm/vec_helper.c b/target/arm/vec_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/vec_helper.c +++ b/target/arm/vec_helper.c @@ -XXX,XX +XXX,XX @@ DO_ABA(gvec_uaba_d, uint64_t) r2 = float16_##OP(m[H2(0)], m[H2(1)], fpst); \ r3 = float16_##OP(m[H2(2)], m[H2(3)], fpst); \ \ - d[H4(0)] = r0; \ - d[H4(1)] = r1; \ - d[H4(2)] = r2; \ - d[H4(3)] = r3; \ + d[H2(0)] = r0; \ + d[H2(1)] = r1; \ + d[H2(2)] = r2; \ + d[H2(3)] = r3; \ } DO_NEON_PAIRWISE(neon_padd, add) -- 2.20.1 The helper functions for performing the udot/sdot operations against a scalar were not using an address-swizzling macro when converting the index of the scalar element into a pointer into the vm array. This had no effect on little-endian hosts but meant we generated incorrect results on big-endian hosts. For these insns, the index is indexing over group of 4 8-bit values, so 32 bits per indexed entity, and H4() is therefore what we want. (For Neon the only possible input indexes are 0 and 1.) Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org> Message-id: 20201028191712.4910-3-peter.maydell@linaro.org --- target/arm/vec_helper.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/target/arm/vec_helper.c b/target/arm/vec_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/vec_helper.c +++ b/target/arm/vec_helper.c @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_sdot_idx_b)(void *vd, void *vn, void *vm, uint32_t desc) intptr_t index = simd_data(desc); uint32_t *d = vd; int8_t *n = vn; - int8_t *m_indexed = (int8_t *)vm + index * 4; + int8_t *m_indexed = (int8_t *)vm + H4(index) * 4; /* Notice the special case of opr_sz == 8, from aa64/aa32 advsimd. * Otherwise opr_sz is a multiple of 16. @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_udot_idx_b)(void *vd, void *vn, void *vm, uint32_t desc) intptr_t index = simd_data(desc); uint32_t *d = vd; uint8_t *n = vn; - uint8_t *m_indexed = (uint8_t *)vm + index * 4; + uint8_t *m_indexed = (uint8_t *)vm + H4(index) * 4; /* Notice the special case of opr_sz == 8, from aa64/aa32 advsimd. * Otherwise opr_sz is a multiple of 16. -- 2.20.1 From: Rémi Denis-Courmont <remi.denis.courmont@huawei.com> HCR should be applied when NS is set, not when it is cleared. Signed-off-by: Rémi Denis-Courmont <remi.denis.courmont@huawei.com> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/helper.c | 5 ++--- 1 file changed, 2 insertions(+), 3 deletions(-) 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 tlbimvaa_is_write(CPUARMState *env, const ARMCPRegInfo *ri, /* * Non-IS variants of TLB operations are upgraded to - * IS versions if we are at NS EL1 and HCR_EL2.FB is set to + * IS versions if we are at EL1 and HCR_EL2.FB is effectively set to * force broadcast of these operations. */ static bool tlb_force_broadcast(CPUARMState *env) { - return (env->cp15.hcr_el2 & HCR_FB) && - arm_current_el(env) == 1 && arm_is_secure_below_el3(env); + return arm_current_el(env) == 1 && (arm_hcr_el2_eff(env) & HCR_FB); } static void tlbiall_write(CPUARMState *env, const ARMCPRegInfo *ri, -- 2.20.1 From: Rémi Denis-Courmont <remi.denis.courmont@huawei.com> Secure mode is not exempted from checking SCR_EL3.TLOR, and in the future HCR_EL2.TLOR when S-EL2 is enabled. Signed-off-by: Rémi Denis-Courmont <remi.denis.courmont@huawei.com> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/helper.c | 19 +++++-------------- 1 file changed, 5 insertions(+), 14 deletions(-) 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 id_aa64pfr0_read(CPUARMState *env, const ARMCPRegInfo *ri) #endif /* Shared logic between LORID and the rest of the LOR* registers. - * Secure state has already been delt with. + * Secure state exclusion has already been dealt with. */ -static CPAccessResult access_lor_ns(CPUARMState *env) +static CPAccessResult access_lor_ns(CPUARMState *env, + const ARMCPRegInfo *ri, bool isread) { int el = arm_current_el(env); @@ -XXX,XX +XXX,XX @@ static CPAccessResult access_lor_ns(CPUARMState *env) return CP_ACCESS_OK; } -static CPAccessResult access_lorid(CPUARMState *env, const ARMCPRegInfo *ri, - bool isread) -{ - if (arm_is_secure_below_el3(env)) { - /* Access ok in secure mode. */ - return CP_ACCESS_OK; - } - return access_lor_ns(env); -} - static CPAccessResult access_lor_other(CPUARMState *env, const ARMCPRegInfo *ri, bool isread) { @@ -XXX,XX +XXX,XX @@ static CPAccessResult access_lor_other(CPUARMState *env, /* Access denied in secure mode. */ return CP_ACCESS_TRAP; } - return access_lor_ns(env); + return access_lor_ns(env, ri, isread); } /* @@ -XXX,XX +XXX,XX @@ static const ARMCPRegInfo lor_reginfo[] = { .type = ARM_CP_CONST, .resetvalue = 0 }, { .name = "LORID_EL1", .state = ARM_CP_STATE_AA64, .opc0 = 3, .opc1 = 0, .crn = 10, .crm = 4, .opc2 = 7, - .access = PL1_R, .accessfn = access_lorid, + .access = PL1_R, .accessfn = access_lor_ns, .type = ARM_CP_CONST, .resetvalue = 0 }, REGINFO_SENTINEL }; -- 2.20.1 If we're using the capstone disassembler, disassembly of a run of instructions more than 32 bytes long disassembles the wrong data for instructions beyond the 32 byte mark: (qemu) xp /16x 0x100 0000000000000100: 0x00000005 0x54410001 0x00000001 0x00001000 0000000000000110: 0x00000000 0x00000004 0x54410002 0x3c000000 0000000000000120: 0x00000000 0x00000004 0x54410009 0x74736574 0000000000000130: 0x00000000 0x00000000 0x00000000 0x00000000 (qemu) xp /16i 0x100 0x00000100: 00000005 andeq r0, r0, r5 0x00000104: 54410001 strbpl r0, [r1], #-1 0x00000108: 00000001 andeq r0, r0, r1 0x0000010c: 00001000 andeq r1, r0, r0 0x00000110: 00000000 andeq r0, r0, r0 0x00000114: 00000004 andeq r0, r0, r4 0x00000118: 54410002 strbpl r0, [r1], #-2 0x0000011c: 3c000000 .byte 0x00, 0x00, 0x00, 0x3c 0x00000120: 54410001 strbpl r0, [r1], #-1 0x00000124: 00000001 andeq r0, r0, r1 0x00000128: 00001000 andeq r1, r0, r0 0x0000012c: 00000000 andeq r0, r0, r0 0x00000130: 00000004 andeq r0, r0, r4 0x00000134: 54410002 strbpl r0, [r1], #-2 0x00000138: 3c000000 .byte 0x00, 0x00, 0x00, 0x3c 0x0000013c: 00000000 andeq r0, r0, r0 Here the disassembly of 0x120..0x13f is using the data that is in 0x104..0x123. This is caused by passing the wrong value to the read_memory_func(). The intention is that at this point in the loop the 'cap_buf' buffer already contains 'csize' bytes of data for the instruction at guest addr 'pc', and we want to read in an extra 'tsize' bytes. Those extra bytes are therefore at 'pc + csize', not 'pc'. On the first time through the loop 'csize' happens to be zero, so the initial read of 32 bytes into cap_buf is correct and as long as the disassembly never needs to read more data we return the correct information. Use the correct guest address in the call to read_memory_func(). Cc: qemu-stable@nongnu.org Fixes: https://bugs.launchpad.net/qemu/+bug/1900779 Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org> Message-id: 20201022132445.25039-1-peter.maydell@linaro.org --- disas/capstone.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/disas/capstone.c b/disas/capstone.c index XXXXXXX..XXXXXXX 100644 --- a/disas/capstone.c +++ b/disas/capstone.c @@ -XXX,XX +XXX,XX @@ bool cap_disas_monitor(disassemble_info *info, uint64_t pc, int count) /* Make certain that we can make progress. */ assert(tsize != 0); - info->read_memory_func(pc, cap_buf + csize, tsize, info); + info->read_memory_func(pc + csize, cap_buf + csize, tsize, info); csize += tsize; if (cs_disasm_iter(handle, &cbuf, &csize, &pc, insn)) { -- 2.20.1 From: Philippe Mathieu-Daudé <philmd@redhat.com> Use the BIT_ULL() macro to ensure we use 64-bit arithmetic. This fixes the following Coverity issue (OVERFLOW_BEFORE_WIDEN): CID 1432363 (#1 of 1): Unintentional integer overflow: overflow_before_widen: Potentially overflowing expression 1 << scale with type int (32 bits, signed) is evaluated using 32-bit arithmetic, and then used in a context that expects an expression of type hwaddr (64 bits, unsigned). Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com> Acked-by: Eric Auger <eric.auger@redhat.com> Message-id: 20201030144617.1535064-1-philmd@redhat.com Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/arm/smmuv3.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) 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/osdep.h" +#include "qemu/bitops.h" #include "hw/irq.h" #include "hw/sysbus.h" #include "migration/vmstate.h" @@ -XXX,XX +XXX,XX @@ static void smmuv3_s1_range_inval(SMMUState *s, Cmd *cmd) scale = CMD_SCALE(cmd); num = CMD_NUM(cmd); ttl = CMD_TTL(cmd); - num_pages = (num + 1) * (1 << (scale)); + num_pages = (num + 1) * BIT_ULL(scale); } if (type == SMMU_CMD_TLBI_NH_VA) { -- 2.20.1 From: Rémi Denis-Courmont <remi.denis.courmont@huawei.com> When booting a CPU with EL3 using the -kernel flag, set up CPTR_EL3 so that SVE will not trap to EL3. Signed-off-by: Rémi Denis-Courmont <remi.denis.courmont@huawei.com> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20201030151541.11976-1-remi@remlab.net Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/arm/boot.c | 3 +++ 1 file changed, 3 insertions(+) diff --git a/hw/arm/boot.c b/hw/arm/boot.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/boot.c +++ b/hw/arm/boot.c @@ -XXX,XX +XXX,XX @@ static void do_cpu_reset(void *opaque) if (cpu_isar_feature(aa64_mte, cpu)) { env->cp15.scr_el3 |= SCR_ATA; } + if (cpu_isar_feature(aa64_sve, cpu)) { + env->cp15.cptr_el[3] |= CPTR_EZ; + } /* AArch64 kernels never boot in secure mode */ assert(!info->secure_boot); /* This hook is only supported for AArch32 currently: -- 2.20.1 From: AlexChen <alex.chen@huawei.com> In omap_lcd_interrupts(), the pointer omap_lcd is dereferinced before being check if it is valid, which may lead to NULL pointer dereference. So move the assignment to surface after checking that the omap_lcd is valid and move surface_bits_per_pixel(surface) to after the surface assignment. Reported-by: Euler Robot <euler.robot@huawei.com> Signed-off-by: AlexChen <alex.chen@huawei.com> Message-id: 5F9CDB8A.9000001@huawei.com Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/display/omap_lcdc.c | 10 +++++++--- 1 file changed, 7 insertions(+), 3 deletions(-) diff --git a/hw/display/omap_lcdc.c b/hw/display/omap_lcdc.c index XXXXXXX..XXXXXXX 100644 --- a/hw/display/omap_lcdc.c +++ b/hw/display/omap_lcdc.c @@ -XXX,XX +XXX,XX @@ static void omap_lcd_interrupts(struct omap_lcd_panel_s *s) static void omap_update_display(void *opaque) { struct omap_lcd_panel_s *omap_lcd = (struct omap_lcd_panel_s *) opaque; - DisplaySurface *surface = qemu_console_surface(omap_lcd->con); + DisplaySurface *surface; draw_line_func draw_line; int size, height, first, last; int width, linesize, step, bpp, frame_offset; hwaddr frame_base; - if (!omap_lcd || omap_lcd->plm == 1 || !omap_lcd->enable || - !surface_bits_per_pixel(surface)) { + if (!omap_lcd || omap_lcd->plm == 1 || !omap_lcd->enable) { + return; + } + + surface = qemu_console_surface(omap_lcd->con); + if (!surface_bits_per_pixel(surface)) { return; } -- 2.20.1 From: AlexChen <alex.chen@huawei.com> In exynos4210_fimd_update(), the pointer s is dereferinced before being check if it is valid, which may lead to NULL pointer dereference. So move the assignment to global_width after checking that the s is valid. Reported-by: Euler Robot <euler.robot@huawei.com> Signed-off-by: Alex Chen <alex.chen@huawei.com> Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com> Message-id: 5F9F8D88.9030102@huawei.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/display/exynos4210_fimd.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/hw/display/exynos4210_fimd.c b/hw/display/exynos4210_fimd.c index XXXXXXX..XXXXXXX 100644 --- a/hw/display/exynos4210_fimd.c +++ b/hw/display/exynos4210_fimd.c @@ -XXX,XX +XXX,XX @@ static void exynos4210_fimd_update(void *opaque) bool blend = false; uint8_t *host_fb_addr; bool is_dirty = false; - const int global_width = (s->vidtcon[2] & FIMD_VIDTCON2_SIZE_MASK) + 1; + int global_width; if (!s || !s->console || !s->enabled || surface_bits_per_pixel(qemu_console_surface(s->console)) == 0) { return; } + + global_width = (s->vidtcon[2] & FIMD_VIDTCON2_SIZE_MASK) + 1; exynos4210_update_resolution(s); surface = qemu_console_surface(s->console); -- 2.20.1 In arm_v7m_mmu_idx_for_secstate() we get the 'priv' level to pass to armv7m_mmu_idx_for_secstate_and_priv() by calling arm_current_el(). This is incorrect when the security state being queried is not the current one, because arm_current_el() uses the current security state to determine which of the banked CONTROL.nPRIV bits to look at. The effect was that if (for instance) Secure state was in privileged mode but Non-Secure was not then we would return the wrong MMU index. The only places where we are using this function in a way that could trigger this bug are for the stack loads during a v8M function-return and for the instruction fetch of a v8M SG insn. Fix the bug by expanding out the M-profile version of the arm_current_el() logic inline so it can use the passed in secstate rather than env->v7m.secure. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20201022164408.13214-1-peter.maydell@linaro.org --- target/arm/m_helper.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/target/arm/m_helper.c b/target/arm/m_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/m_helper.c +++ b/target/arm/m_helper.c @@ -XXX,XX +XXX,XX @@ ARMMMUIdx arm_v7m_mmu_idx_for_secstate_and_priv(CPUARMState *env, /* Return the MMU index for a v7M CPU in the specified security state */ ARMMMUIdx arm_v7m_mmu_idx_for_secstate(CPUARMState *env, bool secstate) { - bool priv = arm_current_el(env) != 0; + bool priv = arm_v7m_is_handler_mode(env) || + !(env->v7m.control[secstate] & 1); return arm_v7m_mmu_idx_for_secstate_and_priv(env, secstate, priv); } -- 2.20.1 On some hosts (eg Ubuntu Bionic) pkg-config returns a set of libraries for gio-2.0 which don't actually work when compiling statically. (Specifically, the returned library string includes -lmount, but not -lblkid which -lmount depends upon, so linking fails due to missing symbols.) Check that the libraries work, and don't enable gio if they don't, in the same way we do for gnutls. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Paolo Bonzini <pbonzini@redhat.com> Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com> Message-id: 20200928160402.7961-1-peter.maydell@linaro.org --- configure | 10 +++++++++- 1 file changed, 9 insertions(+), 1 deletion(-) diff --git a/configure b/configure index XXXXXXX..XXXXXXX 100755 --- a/configure +++ b/configure @@ -XXX,XX +XXX,XX @@ if test "$static" = yes && test "$mingw32" = yes; then fi if $pkg_config --atleast-version=$glib_req_ver gio-2.0; then - gio=yes gio_cflags=$($pkg_config --cflags gio-2.0) gio_libs=$($pkg_config --libs gio-2.0) gdbus_codegen=$($pkg_config --variable=gdbus_codegen gio-2.0) if [ ! -x "$gdbus_codegen" ]; then gdbus_codegen= fi + # Check that the libraries actually work -- Ubuntu 18.04 ships + # with pkg-config --static --libs data for gio-2.0 that is missing + # -lblkid and will give a link error. + write_c_skeleton + if compile_prog "" "gio_libs" ; then + gio=yes + else + gio=no + fi else gio=no fi -- 2.20.1 In gicv3_init_cpuif() we copy the ARMCPU gicv3_maintenance_interrupt into the GICv3CPUState struct's maintenance_irq field. This will only work if the board happens to have already wired up the CPU maintenance IRQ before the GIC was realized. Unfortunately this is not the case for the 'virt' board, and so the value that gets copied is NULL (since a qemu_irq is really a pointer to an IRQState struct under the hood). The effect is that the CPU interface code never actually raises the maintenance interrupt line. Instead, since the GICv3CPUState has a pointer to the CPUState, make the dereference at the point where we want to raise the interrupt, to avoid an implicit requirement on board code to wire things up in a particular order. Reported-by: Jose Martins <josemartins90@gmail.com> Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Message-id: 20201009153904.28529-1-peter.maydell@linaro.org Reviewed-by: Luc Michel <luc@lmichel.fr> --- include/hw/intc/arm_gicv3_common.h | 1 - hw/intc/arm_gicv3_cpuif.c | 5 ++--- 2 files changed, 2 insertions(+), 4 deletions(-) diff --git a/include/hw/intc/arm_gicv3_common.h b/include/hw/intc/arm_gicv3_common.h index XXXXXXX..XXXXXXX 100644 --- a/include/hw/intc/arm_gicv3_common.h +++ b/include/hw/intc/arm_gicv3_common.h @@ -XXX,XX +XXX,XX @@ struct GICv3CPUState { qemu_irq parent_fiq; qemu_irq parent_virq; qemu_irq parent_vfiq; - qemu_irq maintenance_irq; /* Redistributor */ uint32_t level; /* Current IRQ level */ diff --git a/hw/intc/arm_gicv3_cpuif.c b/hw/intc/arm_gicv3_cpuif.c index XXXXXXX..XXXXXXX 100644 --- a/hw/intc/arm_gicv3_cpuif.c +++ b/hw/intc/arm_gicv3_cpuif.c @@ -XXX,XX +XXX,XX @@ static void gicv3_cpuif_virt_update(GICv3CPUState *cs) int irqlevel = 0; int fiqlevel = 0; int maintlevel = 0; + ARMCPU *cpu = ARM_CPU(cs->cpu); idx = hppvi_index(cs); trace_gicv3_cpuif_virt_update(gicv3_redist_affid(cs), idx); @@ -XXX,XX +XXX,XX @@ static void gicv3_cpuif_virt_update(GICv3CPUState *cs) qemu_set_irq(cs->parent_vfiq, fiqlevel); qemu_set_irq(cs->parent_virq, irqlevel); - qemu_set_irq(cs->maintenance_irq, maintlevel); + qemu_set_irq(cpu->gicv3_maintenance_interrupt, maintlevel); } static uint64_t icv_ap_read(CPUARMState *env, const ARMCPRegInfo *ri) @@ -XXX,XX +XXX,XX @@ void gicv3_init_cpuif(GICv3State *s) && cpu->gic_num_lrs) { int j; - cs->maintenance_irq = cpu->gicv3_maintenance_interrupt; - cs->num_list_regs = cpu->gic_num_lrs; cs->vpribits = cpu->gic_vpribits; cs->vprebits = cpu->gic_vprebits; -- 2.20.1 The kerneldoc script currently emits Sphinx markup for a macro with arguments that uses the c:function directive. This is correct for Sphinx versions earlier than Sphinx 3, where c:macro doesn't allow documentation of macros with arguments and c:function is not picky about the syntax of what it is passed. However, in Sphinx 3 the c:macro directive was enhanced to support macros with arguments, and c:function was made more picky about what syntax it accepted. When kerneldoc is told that it needs to produce output for Sphinx 3 or later, make it emit c:function only for functions and c:macro for macros with arguments. We assume that anything with a return type is a function and anything without is a macro. This fixes the Sphinx error: /home/petmay01/linaro/qemu-from-laptop/qemu/docs/../include/qom/object.h:155:Error in declarator If declarator-id with parameters (e.g., 'void f(int arg)'): Invalid C declaration: Expected identifier in nested name. [error at 25] DECLARE_INSTANCE_CHECKER ( InstanceType, OBJ_NAME, TYPENAME) -------------------------^ If parenthesis in noptr-declarator (e.g., 'void (*f(int arg))(double)'): Error in declarator or parameters Invalid C declaration: Expecting "(" in parameters. [error at 39] DECLARE_INSTANCE_CHECKER ( InstanceType, OBJ_NAME, TYPENAME) ---------------------------------------^ Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Daniel P. Berrangé <berrange@redhat.com> Tested-by: Stefan Hajnoczi <stefanha@redhat.com> Message-id: 20201030174700.7204-2-peter.maydell@linaro.org --- scripts/kernel-doc | 18 +++++++++++++++++- 1 file changed, 17 insertions(+), 1 deletion(-) diff --git a/scripts/kernel-doc b/scripts/kernel-doc index XXXXXXX..XXXXXXX 100755 --- a/scripts/kernel-doc +++ b/scripts/kernel-doc @@ -XXX,XX +XXX,XX @@ sub output_function_rst(%) { output_highlight_rst($args{'purpose'}); $start = "\n\n**Syntax**\n\n ``"; } else { - print ".. c:function:: "; + if ((split(/\./, $sphinx_version))[0] >= 3) { + # Sphinx 3 and later distinguish macros and functions and + # complain if you use c:function with something that's not + # syntactically valid as a function declaration. + # We assume that anything with a return type is a function + # and anything without is a macro. + if ($args{'functiontype'} ne "") { + print ".. c:function:: "; + } else { + print ".. c:macro:: "; + } + } else { + # Older Sphinx don't support documenting macros that take + # arguments with c:macro, and don't complain about the use + # of c:function for this. + print ".. c:function:: "; + } } if ($args{'functiontype'} ne "") { $start .= $args{'functiontype'} . " " . $args{'function'} . " ("; -- 2.20.1 Sphinx 3.2 is pickier than earlier versions about the option:: markup, and complains about our usage in qemu-option-trace.rst: ../../docs/qemu-option-trace.rst.inc:4:Malformed option description '[enable=]PATTERN', should look like "opt", "-opt args", "--opt args", "/opt args" or "+opt args" In this file, we're really trying to document the different parts of the top-level --trace option, which qemu-nbd.rst and qemu-img.rst have already introduced with an option:: markup. So it's not right to use option:: here anyway. Switch to a different markup (definition lists) which gives about the same formatted output. (Unlike option::, this markup doesn't produce index entries; but at the moment we don't do anything much with indexes anyway, and in any case I think it doesn't make much sense to have individual index entries for the sub-parts of the --trace option.) Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Daniel P. Berrangé <berrange@redhat.com> Tested-by: Stefan Hajnoczi <stefanha@redhat.com> Message-id: 20201030174700.7204-3-peter.maydell@linaro.org --- docs/qemu-option-trace.rst.inc | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/docs/qemu-option-trace.rst.inc b/docs/qemu-option-trace.rst.inc index XXXXXXX..XXXXXXX 100644 --- a/docs/qemu-option-trace.rst.inc +++ b/docs/qemu-option-trace.rst.inc @@ -XXX,XX +XXX,XX @@ Specify tracing options. -.. option:: [enable=]PATTERN +``[enable=]PATTERN`` Immediately enable events matching *PATTERN* (either event name or a globbing pattern). This option is only @@ -XXX,XX +XXX,XX @@ Specify tracing options. Use :option:`-trace help` to print a list of names of trace points. -.. option:: events=FILE +``events=FILE`` Immediately enable events listed in *FILE*. The file must contain one event name (as listed in the ``trace-events-all`` @@ -XXX,XX +XXX,XX @@ Specify tracing options. available if QEMU has been compiled with the ``simple``, ``log`` or ``ftrace`` tracing backend. -.. option:: file=FILE +``file=FILE`` Log output traces to *FILE*. This option is only available if QEMU has been compiled with -- 2.20.1 The randomness tests in the NPCM7xx RNG test fail intermittently but fairly frequently. On my machine running the test in a loop: while QTEST_QEMU_BINARY=./qemu-system-aarch64 ./tests/qtest/npcm7xx_rng-test; do true; done will fail in less than a minute with an error like: ERROR:../../tests/qtest/npcm7xx_rng-test.c:256:test_first_byte_runs: assertion failed (calc_runs_p(buf.l, sizeof(buf) * BITS_PER_BYTE) > 0.01): (0.00286205989 > 0.01) (Failures have been observed on all 4 of the randomness tests, not just first_byte_runs.) It's not clear why these tests are failing like this, but intermittent failures make CI and merge testing awkward, so disable running them unless a developer specifically sets QEMU_TEST_FLAKY_RNG_TESTS when running the test suite, until we work out the cause. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com> Message-id: 20201102152454.8287-1-peter.maydell@linaro.org Reviewed-by: Havard Skinnemoen <hskinnemoen@google.com> --- tests/qtest/npcm7xx_rng-test.c | 14 ++++++++++---- 1 file changed, 10 insertions(+), 4 deletions(-) diff --git a/tests/qtest/npcm7xx_rng-test.c b/tests/qtest/npcm7xx_rng-test.c index XXXXXXX..XXXXXXX 100644 --- a/tests/qtest/npcm7xx_rng-test.c +++ b/tests/qtest/npcm7xx_rng-test.c @@ -XXX,XX +XXX,XX @@ int main(int argc, char **argv) qtest_add_func("npcm7xx_rng/enable_disable", test_enable_disable); qtest_add_func("npcm7xx_rng/rosel", test_rosel); - qtest_add_func("npcm7xx_rng/continuous/monobit", test_continuous_monobit); - qtest_add_func("npcm7xx_rng/continuous/runs", test_continuous_runs); - qtest_add_func("npcm7xx_rng/first_byte/monobit", test_first_byte_monobit); - qtest_add_func("npcm7xx_rng/first_byte/runs", test_first_byte_runs); + /* + * These tests fail intermittently; only run them on explicit + * request until we figure out why. + */ + if (getenv("QEMU_TEST_FLAKY_RNG_TESTS")) { + qtest_add_func("npcm7xx_rng/continuous/monobit", test_continuous_monobit); + qtest_add_func("npcm7xx_rng/continuous/runs", test_continuous_runs); + qtest_add_func("npcm7xx_rng/first_byte/monobit", test_first_byte_monobit); + qtest_add_func("npcm7xx_rng/first_byte/runs", test_first_byte_runs); + } qtest_start("-machine npcm750-evb"); ret = g_test_run(); -- 2.20.1