First set of arm patches for 6.2. I have a lot more in my to-review queue still... -- PMM The following changes since commit d42685765653ec155fdf60910662f8830bdb2cef: Open 6.2 development tree (2021-08-25 10:25:12 +0100) are available in the Git repository at: https://git.linaro.org/people/pmaydell/qemu-arm.git tags/pull-target-arm-20210825 for you to fetch changes up to 24b1a6aa43615be22c7ee66bd68ec5675f6a6a9a: docs: Document how to use gdb with unix sockets (2021-08-25 10:48:51 +0100) ---------------------------------------------------------------- target-arm queue: * More MVE emulation work * Implement M-profile trapping on division by zero * kvm: use RCU_READ_LOCK_GUARD() in kvm_arch_fixup_msi_route() * hw/char/pl011: add support for sending break * fsl-imx6ul: Instantiate SAI1/2/3 and ASRC as unimplemented devices * hw/dma/pl330: Add memory region to replace default * sbsa-ref: Rename SBSA_GWDT enum value * fsl-imx7: Instantiate SAI1/2/3 as unimplemented devices * docs: Document how to use gdb with unix sockets ---------------------------------------------------------------- Eduardo Habkost (1): sbsa-ref: Rename SBSA_GWDT enum value Guenter Roeck (2): fsl-imx6ul: Instantiate SAI1/2/3 and ASRC as unimplemented devices fsl-imx7: Instantiate SAI1/2/3 as unimplemented devices Hamza Mahfooz (1): target/arm: kvm: use RCU_READ_LOCK_GUARD() in kvm_arch_fixup_msi_route() Jan Luebbe (1): hw/char/pl011: add support for sending break Peter Maydell (37): target/arm: Note that we handle VMOVL as a special case of VSHLL target/arm: Print MVE VPR in CPU dumps target/arm: Fix MVE VSLI by 0 and VSRI by <dt> target/arm: Fix signed VADDV target/arm: Fix mask handling for MVE narrowing operations target/arm: Fix 48-bit saturating shifts target/arm: Fix MVE 48-bit SQRSHRL for small right shifts target/arm: Fix calculation of LTP mask when LR is 0 target/arm: Factor out mve_eci_mask() target/arm: Fix VPT advance when ECI is non-zero target/arm: Fix VLDRB/H/W for predicated elements target/arm: Implement MVE VMULL (polynomial) target/arm: Implement MVE incrementing/decrementing dup insns target/arm: Factor out gen_vpst() target/arm: Implement MVE integer vector comparisons target/arm: Implement MVE integer vector-vs-scalar comparisons target/arm: Implement MVE VPSEL target/arm: Implement MVE VMLAS target/arm: Implement MVE shift-by-scalar target/arm: Move 'x' and 'a' bit definitions into vmlaldav formats target/arm: Implement MVE integer min/max across vector target/arm: Implement MVE VABAV target/arm: Implement MVE narrowing moves target/arm: Rename MVEGenDualAccOpFn to MVEGenLongDualAccOpFn target/arm: Implement MVE VMLADAV and VMLSLDAV target/arm: Implement MVE VMLA target/arm: Implement MVE saturating doubling multiply accumulates target/arm: Implement MVE VQABS, VQNEG target/arm: Implement MVE VMAXA, VMINA target/arm: Implement MVE VMOV to/from 2 general-purpose registers target/arm: Implement MVE VPNOT target/arm: Implement MVE VCTP target/arm: Implement MVE scatter-gather insns target/arm: Implement MVE scatter-gather immediate forms target/arm: Implement MVE interleaving loads/stores target/arm: Re-indent sdiv and udiv helpers target/arm: Implement M-profile trapping on division by zero Sebastian Meyer (1): docs: Document how to use gdb with unix sockets Wen, Jianxian (1): hw/dma/pl330: Add memory region to replace default docs/system/gdb.rst | 26 +- include/hw/arm/fsl-imx7.h | 5 + target/arm/cpu.h | 1 + target/arm/helper-mve.h | 283 ++++++++++ target/arm/helper.h | 4 +- target/arm/translate-a32.h | 2 + target/arm/vec_internal.h | 11 + target/arm/mve.decode | 226 +++++++- target/arm/t32.decode | 1 + hw/arm/exynos4210.c | 3 + hw/arm/fsl-imx6ul.c | 12 + hw/arm/fsl-imx7.c | 7 + hw/arm/sbsa-ref.c | 6 +- hw/arm/xilinx_zynq.c | 3 + hw/char/pl011.c | 6 + hw/dma/pl330.c | 26 +- target/arm/cpu.c | 3 + target/arm/helper.c | 34 +- target/arm/kvm.c | 17 +- target/arm/m_helper.c | 4 + target/arm/mve_helper.c | 1254 ++++++++++++++++++++++++++++++++++++++++++-- target/arm/translate-mve.c | 877 ++++++++++++++++++++++++++++++- target/arm/translate-vfp.c | 2 +- target/arm/translate.c | 37 +- target/arm/vec_helper.c | 14 +- 25 files changed, 2746 insertions(+), 118 deletions(-) Although the architecture doesn't define it as an alias, VMOVL (vector move long) is encoded as a VSHLL with a zero shift. Add a comment in the decode file noting that we handle VMOVL as part of VSHLL. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/mve.decode | 2 ++ 1 file changed, 2 insertions(+) diff --git a/target/arm/mve.decode b/target/arm/mve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve.decode +++ b/target/arm/mve.decode @@ -XXX,XX +XXX,XX @@ VRSHRI_U 111 1 1111 1 . ... ... ... 0 0010 0 1 . 1 ... 0 @2_shr_h VRSHRI_U 111 1 1111 1 . ... ... ... 0 0010 0 1 . 1 ... 0 @2_shr_w # VSHLL T1 encoding; the T2 VSHLL encoding is elsewhere in this file +# Note that VMOVL is encoded as "VSHLL with a zero shift count"; we +# implement it that way rather than special-casing it in the decode. VSHLL_BS 111 0 1110 1 . 1 .. ... ... 0 1111 0 1 . 0 ... 0 @2_shll_b VSHLL_BS 111 0 1110 1 . 1 .. ... ... 0 1111 0 1 . 0 ... 0 @2_shll_h -- 2.20.1 Include the MVE VPR register value in the CPU dumps produced by arm_cpu_dump_state() if we are printing FPU information. This makes it easier to interpret debug logs when predication is active. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/cpu.c | 3 +++ 1 file changed, 3 insertions(+) diff --git a/target/arm/cpu.c b/target/arm/cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/cpu.c +++ b/target/arm/cpu.c @@ -XXX,XX +XXX,XX @@ static void arm_cpu_dump_state(CPUState *cs, FILE *f, int flags) i, v); } qemu_fprintf(f, "FPSCR: %08x\n", vfp_get_fpscr(env)); + if (cpu_isar_feature(aa32_mve, cpu)) { + qemu_fprintf(f, "VPR: %08x\n", env->v7m.vpr); + } } } -- 2.20.1 In the MVE shift-and-insert insns, we special case VSLI by 0 and VSRI by <dt>. VSRI by <dt> means "don't update the destination", which is what we've implemented. However VSLI by 0 is "set destination to the input", so we don't want to use the same special-casing that we do for VSRI by <dt>. Since the generic logic gives the right answer for a shift by 0, just use that. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/mve_helper.c | 9 +++++---- 1 file changed, 5 insertions(+), 4 deletions(-) diff --git a/target/arm/mve_helper.c b/target/arm/mve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve_helper.c +++ b/target/arm/mve_helper.c @@ -XXX,XX +XXX,XX @@ DO_2SHIFT_S(vrshli_s, DO_VRSHLS) uint16_t mask; \ uint64_t shiftmask; \ unsigned e; \ - if (shift == 0 || shift == ESIZE * 8) { \ + if (shift == ESIZE * 8) { \ /* \ - * Only VSLI can shift by 0; only VSRI can shift by <dt>. \ - * The generic logic would give the right answer for 0 but \ - * fails for <dt>. \ + * Only VSRI can shift by <dt>; it should mean "don't \ + * update the destination". The generic logic can't handle \ + * this because it would try to shift by an out-of-range \ + * amount, so special case it here. \ */ \ goto done; \ } \ -- 2.20.1 A cut-and-paste error meant we handled signed VADDV like unsigned VADDV; fix the type used. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/mve_helper.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/target/arm/mve_helper.c b/target/arm/mve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve_helper.c +++ b/target/arm/mve_helper.c @@ -XXX,XX +XXX,XX @@ DO_LDAVH(vrmlsldavhxsw, int32_t, int64_t, true, true) return ra; \ } \ -DO_VADDV(vaddvsb, 1, uint8_t) -DO_VADDV(vaddvsh, 2, uint16_t) -DO_VADDV(vaddvsw, 4, uint32_t) +DO_VADDV(vaddvsb, 1, int8_t) +DO_VADDV(vaddvsh, 2, int16_t) +DO_VADDV(vaddvsw, 4, int32_t) DO_VADDV(vaddvub, 1, uint8_t) DO_VADDV(vaddvuh, 2, uint16_t) DO_VADDV(vaddvuw, 4, uint32_t) -- 2.20.1 In the MVE helpers for the narrowing operations (DO_VSHRN and DO_VSHRN_SAT) we were using the wrong bits of the predicate mask for the 'top' versions of the insn. This is because the loop works over the double-sized input elements and shifts the predicate mask by that many bits each time, but when we write out the half-sized output we must look at the mask bits for whichever half of the element we are writing to. Correct this by shifting the whole mask right by ESIZE bits for the 'top' insns. This allows us also to simplify the saturation bit checking (where we had noticed that we needed to look at a different mask bit for the 'top' insn.) Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/mve_helper.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/target/arm/mve_helper.c b/target/arm/mve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve_helper.c +++ b/target/arm/mve_helper.c @@ -XXX,XX +XXX,XX @@ DO_VSHLL_ALL(vshllt, true) TYPE *d = vd; \ uint16_t mask = mve_element_mask(env); \ unsigned le; \ + mask >>= ESIZE * TOP; \ for (le = 0; le < 16 / LESIZE; le++, mask >>= LESIZE) { \ TYPE r = FN(m[H##LESIZE(le)], shift); \ mergemask(&d[H##ESIZE(le * 2 + TOP)], r, mask); \ @@ -XXX,XX +XXX,XX @@ static inline int32_t do_sat_bhs(int64_t val, int64_t min, int64_t max, uint16_t mask = mve_element_mask(env); \ bool qc = false; \ unsigned le; \ + mask >>= ESIZE * TOP; \ for (le = 0; le < 16 / LESIZE; le++, mask >>= LESIZE) { \ bool sat = false; \ TYPE r = FN(m[H##LESIZE(le)], shift, &sat); \ mergemask(&d[H##ESIZE(le * 2 + TOP)], r, mask); \ - qc |= sat && (mask & 1 << (TOP * ESIZE)); \ + qc |= sat & mask & 1; \ } \ if (qc) { \ env->vfp.qc[0] = qc; \ -- 2.20.1 In do_sqrshl48_d() and do_uqrshl48_d() we got some of the edge cases wrong and failed to saturate correctly: (1) In do_sqrshl48_d() we used the same code that do_shrshl_bhs() does to obtain the saturated most-negative and most-positive 48-bit signed values for the large-shift-left case. This gives (1 << 47) for saturate-to-most-negative, but we weren't sign-extending this value to the 64-bit output as the pseudocode requires. (2) For left shifts by less than 48, we copied the "8/16 bit" code from do_sqrshl_bhs() and do_uqrshl_bhs(). This doesn't do the right thing because it assumes the C type we're working with is at least twice the number of bits we're saturating to (so that a shift left by bits-1 can't shift anything off the top of the value). This isn't true for bits == 48, so we would incorrectly return 0 rather than the most-positive value for situations like "shift (1 << 44) right by 20". Instead check for saturation by doing the shift and signextend and then testing whether shifting back left again gives the original value. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/mve_helper.c | 12 +++++------- 1 file changed, 5 insertions(+), 7 deletions(-) diff --git a/target/arm/mve_helper.c b/target/arm/mve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve_helper.c +++ b/target/arm/mve_helper.c @@ -XXX,XX +XXX,XX @@ static inline int64_t do_sqrshl48_d(int64_t src, int64_t shift, } return src >> -shift; } else if (shift < 48) { - int64_t val = src << shift; - int64_t extval = sextract64(val, 0, 48); - if (!sat || val == extval) { + int64_t extval = sextract64(src << shift, 0, 48); + if (!sat || src == (extval >> shift)) { return extval; } } else if (!sat || src == 0) { @@ -XXX,XX +XXX,XX @@ static inline int64_t do_sqrshl48_d(int64_t src, int64_t shift, } *sat = 1; - return (1ULL << 47) - (src >= 0); + return src >= 0 ? MAKE_64BIT_MASK(0, 47) : MAKE_64BIT_MASK(47, 17); } /* Operate on 64-bit values, but saturate at 48 bits */ @@ -XXX,XX +XXX,XX @@ static inline uint64_t do_uqrshl48_d(uint64_t src, int64_t shift, return extval; } } else if (shift < 48) { - uint64_t val = src << shift; - uint64_t extval = extract64(val, 0, 48); - if (!sat || val == extval) { + uint64_t extval = extract64(src << shift, 0, 48); + if (!sat || src == (extval >> shift)) { return extval; } } else if (!sat || src == 0) { -- 2.20.1 We got an edge case wrong in the 48-bit SQRSHRL implementation: if the shift is to the right, although it always makes the result smaller than the input value it might not be within the 48-bit range the result is supposed to be if the input had some bits in [63..48] set and the shift didn't bring all of those within the [47..0] range. Handle this similarly to the way we already do for this case in do_uqrshl48_d(): extend the calculated result from 48 bits, and return that if not saturating or if it doesn't change the result; otherwise fall through to return a saturated value. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/mve_helper.c | 11 +++++++++-- 1 file changed, 9 insertions(+), 2 deletions(-) diff --git a/target/arm/mve_helper.c b/target/arm/mve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve_helper.c +++ b/target/arm/mve_helper.c @@ -XXX,XX +XXX,XX @@ uint64_t HELPER(mve_uqrshll)(CPUARMState *env, uint64_t n, uint32_t shift) static inline int64_t do_sqrshl48_d(int64_t src, int64_t shift, bool round, uint32_t *sat) { + int64_t val, extval; + if (shift <= -48) { /* Rounding the sign bit always produces 0. */ if (round) { @@ -XXX,XX +XXX,XX @@ static inline int64_t do_sqrshl48_d(int64_t src, int64_t shift, } else if (shift < 0) { if (round) { src >>= -shift - 1; - return (src >> 1) + (src & 1); + val = (src >> 1) + (src & 1); + } else { + val = src >> -shift; + } + extval = sextract64(val, 0, 48); + if (!sat || val == extval) { + return extval; } - return src >> -shift; } else if (shift < 48) { int64_t extval = sextract64(src << shift, 0, 48); if (!sat || src == (extval >> shift)) { -- 2.20.1 In mve_element_mask(), we calculate a mask for tail predication which should have a number of 1 bits based on the value of LR. However, our MAKE_64BIT_MASK() macro has undefined behaviour when passed a zero length. Special case this to give the all-zeroes mask we require. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/mve_helper.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/target/arm/mve_helper.c b/target/arm/mve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve_helper.c +++ b/target/arm/mve_helper.c @@ -XXX,XX +XXX,XX @@ static uint16_t mve_element_mask(CPUARMState *env) */ int masklen = env->regs[14] << env->v7m.ltpsize; assert(masklen <= 16); - mask &= MAKE_64BIT_MASK(0, masklen); + uint16_t ltpmask = masklen ? MAKE_64BIT_MASK(0, masklen) : 0; + mask &= ltpmask; } if ((env->condexec_bits & 0xf) == 0) { -- 2.20.1 In some situations we need a mask telling us which parts of the vector correspond to beats that are not being executed because of ECI, separately from the combined "which bytes are predicated away" mask. Factor this mask calculation out of mve_element_mask() into its own function. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/mve_helper.c | 58 ++++++++++++++++++++++++----------------- 1 file changed, 34 insertions(+), 24 deletions(-) diff --git a/target/arm/mve_helper.c b/target/arm/mve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve_helper.c +++ b/target/arm/mve_helper.c @@ -XXX,XX +XXX,XX @@ #include "exec/exec-all.h" #include "tcg/tcg.h" +static uint16_t mve_eci_mask(CPUARMState *env) +{ + /* + * Return the mask of which elements in the MVE vector correspond + * to beats being executed. The mask has 1 bits for executed lanes + * and 0 bits where ECI says this beat was already executed. + */ + int eci; + + if ((env->condexec_bits & 0xf) != 0) { + return 0xffff; + } + + eci = env->condexec_bits >> 4; + switch (eci) { + case ECI_NONE: + return 0xffff; + case ECI_A0: + return 0xfff0; + case ECI_A0A1: + return 0xff00; + case ECI_A0A1A2: + case ECI_A0A1A2B0: + return 0xf000; + default: + g_assert_not_reached(); + } +} + static uint16_t mve_element_mask(CPUARMState *env) { /* @@ -XXX,XX +XXX,XX @@ static uint16_t mve_element_mask(CPUARMState *env) mask &= ltpmask; } - if ((env->condexec_bits & 0xf) == 0) { - /* - * ECI bits indicate which beats are already executed; - * we handle this by effectively predicating them out. - */ - int eci = env->condexec_bits >> 4; - switch (eci) { - case ECI_NONE: - break; - case ECI_A0: - mask &= 0xfff0; - break; - case ECI_A0A1: - mask &= 0xff00; - break; - case ECI_A0A1A2: - case ECI_A0A1A2B0: - mask &= 0xf000; - break; - default: - g_assert_not_reached(); - } - } - + /* + * ECI bits indicate which beats are already executed; + * we handle this by effectively predicating them out. + */ + mask &= mve_eci_mask(env); return mask; } -- 2.20.1 We were not paying attention to the ECI state when advancing the VPT state. Architecturally, VPT state advance happens for every beat (see the pseudocode VPTAdvance()), so on every beat the 4 bits of VPR.P0 corresponding to the current beat are inverted if required, and at the end of beats 1 and 3 the VPR MASK fields are updated. This means that if the ECI state says we should not be executing all 4 beats then we need to skip some of the updating of the VPR that we currently do in mve_advance_vpt(). Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/mve_helper.c | 24 +++++++++++++++++------- 1 file changed, 17 insertions(+), 7 deletions(-) diff --git a/target/arm/mve_helper.c b/target/arm/mve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve_helper.c +++ b/target/arm/mve_helper.c @@ -XXX,XX +XXX,XX @@ static void mve_advance_vpt(CPUARMState *env) /* Advance the VPT and ECI state if necessary */ uint32_t vpr = env->v7m.vpr; unsigned mask01, mask23; + uint16_t inv_mask; + uint16_t eci_mask = mve_eci_mask(env); if ((env->condexec_bits & 0xf) == 0) { env->condexec_bits = (env->condexec_bits == (ECI_A0A1A2B0 << 4)) ? @@ -XXX,XX +XXX,XX @@ static void mve_advance_vpt(CPUARMState *env) return; } + /* Invert P0 bits if needed, but only for beats we actually executed */ mask01 = FIELD_EX32(vpr, V7M_VPR, MASK01); mask23 = FIELD_EX32(vpr, V7M_VPR, MASK23); - if (mask01 > 8) { - /* high bit set, but not 0b1000: invert the relevant half of P0 */ - vpr ^= 0xff; + /* Start by assuming we invert all bits corresponding to executed beats */ + inv_mask = eci_mask; + if (mask01 <= 8) { + /* MASK01 says don't invert low half of P0 */ + inv_mask &= ~0xff; } - if (mask23 > 8) { - /* high bit set, but not 0b1000: invert the relevant half of P0 */ - vpr ^= 0xff00; + if (mask23 <= 8) { + /* MASK23 says don't invert high half of P0 */ + inv_mask &= ~0xff00; } - vpr = FIELD_DP32(vpr, V7M_VPR, MASK01, mask01 << 1); + vpr ^= inv_mask; + /* Only update MASK01 if beat 1 executed */ + if (eci_mask & 0xf0) { + vpr = FIELD_DP32(vpr, V7M_VPR, MASK01, mask01 << 1); + } + /* Beat 3 always executes, so update MASK23 */ vpr = FIELD_DP32(vpr, V7M_VPR, MASK23, mask23 << 1); env->v7m.vpr = vpr; } -- 2.20.1 For vector loads, predicated elements are zeroed, instead of retaining their previous values (as happens for most data processing operations). This means we need to distinguish "beat not executed due to ECI" (don't touch destination element) from "beat executed but predicated out" (zero destination element). Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/mve_helper.c | 8 +++++--- 1 file changed, 5 insertions(+), 3 deletions(-) diff --git a/target/arm/mve_helper.c b/target/arm/mve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve_helper.c +++ b/target/arm/mve_helper.c @@ -XXX,XX +XXX,XX @@ static void mve_advance_vpt(CPUARMState *env) env->v7m.vpr = vpr; } - +/* For loads, predicated lanes are zeroed instead of keeping their old values */ #define DO_VLDR(OP, MSIZE, LDTYPE, ESIZE, TYPE) \ void HELPER(mve_##OP)(CPUARMState *env, void *vd, uint32_t addr) \ { \ TYPE *d = vd; \ uint16_t mask = mve_element_mask(env); \ + uint16_t eci_mask = mve_eci_mask(env); \ unsigned b, e; \ /* \ * R_SXTM allows the dest reg to become UNKNOWN for abandoned \ @@ -XXX,XX +XXX,XX @@ static void mve_advance_vpt(CPUARMState *env) * then take an exception. \ */ \ for (b = 0, e = 0; b < 16; b += ESIZE, e++) { \ - if (mask & (1 << b)) { \ - d[H##ESIZE(e)] = cpu_##LDTYPE##_data_ra(env, addr, GETPC()); \ + if (eci_mask & (1 << b)) { \ + d[H##ESIZE(e)] = (mask & (1 << b)) ? \ + cpu_##LDTYPE##_data_ra(env, addr, GETPC()) : 0; \ } \ addr += MSIZE; \ } \ -- 2.20.1 Implement the MVE VMULL (polynomial) insn. Unlike Neon, this comes in two flavours: 8x8->16 and a 16x16->32. Also unlike Neon, the inputs are in either the low or the high half of each double-width element. The assembler for this insn indicates the size with "P8" or "P16", encoded into bit 28 as size = 0 or 1. We choose to follow the same encoding as VQDMULL and decode this into a->size as MO_16 or MO_32 indicating the size of the result elements. This then carries through to the helper function names where it then matches up with the existing pmull_h() which does an 8x8->16 operation and a new pmull_w() which does the 16x16->32. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/helper-mve.h | 5 +++++ target/arm/vec_internal.h | 11 +++++++++++ target/arm/mve.decode | 14 ++++++++++---- target/arm/mve_helper.c | 16 ++++++++++++++++ target/arm/translate-mve.c | 28 ++++++++++++++++++++++++++++ target/arm/vec_helper.c | 14 +++++++++++++- 6 files changed, 83 insertions(+), 5 deletions(-) diff --git a/target/arm/helper-mve.h b/target/arm/helper-mve.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/helper-mve.h +++ b/target/arm/helper-mve.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_4(mve_vmulltub, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr) DEF_HELPER_FLAGS_4(mve_vmulltuh, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr) DEF_HELPER_FLAGS_4(mve_vmulltuw, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr) +DEF_HELPER_FLAGS_4(mve_vmullpbh, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr) +DEF_HELPER_FLAGS_4(mve_vmullpth, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr) +DEF_HELPER_FLAGS_4(mve_vmullpbw, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr) +DEF_HELPER_FLAGS_4(mve_vmullptw, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr) + DEF_HELPER_FLAGS_4(mve_vqdmulhb, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr) DEF_HELPER_FLAGS_4(mve_vqdmulhh, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr) DEF_HELPER_FLAGS_4(mve_vqdmulhw, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr) diff --git a/target/arm/vec_internal.h b/target/arm/vec_internal.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/vec_internal.h +++ b/target/arm/vec_internal.h @@ -XXX,XX +XXX,XX @@ int16_t do_sqrdmlah_h(int16_t, int16_t, int16_t, bool, bool, uint32_t *); int32_t do_sqrdmlah_s(int32_t, int32_t, int32_t, bool, bool, uint32_t *); int64_t do_sqrdmlah_d(int64_t, int64_t, int64_t, bool, bool); +/* + * 8 x 8 -> 16 vector polynomial multiply where the inputs are + * in the low 8 bits of each 16-bit element +*/ +uint64_t pmull_h(uint64_t op1, uint64_t op2); +/* + * 16 x 16 -> 32 vector polynomial multiply where the inputs are + * in the low 16 bits of each 32-bit element + */ +uint64_t pmull_w(uint64_t op1, uint64_t op2); + #endif /* TARGET_ARM_VEC_INTERNALS_H */ diff --git a/target/arm/mve.decode b/target/arm/mve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve.decode +++ b/target/arm/mve.decode @@ -XXX,XX +XXX,XX @@ VHADD_U 111 1 1111 0 . .. ... 0 ... 0 0000 . 1 . 0 ... 0 @2op VHSUB_S 111 0 1111 0 . .. ... 0 ... 0 0010 . 1 . 0 ... 0 @2op VHSUB_U 111 1 1111 0 . .. ... 0 ... 0 0010 . 1 . 0 ... 0 @2op -VMULL_BS 111 0 1110 0 . .. ... 1 ... 0 1110 . 0 . 0 ... 0 @2op -VMULL_BU 111 1 1110 0 . .. ... 1 ... 0 1110 . 0 . 0 ... 0 @2op -VMULL_TS 111 0 1110 0 . .. ... 1 ... 1 1110 . 0 . 0 ... 0 @2op -VMULL_TU 111 1 1110 0 . .. ... 1 ... 1 1110 . 0 . 0 ... 0 @2op +{ + VMULLP_B 111 . 1110 0 . 11 ... 1 ... 0 1110 . 0 . 0 ... 0 @2op_sz28 + VMULL_BS 111 0 1110 0 . .. ... 1 ... 0 1110 . 0 . 0 ... 0 @2op + VMULL_BU 111 1 1110 0 . .. ... 1 ... 0 1110 . 0 . 0 ... 0 @2op +} +{ + VMULLP_T 111 . 1110 0 . 11 ... 1 ... 1 1110 . 0 . 0 ... 0 @2op_sz28 + VMULL_TS 111 0 1110 0 . .. ... 1 ... 1 1110 . 0 . 0 ... 0 @2op + VMULL_TU 111 1 1110 0 . .. ... 1 ... 1 1110 . 0 . 0 ... 0 @2op +} VQDMULH 1110 1111 0 . .. ... 0 ... 0 1011 . 1 . 0 ... 0 @2op VQRDMULH 1111 1111 0 . .. ... 0 ... 0 1011 . 1 . 0 ... 0 @2op diff --git a/target/arm/mve_helper.c b/target/arm/mve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve_helper.c +++ b/target/arm/mve_helper.c @@ -XXX,XX +XXX,XX @@ DO_2OP_L(vmulltub, 1, 1, uint8_t, 2, uint16_t, DO_MUL) DO_2OP_L(vmulltuh, 1, 2, uint16_t, 4, uint32_t, DO_MUL) DO_2OP_L(vmulltuw, 1, 4, uint32_t, 8, uint64_t, DO_MUL) +/* + * Polynomial multiply. We can always do this generating 64 bits + * of the result at a time, so we don't need to use DO_2OP_L. + */ +#define VMULLPH_MASK 0x00ff00ff00ff00ffULL +#define VMULLPW_MASK 0x0000ffff0000ffffULL +#define DO_VMULLPBH(N, M) pmull_h((N) & VMULLPH_MASK, (M) & VMULLPH_MASK) +#define DO_VMULLPTH(N, M) DO_VMULLPBH((N) >> 8, (M) >> 8) +#define DO_VMULLPBW(N, M) pmull_w((N) & VMULLPW_MASK, (M) & VMULLPW_MASK) +#define DO_VMULLPTW(N, M) DO_VMULLPBW((N) >> 16, (M) >> 16) + +DO_2OP(vmullpbh, 8, uint64_t, DO_VMULLPBH) +DO_2OP(vmullpth, 8, uint64_t, DO_VMULLPTH) +DO_2OP(vmullpbw, 8, uint64_t, DO_VMULLPBW) +DO_2OP(vmullptw, 8, uint64_t, DO_VMULLPTW) + /* * Because the computation type is at least twice as large as required, * these work for both signed and unsigned source types. diff --git a/target/arm/translate-mve.c b/target/arm/translate-mve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-mve.c +++ b/target/arm/translate-mve.c @@ -XXX,XX +XXX,XX @@ static bool trans_VQDMULLT(DisasContext *s, arg_2op *a) return do_2op(s, a, fns[a->size]); } +static bool trans_VMULLP_B(DisasContext *s, arg_2op *a) +{ + /* + * Note that a->size indicates the output size, ie VMULL.P8 + * is the 8x8->16 operation and a->size is MO_16; VMULL.P16 + * is the 16x16->32 operation and a->size is MO_32. + */ + static MVEGenTwoOpFn * const fns[] = { + NULL, + gen_helper_mve_vmullpbh, + gen_helper_mve_vmullpbw, + NULL, + }; + return do_2op(s, a, fns[a->size]); +} + +static bool trans_VMULLP_T(DisasContext *s, arg_2op *a) +{ + /* a->size is as for trans_VMULLP_B */ + static MVEGenTwoOpFn * const fns[] = { + NULL, + gen_helper_mve_vmullpth, + gen_helper_mve_vmullptw, + NULL, + }; + return do_2op(s, a, fns[a->size]); +} + /* * VADC and VSBC: these perform an add-with-carry or subtract-with-carry * of the 32-bit elements in each lane of the input vectors, where the 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 @@ static uint64_t expand_byte_to_half(uint64_t x) | ((x & 0xff000000) << 24); } -static uint64_t pmull_h(uint64_t op1, uint64_t op2) +uint64_t pmull_w(uint64_t op1, uint64_t op2) { uint64_t result = 0; int i; + for (i = 0; i < 16; ++i) { + uint64_t mask = (op1 & 0x0000000100000001ull) * 0xffffffff; + result ^= op2 & mask; + op1 >>= 1; + op2 <<= 1; + } + return result; +} +uint64_t pmull_h(uint64_t op1, uint64_t op2) +{ + uint64_t result = 0; + int i; for (i = 0; i < 8; ++i) { uint64_t mask = (op1 & 0x0001000100010001ull) * 0xffff; result ^= op2 & mask; -- 2.20.1 Implement the MVE incrementing/decrementing dup insns VIDUP, VDDUP, VIWDUP and VDWDUP. These fill the elements of a vector with successively incrementing values, starting at the offset specified in a general purpose register. The final value of the offset is written back to this register. The wrapping variants take a second general purpose register which specifies the point where the count should wrap back to 0. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/helper-mve.h | 12 ++++ target/arm/mve.decode | 25 ++++++++ target/arm/mve_helper.c | 63 +++++++++++++++++++ target/arm/translate-mve.c | 120 +++++++++++++++++++++++++++++++++++++ 4 files changed, 220 insertions(+) diff --git a/target/arm/helper-mve.h b/target/arm/helper-mve.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/helper-mve.h +++ b/target/arm/helper-mve.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_3(mve_vstrh_w, TCG_CALL_NO_WG, void, env, ptr, i32) DEF_HELPER_FLAGS_3(mve_vdup, TCG_CALL_NO_WG, void, env, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vidupb, TCG_CALL_NO_WG, i32, env, ptr, i32, i32) +DEF_HELPER_FLAGS_4(mve_viduph, TCG_CALL_NO_WG, i32, env, ptr, i32, i32) +DEF_HELPER_FLAGS_4(mve_vidupw, TCG_CALL_NO_WG, i32, env, ptr, i32, i32) + +DEF_HELPER_FLAGS_5(mve_viwdupb, TCG_CALL_NO_WG, i32, env, ptr, i32, i32, i32) +DEF_HELPER_FLAGS_5(mve_viwduph, TCG_CALL_NO_WG, i32, env, ptr, i32, i32, i32) +DEF_HELPER_FLAGS_5(mve_viwdupw, TCG_CALL_NO_WG, i32, env, ptr, i32, i32, i32) + +DEF_HELPER_FLAGS_5(mve_vdwdupb, TCG_CALL_NO_WG, i32, env, ptr, i32, i32, i32) +DEF_HELPER_FLAGS_5(mve_vdwduph, TCG_CALL_NO_WG, i32, env, ptr, i32, i32, i32) +DEF_HELPER_FLAGS_5(mve_vdwdupw, TCG_CALL_NO_WG, i32, env, ptr, i32, i32, i32) + DEF_HELPER_FLAGS_3(mve_vclsb, TCG_CALL_NO_WG, void, env, ptr, ptr) DEF_HELPER_FLAGS_3(mve_vclsh, TCG_CALL_NO_WG, void, env, ptr, ptr) DEF_HELPER_FLAGS_3(mve_vclsw, TCG_CALL_NO_WG, void, env, ptr, ptr) diff --git a/target/arm/mve.decode b/target/arm/mve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve.decode +++ b/target/arm/mve.decode @@ -XXX,XX +XXX,XX @@ &2scalar qd qn rm size &1imm qd imm cmode op &2shift qd qm shift size +&vidup qd rn size imm +&viwdup qd rn rm size imm @vldr_vstr ....... . . . . l:1 rn:4 ... ...... imm:7 &vldr_vstr qd=%qd u=0 # Note that both Rn and Qd are 3 bits only (no D bit) @@ -XXX,XX +XXX,XX @@ VDUP 1110 1110 1 1 10 ... 0 .... 1011 . 0 0 1 0000 @vdup size=0 VDUP 1110 1110 1 0 10 ... 0 .... 1011 . 0 1 1 0000 @vdup size=1 VDUP 1110 1110 1 0 10 ... 0 .... 1011 . 0 0 1 0000 @vdup size=2 +# Incrementing and decrementing dup + +# VIDUP, VDDUP format immediate: 1 << (immh:imml) +%imm_vidup 7:1 0:1 !function=vidup_imm + +# VIDUP, VDDUP registers: Rm bits [3:1] from insn, bit 0 is 1; +# Rn bits [3:1] from insn, bit 0 is 0 +%vidup_rm 1:3 !function=times_2_plus_1 +%vidup_rn 17:3 !function=times_2 + +@vidup .... .... . . size:2 .... .... .... .... .... \ + qd=%qd imm=%imm_vidup rn=%vidup_rn &vidup +@viwdup .... .... . . size:2 .... .... .... .... .... \ + qd=%qd imm=%imm_vidup rm=%vidup_rm rn=%vidup_rn &viwdup +{ + VIDUP 1110 1110 0 . .. ... 1 ... 0 1111 . 110 111 . @vidup + VIWDUP 1110 1110 0 . .. ... 1 ... 0 1111 . 110 ... . @viwdup +} +{ + VDDUP 1110 1110 0 . .. ... 1 ... 1 1111 . 110 111 . @vidup + VDWDUP 1110 1110 0 . .. ... 1 ... 1 1111 . 110 ... . @viwdup +} + # multiply-add long dual accumulate # rdahi: bits [3:1] from insn, bit 0 is 1 # rdalo: bits [3:1] from insn, bit 0 is 0 diff --git a/target/arm/mve_helper.c b/target/arm/mve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve_helper.c +++ b/target/arm/mve_helper.c @@ -XXX,XX +XXX,XX @@ uint32_t HELPER(mve_sqrshr)(CPUARMState *env, uint32_t n, uint32_t shift) { return do_sqrshl_bhs(n, -(int8_t)shift, 32, true, &env->QF); } + +#define DO_VIDUP(OP, ESIZE, TYPE, FN) \ + uint32_t HELPER(mve_##OP)(CPUARMState *env, void *vd, \ + uint32_t offset, uint32_t imm) \ + { \ + TYPE *d = vd; \ + uint16_t mask = mve_element_mask(env); \ + unsigned e; \ + for (e = 0; e < 16 / ESIZE; e++, mask >>= ESIZE) { \ + mergemask(&d[H##ESIZE(e)], offset, mask); \ + offset = FN(offset, imm); \ + } \ + mve_advance_vpt(env); \ + return offset; \ + } + +#define DO_VIWDUP(OP, ESIZE, TYPE, FN) \ + uint32_t HELPER(mve_##OP)(CPUARMState *env, void *vd, \ + uint32_t offset, uint32_t wrap, \ + uint32_t imm) \ + { \ + TYPE *d = vd; \ + uint16_t mask = mve_element_mask(env); \ + unsigned e; \ + for (e = 0; e < 16 / ESIZE; e++, mask >>= ESIZE) { \ + mergemask(&d[H##ESIZE(e)], offset, mask); \ + offset = FN(offset, wrap, imm); \ + } \ + mve_advance_vpt(env); \ + return offset; \ + } + +#define DO_VIDUP_ALL(OP, FN) \ + DO_VIDUP(OP##b, 1, int8_t, FN) \ + DO_VIDUP(OP##h, 2, int16_t, FN) \ + DO_VIDUP(OP##w, 4, int32_t, FN) + +#define DO_VIWDUP_ALL(OP, FN) \ + DO_VIWDUP(OP##b, 1, int8_t, FN) \ + DO_VIWDUP(OP##h, 2, int16_t, FN) \ + DO_VIWDUP(OP##w, 4, int32_t, FN) + +static uint32_t do_add_wrap(uint32_t offset, uint32_t wrap, uint32_t imm) +{ + offset += imm; + if (offset == wrap) { + offset = 0; + } + return offset; +} + +static uint32_t do_sub_wrap(uint32_t offset, uint32_t wrap, uint32_t imm) +{ + if (offset == 0) { + offset = wrap; + } + offset -= imm; + return offset; +} + +DO_VIDUP_ALL(vidup, DO_ADD) +DO_VIWDUP_ALL(viwdup, do_add_wrap) +DO_VIWDUP_ALL(vdwdup, do_sub_wrap) diff --git a/target/arm/translate-mve.c b/target/arm/translate-mve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-mve.c +++ b/target/arm/translate-mve.c @@ -XXX,XX +XXX,XX @@ #include "translate.h" #include "translate-a32.h" +static inline int vidup_imm(DisasContext *s, int x) +{ + return 1 << x; +} + /* Include the generated decoder */ #include "decode-mve.c.inc" @@ -XXX,XX +XXX,XX @@ typedef void MVEGenTwoOpShiftFn(TCGv_ptr, TCGv_ptr, TCGv_ptr, TCGv_i32); typedef void MVEGenDualAccOpFn(TCGv_i64, TCGv_ptr, TCGv_ptr, TCGv_ptr, TCGv_i64); typedef void MVEGenVADDVFn(TCGv_i32, TCGv_ptr, TCGv_ptr, TCGv_i32); typedef void MVEGenOneOpImmFn(TCGv_ptr, TCGv_ptr, TCGv_i64); +typedef void MVEGenVIDUPFn(TCGv_i32, TCGv_ptr, TCGv_ptr, TCGv_i32, TCGv_i32); +typedef void MVEGenVIWDUPFn(TCGv_i32, TCGv_ptr, TCGv_ptr, TCGv_i32, TCGv_i32, TCGv_i32); /* Return the offset of a Qn register (same semantics as aa32_vfp_qreg()) */ static inline long mve_qreg_offset(unsigned reg) @@ -XXX,XX +XXX,XX @@ static bool trans_VSHLC(DisasContext *s, arg_VSHLC *a) mve_update_eci(s); return true; } + +static bool do_vidup(DisasContext *s, arg_vidup *a, MVEGenVIDUPFn *fn) +{ + TCGv_ptr qd; + TCGv_i32 rn; + + /* + * Vector increment/decrement with wrap and duplicate (VIDUP, VDDUP). + * This fills the vector with elements of successively increasing + * or decreasing values, starting from Rn. + */ + if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qd)) { + return false; + } + if (a->size == MO_64) { + /* size 0b11 is another encoding */ + return false; + } + if (!mve_eci_check(s) || !vfp_access_check(s)) { + return true; + } + + qd = mve_qreg_ptr(a->qd); + rn = load_reg(s, a->rn); + fn(rn, cpu_env, qd, rn, tcg_constant_i32(a->imm)); + store_reg(s, a->rn, rn); + tcg_temp_free_ptr(qd); + mve_update_eci(s); + return true; +} + +static bool do_viwdup(DisasContext *s, arg_viwdup *a, MVEGenVIWDUPFn *fn) +{ + TCGv_ptr qd; + TCGv_i32 rn, rm; + + /* + * Vector increment/decrement with wrap and duplicate (VIWDUp, VDWDUP) + * This fills the vector with elements of successively increasing + * or decreasing values, starting from Rn. Rm specifies a point where + * the count wraps back around to 0. The updated offset is written back + * to Rn. + */ + if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qd)) { + return false; + } + if (!fn || a->rm == 13 || a->rm == 15) { + /* + * size 0b11 is another encoding; Rm == 13 is UNPREDICTABLE; + * Rm == 13 is VIWDUP, VDWDUP. + */ + return false; + } + if (!mve_eci_check(s) || !vfp_access_check(s)) { + return true; + } + + qd = mve_qreg_ptr(a->qd); + rn = load_reg(s, a->rn); + rm = load_reg(s, a->rm); + fn(rn, cpu_env, qd, rn, rm, tcg_constant_i32(a->imm)); + store_reg(s, a->rn, rn); + tcg_temp_free_ptr(qd); + tcg_temp_free_i32(rm); + mve_update_eci(s); + return true; +} + +static bool trans_VIDUP(DisasContext *s, arg_vidup *a) +{ + static MVEGenVIDUPFn * const fns[] = { + gen_helper_mve_vidupb, + gen_helper_mve_viduph, + gen_helper_mve_vidupw, + NULL, + }; + return do_vidup(s, a, fns[a->size]); +} + +static bool trans_VDDUP(DisasContext *s, arg_vidup *a) +{ + static MVEGenVIDUPFn * const fns[] = { + gen_helper_mve_vidupb, + gen_helper_mve_viduph, + gen_helper_mve_vidupw, + NULL, + }; + /* VDDUP is just like VIDUP but with a negative immediate */ + a->imm = -a->imm; + return do_vidup(s, a, fns[a->size]); +} + +static bool trans_VIWDUP(DisasContext *s, arg_viwdup *a) +{ + static MVEGenVIWDUPFn * const fns[] = { + gen_helper_mve_viwdupb, + gen_helper_mve_viwduph, + gen_helper_mve_viwdupw, + NULL, + }; + return do_viwdup(s, a, fns[a->size]); +} + +static bool trans_VDWDUP(DisasContext *s, arg_viwdup *a) +{ + static MVEGenVIWDUPFn * const fns[] = { + gen_helper_mve_vdwdupb, + gen_helper_mve_vdwduph, + gen_helper_mve_vdwdupw, + NULL, + }; + return do_viwdup(s, a, fns[a->size]); +} -- 2.20.1 Factor out the "generate code to update VPR.MASK01/MASK23" part of trans_VPST(); we are going to want to reuse it for the VPT insns. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/translate-mve.c | 31 +++++++++++++++++-------------- 1 file changed, 17 insertions(+), 14 deletions(-) diff --git a/target/arm/translate-mve.c b/target/arm/translate-mve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-mve.c +++ b/target/arm/translate-mve.c @@ -XXX,XX +XXX,XX @@ static bool trans_VRMLSLDAVH(DisasContext *s, arg_vmlaldav *a) return do_long_dual_acc(s, a, fns[a->x]); } -static bool trans_VPST(DisasContext *s, arg_VPST *a) +static void gen_vpst(DisasContext *s, uint32_t mask) { - TCGv_i32 vpr; - - /* mask == 0 is a "related encoding" */ - if (!dc_isar_feature(aa32_mve, s) || !a->mask) { - return false; - } - if (!mve_eci_check(s) || !vfp_access_check(s)) { - return true; - } /* * Set the VPR mask fields. We take advantage of MASK01 and MASK23 * being adjacent fields in the register. * - * This insn is not predicated, but it is subject to beat-wise + * Updating the masks is not predicated, but it is subject to beat-wise * execution, and the mask is updated on the odd-numbered beats. * So if PSR.ECI says we should skip beat 1, we mustn't update the * 01 mask field. */ - vpr = load_cpu_field(v7m.vpr); + TCGv_i32 vpr = load_cpu_field(v7m.vpr); switch (s->eci) { case ECI_NONE: case ECI_A0: /* Update both 01 and 23 fields */ tcg_gen_deposit_i32(vpr, vpr, - tcg_constant_i32(a->mask | (a->mask << 4)), + tcg_constant_i32(mask | (mask << 4)), R_V7M_VPR_MASK01_SHIFT, R_V7M_VPR_MASK01_LENGTH + R_V7M_VPR_MASK23_LENGTH); break; @@ -XXX,XX +XXX,XX @@ static bool trans_VPST(DisasContext *s, arg_VPST *a) case ECI_A0A1A2B0: /* Update only the 23 mask field */ tcg_gen_deposit_i32(vpr, vpr, - tcg_constant_i32(a->mask), + tcg_constant_i32(mask), R_V7M_VPR_MASK23_SHIFT, R_V7M_VPR_MASK23_LENGTH); break; default: g_assert_not_reached(); } store_cpu_field(vpr, v7m.vpr); +} + +static bool trans_VPST(DisasContext *s, arg_VPST *a) +{ + /* mask == 0 is a "related encoding" */ + if (!dc_isar_feature(aa32_mve, s) || !a->mask) { + return false; + } + if (!mve_eci_check(s) || !vfp_access_check(s)) { + return true; + } + gen_vpst(s, a->mask); mve_update_and_store_eci(s); return true; } -- 2.20.1 Implement the MVE integer vector comparison instructions. These are "VCMP (vector)" encodings T1, T2 and T3, and "VPT (vector)" encodings T1, T2 and T3. These insns compare corresponding elements in each vector, and update the VPR.P0 predicate bits with the results of the comparison. VPT also sets the VPR.MASK01 and VPR.MASK23 fields -- it is effectively "VCMP then VPST". Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/helper-mve.h | 32 ++++++++++++++++++++++ target/arm/mve.decode | 18 +++++++++++- target/arm/mve_helper.c | 56 ++++++++++++++++++++++++++++++++++++++ target/arm/translate-mve.c | 47 ++++++++++++++++++++++++++++++++ 4 files changed, 152 insertions(+), 1 deletion(-) diff --git a/target/arm/helper-mve.h b/target/arm/helper-mve.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/helper-mve.h +++ b/target/arm/helper-mve.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_3(mve_uqshl, TCG_CALL_NO_RWG, i32, env, i32, i32) DEF_HELPER_FLAGS_3(mve_sqshl, TCG_CALL_NO_RWG, i32, env, i32, i32) DEF_HELPER_FLAGS_3(mve_uqrshl, TCG_CALL_NO_RWG, i32, env, i32, i32) DEF_HELPER_FLAGS_3(mve_sqrshr, TCG_CALL_NO_RWG, i32, env, i32, i32) + +DEF_HELPER_FLAGS_3(mve_vcmpeqb, TCG_CALL_NO_WG, void, env, ptr, ptr) +DEF_HELPER_FLAGS_3(mve_vcmpeqh, TCG_CALL_NO_WG, void, env, ptr, ptr) +DEF_HELPER_FLAGS_3(mve_vcmpeqw, TCG_CALL_NO_WG, void, env, ptr, ptr) + +DEF_HELPER_FLAGS_3(mve_vcmpneb, TCG_CALL_NO_WG, void, env, ptr, ptr) +DEF_HELPER_FLAGS_3(mve_vcmpneh, TCG_CALL_NO_WG, void, env, ptr, ptr) +DEF_HELPER_FLAGS_3(mve_vcmpnew, TCG_CALL_NO_WG, void, env, ptr, ptr) + +DEF_HELPER_FLAGS_3(mve_vcmpcsb, TCG_CALL_NO_WG, void, env, ptr, ptr) +DEF_HELPER_FLAGS_3(mve_vcmpcsh, TCG_CALL_NO_WG, void, env, ptr, ptr) +DEF_HELPER_FLAGS_3(mve_vcmpcsw, TCG_CALL_NO_WG, void, env, ptr, ptr) + +DEF_HELPER_FLAGS_3(mve_vcmphib, TCG_CALL_NO_WG, void, env, ptr, ptr) +DEF_HELPER_FLAGS_3(mve_vcmphih, TCG_CALL_NO_WG, void, env, ptr, ptr) +DEF_HELPER_FLAGS_3(mve_vcmphiw, TCG_CALL_NO_WG, void, env, ptr, ptr) + +DEF_HELPER_FLAGS_3(mve_vcmpgeb, TCG_CALL_NO_WG, void, env, ptr, ptr) +DEF_HELPER_FLAGS_3(mve_vcmpgeh, TCG_CALL_NO_WG, void, env, ptr, ptr) +DEF_HELPER_FLAGS_3(mve_vcmpgew, TCG_CALL_NO_WG, void, env, ptr, ptr) + +DEF_HELPER_FLAGS_3(mve_vcmpltb, TCG_CALL_NO_WG, void, env, ptr, ptr) +DEF_HELPER_FLAGS_3(mve_vcmplth, TCG_CALL_NO_WG, void, env, ptr, ptr) +DEF_HELPER_FLAGS_3(mve_vcmpltw, TCG_CALL_NO_WG, void, env, ptr, ptr) + +DEF_HELPER_FLAGS_3(mve_vcmpgtb, TCG_CALL_NO_WG, void, env, ptr, ptr) +DEF_HELPER_FLAGS_3(mve_vcmpgth, TCG_CALL_NO_WG, void, env, ptr, ptr) +DEF_HELPER_FLAGS_3(mve_vcmpgtw, TCG_CALL_NO_WG, void, env, ptr, ptr) + +DEF_HELPER_FLAGS_3(mve_vcmpleb, TCG_CALL_NO_WG, void, env, ptr, ptr) +DEF_HELPER_FLAGS_3(mve_vcmpleh, TCG_CALL_NO_WG, void, env, ptr, ptr) +DEF_HELPER_FLAGS_3(mve_vcmplew, TCG_CALL_NO_WG, void, env, ptr, ptr) diff --git a/target/arm/mve.decode b/target/arm/mve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve.decode +++ b/target/arm/mve.decode @@ -XXX,XX +XXX,XX @@ &2shift qd qm shift size &vidup qd rn size imm &viwdup qd rn rm size imm +&vcmp qm qn size mask @vldr_vstr ....... . . . . l:1 rn:4 ... ...... imm:7 &vldr_vstr qd=%qd u=0 # Note that both Rn and Qd are 3 bits only (no D bit) @@ -XXX,XX +XXX,XX @@ @2_shr_w .... .... .. 1 ..... .... .... .... .... &2shift qd=%qd qm=%qm \ size=2 shift=%rshift_i5 +# Vector comparison; 4-bit Qm but 3-bit Qn +%mask_22_13 22:1 13:3 +@vcmp .... .... .. size:2 qn:3 . .... .... .... .... &vcmp qm=%qm mask=%mask_22_13 + # Vector loads and stores # Widening loads and narrowing stores: @@ -XXX,XX +XXX,XX @@ VQRDMULH_scalar 1111 1110 0 . .. ... 1 ... 0 1110 . 110 .... @2scalar } # Predicate operations -%mask_22_13 22:1 13:3 VPST 1111 1110 0 . 11 000 1 ... 0 1111 0100 1101 mask=%mask_22_13 # Logical immediate operations (1 reg and modified-immediate) @@ -XXX,XX +XXX,XX @@ VQRSHRUNT 111 1 1110 1 . ... ... ... 1 1111 1 1 . 0 ... 0 @2_shr_b VQRSHRUNT 111 1 1110 1 . ... ... ... 1 1111 1 1 . 0 ... 0 @2_shr_h VSHLC 111 0 1110 1 . 1 imm:5 ... 0 1111 1100 rdm:4 qd=%qd + +# Comparisons. We expand out the conditions which are split across +# encodings T1, T2, T3 and the fc bits. These include VPT, which is +# effectively "VCMP then VPST". A plain "VCMP" has a mask field of zero. +VCMPEQ 1111 1110 0 . .. ... 1 ... 0 1111 0 0 . 0 ... 0 @vcmp +VCMPNE 1111 1110 0 . .. ... 1 ... 0 1111 1 0 . 0 ... 0 @vcmp +VCMPCS 1111 1110 0 . .. ... 1 ... 0 1111 0 0 . 0 ... 1 @vcmp +VCMPHI 1111 1110 0 . .. ... 1 ... 0 1111 1 0 . 0 ... 1 @vcmp +VCMPGE 1111 1110 0 . .. ... 1 ... 1 1111 0 0 . 0 ... 0 @vcmp +VCMPLT 1111 1110 0 . .. ... 1 ... 1 1111 1 0 . 0 ... 0 @vcmp +VCMPGT 1111 1110 0 . .. ... 1 ... 1 1111 0 0 . 0 ... 1 @vcmp +VCMPLE 1111 1110 0 . .. ... 1 ... 1 1111 1 0 . 0 ... 1 @vcmp diff --git a/target/arm/mve_helper.c b/target/arm/mve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve_helper.c +++ b/target/arm/mve_helper.c @@ -XXX,XX +XXX,XX @@ static uint32_t do_sub_wrap(uint32_t offset, uint32_t wrap, uint32_t imm) DO_VIDUP_ALL(vidup, DO_ADD) DO_VIWDUP_ALL(viwdup, do_add_wrap) DO_VIWDUP_ALL(vdwdup, do_sub_wrap) + +/* + * Vector comparison. + * P0 bits for non-executed beats (where eci_mask is 0) are unchanged. + * P0 bits for predicated lanes in executed beats (where mask is 0) are 0. + * P0 bits otherwise are updated with the results of the comparisons. + * We must also keep unchanged the MASK fields at the top of v7m.vpr. + */ +#define DO_VCMP(OP, ESIZE, TYPE, FN) \ + void HELPER(glue(mve_, OP))(CPUARMState *env, void *vn, void *vm) \ + { \ + TYPE *n = vn, *m = vm; \ + uint16_t mask = mve_element_mask(env); \ + uint16_t eci_mask = mve_eci_mask(env); \ + uint16_t beatpred = 0; \ + uint16_t emask = MAKE_64BIT_MASK(0, ESIZE); \ + unsigned e; \ + for (e = 0; e < 16 / ESIZE; e++) { \ + bool r = FN(n[H##ESIZE(e)], m[H##ESIZE(e)]); \ + /* Comparison sets 0/1 bits for each byte in the element */ \ + beatpred |= r * emask; \ + emask <<= ESIZE; \ + } \ + beatpred &= mask; \ + env->v7m.vpr = (env->v7m.vpr & ~(uint32_t)eci_mask) | \ + (beatpred & eci_mask); \ + mve_advance_vpt(env); \ + } + +#define DO_VCMP_S(OP, FN) \ + DO_VCMP(OP##b, 1, int8_t, FN) \ + DO_VCMP(OP##h, 2, int16_t, FN) \ + DO_VCMP(OP##w, 4, int32_t, FN) + +#define DO_VCMP_U(OP, FN) \ + DO_VCMP(OP##b, 1, uint8_t, FN) \ + DO_VCMP(OP##h, 2, uint16_t, FN) \ + DO_VCMP(OP##w, 4, uint32_t, FN) + +#define DO_EQ(N, M) ((N) == (M)) +#define DO_NE(N, M) ((N) != (M)) +#define DO_EQ(N, M) ((N) == (M)) +#define DO_EQ(N, M) ((N) == (M)) +#define DO_GE(N, M) ((N) >= (M)) +#define DO_LT(N, M) ((N) < (M)) +#define DO_GT(N, M) ((N) > (M)) +#define DO_LE(N, M) ((N) <= (M)) + +DO_VCMP_U(vcmpeq, DO_EQ) +DO_VCMP_U(vcmpne, DO_NE) +DO_VCMP_U(vcmpcs, DO_GE) +DO_VCMP_U(vcmphi, DO_GT) +DO_VCMP_S(vcmpge, DO_GE) +DO_VCMP_S(vcmplt, DO_LT) +DO_VCMP_S(vcmpgt, DO_GT) +DO_VCMP_S(vcmple, DO_LE) diff --git a/target/arm/translate-mve.c b/target/arm/translate-mve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-mve.c +++ b/target/arm/translate-mve.c @@ -XXX,XX +XXX,XX @@ typedef void MVEGenVADDVFn(TCGv_i32, TCGv_ptr, TCGv_ptr, TCGv_i32); typedef void MVEGenOneOpImmFn(TCGv_ptr, TCGv_ptr, TCGv_i64); typedef void MVEGenVIDUPFn(TCGv_i32, TCGv_ptr, TCGv_ptr, TCGv_i32, TCGv_i32); typedef void MVEGenVIWDUPFn(TCGv_i32, TCGv_ptr, TCGv_ptr, TCGv_i32, TCGv_i32, TCGv_i32); +typedef void MVEGenCmpFn(TCGv_ptr, TCGv_ptr, TCGv_ptr); /* Return the offset of a Qn register (same semantics as aa32_vfp_qreg()) */ static inline long mve_qreg_offset(unsigned reg) @@ -XXX,XX +XXX,XX @@ static bool trans_VDWDUP(DisasContext *s, arg_viwdup *a) }; return do_viwdup(s, a, fns[a->size]); } + +static bool do_vcmp(DisasContext *s, arg_vcmp *a, MVEGenCmpFn *fn) +{ + TCGv_ptr qn, qm; + + if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qm) || + !fn) { + return false; + } + if (!mve_eci_check(s) || !vfp_access_check(s)) { + return true; + } + + qn = mve_qreg_ptr(a->qn); + qm = mve_qreg_ptr(a->qm); + fn(cpu_env, qn, qm); + tcg_temp_free_ptr(qn); + tcg_temp_free_ptr(qm); + if (a->mask) { + /* VPT */ + gen_vpst(s, a->mask); + } + mve_update_eci(s); + return true; +} + +#define DO_VCMP(INSN, FN) \ + static bool trans_##INSN(DisasContext *s, arg_vcmp *a) \ + { \ + static MVEGenCmpFn * const fns[] = { \ + gen_helper_mve_##FN##b, \ + gen_helper_mve_##FN##h, \ + gen_helper_mve_##FN##w, \ + NULL, \ + }; \ + return do_vcmp(s, a, fns[a->size]); \ + } + +DO_VCMP(VCMPEQ, vcmpeq) +DO_VCMP(VCMPNE, vcmpne) +DO_VCMP(VCMPCS, vcmpcs) +DO_VCMP(VCMPHI, vcmphi) +DO_VCMP(VCMPGE, vcmpge) +DO_VCMP(VCMPLT, vcmplt) +DO_VCMP(VCMPGT, vcmpgt) +DO_VCMP(VCMPLE, vcmple) -- 2.20.1 Implement the MVE integer vector comparison instructions that compare each element against a scalar from a general purpose register. These are "VCMP (vector)" encodings T4, T5 and T6 and "VPT (vector)" encodings T4, T5 and T6. We have to move the decodetree pattern for VPST, because it overlaps with VCMP T4 with size = 0b11. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/helper-mve.h | 32 +++++++++++++++++++++++++++ target/arm/mve.decode | 18 +++++++++++++--- target/arm/mve_helper.c | 44 +++++++++++++++++++++++++++++++------- target/arm/translate-mve.c | 43 +++++++++++++++++++++++++++++++++++++ 4 files changed, 126 insertions(+), 11 deletions(-) diff --git a/target/arm/helper-mve.h b/target/arm/helper-mve.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/helper-mve.h +++ b/target/arm/helper-mve.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_3(mve_vcmpgtw, TCG_CALL_NO_WG, void, env, ptr, ptr) DEF_HELPER_FLAGS_3(mve_vcmpleb, TCG_CALL_NO_WG, void, env, ptr, ptr) DEF_HELPER_FLAGS_3(mve_vcmpleh, TCG_CALL_NO_WG, void, env, ptr, ptr) DEF_HELPER_FLAGS_3(mve_vcmplew, TCG_CALL_NO_WG, void, env, ptr, ptr) + +DEF_HELPER_FLAGS_3(mve_vcmpeq_scalarb, TCG_CALL_NO_WG, void, env, ptr, i32) +DEF_HELPER_FLAGS_3(mve_vcmpeq_scalarh, TCG_CALL_NO_WG, void, env, ptr, i32) +DEF_HELPER_FLAGS_3(mve_vcmpeq_scalarw, TCG_CALL_NO_WG, void, env, ptr, i32) + +DEF_HELPER_FLAGS_3(mve_vcmpne_scalarb, TCG_CALL_NO_WG, void, env, ptr, i32) +DEF_HELPER_FLAGS_3(mve_vcmpne_scalarh, TCG_CALL_NO_WG, void, env, ptr, i32) +DEF_HELPER_FLAGS_3(mve_vcmpne_scalarw, TCG_CALL_NO_WG, void, env, ptr, i32) + +DEF_HELPER_FLAGS_3(mve_vcmpcs_scalarb, TCG_CALL_NO_WG, void, env, ptr, i32) +DEF_HELPER_FLAGS_3(mve_vcmpcs_scalarh, TCG_CALL_NO_WG, void, env, ptr, i32) +DEF_HELPER_FLAGS_3(mve_vcmpcs_scalarw, TCG_CALL_NO_WG, void, env, ptr, i32) + +DEF_HELPER_FLAGS_3(mve_vcmphi_scalarb, TCG_CALL_NO_WG, void, env, ptr, i32) +DEF_HELPER_FLAGS_3(mve_vcmphi_scalarh, TCG_CALL_NO_WG, void, env, ptr, i32) +DEF_HELPER_FLAGS_3(mve_vcmphi_scalarw, TCG_CALL_NO_WG, void, env, ptr, i32) + +DEF_HELPER_FLAGS_3(mve_vcmpge_scalarb, TCG_CALL_NO_WG, void, env, ptr, i32) +DEF_HELPER_FLAGS_3(mve_vcmpge_scalarh, TCG_CALL_NO_WG, void, env, ptr, i32) +DEF_HELPER_FLAGS_3(mve_vcmpge_scalarw, TCG_CALL_NO_WG, void, env, ptr, i32) + +DEF_HELPER_FLAGS_3(mve_vcmplt_scalarb, TCG_CALL_NO_WG, void, env, ptr, i32) +DEF_HELPER_FLAGS_3(mve_vcmplt_scalarh, TCG_CALL_NO_WG, void, env, ptr, i32) +DEF_HELPER_FLAGS_3(mve_vcmplt_scalarw, TCG_CALL_NO_WG, void, env, ptr, i32) + +DEF_HELPER_FLAGS_3(mve_vcmpgt_scalarb, TCG_CALL_NO_WG, void, env, ptr, i32) +DEF_HELPER_FLAGS_3(mve_vcmpgt_scalarh, TCG_CALL_NO_WG, void, env, ptr, i32) +DEF_HELPER_FLAGS_3(mve_vcmpgt_scalarw, TCG_CALL_NO_WG, void, env, ptr, i32) + +DEF_HELPER_FLAGS_3(mve_vcmple_scalarb, TCG_CALL_NO_WG, void, env, ptr, i32) +DEF_HELPER_FLAGS_3(mve_vcmple_scalarh, TCG_CALL_NO_WG, void, env, ptr, i32) +DEF_HELPER_FLAGS_3(mve_vcmple_scalarw, TCG_CALL_NO_WG, void, env, ptr, i32) diff --git a/target/arm/mve.decode b/target/arm/mve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve.decode +++ b/target/arm/mve.decode @@ -XXX,XX +XXX,XX @@ &vidup qd rn size imm &viwdup qd rn rm size imm &vcmp qm qn size mask +&vcmp_scalar qn rm size mask @vldr_vstr ....... . . . . l:1 rn:4 ... ...... imm:7 &vldr_vstr qd=%qd u=0 # Note that both Rn and Qd are 3 bits only (no D bit) @@ -XXX,XX +XXX,XX @@ # Vector comparison; 4-bit Qm but 3-bit Qn %mask_22_13 22:1 13:3 @vcmp .... .... .. size:2 qn:3 . .... .... .... .... &vcmp qm=%qm mask=%mask_22_13 +@vcmp_scalar .... .... .. size:2 qn:3 . .... .... .... rm:4 &vcmp_scalar \ + mask=%mask_22_13 # Vector loads and stores @@ -XXX,XX +XXX,XX @@ VQRDMULH_scalar 1111 1110 0 . .. ... 1 ... 0 1110 . 110 .... @2scalar rdahi=%rdahi rdalo=%rdalo } -# Predicate operations -VPST 1111 1110 0 . 11 000 1 ... 0 1111 0100 1101 mask=%mask_22_13 - # Logical immediate operations (1 reg and modified-immediate) # The cmode/op bits here decode VORR/VBIC/VMOV/VMVN, but @@ -XXX,XX +XXX,XX @@ VCMPGE 1111 1110 0 . .. ... 1 ... 1 1111 0 0 . 0 ... 0 @vcmp VCMPLT 1111 1110 0 . .. ... 1 ... 1 1111 1 0 . 0 ... 0 @vcmp VCMPGT 1111 1110 0 . .. ... 1 ... 1 1111 0 0 . 0 ... 1 @vcmp VCMPLE 1111 1110 0 . .. ... 1 ... 1 1111 1 0 . 0 ... 1 @vcmp + +{ + VPST 1111 1110 0 . 11 000 1 ... 0 1111 0100 1101 mask=%mask_22_13 + VCMPEQ_scalar 1111 1110 0 . .. ... 1 ... 0 1111 0 1 0 0 .... @vcmp_scalar +} +VCMPNE_scalar 1111 1110 0 . .. ... 1 ... 0 1111 1 1 0 0 .... @vcmp_scalar +VCMPCS_scalar 1111 1110 0 . .. ... 1 ... 0 1111 0 1 1 0 .... @vcmp_scalar +VCMPHI_scalar 1111 1110 0 . .. ... 1 ... 0 1111 1 1 1 0 .... @vcmp_scalar +VCMPGE_scalar 1111 1110 0 . .. ... 1 ... 1 1111 0 1 0 0 .... @vcmp_scalar +VCMPLT_scalar 1111 1110 0 . .. ... 1 ... 1 1111 1 1 0 0 .... @vcmp_scalar +VCMPGT_scalar 1111 1110 0 . .. ... 1 ... 1 1111 0 1 1 0 .... @vcmp_scalar +VCMPLE_scalar 1111 1110 0 . .. ... 1 ... 1 1111 1 1 1 0 .... @vcmp_scalar diff --git a/target/arm/mve_helper.c b/target/arm/mve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve_helper.c +++ b/target/arm/mve_helper.c @@ -XXX,XX +XXX,XX @@ DO_VIWDUP_ALL(vdwdup, do_sub_wrap) mve_advance_vpt(env); \ } -#define DO_VCMP_S(OP, FN) \ - DO_VCMP(OP##b, 1, int8_t, FN) \ - DO_VCMP(OP##h, 2, int16_t, FN) \ - DO_VCMP(OP##w, 4, int32_t, FN) +#define DO_VCMP_SCALAR(OP, ESIZE, TYPE, FN) \ + void HELPER(glue(mve_, OP))(CPUARMState *env, void *vn, \ + uint32_t rm) \ + { \ + TYPE *n = vn; \ + uint16_t mask = mve_element_mask(env); \ + uint16_t eci_mask = mve_eci_mask(env); \ + uint16_t beatpred = 0; \ + uint16_t emask = MAKE_64BIT_MASK(0, ESIZE); \ + unsigned e; \ + for (e = 0; e < 16 / ESIZE; e++) { \ + bool r = FN(n[H##ESIZE(e)], (TYPE)rm); \ + /* Comparison sets 0/1 bits for each byte in the element */ \ + beatpred |= r * emask; \ + emask <<= ESIZE; \ + } \ + beatpred &= mask; \ + env->v7m.vpr = (env->v7m.vpr & ~(uint32_t)eci_mask) | \ + (beatpred & eci_mask); \ + mve_advance_vpt(env); \ + } -#define DO_VCMP_U(OP, FN) \ - DO_VCMP(OP##b, 1, uint8_t, FN) \ - DO_VCMP(OP##h, 2, uint16_t, FN) \ - DO_VCMP(OP##w, 4, uint32_t, FN) +#define DO_VCMP_S(OP, FN) \ + DO_VCMP(OP##b, 1, int8_t, FN) \ + DO_VCMP(OP##h, 2, int16_t, FN) \ + DO_VCMP(OP##w, 4, int32_t, FN) \ + DO_VCMP_SCALAR(OP##_scalarb, 1, int8_t, FN) \ + DO_VCMP_SCALAR(OP##_scalarh, 2, int16_t, FN) \ + DO_VCMP_SCALAR(OP##_scalarw, 4, int32_t, FN) + +#define DO_VCMP_U(OP, FN) \ + DO_VCMP(OP##b, 1, uint8_t, FN) \ + DO_VCMP(OP##h, 2, uint16_t, FN) \ + DO_VCMP(OP##w, 4, uint32_t, FN) \ + DO_VCMP_SCALAR(OP##_scalarb, 1, uint8_t, FN) \ + DO_VCMP_SCALAR(OP##_scalarh, 2, uint16_t, FN) \ + DO_VCMP_SCALAR(OP##_scalarw, 4, uint32_t, FN) #define DO_EQ(N, M) ((N) == (M)) #define DO_NE(N, M) ((N) != (M)) diff --git a/target/arm/translate-mve.c b/target/arm/translate-mve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-mve.c +++ b/target/arm/translate-mve.c @@ -XXX,XX +XXX,XX @@ typedef void MVEGenOneOpImmFn(TCGv_ptr, TCGv_ptr, TCGv_i64); typedef void MVEGenVIDUPFn(TCGv_i32, TCGv_ptr, TCGv_ptr, TCGv_i32, TCGv_i32); typedef void MVEGenVIWDUPFn(TCGv_i32, TCGv_ptr, TCGv_ptr, TCGv_i32, TCGv_i32, TCGv_i32); typedef void MVEGenCmpFn(TCGv_ptr, TCGv_ptr, TCGv_ptr); +typedef void MVEGenScalarCmpFn(TCGv_ptr, TCGv_ptr, TCGv_i32); /* Return the offset of a Qn register (same semantics as aa32_vfp_qreg()) */ static inline long mve_qreg_offset(unsigned reg) @@ -XXX,XX +XXX,XX @@ static bool do_vcmp(DisasContext *s, arg_vcmp *a, MVEGenCmpFn *fn) return true; } +static bool do_vcmp_scalar(DisasContext *s, arg_vcmp_scalar *a, + MVEGenScalarCmpFn *fn) +{ + TCGv_ptr qn; + TCGv_i32 rm; + + if (!dc_isar_feature(aa32_mve, s) || !fn || a->rm == 13) { + return false; + } + if (!mve_eci_check(s) || !vfp_access_check(s)) { + return true; + } + + qn = mve_qreg_ptr(a->qn); + if (a->rm == 15) { + /* Encoding Rm=0b1111 means "constant zero" */ + rm = tcg_constant_i32(0); + } else { + rm = load_reg(s, a->rm); + } + fn(cpu_env, qn, rm); + tcg_temp_free_ptr(qn); + tcg_temp_free_i32(rm); + if (a->mask) { + /* VPT */ + gen_vpst(s, a->mask); + } + mve_update_eci(s); + return true; +} + #define DO_VCMP(INSN, FN) \ static bool trans_##INSN(DisasContext *s, arg_vcmp *a) \ { \ @@ -XXX,XX +XXX,XX @@ static bool do_vcmp(DisasContext *s, arg_vcmp *a, MVEGenCmpFn *fn) NULL, \ }; \ return do_vcmp(s, a, fns[a->size]); \ + } \ + static bool trans_##INSN##_scalar(DisasContext *s, \ + arg_vcmp_scalar *a) \ + { \ + static MVEGenScalarCmpFn * const fns[] = { \ + gen_helper_mve_##FN##_scalarb, \ + gen_helper_mve_##FN##_scalarh, \ + gen_helper_mve_##FN##_scalarw, \ + NULL, \ + }; \ + return do_vcmp_scalar(s, a, fns[a->size]); \ } DO_VCMP(VCMPEQ, vcmpeq) -- 2.20.1 Implement the MVE VPSEL insn, which sets each byte of the destination vector Qd to the byte from either Qn or Qm depending on the value of the corresponding bit in VPR.P0. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/helper-mve.h | 2 ++ target/arm/mve.decode | 7 +++++-- target/arm/mve_helper.c | 19 +++++++++++++++++++ target/arm/translate-mve.c | 2 ++ 4 files changed, 28 insertions(+), 2 deletions(-) diff --git a/target/arm/helper-mve.h b/target/arm/helper-mve.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/helper-mve.h +++ b/target/arm/helper-mve.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_4(mve_vorr, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr) DEF_HELPER_FLAGS_4(mve_vorn, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr) DEF_HELPER_FLAGS_4(mve_veor, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr) +DEF_HELPER_FLAGS_4(mve_vpsel, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr) + DEF_HELPER_FLAGS_4(mve_vaddb, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr) DEF_HELPER_FLAGS_4(mve_vaddh, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr) DEF_HELPER_FLAGS_4(mve_vaddw, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr) diff --git a/target/arm/mve.decode b/target/arm/mve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve.decode +++ b/target/arm/mve.decode @@ -XXX,XX +XXX,XX @@ VSHLC 111 0 1110 1 . 1 imm:5 ... 0 1111 1100 rdm:4 qd=%qd # effectively "VCMP then VPST". A plain "VCMP" has a mask field of zero. VCMPEQ 1111 1110 0 . .. ... 1 ... 0 1111 0 0 . 0 ... 0 @vcmp VCMPNE 1111 1110 0 . .. ... 1 ... 0 1111 1 0 . 0 ... 0 @vcmp -VCMPCS 1111 1110 0 . .. ... 1 ... 0 1111 0 0 . 0 ... 1 @vcmp -VCMPHI 1111 1110 0 . .. ... 1 ... 0 1111 1 0 . 0 ... 1 @vcmp +{ + VPSEL 1111 1110 0 . 11 ... 1 ... 0 1111 . 0 . 0 ... 1 @2op_nosz + VCMPCS 1111 1110 0 . .. ... 1 ... 0 1111 0 0 . 0 ... 1 @vcmp + VCMPHI 1111 1110 0 . .. ... 1 ... 0 1111 1 0 . 0 ... 1 @vcmp +} VCMPGE 1111 1110 0 . .. ... 1 ... 1 1111 0 0 . 0 ... 0 @vcmp VCMPLT 1111 1110 0 . .. ... 1 ... 1 1111 1 0 . 0 ... 0 @vcmp VCMPGT 1111 1110 0 . .. ... 1 ... 1 1111 0 0 . 0 ... 1 @vcmp diff --git a/target/arm/mve_helper.c b/target/arm/mve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve_helper.c +++ b/target/arm/mve_helper.c @@ -XXX,XX +XXX,XX @@ DO_VCMP_S(vcmpge, DO_GE) DO_VCMP_S(vcmplt, DO_LT) DO_VCMP_S(vcmpgt, DO_GT) DO_VCMP_S(vcmple, DO_LE) + +void HELPER(mve_vpsel)(CPUARMState *env, void *vd, void *vn, void *vm) +{ + /* + * Qd[n] = VPR.P0[n] ? Qn[n] : Qm[n] + * but note that whether bytes are written to Qd is still subject + * to (all forms of) predication in the usual way. + */ + uint64_t *d = vd, *n = vn, *m = vm; + uint16_t mask = mve_element_mask(env); + uint16_t p0 = FIELD_EX32(env->v7m.vpr, V7M_VPR, P0); + unsigned e; + for (e = 0; e < 16 / 8; e++, mask >>= 8, p0 >>= 8) { + uint64_t r = m[H8(e)]; + mergemask(&r, n[H8(e)], p0); + mergemask(&d[H8(e)], r, mask); + } + mve_advance_vpt(env); +} diff --git a/target/arm/translate-mve.c b/target/arm/translate-mve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-mve.c +++ b/target/arm/translate-mve.c @@ -XXX,XX +XXX,XX @@ DO_LOGIC(VORR, gen_helper_mve_vorr) DO_LOGIC(VORN, gen_helper_mve_vorn) DO_LOGIC(VEOR, gen_helper_mve_veor) +DO_LOGIC(VPSEL, gen_helper_mve_vpsel) + #define DO_2OP(INSN, FN) \ static bool trans_##INSN(DisasContext *s, arg_2op *a) \ { \ -- 2.20.1 Implement the MVE VMLAS insn, which multiplies a vector by a vector and adds a scalar. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/helper-mve.h | 4 ++++ target/arm/mve.decode | 3 +++ target/arm/mve_helper.c | 26 ++++++++++++++++++++++++++ target/arm/translate-mve.c | 1 + 4 files changed, 34 insertions(+) diff --git a/target/arm/helper-mve.h b/target/arm/helper-mve.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/helper-mve.h +++ b/target/arm/helper-mve.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_4(mve_vqdmullb_scalarw, TCG_CALL_NO_WG, void, env, ptr, ptr, i3 DEF_HELPER_FLAGS_4(mve_vqdmullt_scalarh, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) DEF_HELPER_FLAGS_4(mve_vqdmullt_scalarw, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vmlasb, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vmlash, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vmlasw, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) + DEF_HELPER_FLAGS_4(mve_vmlaldavsh, TCG_CALL_NO_WG, i64, env, ptr, ptr, i64) DEF_HELPER_FLAGS_4(mve_vmlaldavsw, TCG_CALL_NO_WG, i64, env, ptr, ptr, i64) DEF_HELPER_FLAGS_4(mve_vmlaldavxsh, TCG_CALL_NO_WG, i64, env, ptr, ptr, i64) diff --git a/target/arm/mve.decode b/target/arm/mve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve.decode +++ b/target/arm/mve.decode @@ -XXX,XX +XXX,XX @@ VBRSR 1111 1110 0 . .. ... 1 ... 1 1110 . 110 .... @2scalar VQDMULH_scalar 1110 1110 0 . .. ... 1 ... 0 1110 . 110 .... @2scalar VQRDMULH_scalar 1111 1110 0 . .. ... 1 ... 0 1110 . 110 .... @2scalar +# The U bit (28) is don't-care because it does not affect the result +VMLAS 111- 1110 0 . .. ... 1 ... 1 1110 . 100 .... @2scalar + # Vector add across vector { VADDV 111 u:1 1110 1111 size:2 01 ... 0 1111 0 0 a:1 0 qm:3 0 rda=%rdalo diff --git a/target/arm/mve_helper.c b/target/arm/mve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve_helper.c +++ b/target/arm/mve_helper.c @@ -XXX,XX +XXX,XX @@ DO_VQDMLADH_OP(vqrdmlsdhxw, 4, int32_t, 1, 1, do_vqdmlsdh_w) mve_advance_vpt(env); \ } +/* "accumulating" version where FN takes d as well as n and m */ +#define DO_2OP_ACC_SCALAR(OP, ESIZE, TYPE, FN) \ + void HELPER(glue(mve_, OP))(CPUARMState *env, void *vd, void *vn, \ + uint32_t rm) \ + { \ + TYPE *d = vd, *n = vn; \ + TYPE m = rm; \ + uint16_t mask = mve_element_mask(env); \ + unsigned e; \ + for (e = 0; e < 16 / ESIZE; e++, mask >>= ESIZE) { \ + mergemask(&d[H##ESIZE(e)], \ + FN(d[H##ESIZE(e)], n[H##ESIZE(e)], m), mask); \ + } \ + mve_advance_vpt(env); \ + } + /* provide unsigned 2-op scalar helpers for all sizes */ #define DO_2OP_SCALAR_U(OP, FN) \ DO_2OP_SCALAR(OP##b, 1, uint8_t, FN) \ @@ -XXX,XX +XXX,XX @@ DO_VQDMLADH_OP(vqrdmlsdhxw, 4, int32_t, 1, 1, do_vqdmlsdh_w) DO_2OP_SCALAR(OP##h, 2, int16_t, FN) \ DO_2OP_SCALAR(OP##w, 4, int32_t, FN) +#define DO_2OP_ACC_SCALAR_U(OP, FN) \ + DO_2OP_ACC_SCALAR(OP##b, 1, uint8_t, FN) \ + DO_2OP_ACC_SCALAR(OP##h, 2, uint16_t, FN) \ + DO_2OP_ACC_SCALAR(OP##w, 4, uint32_t, FN) + DO_2OP_SCALAR_U(vadd_scalar, DO_ADD) DO_2OP_SCALAR_U(vsub_scalar, DO_SUB) DO_2OP_SCALAR_U(vmul_scalar, DO_MUL) @@ -XXX,XX +XXX,XX @@ DO_2OP_SAT_SCALAR(vqrdmulh_scalarb, 1, int8_t, DO_QRDMULH_B) DO_2OP_SAT_SCALAR(vqrdmulh_scalarh, 2, int16_t, DO_QRDMULH_H) DO_2OP_SAT_SCALAR(vqrdmulh_scalarw, 4, int32_t, DO_QRDMULH_W) +/* Vector by vector plus scalar */ +#define DO_VMLAS(D, N, M) ((N) * (D) + (M)) + +DO_2OP_ACC_SCALAR_U(vmlas, DO_VMLAS) + /* * Long saturating scalar ops. As with DO_2OP_L, TYPE and H are for the * input (smaller) type and LESIZE, LTYPE, LH for the output (long) type. diff --git a/target/arm/translate-mve.c b/target/arm/translate-mve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-mve.c +++ b/target/arm/translate-mve.c @@ -XXX,XX +XXX,XX @@ DO_2OP_SCALAR(VQSUB_U_scalar, vqsubu_scalar) DO_2OP_SCALAR(VQDMULH_scalar, vqdmulh_scalar) DO_2OP_SCALAR(VQRDMULH_scalar, vqrdmulh_scalar) DO_2OP_SCALAR(VBRSR, vbrsr) +DO_2OP_SCALAR(VMLAS, vmlas) static bool trans_VQDMULLB_scalar(DisasContext *s, arg_2scalar *a) { -- 2.20.1 Implement the MVE instructions which perform shifts by a scalar. These are VSHL T2, VRSHL T2, VQSHL T1 and VQRSHL T2. They take the shift amount in a general purpose register and shift every element in the vector by that amount. Mostly we can reuse the helper functions for shift-by-immediate; we do need two new helpers for VQRSHL. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/helper-mve.h | 8 +++++++ target/arm/mve.decode | 23 ++++++++++++++++--- target/arm/mve_helper.c | 2 ++ target/arm/translate-mve.c | 46 ++++++++++++++++++++++++++++++++++++++ 4 files changed, 76 insertions(+), 3 deletions(-) diff --git a/target/arm/helper-mve.h b/target/arm/helper-mve.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/helper-mve.h +++ b/target/arm/helper-mve.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_4(mve_vrshli_ub, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) DEF_HELPER_FLAGS_4(mve_vrshli_uh, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) DEF_HELPER_FLAGS_4(mve_vrshli_uw, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vqrshli_sb, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vqrshli_sh, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vqrshli_sw, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) + +DEF_HELPER_FLAGS_4(mve_vqrshli_ub, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vqrshli_uh, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vqrshli_uw, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) + DEF_HELPER_FLAGS_4(mve_vshllbsb, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) DEF_HELPER_FLAGS_4(mve_vshllbsh, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) DEF_HELPER_FLAGS_4(mve_vshllbub, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) diff --git a/target/arm/mve.decode b/target/arm/mve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve.decode +++ b/target/arm/mve.decode @@ -XXX,XX +XXX,XX @@ &viwdup qd rn rm size imm &vcmp qm qn size mask &vcmp_scalar qn rm size mask +&shl_scalar qda rm size @vldr_vstr ....... . . . . l:1 rn:4 ... ...... imm:7 &vldr_vstr qd=%qd u=0 # Note that both Rn and Qd are 3 bits only (no D bit) @@ -XXX,XX +XXX,XX @@ @2_shr_w .... .... .. 1 ..... .... .... .... .... &2shift qd=%qd qm=%qm \ size=2 shift=%rshift_i5 +@shl_scalar .... .... .... size:2 .. .... .... .... rm:4 &shl_scalar qda=%qd + # Vector comparison; 4-bit Qm but 3-bit Qn %mask_22_13 22:1 13:3 @vcmp .... .... .. size:2 qn:3 . .... .... .... .... &vcmp qm=%qm mask=%mask_22_13 @@ -XXX,XX +XXX,XX @@ VRMLSLDAVH 1111 1110 1 ... ... 0 ... x:1 1110 . 0 a:1 0 ... 1 @vmlaldav_no VADD_scalar 1110 1110 0 . .. ... 1 ... 0 1111 . 100 .... @2scalar VSUB_scalar 1110 1110 0 . .. ... 1 ... 1 1111 . 100 .... @2scalar -VMUL_scalar 1110 1110 0 . .. ... 1 ... 1 1110 . 110 .... @2scalar + +{ + VSHL_S_scalar 1110 1110 0 . 11 .. 01 ... 1 1110 0110 .... @shl_scalar + VRSHL_S_scalar 1110 1110 0 . 11 .. 11 ... 1 1110 0110 .... @shl_scalar + VQSHL_S_scalar 1110 1110 0 . 11 .. 01 ... 1 1110 1110 .... @shl_scalar + VQRSHL_S_scalar 1110 1110 0 . 11 .. 11 ... 1 1110 1110 .... @shl_scalar + VMUL_scalar 1110 1110 0 . .. ... 1 ... 1 1110 . 110 .... @2scalar +} + +{ + VSHL_U_scalar 1111 1110 0 . 11 .. 01 ... 1 1110 0110 .... @shl_scalar + VRSHL_U_scalar 1111 1110 0 . 11 .. 11 ... 1 1110 0110 .... @shl_scalar + VQSHL_U_scalar 1111 1110 0 . 11 .. 01 ... 1 1110 1110 .... @shl_scalar + VQRSHL_U_scalar 1111 1110 0 . 11 .. 11 ... 1 1110 1110 .... @shl_scalar + VBRSR 1111 1110 0 . .. ... 1 ... 1 1110 . 110 .... @2scalar +} + VHADD_S_scalar 1110 1110 0 . .. ... 0 ... 0 1111 . 100 .... @2scalar VHADD_U_scalar 1111 1110 0 . .. ... 0 ... 0 1111 . 100 .... @2scalar VHSUB_S_scalar 1110 1110 0 . .. ... 0 ... 1 1111 . 100 .... @2scalar @@ -XXX,XX +XXX,XX @@ VHSUB_U_scalar 1111 1110 0 . .. ... 0 ... 1 1111 . 100 .... @2scalar size=%size_28 } -VBRSR 1111 1110 0 . .. ... 1 ... 1 1110 . 110 .... @2scalar - VQDMULH_scalar 1110 1110 0 . .. ... 1 ... 0 1110 . 110 .... @2scalar VQRDMULH_scalar 1111 1110 0 . .. ... 1 ... 0 1110 . 110 .... @2scalar diff --git a/target/arm/mve_helper.c b/target/arm/mve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve_helper.c +++ b/target/arm/mve_helper.c @@ -XXX,XX +XXX,XX @@ DO_2SHIFT_SAT_S(vqshli_s, DO_SQSHL_OP) DO_2SHIFT_SAT_S(vqshlui_s, DO_SUQSHL_OP) DO_2SHIFT_U(vrshli_u, DO_VRSHLU) DO_2SHIFT_S(vrshli_s, DO_VRSHLS) +DO_2SHIFT_SAT_U(vqrshli_u, DO_UQRSHL_OP) +DO_2SHIFT_SAT_S(vqrshli_s, DO_SQRSHL_OP) /* Shift-and-insert; we always work with 64 bits at a time */ #define DO_2SHIFT_INSERT(OP, ESIZE, SHIFTFN, MASKFN) \ diff --git a/target/arm/translate-mve.c b/target/arm/translate-mve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-mve.c +++ b/target/arm/translate-mve.c @@ -XXX,XX +XXX,XX @@ DO_2SHIFT(VRSHRI_U, vrshli_u, true) DO_2SHIFT(VSRI, vsri, false) DO_2SHIFT(VSLI, vsli, false) +static bool do_2shift_scalar(DisasContext *s, arg_shl_scalar *a, + MVEGenTwoOpShiftFn *fn) +{ + TCGv_ptr qda; + TCGv_i32 rm; + + if (!dc_isar_feature(aa32_mve, s) || + !mve_check_qreg_bank(s, a->qda) || + a->rm == 13 || a->rm == 15 || !fn) { + /* Rm cases are UNPREDICTABLE */ + return false; + } + if (!mve_eci_check(s) || !vfp_access_check(s)) { + return true; + } + + qda = mve_qreg_ptr(a->qda); + rm = load_reg(s, a->rm); + fn(cpu_env, qda, qda, rm); + tcg_temp_free_ptr(qda); + tcg_temp_free_i32(rm); + mve_update_eci(s); + return true; +} + +#define DO_2SHIFT_SCALAR(INSN, FN) \ + static bool trans_##INSN(DisasContext *s, arg_shl_scalar *a) \ + { \ + static MVEGenTwoOpShiftFn * const fns[] = { \ + gen_helper_mve_##FN##b, \ + gen_helper_mve_##FN##h, \ + gen_helper_mve_##FN##w, \ + NULL, \ + }; \ + return do_2shift_scalar(s, a, fns[a->size]); \ + } + +DO_2SHIFT_SCALAR(VSHL_S_scalar, vshli_s) +DO_2SHIFT_SCALAR(VSHL_U_scalar, vshli_u) +DO_2SHIFT_SCALAR(VRSHL_S_scalar, vrshli_s) +DO_2SHIFT_SCALAR(VRSHL_U_scalar, vrshli_u) +DO_2SHIFT_SCALAR(VQSHL_S_scalar, vqshli_s) +DO_2SHIFT_SCALAR(VQSHL_U_scalar, vqshli_u) +DO_2SHIFT_SCALAR(VQRSHL_S_scalar, vqrshli_s) +DO_2SHIFT_SCALAR(VQRSHL_U_scalar, vqrshli_u) + #define DO_VSHLL(INSN, FN) \ static bool trans_##INSN(DisasContext *s, arg_2shift *a) \ { \ -- 2.20.1 All the users of the vmlaldav formats have an 'x bit in bit 12 and an 'a' bit in bit 5; move these to the format rather than specifying them in each insn pattern. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/mve.decode | 16 ++++++++-------- 1 file changed, 8 insertions(+), 8 deletions(-) diff --git a/target/arm/mve.decode b/target/arm/mve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve.decode +++ b/target/arm/mve.decode @@ -XXX,XX +XXX,XX @@ VDUP 1110 1110 1 0 10 ... 0 .... 1011 . 0 0 1 0000 @vdup size=2 &vmlaldav rdahi rdalo size qn qm x a -@vmlaldav .... .... . ... ... . ... . .... .... qm:3 . \ +@vmlaldav .... .... . ... ... . ... x:1 .... .. a:1 . qm:3 . \ qn=%qn rdahi=%rdahi rdalo=%rdalo size=%size_16 &vmlaldav -@vmlaldav_nosz .... .... . ... ... . ... . .... .... qm:3 . \ +@vmlaldav_nosz .... .... . ... ... . ... x:1 .... .. a:1 . qm:3 . \ qn=%qn rdahi=%rdahi rdalo=%rdalo size=0 &vmlaldav -VMLALDAV_S 1110 1110 1 ... ... . ... x:1 1110 . 0 a:1 0 ... 0 @vmlaldav -VMLALDAV_U 1111 1110 1 ... ... . ... x:1 1110 . 0 a:1 0 ... 0 @vmlaldav +VMLALDAV_S 1110 1110 1 ... ... . ... . 1110 . 0 . 0 ... 0 @vmlaldav +VMLALDAV_U 1111 1110 1 ... ... . ... . 1110 . 0 . 0 ... 0 @vmlaldav -VMLSLDAV 1110 1110 1 ... ... . ... x:1 1110 . 0 a:1 0 ... 1 @vmlaldav +VMLSLDAV 1110 1110 1 ... ... . ... . 1110 . 0 . 0 ... 1 @vmlaldav -VRMLALDAVH_S 1110 1110 1 ... ... 0 ... x:1 1111 . 0 a:1 0 ... 0 @vmlaldav_nosz -VRMLALDAVH_U 1111 1110 1 ... ... 0 ... x:1 1111 . 0 a:1 0 ... 0 @vmlaldav_nosz +VRMLALDAVH_S 1110 1110 1 ... ... 0 ... . 1111 . 0 . 0 ... 0 @vmlaldav_nosz +VRMLALDAVH_U 1111 1110 1 ... ... 0 ... . 1111 . 0 . 0 ... 0 @vmlaldav_nosz -VRMLSLDAVH 1111 1110 1 ... ... 0 ... x:1 1110 . 0 a:1 0 ... 1 @vmlaldav_nosz +VRMLSLDAVH 1111 1110 1 ... ... 0 ... . 1110 . 0 . 0 ... 1 @vmlaldav_nosz # Scalar operations -- 2.20.1 Implement the MVE integer min/max across vector insns VMAXV, VMINV, VMAXAV and VMINAV, which find the maximum from the vector elements and a general purpose register, and store the maximum back into the general purpose register. These insns overlap with VRMLALDAVH (they use what would be RdaHi=0b110). Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/helper-mve.h | 20 ++++++++++++ target/arm/mve.decode | 18 +++++++++-- target/arm/mve_helper.c | 66 ++++++++++++++++++++++++++++++++++++++ target/arm/translate-mve.c | 48 +++++++++++++++++++++++++++ 4 files changed, 150 insertions(+), 2 deletions(-) diff --git a/target/arm/helper-mve.h b/target/arm/helper-mve.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/helper-mve.h +++ b/target/arm/helper-mve.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_3(mve_vaddvuh, TCG_CALL_NO_WG, i32, env, ptr, i32) DEF_HELPER_FLAGS_3(mve_vaddvsw, TCG_CALL_NO_WG, i32, env, ptr, i32) DEF_HELPER_FLAGS_3(mve_vaddvuw, TCG_CALL_NO_WG, i32, env, ptr, i32) +DEF_HELPER_FLAGS_3(mve_vmaxvsb, TCG_CALL_NO_WG, i32, env, ptr, i32) +DEF_HELPER_FLAGS_3(mve_vmaxvsh, TCG_CALL_NO_WG, i32, env, ptr, i32) +DEF_HELPER_FLAGS_3(mve_vmaxvsw, TCG_CALL_NO_WG, i32, env, ptr, i32) +DEF_HELPER_FLAGS_3(mve_vmaxvub, TCG_CALL_NO_WG, i32, env, ptr, i32) +DEF_HELPER_FLAGS_3(mve_vmaxvuh, TCG_CALL_NO_WG, i32, env, ptr, i32) +DEF_HELPER_FLAGS_3(mve_vmaxvuw, TCG_CALL_NO_WG, i32, env, ptr, i32) +DEF_HELPER_FLAGS_3(mve_vmaxavb, TCG_CALL_NO_WG, i32, env, ptr, i32) +DEF_HELPER_FLAGS_3(mve_vmaxavh, TCG_CALL_NO_WG, i32, env, ptr, i32) +DEF_HELPER_FLAGS_3(mve_vmaxavw, TCG_CALL_NO_WG, i32, env, ptr, i32) + +DEF_HELPER_FLAGS_3(mve_vminvsb, TCG_CALL_NO_WG, i32, env, ptr, i32) +DEF_HELPER_FLAGS_3(mve_vminvsh, TCG_CALL_NO_WG, i32, env, ptr, i32) +DEF_HELPER_FLAGS_3(mve_vminvsw, TCG_CALL_NO_WG, i32, env, ptr, i32) +DEF_HELPER_FLAGS_3(mve_vminvub, TCG_CALL_NO_WG, i32, env, ptr, i32) +DEF_HELPER_FLAGS_3(mve_vminvuh, TCG_CALL_NO_WG, i32, env, ptr, i32) +DEF_HELPER_FLAGS_3(mve_vminvuw, TCG_CALL_NO_WG, i32, env, ptr, i32) +DEF_HELPER_FLAGS_3(mve_vminavb, TCG_CALL_NO_WG, i32, env, ptr, i32) +DEF_HELPER_FLAGS_3(mve_vminavh, TCG_CALL_NO_WG, i32, env, ptr, i32) +DEF_HELPER_FLAGS_3(mve_vminavw, TCG_CALL_NO_WG, i32, env, ptr, i32) + DEF_HELPER_FLAGS_3(mve_vaddlv_s, TCG_CALL_NO_WG, i64, env, ptr, i64) DEF_HELPER_FLAGS_3(mve_vaddlv_u, TCG_CALL_NO_WG, i64, env, ptr, i64) diff --git a/target/arm/mve.decode b/target/arm/mve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve.decode +++ b/target/arm/mve.decode @@ -XXX,XX +XXX,XX @@ &vcmp qm qn size mask &vcmp_scalar qn rm size mask &shl_scalar qda rm size +&vmaxv qm rda size @vldr_vstr ....... . . . . l:1 rn:4 ... ...... imm:7 &vldr_vstr qd=%qd u=0 # Note that both Rn and Qd are 3 bits only (no D bit) @@ -XXX,XX +XXX,XX @@ @vcmp_scalar .... .... .. size:2 qn:3 . .... .... .... rm:4 &vcmp_scalar \ mask=%mask_22_13 +@vmaxv .... .... .... size:2 .. rda:4 .... .... .... &vmaxv qm=%qm + # Vector loads and stores # Widening loads and narrowing stores: @@ -XXX,XX +XXX,XX @@ VMLALDAV_U 1111 1110 1 ... ... . ... . 1110 . 0 . 0 ... 0 @vmlaldav VMLSLDAV 1110 1110 1 ... ... . ... . 1110 . 0 . 0 ... 1 @vmlaldav -VRMLALDAVH_S 1110 1110 1 ... ... 0 ... . 1111 . 0 . 0 ... 0 @vmlaldav_nosz -VRMLALDAVH_U 1111 1110 1 ... ... 0 ... . 1111 . 0 . 0 ... 0 @vmlaldav_nosz +{ + VMAXV_S 1110 1110 1110 .. 10 .... 1111 0 0 . 0 ... 0 @vmaxv + VMINV_S 1110 1110 1110 .. 10 .... 1111 1 0 . 0 ... 0 @vmaxv + VMAXAV 1110 1110 1110 .. 00 .... 1111 0 0 . 0 ... 0 @vmaxv + VMINAV 1110 1110 1110 .. 00 .... 1111 1 0 . 0 ... 0 @vmaxv + VRMLALDAVH_S 1110 1110 1 ... ... 0 ... . 1111 . 0 . 0 ... 0 @vmlaldav_nosz +} + +{ + VMAXV_U 1111 1110 1110 .. 10 .... 1111 0 0 . 0 ... 0 @vmaxv + VMINV_U 1111 1110 1110 .. 10 .... 1111 1 0 . 0 ... 0 @vmaxv + VRMLALDAVH_U 1111 1110 1 ... ... 0 ... . 1111 . 0 . 0 ... 0 @vmlaldav_nosz +} VRMLSLDAVH 1111 1110 1 ... ... 0 ... . 1110 . 0 . 0 ... 1 @vmlaldav_nosz diff --git a/target/arm/mve_helper.c b/target/arm/mve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve_helper.c +++ b/target/arm/mve_helper.c @@ -XXX,XX +XXX,XX @@ DO_VADDV(vaddvub, 1, uint8_t) DO_VADDV(vaddvuh, 2, uint16_t) DO_VADDV(vaddvuw, 4, uint32_t) +/* + * Vector max/min across vector. Unlike VADDV, we must + * read ra as the element size, not its full width. + * We work with int64_t internally for simplicity. + */ +#define DO_VMAXMINV(OP, ESIZE, TYPE, RATYPE, FN) \ + uint32_t HELPER(glue(mve_, OP))(CPUARMState *env, void *vm, \ + uint32_t ra_in) \ + { \ + uint16_t mask = mve_element_mask(env); \ + unsigned e; \ + TYPE *m = vm; \ + int64_t ra = (RATYPE)ra_in; \ + for (e = 0; e < 16 / ESIZE; e++, mask >>= ESIZE) { \ + if (mask & 1) { \ + ra = FN(ra, m[H##ESIZE(e)]); \ + } \ + } \ + mve_advance_vpt(env); \ + return ra; \ + } \ + +#define DO_VMAXMINV_U(INSN, FN) \ + DO_VMAXMINV(INSN##b, 1, uint8_t, uint8_t, FN) \ + DO_VMAXMINV(INSN##h, 2, uint16_t, uint16_t, FN) \ + DO_VMAXMINV(INSN##w, 4, uint32_t, uint32_t, FN) +#define DO_VMAXMINV_S(INSN, FN) \ + DO_VMAXMINV(INSN##b, 1, int8_t, int8_t, FN) \ + DO_VMAXMINV(INSN##h, 2, int16_t, int16_t, FN) \ + DO_VMAXMINV(INSN##w, 4, int32_t, int32_t, FN) + +/* + * Helpers for max and min of absolute values across vector: + * note that we only take the absolute value of 'm', not 'n' + */ +static int64_t do_maxa(int64_t n, int64_t m) +{ + if (m < 0) { + m = -m; + } + return MAX(n, m); +} + +static int64_t do_mina(int64_t n, int64_t m) +{ + if (m < 0) { + m = -m; + } + return MIN(n, m); +} + +DO_VMAXMINV_S(vmaxvs, DO_MAX) +DO_VMAXMINV_U(vmaxvu, DO_MAX) +DO_VMAXMINV_S(vminvs, DO_MIN) +DO_VMAXMINV_U(vminvu, DO_MIN) +/* + * VMAXAV, VMINAV treat the general purpose input as unsigned + * and the vector elements as signed. + */ +DO_VMAXMINV(vmaxavb, 1, int8_t, uint8_t, do_maxa) +DO_VMAXMINV(vmaxavh, 2, int16_t, uint16_t, do_maxa) +DO_VMAXMINV(vmaxavw, 4, int32_t, uint32_t, do_maxa) +DO_VMAXMINV(vminavb, 1, int8_t, uint8_t, do_mina) +DO_VMAXMINV(vminavh, 2, int16_t, uint16_t, do_mina) +DO_VMAXMINV(vminavw, 4, int32_t, uint32_t, do_mina) + #define DO_VADDLV(OP, TYPE, LTYPE) \ uint64_t HELPER(glue(mve_, OP))(CPUARMState *env, void *vm, \ uint64_t ra) \ diff --git a/target/arm/translate-mve.c b/target/arm/translate-mve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-mve.c +++ b/target/arm/translate-mve.c @@ -XXX,XX +XXX,XX @@ DO_VCMP(VCMPGE, vcmpge) DO_VCMP(VCMPLT, vcmplt) DO_VCMP(VCMPGT, vcmpgt) DO_VCMP(VCMPLE, vcmple) + +static bool do_vmaxv(DisasContext *s, arg_vmaxv *a, MVEGenVADDVFn fn) +{ + /* + * MIN/MAX operations across a vector: compute the min or + * max of the initial value in a general purpose register + * and all the elements in the vector, and store it back + * into the general purpose register. + */ + TCGv_ptr qm; + TCGv_i32 rda; + + if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qm) || + !fn || a->rda == 13 || a->rda == 15) { + /* Rda cases are UNPREDICTABLE */ + return false; + } + if (!mve_eci_check(s) || !vfp_access_check(s)) { + return true; + } + + qm = mve_qreg_ptr(a->qm); + rda = load_reg(s, a->rda); + fn(rda, cpu_env, qm, rda); + store_reg(s, a->rda, rda); + tcg_temp_free_ptr(qm); + mve_update_eci(s); + return true; +} + +#define DO_VMAXV(INSN, FN) \ + static bool trans_##INSN(DisasContext *s, arg_vmaxv *a) \ + { \ + static MVEGenVADDVFn * const fns[] = { \ + gen_helper_mve_##FN##b, \ + gen_helper_mve_##FN##h, \ + gen_helper_mve_##FN##w, \ + NULL, \ + }; \ + return do_vmaxv(s, a, fns[a->size]); \ + } + +DO_VMAXV(VMAXV_S, vmaxvs) +DO_VMAXV(VMAXV_U, vmaxvu) +DO_VMAXV(VMAXAV, vmaxav) +DO_VMAXV(VMINV_S, vminvs) +DO_VMAXV(VMINV_U, vminvu) +DO_VMAXV(VMINAV, vminav) -- 2.20.1 Implement the MVE VABAV insn, which computes absolute differences between elements of two vectors and accumulates the result into a general purpose register. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/helper-mve.h | 7 +++++++ target/arm/mve.decode | 6 ++++++ target/arm/mve_helper.c | 26 +++++++++++++++++++++++ target/arm/translate-mve.c | 43 ++++++++++++++++++++++++++++++++++++++ 4 files changed, 82 insertions(+) diff --git a/target/arm/helper-mve.h b/target/arm/helper-mve.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/helper-mve.h +++ b/target/arm/helper-mve.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_3(mve_vminavw, TCG_CALL_NO_WG, i32, env, ptr, i32) DEF_HELPER_FLAGS_3(mve_vaddlv_s, TCG_CALL_NO_WG, i64, env, ptr, i64) DEF_HELPER_FLAGS_3(mve_vaddlv_u, TCG_CALL_NO_WG, i64, env, ptr, i64) +DEF_HELPER_FLAGS_4(mve_vabavsb, TCG_CALL_NO_WG, i32, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vabavsh, TCG_CALL_NO_WG, i32, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vabavsw, TCG_CALL_NO_WG, i32, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vabavub, TCG_CALL_NO_WG, i32, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vabavuh, TCG_CALL_NO_WG, i32, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vabavuw, TCG_CALL_NO_WG, i32, env, ptr, ptr, i32) + DEF_HELPER_FLAGS_3(mve_vmovi, TCG_CALL_NO_WG, void, env, ptr, i64) DEF_HELPER_FLAGS_3(mve_vandi, TCG_CALL_NO_WG, void, env, ptr, i64) DEF_HELPER_FLAGS_3(mve_vorri, TCG_CALL_NO_WG, void, env, ptr, i64) diff --git a/target/arm/mve.decode b/target/arm/mve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve.decode +++ b/target/arm/mve.decode @@ -XXX,XX +XXX,XX @@ &vcmp_scalar qn rm size mask &shl_scalar qda rm size &vmaxv qm rda size +&vabav qn qm rda size @vldr_vstr ....... . . . . l:1 rn:4 ... ...... imm:7 &vldr_vstr qd=%qd u=0 # Note that both Rn and Qd are 3 bits only (no D bit) @@ -XXX,XX +XXX,XX @@ VMLAS 111- 1110 0 . .. ... 1 ... 1 1110 . 100 .... @2scalar rdahi=%rdahi rdalo=%rdalo } +@vabav .... .... .. size:2 .... rda:4 .... .... .... &vabav qn=%qn qm=%qm + +VABAV_S 111 0 1110 10 .. ... 0 .... 1111 . 0 . 0 ... 1 @vabav +VABAV_U 111 1 1110 10 .. ... 0 .... 1111 . 0 . 0 ... 1 @vabav + # Logical immediate operations (1 reg and modified-immediate) # The cmode/op bits here decode VORR/VBIC/VMOV/VMVN, but diff --git a/target/arm/mve_helper.c b/target/arm/mve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve_helper.c +++ b/target/arm/mve_helper.c @@ -XXX,XX +XXX,XX @@ DO_VMAXMINV(vminavb, 1, int8_t, uint8_t, do_mina) DO_VMAXMINV(vminavh, 2, int16_t, uint16_t, do_mina) DO_VMAXMINV(vminavw, 4, int32_t, uint32_t, do_mina) +#define DO_VABAV(OP, ESIZE, TYPE) \ + uint32_t HELPER(glue(mve_, OP))(CPUARMState *env, void *vn, \ + void *vm, uint32_t ra) \ + { \ + uint16_t mask = mve_element_mask(env); \ + unsigned e; \ + TYPE *m = vm, *n = vn; \ + for (e = 0; e < 16 / ESIZE; e++, mask >>= ESIZE) { \ + if (mask & 1) { \ + int64_t n0 = n[H##ESIZE(e)]; \ + int64_t m0 = m[H##ESIZE(e)]; \ + uint32_t r = n0 >= m0 ? (n0 - m0) : (m0 - n0); \ + ra += r; \ + } \ + } \ + mve_advance_vpt(env); \ + return ra; \ + } + +DO_VABAV(vabavsb, 1, int8_t) +DO_VABAV(vabavsh, 2, int16_t) +DO_VABAV(vabavsw, 4, int32_t) +DO_VABAV(vabavub, 1, uint8_t) +DO_VABAV(vabavuh, 2, uint16_t) +DO_VABAV(vabavuw, 4, uint32_t) + #define DO_VADDLV(OP, TYPE, LTYPE) \ uint64_t HELPER(glue(mve_, OP))(CPUARMState *env, void *vm, \ uint64_t ra) \ diff --git a/target/arm/translate-mve.c b/target/arm/translate-mve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-mve.c +++ b/target/arm/translate-mve.c @@ -XXX,XX +XXX,XX @@ typedef void MVEGenVIDUPFn(TCGv_i32, TCGv_ptr, TCGv_ptr, TCGv_i32, TCGv_i32); typedef void MVEGenVIWDUPFn(TCGv_i32, TCGv_ptr, TCGv_ptr, TCGv_i32, TCGv_i32, TCGv_i32); typedef void MVEGenCmpFn(TCGv_ptr, TCGv_ptr, TCGv_ptr); typedef void MVEGenScalarCmpFn(TCGv_ptr, TCGv_ptr, TCGv_i32); +typedef void MVEGenVABAVFn(TCGv_i32, TCGv_ptr, TCGv_ptr, TCGv_ptr, TCGv_i32); /* Return the offset of a Qn register (same semantics as aa32_vfp_qreg()) */ static inline long mve_qreg_offset(unsigned reg) @@ -XXX,XX +XXX,XX @@ DO_VMAXV(VMAXAV, vmaxav) DO_VMAXV(VMINV_S, vminvs) DO_VMAXV(VMINV_U, vminvu) DO_VMAXV(VMINAV, vminav) + +static bool do_vabav(DisasContext *s, arg_vabav *a, MVEGenVABAVFn *fn) +{ + /* Absolute difference accumulated across vector */ + TCGv_ptr qn, qm; + TCGv_i32 rda; + + if (!dc_isar_feature(aa32_mve, s) || + !mve_check_qreg_bank(s, a->qm | a->qn) || + !fn || a->rda == 13 || a->rda == 15) { + /* Rda cases are UNPREDICTABLE */ + return false; + } + if (!mve_eci_check(s) || !vfp_access_check(s)) { + return true; + } + + qm = mve_qreg_ptr(a->qm); + qn = mve_qreg_ptr(a->qn); + rda = load_reg(s, a->rda); + fn(rda, cpu_env, qn, qm, rda); + store_reg(s, a->rda, rda); + tcg_temp_free_ptr(qm); + tcg_temp_free_ptr(qn); + mve_update_eci(s); + return true; +} + +#define DO_VABAV(INSN, FN) \ + static bool trans_##INSN(DisasContext *s, arg_vabav *a) \ + { \ + static MVEGenVABAVFn * const fns[] = { \ + gen_helper_mve_##FN##b, \ + gen_helper_mve_##FN##h, \ + gen_helper_mve_##FN##w, \ + NULL, \ + }; \ + return do_vabav(s, a, fns[a->size]); \ + } + +DO_VABAV(VABAV_S, vabavs) +DO_VABAV(VABAV_U, vabavu) -- 2.20.1 Implement the MVE narrowing move insns VMOVN, VQMOVN and VQMOVUN. These take a double-width input, narrow it (possibly saturating) and store the result to either the top or bottom half of the output element. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/helper-mve.h | 20 ++++++++++ target/arm/mve.decode | 12 ++++++ target/arm/mve_helper.c | 78 ++++++++++++++++++++++++++++++++++++++ target/arm/translate-mve.c | 22 +++++++++++ 4 files changed, 132 insertions(+) diff --git a/target/arm/helper-mve.h b/target/arm/helper-mve.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/helper-mve.h +++ b/target/arm/helper-mve.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_3(mve_vnegw, TCG_CALL_NO_WG, void, env, ptr, ptr) DEF_HELPER_FLAGS_3(mve_vfnegh, TCG_CALL_NO_WG, void, env, ptr, ptr) DEF_HELPER_FLAGS_3(mve_vfnegs, TCG_CALL_NO_WG, void, env, ptr, ptr) +DEF_HELPER_FLAGS_3(mve_vmovnbb, TCG_CALL_NO_WG, void, env, ptr, ptr) +DEF_HELPER_FLAGS_3(mve_vmovnbh, TCG_CALL_NO_WG, void, env, ptr, ptr) +DEF_HELPER_FLAGS_3(mve_vmovntb, TCG_CALL_NO_WG, void, env, ptr, ptr) +DEF_HELPER_FLAGS_3(mve_vmovnth, TCG_CALL_NO_WG, void, env, ptr, ptr) + +DEF_HELPER_FLAGS_3(mve_vqmovunbb, TCG_CALL_NO_WG, void, env, ptr, ptr) +DEF_HELPER_FLAGS_3(mve_vqmovunbh, TCG_CALL_NO_WG, void, env, ptr, ptr) +DEF_HELPER_FLAGS_3(mve_vqmovuntb, TCG_CALL_NO_WG, void, env, ptr, ptr) +DEF_HELPER_FLAGS_3(mve_vqmovunth, TCG_CALL_NO_WG, void, env, ptr, ptr) + +DEF_HELPER_FLAGS_3(mve_vqmovnbsb, TCG_CALL_NO_WG, void, env, ptr, ptr) +DEF_HELPER_FLAGS_3(mve_vqmovnbsh, TCG_CALL_NO_WG, void, env, ptr, ptr) +DEF_HELPER_FLAGS_3(mve_vqmovntsb, TCG_CALL_NO_WG, void, env, ptr, ptr) +DEF_HELPER_FLAGS_3(mve_vqmovntsh, TCG_CALL_NO_WG, void, env, ptr, ptr) + +DEF_HELPER_FLAGS_3(mve_vqmovnbub, TCG_CALL_NO_WG, void, env, ptr, ptr) +DEF_HELPER_FLAGS_3(mve_vqmovnbuh, TCG_CALL_NO_WG, void, env, ptr, ptr) +DEF_HELPER_FLAGS_3(mve_vqmovntub, TCG_CALL_NO_WG, void, env, ptr, ptr) +DEF_HELPER_FLAGS_3(mve_vqmovntuh, TCG_CALL_NO_WG, void, env, ptr, ptr) + DEF_HELPER_FLAGS_4(mve_vand, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr) DEF_HELPER_FLAGS_4(mve_vbic, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr) DEF_HELPER_FLAGS_4(mve_vorr, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr) diff --git a/target/arm/mve.decode b/target/arm/mve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve.decode +++ b/target/arm/mve.decode @@ -XXX,XX +XXX,XX @@ VMUL 1110 1111 0 . .. ... 0 ... 0 1001 . 1 . 1 ... 0 @2op VSHLL_BS 111 0 1110 0 . 11 .. 01 ... 0 1110 0 0 . 0 ... 1 @2_shll_esize_b VSHLL_BS 111 0 1110 0 . 11 .. 01 ... 0 1110 0 0 . 0 ... 1 @2_shll_esize_h + VQMOVUNB 111 0 1110 0 . 11 .. 01 ... 0 1110 1 0 . 0 ... 1 @1op + VQMOVN_BS 111 0 1110 0 . 11 .. 11 ... 0 1110 0 0 . 0 ... 1 @1op + VMULH_S 111 0 1110 0 . .. ...1 ... 0 1110 . 0 . 0 ... 1 @2op } @@ -XXX,XX +XXX,XX @@ VMUL 1110 1111 0 . .. ... 0 ... 0 1001 . 1 . 1 ... 0 @2op VSHLL_BU 111 1 1110 0 . 11 .. 01 ... 0 1110 0 0 . 0 ... 1 @2_shll_esize_b VSHLL_BU 111 1 1110 0 . 11 .. 01 ... 0 1110 0 0 . 0 ... 1 @2_shll_esize_h + VMOVNB 111 1 1110 0 . 11 .. 01 ... 0 1110 1 0 . 0 ... 1 @1op + VQMOVN_BU 111 1 1110 0 . 11 .. 11 ... 0 1110 0 0 . 0 ... 1 @1op + VMULH_U 111 1 1110 0 . .. ...1 ... 0 1110 . 0 . 0 ... 1 @2op } @@ -XXX,XX +XXX,XX @@ VMUL 1110 1111 0 . .. ... 0 ... 0 1001 . 1 . 1 ... 0 @2op VSHLL_TS 111 0 1110 0 . 11 .. 01 ... 1 1110 0 0 . 0 ... 1 @2_shll_esize_b VSHLL_TS 111 0 1110 0 . 11 .. 01 ... 1 1110 0 0 . 0 ... 1 @2_shll_esize_h + VQMOVUNT 111 0 1110 0 . 11 .. 01 ... 1 1110 1 0 . 0 ... 1 @1op + VQMOVN_TS 111 0 1110 0 . 11 .. 11 ... 1 1110 0 0 . 0 ... 1 @1op + VRMULH_S 111 0 1110 0 . .. ...1 ... 1 1110 . 0 . 0 ... 1 @2op } @@ -XXX,XX +XXX,XX @@ VMUL 1110 1111 0 . .. ... 0 ... 0 1001 . 1 . 1 ... 0 @2op VSHLL_TU 111 1 1110 0 . 11 .. 01 ... 1 1110 0 0 . 0 ... 1 @2_shll_esize_b VSHLL_TU 111 1 1110 0 . 11 .. 01 ... 1 1110 0 0 . 0 ... 1 @2_shll_esize_h + VMOVNT 111 1 1110 0 . 11 .. 01 ... 1 1110 1 0 . 0 ... 1 @1op + VQMOVN_TU 111 1 1110 0 . 11 .. 11 ... 1 1110 0 0 . 0 ... 1 @1op + VRMULH_U 111 1 1110 0 . .. ...1 ... 1 1110 . 0 . 0 ... 1 @2op } diff --git a/target/arm/mve_helper.c b/target/arm/mve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve_helper.c +++ b/target/arm/mve_helper.c @@ -XXX,XX +XXX,XX @@ DO_VSHRN_SAT_UH(vqrshrnb_uh, vqrshrnt_uh, DO_RSHRN_UH) DO_VSHRN_SAT_SB(vqrshrunbb, vqrshruntb, DO_RSHRUN_B) DO_VSHRN_SAT_SH(vqrshrunbh, vqrshrunth, DO_RSHRUN_H) +#define DO_VMOVN(OP, TOP, ESIZE, TYPE, LESIZE, LTYPE) \ + void HELPER(mve_##OP)(CPUARMState *env, void *vd, void *vm) \ + { \ + LTYPE *m = vm; \ + TYPE *d = vd; \ + uint16_t mask = mve_element_mask(env); \ + unsigned le; \ + mask >>= ESIZE * TOP; \ + for (le = 0; le < 16 / LESIZE; le++, mask >>= LESIZE) { \ + mergemask(&d[H##ESIZE(le * 2 + TOP)], \ + m[H##LESIZE(le)], mask); \ + } \ + mve_advance_vpt(env); \ + } + +DO_VMOVN(vmovnbb, false, 1, uint8_t, 2, uint16_t) +DO_VMOVN(vmovnbh, false, 2, uint16_t, 4, uint32_t) +DO_VMOVN(vmovntb, true, 1, uint8_t, 2, uint16_t) +DO_VMOVN(vmovnth, true, 2, uint16_t, 4, uint32_t) + +#define DO_VMOVN_SAT(OP, TOP, ESIZE, TYPE, LESIZE, LTYPE, FN) \ + void HELPER(mve_##OP)(CPUARMState *env, void *vd, void *vm) \ + { \ + LTYPE *m = vm; \ + TYPE *d = vd; \ + uint16_t mask = mve_element_mask(env); \ + bool qc = false; \ + unsigned le; \ + mask >>= ESIZE * TOP; \ + for (le = 0; le < 16 / LESIZE; le++, mask >>= LESIZE) { \ + bool sat = false; \ + TYPE r = FN(m[H##LESIZE(le)], &sat); \ + mergemask(&d[H##ESIZE(le * 2 + TOP)], r, mask); \ + qc |= sat & mask & 1; \ + } \ + if (qc) { \ + env->vfp.qc[0] = qc; \ + } \ + mve_advance_vpt(env); \ + } + +#define DO_VMOVN_SAT_UB(BOP, TOP, FN) \ + DO_VMOVN_SAT(BOP, false, 1, uint8_t, 2, uint16_t, FN) \ + DO_VMOVN_SAT(TOP, true, 1, uint8_t, 2, uint16_t, FN) + +#define DO_VMOVN_SAT_UH(BOP, TOP, FN) \ + DO_VMOVN_SAT(BOP, false, 2, uint16_t, 4, uint32_t, FN) \ + DO_VMOVN_SAT(TOP, true, 2, uint16_t, 4, uint32_t, FN) + +#define DO_VMOVN_SAT_SB(BOP, TOP, FN) \ + DO_VMOVN_SAT(BOP, false, 1, int8_t, 2, int16_t, FN) \ + DO_VMOVN_SAT(TOP, true, 1, int8_t, 2, int16_t, FN) + +#define DO_VMOVN_SAT_SH(BOP, TOP, FN) \ + DO_VMOVN_SAT(BOP, false, 2, int16_t, 4, int32_t, FN) \ + DO_VMOVN_SAT(TOP, true, 2, int16_t, 4, int32_t, FN) + +#define DO_VQMOVN_SB(N, SATP) \ + do_sat_bhs((int64_t)(N), INT8_MIN, INT8_MAX, SATP) +#define DO_VQMOVN_UB(N, SATP) \ + do_sat_bhs((uint64_t)(N), 0, UINT8_MAX, SATP) +#define DO_VQMOVUN_B(N, SATP) \ + do_sat_bhs((int64_t)(N), 0, UINT8_MAX, SATP) + +#define DO_VQMOVN_SH(N, SATP) \ + do_sat_bhs((int64_t)(N), INT16_MIN, INT16_MAX, SATP) +#define DO_VQMOVN_UH(N, SATP) \ + do_sat_bhs((uint64_t)(N), 0, UINT16_MAX, SATP) +#define DO_VQMOVUN_H(N, SATP) \ + do_sat_bhs((int64_t)(N), 0, UINT16_MAX, SATP) + +DO_VMOVN_SAT_SB(vqmovnbsb, vqmovntsb, DO_VQMOVN_SB) +DO_VMOVN_SAT_SH(vqmovnbsh, vqmovntsh, DO_VQMOVN_SH) +DO_VMOVN_SAT_UB(vqmovnbub, vqmovntub, DO_VQMOVN_UB) +DO_VMOVN_SAT_UH(vqmovnbuh, vqmovntuh, DO_VQMOVN_UH) +DO_VMOVN_SAT_SB(vqmovunbb, vqmovuntb, DO_VQMOVUN_B) +DO_VMOVN_SAT_SH(vqmovunbh, vqmovunth, DO_VQMOVUN_H) + uint32_t HELPER(mve_vshlc)(CPUARMState *env, void *vd, uint32_t rdm, uint32_t shift) { diff --git a/target/arm/translate-mve.c b/target/arm/translate-mve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-mve.c +++ b/target/arm/translate-mve.c @@ -XXX,XX +XXX,XX @@ DO_1OP(VCLS, vcls) DO_1OP(VABS, vabs) DO_1OP(VNEG, vneg) +/* Narrowing moves: only size 0 and 1 are valid */ +#define DO_VMOVN(INSN, FN) \ + static bool trans_##INSN(DisasContext *s, arg_1op *a) \ + { \ + static MVEGenOneOpFn * const fns[] = { \ + gen_helper_mve_##FN##b, \ + gen_helper_mve_##FN##h, \ + NULL, \ + NULL, \ + }; \ + return do_1op(s, a, fns[a->size]); \ + } + +DO_VMOVN(VMOVNB, vmovnb) +DO_VMOVN(VMOVNT, vmovnt) +DO_VMOVN(VQMOVUNB, vqmovunb) +DO_VMOVN(VQMOVUNT, vqmovunt) +DO_VMOVN(VQMOVN_BS, vqmovnbs) +DO_VMOVN(VQMOVN_TS, vqmovnts) +DO_VMOVN(VQMOVN_BU, vqmovnbu) +DO_VMOVN(VQMOVN_TU, vqmovntu) + static bool trans_VREV16(DisasContext *s, arg_1op *a) { static MVEGenOneOpFn * const fns[] = { -- 2.20.1 The MVEGenDualAccOpFn is a bit misnamed, since it is used for the "long dual accumulate" operations that use a 64-bit accumulator. Rename it to MVEGenLongDualAccOpFn so we can use the former name for the 32-bit accumulator insns. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/translate-mve.c | 16 ++++++++-------- 1 file changed, 8 insertions(+), 8 deletions(-) diff --git a/target/arm/translate-mve.c b/target/arm/translate-mve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-mve.c +++ b/target/arm/translate-mve.c @@ -XXX,XX +XXX,XX @@ typedef void MVEGenOneOpFn(TCGv_ptr, TCGv_ptr, TCGv_ptr); typedef void MVEGenTwoOpFn(TCGv_ptr, TCGv_ptr, TCGv_ptr, TCGv_ptr); typedef void MVEGenTwoOpScalarFn(TCGv_ptr, TCGv_ptr, TCGv_ptr, TCGv_i32); typedef void MVEGenTwoOpShiftFn(TCGv_ptr, TCGv_ptr, TCGv_ptr, TCGv_i32); -typedef void MVEGenDualAccOpFn(TCGv_i64, TCGv_ptr, TCGv_ptr, TCGv_ptr, TCGv_i64); +typedef void MVEGenLongDualAccOpFn(TCGv_i64, TCGv_ptr, TCGv_ptr, TCGv_ptr, TCGv_i64); typedef void MVEGenVADDVFn(TCGv_i32, TCGv_ptr, TCGv_ptr, TCGv_i32); typedef void MVEGenOneOpImmFn(TCGv_ptr, TCGv_ptr, TCGv_i64); typedef void MVEGenVIDUPFn(TCGv_i32, TCGv_ptr, TCGv_ptr, TCGv_i32, TCGv_i32); @@ -XXX,XX +XXX,XX @@ static bool trans_VQDMULLT_scalar(DisasContext *s, arg_2scalar *a) } static bool do_long_dual_acc(DisasContext *s, arg_vmlaldav *a, - MVEGenDualAccOpFn *fn) + MVEGenLongDualAccOpFn *fn) { TCGv_ptr qn, qm; TCGv_i64 rda; @@ -XXX,XX +XXX,XX @@ static bool do_long_dual_acc(DisasContext *s, arg_vmlaldav *a, static bool trans_VMLALDAV_S(DisasContext *s, arg_vmlaldav *a) { - static MVEGenDualAccOpFn * const fns[4][2] = { + static MVEGenLongDualAccOpFn * const fns[4][2] = { { NULL, NULL }, { gen_helper_mve_vmlaldavsh, gen_helper_mve_vmlaldavxsh }, { gen_helper_mve_vmlaldavsw, gen_helper_mve_vmlaldavxsw }, @@ -XXX,XX +XXX,XX @@ static bool trans_VMLALDAV_S(DisasContext *s, arg_vmlaldav *a) static bool trans_VMLALDAV_U(DisasContext *s, arg_vmlaldav *a) { - static MVEGenDualAccOpFn * const fns[4][2] = { + static MVEGenLongDualAccOpFn * const fns[4][2] = { { NULL, NULL }, { gen_helper_mve_vmlaldavuh, NULL }, { gen_helper_mve_vmlaldavuw, NULL }, @@ -XXX,XX +XXX,XX @@ static bool trans_VMLALDAV_U(DisasContext *s, arg_vmlaldav *a) static bool trans_VMLSLDAV(DisasContext *s, arg_vmlaldav *a) { - static MVEGenDualAccOpFn * const fns[4][2] = { + static MVEGenLongDualAccOpFn * const fns[4][2] = { { NULL, NULL }, { gen_helper_mve_vmlsldavsh, gen_helper_mve_vmlsldavxsh }, { gen_helper_mve_vmlsldavsw, gen_helper_mve_vmlsldavxsw }, @@ -XXX,XX +XXX,XX @@ static bool trans_VMLSLDAV(DisasContext *s, arg_vmlaldav *a) static bool trans_VRMLALDAVH_S(DisasContext *s, arg_vmlaldav *a) { - static MVEGenDualAccOpFn * const fns[] = { + static MVEGenLongDualAccOpFn * const fns[] = { gen_helper_mve_vrmlaldavhsw, gen_helper_mve_vrmlaldavhxsw, }; return do_long_dual_acc(s, a, fns[a->x]); @@ -XXX,XX +XXX,XX @@ static bool trans_VRMLALDAVH_S(DisasContext *s, arg_vmlaldav *a) static bool trans_VRMLALDAVH_U(DisasContext *s, arg_vmlaldav *a) { - static MVEGenDualAccOpFn * const fns[] = { + static MVEGenLongDualAccOpFn * const fns[] = { gen_helper_mve_vrmlaldavhuw, NULL, }; return do_long_dual_acc(s, a, fns[a->x]); @@ -XXX,XX +XXX,XX @@ static bool trans_VRMLALDAVH_U(DisasContext *s, arg_vmlaldav *a) static bool trans_VRMLSLDAVH(DisasContext *s, arg_vmlaldav *a) { - static MVEGenDualAccOpFn * const fns[] = { + static MVEGenLongDualAccOpFn * const fns[] = { gen_helper_mve_vrmlsldavhsw, gen_helper_mve_vrmlsldavhxsw, }; return do_long_dual_acc(s, a, fns[a->x]); -- 2.20.1 Implement the MVE VMLADAV and VMLSLDAV insns. Like the VMLALDAV and VMLSLDAV insns already implemented, these accumulate multiplied vector elements; but they accumulate a 32-bit result rather than a 64-bit one. Note that these encodings overlap with what would be RdaHi=0b111 for VMLALDAV, VMLSLDAV, VRMLALDAVH and VRMLSLDAVH. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/helper-mve.h | 17 ++++++++++ target/arm/mve.decode | 33 +++++++++++++++++--- target/arm/mve_helper.c | 41 ++++++++++++++++++++++++ target/arm/translate-mve.c | 64 ++++++++++++++++++++++++++++++++++++++ 4 files changed, 150 insertions(+), 5 deletions(-) diff --git a/target/arm/helper-mve.h b/target/arm/helper-mve.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/helper-mve.h +++ b/target/arm/helper-mve.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_4(mve_vrmlaldavhuw, TCG_CALL_NO_WG, i64, env, ptr, ptr, i64) DEF_HELPER_FLAGS_4(mve_vrmlsldavhsw, TCG_CALL_NO_WG, i64, env, ptr, ptr, i64) DEF_HELPER_FLAGS_4(mve_vrmlsldavhxsw, TCG_CALL_NO_WG, i64, env, ptr, ptr, i64) +DEF_HELPER_FLAGS_4(mve_vmladavsb, TCG_CALL_NO_WG, i32, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vmladavsh, TCG_CALL_NO_WG, i32, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vmladavsw, TCG_CALL_NO_WG, i32, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vmladavub, TCG_CALL_NO_WG, i32, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vmladavuh, TCG_CALL_NO_WG, i32, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vmladavuw, TCG_CALL_NO_WG, i32, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vmlsdavb, TCG_CALL_NO_WG, i32, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vmlsdavh, TCG_CALL_NO_WG, i32, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vmlsdavw, TCG_CALL_NO_WG, i32, env, ptr, ptr, i32) + +DEF_HELPER_FLAGS_4(mve_vmladavsxb, TCG_CALL_NO_WG, i32, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vmladavsxh, TCG_CALL_NO_WG, i32, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vmladavsxw, TCG_CALL_NO_WG, i32, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vmlsdavxb, TCG_CALL_NO_WG, i32, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vmlsdavxh, TCG_CALL_NO_WG, i32, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vmlsdavxw, TCG_CALL_NO_WG, i32, env, ptr, ptr, i32) + DEF_HELPER_FLAGS_3(mve_vaddvsb, TCG_CALL_NO_WG, i32, env, ptr, i32) DEF_HELPER_FLAGS_3(mve_vaddvub, TCG_CALL_NO_WG, i32, env, ptr, i32) DEF_HELPER_FLAGS_3(mve_vaddvsh, TCG_CALL_NO_WG, i32, env, ptr, i32) diff --git a/target/arm/mve.decode b/target/arm/mve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve.decode +++ b/target/arm/mve.decode @@ -XXX,XX +XXX,XX @@ VDUP 1110 1110 1 0 10 ... 0 .... 1011 . 0 0 1 0000 @vdup size=2 %size_16 16:1 !function=plus_1 &vmlaldav rdahi rdalo size qn qm x a +&vmladav rda size qn qm x a @vmlaldav .... .... . ... ... . ... x:1 .... .. a:1 . qm:3 . \ qn=%qn rdahi=%rdahi rdalo=%rdalo size=%size_16 &vmlaldav @vmlaldav_nosz .... .... . ... ... . ... x:1 .... .. a:1 . qm:3 . \ qn=%qn rdahi=%rdahi rdalo=%rdalo size=0 &vmlaldav -VMLALDAV_S 1110 1110 1 ... ... . ... . 1110 . 0 . 0 ... 0 @vmlaldav -VMLALDAV_U 1111 1110 1 ... ... . ... . 1110 . 0 . 0 ... 0 @vmlaldav +@vmladav .... .... .... ... . ... x:1 .... . . a:1 . qm:3 . \ + qn=%qn rda=%rdalo size=%size_16 &vmladav +@vmladav_nosz .... .... .... ... . ... x:1 .... . . a:1 . qm:3 . \ + qn=%qn rda=%rdalo size=0 &vmladav -VMLSLDAV 1110 1110 1 ... ... . ... . 1110 . 0 . 0 ... 1 @vmlaldav +{ + VMLADAV_S 1110 1110 1111 ... . ... . 1110 . 0 . 0 ... 0 @vmladav + VMLALDAV_S 1110 1110 1 ... ... . ... . 1110 . 0 . 0 ... 0 @vmlaldav +} +{ + VMLADAV_U 1111 1110 1111 ... . ... . 1110 . 0 . 0 ... 0 @vmladav + VMLALDAV_U 1111 1110 1 ... ... . ... . 1110 . 0 . 0 ... 0 @vmlaldav +} + +{ + VMLSDAV 1110 1110 1111 ... . ... . 1110 . 0 . 0 ... 1 @vmladav + VMLSLDAV 1110 1110 1 ... ... . ... . 1110 . 0 . 0 ... 1 @vmlaldav +} + +{ + VMLSDAV 1111 1110 1111 ... 0 ... . 1110 . 0 . 0 ... 1 @vmladav_nosz + VRMLSLDAVH 1111 1110 1 ... ... 0 ... . 1110 . 0 . 0 ... 1 @vmlaldav_nosz +} + +VMLADAV_S 1110 1110 1111 ... 0 ... . 1111 . 0 . 0 ... 1 @vmladav_nosz +VMLADAV_U 1111 1110 1111 ... 0 ... . 1111 . 0 . 0 ... 1 @vmladav_nosz { VMAXV_S 1110 1110 1110 .. 10 .... 1111 0 0 . 0 ... 0 @vmaxv VMINV_S 1110 1110 1110 .. 10 .... 1111 1 0 . 0 ... 0 @vmaxv VMAXAV 1110 1110 1110 .. 00 .... 1111 0 0 . 0 ... 0 @vmaxv VMINAV 1110 1110 1110 .. 00 .... 1111 1 0 . 0 ... 0 @vmaxv + VMLADAV_S 1110 1110 1111 ... 0 ... . 1111 . 0 . 0 ... 0 @vmladav_nosz VRMLALDAVH_S 1110 1110 1 ... ... 0 ... . 1111 . 0 . 0 ... 0 @vmlaldav_nosz } { VMAXV_U 1111 1110 1110 .. 10 .... 1111 0 0 . 0 ... 0 @vmaxv VMINV_U 1111 1110 1110 .. 10 .... 1111 1 0 . 0 ... 0 @vmaxv + VMLADAV_U 1111 1110 1111 ... 0 ... . 1111 . 0 . 0 ... 0 @vmladav_nosz VRMLALDAVH_U 1111 1110 1 ... ... 0 ... . 1111 . 0 . 0 ... 0 @vmlaldav_nosz } -VRMLSLDAVH 1111 1110 1 ... ... 0 ... . 1110 . 0 . 0 ... 1 @vmlaldav_nosz - # Scalar operations VADD_scalar 1110 1110 0 . .. ... 1 ... 0 1111 . 100 .... @2scalar diff --git a/target/arm/mve_helper.c b/target/arm/mve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve_helper.c +++ b/target/arm/mve_helper.c @@ -XXX,XX +XXX,XX @@ DO_LDAV(vmlsldavxsh, 2, int16_t, true, +=, -=) DO_LDAV(vmlsldavsw, 4, int32_t, false, +=, -=) DO_LDAV(vmlsldavxsw, 4, int32_t, true, +=, -=) +/* + * Multiply add dual accumulate ops + */ +#define DO_DAV(OP, ESIZE, TYPE, XCHG, EVENACC, ODDACC) \ + uint32_t HELPER(glue(mve_, OP))(CPUARMState *env, void *vn, \ + void *vm, uint32_t a) \ + { \ + uint16_t mask = mve_element_mask(env); \ + unsigned e; \ + TYPE *n = vn, *m = vm; \ + for (e = 0; e < 16 / ESIZE; e++, mask >>= ESIZE) { \ + if (mask & 1) { \ + if (e & 1) { \ + a ODDACC \ + n[H##ESIZE(e - 1 * XCHG)] * m[H##ESIZE(e)]; \ + } else { \ + a EVENACC \ + n[H##ESIZE(e + 1 * XCHG)] * m[H##ESIZE(e)]; \ + } \ + } \ + } \ + mve_advance_vpt(env); \ + return a; \ + } + +#define DO_DAV_S(INSN, XCHG, EVENACC, ODDACC) \ + DO_DAV(INSN##b, 1, int8_t, XCHG, EVENACC, ODDACC) \ + DO_DAV(INSN##h, 2, int16_t, XCHG, EVENACC, ODDACC) \ + DO_DAV(INSN##w, 4, int32_t, XCHG, EVENACC, ODDACC) + +#define DO_DAV_U(INSN, XCHG, EVENACC, ODDACC) \ + DO_DAV(INSN##b, 1, uint8_t, XCHG, EVENACC, ODDACC) \ + DO_DAV(INSN##h, 2, uint16_t, XCHG, EVENACC, ODDACC) \ + DO_DAV(INSN##w, 4, uint32_t, XCHG, EVENACC, ODDACC) + +DO_DAV_S(vmladavs, false, +=, +=) +DO_DAV_U(vmladavu, false, +=, +=) +DO_DAV_S(vmlsdav, false, +=, -=) +DO_DAV_S(vmladavsx, true, +=, +=) +DO_DAV_S(vmlsdavx, true, +=, -=) + /* * Rounding multiply add long dual accumulate high. In the pseudocode * this is implemented with a 72-bit internal accumulator value of which diff --git a/target/arm/translate-mve.c b/target/arm/translate-mve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-mve.c +++ b/target/arm/translate-mve.c @@ -XXX,XX +XXX,XX @@ typedef void MVEGenVIWDUPFn(TCGv_i32, TCGv_ptr, TCGv_ptr, TCGv_i32, TCGv_i32, TC typedef void MVEGenCmpFn(TCGv_ptr, TCGv_ptr, TCGv_ptr); typedef void MVEGenScalarCmpFn(TCGv_ptr, TCGv_ptr, TCGv_i32); typedef void MVEGenVABAVFn(TCGv_i32, TCGv_ptr, TCGv_ptr, TCGv_ptr, TCGv_i32); +typedef void MVEGenDualAccOpFn(TCGv_i32, TCGv_ptr, TCGv_ptr, TCGv_ptr, TCGv_i32); /* Return the offset of a Qn register (same semantics as aa32_vfp_qreg()) */ static inline long mve_qreg_offset(unsigned reg) @@ -XXX,XX +XXX,XX @@ static bool trans_VRMLSLDAVH(DisasContext *s, arg_vmlaldav *a) return do_long_dual_acc(s, a, fns[a->x]); } +static bool do_dual_acc(DisasContext *s, arg_vmladav *a, MVEGenDualAccOpFn *fn) +{ + TCGv_ptr qn, qm; + TCGv_i32 rda; + + if (!dc_isar_feature(aa32_mve, s) || + !mve_check_qreg_bank(s, a->qn) || + !fn) { + return false; + } + if (!mve_eci_check(s) || !vfp_access_check(s)) { + return true; + } + + qn = mve_qreg_ptr(a->qn); + qm = mve_qreg_ptr(a->qm); + + /* + * This insn is subject to beat-wise execution. Partial execution + * of an A=0 (no-accumulate) insn which does not execute the first + * beat must start with the current rda value, not 0. + */ + if (a->a || mve_skip_first_beat(s)) { + rda = load_reg(s, a->rda); + } else { + rda = tcg_const_i32(0); + } + + fn(rda, cpu_env, qn, qm, rda); + store_reg(s, a->rda, rda); + tcg_temp_free_ptr(qn); + tcg_temp_free_ptr(qm); + + mve_update_eci(s); + return true; +} + +#define DO_DUAL_ACC(INSN, FN) \ + static bool trans_##INSN(DisasContext *s, arg_vmladav *a) \ + { \ + static MVEGenDualAccOpFn * const fns[4][2] = { \ + { gen_helper_mve_##FN##b, gen_helper_mve_##FN##xb }, \ + { gen_helper_mve_##FN##h, gen_helper_mve_##FN##xh }, \ + { gen_helper_mve_##FN##w, gen_helper_mve_##FN##xw }, \ + { NULL, NULL }, \ + }; \ + return do_dual_acc(s, a, fns[a->size][a->x]); \ + } + +DO_DUAL_ACC(VMLADAV_S, vmladavs) +DO_DUAL_ACC(VMLSDAV, vmlsdav) + +static bool trans_VMLADAV_U(DisasContext *s, arg_vmladav *a) +{ + static MVEGenDualAccOpFn * const fns[4][2] = { + { gen_helper_mve_vmladavub, NULL }, + { gen_helper_mve_vmladavuh, NULL }, + { gen_helper_mve_vmladavuw, NULL }, + { NULL, NULL }, + }; + return do_dual_acc(s, a, fns[a->size][a->x]); +} + static void gen_vpst(DisasContext *s, uint32_t mask) { /* -- 2.20.1 Implement the MVE VMLA insn, which multiplies a vector by a scalar and accumulates into another vector. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/helper-mve.h | 4 ++++ target/arm/mve.decode | 1 + target/arm/mve_helper.c | 5 +++++ target/arm/translate-mve.c | 1 + 4 files changed, 11 insertions(+) diff --git a/target/arm/helper-mve.h b/target/arm/helper-mve.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/helper-mve.h +++ b/target/arm/helper-mve.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_4(mve_vqdmullb_scalarw, TCG_CALL_NO_WG, void, env, ptr, ptr, i3 DEF_HELPER_FLAGS_4(mve_vqdmullt_scalarh, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) DEF_HELPER_FLAGS_4(mve_vqdmullt_scalarw, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vmlab, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vmlah, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vmlaw, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) + DEF_HELPER_FLAGS_4(mve_vmlasb, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) DEF_HELPER_FLAGS_4(mve_vmlash, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) DEF_HELPER_FLAGS_4(mve_vmlasw, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) diff --git a/target/arm/mve.decode b/target/arm/mve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve.decode +++ b/target/arm/mve.decode @@ -XXX,XX +XXX,XX @@ VQDMULH_scalar 1110 1110 0 . .. ... 1 ... 0 1110 . 110 .... @2scalar VQRDMULH_scalar 1111 1110 0 . .. ... 1 ... 0 1110 . 110 .... @2scalar # The U bit (28) is don't-care because it does not affect the result +VMLA 111- 1110 0 . .. ... 1 ... 0 1110 . 100 .... @2scalar VMLAS 111- 1110 0 . .. ... 1 ... 1 1110 . 100 .... @2scalar # Vector add across vector diff --git a/target/arm/mve_helper.c b/target/arm/mve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve_helper.c +++ b/target/arm/mve_helper.c @@ -XXX,XX +XXX,XX @@ DO_2OP_SAT_SCALAR(vqrdmulh_scalarb, 1, int8_t, DO_QRDMULH_B) DO_2OP_SAT_SCALAR(vqrdmulh_scalarh, 2, int16_t, DO_QRDMULH_H) DO_2OP_SAT_SCALAR(vqrdmulh_scalarw, 4, int32_t, DO_QRDMULH_W) +/* Vector by scalar plus vector */ +#define DO_VMLA(D, N, M) ((N) * (M) + (D)) + +DO_2OP_ACC_SCALAR_U(vmla, DO_VMLA) + /* Vector by vector plus scalar */ #define DO_VMLAS(D, N, M) ((N) * (D) + (M)) diff --git a/target/arm/translate-mve.c b/target/arm/translate-mve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-mve.c +++ b/target/arm/translate-mve.c @@ -XXX,XX +XXX,XX @@ DO_2OP_SCALAR(VQSUB_U_scalar, vqsubu_scalar) DO_2OP_SCALAR(VQDMULH_scalar, vqdmulh_scalar) DO_2OP_SCALAR(VQRDMULH_scalar, vqrdmulh_scalar) DO_2OP_SCALAR(VBRSR, vbrsr) +DO_2OP_SCALAR(VMLA, vmla) DO_2OP_SCALAR(VMLAS, vmlas) static bool trans_VQDMULLB_scalar(DisasContext *s, arg_2scalar *a) -- 2.20.1 Implement the MVE saturating doubling multiply accumulate insns VQDMLAH, VQRDMLAH, VQDMLASH and VQRDMLASH. These perform a multiply, double, add the accumulator shifted by the element size, possibly round, saturate to twice the element size, then take the high half of the result. The *MLAH insns do vector * scalar + vector, and the *MLASH insns do vector * vector + scalar. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/helper-mve.h | 16 +++++++ target/arm/mve.decode | 5 ++ target/arm/mve_helper.c | 95 ++++++++++++++++++++++++++++++++++++++ target/arm/translate-mve.c | 4 ++ 4 files changed, 120 insertions(+) diff --git a/target/arm/helper-mve.h b/target/arm/helper-mve.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/helper-mve.h +++ b/target/arm/helper-mve.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_4(mve_vmlasb, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) DEF_HELPER_FLAGS_4(mve_vmlash, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) DEF_HELPER_FLAGS_4(mve_vmlasw, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vqdmlahb, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vqdmlahh, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vqdmlahw, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) + +DEF_HELPER_FLAGS_4(mve_vqrdmlahb, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vqrdmlahh, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vqrdmlahw, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) + +DEF_HELPER_FLAGS_4(mve_vqdmlashb, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vqdmlashh, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vqdmlashw, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) + +DEF_HELPER_FLAGS_4(mve_vqrdmlashb, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vqrdmlashh, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vqrdmlashw, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) + DEF_HELPER_FLAGS_4(mve_vmlaldavsh, TCG_CALL_NO_WG, i64, env, ptr, ptr, i64) DEF_HELPER_FLAGS_4(mve_vmlaldavsw, TCG_CALL_NO_WG, i64, env, ptr, ptr, i64) DEF_HELPER_FLAGS_4(mve_vmlaldavxsh, TCG_CALL_NO_WG, i64, env, ptr, ptr, i64) diff --git a/target/arm/mve.decode b/target/arm/mve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve.decode +++ b/target/arm/mve.decode @@ -XXX,XX +XXX,XX @@ VQRDMULH_scalar 1111 1110 0 . .. ... 1 ... 0 1110 . 110 .... @2scalar VMLA 111- 1110 0 . .. ... 1 ... 0 1110 . 100 .... @2scalar VMLAS 111- 1110 0 . .. ... 1 ... 1 1110 . 100 .... @2scalar +VQRDMLAH 1110 1110 0 . .. ... 0 ... 0 1110 . 100 .... @2scalar +VQRDMLASH 1110 1110 0 . .. ... 0 ... 1 1110 . 100 .... @2scalar +VQDMLAH 1110 1110 0 . .. ... 0 ... 0 1110 . 110 .... @2scalar +VQDMLASH 1110 1110 0 . .. ... 0 ... 1 1110 . 110 .... @2scalar + # Vector add across vector { VADDV 111 u:1 1110 1111 size:2 01 ... 0 1111 0 0 a:1 0 qm:3 0 rda=%rdalo diff --git a/target/arm/mve_helper.c b/target/arm/mve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve_helper.c +++ b/target/arm/mve_helper.c @@ -XXX,XX +XXX,XX @@ DO_VQDMLADH_OP(vqrdmlsdhxw, 4, int32_t, 1, 1, do_vqdmlsdh_w) mve_advance_vpt(env); \ } +#define DO_2OP_SAT_ACC_SCALAR(OP, ESIZE, TYPE, FN) \ + void HELPER(glue(mve_, OP))(CPUARMState *env, void *vd, void *vn, \ + uint32_t rm) \ + { \ + TYPE *d = vd, *n = vn; \ + TYPE m = rm; \ + uint16_t mask = mve_element_mask(env); \ + unsigned e; \ + bool qc = false; \ + for (e = 0; e < 16 / ESIZE; e++, mask >>= ESIZE) { \ + bool sat = false; \ + mergemask(&d[H##ESIZE(e)], \ + FN(d[H##ESIZE(e)], n[H##ESIZE(e)], m, &sat), \ + mask); \ + qc |= sat & mask & 1; \ + } \ + if (qc) { \ + env->vfp.qc[0] = qc; \ + } \ + mve_advance_vpt(env); \ + } + /* provide unsigned 2-op scalar helpers for all sizes */ #define DO_2OP_SCALAR_U(OP, FN) \ DO_2OP_SCALAR(OP##b, 1, uint8_t, FN) \ @@ -XXX,XX +XXX,XX @@ DO_2OP_SAT_SCALAR(vqrdmulh_scalarb, 1, int8_t, DO_QRDMULH_B) DO_2OP_SAT_SCALAR(vqrdmulh_scalarh, 2, int16_t, DO_QRDMULH_H) DO_2OP_SAT_SCALAR(vqrdmulh_scalarw, 4, int32_t, DO_QRDMULH_W) +static int8_t do_vqdmlah_b(int8_t a, int8_t b, int8_t c, int round, bool *sat) +{ + int64_t r = (int64_t)a * b * 2 + ((int64_t)c << 8) + (round << 7); + return do_sat_bhw(r, INT16_MIN, INT16_MAX, sat) >> 8; +} + +static int16_t do_vqdmlah_h(int16_t a, int16_t b, int16_t c, + int round, bool *sat) +{ + int64_t r = (int64_t)a * b * 2 + ((int64_t)c << 16) + (round << 15); + return do_sat_bhw(r, INT32_MIN, INT32_MAX, sat) >> 16; +} + +static int32_t do_vqdmlah_w(int32_t a, int32_t b, int32_t c, + int round, bool *sat) +{ + /* + * Architecturally we should do the entire add, double, round + * and then check for saturation. We do three saturating adds, + * but we need to be careful about the order. If the first + * m1 + m2 saturates then it's impossible for the *2+rc to + * bring it back into the non-saturated range. However, if + * m1 + m2 is negative then it's possible that doing the doubling + * would take the intermediate result below INT64_MAX and the + * addition of the rounding constant then brings it back in range. + * So we add half the rounding constant and half the "c << esize" + * before doubling rather than adding the rounding constant after + * the doubling. + */ + int64_t m1 = (int64_t)a * b; + int64_t m2 = (int64_t)c << 31; + int64_t r; + if (sadd64_overflow(m1, m2, &r) || + sadd64_overflow(r, (round << 30), &r) || + sadd64_overflow(r, r, &r)) { + *sat = true; + return r < 0 ? INT32_MAX : INT32_MIN; + } + return r >> 32; +} + +/* + * The *MLAH insns are vector * scalar + vector; + * the *MLASH insns are vector * vector + scalar + */ +#define DO_VQDMLAH_B(D, N, M, S) do_vqdmlah_b(N, M, D, 0, S) +#define DO_VQDMLAH_H(D, N, M, S) do_vqdmlah_h(N, M, D, 0, S) +#define DO_VQDMLAH_W(D, N, M, S) do_vqdmlah_w(N, M, D, 0, S) +#define DO_VQRDMLAH_B(D, N, M, S) do_vqdmlah_b(N, M, D, 1, S) +#define DO_VQRDMLAH_H(D, N, M, S) do_vqdmlah_h(N, M, D, 1, S) +#define DO_VQRDMLAH_W(D, N, M, S) do_vqdmlah_w(N, M, D, 1, S) + +#define DO_VQDMLASH_B(D, N, M, S) do_vqdmlah_b(N, D, M, 0, S) +#define DO_VQDMLASH_H(D, N, M, S) do_vqdmlah_h(N, D, M, 0, S) +#define DO_VQDMLASH_W(D, N, M, S) do_vqdmlah_w(N, D, M, 0, S) +#define DO_VQRDMLASH_B(D, N, M, S) do_vqdmlah_b(N, D, M, 1, S) +#define DO_VQRDMLASH_H(D, N, M, S) do_vqdmlah_h(N, D, M, 1, S) +#define DO_VQRDMLASH_W(D, N, M, S) do_vqdmlah_w(N, D, M, 1, S) + +DO_2OP_SAT_ACC_SCALAR(vqdmlahb, 1, int8_t, DO_VQDMLAH_B) +DO_2OP_SAT_ACC_SCALAR(vqdmlahh, 2, int16_t, DO_VQDMLAH_H) +DO_2OP_SAT_ACC_SCALAR(vqdmlahw, 4, int32_t, DO_VQDMLAH_W) +DO_2OP_SAT_ACC_SCALAR(vqrdmlahb, 1, int8_t, DO_VQRDMLAH_B) +DO_2OP_SAT_ACC_SCALAR(vqrdmlahh, 2, int16_t, DO_VQRDMLAH_H) +DO_2OP_SAT_ACC_SCALAR(vqrdmlahw, 4, int32_t, DO_VQRDMLAH_W) + +DO_2OP_SAT_ACC_SCALAR(vqdmlashb, 1, int8_t, DO_VQDMLASH_B) +DO_2OP_SAT_ACC_SCALAR(vqdmlashh, 2, int16_t, DO_VQDMLASH_H) +DO_2OP_SAT_ACC_SCALAR(vqdmlashw, 4, int32_t, DO_VQDMLASH_W) +DO_2OP_SAT_ACC_SCALAR(vqrdmlashb, 1, int8_t, DO_VQRDMLASH_B) +DO_2OP_SAT_ACC_SCALAR(vqrdmlashh, 2, int16_t, DO_VQRDMLASH_H) +DO_2OP_SAT_ACC_SCALAR(vqrdmlashw, 4, int32_t, DO_VQRDMLASH_W) + /* Vector by scalar plus vector */ #define DO_VMLA(D, N, M) ((N) * (M) + (D)) diff --git a/target/arm/translate-mve.c b/target/arm/translate-mve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-mve.c +++ b/target/arm/translate-mve.c @@ -XXX,XX +XXX,XX @@ DO_2OP_SCALAR(VQRDMULH_scalar, vqrdmulh_scalar) DO_2OP_SCALAR(VBRSR, vbrsr) DO_2OP_SCALAR(VMLA, vmla) DO_2OP_SCALAR(VMLAS, vmlas) +DO_2OP_SCALAR(VQDMLAH, vqdmlah) +DO_2OP_SCALAR(VQRDMLAH, vqrdmlah) +DO_2OP_SCALAR(VQDMLASH, vqdmlash) +DO_2OP_SCALAR(VQRDMLASH, vqrdmlash) static bool trans_VQDMULLB_scalar(DisasContext *s, arg_2scalar *a) { -- 2.20.1 Implement the MVE 1-operand saturating operations VQABS and VQNEG. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/helper-mve.h | 8 ++++++++ target/arm/mve.decode | 3 +++ target/arm/mve_helper.c | 37 +++++++++++++++++++++++++++++++++++++ target/arm/translate-mve.c | 2 ++ 4 files changed, 50 insertions(+) diff --git a/target/arm/helper-mve.h b/target/arm/helper-mve.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/helper-mve.h +++ b/target/arm/helper-mve.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_3(mve_vnegw, TCG_CALL_NO_WG, void, env, ptr, ptr) DEF_HELPER_FLAGS_3(mve_vfnegh, TCG_CALL_NO_WG, void, env, ptr, ptr) DEF_HELPER_FLAGS_3(mve_vfnegs, TCG_CALL_NO_WG, void, env, ptr, ptr) +DEF_HELPER_FLAGS_3(mve_vqabsb, TCG_CALL_NO_WG, void, env, ptr, ptr) +DEF_HELPER_FLAGS_3(mve_vqabsh, TCG_CALL_NO_WG, void, env, ptr, ptr) +DEF_HELPER_FLAGS_3(mve_vqabsw, TCG_CALL_NO_WG, void, env, ptr, ptr) + +DEF_HELPER_FLAGS_3(mve_vqnegb, TCG_CALL_NO_WG, void, env, ptr, ptr) +DEF_HELPER_FLAGS_3(mve_vqnegh, TCG_CALL_NO_WG, void, env, ptr, ptr) +DEF_HELPER_FLAGS_3(mve_vqnegw, TCG_CALL_NO_WG, void, env, ptr, ptr) + DEF_HELPER_FLAGS_3(mve_vmovnbb, TCG_CALL_NO_WG, void, env, ptr, ptr) DEF_HELPER_FLAGS_3(mve_vmovnbh, TCG_CALL_NO_WG, void, env, ptr, ptr) DEF_HELPER_FLAGS_3(mve_vmovntb, TCG_CALL_NO_WG, void, env, ptr, ptr) diff --git a/target/arm/mve.decode b/target/arm/mve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve.decode +++ b/target/arm/mve.decode @@ -XXX,XX +XXX,XX @@ VABS_fp 1111 1111 1 . 11 .. 01 ... 0 0111 01 . 0 ... 0 @1op VNEG 1111 1111 1 . 11 .. 01 ... 0 0011 11 . 0 ... 0 @1op VNEG_fp 1111 1111 1 . 11 .. 01 ... 0 0111 11 . 0 ... 0 @1op +VQABS 1111 1111 1 . 11 .. 00 ... 0 0111 01 . 0 ... 0 @1op +VQNEG 1111 1111 1 . 11 .. 00 ... 0 0111 11 . 0 ... 0 @1op + &vdup qd rt size # Qd is in the fields usually named Qn @vdup .... .... . . .. ... . rt:4 .... . . . . .... qd=%qn &vdup diff --git a/target/arm/mve_helper.c b/target/arm/mve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve_helper.c +++ b/target/arm/mve_helper.c @@ -XXX,XX +XXX,XX @@ void HELPER(mve_vpsel)(CPUARMState *env, void *vd, void *vn, void *vm) } mve_advance_vpt(env); } + +#define DO_1OP_SAT(OP, ESIZE, TYPE, FN) \ + void HELPER(mve_##OP)(CPUARMState *env, void *vd, void *vm) \ + { \ + TYPE *d = vd, *m = vm; \ + uint16_t mask = mve_element_mask(env); \ + unsigned e; \ + bool qc = false; \ + for (e = 0; e < 16 / ESIZE; e++, mask >>= ESIZE) { \ + bool sat = false; \ + mergemask(&d[H##ESIZE(e)], FN(m[H##ESIZE(e)], &sat), mask); \ + qc |= sat & mask & 1; \ + } \ + if (qc) { \ + env->vfp.qc[0] = qc; \ + } \ + mve_advance_vpt(env); \ + } + +#define DO_VQABS_B(N, SATP) \ + do_sat_bhs(DO_ABS((int64_t)N), INT8_MIN, INT8_MAX, SATP) +#define DO_VQABS_H(N, SATP) \ + do_sat_bhs(DO_ABS((int64_t)N), INT16_MIN, INT16_MAX, SATP) +#define DO_VQABS_W(N, SATP) \ + do_sat_bhs(DO_ABS((int64_t)N), INT32_MIN, INT32_MAX, SATP) + +#define DO_VQNEG_B(N, SATP) do_sat_bhs(-(int64_t)N, INT8_MIN, INT8_MAX, SATP) +#define DO_VQNEG_H(N, SATP) do_sat_bhs(-(int64_t)N, INT16_MIN, INT16_MAX, SATP) +#define DO_VQNEG_W(N, SATP) do_sat_bhs(-(int64_t)N, INT32_MIN, INT32_MAX, SATP) + +DO_1OP_SAT(vqabsb, 1, int8_t, DO_VQABS_B) +DO_1OP_SAT(vqabsh, 2, int16_t, DO_VQABS_H) +DO_1OP_SAT(vqabsw, 4, int32_t, DO_VQABS_W) + +DO_1OP_SAT(vqnegb, 1, int8_t, DO_VQNEG_B) +DO_1OP_SAT(vqnegh, 2, int16_t, DO_VQNEG_H) +DO_1OP_SAT(vqnegw, 4, int32_t, DO_VQNEG_W) diff --git a/target/arm/translate-mve.c b/target/arm/translate-mve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-mve.c +++ b/target/arm/translate-mve.c @@ -XXX,XX +XXX,XX @@ DO_1OP(VCLZ, vclz) DO_1OP(VCLS, vcls) DO_1OP(VABS, vabs) DO_1OP(VNEG, vneg) +DO_1OP(VQABS, vqabs) +DO_1OP(VQNEG, vqneg) /* Narrowing moves: only size 0 and 1 are valid */ #define DO_VMOVN(INSN, FN) \ -- 2.20.1 Implement the MVE VMAXA and VMINA insns, which take the absolute value of the signed elements in the input vector and then accumulate the unsigned max or min into the destination vector. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/helper-mve.h | 8 ++++++++ target/arm/mve.decode | 4 ++++ target/arm/mve_helper.c | 26 ++++++++++++++++++++++++++ target/arm/translate-mve.c | 2 ++ 4 files changed, 40 insertions(+) diff --git a/target/arm/helper-mve.h b/target/arm/helper-mve.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/helper-mve.h +++ b/target/arm/helper-mve.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_3(mve_vqnegb, TCG_CALL_NO_WG, void, env, ptr, ptr) DEF_HELPER_FLAGS_3(mve_vqnegh, TCG_CALL_NO_WG, void, env, ptr, ptr) DEF_HELPER_FLAGS_3(mve_vqnegw, TCG_CALL_NO_WG, void, env, ptr, ptr) +DEF_HELPER_FLAGS_3(mve_vmaxab, TCG_CALL_NO_WG, void, env, ptr, ptr) +DEF_HELPER_FLAGS_3(mve_vmaxah, TCG_CALL_NO_WG, void, env, ptr, ptr) +DEF_HELPER_FLAGS_3(mve_vmaxaw, TCG_CALL_NO_WG, void, env, ptr, ptr) + +DEF_HELPER_FLAGS_3(mve_vminab, TCG_CALL_NO_WG, void, env, ptr, ptr) +DEF_HELPER_FLAGS_3(mve_vminah, TCG_CALL_NO_WG, void, env, ptr, ptr) +DEF_HELPER_FLAGS_3(mve_vminaw, TCG_CALL_NO_WG, void, env, ptr, ptr) + DEF_HELPER_FLAGS_3(mve_vmovnbb, TCG_CALL_NO_WG, void, env, ptr, ptr) DEF_HELPER_FLAGS_3(mve_vmovnbh, TCG_CALL_NO_WG, void, env, ptr, ptr) DEF_HELPER_FLAGS_3(mve_vmovntb, TCG_CALL_NO_WG, void, env, ptr, ptr) diff --git a/target/arm/mve.decode b/target/arm/mve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve.decode +++ b/target/arm/mve.decode @@ -XXX,XX +XXX,XX @@ VMUL 1110 1111 0 . .. ... 0 ... 0 1001 . 1 . 1 ... 0 @2op VQMOVUNB 111 0 1110 0 . 11 .. 01 ... 0 1110 1 0 . 0 ... 1 @1op VQMOVN_BS 111 0 1110 0 . 11 .. 11 ... 0 1110 0 0 . 0 ... 1 @1op + VMAXA 111 0 1110 0 . 11 .. 11 ... 0 1110 1 0 . 0 ... 1 @1op + VMULH_S 111 0 1110 0 . .. ...1 ... 0 1110 . 0 . 0 ... 1 @2op } @@ -XXX,XX +XXX,XX @@ VMUL 1110 1111 0 . .. ... 0 ... 0 1001 . 1 . 1 ... 0 @2op VQMOVUNT 111 0 1110 0 . 11 .. 01 ... 1 1110 1 0 . 0 ... 1 @1op VQMOVN_TS 111 0 1110 0 . 11 .. 11 ... 1 1110 0 0 . 0 ... 1 @1op + VMINA 111 0 1110 0 . 11 .. 11 ... 1 1110 1 0 . 0 ... 1 @1op + VRMULH_S 111 0 1110 0 . .. ...1 ... 1 1110 . 0 . 0 ... 1 @2op } diff --git a/target/arm/mve_helper.c b/target/arm/mve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve_helper.c +++ b/target/arm/mve_helper.c @@ -XXX,XX +XXX,XX @@ DO_1OP_SAT(vqabsw, 4, int32_t, DO_VQABS_W) DO_1OP_SAT(vqnegb, 1, int8_t, DO_VQNEG_B) DO_1OP_SAT(vqnegh, 2, int16_t, DO_VQNEG_H) DO_1OP_SAT(vqnegw, 4, int32_t, DO_VQNEG_W) + +/* + * VMAXA, VMINA: vd is unsigned; vm is signed, and we take its + * absolute value; we then do an unsigned comparison. + */ +#define DO_VMAXMINA(OP, ESIZE, STYPE, UTYPE, FN) \ + void HELPER(mve_##OP)(CPUARMState *env, void *vd, void *vm) \ + { \ + UTYPE *d = vd; \ + STYPE *m = vm; \ + uint16_t mask = mve_element_mask(env); \ + unsigned e; \ + for (e = 0; e < 16 / ESIZE; e++, mask >>= ESIZE) { \ + UTYPE r = DO_ABS(m[H##ESIZE(e)]); \ + r = FN(d[H##ESIZE(e)], r); \ + mergemask(&d[H##ESIZE(e)], r, mask); \ + } \ + mve_advance_vpt(env); \ + } + +DO_VMAXMINA(vmaxab, 1, int8_t, uint8_t, DO_MAX) +DO_VMAXMINA(vmaxah, 2, int16_t, uint16_t, DO_MAX) +DO_VMAXMINA(vmaxaw, 4, int32_t, uint32_t, DO_MAX) +DO_VMAXMINA(vminab, 1, int8_t, uint8_t, DO_MIN) +DO_VMAXMINA(vminah, 2, int16_t, uint16_t, DO_MIN) +DO_VMAXMINA(vminaw, 4, int32_t, uint32_t, DO_MIN) diff --git a/target/arm/translate-mve.c b/target/arm/translate-mve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-mve.c +++ b/target/arm/translate-mve.c @@ -XXX,XX +XXX,XX @@ DO_1OP(VABS, vabs) DO_1OP(VNEG, vneg) DO_1OP(VQABS, vqabs) DO_1OP(VQNEG, vqneg) +DO_1OP(VMAXA, vmaxa) +DO_1OP(VMINA, vmina) /* Narrowing moves: only size 0 and 1 are valid */ #define DO_VMOVN(INSN, FN) \ -- 2.20.1 Implement the MVE VMOV forms that move data between 2 general-purpose registers and 2 32-bit lanes in a vector register. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/translate-a32.h | 1 + target/arm/mve.decode | 4 ++ target/arm/translate-mve.c | 85 ++++++++++++++++++++++++++++++++++++++ target/arm/translate-vfp.c | 2 +- 4 files changed, 91 insertions(+), 1 deletion(-) diff --git a/target/arm/translate-a32.h b/target/arm/translate-a32.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-a32.h +++ b/target/arm/translate-a32.h @@ -XXX,XX +XXX,XX @@ void gen_rev16(TCGv_i32 dest, TCGv_i32 var); void clear_eci_state(DisasContext *s); bool mve_eci_check(DisasContext *s); void mve_update_and_store_eci(DisasContext *s); +bool mve_skip_vmov(DisasContext *s, int vn, int index, int size); static inline TCGv_i32 load_cpu_offset(int offset) { diff --git a/target/arm/mve.decode b/target/arm/mve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve.decode +++ b/target/arm/mve.decode @@ -XXX,XX +XXX,XX @@ VLDR_VSTR 1110110 1 a:1 . w:1 . .... ... 111101 ....... @vldr_vstr \ VLDR_VSTR 1110110 1 a:1 . w:1 . .... ... 111110 ....... @vldr_vstr \ size=2 p=1 +# Moves between 2 32-bit vector lanes and 2 general purpose registers +VMOV_to_2gp 1110 1100 0 . 00 rt2:4 ... 0 1111 000 idx:1 rt:4 qd=%qd +VMOV_from_2gp 1110 1100 0 . 01 rt2:4 ... 0 1111 000 idx:1 rt:4 qd=%qd + # Vector 2-op VAND 1110 1111 0 . 00 ... 0 ... 0 0001 . 1 . 1 ... 0 @2op_nosz VBIC 1110 1111 0 . 01 ... 0 ... 0 0001 . 1 . 1 ... 0 @2op_nosz diff --git a/target/arm/translate-mve.c b/target/arm/translate-mve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-mve.c +++ b/target/arm/translate-mve.c @@ -XXX,XX +XXX,XX @@ static bool do_vabav(DisasContext *s, arg_vabav *a, MVEGenVABAVFn *fn) DO_VABAV(VABAV_S, vabavs) DO_VABAV(VABAV_U, vabavu) + +static bool trans_VMOV_to_2gp(DisasContext *s, arg_VMOV_to_2gp *a) +{ + /* + * VMOV two 32-bit vector lanes to two general-purpose registers. + * This insn is not predicated but it is subject to beat-wise + * execution if it is not in an IT block. For us this means + * only that if PSR.ECI says we should not be executing the beat + * corresponding to the lane of the vector register being accessed + * then we should skip perfoming the move, and that we need to do + * the usual check for bad ECI state and advance of ECI state. + * (If PSR.ECI is non-zero then we cannot be in an IT block.) + */ + TCGv_i32 tmp; + int vd; + + if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qd) || + a->rt == 13 || a->rt == 15 || a->rt2 == 13 || a->rt2 == 15 || + a->rt == a->rt2) { + /* Rt/Rt2 cases are UNPREDICTABLE */ + return false; + } + if (!mve_eci_check(s) || !vfp_access_check(s)) { + return true; + } + + /* Convert Qreg index to Dreg for read_neon_element32() etc */ + vd = a->qd * 2; + + if (!mve_skip_vmov(s, vd, a->idx, MO_32)) { + tmp = tcg_temp_new_i32(); + read_neon_element32(tmp, vd, a->idx, MO_32); + store_reg(s, a->rt, tmp); + } + if (!mve_skip_vmov(s, vd + 1, a->idx, MO_32)) { + tmp = tcg_temp_new_i32(); + read_neon_element32(tmp, vd + 1, a->idx, MO_32); + store_reg(s, a->rt2, tmp); + } + + mve_update_and_store_eci(s); + return true; +} + +static bool trans_VMOV_from_2gp(DisasContext *s, arg_VMOV_to_2gp *a) +{ + /* + * VMOV two general-purpose registers to two 32-bit vector lanes. + * This insn is not predicated but it is subject to beat-wise + * execution if it is not in an IT block. For us this means + * only that if PSR.ECI says we should not be executing the beat + * corresponding to the lane of the vector register being accessed + * then we should skip perfoming the move, and that we need to do + * the usual check for bad ECI state and advance of ECI state. + * (If PSR.ECI is non-zero then we cannot be in an IT block.) + */ + TCGv_i32 tmp; + int vd; + + if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qd) || + a->rt == 13 || a->rt == 15 || a->rt2 == 13 || a->rt2 == 15) { + /* Rt/Rt2 cases are UNPREDICTABLE */ + return false; + } + if (!mve_eci_check(s) || !vfp_access_check(s)) { + return true; + } + + /* Convert Qreg idx to Dreg for read_neon_element32() etc */ + vd = a->qd * 2; + + if (!mve_skip_vmov(s, vd, a->idx, MO_32)) { + tmp = load_reg(s, a->rt); + write_neon_element32(tmp, vd, a->idx, MO_32); + tcg_temp_free_i32(tmp); + } + if (!mve_skip_vmov(s, vd + 1, a->idx, MO_32)) { + tmp = load_reg(s, a->rt2); + write_neon_element32(tmp, vd + 1, a->idx, MO_32); + tcg_temp_free_i32(tmp); + } + + mve_update_and_store_eci(s); + return true; +} diff --git a/target/arm/translate-vfp.c b/target/arm/translate-vfp.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-vfp.c +++ b/target/arm/translate-vfp.c @@ -XXX,XX +XXX,XX @@ static bool trans_VCVT(DisasContext *s, arg_VCVT *a) return true; } -static bool mve_skip_vmov(DisasContext *s, int vn, int index, int size) +bool mve_skip_vmov(DisasContext *s, int vn, int index, int size) { /* * In a CPU with MVE, the VMOV (vector lane to general-purpose register) -- 2.20.1 Implement the MVE VPNOT insn, which inverts the bits in VPR.P0 (subject to both predication and to beatwise execution). Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/helper-mve.h | 1 + target/arm/mve.decode | 1 + target/arm/mve_helper.c | 17 +++++++++++++++++ target/arm/translate-mve.c | 19 +++++++++++++++++++ 4 files changed, 38 insertions(+) diff --git a/target/arm/helper-mve.h b/target/arm/helper-mve.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/helper-mve.h +++ b/target/arm/helper-mve.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_4(mve_vorn, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr) DEF_HELPER_FLAGS_4(mve_veor, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr) DEF_HELPER_FLAGS_4(mve_vpsel, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr) +DEF_HELPER_FLAGS_1(mve_vpnot, TCG_CALL_NO_WG, void, env) DEF_HELPER_FLAGS_4(mve_vaddb, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr) DEF_HELPER_FLAGS_4(mve_vaddh, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr) diff --git a/target/arm/mve.decode b/target/arm/mve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve.decode +++ b/target/arm/mve.decode @@ -XXX,XX +XXX,XX @@ VCMPGT 1111 1110 0 . .. ... 1 ... 1 1111 0 0 . 0 ... 1 @vcmp VCMPLE 1111 1110 0 . .. ... 1 ... 1 1111 1 0 . 0 ... 1 @vcmp { + VPNOT 1111 1110 0 0 11 000 1 000 0 1111 0100 1101 VPST 1111 1110 0 . 11 000 1 ... 0 1111 0100 1101 mask=%mask_22_13 VCMPEQ_scalar 1111 1110 0 . .. ... 1 ... 0 1111 0 1 0 0 .... @vcmp_scalar } diff --git a/target/arm/mve_helper.c b/target/arm/mve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve_helper.c +++ b/target/arm/mve_helper.c @@ -XXX,XX +XXX,XX @@ void HELPER(mve_vpsel)(CPUARMState *env, void *vd, void *vn, void *vm) mve_advance_vpt(env); } +void HELPER(mve_vpnot)(CPUARMState *env) +{ + /* + * P0 bits for unexecuted beats (where eci_mask is 0) are unchanged. + * P0 bits for predicated lanes in executed bits (where mask is 0) are 0. + * P0 bits otherwise are inverted. + * (This is the same logic as VCMP.) + * This insn is itself subject to predication and to beat-wise execution, + * and after it executes VPT state advances in the usual way. + */ + uint16_t mask = mve_element_mask(env); + uint16_t eci_mask = mve_eci_mask(env); + uint16_t beatpred = ~env->v7m.vpr & mask; + env->v7m.vpr = (env->v7m.vpr & ~(uint32_t)eci_mask) | (beatpred & eci_mask); + mve_advance_vpt(env); +} + #define DO_1OP_SAT(OP, ESIZE, TYPE, FN) \ void HELPER(mve_##OP)(CPUARMState *env, void *vd, void *vm) \ { \ diff --git a/target/arm/translate-mve.c b/target/arm/translate-mve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-mve.c +++ b/target/arm/translate-mve.c @@ -XXX,XX +XXX,XX @@ static bool trans_VPST(DisasContext *s, arg_VPST *a) return true; } +static bool trans_VPNOT(DisasContext *s, arg_VPNOT *a) +{ + /* + * Invert the predicate in VPR.P0. We have call out to + * a helper because this insn itself is beatwise and can + * be predicated. + */ + if (!dc_isar_feature(aa32_mve, s)) { + return false; + } + if (!mve_eci_check(s) || !vfp_access_check(s)) { + return true; + } + + gen_helper_mve_vpnot(cpu_env); + mve_update_eci(s); + return true; +} + static bool trans_VADDV(DisasContext *s, arg_VADDV *a) { /* VADDV: vector add across vector */ -- 2.20.1 Implement the MVE VCTP insn, which sets the VPR.P0 predicate bits so as to predicate any element at index Rn or greater is predicated. As with VPNOT, this insn itself is predicable and subject to beatwise execution. The calculation of the mask is the same as is used to determine ltpmask in mve_element_mask(), but we precalculate masklen in generated code to avoid having to have 4 helpers specialized by size. We put the decode line in with the low-overhead-loop insns in t32.decode because it's logically part of that collection of insn patterns, even though it is an MVE only insn. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/helper-mve.h | 2 ++ target/arm/translate-a32.h | 1 + target/arm/t32.decode | 1 + target/arm/mve_helper.c | 20 ++++++++++++++++++++ target/arm/translate-mve.c | 2 +- target/arm/translate.c | 33 +++++++++++++++++++++++++++++++++ 6 files changed, 58 insertions(+), 1 deletion(-) diff --git a/target/arm/helper-mve.h b/target/arm/helper-mve.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/helper-mve.h +++ b/target/arm/helper-mve.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_4(mve_veor, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr) DEF_HELPER_FLAGS_4(mve_vpsel, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr) DEF_HELPER_FLAGS_1(mve_vpnot, TCG_CALL_NO_WG, void, env) +DEF_HELPER_FLAGS_2(mve_vctp, TCG_CALL_NO_WG, void, env, i32) + DEF_HELPER_FLAGS_4(mve_vaddb, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr) DEF_HELPER_FLAGS_4(mve_vaddh, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr) DEF_HELPER_FLAGS_4(mve_vaddw, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr) diff --git a/target/arm/translate-a32.h b/target/arm/translate-a32.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-a32.h +++ b/target/arm/translate-a32.h @@ -XXX,XX +XXX,XX @@ long neon_element_offset(int reg, int element, MemOp memop); void gen_rev16(TCGv_i32 dest, TCGv_i32 var); void clear_eci_state(DisasContext *s); bool mve_eci_check(DisasContext *s); +void mve_update_eci(DisasContext *s); void mve_update_and_store_eci(DisasContext *s); bool mve_skip_vmov(DisasContext *s, int vn, int index, int size); diff --git a/target/arm/t32.decode b/target/arm/t32.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/t32.decode +++ b/target/arm/t32.decode @@ -XXX,XX +XXX,XX @@ BL 1111 0. .......... 11.1 ............ @branch24 # This is DLSTP DLS 1111 0 0000 0 size:2 rn:4 1110 0000 0000 0001 } + VCTP 1111 0 0000 0 size:2 rn:4 1110 1000 0000 0001 ] } diff --git a/target/arm/mve_helper.c b/target/arm/mve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve_helper.c +++ b/target/arm/mve_helper.c @@ -XXX,XX +XXX,XX @@ void HELPER(mve_vpnot)(CPUARMState *env) mve_advance_vpt(env); } +/* + * VCTP: P0 unexecuted bits unchanged, predicated bits zeroed, + * otherwise set according to value of Rn. The calculation of + * newmask here works in the same way as the calculation of the + * ltpmask in mve_element_mask(), but we have pre-calculated + * the masklen in the generated code. + */ +void HELPER(mve_vctp)(CPUARMState *env, uint32_t masklen) +{ + uint16_t mask = mve_element_mask(env); + uint16_t eci_mask = mve_eci_mask(env); + uint16_t newmask; + + assert(masklen <= 16); + newmask = masklen ? MAKE_64BIT_MASK(0, masklen) : 0; + newmask &= mask; + env->v7m.vpr = (env->v7m.vpr & ~(uint32_t)eci_mask) | (newmask & eci_mask); + mve_advance_vpt(env); +} + #define DO_1OP_SAT(OP, ESIZE, TYPE, FN) \ void HELPER(mve_##OP)(CPUARMState *env, void *vd, void *vm) \ { \ diff --git a/target/arm/translate-mve.c b/target/arm/translate-mve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-mve.c +++ b/target/arm/translate-mve.c @@ -XXX,XX +XXX,XX @@ bool mve_eci_check(DisasContext *s) } } -static void mve_update_eci(DisasContext *s) +void mve_update_eci(DisasContext *s) { /* * The helper function will always update the CPUState field, 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 bool trans_LCTP(DisasContext *s, arg_LCTP *a) return true; } +static bool trans_VCTP(DisasContext *s, arg_VCTP *a) +{ + /* + * M-profile Create Vector Tail Predicate. This insn is itself + * predicated and is subject to beatwise execution. + */ + TCGv_i32 rn_shifted, masklen; + + if (!dc_isar_feature(aa32_mve, s) || a->rn == 13 || a->rn == 15) { + return false; + } + + if (!mve_eci_check(s) || !vfp_access_check(s)) { + return true; + } + + /* + * We pre-calculate the mask length here to avoid having + * to have multiple helpers specialized for size. + * We pass the helper "rn <= (1 << (4 - size)) ? (rn << size) : 16". + */ + rn_shifted = tcg_temp_new_i32(); + masklen = load_reg(s, a->rn); + tcg_gen_shli_i32(rn_shifted, masklen, a->size); + tcg_gen_movcond_i32(TCG_COND_LEU, masklen, + masklen, tcg_constant_i32(1 << (4 - a->size)), + rn_shifted, tcg_constant_i32(16)); + gen_helper_mve_vctp(cpu_env, masklen); + tcg_temp_free_i32(masklen); + tcg_temp_free_i32(rn_shifted); + mve_update_eci(s); + return true; +} static bool op_tbranch(DisasContext *s, arg_tbranch *a, bool half) { -- 2.20.1 Implement the MVE gather-loads and scatter-stores which form the address by adding a base value from a scalar register to an offset in each element of a vector. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/helper-mve.h | 32 +++++++++ target/arm/mve.decode | 12 ++++ target/arm/mve_helper.c | 129 +++++++++++++++++++++++++++++++++++++ target/arm/translate-mve.c | 97 ++++++++++++++++++++++++++++ 4 files changed, 270 insertions(+) diff --git a/target/arm/helper-mve.h b/target/arm/helper-mve.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/helper-mve.h +++ b/target/arm/helper-mve.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_3(mve_vstrb_h, TCG_CALL_NO_WG, void, env, ptr, i32) DEF_HELPER_FLAGS_3(mve_vstrb_w, TCG_CALL_NO_WG, void, env, ptr, i32) DEF_HELPER_FLAGS_3(mve_vstrh_w, TCG_CALL_NO_WG, void, env, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vldrb_sg_sh, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vldrb_sg_sw, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vldrh_sg_sw, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) + +DEF_HELPER_FLAGS_4(mve_vldrb_sg_ub, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vldrb_sg_uh, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vldrb_sg_uw, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vldrh_sg_uh, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vldrh_sg_uw, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vldrw_sg_uw, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vldrd_sg_ud, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) + +DEF_HELPER_FLAGS_4(mve_vstrb_sg_ub, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vstrb_sg_uh, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vstrb_sg_uw, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vstrh_sg_uh, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vstrh_sg_uw, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vstrw_sg_uw, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vstrd_sg_ud, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) + +DEF_HELPER_FLAGS_4(mve_vldrh_sg_os_sw, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) + +DEF_HELPER_FLAGS_4(mve_vldrh_sg_os_uh, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vldrh_sg_os_uw, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vldrw_sg_os_uw, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vldrd_sg_os_ud, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) + +DEF_HELPER_FLAGS_4(mve_vstrh_sg_os_uh, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vstrh_sg_os_uw, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vstrw_sg_os_uw, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vstrd_sg_os_ud, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) + DEF_HELPER_FLAGS_3(mve_vdup, TCG_CALL_NO_WG, void, env, ptr, i32) DEF_HELPER_FLAGS_4(mve_vidupb, TCG_CALL_NO_WG, i32, env, ptr, i32, i32) diff --git a/target/arm/mve.decode b/target/arm/mve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve.decode +++ b/target/arm/mve.decode @@ -XXX,XX +XXX,XX @@ &shl_scalar qda rm size &vmaxv qm rda size &vabav qn qm rda size +&vldst_sg qd qm rn size msize os + +# scatter-gather memory size is in bits 6:4 +%sg_msize 6:1 4:1 @vldr_vstr ....... . . . . l:1 rn:4 ... ...... imm:7 &vldr_vstr qd=%qd u=0 # Note that both Rn and Qd are 3 bits only (no D bit) @vldst_wn ... u:1 ... . . . . l:1 . rn:3 qd:3 . ... .. imm:7 &vldr_vstr +@vldst_sg .... .... .... rn:4 .... ... size:2 ... ... os:1 &vldst_sg \ + qd=%qd qm=%qm msize=%sg_msize + @1op .... .... .... size:2 .. .... .... .... .... &1op qd=%qd qm=%qm @1op_nosz .... .... .... .... .... .... .... .... &1op qd=%qd qm=%qm size=0 @2op .... .... .. size:2 .... .... .... .... .... &2op qd=%qd qm=%qm qn=%qn @@ -XXX,XX +XXX,XX @@ VLDR_VSTR 1110110 1 a:1 . w:1 . .... ... 111101 ....... @vldr_vstr \ VLDR_VSTR 1110110 1 a:1 . w:1 . .... ... 111110 ....... @vldr_vstr \ size=2 p=1 +# gather loads/scatter stores +VLDR_S_sg 111 0 1100 1 . 01 .... ... 0 111 . .... .... @vldst_sg +VLDR_U_sg 111 1 1100 1 . 01 .... ... 0 111 . .... .... @vldst_sg +VSTR_sg 111 0 1100 1 . 00 .... ... 0 111 . .... .... @vldst_sg + # Moves between 2 32-bit vector lanes and 2 general purpose registers VMOV_to_2gp 1110 1100 0 . 00 rt2:4 ... 0 1111 000 idx:1 rt:4 qd=%qd VMOV_from_2gp 1110 1100 0 . 01 rt2:4 ... 0 1111 000 idx:1 rt:4 qd=%qd diff --git a/target/arm/mve_helper.c b/target/arm/mve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve_helper.c +++ b/target/arm/mve_helper.c @@ -XXX,XX +XXX,XX @@ DO_VSTR(vstrh_w, 2, stw, 4, int32_t) #undef DO_VLDR #undef DO_VSTR +/* + * Gather loads/scatter stores. Here each element of Qm specifies + * an offset to use from the base register Rm. In the _os_ versions + * that offset is scaled by the element size. + * For loads, predicated lanes are zeroed instead of retaining + * their previous values. + */ +#define DO_VLDR_SG(OP, LDTYPE, ESIZE, TYPE, OFFTYPE, ADDRFN) \ + void HELPER(mve_##OP)(CPUARMState *env, void *vd, void *vm, \ + uint32_t base) \ + { \ + TYPE *d = vd; \ + OFFTYPE *m = vm; \ + uint16_t mask = mve_element_mask(env); \ + uint16_t eci_mask = mve_eci_mask(env); \ + unsigned e; \ + uint32_t addr; \ + for (e = 0; e < 16 / ESIZE; e++, mask >>= ESIZE, eci_mask >>= ESIZE) { \ + if (!(eci_mask & 1)) { \ + continue; \ + } \ + addr = ADDRFN(base, m[H##ESIZE(e)]); \ + d[H##ESIZE(e)] = (mask & 1) ? \ + cpu_##LDTYPE##_data_ra(env, addr, GETPC()) : 0; \ + } \ + mve_advance_vpt(env); \ + } + +/* We know here TYPE is unsigned so always the same as the offset type */ +#define DO_VSTR_SG(OP, STTYPE, ESIZE, TYPE, ADDRFN) \ + void HELPER(mve_##OP)(CPUARMState *env, void *vd, void *vm, \ + uint32_t base) \ + { \ + TYPE *d = vd; \ + TYPE *m = vm; \ + uint16_t mask = mve_element_mask(env); \ + unsigned e; \ + uint32_t addr; \ + for (e = 0; e < 16 / ESIZE; e++, mask >>= ESIZE) { \ + addr = ADDRFN(base, m[H##ESIZE(e)]); \ + if (mask & 1) { \ + cpu_##STTYPE##_data_ra(env, addr, d[H##ESIZE(e)], GETPC()); \ + } \ + } \ + mve_advance_vpt(env); \ + } + +/* + * 64-bit accesses are slightly different: they are done as two 32-bit + * accesses, controlled by the predicate mask for the relevant beat, + * and with a single 32-bit offset in the first of the two Qm elements. + * Note that for QEMU our IMPDEF AIRCR.ENDIANNESS is always 0 (little). + */ +#define DO_VLDR64_SG(OP, ADDRFN) \ + void HELPER(mve_##OP)(CPUARMState *env, void *vd, void *vm, \ + uint32_t base) \ + { \ + uint32_t *d = vd; \ + uint32_t *m = vm; \ + uint16_t mask = mve_element_mask(env); \ + uint16_t eci_mask = mve_eci_mask(env); \ + unsigned e; \ + uint32_t addr; \ + for (e = 0; e < 16 / 4; e++, mask >>= 4, eci_mask >>= 4) { \ + if (!(eci_mask & 1)) { \ + continue; \ + } \ + addr = ADDRFN(base, m[H4(e & ~1)]); \ + addr += 4 * (e & 1); \ + d[H4(e)] = (mask & 1) ? cpu_ldl_data_ra(env, addr, GETPC()) : 0; \ + } \ + mve_advance_vpt(env); \ + } + +#define DO_VSTR64_SG(OP, ADDRFN) \ + void HELPER(mve_##OP)(CPUARMState *env, void *vd, void *vm, \ + uint32_t base) \ + { \ + uint32_t *d = vd; \ + uint32_t *m = vm; \ + uint16_t mask = mve_element_mask(env); \ + unsigned e; \ + uint32_t addr; \ + for (e = 0; e < 16 / 4; e++, mask >>= 4) { \ + addr = ADDRFN(base, m[H4(e & ~1)]); \ + addr += 4 * (e & 1); \ + if (mask & 1) { \ + cpu_stl_data_ra(env, addr, d[H4(e)], GETPC()); \ + } \ + } \ + mve_advance_vpt(env); \ + } + +#define ADDR_ADD(BASE, OFFSET) ((BASE) + (OFFSET)) +#define ADDR_ADD_OSH(BASE, OFFSET) ((BASE) + ((OFFSET) << 1)) +#define ADDR_ADD_OSW(BASE, OFFSET) ((BASE) + ((OFFSET) << 2)) +#define ADDR_ADD_OSD(BASE, OFFSET) ((BASE) + ((OFFSET) << 3)) + +DO_VLDR_SG(vldrb_sg_sh, ldsb, 2, int16_t, uint16_t, ADDR_ADD) +DO_VLDR_SG(vldrb_sg_sw, ldsb, 4, int32_t, uint32_t, ADDR_ADD) +DO_VLDR_SG(vldrh_sg_sw, ldsw, 4, int32_t, uint32_t, ADDR_ADD) + +DO_VLDR_SG(vldrb_sg_ub, ldub, 1, uint8_t, uint8_t, ADDR_ADD) +DO_VLDR_SG(vldrb_sg_uh, ldub, 2, uint16_t, uint16_t, ADDR_ADD) +DO_VLDR_SG(vldrb_sg_uw, ldub, 4, uint32_t, uint32_t, ADDR_ADD) +DO_VLDR_SG(vldrh_sg_uh, lduw, 2, uint16_t, uint16_t, ADDR_ADD) +DO_VLDR_SG(vldrh_sg_uw, lduw, 4, uint32_t, uint32_t, ADDR_ADD) +DO_VLDR_SG(vldrw_sg_uw, ldl, 4, uint32_t, uint32_t, ADDR_ADD) +DO_VLDR64_SG(vldrd_sg_ud, ADDR_ADD) + +DO_VLDR_SG(vldrh_sg_os_sw, ldsw, 4, int32_t, uint32_t, ADDR_ADD_OSH) +DO_VLDR_SG(vldrh_sg_os_uh, lduw, 2, uint16_t, uint16_t, ADDR_ADD_OSH) +DO_VLDR_SG(vldrh_sg_os_uw, lduw, 4, uint32_t, uint32_t, ADDR_ADD_OSH) +DO_VLDR_SG(vldrw_sg_os_uw, ldl, 4, uint32_t, uint32_t, ADDR_ADD_OSW) +DO_VLDR64_SG(vldrd_sg_os_ud, ADDR_ADD_OSD) + +DO_VSTR_SG(vstrb_sg_ub, stb, 1, uint8_t, ADDR_ADD) +DO_VSTR_SG(vstrb_sg_uh, stb, 2, uint16_t, ADDR_ADD) +DO_VSTR_SG(vstrb_sg_uw, stb, 4, uint32_t, ADDR_ADD) +DO_VSTR_SG(vstrh_sg_uh, stw, 2, uint16_t, ADDR_ADD) +DO_VSTR_SG(vstrh_sg_uw, stw, 4, uint32_t, ADDR_ADD) +DO_VSTR_SG(vstrw_sg_uw, stl, 4, uint32_t, ADDR_ADD) +DO_VSTR64_SG(vstrd_sg_ud, ADDR_ADD) + +DO_VSTR_SG(vstrh_sg_os_uh, stw, 2, uint16_t, ADDR_ADD_OSH) +DO_VSTR_SG(vstrh_sg_os_uw, stw, 4, uint32_t, ADDR_ADD_OSH) +DO_VSTR_SG(vstrw_sg_os_uw, stl, 4, uint32_t, ADDR_ADD_OSW) +DO_VSTR64_SG(vstrd_sg_os_ud, ADDR_ADD_OSD) + /* * The mergemask(D, R, M) macro performs the operation "*D = R" but * storing only the bytes which correspond to 1 bits in M, diff --git a/target/arm/translate-mve.c b/target/arm/translate-mve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-mve.c +++ b/target/arm/translate-mve.c @@ -XXX,XX +XXX,XX @@ static inline int vidup_imm(DisasContext *s, int x) #include "decode-mve.c.inc" typedef void MVEGenLdStFn(TCGv_ptr, TCGv_ptr, TCGv_i32); +typedef void MVEGenLdStSGFn(TCGv_ptr, TCGv_ptr, TCGv_ptr, TCGv_i32); typedef void MVEGenOneOpFn(TCGv_ptr, TCGv_ptr, TCGv_ptr); typedef void MVEGenTwoOpFn(TCGv_ptr, TCGv_ptr, TCGv_ptr, TCGv_ptr); typedef void MVEGenTwoOpScalarFn(TCGv_ptr, TCGv_ptr, TCGv_ptr, TCGv_i32); @@ -XXX,XX +XXX,XX @@ DO_VLDST_WIDE_NARROW(VLDSTB_H, vldrb_sh, vldrb_uh, vstrb_h, MO_8) DO_VLDST_WIDE_NARROW(VLDSTB_W, vldrb_sw, vldrb_uw, vstrb_w, MO_8) DO_VLDST_WIDE_NARROW(VLDSTH_W, vldrh_sw, vldrh_uw, vstrh_w, MO_16) +static bool do_ldst_sg(DisasContext *s, arg_vldst_sg *a, MVEGenLdStSGFn fn) +{ + TCGv_i32 addr; + TCGv_ptr qd, qm; + + if (!dc_isar_feature(aa32_mve, s) || + !mve_check_qreg_bank(s, a->qd | a->qm) || + !fn || a->rn == 15) { + /* Rn case is UNPREDICTABLE */ + return false; + } + + if (!mve_eci_check(s) || !vfp_access_check(s)) { + return true; + } + + addr = load_reg(s, a->rn); + + qd = mve_qreg_ptr(a->qd); + qm = mve_qreg_ptr(a->qm); + fn(cpu_env, qd, qm, addr); + tcg_temp_free_ptr(qd); + tcg_temp_free_ptr(qm); + tcg_temp_free_i32(addr); + mve_update_eci(s); + return true; +} + +/* + * The naming scheme here is "vldrb_sg_sh == in-memory byte loads + * signextended to halfword elements in register". _os_ indicates that + * the offsets in Qm should be scaled by the element size. + */ +/* This macro is just to make the arrays more compact in these functions */ +#define F(N) gen_helper_mve_##N + +/* VLDRB/VSTRB (ie msize 1) with OS=1 is UNPREDICTABLE; we UNDEF */ +static bool trans_VLDR_S_sg(DisasContext *s, arg_vldst_sg *a) +{ + static MVEGenLdStSGFn * const fns[2][4][4] = { { + { NULL, F(vldrb_sg_sh), F(vldrb_sg_sw), NULL }, + { NULL, NULL, F(vldrh_sg_sw), NULL }, + { NULL, NULL, NULL, NULL }, + { NULL, NULL, NULL, NULL } + }, { + { NULL, NULL, NULL, NULL }, + { NULL, NULL, F(vldrh_sg_os_sw), NULL }, + { NULL, NULL, NULL, NULL }, + { NULL, NULL, NULL, NULL } + } + }; + if (a->qd == a->qm) { + return false; /* UNPREDICTABLE */ + } + return do_ldst_sg(s, a, fns[a->os][a->msize][a->size]); +} + +static bool trans_VLDR_U_sg(DisasContext *s, arg_vldst_sg *a) +{ + static MVEGenLdStSGFn * const fns[2][4][4] = { { + { F(vldrb_sg_ub), F(vldrb_sg_uh), F(vldrb_sg_uw), NULL }, + { NULL, F(vldrh_sg_uh), F(vldrh_sg_uw), NULL }, + { NULL, NULL, F(vldrw_sg_uw), NULL }, + { NULL, NULL, NULL, F(vldrd_sg_ud) } + }, { + { NULL, NULL, NULL, NULL }, + { NULL, F(vldrh_sg_os_uh), F(vldrh_sg_os_uw), NULL }, + { NULL, NULL, F(vldrw_sg_os_uw), NULL }, + { NULL, NULL, NULL, F(vldrd_sg_os_ud) } + } + }; + if (a->qd == a->qm) { + return false; /* UNPREDICTABLE */ + } + return do_ldst_sg(s, a, fns[a->os][a->msize][a->size]); +} + +static bool trans_VSTR_sg(DisasContext *s, arg_vldst_sg *a) +{ + static MVEGenLdStSGFn * const fns[2][4][4] = { { + { F(vstrb_sg_ub), F(vstrb_sg_uh), F(vstrb_sg_uw), NULL }, + { NULL, F(vstrh_sg_uh), F(vstrh_sg_uw), NULL }, + { NULL, NULL, F(vstrw_sg_uw), NULL }, + { NULL, NULL, NULL, F(vstrd_sg_ud) } + }, { + { NULL, NULL, NULL, NULL }, + { NULL, F(vstrh_sg_os_uh), F(vstrh_sg_os_uw), NULL }, + { NULL, NULL, F(vstrw_sg_os_uw), NULL }, + { NULL, NULL, NULL, F(vstrd_sg_os_ud) } + } + }; + return do_ldst_sg(s, a, fns[a->os][a->msize][a->size]); +} + +#undef F + static bool trans_VDUP(DisasContext *s, arg_VDUP *a) { TCGv_ptr qd; -- 2.20.1 Implement the MVE VLDR/VSTR insns which do scatter-gather using base addresses from Qm plus or minus an immediate offset (possibly with writeback). Note that writeback is not predicated but it does have to honour ECI state, so we have to add an eci_mask check to the VSTR_SG macros (the VLDR_SG macros already needed this to be able to distinguish "skip beat" from "set predicated element to 0"). Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/helper-mve.h | 5 +++ target/arm/mve.decode | 10 +++++ target/arm/mve_helper.c | 91 ++++++++++++++++++++++++-------------- target/arm/translate-mve.c | 72 ++++++++++++++++++++++++++++++ 4 files changed, 146 insertions(+), 32 deletions(-) diff --git a/target/arm/helper-mve.h b/target/arm/helper-mve.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/helper-mve.h +++ b/target/arm/helper-mve.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_4(mve_vstrh_sg_os_uw, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) DEF_HELPER_FLAGS_4(mve_vstrw_sg_os_uw, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) DEF_HELPER_FLAGS_4(mve_vstrd_sg_os_ud, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vldrw_sg_wb_uw, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vldrd_sg_wb_ud, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vstrw_sg_wb_uw, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_4(mve_vstrd_sg_wb_ud, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) + DEF_HELPER_FLAGS_3(mve_vdup, TCG_CALL_NO_WG, void, env, ptr, i32) DEF_HELPER_FLAGS_4(mve_vidupb, TCG_CALL_NO_WG, i32, env, ptr, i32, i32) diff --git a/target/arm/mve.decode b/target/arm/mve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve.decode +++ b/target/arm/mve.decode @@ -XXX,XX +XXX,XX @@ &vmaxv qm rda size &vabav qn qm rda size &vldst_sg qd qm rn size msize os +&vldst_sg_imm qd qm a w imm # scatter-gather memory size is in bits 6:4 %sg_msize 6:1 4:1 @@ -XXX,XX +XXX,XX @@ @vldst_sg .... .... .... rn:4 .... ... size:2 ... ... os:1 &vldst_sg \ qd=%qd qm=%qm msize=%sg_msize +# Qm is in the fields usually labeled Qn +@vldst_sg_imm .... .... a:1 . w:1 . .... .... .... . imm:7 &vldst_sg_imm \ + qd=%qd qm=%qn + @1op .... .... .... size:2 .. .... .... .... .... &1op qd=%qd qm=%qm @1op_nosz .... .... .... .... .... .... .... .... &1op qd=%qd qm=%qm size=0 @2op .... .... .. size:2 .... .... .... .... .... &2op qd=%qd qm=%qm qn=%qn @@ -XXX,XX +XXX,XX @@ VLDR_S_sg 111 0 1100 1 . 01 .... ... 0 111 . .... .... @vldst_sg VLDR_U_sg 111 1 1100 1 . 01 .... ... 0 111 . .... .... @vldst_sg VSTR_sg 111 0 1100 1 . 00 .... ... 0 111 . .... .... @vldst_sg +VLDRW_sg_imm 111 1 1101 ... 1 ... 0 ... 1 1110 .... .... @vldst_sg_imm +VLDRD_sg_imm 111 1 1101 ... 1 ... 0 ... 1 1111 .... .... @vldst_sg_imm +VSTRW_sg_imm 111 1 1101 ... 0 ... 0 ... 1 1110 .... .... @vldst_sg_imm +VSTRD_sg_imm 111 1 1101 ... 0 ... 0 ... 1 1111 .... .... @vldst_sg_imm + # Moves between 2 32-bit vector lanes and 2 general purpose registers VMOV_to_2gp 1110 1100 0 . 00 rt2:4 ... 0 1111 000 idx:1 rt:4 qd=%qd VMOV_from_2gp 1110 1100 0 . 01 rt2:4 ... 0 1111 000 idx:1 rt:4 qd=%qd diff --git a/target/arm/mve_helper.c b/target/arm/mve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve_helper.c +++ b/target/arm/mve_helper.c @@ -XXX,XX +XXX,XX @@ DO_VSTR(vstrh_w, 2, stw, 4, int32_t) * For loads, predicated lanes are zeroed instead of retaining * their previous values. */ -#define DO_VLDR_SG(OP, LDTYPE, ESIZE, TYPE, OFFTYPE, ADDRFN) \ +#define DO_VLDR_SG(OP, LDTYPE, ESIZE, TYPE, OFFTYPE, ADDRFN, WB) \ void HELPER(mve_##OP)(CPUARMState *env, void *vd, void *vm, \ uint32_t base) \ { \ @@ -XXX,XX +XXX,XX @@ DO_VSTR(vstrh_w, 2, stw, 4, int32_t) addr = ADDRFN(base, m[H##ESIZE(e)]); \ d[H##ESIZE(e)] = (mask & 1) ? \ cpu_##LDTYPE##_data_ra(env, addr, GETPC()) : 0; \ + if (WB) { \ + m[H##ESIZE(e)] = addr; \ + } \ } \ mve_advance_vpt(env); \ } /* We know here TYPE is unsigned so always the same as the offset type */ -#define DO_VSTR_SG(OP, STTYPE, ESIZE, TYPE, ADDRFN) \ +#define DO_VSTR_SG(OP, STTYPE, ESIZE, TYPE, ADDRFN, WB) \ void HELPER(mve_##OP)(CPUARMState *env, void *vd, void *vm, \ uint32_t base) \ { \ TYPE *d = vd; \ TYPE *m = vm; \ uint16_t mask = mve_element_mask(env); \ + uint16_t eci_mask = mve_eci_mask(env); \ unsigned e; \ uint32_t addr; \ - for (e = 0; e < 16 / ESIZE; e++, mask >>= ESIZE) { \ + for (e = 0; e < 16 / ESIZE; e++, mask >>= ESIZE, eci_mask >>= ESIZE) { \ + if (!(eci_mask & 1)) { \ + continue; \ + } \ addr = ADDRFN(base, m[H##ESIZE(e)]); \ if (mask & 1) { \ cpu_##STTYPE##_data_ra(env, addr, d[H##ESIZE(e)], GETPC()); \ } \ + if (WB) { \ + m[H##ESIZE(e)] = addr; \ + } \ } \ mve_advance_vpt(env); \ } @@ -XXX,XX +XXX,XX @@ DO_VSTR(vstrh_w, 2, stw, 4, int32_t) * accesses, controlled by the predicate mask for the relevant beat, * and with a single 32-bit offset in the first of the two Qm elements. * Note that for QEMU our IMPDEF AIRCR.ENDIANNESS is always 0 (little). + * Address writeback happens on the odd beats and updates the address + * stored in the even-beat element. */ -#define DO_VLDR64_SG(OP, ADDRFN) \ +#define DO_VLDR64_SG(OP, ADDRFN, WB) \ void HELPER(mve_##OP)(CPUARMState *env, void *vd, void *vm, \ uint32_t base) \ { \ @@ -XXX,XX +XXX,XX @@ DO_VSTR(vstrh_w, 2, stw, 4, int32_t) addr = ADDRFN(base, m[H4(e & ~1)]); \ addr += 4 * (e & 1); \ d[H4(e)] = (mask & 1) ? cpu_ldl_data_ra(env, addr, GETPC()) : 0; \ + if (WB && (e & 1)) { \ + m[H4(e & ~1)] = addr - 4; \ + } \ } \ mve_advance_vpt(env); \ } -#define DO_VSTR64_SG(OP, ADDRFN) \ +#define DO_VSTR64_SG(OP, ADDRFN, WB) \ void HELPER(mve_##OP)(CPUARMState *env, void *vd, void *vm, \ uint32_t base) \ { \ uint32_t *d = vd; \ uint32_t *m = vm; \ uint16_t mask = mve_element_mask(env); \ + uint16_t eci_mask = mve_eci_mask(env); \ unsigned e; \ uint32_t addr; \ - for (e = 0; e < 16 / 4; e++, mask >>= 4) { \ + for (e = 0; e < 16 / 4; e++, mask >>= 4, eci_mask >>= 4) { \ + if (!(eci_mask & 1)) { \ + continue; \ + } \ addr = ADDRFN(base, m[H4(e & ~1)]); \ addr += 4 * (e & 1); \ if (mask & 1) { \ cpu_stl_data_ra(env, addr, d[H4(e)], GETPC()); \ } \ + if (WB && (e & 1)) { \ + m[H4(e & ~1)] = addr - 4; \ + } \ } \ mve_advance_vpt(env); \ } @@ -XXX,XX +XXX,XX @@ DO_VSTR(vstrh_w, 2, stw, 4, int32_t) #define ADDR_ADD_OSW(BASE, OFFSET) ((BASE) + ((OFFSET) << 2)) #define ADDR_ADD_OSD(BASE, OFFSET) ((BASE) + ((OFFSET) << 3)) -DO_VLDR_SG(vldrb_sg_sh, ldsb, 2, int16_t, uint16_t, ADDR_ADD) -DO_VLDR_SG(vldrb_sg_sw, ldsb, 4, int32_t, uint32_t, ADDR_ADD) -DO_VLDR_SG(vldrh_sg_sw, ldsw, 4, int32_t, uint32_t, ADDR_ADD) +DO_VLDR_SG(vldrb_sg_sh, ldsb, 2, int16_t, uint16_t, ADDR_ADD, false) +DO_VLDR_SG(vldrb_sg_sw, ldsb, 4, int32_t, uint32_t, ADDR_ADD, false) +DO_VLDR_SG(vldrh_sg_sw, ldsw, 4, int32_t, uint32_t, ADDR_ADD, false) -DO_VLDR_SG(vldrb_sg_ub, ldub, 1, uint8_t, uint8_t, ADDR_ADD) -DO_VLDR_SG(vldrb_sg_uh, ldub, 2, uint16_t, uint16_t, ADDR_ADD) -DO_VLDR_SG(vldrb_sg_uw, ldub, 4, uint32_t, uint32_t, ADDR_ADD) -DO_VLDR_SG(vldrh_sg_uh, lduw, 2, uint16_t, uint16_t, ADDR_ADD) -DO_VLDR_SG(vldrh_sg_uw, lduw, 4, uint32_t, uint32_t, ADDR_ADD) -DO_VLDR_SG(vldrw_sg_uw, ldl, 4, uint32_t, uint32_t, ADDR_ADD) -DO_VLDR64_SG(vldrd_sg_ud, ADDR_ADD) +DO_VLDR_SG(vldrb_sg_ub, ldub, 1, uint8_t, uint8_t, ADDR_ADD, false) +DO_VLDR_SG(vldrb_sg_uh, ldub, 2, uint16_t, uint16_t, ADDR_ADD, false) +DO_VLDR_SG(vldrb_sg_uw, ldub, 4, uint32_t, uint32_t, ADDR_ADD, false) +DO_VLDR_SG(vldrh_sg_uh, lduw, 2, uint16_t, uint16_t, ADDR_ADD, false) +DO_VLDR_SG(vldrh_sg_uw, lduw, 4, uint32_t, uint32_t, ADDR_ADD, false) +DO_VLDR_SG(vldrw_sg_uw, ldl, 4, uint32_t, uint32_t, ADDR_ADD, false) +DO_VLDR64_SG(vldrd_sg_ud, ADDR_ADD, false) -DO_VLDR_SG(vldrh_sg_os_sw, ldsw, 4, int32_t, uint32_t, ADDR_ADD_OSH) -DO_VLDR_SG(vldrh_sg_os_uh, lduw, 2, uint16_t, uint16_t, ADDR_ADD_OSH) -DO_VLDR_SG(vldrh_sg_os_uw, lduw, 4, uint32_t, uint32_t, ADDR_ADD_OSH) -DO_VLDR_SG(vldrw_sg_os_uw, ldl, 4, uint32_t, uint32_t, ADDR_ADD_OSW) -DO_VLDR64_SG(vldrd_sg_os_ud, ADDR_ADD_OSD) +DO_VLDR_SG(vldrh_sg_os_sw, ldsw, 4, int32_t, uint32_t, ADDR_ADD_OSH, false) +DO_VLDR_SG(vldrh_sg_os_uh, lduw, 2, uint16_t, uint16_t, ADDR_ADD_OSH, false) +DO_VLDR_SG(vldrh_sg_os_uw, lduw, 4, uint32_t, uint32_t, ADDR_ADD_OSH, false) +DO_VLDR_SG(vldrw_sg_os_uw, ldl, 4, uint32_t, uint32_t, ADDR_ADD_OSW, false) +DO_VLDR64_SG(vldrd_sg_os_ud, ADDR_ADD_OSD, false) -DO_VSTR_SG(vstrb_sg_ub, stb, 1, uint8_t, ADDR_ADD) -DO_VSTR_SG(vstrb_sg_uh, stb, 2, uint16_t, ADDR_ADD) -DO_VSTR_SG(vstrb_sg_uw, stb, 4, uint32_t, ADDR_ADD) -DO_VSTR_SG(vstrh_sg_uh, stw, 2, uint16_t, ADDR_ADD) -DO_VSTR_SG(vstrh_sg_uw, stw, 4, uint32_t, ADDR_ADD) -DO_VSTR_SG(vstrw_sg_uw, stl, 4, uint32_t, ADDR_ADD) -DO_VSTR64_SG(vstrd_sg_ud, ADDR_ADD) +DO_VSTR_SG(vstrb_sg_ub, stb, 1, uint8_t, ADDR_ADD, false) +DO_VSTR_SG(vstrb_sg_uh, stb, 2, uint16_t, ADDR_ADD, false) +DO_VSTR_SG(vstrb_sg_uw, stb, 4, uint32_t, ADDR_ADD, false) +DO_VSTR_SG(vstrh_sg_uh, stw, 2, uint16_t, ADDR_ADD, false) +DO_VSTR_SG(vstrh_sg_uw, stw, 4, uint32_t, ADDR_ADD, false) +DO_VSTR_SG(vstrw_sg_uw, stl, 4, uint32_t, ADDR_ADD, false) +DO_VSTR64_SG(vstrd_sg_ud, ADDR_ADD, false) -DO_VSTR_SG(vstrh_sg_os_uh, stw, 2, uint16_t, ADDR_ADD_OSH) -DO_VSTR_SG(vstrh_sg_os_uw, stw, 4, uint32_t, ADDR_ADD_OSH) -DO_VSTR_SG(vstrw_sg_os_uw, stl, 4, uint32_t, ADDR_ADD_OSW) -DO_VSTR64_SG(vstrd_sg_os_ud, ADDR_ADD_OSD) +DO_VSTR_SG(vstrh_sg_os_uh, stw, 2, uint16_t, ADDR_ADD_OSH, false) +DO_VSTR_SG(vstrh_sg_os_uw, stw, 4, uint32_t, ADDR_ADD_OSH, false) +DO_VSTR_SG(vstrw_sg_os_uw, stl, 4, uint32_t, ADDR_ADD_OSW, false) +DO_VSTR64_SG(vstrd_sg_os_ud, ADDR_ADD_OSD, false) + +DO_VLDR_SG(vldrw_sg_wb_uw, ldl, 4, uint32_t, uint32_t, ADDR_ADD, true) +DO_VLDR64_SG(vldrd_sg_wb_ud, ADDR_ADD, true) +DO_VSTR_SG(vstrw_sg_wb_uw, stl, 4, uint32_t, ADDR_ADD, true) +DO_VSTR64_SG(vstrd_sg_wb_ud, ADDR_ADD, true) /* * The mergemask(D, R, M) macro performs the operation "*D = R" but diff --git a/target/arm/translate-mve.c b/target/arm/translate-mve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-mve.c +++ b/target/arm/translate-mve.c @@ -XXX,XX +XXX,XX @@ static bool trans_VSTR_sg(DisasContext *s, arg_vldst_sg *a) #undef F +static bool do_ldst_sg_imm(DisasContext *s, arg_vldst_sg_imm *a, + MVEGenLdStSGFn *fn, unsigned msize) +{ + uint32_t offset; + TCGv_ptr qd, qm; + + if (!dc_isar_feature(aa32_mve, s) || + !mve_check_qreg_bank(s, a->qd | a->qm) || + !fn) { + return false; + } + + if (!mve_eci_check(s) || !vfp_access_check(s)) { + return true; + } + + offset = a->imm << msize; + if (!a->a) { + offset = -offset; + } + + qd = mve_qreg_ptr(a->qd); + qm = mve_qreg_ptr(a->qm); + fn(cpu_env, qd, qm, tcg_constant_i32(offset)); + tcg_temp_free_ptr(qd); + tcg_temp_free_ptr(qm); + mve_update_eci(s); + return true; +} + +static bool trans_VLDRW_sg_imm(DisasContext *s, arg_vldst_sg_imm *a) +{ + static MVEGenLdStSGFn * const fns[] = { + gen_helper_mve_vldrw_sg_uw, + gen_helper_mve_vldrw_sg_wb_uw, + }; + if (a->qd == a->qm) { + return false; /* UNPREDICTABLE */ + } + return do_ldst_sg_imm(s, a, fns[a->w], MO_32); +} + +static bool trans_VLDRD_sg_imm(DisasContext *s, arg_vldst_sg_imm *a) +{ + static MVEGenLdStSGFn * const fns[] = { + gen_helper_mve_vldrd_sg_ud, + gen_helper_mve_vldrd_sg_wb_ud, + }; + if (a->qd == a->qm) { + return false; /* UNPREDICTABLE */ + } + return do_ldst_sg_imm(s, a, fns[a->w], MO_64); +} + +static bool trans_VSTRW_sg_imm(DisasContext *s, arg_vldst_sg_imm *a) +{ + static MVEGenLdStSGFn * const fns[] = { + gen_helper_mve_vstrw_sg_uw, + gen_helper_mve_vstrw_sg_wb_uw, + }; + return do_ldst_sg_imm(s, a, fns[a->w], MO_32); +} + +static bool trans_VSTRD_sg_imm(DisasContext *s, arg_vldst_sg_imm *a) +{ + static MVEGenLdStSGFn * const fns[] = { + gen_helper_mve_vstrd_sg_ud, + gen_helper_mve_vstrd_sg_wb_ud, + }; + return do_ldst_sg_imm(s, a, fns[a->w], MO_64); +} + static bool trans_VDUP(DisasContext *s, arg_VDUP *a) { TCGv_ptr qd; -- 2.20.1 Implement the MVE interleaving load/store functions VLD2, VLD4, VST2 and VST4. VLD2 loads 16 bytes of data from memory and writes to 2 consecutive Qregs; VLD4 loads 16 bytes of data from memory and writes to 4 consecutive Qregs. The 'pattern' field in the encoding determines the offset into memory which is accessed and also which elements in the Qregs are written to. (The intention is that a sequence of four consecutive VLD4 with different pattern values performs a complete de-interleaving load of 64 bytes into all elements of the 4 Qregs.) VST2 and VST4 do the same, but for stores. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/helper-mve.h | 48 ++++++ target/arm/mve.decode | 11 ++ target/arm/mve_helper.c | 342 +++++++++++++++++++++++++++++++++++++ target/arm/translate-mve.c | 94 ++++++++++ 4 files changed, 495 insertions(+) diff --git a/target/arm/helper-mve.h b/target/arm/helper-mve.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/helper-mve.h +++ b/target/arm/helper-mve.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_4(mve_vldrd_sg_wb_ud, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) DEF_HELPER_FLAGS_4(mve_vstrw_sg_wb_uw, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) DEF_HELPER_FLAGS_4(mve_vstrd_sg_wb_ud, TCG_CALL_NO_WG, void, env, ptr, ptr, i32) +DEF_HELPER_FLAGS_3(mve_vld20b, TCG_CALL_NO_WG, void, env, i32, i32) +DEF_HELPER_FLAGS_3(mve_vld20h, TCG_CALL_NO_WG, void, env, i32, i32) +DEF_HELPER_FLAGS_3(mve_vld20w, TCG_CALL_NO_WG, void, env, i32, i32) + +DEF_HELPER_FLAGS_3(mve_vld21b, TCG_CALL_NO_WG, void, env, i32, i32) +DEF_HELPER_FLAGS_3(mve_vld21h, TCG_CALL_NO_WG, void, env, i32, i32) +DEF_HELPER_FLAGS_3(mve_vld21w, TCG_CALL_NO_WG, void, env, i32, i32) + +DEF_HELPER_FLAGS_3(mve_vld40b, TCG_CALL_NO_WG, void, env, i32, i32) +DEF_HELPER_FLAGS_3(mve_vld40h, TCG_CALL_NO_WG, void, env, i32, i32) +DEF_HELPER_FLAGS_3(mve_vld40w, TCG_CALL_NO_WG, void, env, i32, i32) + +DEF_HELPER_FLAGS_3(mve_vld41b, TCG_CALL_NO_WG, void, env, i32, i32) +DEF_HELPER_FLAGS_3(mve_vld41h, TCG_CALL_NO_WG, void, env, i32, i32) +DEF_HELPER_FLAGS_3(mve_vld41w, TCG_CALL_NO_WG, void, env, i32, i32) + +DEF_HELPER_FLAGS_3(mve_vld42b, TCG_CALL_NO_WG, void, env, i32, i32) +DEF_HELPER_FLAGS_3(mve_vld42h, TCG_CALL_NO_WG, void, env, i32, i32) +DEF_HELPER_FLAGS_3(mve_vld42w, TCG_CALL_NO_WG, void, env, i32, i32) + +DEF_HELPER_FLAGS_3(mve_vld43b, TCG_CALL_NO_WG, void, env, i32, i32) +DEF_HELPER_FLAGS_3(mve_vld43h, TCG_CALL_NO_WG, void, env, i32, i32) +DEF_HELPER_FLAGS_3(mve_vld43w, TCG_CALL_NO_WG, void, env, i32, i32) + +DEF_HELPER_FLAGS_3(mve_vst20b, TCG_CALL_NO_WG, void, env, i32, i32) +DEF_HELPER_FLAGS_3(mve_vst20h, TCG_CALL_NO_WG, void, env, i32, i32) +DEF_HELPER_FLAGS_3(mve_vst20w, TCG_CALL_NO_WG, void, env, i32, i32) + +DEF_HELPER_FLAGS_3(mve_vst21b, TCG_CALL_NO_WG, void, env, i32, i32) +DEF_HELPER_FLAGS_3(mve_vst21h, TCG_CALL_NO_WG, void, env, i32, i32) +DEF_HELPER_FLAGS_3(mve_vst21w, TCG_CALL_NO_WG, void, env, i32, i32) + +DEF_HELPER_FLAGS_3(mve_vst40b, TCG_CALL_NO_WG, void, env, i32, i32) +DEF_HELPER_FLAGS_3(mve_vst40h, TCG_CALL_NO_WG, void, env, i32, i32) +DEF_HELPER_FLAGS_3(mve_vst40w, TCG_CALL_NO_WG, void, env, i32, i32) + +DEF_HELPER_FLAGS_3(mve_vst41b, TCG_CALL_NO_WG, void, env, i32, i32) +DEF_HELPER_FLAGS_3(mve_vst41h, TCG_CALL_NO_WG, void, env, i32, i32) +DEF_HELPER_FLAGS_3(mve_vst41w, TCG_CALL_NO_WG, void, env, i32, i32) + +DEF_HELPER_FLAGS_3(mve_vst42b, TCG_CALL_NO_WG, void, env, i32, i32) +DEF_HELPER_FLAGS_3(mve_vst42h, TCG_CALL_NO_WG, void, env, i32, i32) +DEF_HELPER_FLAGS_3(mve_vst42w, TCG_CALL_NO_WG, void, env, i32, i32) + +DEF_HELPER_FLAGS_3(mve_vst43b, TCG_CALL_NO_WG, void, env, i32, i32) +DEF_HELPER_FLAGS_3(mve_vst43h, TCG_CALL_NO_WG, void, env, i32, i32) +DEF_HELPER_FLAGS_3(mve_vst43w, TCG_CALL_NO_WG, void, env, i32, i32) + DEF_HELPER_FLAGS_3(mve_vdup, TCG_CALL_NO_WG, void, env, ptr, i32) DEF_HELPER_FLAGS_4(mve_vidupb, TCG_CALL_NO_WG, i32, env, ptr, i32, i32) diff --git a/target/arm/mve.decode b/target/arm/mve.decode index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve.decode +++ b/target/arm/mve.decode @@ -XXX,XX +XXX,XX @@ &vabav qn qm rda size &vldst_sg qd qm rn size msize os &vldst_sg_imm qd qm a w imm +&vldst_il qd rn size pat w # scatter-gather memory size is in bits 6:4 %sg_msize 6:1 4:1 @@ -XXX,XX +XXX,XX @@ @vldst_sg_imm .... .... a:1 . w:1 . .... .... .... . imm:7 &vldst_sg_imm \ qd=%qd qm=%qn +# Deinterleaving load/interleaving store +@vldst_il .... .... .. w:1 . rn:4 .... ... size:2 pat:2 ..... &vldst_il \ + qd=%qd + @1op .... .... .... size:2 .. .... .... .... .... &1op qd=%qd qm=%qm @1op_nosz .... .... .... .... .... .... .... .... &1op qd=%qd qm=%qm size=0 @2op .... .... .. size:2 .... .... .... .... .... &2op qd=%qd qm=%qm qn=%qn @@ -XXX,XX +XXX,XX @@ VLDRD_sg_imm 111 1 1101 ... 1 ... 0 ... 1 1111 .... .... @vldst_sg_imm VSTRW_sg_imm 111 1 1101 ... 0 ... 0 ... 1 1110 .... .... @vldst_sg_imm VSTRD_sg_imm 111 1 1101 ... 0 ... 0 ... 1 1111 .... .... @vldst_sg_imm +# deinterleaving loads/interleaving stores +VLD2 1111 1100 1 .. 1 .... ... 1 111 .. .. 00000 @vldst_il +VLD4 1111 1100 1 .. 1 .... ... 1 111 .. .. 00001 @vldst_il +VST2 1111 1100 1 .. 0 .... ... 1 111 .. .. 00000 @vldst_il +VST4 1111 1100 1 .. 0 .... ... 1 111 .. .. 00001 @vldst_il + # Moves between 2 32-bit vector lanes and 2 general purpose registers VMOV_to_2gp 1110 1100 0 . 00 rt2:4 ... 0 1111 000 idx:1 rt:4 qd=%qd VMOV_from_2gp 1110 1100 0 . 01 rt2:4 ... 0 1111 000 idx:1 rt:4 qd=%qd diff --git a/target/arm/mve_helper.c b/target/arm/mve_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/mve_helper.c +++ b/target/arm/mve_helper.c @@ -XXX,XX +XXX,XX @@ DO_VLDR64_SG(vldrd_sg_wb_ud, ADDR_ADD, true) DO_VSTR_SG(vstrw_sg_wb_uw, stl, 4, uint32_t, ADDR_ADD, true) DO_VSTR64_SG(vstrd_sg_wb_ud, ADDR_ADD, true) +/* + * Deinterleaving loads/interleaving stores. + * + * For these helpers we are passed the index of the first Qreg + * (VLD2/VST2 will also access Qn+1, VLD4/VST4 access Qn .. Qn+3) + * and the value of the base address register Rn. + * The helpers are specialized for pattern and element size, so + * for instance vld42h is VLD4 with pattern 2, element size MO_16. + * + * These insns are beatwise but not predicated, so we must honour ECI, + * but need not look at mve_element_mask(). + * + * The pseudocode implements these insns with multiple memory accesses + * of the element size, but rules R_VVVG and R_FXDM permit us to make + * one 32-bit memory access per beat. + */ +#define DO_VLD4B(OP, O1, O2, O3, O4) \ + void HELPER(mve_##OP)(CPUARMState *env, uint32_t qnidx, \ + uint32_t base) \ + { \ + int beat, e; \ + uint16_t mask = mve_eci_mask(env); \ + static const uint8_t off[4] = { O1, O2, O3, O4 }; \ + uint32_t addr, data; \ + for (beat = 0; beat < 4; beat++, mask >>= 4) { \ + if ((mask & 1) == 0) { \ + /* ECI says skip this beat */ \ + continue; \ + } \ + addr = base + off[beat] * 4; \ + data = cpu_ldl_le_data_ra(env, addr, GETPC()); \ + for (e = 0; e < 4; e++, data >>= 8) { \ + uint8_t *qd = (uint8_t *)aa32_vfp_qreg(env, qnidx + e); \ + qd[H1(off[beat])] = data; \ + } \ + } \ + } + +#define DO_VLD4H(OP, O1, O2) \ + void HELPER(mve_##OP)(CPUARMState *env, uint32_t qnidx, \ + uint32_t base) \ + { \ + int beat; \ + uint16_t mask = mve_eci_mask(env); \ + static const uint8_t off[4] = { O1, O1, O2, O2 }; \ + uint32_t addr, data; \ + int y; /* y counts 0 2 0 2 */ \ + uint16_t *qd; \ + for (beat = 0, y = 0; beat < 4; beat++, mask >>= 4, y ^= 2) { \ + if ((mask & 1) == 0) { \ + /* ECI says skip this beat */ \ + continue; \ + } \ + addr = base + off[beat] * 8 + (beat & 1) * 4; \ + data = cpu_ldl_le_data_ra(env, addr, GETPC()); \ + qd = (uint16_t *)aa32_vfp_qreg(env, qnidx + y); \ + qd[H2(off[beat])] = data; \ + data >>= 16; \ + qd = (uint16_t *)aa32_vfp_qreg(env, qnidx + y + 1); \ + qd[H2(off[beat])] = data; \ + } \ + } + +#define DO_VLD4W(OP, O1, O2, O3, O4) \ + void HELPER(mve_##OP)(CPUARMState *env, uint32_t qnidx, \ + uint32_t base) \ + { \ + int beat; \ + uint16_t mask = mve_eci_mask(env); \ + static const uint8_t off[4] = { O1, O2, O3, O4 }; \ + uint32_t addr, data; \ + uint32_t *qd; \ + int y; \ + for (beat = 0; beat < 4; beat++, mask >>= 4) { \ + if ((mask & 1) == 0) { \ + /* ECI says skip this beat */ \ + continue; \ + } \ + addr = base + off[beat] * 4; \ + data = cpu_ldl_le_data_ra(env, addr, GETPC()); \ + y = (beat + (O1 & 2)) & 3; \ + qd = (uint32_t *)aa32_vfp_qreg(env, qnidx + y); \ + qd[H4(off[beat] >> 2)] = data; \ + } \ + } + +DO_VLD4B(vld40b, 0, 1, 10, 11) +DO_VLD4B(vld41b, 2, 3, 12, 13) +DO_VLD4B(vld42b, 4, 5, 14, 15) +DO_VLD4B(vld43b, 6, 7, 8, 9) + +DO_VLD4H(vld40h, 0, 5) +DO_VLD4H(vld41h, 1, 6) +DO_VLD4H(vld42h, 2, 7) +DO_VLD4H(vld43h, 3, 4) + +DO_VLD4W(vld40w, 0, 1, 10, 11) +DO_VLD4W(vld41w, 2, 3, 12, 13) +DO_VLD4W(vld42w, 4, 5, 14, 15) +DO_VLD4W(vld43w, 6, 7, 8, 9) + +#define DO_VLD2B(OP, O1, O2, O3, O4) \ + void HELPER(mve_##OP)(CPUARMState *env, uint32_t qnidx, \ + uint32_t base) \ + { \ + int beat, e; \ + uint16_t mask = mve_eci_mask(env); \ + static const uint8_t off[4] = { O1, O2, O3, O4 }; \ + uint32_t addr, data; \ + uint8_t *qd; \ + for (beat = 0; beat < 4; beat++, mask >>= 4) { \ + if ((mask & 1) == 0) { \ + /* ECI says skip this beat */ \ + continue; \ + } \ + addr = base + off[beat] * 2; \ + data = cpu_ldl_le_data_ra(env, addr, GETPC()); \ + for (e = 0; e < 4; e++, data >>= 8) { \ + qd = (uint8_t *)aa32_vfp_qreg(env, qnidx + (e & 1)); \ + qd[H1(off[beat] + (e >> 1))] = data; \ + } \ + } \ + } + +#define DO_VLD2H(OP, O1, O2, O3, O4) \ + void HELPER(mve_##OP)(CPUARMState *env, uint32_t qnidx, \ + uint32_t base) \ + { \ + int beat; \ + uint16_t mask = mve_eci_mask(env); \ + static const uint8_t off[4] = { O1, O2, O3, O4 }; \ + uint32_t addr, data; \ + int e; \ + uint16_t *qd; \ + for (beat = 0; beat < 4; beat++, mask >>= 4) { \ + if ((mask & 1) == 0) { \ + /* ECI says skip this beat */ \ + continue; \ + } \ + addr = base + off[beat] * 4; \ + data = cpu_ldl_le_data_ra(env, addr, GETPC()); \ + for (e = 0; e < 2; e++, data >>= 16) { \ + qd = (uint16_t *)aa32_vfp_qreg(env, qnidx + e); \ + qd[H2(off[beat])] = data; \ + } \ + } \ + } + +#define DO_VLD2W(OP, O1, O2, O3, O4) \ + void HELPER(mve_##OP)(CPUARMState *env, uint32_t qnidx, \ + uint32_t base) \ + { \ + int beat; \ + uint16_t mask = mve_eci_mask(env); \ + static const uint8_t off[4] = { O1, O2, O3, O4 }; \ + uint32_t addr, data; \ + uint32_t *qd; \ + for (beat = 0; beat < 4; beat++, mask >>= 4) { \ + if ((mask & 1) == 0) { \ + /* ECI says skip this beat */ \ + continue; \ + } \ + addr = base + off[beat]; \ + data = cpu_ldl_le_data_ra(env, addr, GETPC()); \ + qd = (uint32_t *)aa32_vfp_qreg(env, qnidx + (beat & 1)); \ + qd[H4(off[beat] >> 3)] = data; \ + } \ + } + +DO_VLD2B(vld20b, 0, 2, 12, 14) +DO_VLD2B(vld21b, 4, 6, 8, 10) + +DO_VLD2H(vld20h, 0, 1, 6, 7) +DO_VLD2H(vld21h, 2, 3, 4, 5) + +DO_VLD2W(vld20w, 0, 4, 24, 28) +DO_VLD2W(vld21w, 8, 12, 16, 20) + +#define DO_VST4B(OP, O1, O2, O3, O4) \ + void HELPER(mve_##OP)(CPUARMState *env, uint32_t qnidx, \ + uint32_t base) \ + { \ + int beat, e; \ + uint16_t mask = mve_eci_mask(env); \ + static const uint8_t off[4] = { O1, O2, O3, O4 }; \ + uint32_t addr, data; \ + for (beat = 0; beat < 4; beat++, mask >>= 4) { \ + if ((mask & 1) == 0) { \ + /* ECI says skip this beat */ \ + continue; \ + } \ + addr = base + off[beat] * 4; \ + data = 0; \ + for (e = 3; e >= 0; e--) { \ + uint8_t *qd = (uint8_t *)aa32_vfp_qreg(env, qnidx + e); \ + data = (data << 8) | qd[H1(off[beat])]; \ + } \ + cpu_stl_le_data_ra(env, addr, data, GETPC()); \ + } \ + } + +#define DO_VST4H(OP, O1, O2) \ + void HELPER(mve_##OP)(CPUARMState *env, uint32_t qnidx, \ + uint32_t base) \ + { \ + int beat; \ + uint16_t mask = mve_eci_mask(env); \ + static const uint8_t off[4] = { O1, O1, O2, O2 }; \ + uint32_t addr, data; \ + int y; /* y counts 0 2 0 2 */ \ + uint16_t *qd; \ + for (beat = 0, y = 0; beat < 4; beat++, mask >>= 4, y ^= 2) { \ + if ((mask & 1) == 0) { \ + /* ECI says skip this beat */ \ + continue; \ + } \ + addr = base + off[beat] * 8 + (beat & 1) * 4; \ + qd = (uint16_t *)aa32_vfp_qreg(env, qnidx + y); \ + data = qd[H2(off[beat])]; \ + qd = (uint16_t *)aa32_vfp_qreg(env, qnidx + y + 1); \ + data |= qd[H2(off[beat])] << 16; \ + cpu_stl_le_data_ra(env, addr, data, GETPC()); \ + } \ + } + +#define DO_VST4W(OP, O1, O2, O3, O4) \ + void HELPER(mve_##OP)(CPUARMState *env, uint32_t qnidx, \ + uint32_t base) \ + { \ + int beat; \ + uint16_t mask = mve_eci_mask(env); \ + static const uint8_t off[4] = { O1, O2, O3, O4 }; \ + uint32_t addr, data; \ + uint32_t *qd; \ + int y; \ + for (beat = 0; beat < 4; beat++, mask >>= 4) { \ + if ((mask & 1) == 0) { \ + /* ECI says skip this beat */ \ + continue; \ + } \ + addr = base + off[beat] * 4; \ + y = (beat + (O1 & 2)) & 3; \ + qd = (uint32_t *)aa32_vfp_qreg(env, qnidx + y); \ + data = qd[H4(off[beat] >> 2)]; \ + cpu_stl_le_data_ra(env, addr, data, GETPC()); \ + } \ + } + +DO_VST4B(vst40b, 0, 1, 10, 11) +DO_VST4B(vst41b, 2, 3, 12, 13) +DO_VST4B(vst42b, 4, 5, 14, 15) +DO_VST4B(vst43b, 6, 7, 8, 9) + +DO_VST4H(vst40h, 0, 5) +DO_VST4H(vst41h, 1, 6) +DO_VST4H(vst42h, 2, 7) +DO_VST4H(vst43h, 3, 4) + +DO_VST4W(vst40w, 0, 1, 10, 11) +DO_VST4W(vst41w, 2, 3, 12, 13) +DO_VST4W(vst42w, 4, 5, 14, 15) +DO_VST4W(vst43w, 6, 7, 8, 9) + +#define DO_VST2B(OP, O1, O2, O3, O4) \ + void HELPER(mve_##OP)(CPUARMState *env, uint32_t qnidx, \ + uint32_t base) \ + { \ + int beat, e; \ + uint16_t mask = mve_eci_mask(env); \ + static const uint8_t off[4] = { O1, O2, O3, O4 }; \ + uint32_t addr, data; \ + uint8_t *qd; \ + for (beat = 0; beat < 4; beat++, mask >>= 4) { \ + if ((mask & 1) == 0) { \ + /* ECI says skip this beat */ \ + continue; \ + } \ + addr = base + off[beat] * 2; \ + data = 0; \ + for (e = 3; e >= 0; e--) { \ + qd = (uint8_t *)aa32_vfp_qreg(env, qnidx + (e & 1)); \ + data = (data << 8) | qd[H1(off[beat] + (e >> 1))]; \ + } \ + cpu_stl_le_data_ra(env, addr, data, GETPC()); \ + } \ + } + +#define DO_VST2H(OP, O1, O2, O3, O4) \ + void HELPER(mve_##OP)(CPUARMState *env, uint32_t qnidx, \ + uint32_t base) \ + { \ + int beat; \ + uint16_t mask = mve_eci_mask(env); \ + static const uint8_t off[4] = { O1, O2, O3, O4 }; \ + uint32_t addr, data; \ + int e; \ + uint16_t *qd; \ + for (beat = 0; beat < 4; beat++, mask >>= 4) { \ + if ((mask & 1) == 0) { \ + /* ECI says skip this beat */ \ + continue; \ + } \ + addr = base + off[beat] * 4; \ + data = 0; \ + for (e = 1; e >= 0; e--) { \ + qd = (uint16_t *)aa32_vfp_qreg(env, qnidx + e); \ + data = (data << 16) | qd[H2(off[beat])]; \ + } \ + cpu_stl_le_data_ra(env, addr, data, GETPC()); \ + } \ + } + +#define DO_VST2W(OP, O1, O2, O3, O4) \ + void HELPER(mve_##OP)(CPUARMState *env, uint32_t qnidx, \ + uint32_t base) \ + { \ + int beat; \ + uint16_t mask = mve_eci_mask(env); \ + static const uint8_t off[4] = { O1, O2, O3, O4 }; \ + uint32_t addr, data; \ + uint32_t *qd; \ + for (beat = 0; beat < 4; beat++, mask >>= 4) { \ + if ((mask & 1) == 0) { \ + /* ECI says skip this beat */ \ + continue; \ + } \ + addr = base + off[beat]; \ + qd = (uint32_t *)aa32_vfp_qreg(env, qnidx + (beat & 1)); \ + data = qd[H4(off[beat] >> 3)]; \ + cpu_stl_le_data_ra(env, addr, data, GETPC()); \ + } \ + } + +DO_VST2B(vst20b, 0, 2, 12, 14) +DO_VST2B(vst21b, 4, 6, 8, 10) + +DO_VST2H(vst20h, 0, 1, 6, 7) +DO_VST2H(vst21h, 2, 3, 4, 5) + +DO_VST2W(vst20w, 0, 4, 24, 28) +DO_VST2W(vst21w, 8, 12, 16, 20) + /* * The mergemask(D, R, M) macro performs the operation "*D = R" but * storing only the bytes which correspond to 1 bits in M, diff --git a/target/arm/translate-mve.c b/target/arm/translate-mve.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/translate-mve.c +++ b/target/arm/translate-mve.c @@ -XXX,XX +XXX,XX @@ static inline int vidup_imm(DisasContext *s, int x) typedef void MVEGenLdStFn(TCGv_ptr, TCGv_ptr, TCGv_i32); typedef void MVEGenLdStSGFn(TCGv_ptr, TCGv_ptr, TCGv_ptr, TCGv_i32); +typedef void MVEGenLdStIlFn(TCGv_ptr, TCGv_i32, TCGv_i32); typedef void MVEGenOneOpFn(TCGv_ptr, TCGv_ptr, TCGv_ptr); typedef void MVEGenTwoOpFn(TCGv_ptr, TCGv_ptr, TCGv_ptr, TCGv_ptr); typedef void MVEGenTwoOpScalarFn(TCGv_ptr, TCGv_ptr, TCGv_ptr, TCGv_i32); @@ -XXX,XX +XXX,XX @@ static bool trans_VSTRD_sg_imm(DisasContext *s, arg_vldst_sg_imm *a) return do_ldst_sg_imm(s, a, fns[a->w], MO_64); } +static bool do_vldst_il(DisasContext *s, arg_vldst_il *a, MVEGenLdStIlFn *fn, + int addrinc) +{ + TCGv_i32 rn; + + if (!dc_isar_feature(aa32_mve, s) || + !mve_check_qreg_bank(s, a->qd) || + !fn || (a->rn == 13 && a->w) || a->rn == 15) { + /* Variously UNPREDICTABLE or UNDEF or related-encoding */ + return false; + } + if (!mve_eci_check(s) || !vfp_access_check(s)) { + return true; + } + + rn = load_reg(s, a->rn); + /* + * We pass the index of Qd, not a pointer, because the helper must + * access multiple Q registers starting at Qd and working up. + */ + fn(cpu_env, tcg_constant_i32(a->qd), rn); + + if (a->w) { + tcg_gen_addi_i32(rn, rn, addrinc); + store_reg(s, a->rn, rn); + } else { + tcg_temp_free_i32(rn); + } + mve_update_and_store_eci(s); + return true; +} + +/* This macro is just to make the arrays more compact in these functions */ +#define F(N) gen_helper_mve_##N + +static bool trans_VLD2(DisasContext *s, arg_vldst_il *a) +{ + static MVEGenLdStIlFn * const fns[4][4] = { + { F(vld20b), F(vld20h), F(vld20w), NULL, }, + { F(vld21b), F(vld21h), F(vld21w), NULL, }, + { NULL, NULL, NULL, NULL }, + { NULL, NULL, NULL, NULL }, + }; + if (a->qd > 6) { + return false; + } + return do_vldst_il(s, a, fns[a->pat][a->size], 32); +} + +static bool trans_VLD4(DisasContext *s, arg_vldst_il *a) +{ + static MVEGenLdStIlFn * const fns[4][4] = { + { F(vld40b), F(vld40h), F(vld40w), NULL, }, + { F(vld41b), F(vld41h), F(vld41w), NULL, }, + { F(vld42b), F(vld42h), F(vld42w), NULL, }, + { F(vld43b), F(vld43h), F(vld43w), NULL, }, + }; + if (a->qd > 4) { + return false; + } + return do_vldst_il(s, a, fns[a->pat][a->size], 64); +} + +static bool trans_VST2(DisasContext *s, arg_vldst_il *a) +{ + static MVEGenLdStIlFn * const fns[4][4] = { + { F(vst20b), F(vst20h), F(vst20w), NULL, }, + { F(vst21b), F(vst21h), F(vst21w), NULL, }, + { NULL, NULL, NULL, NULL }, + { NULL, NULL, NULL, NULL }, + }; + if (a->qd > 6) { + return false; + } + return do_vldst_il(s, a, fns[a->pat][a->size], 32); +} + +static bool trans_VST4(DisasContext *s, arg_vldst_il *a) +{ + static MVEGenLdStIlFn * const fns[4][4] = { + { F(vst40b), F(vst40h), F(vst40w), NULL, }, + { F(vst41b), F(vst41h), F(vst41w), NULL, }, + { F(vst42b), F(vst42h), F(vst42w), NULL, }, + { F(vst43b), F(vst43h), F(vst43w), NULL, }, + }; + if (a->qd > 4) { + return false; + } + return do_vldst_il(s, a, fns[a->pat][a->size], 64); +} + +#undef F + static bool trans_VDUP(DisasContext *s, arg_VDUP *a) { TCGv_ptr qd; -- 2.20.1 We're about to make a code change to the sdiv and udiv helper functions, so first fix their indentation and coding style. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210730151636.17254-2-peter.maydell@linaro.org --- target/arm/helper.c | 15 +++++++++------ 1 file changed, 9 insertions(+), 6 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 @@ uint32_t HELPER(uxtb16)(uint32_t x) int32_t HELPER(sdiv)(int32_t num, int32_t den) { - if (den == 0) - return 0; - if (num == INT_MIN && den == -1) - return INT_MIN; + if (den == 0) { + return 0; + } + if (num == INT_MIN && den == -1) { + return INT_MIN; + } return num / den; } uint32_t HELPER(udiv)(uint32_t num, uint32_t den) { - if (den == 0) - return 0; + if (den == 0) { + return 0; + } return num / den; } -- 2.20.1 Unlike A-profile, for M-profile the UDIV and SDIV insns can be configured to raise an exception on division by zero, using the CCR DIV_0_TRP bit. Implement support for setting this bit by making the helper functions raise the appropriate exception. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210730151636.17254-3-peter.maydell@linaro.org --- target/arm/cpu.h | 1 + target/arm/helper.h | 4 ++-- target/arm/helper.c | 19 +++++++++++++++++-- target/arm/m_helper.c | 4 ++++ target/arm/translate.c | 4 ++-- 5 files changed, 26 insertions(+), 6 deletions(-) diff --git a/target/arm/cpu.h b/target/arm/cpu.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/cpu.h +++ b/target/arm/cpu.h @@ -XXX,XX +XXX,XX @@ #define EXCP_LAZYFP 20 /* v7M fault during lazy FP stacking */ #define EXCP_LSERR 21 /* v8M LSERR SecureFault */ #define EXCP_UNALIGNED 22 /* v7M UNALIGNED UsageFault */ +#define EXCP_DIVBYZERO 23 /* v7M DIVBYZERO UsageFault */ /* NB: add new EXCP_ defines to the array in arm_log_exception() too */ #define ARMV7M_EXCP_RESET 1 diff --git a/target/arm/helper.h b/target/arm/helper.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/helper.h +++ b/target/arm/helper.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_3(add_saturate, i32, env, i32, i32) DEF_HELPER_3(sub_saturate, i32, env, i32, i32) DEF_HELPER_3(add_usaturate, i32, env, i32, i32) DEF_HELPER_3(sub_usaturate, i32, env, i32, i32) -DEF_HELPER_FLAGS_2(sdiv, TCG_CALL_NO_RWG_SE, s32, s32, s32) -DEF_HELPER_FLAGS_2(udiv, TCG_CALL_NO_RWG_SE, i32, i32, i32) +DEF_HELPER_FLAGS_3(sdiv, TCG_CALL_NO_RWG, s32, env, s32, s32) +DEF_HELPER_FLAGS_3(udiv, TCG_CALL_NO_RWG, i32, env, i32, i32) DEF_HELPER_FLAGS_1(rbit, TCG_CALL_NO_RWG_SE, i32, i32) #define PAS_OP(pfx) \ 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 @@ uint32_t HELPER(sxtb16)(uint32_t x) return res; } +static void handle_possible_div0_trap(CPUARMState *env, uintptr_t ra) +{ + /* + * Take a division-by-zero exception if necessary; otherwise return + * to get the usual non-trapping division behaviour (result of 0) + */ + if (arm_feature(env, ARM_FEATURE_M) + && (env->v7m.ccr[env->v7m.secure] & R_V7M_CCR_DIV_0_TRP_MASK)) { + raise_exception_ra(env, EXCP_DIVBYZERO, 0, 1, ra); + } +} + uint32_t HELPER(uxtb16)(uint32_t x) { uint32_t res; @@ -XXX,XX +XXX,XX @@ uint32_t HELPER(uxtb16)(uint32_t x) return res; } -int32_t HELPER(sdiv)(int32_t num, int32_t den) +int32_t HELPER(sdiv)(CPUARMState *env, int32_t num, int32_t den) { if (den == 0) { + handle_possible_div0_trap(env, GETPC()); return 0; } if (num == INT_MIN && den == -1) { @@ -XXX,XX +XXX,XX @@ int32_t HELPER(sdiv)(int32_t num, int32_t den) return num / den; } -uint32_t HELPER(udiv)(uint32_t num, uint32_t den) +uint32_t HELPER(udiv)(CPUARMState *env, uint32_t num, uint32_t den) { if (den == 0) { + handle_possible_div0_trap(env, GETPC()); return 0; } return num / den; @@ -XXX,XX +XXX,XX @@ void arm_log_exception(int idx) [EXCP_LAZYFP] = "v7M exception during lazy FP stacking", [EXCP_LSERR] = "v8M LSERR UsageFault", [EXCP_UNALIGNED] = "v7M UNALIGNED UsageFault", + [EXCP_DIVBYZERO] = "v7M DIVBYZERO UsageFault", }; if (idx >= 0 && idx < ARRAY_SIZE(excnames)) { 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 @@ void arm_v7m_cpu_do_interrupt(CPUState *cs) armv7m_nvic_set_pending(env->nvic, ARMV7M_EXCP_USAGE, env->v7m.secure); env->v7m.cfsr[env->v7m.secure] |= R_V7M_CFSR_UNALIGNED_MASK; break; + case EXCP_DIVBYZERO: + armv7m_nvic_set_pending(env->nvic, ARMV7M_EXCP_USAGE, env->v7m.secure); + env->v7m.cfsr[env->v7m.secure] |= R_V7M_CFSR_DIVBYZERO_MASK; + break; case EXCP_SWI: /* The PC already points to the next instruction. */ armv7m_nvic_set_pending(env->nvic, ARMV7M_EXCP_SVC, env->v7m.secure); 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 bool op_div(DisasContext *s, arg_rrr *a, bool u) t1 = load_reg(s, a->rn); t2 = load_reg(s, a->rm); if (u) { - gen_helper_udiv(t1, t1, t2); + gen_helper_udiv(t1, cpu_env, t1, t2); } else { - gen_helper_sdiv(t1, t1, t2); + gen_helper_sdiv(t1, cpu_env, t1, t2); } tcg_temp_free_i32(t2); store_reg(s, a->rd, t1); -- 2.20.1 From: Hamza Mahfooz <someguy@effective-light.com> As per commit 5626f8c6d468 ("rcu: Add automatically released rcu_read_lock variants"), RCU_READ_LOCK_GUARD() should be used instead of rcu_read_{un}lock(). Signed-off-by: Hamza Mahfooz <someguy@effective-light.com> Reviewed-by: Paolo Bonzini <pbonzini@redhat.com> Message-id: 20210727235201.11491-1-someguy@effective-light.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/kvm.c | 17 ++++++++--------- 1 file changed, 8 insertions(+), 9 deletions(-) diff --git a/target/arm/kvm.c b/target/arm/kvm.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/kvm.c +++ b/target/arm/kvm.c @@ -XXX,XX +XXX,XX @@ int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route, hwaddr xlat, len, doorbell_gpa; MemoryRegionSection mrs; MemoryRegion *mr; - int ret = 1; if (as == &address_space_memory) { return 0; @@ -XXX,XX +XXX,XX @@ int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route, /* MSI doorbell address is translated by an IOMMU */ - rcu_read_lock(); + RCU_READ_LOCK_GUARD(); + mr = address_space_translate(as, address, &xlat, &len, true, MEMTXATTRS_UNSPECIFIED); + if (!mr) { - goto unlock; + return 1; } + mrs = memory_region_find(mr, xlat, 1); + if (!mrs.mr) { - goto unlock; + return 1; } doorbell_gpa = mrs.offset_within_address_space; @@ -XXX,XX +XXX,XX @@ int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route, trace_kvm_arm_fixup_msi_route(address, doorbell_gpa); - ret = 0; - -unlock: - rcu_read_unlock(); - return ret; + return 0; } int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route, -- 2.20.1 From: Jan Luebbe <jlu@pengutronix.de> Break events are currently only handled by chardev/char-serial.c, so we just ignore errors, which results in no behaviour change for other chardevs. Signed-off-by: Jan Luebbe <jlu@pengutronix.de> Message-id: 20210806144700.3751979-1-jlu@pengutronix.de Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/char/pl011.c | 6 ++++++ 1 file changed, 6 insertions(+) diff --git a/hw/char/pl011.c b/hw/char/pl011.c index XXXXXXX..XXXXXXX 100644 --- a/hw/char/pl011.c +++ b/hw/char/pl011.c @@ -XXX,XX +XXX,XX @@ #include "hw/qdev-properties-system.h" #include "migration/vmstate.h" #include "chardev/char-fe.h" +#include "chardev/char-serial.h" #include "qemu/log.h" #include "qemu/module.h" #include "trace.h" @@ -XXX,XX +XXX,XX @@ static void pl011_write(void *opaque, hwaddr offset, s->read_count = 0; s->read_pos = 0; } + if ((s->lcr ^ value) & 0x1) { + int break_enable = value & 0x1; + qemu_chr_fe_ioctl(&s->chr, CHR_IOCTL_SERIAL_SET_BREAK, + &break_enable); + } s->lcr = value; pl011_set_read_trigger(s); break; -- 2.20.1 From: Guenter Roeck <linux@roeck-us.net> Instantiate SAI1/2/3 and ASRC as unimplemented devices to avoid random Linux kernel crashes, such as Unhandled fault: external abort on non-linefetch (0x808) at 0xd1580010 pgd = (ptrval) [d1580010] *pgd=8231b811, *pte=02034653, *ppte=02034453 Internal error: : 808 [#1] SMP ARM ... [<c095e974>] (regmap_mmio_write32le) from [<c095eb48>] (regmap_mmio_write+0x3c/0x54) [<c095eb48>] (regmap_mmio_write) from [<c09580f4>] (_regmap_write+0x4c/0x1f0) [<c09580f4>] (_regmap_write) from [<c095837c>] (_regmap_update_bits+0xe4/0xec) [<c095837c>] (_regmap_update_bits) from [<c09599b4>] (regmap_update_bits_base+0x50/0x74) [<c09599b4>] (regmap_update_bits_base) from [<c0d3e9e4>] (fsl_asrc_runtime_resume+0x1e4/0x21c) [<c0d3e9e4>] (fsl_asrc_runtime_resume) from [<c0942464>] (__rpm_callback+0x3c/0x108) [<c0942464>] (__rpm_callback) from [<c0942590>] (rpm_callback+0x60/0x64) [<c0942590>] (rpm_callback) from [<c0942b60>] (rpm_resume+0x5cc/0x808) [<c0942b60>] (rpm_resume) from [<c0942dfc>] (__pm_runtime_resume+0x60/0xa0) [<c0942dfc>] (__pm_runtime_resume) from [<c0d3ecc4>] (fsl_asrc_probe+0x2a8/0x708) [<c0d3ecc4>] (fsl_asrc_probe) from [<c0935b08>] (platform_probe+0x58/0xb8) [<c0935b08>] (platform_probe) from [<c0933264>] (really_probe.part.0+0x9c/0x334) [<c0933264>] (really_probe.part.0) from [<c093359c>] (__driver_probe_device+0xa0/0x138) [<c093359c>] (__driver_probe_device) from [<c0933664>] (driver_probe_device+0x30/0xc8) [<c0933664>] (driver_probe_device) from [<c0933c88>] (__driver_attach+0x90/0x130) [<c0933c88>] (__driver_attach) from [<c0931060>] (bus_for_each_dev+0x78/0xb8) [<c0931060>] (bus_for_each_dev) from [<c093254c>] (bus_add_driver+0xf0/0x1d8) [<c093254c>] (bus_add_driver) from [<c0934a30>] (driver_register+0x88/0x118) [<c0934a30>] (driver_register) from [<c01022c0>] (do_one_initcall+0x7c/0x3a4) [<c01022c0>] (do_one_initcall) from [<c1601204>] (kernel_init_freeable+0x198/0x22c) [<c1601204>] (kernel_init_freeable) from [<c0f5ff2c>] (kernel_init+0x10/0x128) [<c0f5ff2c>] (kernel_init) from [<c010013c>] (ret_from_fork+0x14/0x38) or Unhandled fault: external abort on non-linefetch (0x808) at 0xd19b0000 pgd = (ptrval) [d19b0000] *pgd=82711811, *pte=308a0653, *ppte=308a0453 Internal error: : 808 [#1] SMP ARM ... [<c095e974>] (regmap_mmio_write32le) from [<c095eb48>] (regmap_mmio_write+0x3c/0x54) [<c095eb48>] (regmap_mmio_write) from [<c09580f4>] (_regmap_write+0x4c/0x1f0) [<c09580f4>] (_regmap_write) from [<c0959b28>] (regmap_write+0x3c/0x60) [<c0959b28>] (regmap_write) from [<c0d41130>] (fsl_sai_runtime_resume+0x9c/0x1ec) [<c0d41130>] (fsl_sai_runtime_resume) from [<c0942464>] (__rpm_callback+0x3c/0x108) [<c0942464>] (__rpm_callback) from [<c0942590>] (rpm_callback+0x60/0x64) [<c0942590>] (rpm_callback) from [<c0942b60>] (rpm_resume+0x5cc/0x808) [<c0942b60>] (rpm_resume) from [<c0942dfc>] (__pm_runtime_resume+0x60/0xa0) [<c0942dfc>] (__pm_runtime_resume) from [<c0d4231c>] (fsl_sai_probe+0x2b8/0x65c) [<c0d4231c>] (fsl_sai_probe) from [<c0935b08>] (platform_probe+0x58/0xb8) [<c0935b08>] (platform_probe) from [<c0933264>] (really_probe.part.0+0x9c/0x334) [<c0933264>] (really_probe.part.0) from [<c093359c>] (__driver_probe_device+0xa0/0x138) [<c093359c>] (__driver_probe_device) from [<c0933664>] (driver_probe_device+0x30/0xc8) [<c0933664>] (driver_probe_device) from [<c0933c88>] (__driver_attach+0x90/0x130) [<c0933c88>] (__driver_attach) from [<c0931060>] (bus_for_each_dev+0x78/0xb8) [<c0931060>] (bus_for_each_dev) from [<c093254c>] (bus_add_driver+0xf0/0x1d8) [<c093254c>] (bus_add_driver) from [<c0934a30>] (driver_register+0x88/0x118) [<c0934a30>] (driver_register) from [<c01022c0>] (do_one_initcall+0x7c/0x3a4) [<c01022c0>] (do_one_initcall) from [<c1601204>] (kernel_init_freeable+0x198/0x22c) [<c1601204>] (kernel_init_freeable) from [<c0f5ff2c>] (kernel_init+0x10/0x128) [<c0f5ff2c>] (kernel_init) from [<c010013c>] (ret_from_fork+0x14/0x38) Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org> Signed-off-by: Guenter Roeck <linux@roeck-us.net> Message-id: 20210810160318.87376-1-linux@roeck-us.net Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/arm/fsl-imx6ul.c | 12 ++++++++++++ 1 file changed, 12 insertions(+) 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) */ create_unimplemented_device("sdma", FSL_IMX6UL_SDMA_ADDR, 0x4000); + /* + * SAI (Audio SSI (Synchronous Serial Interface)) + */ + create_unimplemented_device("sai1", FSL_IMX6UL_SAI1_ADDR, 0x4000); + create_unimplemented_device("sai2", FSL_IMX6UL_SAI2_ADDR, 0x4000); + create_unimplemented_device("sai3", FSL_IMX6UL_SAI3_ADDR, 0x4000); + /* * PWM */ @@ -XXX,XX +XXX,XX @@ static void fsl_imx6ul_realize(DeviceState *dev, Error **errp) create_unimplemented_device("pwm3", FSL_IMX6UL_PWM3_ADDR, 0x4000); create_unimplemented_device("pwm4", FSL_IMX6UL_PWM4_ADDR, 0x4000); + /* + * Audio ASRC (asynchronous sample rate converter) + */ + create_unimplemented_device("asrc", FSL_IMX6UL_ASRC_ADDR, 0x4000); + /* * CAN */ -- 2.20.1 From: "Wen, Jianxian" <Jianxian.Wen@verisilicon.com> Add property memory region which can connect with IOMMU region to support SMMU translate. Signed-off-by: Jianxian Wen <jianxian.wen@verisilicon.com> Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org> Message-id: 4C23C17B8E87E74E906A25A3254A03F4FA1FEC31@SHASXM03.verisilicon.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/arm/exynos4210.c | 3 +++ hw/arm/xilinx_zynq.c | 3 +++ hw/dma/pl330.c | 26 ++++++++++++++++++++++---- 3 files changed, 28 insertions(+), 4 deletions(-) diff --git a/hw/arm/exynos4210.c b/hw/arm/exynos4210.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/exynos4210.c +++ b/hw/arm/exynos4210.c @@ -XXX,XX +XXX,XX @@ static DeviceState *pl330_create(uint32_t base, qemu_or_irq *orgate, int i; dev = qdev_new("pl330"); + object_property_set_link(OBJECT(dev), "memory", + OBJECT(get_system_memory()), + &error_fatal); qdev_prop_set_uint8(dev, "num_events", nevents); qdev_prop_set_uint8(dev, "num_chnls", 8); qdev_prop_set_uint8(dev, "num_periph_req", nreq); diff --git a/hw/arm/xilinx_zynq.c b/hw/arm/xilinx_zynq.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/xilinx_zynq.c +++ b/hw/arm/xilinx_zynq.c @@ -XXX,XX +XXX,XX @@ static void zynq_init(MachineState *machine) sysbus_connect_irq(SYS_BUS_DEVICE(dev), 0, pic[39-IRQ_OFFSET]); dev = qdev_new("pl330"); + object_property_set_link(OBJECT(dev), "memory", + OBJECT(address_space_mem), + &error_fatal); qdev_prop_set_uint8(dev, "num_chnls", 8); qdev_prop_set_uint8(dev, "num_periph_req", 4); qdev_prop_set_uint8(dev, "num_events", 16); diff --git a/hw/dma/pl330.c b/hw/dma/pl330.c index XXXXXXX..XXXXXXX 100644 --- a/hw/dma/pl330.c +++ b/hw/dma/pl330.c @@ -XXX,XX +XXX,XX @@ struct PL330State { uint8_t num_faulting; uint8_t periph_busy[PL330_PERIPH_NUM]; + /* Memory region that DMA operation access */ + MemoryRegion *mem_mr; + AddressSpace *mem_as; }; #define TYPE_PL330 "pl330" @@ -XXX,XX +XXX,XX @@ static inline const PL330InsnDesc *pl330_fetch_insn(PL330Chan *ch) uint8_t opcode; int i; - dma_memory_read(&address_space_memory, ch->pc, &opcode, 1); + dma_memory_read(ch->parent->mem_as, ch->pc, &opcode, 1); for (i = 0; insn_desc[i].size; i++) { if ((opcode & insn_desc[i].opmask) == insn_desc[i].opcode) { return &insn_desc[i]; @@ -XXX,XX +XXX,XX @@ static inline void pl330_exec_insn(PL330Chan *ch, const PL330InsnDesc *insn) uint8_t buf[PL330_INSN_MAXSIZE]; assert(insn->size <= PL330_INSN_MAXSIZE); - dma_memory_read(&address_space_memory, ch->pc, buf, insn->size); + dma_memory_read(ch->parent->mem_as, ch->pc, buf, insn->size); insn->exec(ch, buf[0], &buf[1], insn->size - 1); } @@ -XXX,XX +XXX,XX @@ static int pl330_exec_cycle(PL330Chan *channel) if (q != NULL && q->len <= pl330_fifo_num_free(&s->fifo)) { int len = q->len - (q->addr & (q->len - 1)); - dma_memory_read(&address_space_memory, q->addr, buf, len); + dma_memory_read(s->mem_as, q->addr, buf, len); trace_pl330_exec_cycle(q->addr, len); if (trace_event_get_state_backends(TRACE_PL330_HEXDUMP)) { pl330_hexdump(buf, len); @@ -XXX,XX +XXX,XX @@ static int pl330_exec_cycle(PL330Chan *channel) fifo_res = pl330_fifo_get(&s->fifo, buf, len, q->tag); } if (fifo_res == PL330_FIFO_OK || q->z) { - dma_memory_write(&address_space_memory, q->addr, buf, len); + dma_memory_write(s->mem_as, q->addr, buf, len); trace_pl330_exec_cycle(q->addr, len); if (trace_event_get_state_backends(TRACE_PL330_HEXDUMP)) { pl330_hexdump(buf, len); @@ -XXX,XX +XXX,XX @@ static void pl330_realize(DeviceState *dev, Error **errp) "dma", PL330_IOMEM_SIZE); sysbus_init_mmio(SYS_BUS_DEVICE(dev), &s->iomem); + if (!s->mem_mr) { + error_setg(errp, "'memory' link is not set"); + return; + } else if (s->mem_mr == get_system_memory()) { + /* Avoid creating new AS for system memory. */ + s->mem_as = &address_space_memory; + } else { + s->mem_as = g_new0(AddressSpace, 1); + address_space_init(s->mem_as, s->mem_mr, + memory_region_name(s->mem_mr)); + } + s->timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, pl330_exec_cycle_timer, s); s->cfg[0] = (s->mgr_ns_at_rst ? 0x4 : 0) | @@ -XXX,XX +XXX,XX @@ static Property pl330_properties[] = { DEFINE_PROP_UINT8("rd_q_dep", PL330State, rd_q_dep, 16), DEFINE_PROP_UINT16("data_buffer_dep", PL330State, data_buffer_dep, 256), + DEFINE_PROP_LINK("memory", PL330State, mem_mr, + TYPE_MEMORY_REGION, MemoryRegion *), + DEFINE_PROP_END_OF_LIST(), }; -- 2.20.1 From: Eduardo Habkost <ehabkost@redhat.com> The SBSA_GWDT enum value conflicts with the SBSA_GWDT() QOM type checking helper, preventing us from using a OBJECT_DEFINE* or DEFINE_INSTANCE_CHECKER macro for the SBSA_GWDT() wrapper. If I understand the SBSA 6.0 specification correctly, the signal being connected to IRQ 16 is the WS0 output signal from the Generic Watchdog. Rename the enum value to SBSA_GWDT_WS0 to be more explicit and avoid the name conflict. Signed-off-by: Eduardo Habkost <ehabkost@redhat.com> Message-id: 20210806023119.431680-1-ehabkost@redhat.com Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/arm/sbsa-ref.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/hw/arm/sbsa-ref.c b/hw/arm/sbsa-ref.c index XXXXXXX..XXXXXXX 100644 --- a/hw/arm/sbsa-ref.c +++ b/hw/arm/sbsa-ref.c @@ -XXX,XX +XXX,XX @@ enum { SBSA_GIC_DIST, SBSA_GIC_REDIST, SBSA_SECURE_EC, - SBSA_GWDT, + SBSA_GWDT_WS0, SBSA_GWDT_REFRESH, SBSA_GWDT_CONTROL, SBSA_SMMU, @@ -XXX,XX +XXX,XX @@ static const int sbsa_ref_irqmap[] = { [SBSA_AHCI] = 10, [SBSA_EHCI] = 11, [SBSA_SMMU] = 12, /* ... to 15 */ - [SBSA_GWDT] = 16, + [SBSA_GWDT_WS0] = 16, }; static const char * const valid_cpus[] = { @@ -XXX,XX +XXX,XX @@ static void create_wdt(const SBSAMachineState *sms) hwaddr cbase = sbsa_ref_memmap[SBSA_GWDT_CONTROL].base; DeviceState *dev = qdev_new(TYPE_WDT_SBSA); SysBusDevice *s = SYS_BUS_DEVICE(dev); - int irq = sbsa_ref_irqmap[SBSA_GWDT]; + int irq = sbsa_ref_irqmap[SBSA_GWDT_WS0]; sysbus_realize_and_unref(s, &error_fatal); sysbus_mmio_map(s, 0, rbase); -- 2.20.1 From: Guenter Roeck <linux@roeck-us.net> Instantiate SAI1/2/3 as unimplemented devices to avoid Linux kernel crashes such as the following. Unhandled fault: external abort on non-linefetch (0x808) at 0xd19b0000 pgd = (ptrval) [d19b0000] *pgd=82711811, *pte=308a0653, *ppte=308a0453 Internal error: : 808 [#1] SMP ARM Modules linked in: CPU: 0 PID: 1 Comm: swapper/0 Not tainted 5.14.0-rc5 #1 ... [<c095e974>] (regmap_mmio_write32le) from [<c095eb48>] (regmap_mmio_write+0x3c/0x54) [<c095eb48>] (regmap_mmio_write) from [<c09580f4>] (_regmap_write+0x4c/0x1f0) [<c09580f4>] (_regmap_write) from [<c0959b28>] (regmap_write+0x3c/0x60) [<c0959b28>] (regmap_write) from [<c0d41130>] (fsl_sai_runtime_resume+0x9c/0x1ec) [<c0d41130>] (fsl_sai_runtime_resume) from [<c0942464>] (__rpm_callback+0x3c/0x108) [<c0942464>] (__rpm_callback) from [<c0942590>] (rpm_callback+0x60/0x64) [<c0942590>] (rpm_callback) from [<c0942b60>] (rpm_resume+0x5cc/0x808) [<c0942b60>] (rpm_resume) from [<c0942dfc>] (__pm_runtime_resume+0x60/0xa0) [<c0942dfc>] (__pm_runtime_resume) from [<c0d4231c>] (fsl_sai_probe+0x2b8/0x65c) [<c0d4231c>] (fsl_sai_probe) from [<c0935b08>] (platform_probe+0x58/0xb8) [<c0935b08>] (platform_probe) from [<c0933264>] (really_probe.part.0+0x9c/0x334) [<c0933264>] (really_probe.part.0) from [<c093359c>] (__driver_probe_device+0xa0/0x138) [<c093359c>] (__driver_probe_device) from [<c0933664>] (driver_probe_device+0x30/0xc8) [<c0933664>] (driver_probe_device) from [<c0933c88>] (__driver_attach+0x90/0x130) [<c0933c88>] (__driver_attach) from [<c0931060>] (bus_for_each_dev+0x78/0xb8) [<c0931060>] (bus_for_each_dev) from [<c093254c>] (bus_add_driver+0xf0/0x1d8) [<c093254c>] (bus_add_driver) from [<c0934a30>] (driver_register+0x88/0x118) [<c0934a30>] (driver_register) from [<c01022c0>] (do_one_initcall+0x7c/0x3a4) [<c01022c0>] (do_one_initcall) from [<c1601204>] (kernel_init_freeable+0x198/0x22c) [<c1601204>] (kernel_init_freeable) from [<c0f5ff2c>] (kernel_init+0x10/0x128) [<c0f5ff2c>] (kernel_init) from [<c010013c>] (ret_from_fork+0x14/0x38) Signed-off-by: Guenter Roeck <linux@roeck-us.net> Message-id: 20210810175607.538090-1-linux@roeck-us.net Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- include/hw/arm/fsl-imx7.h | 5 +++++ hw/arm/fsl-imx7.c | 7 +++++++ 2 files changed, 12 insertions(+) diff --git a/include/hw/arm/fsl-imx7.h b/include/hw/arm/fsl-imx7.h index XXXXXXX..XXXXXXX 100644 --- a/include/hw/arm/fsl-imx7.h +++ b/include/hw/arm/fsl-imx7.h @@ -XXX,XX +XXX,XX @@ enum FslIMX7MemoryMap { FSL_IMX7_UART6_ADDR = 0x30A80000, FSL_IMX7_UART7_ADDR = 0x30A90000, + FSL_IMX7_SAI1_ADDR = 0x308A0000, + FSL_IMX7_SAI2_ADDR = 0x308B0000, + FSL_IMX7_SAI3_ADDR = 0x308C0000, + FSL_IMX7_SAIn_SIZE = 0x10000, + FSL_IMX7_ENET1_ADDR = 0x30BE0000, FSL_IMX7_ENET2_ADDR = 0x30BF0000, 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) create_unimplemented_device("can1", FSL_IMX7_CAN1_ADDR, FSL_IMX7_CANn_SIZE); create_unimplemented_device("can2", FSL_IMX7_CAN2_ADDR, FSL_IMX7_CANn_SIZE); + /* + * SAI (Audio SSI (Synchronous Serial Interface)) + */ + create_unimplemented_device("sai1", FSL_IMX7_SAI1_ADDR, FSL_IMX7_SAIn_SIZE); + create_unimplemented_device("sai2", FSL_IMX7_SAI2_ADDR, FSL_IMX7_SAIn_SIZE); + create_unimplemented_device("sai2", FSL_IMX7_SAI3_ADDR, FSL_IMX7_SAIn_SIZE); + /* * OCOTP */ -- 2.20.1 From: Sebastian Meyer <meyer@absint.com> With gdb 9.0 and better it is possible to connect to a gdbstub over unix sockets, which is better than a TCP socket connection in some situations. The QEMU command line to set this up is non-obvious; document it. Signed-off-by: Sebastian Meyer <meyer@absint.com> Message-id: 162867284829.27377.4784930719350564918-0@git.sr.ht [PMM: Tweaked commit message; adjusted wording in a couple of places; fixed rST formatting issue; moved section up out of the 'advanced debugging options' subsection] Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com> Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- docs/system/gdb.rst | 26 +++++++++++++++++++++++++- 1 file changed, 25 insertions(+), 1 deletion(-) diff --git a/docs/system/gdb.rst b/docs/system/gdb.rst index XXXXXXX..XXXXXXX 100644 --- a/docs/system/gdb.rst +++ b/docs/system/gdb.rst @@ -XXX,XX +XXX,XX @@ The ``-s`` option will make QEMU listen for an incoming connection from gdb on TCP port 1234, and ``-S`` will make QEMU not start the guest until you tell it to from gdb. (If you want to specify which TCP port to use or to use something other than TCP for the gdbstub -connection, use the ``-gdb dev`` option instead of ``-s``.) +connection, use the ``-gdb dev`` option instead of ``-s``. See +`Using unix sockets`_ for an example.) .. parsed-literal:: @@ -XXX,XX +XXX,XX @@ not just those in the cluster you are currently working on:: (gdb) set schedule-multiple on +Using unix sockets +================== + +An alternate method for connecting gdb to the QEMU gdbstub is to use +a unix socket (if supported by your operating system). This is useful when +running several tests in parallel, or if you do not have a known free TCP +port (e.g. when running automated tests). + +First create a chardev with the appropriate options, then +instruct the gdbserver to use that device: + +.. parsed-literal:: + + |qemu_system| -chardev socket,path=/tmp/gdb-socket,server=on,wait=off,id=gdb0 -gdb chardev:gdb0 -S ... + +Start gdb as before, but this time connect using the path to +the socket:: + + (gdb) target remote /tmp/gdb-socket + +Note that to use a unix socket for the connection you will need +gdb version 9.0 or newer. + Advanced debugging options ========================== -- 2.20.1 First arm pullreq of the cycle; this is mostly my softfloat NaN handling series. (Lots more in my to-review queue, but I don't like pullreqs growing too close to a hundred patches at a time :-)) thanks -- PMM The following changes since commit 97f2796a3736ed37a1b85dc1c76a6c45b829dd17: Open 10.0 development tree (2024-12-10 17:41:17 +0000) are available in the Git repository at: https://git.linaro.org/people/pmaydell/qemu-arm.git tags/pull-target-arm-20241211 for you to fetch changes up to 1abe28d519239eea5cf9620bb13149423e5665f8: MAINTAINERS: Add correct email address for Vikram Garhwal (2024-12-11 15:31:09 +0000) ---------------------------------------------------------------- target-arm queue: * hw/net/lan9118: Extract PHY model, reuse with imx_fec, fix bugs * fpu: Make muladd NaN handling runtime-selected, not compile-time * fpu: Make default NaN pattern runtime-selected, not compile-time * fpu: Minor NaN-related cleanups * MAINTAINERS: email address updates ---------------------------------------------------------------- Bernhard Beschow (5): hw/net/lan9118: Extract lan9118_phy hw/net/lan9118_phy: Reuse in imx_fec and consolidate implementations hw/net/lan9118_phy: Fix off-by-one error in MII_ANLPAR register hw/net/lan9118_phy: Reuse MII constants hw/net/lan9118_phy: Add missing 100 mbps full duplex advertisement Leif Lindholm (1): MAINTAINERS: update email address for Leif Lindholm Peter Maydell (54): fpu: handle raising Invalid for infzero in pick_nan_muladd fpu: Check for default_nan_mode before calling pickNaNMulAdd softfloat: Allow runtime choice of inf * 0 + NaN result tests/fp: Explicitly set inf-zero-nan rule target/arm: Set FloatInfZeroNaNRule explicitly target/s390: Set FloatInfZeroNaNRule explicitly target/ppc: Set FloatInfZeroNaNRule explicitly target/mips: Set FloatInfZeroNaNRule explicitly target/sparc: Set FloatInfZeroNaNRule explicitly target/xtensa: Set FloatInfZeroNaNRule explicitly target/x86: Set FloatInfZeroNaNRule explicitly target/loongarch: Set FloatInfZeroNaNRule explicitly target/hppa: Set FloatInfZeroNaNRule explicitly softfloat: Pass have_snan to pickNaNMulAdd softfloat: Allow runtime choice of NaN propagation for muladd tests/fp: Explicitly set 3-NaN propagation rule target/arm: Set Float3NaNPropRule explicitly target/loongarch: Set Float3NaNPropRule explicitly target/ppc: Set Float3NaNPropRule explicitly target/s390x: Set Float3NaNPropRule explicitly target/sparc: Set Float3NaNPropRule explicitly target/mips: Set Float3NaNPropRule explicitly target/xtensa: Set Float3NaNPropRule explicitly target/i386: Set Float3NaNPropRule explicitly target/hppa: Set Float3NaNPropRule explicitly fpu: Remove use_first_nan field from float_status target/m68k: Don't pass NULL float_status to floatx80_default_nan() softfloat: Create floatx80 default NaN from parts64_default_nan target/loongarch: Use normal float_status in fclass_s and fclass_d helpers target/m68k: In frem helper, initialize local float_status from env->fp_status target/m68k: Init local float_status from env fp_status in gdb get/set reg target/sparc: Initialize local scratch float_status from env->fp_status target/ppc: Use env->fp_status in helper_compute_fprf functions fpu: Allow runtime choice of default NaN value tests/fp: Set default NaN pattern explicitly target/microblaze: Set default NaN pattern explicitly target/i386: Set default NaN pattern explicitly target/hppa: Set default NaN pattern explicitly target/alpha: Set default NaN pattern explicitly target/arm: Set default NaN pattern explicitly target/loongarch: Set default NaN pattern explicitly target/m68k: Set default NaN pattern explicitly target/mips: Set default NaN pattern explicitly target/openrisc: Set default NaN pattern explicitly target/ppc: Set default NaN pattern explicitly target/sh4: Set default NaN pattern explicitly target/rx: Set default NaN pattern explicitly target/s390x: Set default NaN pattern explicitly target/sparc: Set default NaN pattern explicitly target/xtensa: Set default NaN pattern explicitly target/hexagon: Set default NaN pattern explicitly target/riscv: Set default NaN pattern explicitly target/tricore: Set default NaN pattern explicitly fpu: Remove default handling for dnan_pattern Richard Henderson (11): target/arm: Copy entire float_status in is_ebf softfloat: Inline pickNaNMulAdd softfloat: Use goto for default nan case in pick_nan_muladd softfloat: Remove which from parts_pick_nan_muladd softfloat: Pad array size in pick_nan_muladd softfloat: Move propagateFloatx80NaN to softfloat.c softfloat: Use parts_pick_nan in propagateFloatx80NaN softfloat: Inline pickNaN softfloat: Share code between parts_pick_nan cases softfloat: Sink frac_cmp in parts_pick_nan until needed softfloat: Replace WHICH with RET in parts_pick_nan Vikram Garhwal (1): MAINTAINERS: Add correct email address for Vikram Garhwal MAINTAINERS | 4 +- include/fpu/softfloat-helpers.h | 38 +++- include/fpu/softfloat-types.h | 89 +++++++- include/hw/net/imx_fec.h | 9 +- include/hw/net/lan9118_phy.h | 37 ++++ include/hw/net/mii.h | 6 + target/mips/fpu_helper.h | 20 ++ target/sparc/helper.h | 4 +- fpu/softfloat.c | 19 ++ hw/net/imx_fec.c | 146 ++------------ hw/net/lan9118.c | 137 ++----------- hw/net/lan9118_phy.c | 222 ++++++++++++++++++++ linux-user/arm/nwfpe/fpa11.c | 5 + target/alpha/cpu.c | 2 + target/arm/cpu.c | 10 + target/arm/tcg/vec_helper.c | 20 +- target/hexagon/cpu.c | 2 + target/hppa/fpu_helper.c | 12 ++ target/i386/tcg/fpu_helper.c | 12 ++ target/loongarch/tcg/fpu_helper.c | 14 +- target/m68k/cpu.c | 14 +- target/m68k/fpu_helper.c | 6 +- target/m68k/helper.c | 6 +- target/microblaze/cpu.c | 2 + target/mips/msa.c | 10 + target/openrisc/cpu.c | 2 + target/ppc/cpu_init.c | 19 ++ target/ppc/fpu_helper.c | 3 +- target/riscv/cpu.c | 2 + target/rx/cpu.c | 2 + target/s390x/cpu.c | 5 + target/sh4/cpu.c | 2 + target/sparc/cpu.c | 6 + target/sparc/fop_helper.c | 8 +- target/sparc/translate.c | 4 +- target/tricore/helper.c | 2 + target/xtensa/cpu.c | 4 + target/xtensa/fpu_helper.c | 3 +- tests/fp/fp-bench.c | 7 + tests/fp/fp-test-log2.c | 1 + tests/fp/fp-test.c | 7 + fpu/softfloat-parts.c.inc | 152 +++++++++++--- fpu/softfloat-specialize.c.inc | 412 ++------------------------------------ .mailmap | 5 +- hw/net/Kconfig | 5 + hw/net/meson.build | 1 + hw/net/trace-events | 10 +- 47 files changed, 778 insertions(+), 730 deletions(-) create mode 100644 include/hw/net/lan9118_phy.h create mode 100644 hw/net/lan9118_phy.c From: Bernhard Beschow <shentey@gmail.com> A very similar implementation of the same device exists in imx_fec. Prepare for a common implementation by extracting a device model into its own files. Some migration state has been moved into the new device model which breaks migration compatibility for the following machines: * smdkc210 * realview-* * vexpress-* * kzm * mps2-* While breaking migration ABI, fix the size of the MII registers to be 16 bit, as defined by IEEE 802.3u. Signed-off-by: Bernhard Beschow <shentey@gmail.com> Tested-by: Guenter Roeck <linux@roeck-us.net> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Message-id: 20241102125724.532843-2-shentey@gmail.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- include/hw/net/lan9118_phy.h | 37 ++++++++ hw/net/lan9118.c | 137 +++++----------------------- hw/net/lan9118_phy.c | 169 +++++++++++++++++++++++++++++++++++ hw/net/Kconfig | 4 + hw/net/meson.build | 1 + 5 files changed, 233 insertions(+), 115 deletions(-) create mode 100644 include/hw/net/lan9118_phy.h create mode 100644 hw/net/lan9118_phy.c diff --git a/include/hw/net/lan9118_phy.h b/include/hw/net/lan9118_phy.h new file mode 100644 index XXXXXXX..XXXXXXX --- /dev/null +++ b/include/hw/net/lan9118_phy.h @@ -XXX,XX +XXX,XX @@ +/* + * SMSC LAN9118 PHY emulation + * + * Copyright (c) 2009 CodeSourcery, LLC. + * Written by Paul Brook + * + * This work is licensed under the terms of the GNU GPL, version 2 or later. + * See the COPYING file in the top-level directory. + */ + +#ifndef HW_NET_LAN9118_PHY_H +#define HW_NET_LAN9118_PHY_H + +#include "qom/object.h" +#include "hw/sysbus.h" + +#define TYPE_LAN9118_PHY "lan9118-phy" +OBJECT_DECLARE_SIMPLE_TYPE(Lan9118PhyState, LAN9118_PHY) + +typedef struct Lan9118PhyState { + SysBusDevice parent_obj; + + uint16_t status; + uint16_t control; + uint16_t advertise; + uint16_t ints; + uint16_t int_mask; + qemu_irq irq; + bool link_down; +} Lan9118PhyState; + +void lan9118_phy_update_link(Lan9118PhyState *s, bool link_down); +void lan9118_phy_reset(Lan9118PhyState *s); +uint16_t lan9118_phy_read(Lan9118PhyState *s, int reg); +void lan9118_phy_write(Lan9118PhyState *s, int reg, uint16_t val); + +#endif diff --git a/hw/net/lan9118.c b/hw/net/lan9118.c index XXXXXXX..XXXXXXX 100644 --- a/hw/net/lan9118.c +++ b/hw/net/lan9118.c @@ -XXX,XX +XXX,XX @@ #include "net/net.h" #include "net/eth.h" #include "hw/irq.h" +#include "hw/net/lan9118_phy.h" #include "hw/net/lan9118.h" #include "hw/ptimer.h" #include "hw/qdev-properties.h" @@ -XXX,XX +XXX,XX @@ do { printf("lan9118: " fmt , ## __VA_ARGS__); } while (0) #define MAC_CR_RXEN 0x00000004 #define MAC_CR_RESERVED 0x7f404213 -#define PHY_INT_ENERGYON 0x80 -#define PHY_INT_AUTONEG_COMPLETE 0x40 -#define PHY_INT_FAULT 0x20 -#define PHY_INT_DOWN 0x10 -#define PHY_INT_AUTONEG_LP 0x08 -#define PHY_INT_PARFAULT 0x04 -#define PHY_INT_AUTONEG_PAGE 0x02 - #define GPT_TIMER_EN 0x20000000 /* @@ -XXX,XX +XXX,XX @@ struct lan9118_state { uint32_t mac_mii_data; uint32_t mac_flow; - uint32_t phy_status; - uint32_t phy_control; - uint32_t phy_advertise; - uint32_t phy_int; - uint32_t phy_int_mask; + Lan9118PhyState mii; + IRQState mii_irq; int32_t eeprom_writable; uint8_t eeprom[128]; @@ -XXX,XX +XXX,XX @@ struct lan9118_state { static const VMStateDescription vmstate_lan9118 = { .name = "lan9118", - .version_id = 2, - .minimum_version_id = 1, + .version_id = 3, + .minimum_version_id = 3, .fields = (const VMStateField[]) { VMSTATE_PTIMER(timer, lan9118_state), VMSTATE_UINT32(irq_cfg, lan9118_state), @@ -XXX,XX +XXX,XX @@ static const VMStateDescription vmstate_lan9118 = { VMSTATE_UINT32(mac_mii_acc, lan9118_state), VMSTATE_UINT32(mac_mii_data, lan9118_state), VMSTATE_UINT32(mac_flow, lan9118_state), - VMSTATE_UINT32(phy_status, lan9118_state), - VMSTATE_UINT32(phy_control, lan9118_state), - VMSTATE_UINT32(phy_advertise, lan9118_state), - VMSTATE_UINT32(phy_int, lan9118_state), - VMSTATE_UINT32(phy_int_mask, lan9118_state), VMSTATE_INT32(eeprom_writable, lan9118_state), VMSTATE_UINT8_ARRAY(eeprom, lan9118_state, 128), VMSTATE_INT32(tx_fifo_size, lan9118_state), @@ -XXX,XX +XXX,XX @@ static void lan9118_reload_eeprom(lan9118_state *s) lan9118_mac_changed(s); } -static void phy_update_irq(lan9118_state *s) +static void lan9118_update_irq(void *opaque, int n, int level) { - if (s->phy_int & s->phy_int_mask) { + lan9118_state *s = opaque; + + if (level) { s->int_sts |= PHY_INT; } else { s->int_sts &= ~PHY_INT; @@ -XXX,XX +XXX,XX @@ static void phy_update_irq(lan9118_state *s) lan9118_update(s); } -static void phy_update_link(lan9118_state *s) -{ - /* Autonegotiation status mirrors link status. */ - if (qemu_get_queue(s->nic)->link_down) { - s->phy_status &= ~0x0024; - s->phy_int |= PHY_INT_DOWN; - } else { - s->phy_status |= 0x0024; - s->phy_int |= PHY_INT_ENERGYON; - s->phy_int |= PHY_INT_AUTONEG_COMPLETE; - } - phy_update_irq(s); -} - static void lan9118_set_link(NetClientState *nc) { - phy_update_link(qemu_get_nic_opaque(nc)); -} - -static void phy_reset(lan9118_state *s) -{ - 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); + lan9118_phy_update_link(&LAN9118(qemu_get_nic_opaque(nc))->mii, + nc->link_down); } static void lan9118_reset(DeviceState *d) @@ -XXX,XX +XXX,XX @@ static void lan9118_reset(DeviceState *d) s->read_word_n = 0; s->write_word_n = 0; - phy_reset(s); - s->eeprom_writable = 0; lan9118_reload_eeprom(s); } @@ -XXX,XX +XXX,XX @@ static void do_tx_packet(lan9118_state *s) uint32_t status; /* FIXME: Honor TX disable, and allow queueing of packets. */ - if (s->phy_control & 0x4000) { + if (s->mii.control & 0x4000) { /* This assumes the receive routine doesn't touch the VLANClient. */ qemu_receive_packet(qemu_get_queue(s->nic), s->txp->data, s->txp->len); } else { @@ -XXX,XX +XXX,XX @@ static void tx_fifo_push(lan9118_state *s, uint32_t val) } } -static uint32_t do_phy_read(lan9118_state *s, int reg) -{ - uint32_t val; - - switch (reg) { - case 0: /* Basic Control */ - return s->phy_control; - case 1: /* Basic Status */ - return s->phy_status; - case 2: /* ID1 */ - return 0x0007; - case 3: /* ID2 */ - return 0xc0d1; - case 4: /* Auto-neg advertisement */ - return s->phy_advertise; - case 5: /* Auto-neg Link Partner Ability */ - return 0x0f71; - case 6: /* Auto-neg Expansion */ - return 1; - /* TODO 17, 18, 27, 29, 30, 31 */ - case 29: /* Interrupt source. */ - val = s->phy_int; - s->phy_int = 0; - phy_update_irq(s); - return val; - case 30: /* Interrupt mask */ - return s->phy_int_mask; - default: - qemu_log_mask(LOG_GUEST_ERROR, - "do_phy_read: PHY read reg %d\n", reg); - return 0; - } -} - -static void do_phy_write(lan9118_state *s, int reg, uint32_t val) -{ - switch (reg) { - case 0: /* Basic Control */ - if (val & 0x8000) { - phy_reset(s); - break; - } - s->phy_control = val & 0x7980; - /* Complete autonegotiation immediately. */ - if (val & 0x1000) { - s->phy_status |= 0x0020; - } - break; - case 4: /* Auto-neg advertisement */ - s->phy_advertise = (val & 0x2d7f) | 0x80; - break; - /* TODO 17, 18, 27, 31 */ - case 30: /* Interrupt mask */ - s->phy_int_mask = val & 0xff; - phy_update_irq(s); - break; - default: - qemu_log_mask(LOG_GUEST_ERROR, - "do_phy_write: PHY write reg %d = 0x%04x\n", reg, val); - } -} - static void do_mac_write(lan9118_state *s, int reg, uint32_t val) { switch (reg) { @@ -XXX,XX +XXX,XX @@ static void do_mac_write(lan9118_state *s, int reg, uint32_t val) if (val & 2) { DPRINTF("PHY write %d = 0x%04x\n", (val >> 6) & 0x1f, s->mac_mii_data); - do_phy_write(s, (val >> 6) & 0x1f, s->mac_mii_data); + lan9118_phy_write(&s->mii, (val >> 6) & 0x1f, s->mac_mii_data); } else { - s->mac_mii_data = do_phy_read(s, (val >> 6) & 0x1f); + s->mac_mii_data = lan9118_phy_read(&s->mii, (val >> 6) & 0x1f); DPRINTF("PHY read %d = 0x%04x\n", (val >> 6) & 0x1f, s->mac_mii_data); } @@ -XXX,XX +XXX,XX @@ static void lan9118_writel(void *opaque, hwaddr offset, break; case CSR_PMT_CTRL: if (val & 0x400) { - phy_reset(s); + lan9118_phy_reset(&s->mii); } s->pmt_ctrl &= ~0x34e; s->pmt_ctrl |= (val & 0x34e); @@ -XXX,XX +XXX,XX @@ static void lan9118_realize(DeviceState *dev, Error **errp) const MemoryRegionOps *mem_ops = s->mode_16bit ? &lan9118_16bit_mem_ops : &lan9118_mem_ops; + qemu_init_irq(&s->mii_irq, lan9118_update_irq, s, 0); + object_initialize_child(OBJECT(s), "mii", &s->mii, TYPE_LAN9118_PHY); + if (!sysbus_realize_and_unref(SYS_BUS_DEVICE(&s->mii), errp)) { + return; + } + qdev_connect_gpio_out(DEVICE(&s->mii), 0, &s->mii_irq); + memory_region_init_io(&s->mmio, OBJECT(dev), mem_ops, s, "lan9118-mmio", 0x100); sysbus_init_mmio(sbd, &s->mmio); diff --git a/hw/net/lan9118_phy.c b/hw/net/lan9118_phy.c new file mode 100644 index XXXXXXX..XXXXXXX --- /dev/null +++ b/hw/net/lan9118_phy.c @@ -XXX,XX +XXX,XX @@ +/* + * SMSC LAN9118 PHY emulation + * + * Copyright (c) 2009 CodeSourcery, LLC. + * Written by Paul Brook + * + * This code is licensed under the GNU GPL v2 + * + * Contributions after 2012-01-13 are licensed under the terms of the + * GNU GPL, version 2 or (at your option) any later version. + */ + +#include "qemu/osdep.h" +#include "hw/net/lan9118_phy.h" +#include "hw/irq.h" +#include "hw/resettable.h" +#include "migration/vmstate.h" +#include "qemu/log.h" + +#define PHY_INT_ENERGYON (1 << 7) +#define PHY_INT_AUTONEG_COMPLETE (1 << 6) +#define PHY_INT_FAULT (1 << 5) +#define PHY_INT_DOWN (1 << 4) +#define PHY_INT_AUTONEG_LP (1 << 3) +#define PHY_INT_PARFAULT (1 << 2) +#define PHY_INT_AUTONEG_PAGE (1 << 1) + +static void lan9118_phy_update_irq(Lan9118PhyState *s) +{ + qemu_set_irq(s->irq, !!(s->ints & s->int_mask)); +} + +uint16_t lan9118_phy_read(Lan9118PhyState *s, int reg) +{ + uint16_t val; + + switch (reg) { + case 0: /* Basic Control */ + return s->control; + case 1: /* Basic Status */ + return s->status; + case 2: /* ID1 */ + return 0x0007; + case 3: /* ID2 */ + return 0xc0d1; + case 4: /* Auto-neg advertisement */ + return s->advertise; + case 5: /* Auto-neg Link Partner Ability */ + return 0x0f71; + case 6: /* Auto-neg Expansion */ + return 1; + /* TODO 17, 18, 27, 29, 30, 31 */ + case 29: /* Interrupt source. */ + val = s->ints; + s->ints = 0; + lan9118_phy_update_irq(s); + return val; + case 30: /* Interrupt mask */ + return s->int_mask; + default: + qemu_log_mask(LOG_GUEST_ERROR, + "lan9118_phy_read: PHY read reg %d\n", reg); + return 0; + } +} + +void lan9118_phy_write(Lan9118PhyState *s, int reg, uint16_t val) +{ + switch (reg) { + case 0: /* Basic Control */ + if (val & 0x8000) { + lan9118_phy_reset(s); + break; + } + s->control = val & 0x7980; + /* Complete autonegotiation immediately. */ + if (val & 0x1000) { + s->status |= 0x0020; + } + break; + case 4: /* Auto-neg advertisement */ + s->advertise = (val & 0x2d7f) | 0x80; + break; + /* TODO 17, 18, 27, 31 */ + case 30: /* Interrupt mask */ + s->int_mask = val & 0xff; + lan9118_phy_update_irq(s); + break; + default: + qemu_log_mask(LOG_GUEST_ERROR, + "lan9118_phy_write: PHY write reg %d = 0x%04x\n", reg, val); + } +} + +void lan9118_phy_update_link(Lan9118PhyState *s, bool link_down) +{ + s->link_down = link_down; + + /* Autonegotiation status mirrors link status. */ + if (link_down) { + s->status &= ~0x0024; + s->ints |= PHY_INT_DOWN; + } else { + s->status |= 0x0024; + s->ints |= PHY_INT_ENERGYON; + s->ints |= PHY_INT_AUTONEG_COMPLETE; + } + lan9118_phy_update_irq(s); +} + +void lan9118_phy_reset(Lan9118PhyState *s) +{ + s->control = 0x3000; + s->status = 0x7809; + s->advertise = 0x01e1; + s->int_mask = 0; + s->ints = 0; + lan9118_phy_update_link(s, s->link_down); +} + +static void lan9118_phy_reset_hold(Object *obj, ResetType type) +{ + Lan9118PhyState *s = LAN9118_PHY(obj); + + lan9118_phy_reset(s); +} + +static void lan9118_phy_init(Object *obj) +{ + Lan9118PhyState *s = LAN9118_PHY(obj); + + qdev_init_gpio_out(DEVICE(s), &s->irq, 1); +} + +static const VMStateDescription vmstate_lan9118_phy = { + .name = "lan9118-phy", + .version_id = 1, + .minimum_version_id = 1, + .fields = (const VMStateField[]) { + VMSTATE_UINT16(control, Lan9118PhyState), + VMSTATE_UINT16(status, Lan9118PhyState), + VMSTATE_UINT16(advertise, Lan9118PhyState), + VMSTATE_UINT16(ints, Lan9118PhyState), + VMSTATE_UINT16(int_mask, Lan9118PhyState), + VMSTATE_BOOL(link_down, Lan9118PhyState), + VMSTATE_END_OF_LIST() + } +}; + +static void lan9118_phy_class_init(ObjectClass *klass, void *data) +{ + ResettableClass *rc = RESETTABLE_CLASS(klass); + DeviceClass *dc = DEVICE_CLASS(klass); + + rc->phases.hold = lan9118_phy_reset_hold; + dc->vmsd = &vmstate_lan9118_phy; +} + +static const TypeInfo types[] = { + { + .name = TYPE_LAN9118_PHY, + .parent = TYPE_SYS_BUS_DEVICE, + .instance_size = sizeof(Lan9118PhyState), + .instance_init = lan9118_phy_init, + .class_init = lan9118_phy_class_init, + } +}; + +DEFINE_TYPES(types) diff --git a/hw/net/Kconfig b/hw/net/Kconfig index XXXXXXX..XXXXXXX 100644 --- a/hw/net/Kconfig +++ b/hw/net/Kconfig @@ -XXX,XX +XXX,XX @@ config VMXNET3_PCI config SMC91C111 bool +config LAN9118_PHY + bool + config LAN9118 bool + select LAN9118_PHY select PTIMER config NE2000_ISA diff --git a/hw/net/meson.build b/hw/net/meson.build index XXXXXXX..XXXXXXX 100644 --- a/hw/net/meson.build +++ b/hw/net/meson.build @@ -XXX,XX +XXX,XX @@ system_ss.add(when: 'CONFIG_VMXNET3_PCI', if_true: files('vmxnet3.c')) system_ss.add(when: 'CONFIG_SMC91C111', if_true: files('smc91c111.c')) system_ss.add(when: 'CONFIG_LAN9118', if_true: files('lan9118.c')) +system_ss.add(when: 'CONFIG_LAN9118_PHY', if_true: files('lan9118_phy.c')) system_ss.add(when: 'CONFIG_NE2000_ISA', if_true: files('ne2000-isa.c')) system_ss.add(when: 'CONFIG_OPENCORES_ETH', if_true: files('opencores_eth.c')) system_ss.add(when: 'CONFIG_XGMAC', if_true: files('xgmac.c')) -- 2.34.1 From: Bernhard Beschow <shentey@gmail.com> imx_fec models the same PHY as lan9118_phy. The code is almost the same with imx_fec having more logging and tracing. Merge these improvements into lan9118_phy and reuse in imx_fec to fix the code duplication. Some migration state how resides in the new device model which breaks migration compatibility for the following machines: * imx25-pdk * sabrelite * mcimx7d-sabre * mcimx6ul-evk Signed-off-by: Bernhard Beschow <shentey@gmail.com> Tested-by: Guenter Roeck <linux@roeck-us.net> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Message-id: 20241102125724.532843-3-shentey@gmail.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- include/hw/net/imx_fec.h | 9 ++- hw/net/imx_fec.c | 146 ++++----------------------------------- hw/net/lan9118_phy.c | 82 ++++++++++++++++------ hw/net/Kconfig | 1 + hw/net/trace-events | 10 +-- 5 files changed, 85 insertions(+), 163 deletions(-) diff --git a/include/hw/net/imx_fec.h b/include/hw/net/imx_fec.h index XXXXXXX..XXXXXXX 100644 --- a/include/hw/net/imx_fec.h +++ b/include/hw/net/imx_fec.h @@ -XXX,XX +XXX,XX @@ OBJECT_DECLARE_SIMPLE_TYPE(IMXFECState, IMX_FEC) #define TYPE_IMX_ENET "imx.enet" #include "hw/sysbus.h" +#include "hw/net/lan9118_phy.h" +#include "hw/irq.h" #include "net/net.h" #define ENET_EIR 1 @@ -XXX,XX +XXX,XX @@ struct IMXFECState { uint32_t tx_descriptor[ENET_TX_RING_NUM]; uint32_t tx_ring_num; - uint32_t phy_status; - uint32_t phy_control; - uint32_t phy_advertise; - uint32_t phy_int; - uint32_t phy_int_mask; + Lan9118PhyState mii; + IRQState mii_irq; uint32_t phy_num; bool phy_connected; struct IMXFECState *phy_consumer; 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 @@ static const VMStateDescription vmstate_imx_eth_txdescs = { static const VMStateDescription vmstate_imx_eth = { .name = TYPE_IMX_FEC, - .version_id = 2, - .minimum_version_id = 2, + .version_id = 3, + .minimum_version_id = 3, .fields = (const VMStateField[]) { VMSTATE_UINT32_ARRAY(regs, IMXFECState, ENET_MAX), VMSTATE_UINT32(rx_descriptor, IMXFECState), VMSTATE_UINT32(tx_descriptor[0], IMXFECState), - VMSTATE_UINT32(phy_status, IMXFECState), - VMSTATE_UINT32(phy_control, IMXFECState), - VMSTATE_UINT32(phy_advertise, IMXFECState), - VMSTATE_UINT32(phy_int, IMXFECState), - VMSTATE_UINT32(phy_int_mask, IMXFECState), VMSTATE_END_OF_LIST() }, .subsections = (const VMStateDescription * const []) { @@ -XXX,XX +XXX,XX @@ static const VMStateDescription vmstate_imx_eth = { }, }; -#define PHY_INT_ENERGYON (1 << 7) -#define PHY_INT_AUTONEG_COMPLETE (1 << 6) -#define PHY_INT_FAULT (1 << 5) -#define PHY_INT_DOWN (1 << 4) -#define PHY_INT_AUTONEG_LP (1 << 3) -#define PHY_INT_PARFAULT (1 << 2) -#define PHY_INT_AUTONEG_PAGE (1 << 1) - static void imx_eth_update(IMXFECState *s); /* @@ -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 imx_phy_update_irq(IMXFECState *s) +static void imx_phy_update_irq(void *opaque, int n, int level) { - imx_eth_update(s); -} - -static void imx_phy_update_link(IMXFECState *s) -{ - /* Autonegotiation status mirrors link status. */ - if (qemu_get_queue(s->nic)->link_down) { - trace_imx_phy_update_link("down"); - s->phy_status &= ~0x0024; - s->phy_int |= PHY_INT_DOWN; - } else { - trace_imx_phy_update_link("up"); - s->phy_status |= 0x0024; - s->phy_int |= PHY_INT_ENERGYON; - s->phy_int |= PHY_INT_AUTONEG_COMPLETE; - } - imx_phy_update_irq(s); + imx_eth_update(opaque); } static void imx_eth_set_link(NetClientState *nc) { - imx_phy_update_link(IMX_FEC(qemu_get_nic_opaque(nc))); -} - -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; - imx_phy_update_link(s); + lan9118_phy_update_link(&IMX_FEC(qemu_get_nic_opaque(nc))->mii, + nc->link_down); } static uint32_t imx_phy_read(IMXFECState *s, int reg) { - uint32_t val; uint32_t phy = reg / 32; if (!s->phy_connected) { @@ -XXX,XX +XXX,XX @@ static uint32_t imx_phy_read(IMXFECState *s, int reg) reg %= 32; - switch (reg) { - case 0: /* Basic Control */ - val = s->phy_control; - break; - case 1: /* Basic Status */ - val = s->phy_status; - break; - case 2: /* ID1 */ - val = 0x0007; - break; - case 3: /* ID2 */ - val = 0xc0d1; - break; - case 4: /* Auto-neg advertisement */ - val = s->phy_advertise; - break; - case 5: /* Auto-neg Link Partner Ability */ - val = 0x0f71; - break; - case 6: /* Auto-neg Expansion */ - val = 1; - break; - case 29: /* Interrupt source. */ - val = s->phy_int; - s->phy_int = 0; - imx_phy_update_irq(s); - break; - case 30: /* Interrupt mask */ - val = s->phy_int_mask; - break; - case 17: - case 18: - case 27: - case 31: - qemu_log_mask(LOG_UNIMP, "[%s.phy]%s: reg %d not implemented\n", - TYPE_IMX_FEC, __func__, reg); - val = 0; - break; - default: - qemu_log_mask(LOG_GUEST_ERROR, "[%s.phy]%s: Bad address at offset %d\n", - TYPE_IMX_FEC, __func__, reg); - val = 0; - break; - } - - trace_imx_phy_read(val, phy, reg); - - return val; + return lan9118_phy_read(&s->mii, reg); } static void imx_phy_write(IMXFECState *s, int reg, uint32_t val) @@ -XXX,XX +XXX,XX @@ static void imx_phy_write(IMXFECState *s, int reg, uint32_t val) reg %= 32; - trace_imx_phy_write(val, phy, reg); - - switch (reg) { - case 0: /* Basic Control */ - if (val & 0x8000) { - imx_phy_reset(s); - } else { - s->phy_control = val & 0x7980; - /* Complete autonegotiation immediately. */ - if (val & 0x1000) { - s->phy_status |= 0x0020; - } - } - break; - case 4: /* Auto-neg advertisement */ - s->phy_advertise = (val & 0x2d7f) | 0x80; - break; - case 30: /* Interrupt mask */ - s->phy_int_mask = val & 0xff; - imx_phy_update_irq(s); - break; - case 17: - case 18: - case 27: - case 31: - qemu_log_mask(LOG_UNIMP, "[%s.phy)%s: reg %d not implemented\n", - TYPE_IMX_FEC, __func__, reg); - break; - default: - qemu_log_mask(LOG_GUEST_ERROR, "[%s.phy]%s: Bad address at offset %d\n", - TYPE_IMX_FEC, __func__, reg); - break; - } + lan9118_phy_write(&s->mii, reg, val); } static void imx_fec_read_bd(IMXFECBufDesc *bd, dma_addr_t addr) @@ -XXX,XX +XXX,XX @@ static void imx_eth_reset(DeviceState *d) s->rx_descriptor = 0; memset(s->tx_descriptor, 0, sizeof(s->tx_descriptor)); - - /* We also reset the PHY */ - imx_phy_reset(s); } static uint32_t imx_default_read(IMXFECState *s, uint32_t index) @@ -XXX,XX +XXX,XX @@ static void imx_eth_realize(DeviceState *dev, Error **errp) sysbus_init_irq(sbd, &s->irq[0]); sysbus_init_irq(sbd, &s->irq[1]); + qemu_init_irq(&s->mii_irq, imx_phy_update_irq, s, 0); + object_initialize_child(OBJECT(s), "mii", &s->mii, TYPE_LAN9118_PHY); + if (!sysbus_realize_and_unref(SYS_BUS_DEVICE(&s->mii), errp)) { + return; + } + qdev_connect_gpio_out(DEVICE(&s->mii), 0, &s->mii_irq); + qemu_macaddr_default_if_unset(&s->conf.macaddr); s->nic = qemu_new_nic(&imx_eth_net_info, &s->conf, diff --git a/hw/net/lan9118_phy.c b/hw/net/lan9118_phy.c index XXXXXXX..XXXXXXX 100644 --- a/hw/net/lan9118_phy.c +++ b/hw/net/lan9118_phy.c @@ -XXX,XX +XXX,XX @@ * Copyright (c) 2009 CodeSourcery, LLC. * Written by Paul Brook * + * Copyright (c) 2013 Jean-Christophe Dubois. <jcd@tribudubois.net> + * * This code is licensed under the GNU GPL v2 * * Contributions after 2012-01-13 are licensed under the terms of the @@ -XXX,XX +XXX,XX @@ #include "hw/resettable.h" #include "migration/vmstate.h" #include "qemu/log.h" +#include "trace.h" #define PHY_INT_ENERGYON (1 << 7) #define PHY_INT_AUTONEG_COMPLETE (1 << 6) @@ -XXX,XX +XXX,XX @@ uint16_t lan9118_phy_read(Lan9118PhyState *s, int reg) switch (reg) { case 0: /* Basic Control */ - return s->control; + val = s->control; + break; case 1: /* Basic Status */ - return s->status; + val = s->status; + break; case 2: /* ID1 */ - return 0x0007; + val = 0x0007; + break; case 3: /* ID2 */ - return 0xc0d1; + val = 0xc0d1; + break; case 4: /* Auto-neg advertisement */ - return s->advertise; + val = s->advertise; + break; case 5: /* Auto-neg Link Partner Ability */ - return 0x0f71; + val = 0x0f71; + break; case 6: /* Auto-neg Expansion */ - return 1; - /* TODO 17, 18, 27, 29, 30, 31 */ + val = 1; + break; case 29: /* Interrupt source. */ val = s->ints; s->ints = 0; lan9118_phy_update_irq(s); - return val; + break; case 30: /* Interrupt mask */ - return s->int_mask; + val = s->int_mask; + break; + case 17: + case 18: + case 27: + case 31: + qemu_log_mask(LOG_UNIMP, "%s: reg %d not implemented\n", + __func__, reg); + val = 0; + break; default: - qemu_log_mask(LOG_GUEST_ERROR, - "lan9118_phy_read: PHY read reg %d\n", reg); - return 0; + qemu_log_mask(LOG_GUEST_ERROR, "%s: Bad address at offset %d\n", + __func__, reg); + val = 0; + break; } + + trace_lan9118_phy_read(val, reg); + + return val; } void lan9118_phy_write(Lan9118PhyState *s, int reg, uint16_t val) { + trace_lan9118_phy_write(val, reg); + switch (reg) { case 0: /* Basic Control */ if (val & 0x8000) { lan9118_phy_reset(s); - break; - } - s->control = val & 0x7980; - /* Complete autonegotiation immediately. */ - if (val & 0x1000) { - s->status |= 0x0020; + } else { + s->control = val & 0x7980; + /* Complete autonegotiation immediately. */ + if (val & 0x1000) { + s->status |= 0x0020; + } } break; case 4: /* Auto-neg advertisement */ s->advertise = (val & 0x2d7f) | 0x80; break; - /* TODO 17, 18, 27, 31 */ case 30: /* Interrupt mask */ s->int_mask = val & 0xff; lan9118_phy_update_irq(s); break; + case 17: + case 18: + case 27: + case 31: + qemu_log_mask(LOG_UNIMP, "%s: reg %d not implemented\n", + __func__, reg); + break; default: - qemu_log_mask(LOG_GUEST_ERROR, - "lan9118_phy_write: PHY write reg %d = 0x%04x\n", reg, val); + qemu_log_mask(LOG_GUEST_ERROR, "%s: Bad address at offset %d\n", + __func__, reg); + break; } } @@ -XXX,XX +XXX,XX @@ void lan9118_phy_update_link(Lan9118PhyState *s, bool link_down) /* Autonegotiation status mirrors link status. */ if (link_down) { + trace_lan9118_phy_update_link("down"); s->status &= ~0x0024; s->ints |= PHY_INT_DOWN; } else { + trace_lan9118_phy_update_link("up"); s->status |= 0x0024; s->ints |= PHY_INT_ENERGYON; s->ints |= PHY_INT_AUTONEG_COMPLETE; @@ -XXX,XX +XXX,XX @@ void lan9118_phy_update_link(Lan9118PhyState *s, bool link_down) void lan9118_phy_reset(Lan9118PhyState *s) { + trace_lan9118_phy_reset(); + s->control = 0x3000; s->status = 0x7809; s->advertise = 0x01e1; @@ -XXX,XX +XXX,XX @@ static const VMStateDescription vmstate_lan9118_phy = { .version_id = 1, .minimum_version_id = 1, .fields = (const VMStateField[]) { - VMSTATE_UINT16(control, Lan9118PhyState), VMSTATE_UINT16(status, Lan9118PhyState), + VMSTATE_UINT16(control, Lan9118PhyState), VMSTATE_UINT16(advertise, Lan9118PhyState), VMSTATE_UINT16(ints, Lan9118PhyState), VMSTATE_UINT16(int_mask, Lan9118PhyState), diff --git a/hw/net/Kconfig b/hw/net/Kconfig index XXXXXXX..XXXXXXX 100644 --- a/hw/net/Kconfig +++ b/hw/net/Kconfig @@ -XXX,XX +XXX,XX @@ config ALLWINNER_SUN8I_EMAC config IMX_FEC bool + select LAN9118_PHY config CADENCE bool 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 @@ allwinner_sun8i_emac_set_link(bool active) "Set link: active=%u" allwinner_sun8i_emac_read(uint64_t offset, uint64_t val) "MMIO read: offset=0x%" PRIx64 " value=0x%" PRIx64 allwinner_sun8i_emac_write(uint64_t offset, uint64_t val) "MMIO write: offset=0x%" PRIx64 " value=0x%" PRIx64 +# lan9118_phy.c +lan9118_phy_read(uint16_t val, int reg) "[0x%02x] -> 0x%04" PRIx16 +lan9118_phy_write(uint16_t val, int reg) "[0x%02x] <- 0x%04" PRIx16 +lan9118_phy_update_link(const char *s) "%s" +lan9118_phy_reset(void) "" + # lance.c lance_mem_readw(uint64_t addr, uint32_t ret) "addr=0x%"PRIx64"val=0x%04x" lance_mem_writew(uint64_t addr, uint32_t val) "addr=0x%"PRIx64"val=0x%04x" @@ -XXX,XX +XXX,XX @@ 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 phy, int reg) "0x%04"PRIx32" <= phy[%d].reg[%d]" imx_phy_read_num(int phy, int configured) "read request from unconfigured phy %d (configured %d)" -imx_phy_write(uint32_t val, int phy, int reg) "0x%04"PRIx32" => phy[%d].reg[%d]" imx_phy_write_num(int phy, int configured) "write request to unconfigured phy %d (configured %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" -- 2.34.1 From: Bernhard Beschow <shentey@gmail.com> Turns 0x70 into 0xe0 (== 0x70 << 1) which adds the missing MII_ANLPAR_TX and fixes the MSB of selector field to be zero, as specified in the datasheet. Fixes: 2a424990170b "LAN9118 emulation" Signed-off-by: Bernhard Beschow <shentey@gmail.com> Tested-by: Guenter Roeck <linux@roeck-us.net> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Message-id: 20241102125724.532843-4-shentey@gmail.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/net/lan9118_phy.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/hw/net/lan9118_phy.c b/hw/net/lan9118_phy.c index XXXXXXX..XXXXXXX 100644 --- a/hw/net/lan9118_phy.c +++ b/hw/net/lan9118_phy.c @@ -XXX,XX +XXX,XX @@ uint16_t lan9118_phy_read(Lan9118PhyState *s, int reg) val = s->advertise; break; case 5: /* Auto-neg Link Partner Ability */ - val = 0x0f71; + val = 0x0fe1; break; case 6: /* Auto-neg Expansion */ val = 1; -- 2.34.1 From: Bernhard Beschow <shentey@gmail.com> Prefer named constants over magic values for better readability. Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Bernhard Beschow <shentey@gmail.com> Tested-by: Guenter Roeck <linux@roeck-us.net> Message-id: 20241102125724.532843-5-shentey@gmail.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- include/hw/net/mii.h | 6 +++++ hw/net/lan9118_phy.c | 63 ++++++++++++++++++++++++++++---------------- 2 files changed, 46 insertions(+), 23 deletions(-) diff --git a/include/hw/net/mii.h b/include/hw/net/mii.h index XXXXXXX..XXXXXXX 100644 --- a/include/hw/net/mii.h +++ b/include/hw/net/mii.h @@ -XXX,XX +XXX,XX @@ #define MII_BMSR_JABBER (1 << 1) /* Jabber detected */ #define MII_BMSR_EXTCAP (1 << 0) /* Ext-reg capability */ +#define MII_ANAR_RFAULT (1 << 13) /* Say we can detect faults */ #define MII_ANAR_PAUSE_ASYM (1 << 11) /* Try for asymmetric pause */ #define MII_ANAR_PAUSE (1 << 10) /* Try for pause */ #define MII_ANAR_TXFD (1 << 8) @@ -XXX,XX +XXX,XX @@ #define MII_ANAR_10FD (1 << 6) #define MII_ANAR_10 (1 << 5) #define MII_ANAR_CSMACD (1 << 0) +#define MII_ANAR_SELECT (0x001f) /* Selector bits */ #define MII_ANLPAR_ACK (1 << 14) #define MII_ANLPAR_PAUSEASY (1 << 11) /* can pause asymmetrically */ @@ -XXX,XX +XXX,XX @@ #define RTL8201CP_PHYID1 0x0000 #define RTL8201CP_PHYID2 0x8201 +/* SMSC LAN9118 */ +#define SMSCLAN9118_PHYID1 0x0007 +#define SMSCLAN9118_PHYID2 0xc0d1 + /* RealTek 8211E */ #define RTL8211E_PHYID1 0x001c #define RTL8211E_PHYID2 0xc915 diff --git a/hw/net/lan9118_phy.c b/hw/net/lan9118_phy.c index XXXXXXX..XXXXXXX 100644 --- a/hw/net/lan9118_phy.c +++ b/hw/net/lan9118_phy.c @@ -XXX,XX +XXX,XX @@ #include "qemu/osdep.h" #include "hw/net/lan9118_phy.h" +#include "hw/net/mii.h" #include "hw/irq.h" #include "hw/resettable.h" #include "migration/vmstate.h" @@ -XXX,XX +XXX,XX @@ uint16_t lan9118_phy_read(Lan9118PhyState *s, int reg) uint16_t val; switch (reg) { - case 0: /* Basic Control */ + case MII_BMCR: val = s->control; break; - case 1: /* Basic Status */ + case MII_BMSR: val = s->status; break; - case 2: /* ID1 */ - val = 0x0007; + case MII_PHYID1: + val = SMSCLAN9118_PHYID1; break; - case 3: /* ID2 */ - val = 0xc0d1; + case MII_PHYID2: + val = SMSCLAN9118_PHYID2; break; - case 4: /* Auto-neg advertisement */ + case MII_ANAR: val = s->advertise; break; - case 5: /* Auto-neg Link Partner Ability */ - val = 0x0fe1; + case MII_ANLPAR: + val = MII_ANLPAR_PAUSEASY | MII_ANLPAR_PAUSE | MII_ANLPAR_T4 | + MII_ANLPAR_TXFD | MII_ANLPAR_TX | MII_ANLPAR_10FD | + MII_ANLPAR_10 | MII_ANLPAR_CSMACD; break; - case 6: /* Auto-neg Expansion */ - val = 1; + case MII_ANER: + val = MII_ANER_NWAY; break; case 29: /* Interrupt source. */ val = s->ints; @@ -XXX,XX +XXX,XX @@ void lan9118_phy_write(Lan9118PhyState *s, int reg, uint16_t val) trace_lan9118_phy_write(val, reg); switch (reg) { - case 0: /* Basic Control */ - if (val & 0x8000) { + case MII_BMCR: + if (val & MII_BMCR_RESET) { lan9118_phy_reset(s); } else { - s->control = val & 0x7980; + s->control = val & (MII_BMCR_LOOPBACK | MII_BMCR_SPEED100 | + MII_BMCR_AUTOEN | MII_BMCR_PDOWN | MII_BMCR_FD | + MII_BMCR_CTST); /* Complete autonegotiation immediately. */ - if (val & 0x1000) { - s->status |= 0x0020; + if (val & MII_BMCR_AUTOEN) { + s->status |= MII_BMSR_AN_COMP; } } break; - case 4: /* Auto-neg advertisement */ - s->advertise = (val & 0x2d7f) | 0x80; + case MII_ANAR: + s->advertise = (val & (MII_ANAR_RFAULT | MII_ANAR_PAUSE_ASYM | + MII_ANAR_PAUSE | MII_ANAR_10FD | MII_ANAR_10 | + MII_ANAR_SELECT)) + | MII_ANAR_TX; break; case 30: /* Interrupt mask */ s->int_mask = val & 0xff; @@ -XXX,XX +XXX,XX @@ void lan9118_phy_update_link(Lan9118PhyState *s, bool link_down) /* Autonegotiation status mirrors link status. */ if (link_down) { trace_lan9118_phy_update_link("down"); - s->status &= ~0x0024; + s->status &= ~(MII_BMSR_AN_COMP | MII_BMSR_LINK_ST); s->ints |= PHY_INT_DOWN; } else { trace_lan9118_phy_update_link("up"); - s->status |= 0x0024; + s->status |= MII_BMSR_AN_COMP | MII_BMSR_LINK_ST; s->ints |= PHY_INT_ENERGYON; s->ints |= PHY_INT_AUTONEG_COMPLETE; } @@ -XXX,XX +XXX,XX @@ void lan9118_phy_reset(Lan9118PhyState *s) { trace_lan9118_phy_reset(); - s->control = 0x3000; - s->status = 0x7809; - s->advertise = 0x01e1; + s->control = MII_BMCR_AUTOEN | MII_BMCR_SPEED100; + s->status = MII_BMSR_100TX_FD + | MII_BMSR_100TX_HD + | MII_BMSR_10T_FD + | MII_BMSR_10T_HD + | MII_BMSR_AUTONEG + | MII_BMSR_EXTCAP; + s->advertise = MII_ANAR_TXFD + | MII_ANAR_TX + | MII_ANAR_10FD + | MII_ANAR_10 + | MII_ANAR_CSMACD; s->int_mask = 0; s->ints = 0; lan9118_phy_update_link(s, s->link_down); -- 2.34.1 From: Bernhard Beschow <shentey@gmail.com> The real device advertises this mode and the device model already advertises 100 mbps half duplex and 10 mbps full+half duplex. So advertise this mode to make the model more realistic. Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Bernhard Beschow <shentey@gmail.com> Tested-by: Guenter Roeck <linux@roeck-us.net> Message-id: 20241102125724.532843-6-shentey@gmail.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- hw/net/lan9118_phy.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/hw/net/lan9118_phy.c b/hw/net/lan9118_phy.c index XXXXXXX..XXXXXXX 100644 --- a/hw/net/lan9118_phy.c +++ b/hw/net/lan9118_phy.c @@ -XXX,XX +XXX,XX @@ void lan9118_phy_write(Lan9118PhyState *s, int reg, uint16_t val) break; case MII_ANAR: s->advertise = (val & (MII_ANAR_RFAULT | MII_ANAR_PAUSE_ASYM | - MII_ANAR_PAUSE | MII_ANAR_10FD | MII_ANAR_10 | - MII_ANAR_SELECT)) + MII_ANAR_PAUSE | MII_ANAR_TXFD | MII_ANAR_10FD | + MII_ANAR_10 | MII_ANAR_SELECT)) | MII_ANAR_TX; break; case 30: /* Interrupt mask */ -- 2.34.1 For IEEE fused multiply-add, the (0 * inf) + NaN case should raise Invalid for the multiplication of 0 by infinity. Currently we handle this in the per-architecture ifdef ladder in pickNaNMulAdd(). However, since this isn't really architecture specific we can hoist it up to the generic code. For the cases where the infzero test in pickNaNMulAdd was returning 2, we can delete the check entirely and allow the code to fall into the normal pick-a-NaN handling, because this will return 2 anyway (input 'c' being the only NaN in this case). For the cases where infzero was returning 3 to indicate "return the default NaN", we must retain that "return 3". For Arm, this looks like it might be a behaviour change because we used to set float_flag_invalid | float_flag_invalid_imz only if C is a quiet NaN. However, it is not, because Arm target code never looks at float_flag_invalid_imz, and for the (0 * inf) + SNaN case we already raised float_flag_invalid via the "abc_mask & float_cmask_snan" check in pick_nan_muladd. For any target architecture using the "default implementation" at the bottom of the ifdef, this is a behaviour change but will be fixing a bug (where we failed to raise the Invalid exception for (0 * inf + QNaN). The architectures using the default case are: * hppa * i386 * sh4 * tricore The x86, Tricore and SH4 CPU architecture manuals are clear that this should have raised Invalid; HPPA is a bit vaguer but still seems clear enough. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-2-peter.maydell@linaro.org --- fpu/softfloat-parts.c.inc | 13 +++++++------ fpu/softfloat-specialize.c.inc | 29 +---------------------------- 2 files changed, 8 insertions(+), 34 deletions(-) diff --git a/fpu/softfloat-parts.c.inc b/fpu/softfloat-parts.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-parts.c.inc +++ b/fpu/softfloat-parts.c.inc @@ -XXX,XX +XXX,XX @@ static FloatPartsN *partsN(pick_nan_muladd)(FloatPartsN *a, FloatPartsN *b, int ab_mask, int abc_mask) { int which; + bool infzero = (ab_mask == float_cmask_infzero); if (unlikely(abc_mask & float_cmask_snan)) { float_raise(float_flag_invalid | float_flag_invalid_snan, s); } - which = pickNaNMulAdd(a->cls, b->cls, c->cls, - ab_mask == float_cmask_infzero, s); + if (infzero) { + /* This is (0 * inf) + NaN or (inf * 0) + NaN */ + float_raise(float_flag_invalid | float_flag_invalid_imz, s); + } + + which = pickNaNMulAdd(a->cls, b->cls, c->cls, infzero, s); if (s->default_nan_mode || which == 3) { - /* - * Note that this check is after pickNaNMulAdd so that function - * has an opportunity to set the Invalid flag for infzero. - */ parts_default_nan(a, s); return a; } diff --git a/fpu/softfloat-specialize.c.inc b/fpu/softfloat-specialize.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-specialize.c.inc +++ b/fpu/softfloat-specialize.c.inc @@ -XXX,XX +XXX,XX @@ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls, * the default NaN */ if (infzero && is_qnan(c_cls)) { - float_raise(float_flag_invalid | float_flag_invalid_imz, status); return 3; } @@ -XXX,XX +XXX,XX @@ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls, * case sets InvalidOp and returns the default NaN */ if (infzero) { - float_raise(float_flag_invalid | float_flag_invalid_imz, status); return 3; } /* Prefer sNaN over qNaN, in the a, b, c order. */ @@ -XXX,XX +XXX,XX @@ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls, * For MIPS systems that conform to IEEE754-2008, the (inf,zero,nan) * case sets InvalidOp and returns the input value 'c' */ - if (infzero) { - float_raise(float_flag_invalid | float_flag_invalid_imz, status); - return 2; - } /* Prefer sNaN over qNaN, in the c, a, b order. */ if (is_snan(c_cls)) { return 2; @@ -XXX,XX +XXX,XX @@ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls, * For LoongArch systems that conform to IEEE754-2008, the (inf,zero,nan) * case sets InvalidOp and returns the input value 'c' */ - if (infzero) { - float_raise(float_flag_invalid | float_flag_invalid_imz, status); - return 2; - } + /* Prefer sNaN over qNaN, in the c, a, b order. */ if (is_snan(c_cls)) { return 2; @@ -XXX,XX +XXX,XX @@ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls, * to return an input NaN if we have one (ie c) rather than generating * a default NaN */ - if (infzero) { - float_raise(float_flag_invalid | float_flag_invalid_imz, status); - return 2; - } /* If fRA is a NaN return it; otherwise if fRB is a NaN return it; * otherwise return fRC. Note that muladd on PPC is (fRA * fRC) + frB @@ -XXX,XX +XXX,XX @@ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls, return 1; } #elif defined(TARGET_RISCV) - /* For RISC-V, InvalidOp is set when multiplicands are Inf and zero */ - if (infzero) { - float_raise(float_flag_invalid | float_flag_invalid_imz, status); - } return 3; /* default NaN */ #elif defined(TARGET_S390X) if (infzero) { - float_raise(float_flag_invalid | float_flag_invalid_imz, status); return 3; } @@ -XXX,XX +XXX,XX @@ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls, return 2; } #elif defined(TARGET_SPARC) - /* For (inf,0,nan) return c. */ - if (infzero) { - float_raise(float_flag_invalid | float_flag_invalid_imz, status); - return 2; - } /* Prefer SNaN over QNaN, order C, B, A. */ if (is_snan(c_cls)) { return 2; @@ -XXX,XX +XXX,XX @@ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls, * For Xtensa, the (inf,zero,nan) case sets InvalidOp and returns * an input NaN if we have one (ie c). */ - if (infzero) { - float_raise(float_flag_invalid | float_flag_invalid_imz, status); - return 2; - } if (status->use_first_nan) { if (is_nan(a_cls)) { return 0; -- 2.34.1 If the target sets default_nan_mode then we're always going to return the default NaN, and pickNaNMulAdd() no longer has any side effects. For consistency with pickNaN(), check for default_nan_mode before calling pickNaNMulAdd(). When we convert pickNaNMulAdd() to allow runtime selection of the NaN propagation rule, this means we won't have to make the targets which use default_nan_mode also set a propagation rule. Since RiscV always uses default_nan_mode, this allows us to remove its ifdef case from pickNaNMulAdd(). Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-3-peter.maydell@linaro.org --- fpu/softfloat-parts.c.inc | 8 ++++++-- fpu/softfloat-specialize.c.inc | 9 +++++++-- 2 files changed, 13 insertions(+), 4 deletions(-) diff --git a/fpu/softfloat-parts.c.inc b/fpu/softfloat-parts.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-parts.c.inc +++ b/fpu/softfloat-parts.c.inc @@ -XXX,XX +XXX,XX @@ static FloatPartsN *partsN(pick_nan_muladd)(FloatPartsN *a, FloatPartsN *b, float_raise(float_flag_invalid | float_flag_invalid_imz, s); } - which = pickNaNMulAdd(a->cls, b->cls, c->cls, infzero, s); + if (s->default_nan_mode) { + which = 3; + } else { + which = pickNaNMulAdd(a->cls, b->cls, c->cls, infzero, s); + } - if (s->default_nan_mode || which == 3) { + if (which == 3) { parts_default_nan(a, s); return a; } diff --git a/fpu/softfloat-specialize.c.inc b/fpu/softfloat-specialize.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-specialize.c.inc +++ b/fpu/softfloat-specialize.c.inc @@ -XXX,XX +XXX,XX @@ static int pickNaN(FloatClass a_cls, FloatClass b_cls, static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls, bool infzero, float_status *status) { + /* + * We guarantee not to require the target to tell us how to + * pick a NaN if we're always returning the default NaN. + * But if we're not in default-NaN mode then the target must + * specify. + */ + assert(!status->default_nan_mode); #if defined(TARGET_ARM) /* For ARM, the (inf,zero,qnan) case sets InvalidOp and returns * the default NaN @@ -XXX,XX +XXX,XX @@ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls, } else { return 1; } -#elif defined(TARGET_RISCV) - return 3; /* default NaN */ #elif defined(TARGET_S390X) if (infzero) { return 3; -- 2.34.1 IEEE 758 does not define a fixed rule for what NaN to return in the case of a fused multiply-add of inf * 0 + NaN. Different architectures thus do different things: * some return the default NaN * some return the input NaN * Arm returns the default NaN if the input NaN is quiet, and the input NaN if it is signalling We want to make this logic be runtime selected rather than hardcoded into the binary, because: * this will let us have multiple targets in one QEMU binary * the Arm FEAT_AFP architectural feature includes letting the guest select a NaN propagation rule at runtime In this commit we add an enum for the propagation rule, the field in float_status, and the corresponding getters and setters. We change pickNaNMulAdd to honour this, but because all targets still leave this field at its default 0 value, the fallback logic will pick the rule type with the old ifdef ladder. Note that four architectures both use the muladd softfloat functions and did not have a branch of the ifdef ladder to specify their behaviour (and so were ending up with the "default" case, probably wrongly): i386, HPPA, SH4 and Tricore. SH4 and Tricore both set default_nan_mode, and so will never get into pickNaNMulAdd(). For HPPA and i386 we retain the same behaviour as the old default-case, which is to not ever return the default NaN. This might not be correct but it is not a behaviour change. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-4-peter.maydell@linaro.org --- include/fpu/softfloat-helpers.h | 11 ++++ include/fpu/softfloat-types.h | 23 +++++++++ fpu/softfloat-specialize.c.inc | 91 ++++++++++++++++++++++----------- 3 files changed, 95 insertions(+), 30 deletions(-) diff --git a/include/fpu/softfloat-helpers.h b/include/fpu/softfloat-helpers.h index XXXXXXX..XXXXXXX 100644 --- a/include/fpu/softfloat-helpers.h +++ b/include/fpu/softfloat-helpers.h @@ -XXX,XX +XXX,XX @@ static inline void set_float_2nan_prop_rule(Float2NaNPropRule rule, status->float_2nan_prop_rule = rule; } +static inline void set_float_infzeronan_rule(FloatInfZeroNaNRule rule, + float_status *status) +{ + status->float_infzeronan_rule = rule; +} + static inline void set_flush_to_zero(bool val, float_status *status) { status->flush_to_zero = val; @@ -XXX,XX +XXX,XX @@ static inline Float2NaNPropRule get_float_2nan_prop_rule(float_status *status) return status->float_2nan_prop_rule; } +static inline FloatInfZeroNaNRule get_float_infzeronan_rule(float_status *status) +{ + return status->float_infzeronan_rule; +} + static inline bool get_flush_to_zero(float_status *status) { return status->flush_to_zero; diff --git a/include/fpu/softfloat-types.h b/include/fpu/softfloat-types.h index XXXXXXX..XXXXXXX 100644 --- a/include/fpu/softfloat-types.h +++ b/include/fpu/softfloat-types.h @@ -XXX,XX +XXX,XX @@ typedef enum __attribute__((__packed__)) { float_2nan_prop_x87, } Float2NaNPropRule; +/* + * Rule for result of fused multiply-add 0 * Inf + NaN. + * This must be a NaN, but implementations differ on whether this + * is the input NaN or the default NaN. + * + * You don't need to set this if default_nan_mode is enabled. + * When not in default-NaN mode, it is an error for the target + * not to set the rule in float_status if it uses muladd, and we + * will assert if we need to handle an input NaN and no rule was + * selected. + */ +typedef enum __attribute__((__packed__)) { + /* No propagation rule specified */ + float_infzeronan_none = 0, + /* Result is never the default NaN (so always the input NaN) */ + float_infzeronan_dnan_never, + /* Result is always the default NaN */ + float_infzeronan_dnan_always, + /* Result is the default NaN if the input NaN is quiet */ + float_infzeronan_dnan_if_qnan, +} FloatInfZeroNaNRule; + /* * Floating Point Status. Individual architectures may maintain * several versions of float_status for different functions. The @@ -XXX,XX +XXX,XX @@ typedef struct float_status { FloatRoundMode float_rounding_mode; FloatX80RoundPrec floatx80_rounding_precision; Float2NaNPropRule float_2nan_prop_rule; + FloatInfZeroNaNRule float_infzeronan_rule; bool tininess_before_rounding; /* should denormalised results go to zero and set the inexact flag? */ bool flush_to_zero; diff --git a/fpu/softfloat-specialize.c.inc b/fpu/softfloat-specialize.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-specialize.c.inc +++ b/fpu/softfloat-specialize.c.inc @@ -XXX,XX +XXX,XX @@ static int pickNaN(FloatClass a_cls, FloatClass b_cls, static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls, bool infzero, float_status *status) { + FloatInfZeroNaNRule rule = status->float_infzeronan_rule; + /* * We guarantee not to require the target to tell us how to * pick a NaN if we're always returning the default NaN. @@ -XXX,XX +XXX,XX @@ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls, * specify. */ assert(!status->default_nan_mode); + + if (rule == float_infzeronan_none) { + /* + * Temporarily fall back to ifdef ladder + */ #if defined(TARGET_ARM) - /* For ARM, the (inf,zero,qnan) case sets InvalidOp and returns - * the default NaN - */ - if (infzero && is_qnan(c_cls)) { - return 3; + /* + * For ARM, the (inf,zero,qnan) case returns the default NaN, + * but (inf,zero,snan) returns the input NaN. + */ + rule = float_infzeronan_dnan_if_qnan; +#elif defined(TARGET_MIPS) + if (snan_bit_is_one(status)) { + /* + * For MIPS systems that conform to IEEE754-1985, the (inf,zero,nan) + * case sets InvalidOp and returns the default NaN + */ + rule = float_infzeronan_dnan_always; + } else { + /* + * For MIPS systems that conform to IEEE754-2008, the (inf,zero,nan) + * case sets InvalidOp and returns the input value 'c' + */ + rule = float_infzeronan_dnan_never; + } +#elif defined(TARGET_PPC) || defined(TARGET_SPARC) || \ + defined(TARGET_XTENSA) || defined(TARGET_HPPA) || \ + defined(TARGET_I386) || defined(TARGET_LOONGARCH) + /* + * For LoongArch systems that conform to IEEE754-2008, the (inf,zero,nan) + * case sets InvalidOp and returns the input value 'c' + */ + /* + * For PPC, the (inf,zero,qnan) case sets InvalidOp, but we prefer + * to return an input NaN if we have one (ie c) rather than generating + * a default NaN + */ + rule = float_infzeronan_dnan_never; +#elif defined(TARGET_S390X) + rule = float_infzeronan_dnan_always; +#endif } + if (infzero) { + /* + * Inf * 0 + NaN -- some implementations return the default NaN here, + * and some return the input NaN. + */ + switch (rule) { + case float_infzeronan_dnan_never: + return 2; + case float_infzeronan_dnan_always: + return 3; + case float_infzeronan_dnan_if_qnan: + return is_qnan(c_cls) ? 3 : 2; + default: + g_assert_not_reached(); + } + } + +#if defined(TARGET_ARM) + /* This looks different from the ARM ARM pseudocode, because the ARM ARM * puts the operands to a fused mac operation (a*b)+c in the order c,a,b. */ @@ -XXX,XX +XXX,XX @@ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls, } #elif defined(TARGET_MIPS) if (snan_bit_is_one(status)) { - /* - * For MIPS systems that conform to IEEE754-1985, the (inf,zero,nan) - * case sets InvalidOp and returns the default NaN - */ - if (infzero) { - return 3; - } /* Prefer sNaN over qNaN, in the a, b, c order. */ if (is_snan(a_cls)) { return 0; @@ -XXX,XX +XXX,XX @@ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls, return 2; } } else { - /* - * For MIPS systems that conform to IEEE754-2008, the (inf,zero,nan) - * case sets InvalidOp and returns the input value 'c' - */ /* Prefer sNaN over qNaN, in the c, a, b order. */ if (is_snan(c_cls)) { return 2; @@ -XXX,XX +XXX,XX @@ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls, } } #elif defined(TARGET_LOONGARCH64) - /* - * For LoongArch systems that conform to IEEE754-2008, the (inf,zero,nan) - * case sets InvalidOp and returns the input value 'c' - */ - /* Prefer sNaN over qNaN, in the c, a, b order. */ if (is_snan(c_cls)) { return 2; @@ -XXX,XX +XXX,XX @@ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls, return 1; } #elif defined(TARGET_PPC) - /* For PPC, the (inf,zero,qnan) case sets InvalidOp, but we prefer - * to return an input NaN if we have one (ie c) rather than generating - * a default NaN - */ - /* If fRA is a NaN return it; otherwise if fRB is a NaN return it; * otherwise return fRC. Note that muladd on PPC is (fRA * fRC) + frB */ @@ -XXX,XX +XXX,XX @@ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls, return 1; } #elif defined(TARGET_S390X) - if (infzero) { - return 3; - } - if (is_snan(a_cls)) { return 0; } else if (is_snan(b_cls)) { -- 2.34.1 Explicitly set a rule in the softfloat tests for the inf-zero-nan muladd special case. In meson.build we put -DTARGET_ARM in fpcflags, and so we should select here the Arm rule of float_infzeronan_dnan_if_qnan. Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Message-id: 20241202131347.498124-5-peter.maydell@linaro.org --- tests/fp/fp-bench.c | 5 +++++ tests/fp/fp-test.c | 5 +++++ 2 files changed, 10 insertions(+) diff --git a/tests/fp/fp-bench.c b/tests/fp/fp-bench.c index XXXXXXX..XXXXXXX 100644 --- a/tests/fp/fp-bench.c +++ b/tests/fp/fp-bench.c @@ -XXX,XX +XXX,XX @@ static void run_bench(void) { bench_func_t f; + /* + * These implementation-defined choices for various things IEEE + * doesn't specify match those used by the Arm architecture. + */ set_float_2nan_prop_rule(float_2nan_prop_s_ab, &soft_status); + set_float_infzeronan_rule(float_infzeronan_dnan_if_qnan, &soft_status); f = bench_funcs[operation][precision]; g_assert(f); diff --git a/tests/fp/fp-test.c b/tests/fp/fp-test.c index XXXXXXX..XXXXXXX 100644 --- a/tests/fp/fp-test.c +++ b/tests/fp/fp-test.c @@ -XXX,XX +XXX,XX @@ void run_test(void) { unsigned int i; + /* + * These implementation-defined choices for various things IEEE + * doesn't specify match those used by the Arm architecture. + */ set_float_2nan_prop_rule(float_2nan_prop_s_ab, &qsf); + set_float_infzeronan_rule(float_infzeronan_dnan_if_qnan, &qsf); genCases_setLevel(test_level); verCases_maxErrorCount = n_max_errors; -- 2.34.1 Set the FloatInfZeroNaNRule explicitly for the Arm target, so we can remove the ifdef from pickNaNMulAdd(). Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-6-peter.maydell@linaro.org --- target/arm/cpu.c | 3 +++ fpu/softfloat-specialize.c.inc | 8 +------- 2 files changed, 4 insertions(+), 7 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_register_el_change_hook(ARMCPU *cpu, ARMELChangeHookFn *hook, * * tininess-before-rounding * * 2-input NaN propagation prefers SNaN over QNaN, and then * operand A over operand B (see FPProcessNaNs() pseudocode) + * * 0 * Inf + NaN returns the default NaN if the input NaN is quiet, + * and the input NaN if it is signalling */ static void arm_set_default_fp_behaviours(float_status *s) { set_float_detect_tininess(float_tininess_before_rounding, s); set_float_2nan_prop_rule(float_2nan_prop_s_ab, s); + set_float_infzeronan_rule(float_infzeronan_dnan_if_qnan, s); } static void cp_reg_reset(gpointer key, gpointer value, gpointer opaque) diff --git a/fpu/softfloat-specialize.c.inc b/fpu/softfloat-specialize.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-specialize.c.inc +++ b/fpu/softfloat-specialize.c.inc @@ -XXX,XX +XXX,XX @@ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls, /* * Temporarily fall back to ifdef ladder */ -#if defined(TARGET_ARM) - /* - * For ARM, the (inf,zero,qnan) case returns the default NaN, - * but (inf,zero,snan) returns the input NaN. - */ - rule = float_infzeronan_dnan_if_qnan; -#elif defined(TARGET_MIPS) +#if defined(TARGET_MIPS) if (snan_bit_is_one(status)) { /* * For MIPS systems that conform to IEEE754-1985, the (inf,zero,nan) -- 2.34.1 Set the FloatInfZeroNaNRule explicitly for s390, so we can remove the ifdef from pickNaNMulAdd(). Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-7-peter.maydell@linaro.org --- target/s390x/cpu.c | 2 ++ fpu/softfloat-specialize.c.inc | 2 -- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/target/s390x/cpu.c b/target/s390x/cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/s390x/cpu.c +++ b/target/s390x/cpu.c @@ -XXX,XX +XXX,XX @@ static void s390_cpu_reset_hold(Object *obj, ResetType type) set_float_detect_tininess(float_tininess_before_rounding, &env->fpu_status); set_float_2nan_prop_rule(float_2nan_prop_s_ab, &env->fpu_status); + set_float_infzeronan_rule(float_infzeronan_dnan_always, + &env->fpu_status); /* fall through */ case RESET_TYPE_S390_CPU_NORMAL: env->psw.mask &= ~PSW_MASK_RI; diff --git a/fpu/softfloat-specialize.c.inc b/fpu/softfloat-specialize.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-specialize.c.inc +++ b/fpu/softfloat-specialize.c.inc @@ -XXX,XX +XXX,XX @@ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls, * a default NaN */ rule = float_infzeronan_dnan_never; -#elif defined(TARGET_S390X) - rule = float_infzeronan_dnan_always; #endif } -- 2.34.1 Set the FloatInfZeroNaNRule explicitly for the PPC target, so we can remove the ifdef from pickNaNMulAdd(). Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-8-peter.maydell@linaro.org --- target/ppc/cpu_init.c | 7 +++++++ fpu/softfloat-specialize.c.inc | 7 +------ 2 files changed, 8 insertions(+), 6 deletions(-) diff --git a/target/ppc/cpu_init.c b/target/ppc/cpu_init.c index XXXXXXX..XXXXXXX 100644 --- a/target/ppc/cpu_init.c +++ b/target/ppc/cpu_init.c @@ -XXX,XX +XXX,XX @@ static void ppc_cpu_reset_hold(Object *obj, ResetType type) */ set_float_2nan_prop_rule(float_2nan_prop_ab, &env->fp_status); set_float_2nan_prop_rule(float_2nan_prop_ab, &env->vec_status); + /* + * For PPC, the (inf,zero,qnan) case sets InvalidOp, but we prefer + * to return an input NaN if we have one (ie c) rather than generating + * a default NaN + */ + set_float_infzeronan_rule(float_infzeronan_dnan_never, &env->fp_status); + set_float_infzeronan_rule(float_infzeronan_dnan_never, &env->vec_status); for (i = 0; i < ARRAY_SIZE(env->spr_cb); i++) { ppc_spr_t *spr = &env->spr_cb[i]; diff --git a/fpu/softfloat-specialize.c.inc b/fpu/softfloat-specialize.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-specialize.c.inc +++ b/fpu/softfloat-specialize.c.inc @@ -XXX,XX +XXX,XX @@ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls, */ rule = float_infzeronan_dnan_never; } -#elif defined(TARGET_PPC) || defined(TARGET_SPARC) || \ +#elif defined(TARGET_SPARC) || \ defined(TARGET_XTENSA) || defined(TARGET_HPPA) || \ defined(TARGET_I386) || defined(TARGET_LOONGARCH) /* * For LoongArch systems that conform to IEEE754-2008, the (inf,zero,nan) * case sets InvalidOp and returns the input value 'c' */ - /* - * For PPC, the (inf,zero,qnan) case sets InvalidOp, but we prefer - * to return an input NaN if we have one (ie c) rather than generating - * a default NaN - */ rule = float_infzeronan_dnan_never; #endif } -- 2.34.1 Set the FloatInfZeroNaNRule explicitly for the MIPS target, so we can remove the ifdef from pickNaNMulAdd(). Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-9-peter.maydell@linaro.org --- target/mips/fpu_helper.h | 9 +++++++++ target/mips/msa.c | 4 ++++ fpu/softfloat-specialize.c.inc | 16 +--------------- 3 files changed, 14 insertions(+), 15 deletions(-) diff --git a/target/mips/fpu_helper.h b/target/mips/fpu_helper.h index XXXXXXX..XXXXXXX 100644 --- a/target/mips/fpu_helper.h +++ b/target/mips/fpu_helper.h @@ -XXX,XX +XXX,XX @@ static inline void restore_flush_mode(CPUMIPSState *env) static inline void restore_snan_bit_mode(CPUMIPSState *env) { bool nan2008 = env->active_fpu.fcr31 & (1 << FCR31_NAN2008); + FloatInfZeroNaNRule izn_rule; /* * With nan2008, SNaNs are silenced in the usual way. @@ -XXX,XX +XXX,XX @@ static inline void restore_snan_bit_mode(CPUMIPSState *env) */ set_snan_bit_is_one(!nan2008, &env->active_fpu.fp_status); set_default_nan_mode(!nan2008, &env->active_fpu.fp_status); + /* + * For MIPS systems that conform to IEEE754-1985, the (inf,zero,nan) + * case sets InvalidOp and returns the default NaN. + * For MIPS systems that conform to IEEE754-2008, the (inf,zero,nan) + * case sets InvalidOp and returns the input value 'c'. + */ + izn_rule = nan2008 ? float_infzeronan_dnan_never : float_infzeronan_dnan_always; + set_float_infzeronan_rule(izn_rule, &env->active_fpu.fp_status); } static inline void restore_fp_status(CPUMIPSState *env) diff --git a/target/mips/msa.c b/target/mips/msa.c index XXXXXXX..XXXXXXX 100644 --- a/target/mips/msa.c +++ b/target/mips/msa.c @@ -XXX,XX +XXX,XX @@ void msa_reset(CPUMIPSState *env) /* set proper signanling bit meaning ("1" means "quiet") */ set_snan_bit_is_one(0, &env->active_tc.msa_fp_status); + + /* Inf * 0 + NaN returns the input NaN */ + set_float_infzeronan_rule(float_infzeronan_dnan_never, + &env->active_tc.msa_fp_status); } diff --git a/fpu/softfloat-specialize.c.inc b/fpu/softfloat-specialize.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-specialize.c.inc +++ b/fpu/softfloat-specialize.c.inc @@ -XXX,XX +XXX,XX @@ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls, /* * Temporarily fall back to ifdef ladder */ -#if defined(TARGET_MIPS) - if (snan_bit_is_one(status)) { - /* - * For MIPS systems that conform to IEEE754-1985, the (inf,zero,nan) - * case sets InvalidOp and returns the default NaN - */ - rule = float_infzeronan_dnan_always; - } else { - /* - * For MIPS systems that conform to IEEE754-2008, the (inf,zero,nan) - * case sets InvalidOp and returns the input value 'c' - */ - rule = float_infzeronan_dnan_never; - } -#elif defined(TARGET_SPARC) || \ +#if defined(TARGET_SPARC) || \ defined(TARGET_XTENSA) || defined(TARGET_HPPA) || \ defined(TARGET_I386) || defined(TARGET_LOONGARCH) /* -- 2.34.1 Set the FloatInfZeroNaNRule explicitly for the SPARC target, so we can remove the ifdef from pickNaNMulAdd(). Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-10-peter.maydell@linaro.org --- target/sparc/cpu.c | 2 ++ fpu/softfloat-specialize.c.inc | 3 +-- 2 files changed, 3 insertions(+), 2 deletions(-) diff --git a/target/sparc/cpu.c b/target/sparc/cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/sparc/cpu.c +++ b/target/sparc/cpu.c @@ -XXX,XX +XXX,XX @@ static void sparc_cpu_realizefn(DeviceState *dev, Error **errp) * the CPU state struct so it won't get zeroed on reset. */ set_float_2nan_prop_rule(float_2nan_prop_s_ba, &env->fp_status); + /* For inf * 0 + NaN, return the input NaN */ + set_float_infzeronan_rule(float_infzeronan_dnan_never, &env->fp_status); cpu_exec_realizefn(cs, &local_err); if (local_err != NULL) { diff --git a/fpu/softfloat-specialize.c.inc b/fpu/softfloat-specialize.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-specialize.c.inc +++ b/fpu/softfloat-specialize.c.inc @@ -XXX,XX +XXX,XX @@ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls, /* * Temporarily fall back to ifdef ladder */ -#if defined(TARGET_SPARC) || \ - defined(TARGET_XTENSA) || defined(TARGET_HPPA) || \ +#if defined(TARGET_XTENSA) || defined(TARGET_HPPA) || \ defined(TARGET_I386) || defined(TARGET_LOONGARCH) /* * For LoongArch systems that conform to IEEE754-2008, the (inf,zero,nan) -- 2.34.1 Set the FloatInfZeroNaNRule explicitly for the xtensa target, so we can remove the ifdef from pickNaNMulAdd(). Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-11-peter.maydell@linaro.org --- target/xtensa/cpu.c | 2 ++ fpu/softfloat-specialize.c.inc | 2 +- 2 files changed, 3 insertions(+), 1 deletion(-) diff --git a/target/xtensa/cpu.c b/target/xtensa/cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/xtensa/cpu.c +++ b/target/xtensa/cpu.c @@ -XXX,XX +XXX,XX @@ static void xtensa_cpu_reset_hold(Object *obj, ResetType type) reset_mmu(env); cs->halted = env->runstall; #endif + /* For inf * 0 + NaN, return the input NaN */ + set_float_infzeronan_rule(float_infzeronan_dnan_never, &env->fp_status); set_no_signaling_nans(!dfpu, &env->fp_status); xtensa_use_first_nan(env, !dfpu); } diff --git a/fpu/softfloat-specialize.c.inc b/fpu/softfloat-specialize.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-specialize.c.inc +++ b/fpu/softfloat-specialize.c.inc @@ -XXX,XX +XXX,XX @@ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls, /* * Temporarily fall back to ifdef ladder */ -#if defined(TARGET_XTENSA) || defined(TARGET_HPPA) || \ +#if defined(TARGET_HPPA) || \ defined(TARGET_I386) || defined(TARGET_LOONGARCH) /* * For LoongArch systems that conform to IEEE754-2008, the (inf,zero,nan) -- 2.34.1 Set the FloatInfZeroNaNRule explicitly for the x86 target. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-12-peter.maydell@linaro.org --- target/i386/tcg/fpu_helper.c | 7 +++++++ fpu/softfloat-specialize.c.inc | 2 +- 2 files changed, 8 insertions(+), 1 deletion(-) diff --git a/target/i386/tcg/fpu_helper.c b/target/i386/tcg/fpu_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/i386/tcg/fpu_helper.c +++ b/target/i386/tcg/fpu_helper.c @@ -XXX,XX +XXX,XX @@ void cpu_init_fp_statuses(CPUX86State *env) */ set_float_2nan_prop_rule(float_2nan_prop_x87, &env->mmx_status); set_float_2nan_prop_rule(float_2nan_prop_x87, &env->sse_status); + /* + * Only SSE has multiply-add instructions. In the SDM Section 14.5.2 + * "Fused-Multiply-ADD (FMA) Numeric Behavior" the NaN handling is + * specified -- for 0 * inf + NaN the input NaN is selected, and if + * there are multiple input NaNs they are selected in the order a, b, c. + */ + set_float_infzeronan_rule(float_infzeronan_dnan_never, &env->sse_status); } static inline uint8_t save_exception_flags(CPUX86State *env) diff --git a/fpu/softfloat-specialize.c.inc b/fpu/softfloat-specialize.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-specialize.c.inc +++ b/fpu/softfloat-specialize.c.inc @@ -XXX,XX +XXX,XX @@ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls, * Temporarily fall back to ifdef ladder */ #if defined(TARGET_HPPA) || \ - defined(TARGET_I386) || defined(TARGET_LOONGARCH) + defined(TARGET_LOONGARCH) /* * For LoongArch systems that conform to IEEE754-2008, the (inf,zero,nan) * case sets InvalidOp and returns the input value 'c' -- 2.34.1 Set the FloatInfZeroNaNRule explicitly for the loongarch target. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-13-peter.maydell@linaro.org --- target/loongarch/tcg/fpu_helper.c | 5 +++++ fpu/softfloat-specialize.c.inc | 7 +------ 2 files changed, 6 insertions(+), 6 deletions(-) diff --git a/target/loongarch/tcg/fpu_helper.c b/target/loongarch/tcg/fpu_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/loongarch/tcg/fpu_helper.c +++ b/target/loongarch/tcg/fpu_helper.c @@ -XXX,XX +XXX,XX @@ void restore_fp_status(CPULoongArchState *env) &env->fp_status); set_flush_to_zero(0, &env->fp_status); set_float_2nan_prop_rule(float_2nan_prop_s_ab, &env->fp_status); + /* + * For LoongArch systems that conform to IEEE754-2008, the (inf,zero,nan) + * case sets InvalidOp and returns the input value 'c' + */ + set_float_infzeronan_rule(float_infzeronan_dnan_never, &env->fp_status); } int ieee_ex_to_loongarch(int xcpt) diff --git a/fpu/softfloat-specialize.c.inc b/fpu/softfloat-specialize.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-specialize.c.inc +++ b/fpu/softfloat-specialize.c.inc @@ -XXX,XX +XXX,XX @@ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls, /* * Temporarily fall back to ifdef ladder */ -#if defined(TARGET_HPPA) || \ - defined(TARGET_LOONGARCH) - /* - * For LoongArch systems that conform to IEEE754-2008, the (inf,zero,nan) - * case sets InvalidOp and returns the input value 'c' - */ +#if defined(TARGET_HPPA) rule = float_infzeronan_dnan_never; #endif } -- 2.34.1 Set the FloatInfZeroNaNRule explicitly for the HPPA target, so we can remove the ifdef from pickNaNMulAdd(). As this is the last target to be converted to explicitly setting the rule, we can remove the fallback code in pickNaNMulAdd() entirely. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-14-peter.maydell@linaro.org --- target/hppa/fpu_helper.c | 2 ++ fpu/softfloat-specialize.c.inc | 13 +------------ 2 files changed, 3 insertions(+), 12 deletions(-) diff --git a/target/hppa/fpu_helper.c b/target/hppa/fpu_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/hppa/fpu_helper.c +++ b/target/hppa/fpu_helper.c @@ -XXX,XX +XXX,XX @@ void HELPER(loaded_fr0)(CPUHPPAState *env) * HPPA does note implement a CPU reset method at all... */ set_float_2nan_prop_rule(float_2nan_prop_s_ab, &env->fp_status); + /* For inf * 0 + NaN, return the input NaN */ + set_float_infzeronan_rule(float_infzeronan_dnan_never, &env->fp_status); } void cpu_hppa_loaded_fr0(CPUHPPAState *env) diff --git a/fpu/softfloat-specialize.c.inc b/fpu/softfloat-specialize.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-specialize.c.inc +++ b/fpu/softfloat-specialize.c.inc @@ -XXX,XX +XXX,XX @@ static int pickNaN(FloatClass a_cls, FloatClass b_cls, static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls, bool infzero, float_status *status) { - FloatInfZeroNaNRule rule = status->float_infzeronan_rule; - /* * We guarantee not to require the target to tell us how to * pick a NaN if we're always returning the default NaN. @@ -XXX,XX +XXX,XX @@ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls, */ assert(!status->default_nan_mode); - if (rule == float_infzeronan_none) { - /* - * Temporarily fall back to ifdef ladder - */ -#if defined(TARGET_HPPA) - rule = float_infzeronan_dnan_never; -#endif - } - if (infzero) { /* * Inf * 0 + NaN -- some implementations return the default NaN here, * and some return the input NaN. */ - switch (rule) { + switch (status->float_infzeronan_rule) { case float_infzeronan_dnan_never: return 2; case float_infzeronan_dnan_always: -- 2.34.1 The new implementation of pickNaNMulAdd() will find it convenient to know whether at least one of the three arguments to the muladd was a signaling NaN. We already calculate that in the caller, so pass it in as a new bool have_snan. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-15-peter.maydell@linaro.org --- fpu/softfloat-parts.c.inc | 5 +++-- fpu/softfloat-specialize.c.inc | 2 +- 2 files changed, 4 insertions(+), 3 deletions(-) diff --git a/fpu/softfloat-parts.c.inc b/fpu/softfloat-parts.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-parts.c.inc +++ b/fpu/softfloat-parts.c.inc @@ -XXX,XX +XXX,XX @@ static FloatPartsN *partsN(pick_nan_muladd)(FloatPartsN *a, FloatPartsN *b, { int which; bool infzero = (ab_mask == float_cmask_infzero); + bool have_snan = (abc_mask & float_cmask_snan); - if (unlikely(abc_mask & float_cmask_snan)) { + if (unlikely(have_snan)) { float_raise(float_flag_invalid | float_flag_invalid_snan, s); } @@ -XXX,XX +XXX,XX @@ static FloatPartsN *partsN(pick_nan_muladd)(FloatPartsN *a, FloatPartsN *b, if (s->default_nan_mode) { which = 3; } else { - which = pickNaNMulAdd(a->cls, b->cls, c->cls, infzero, s); + which = pickNaNMulAdd(a->cls, b->cls, c->cls, infzero, have_snan, s); } if (which == 3) { diff --git a/fpu/softfloat-specialize.c.inc b/fpu/softfloat-specialize.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-specialize.c.inc +++ b/fpu/softfloat-specialize.c.inc @@ -XXX,XX +XXX,XX @@ static int pickNaN(FloatClass a_cls, FloatClass b_cls, | Return values : 0 : a; 1 : b; 2 : c; 3 : default-NaN *----------------------------------------------------------------------------*/ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls, - bool infzero, float_status *status) + bool infzero, bool have_snan, float_status *status) { /* * We guarantee not to require the target to tell us how to -- 2.34.1 IEEE 758 does not define a fixed rule for which NaN to pick as the result if both operands of a 3-operand fused multiply-add operation are NaNs. As a result different architectures have ended up with different rules for propagating NaNs. QEMU currently hardcodes the NaN propagation logic into the binary because pickNaNMulAdd() has an ifdef ladder for different targets. We want to make the propagation rule instead be selectable at runtime, because: * this will let us have multiple targets in one QEMU binary * the Arm FEAT_AFP architectural feature includes letting the guest select a NaN propagation rule at runtime In this commit we add an enum for the propagation rule, the field in float_status, and the corresponding getters and setters. We change pickNaNMulAdd to honour this, but because all targets still leave this field at its default 0 value, the fallback logic will pick the rule type with the old ifdef ladder. It's valid not to set a propagation rule if default_nan_mode is enabled, because in that case there's no need to pick a NaN; all the callers of pickNaNMulAdd() catch this case and skip calling it. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-16-peter.maydell@linaro.org --- include/fpu/softfloat-helpers.h | 11 +++ include/fpu/softfloat-types.h | 55 +++++++++++ fpu/softfloat-specialize.c.inc | 167 ++++++++------------------------ 3 files changed, 107 insertions(+), 126 deletions(-) diff --git a/include/fpu/softfloat-helpers.h b/include/fpu/softfloat-helpers.h index XXXXXXX..XXXXXXX 100644 --- a/include/fpu/softfloat-helpers.h +++ b/include/fpu/softfloat-helpers.h @@ -XXX,XX +XXX,XX @@ static inline void set_float_2nan_prop_rule(Float2NaNPropRule rule, status->float_2nan_prop_rule = rule; } +static inline void set_float_3nan_prop_rule(Float3NaNPropRule rule, + float_status *status) +{ + status->float_3nan_prop_rule = rule; +} + static inline void set_float_infzeronan_rule(FloatInfZeroNaNRule rule, float_status *status) { @@ -XXX,XX +XXX,XX @@ static inline Float2NaNPropRule get_float_2nan_prop_rule(float_status *status) return status->float_2nan_prop_rule; } +static inline Float3NaNPropRule get_float_3nan_prop_rule(float_status *status) +{ + return status->float_3nan_prop_rule; +} + static inline FloatInfZeroNaNRule get_float_infzeronan_rule(float_status *status) { return status->float_infzeronan_rule; diff --git a/include/fpu/softfloat-types.h b/include/fpu/softfloat-types.h index XXXXXXX..XXXXXXX 100644 --- a/include/fpu/softfloat-types.h +++ b/include/fpu/softfloat-types.h @@ -XXX,XX +XXX,XX @@ this code that are retained. #ifndef SOFTFLOAT_TYPES_H #define SOFTFLOAT_TYPES_H +#include "hw/registerfields.h" + /* * Software IEC/IEEE floating-point types. */ @@ -XXX,XX +XXX,XX @@ typedef enum __attribute__((__packed__)) { float_2nan_prop_x87, } Float2NaNPropRule; +/* + * 3-input NaN propagation rule, for fused multiply-add. Individual + * architectures have different rules for which input NaN is + * propagated to the output when there is more than one NaN on the + * input. + * + * If default_nan_mode is enabled then it is valid not to set a NaN + * propagation rule, because the softfloat code guarantees not to try + * to pick a NaN to propagate in default NaN mode. When not in + * default-NaN mode, it is an error for the target not to set the rule + * in float_status if it uses a muladd, and we will assert if we need + * to handle an input NaN and no rule was selected. + * + * The naming scheme for Float3NaNPropRule values is: + * float_3nan_prop_s_abc: + * = "Prefer SNaN over QNaN, then operand A over B over C" + * float_3nan_prop_abc: + * = "Prefer A over B over C regardless of SNaN vs QNAN" + * + * For QEMU, the multiply-add operation is A * B + C. + */ + +/* + * We set the Float3NaNPropRule enum values up so we can select the + * right value in pickNaNMulAdd in a data driven way. + */ +FIELD(3NAN, 1ST, 0, 2) /* which operand is most preferred ? */ +FIELD(3NAN, 2ND, 2, 2) /* which operand is next most preferred ? */ +FIELD(3NAN, 3RD, 4, 2) /* which operand is least preferred ? */ +FIELD(3NAN, SNAN, 6, 1) /* do we prefer SNaN over QNaN ? */ + +#define PROPRULE(X, Y, Z) \ + ((X << R_3NAN_1ST_SHIFT) | (Y << R_3NAN_2ND_SHIFT) | (Z << R_3NAN_3RD_SHIFT)) + +typedef enum __attribute__((__packed__)) { + float_3nan_prop_none = 0, /* No propagation rule specified */ + float_3nan_prop_abc = PROPRULE(0, 1, 2), + float_3nan_prop_acb = PROPRULE(0, 2, 1), + float_3nan_prop_bac = PROPRULE(1, 0, 2), + float_3nan_prop_bca = PROPRULE(1, 2, 0), + float_3nan_prop_cab = PROPRULE(2, 0, 1), + float_3nan_prop_cba = PROPRULE(2, 1, 0), + float_3nan_prop_s_abc = float_3nan_prop_abc | R_3NAN_SNAN_MASK, + float_3nan_prop_s_acb = float_3nan_prop_acb | R_3NAN_SNAN_MASK, + float_3nan_prop_s_bac = float_3nan_prop_bac | R_3NAN_SNAN_MASK, + float_3nan_prop_s_bca = float_3nan_prop_bca | R_3NAN_SNAN_MASK, + float_3nan_prop_s_cab = float_3nan_prop_cab | R_3NAN_SNAN_MASK, + float_3nan_prop_s_cba = float_3nan_prop_cba | R_3NAN_SNAN_MASK, +} Float3NaNPropRule; + +#undef PROPRULE + /* * Rule for result of fused multiply-add 0 * Inf + NaN. * This must be a NaN, but implementations differ on whether this @@ -XXX,XX +XXX,XX @@ typedef struct float_status { FloatRoundMode float_rounding_mode; FloatX80RoundPrec floatx80_rounding_precision; Float2NaNPropRule float_2nan_prop_rule; + Float3NaNPropRule float_3nan_prop_rule; FloatInfZeroNaNRule float_infzeronan_rule; bool tininess_before_rounding; /* should denormalised results go to zero and set the inexact flag? */ diff --git a/fpu/softfloat-specialize.c.inc b/fpu/softfloat-specialize.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-specialize.c.inc +++ b/fpu/softfloat-specialize.c.inc @@ -XXX,XX +XXX,XX @@ static int pickNaN(FloatClass a_cls, FloatClass b_cls, static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls, bool infzero, bool have_snan, float_status *status) { + FloatClass cls[3] = { a_cls, b_cls, c_cls }; + Float3NaNPropRule rule = status->float_3nan_prop_rule; + int which; + /* * We guarantee not to require the target to tell us how to * pick a NaN if we're always returning the default NaN. @@ -XXX,XX +XXX,XX @@ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls, } } + if (rule == float_3nan_prop_none) { #if defined(TARGET_ARM) - - /* This looks different from the ARM ARM pseudocode, because the ARM ARM - * puts the operands to a fused mac operation (a*b)+c in the order c,a,b. - */ - if (is_snan(c_cls)) { - return 2; - } else if (is_snan(a_cls)) { - return 0; - } else if (is_snan(b_cls)) { - return 1; - } else if (is_qnan(c_cls)) { - return 2; - } else if (is_qnan(a_cls)) { - return 0; - } else { - return 1; - } + /* + * This looks different from the ARM ARM pseudocode, because the ARM ARM + * puts the operands to a fused mac operation (a*b)+c in the order c,a,b + */ + rule = float_3nan_prop_s_cab; #elif defined(TARGET_MIPS) - if (snan_bit_is_one(status)) { - /* Prefer sNaN over qNaN, in the a, b, c order. */ - if (is_snan(a_cls)) { - return 0; - } else if (is_snan(b_cls)) { - return 1; - } else if (is_snan(c_cls)) { - return 2; - } else if (is_qnan(a_cls)) { - return 0; - } else if (is_qnan(b_cls)) { - return 1; + if (snan_bit_is_one(status)) { + rule = float_3nan_prop_s_abc; } else { - return 2; + rule = float_3nan_prop_s_cab; } - } else { - /* Prefer sNaN over qNaN, in the c, a, b order. */ - if (is_snan(c_cls)) { - return 2; - } else if (is_snan(a_cls)) { - return 0; - } else if (is_snan(b_cls)) { - return 1; - } else if (is_qnan(c_cls)) { - return 2; - } else if (is_qnan(a_cls)) { - return 0; - } else { - return 1; - } - } #elif defined(TARGET_LOONGARCH64) - /* Prefer sNaN over qNaN, in the c, a, b order. */ - if (is_snan(c_cls)) { - return 2; - } else if (is_snan(a_cls)) { - return 0; - } else if (is_snan(b_cls)) { - return 1; - } else if (is_qnan(c_cls)) { - return 2; - } else if (is_qnan(a_cls)) { - return 0; - } else { - return 1; - } + rule = float_3nan_prop_s_cab; #elif defined(TARGET_PPC) - /* If fRA is a NaN return it; otherwise if fRB is a NaN return it; - * otherwise return fRC. Note that muladd on PPC is (fRA * fRC) + frB - */ - if (is_nan(a_cls)) { - return 0; - } else if (is_nan(c_cls)) { - return 2; - } else { - return 1; - } + /* + * If fRA is a NaN return it; otherwise if fRB is a NaN return it; + * otherwise return fRC. Note that muladd on PPC is (fRA * fRC) + frB + */ + rule = float_3nan_prop_acb; #elif defined(TARGET_S390X) - if (is_snan(a_cls)) { - return 0; - } else if (is_snan(b_cls)) { - return 1; - } else if (is_snan(c_cls)) { - return 2; - } else if (is_qnan(a_cls)) { - return 0; - } else if (is_qnan(b_cls)) { - return 1; - } else { - return 2; - } + rule = float_3nan_prop_s_abc; #elif defined(TARGET_SPARC) - /* Prefer SNaN over QNaN, order C, B, A. */ - if (is_snan(c_cls)) { - return 2; - } else if (is_snan(b_cls)) { - return 1; - } else if (is_snan(a_cls)) { - return 0; - } else if (is_qnan(c_cls)) { - return 2; - } else if (is_qnan(b_cls)) { - return 1; - } else { - return 0; - } + rule = float_3nan_prop_s_cba; #elif defined(TARGET_XTENSA) - /* - * For Xtensa, the (inf,zero,nan) case sets InvalidOp and returns - * an input NaN if we have one (ie c). - */ - if (status->use_first_nan) { - if (is_nan(a_cls)) { - return 0; - } else if (is_nan(b_cls)) { - return 1; + if (status->use_first_nan) { + rule = float_3nan_prop_abc; } else { - return 2; + rule = float_3nan_prop_cba; } - } else { - if (is_nan(c_cls)) { - return 2; - } else if (is_nan(b_cls)) { - return 1; - } else { - return 0; - } - } #else - /* A default implementation: prefer a to b to c. - * This is unlikely to actually match any real implementation. - */ - if (is_nan(a_cls)) { - return 0; - } else if (is_nan(b_cls)) { - return 1; - } else { - return 2; - } + rule = float_3nan_prop_abc; #endif + } + + assert(rule != float_3nan_prop_none); + if (have_snan && (rule & R_3NAN_SNAN_MASK)) { + /* We have at least one SNaN input and should prefer it */ + do { + which = rule & R_3NAN_1ST_MASK; + rule >>= R_3NAN_1ST_LENGTH; + } while (!is_snan(cls[which])); + } else { + do { + which = rule & R_3NAN_1ST_MASK; + rule >>= R_3NAN_1ST_LENGTH; + } while (!is_nan(cls[which])); + } + return which; } /*---------------------------------------------------------------------------- -- 2.34.1 Explicitly set a rule in the softfloat tests for propagating NaNs in the muladd case. In meson.build we put -DTARGET_ARM in fpcflags, and so we should select here the Arm rule of float_3nan_prop_s_cab. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-17-peter.maydell@linaro.org --- tests/fp/fp-bench.c | 1 + tests/fp/fp-test.c | 1 + 2 files changed, 2 insertions(+) diff --git a/tests/fp/fp-bench.c b/tests/fp/fp-bench.c index XXXXXXX..XXXXXXX 100644 --- a/tests/fp/fp-bench.c +++ b/tests/fp/fp-bench.c @@ -XXX,XX +XXX,XX @@ static void run_bench(void) * doesn't specify match those used by the Arm architecture. */ set_float_2nan_prop_rule(float_2nan_prop_s_ab, &soft_status); + set_float_3nan_prop_rule(float_3nan_prop_s_cab, &soft_status); set_float_infzeronan_rule(float_infzeronan_dnan_if_qnan, &soft_status); f = bench_funcs[operation][precision]; diff --git a/tests/fp/fp-test.c b/tests/fp/fp-test.c index XXXXXXX..XXXXXXX 100644 --- a/tests/fp/fp-test.c +++ b/tests/fp/fp-test.c @@ -XXX,XX +XXX,XX @@ void run_test(void) * doesn't specify match those used by the Arm architecture. */ set_float_2nan_prop_rule(float_2nan_prop_s_ab, &qsf); + set_float_3nan_prop_rule(float_3nan_prop_s_cab, &qsf); set_float_infzeronan_rule(float_infzeronan_dnan_if_qnan, &qsf); genCases_setLevel(test_level); -- 2.34.1 Set the Float3NaNPropRule explicitly for Arm, and remove the ifdef from pickNaNMulAdd(). Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-18-peter.maydell@linaro.org --- target/arm/cpu.c | 5 +++++ fpu/softfloat-specialize.c.inc | 8 +------- 2 files changed, 6 insertions(+), 7 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_register_el_change_hook(ARMCPU *cpu, ARMELChangeHookFn *hook, * * tininess-before-rounding * * 2-input NaN propagation prefers SNaN over QNaN, and then * operand A over operand B (see FPProcessNaNs() pseudocode) + * * 3-input NaN propagation prefers SNaN over QNaN, and then + * operand C over A over B (see FPProcessNaNs3() pseudocode, + * but note that for QEMU muladd is a * b + c, whereas for + * the pseudocode function the arguments are in the order c, a, b. * * 0 * Inf + NaN returns the default NaN if the input NaN is quiet, * and the input NaN if it is signalling */ @@ -XXX,XX +XXX,XX @@ static void arm_set_default_fp_behaviours(float_status *s) { set_float_detect_tininess(float_tininess_before_rounding, s); set_float_2nan_prop_rule(float_2nan_prop_s_ab, s); + set_float_3nan_prop_rule(float_3nan_prop_s_cab, s); set_float_infzeronan_rule(float_infzeronan_dnan_if_qnan, s); } diff --git a/fpu/softfloat-specialize.c.inc b/fpu/softfloat-specialize.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-specialize.c.inc +++ b/fpu/softfloat-specialize.c.inc @@ -XXX,XX +XXX,XX @@ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls, } if (rule == float_3nan_prop_none) { -#if defined(TARGET_ARM) - /* - * This looks different from the ARM ARM pseudocode, because the ARM ARM - * puts the operands to a fused mac operation (a*b)+c in the order c,a,b - */ - rule = float_3nan_prop_s_cab; -#elif defined(TARGET_MIPS) +#if defined(TARGET_MIPS) if (snan_bit_is_one(status)) { rule = float_3nan_prop_s_abc; } else { -- 2.34.1 Set the Float3NaNPropRule explicitly for loongarch, and remove the ifdef from pickNaNMulAdd(). Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-19-peter.maydell@linaro.org --- target/loongarch/tcg/fpu_helper.c | 1 + fpu/softfloat-specialize.c.inc | 2 -- 2 files changed, 1 insertion(+), 2 deletions(-) diff --git a/target/loongarch/tcg/fpu_helper.c b/target/loongarch/tcg/fpu_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/loongarch/tcg/fpu_helper.c +++ b/target/loongarch/tcg/fpu_helper.c @@ -XXX,XX +XXX,XX @@ void restore_fp_status(CPULoongArchState *env) * case sets InvalidOp and returns the input value 'c' */ set_float_infzeronan_rule(float_infzeronan_dnan_never, &env->fp_status); + set_float_3nan_prop_rule(float_3nan_prop_s_cab, &env->fp_status); } int ieee_ex_to_loongarch(int xcpt) diff --git a/fpu/softfloat-specialize.c.inc b/fpu/softfloat-specialize.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-specialize.c.inc +++ b/fpu/softfloat-specialize.c.inc @@ -XXX,XX +XXX,XX @@ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls, } else { rule = float_3nan_prop_s_cab; } -#elif defined(TARGET_LOONGARCH64) - rule = float_3nan_prop_s_cab; #elif defined(TARGET_PPC) /* * If fRA is a NaN return it; otherwise if fRB is a NaN return it; -- 2.34.1 Set the Float3NaNPropRule explicitly for PPC, and remove the ifdef from pickNaNMulAdd(). Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-20-peter.maydell@linaro.org --- target/ppc/cpu_init.c | 8 ++++++++ fpu/softfloat-specialize.c.inc | 6 ------ 2 files changed, 8 insertions(+), 6 deletions(-) diff --git a/target/ppc/cpu_init.c b/target/ppc/cpu_init.c index XXXXXXX..XXXXXXX 100644 --- a/target/ppc/cpu_init.c +++ b/target/ppc/cpu_init.c @@ -XXX,XX +XXX,XX @@ static void ppc_cpu_reset_hold(Object *obj, ResetType type) */ set_float_2nan_prop_rule(float_2nan_prop_ab, &env->fp_status); set_float_2nan_prop_rule(float_2nan_prop_ab, &env->vec_status); + /* + * NaN propagation for fused multiply-add: + * if fRA is a NaN return it; otherwise if fRB is a NaN return it; + * otherwise return fRC. Note that muladd on PPC is (fRA * fRC) + frB + * whereas QEMU labels the operands as (a * b) + c. + */ + set_float_3nan_prop_rule(float_3nan_prop_acb, &env->fp_status); + set_float_3nan_prop_rule(float_3nan_prop_acb, &env->vec_status); /* * For PPC, the (inf,zero,qnan) case sets InvalidOp, but we prefer * to return an input NaN if we have one (ie c) rather than generating diff --git a/fpu/softfloat-specialize.c.inc b/fpu/softfloat-specialize.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-specialize.c.inc +++ b/fpu/softfloat-specialize.c.inc @@ -XXX,XX +XXX,XX @@ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls, } else { rule = float_3nan_prop_s_cab; } -#elif defined(TARGET_PPC) - /* - * If fRA is a NaN return it; otherwise if fRB is a NaN return it; - * otherwise return fRC. Note that muladd on PPC is (fRA * fRC) + frB - */ - rule = float_3nan_prop_acb; #elif defined(TARGET_S390X) rule = float_3nan_prop_s_abc; #elif defined(TARGET_SPARC) -- 2.34.1 Set the Float3NaNPropRule explicitly for s390x, and remove the ifdef from pickNaNMulAdd(). Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-21-peter.maydell@linaro.org --- target/s390x/cpu.c | 1 + fpu/softfloat-specialize.c.inc | 2 -- 2 files changed, 1 insertion(+), 2 deletions(-) diff --git a/target/s390x/cpu.c b/target/s390x/cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/s390x/cpu.c +++ b/target/s390x/cpu.c @@ -XXX,XX +XXX,XX @@ static void s390_cpu_reset_hold(Object *obj, ResetType type) set_float_detect_tininess(float_tininess_before_rounding, &env->fpu_status); set_float_2nan_prop_rule(float_2nan_prop_s_ab, &env->fpu_status); + set_float_3nan_prop_rule(float_3nan_prop_s_abc, &env->fpu_status); set_float_infzeronan_rule(float_infzeronan_dnan_always, &env->fpu_status); /* fall through */ diff --git a/fpu/softfloat-specialize.c.inc b/fpu/softfloat-specialize.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-specialize.c.inc +++ b/fpu/softfloat-specialize.c.inc @@ -XXX,XX +XXX,XX @@ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls, } else { rule = float_3nan_prop_s_cab; } -#elif defined(TARGET_S390X) - rule = float_3nan_prop_s_abc; #elif defined(TARGET_SPARC) rule = float_3nan_prop_s_cba; #elif defined(TARGET_XTENSA) -- 2.34.1 Set the Float3NaNPropRule explicitly for SPARC, and remove the ifdef from pickNaNMulAdd(). Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-22-peter.maydell@linaro.org --- target/sparc/cpu.c | 2 ++ fpu/softfloat-specialize.c.inc | 2 -- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/target/sparc/cpu.c b/target/sparc/cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/sparc/cpu.c +++ b/target/sparc/cpu.c @@ -XXX,XX +XXX,XX @@ static void sparc_cpu_realizefn(DeviceState *dev, Error **errp) * the CPU state struct so it won't get zeroed on reset. */ set_float_2nan_prop_rule(float_2nan_prop_s_ba, &env->fp_status); + /* For fused-multiply add, prefer SNaN over QNaN, then C->B->A */ + set_float_3nan_prop_rule(float_3nan_prop_s_cba, &env->fp_status); /* For inf * 0 + NaN, return the input NaN */ set_float_infzeronan_rule(float_infzeronan_dnan_never, &env->fp_status); diff --git a/fpu/softfloat-specialize.c.inc b/fpu/softfloat-specialize.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-specialize.c.inc +++ b/fpu/softfloat-specialize.c.inc @@ -XXX,XX +XXX,XX @@ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls, } else { rule = float_3nan_prop_s_cab; } -#elif defined(TARGET_SPARC) - rule = float_3nan_prop_s_cba; #elif defined(TARGET_XTENSA) if (status->use_first_nan) { rule = float_3nan_prop_abc; -- 2.34.1 Set the Float3NaNPropRule explicitly for Arm, and remove the ifdef from pickNaNMulAdd(). Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-23-peter.maydell@linaro.org --- target/mips/fpu_helper.h | 4 ++++ target/mips/msa.c | 3 +++ fpu/softfloat-specialize.c.inc | 8 +------- 3 files changed, 8 insertions(+), 7 deletions(-) diff --git a/target/mips/fpu_helper.h b/target/mips/fpu_helper.h index XXXXXXX..XXXXXXX 100644 --- a/target/mips/fpu_helper.h +++ b/target/mips/fpu_helper.h @@ -XXX,XX +XXX,XX @@ static inline void restore_snan_bit_mode(CPUMIPSState *env) { bool nan2008 = env->active_fpu.fcr31 & (1 << FCR31_NAN2008); FloatInfZeroNaNRule izn_rule; + Float3NaNPropRule nan3_rule; /* * With nan2008, SNaNs are silenced in the usual way. @@ -XXX,XX +XXX,XX @@ static inline void restore_snan_bit_mode(CPUMIPSState *env) */ izn_rule = nan2008 ? float_infzeronan_dnan_never : float_infzeronan_dnan_always; set_float_infzeronan_rule(izn_rule, &env->active_fpu.fp_status); + nan3_rule = nan2008 ? float_3nan_prop_s_cab : float_3nan_prop_s_abc; + set_float_3nan_prop_rule(nan3_rule, &env->active_fpu.fp_status); + } static inline void restore_fp_status(CPUMIPSState *env) diff --git a/target/mips/msa.c b/target/mips/msa.c index XXXXXXX..XXXXXXX 100644 --- a/target/mips/msa.c +++ b/target/mips/msa.c @@ -XXX,XX +XXX,XX @@ void msa_reset(CPUMIPSState *env) set_float_2nan_prop_rule(float_2nan_prop_s_ab, &env->active_tc.msa_fp_status); + set_float_3nan_prop_rule(float_3nan_prop_s_cab, + &env->active_tc.msa_fp_status); + /* clear float_status exception flags */ set_float_exception_flags(0, &env->active_tc.msa_fp_status); diff --git a/fpu/softfloat-specialize.c.inc b/fpu/softfloat-specialize.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-specialize.c.inc +++ b/fpu/softfloat-specialize.c.inc @@ -XXX,XX +XXX,XX @@ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls, } if (rule == float_3nan_prop_none) { -#if defined(TARGET_MIPS) - if (snan_bit_is_one(status)) { - rule = float_3nan_prop_s_abc; - } else { - rule = float_3nan_prop_s_cab; - } -#elif defined(TARGET_XTENSA) +#if defined(TARGET_XTENSA) if (status->use_first_nan) { rule = float_3nan_prop_abc; } else { -- 2.34.1 Set the Float3NaNPropRule explicitly for xtensa, and remove the ifdef from pickNaNMulAdd(). Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-24-peter.maydell@linaro.org --- target/xtensa/fpu_helper.c | 2 ++ fpu/softfloat-specialize.c.inc | 8 -------- 2 files changed, 2 insertions(+), 8 deletions(-) diff --git a/target/xtensa/fpu_helper.c b/target/xtensa/fpu_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/xtensa/fpu_helper.c +++ b/target/xtensa/fpu_helper.c @@ -XXX,XX +XXX,XX @@ void xtensa_use_first_nan(CPUXtensaState *env, bool use_first) set_use_first_nan(use_first, &env->fp_status); set_float_2nan_prop_rule(use_first ? float_2nan_prop_ab : float_2nan_prop_ba, &env->fp_status); + set_float_3nan_prop_rule(use_first ? float_3nan_prop_abc : float_3nan_prop_cba, + &env->fp_status); } void HELPER(wur_fpu2k_fcr)(CPUXtensaState *env, uint32_t v) diff --git a/fpu/softfloat-specialize.c.inc b/fpu/softfloat-specialize.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-specialize.c.inc +++ b/fpu/softfloat-specialize.c.inc @@ -XXX,XX +XXX,XX @@ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls, } if (rule == float_3nan_prop_none) { -#if defined(TARGET_XTENSA) - if (status->use_first_nan) { - rule = float_3nan_prop_abc; - } else { - rule = float_3nan_prop_cba; - } -#else rule = float_3nan_prop_abc; -#endif } assert(rule != float_3nan_prop_none); -- 2.34.1 Set the Float3NaNPropRule explicitly for i386. We had no i386-specific behaviour in the old ifdef ladder, so we were using the default "prefer a then b then c" fallback; this is actually the correct per-the-spec handling for i386. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-25-peter.maydell@linaro.org --- target/i386/tcg/fpu_helper.c | 1 + 1 file changed, 1 insertion(+) diff --git a/target/i386/tcg/fpu_helper.c b/target/i386/tcg/fpu_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/i386/tcg/fpu_helper.c +++ b/target/i386/tcg/fpu_helper.c @@ -XXX,XX +XXX,XX @@ void cpu_init_fp_statuses(CPUX86State *env) * there are multiple input NaNs they are selected in the order a, b, c. */ set_float_infzeronan_rule(float_infzeronan_dnan_never, &env->sse_status); + set_float_3nan_prop_rule(float_3nan_prop_abc, &env->sse_status); } static inline uint8_t save_exception_flags(CPUX86State *env) -- 2.34.1 Set the Float3NaNPropRule explicitly for HPPA, and remove the ifdef from pickNaNMulAdd(). HPPA is the only target that was using the default branch of the ifdef ladder (other targets either do not use muladd or set default_nan_mode), so we can remove the ifdef fallback entirely now (allowing the "rule not set" case to fall into the default of the switch statement and assert). We add a TODO note that the HPPA rule is probably wrong; this is not a behavioural change for this refactoring. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-26-peter.maydell@linaro.org --- target/hppa/fpu_helper.c | 8 ++++++++ fpu/softfloat-specialize.c.inc | 4 ---- 2 files changed, 8 insertions(+), 4 deletions(-) diff --git a/target/hppa/fpu_helper.c b/target/hppa/fpu_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/hppa/fpu_helper.c +++ b/target/hppa/fpu_helper.c @@ -XXX,XX +XXX,XX @@ void HELPER(loaded_fr0)(CPUHPPAState *env) * HPPA does note implement a CPU reset method at all... */ set_float_2nan_prop_rule(float_2nan_prop_s_ab, &env->fp_status); + /* + * TODO: The HPPA architecture reference only documents its NaN + * propagation rule for 2-operand operations. Testing on real hardware + * might be necessary to confirm whether this order for muladd is correct. + * Not preferring the SNaN is almost certainly incorrect as it diverges + * from the documented rules for 2-operand operations. + */ + set_float_3nan_prop_rule(float_3nan_prop_abc, &env->fp_status); /* For inf * 0 + NaN, return the input NaN */ set_float_infzeronan_rule(float_infzeronan_dnan_never, &env->fp_status); } diff --git a/fpu/softfloat-specialize.c.inc b/fpu/softfloat-specialize.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-specialize.c.inc +++ b/fpu/softfloat-specialize.c.inc @@ -XXX,XX +XXX,XX @@ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls, } } - if (rule == float_3nan_prop_none) { - rule = float_3nan_prop_abc; - } - assert(rule != float_3nan_prop_none); if (have_snan && (rule & R_3NAN_SNAN_MASK)) { /* We have at least one SNaN input and should prefer it */ -- 2.34.1 The use_first_nan field in float_status was an xtensa-specific way to select at runtime from two different NaN propagation rules. Now that xtensa is using the target-agnostic NaN propagation rule selection that we've just added, we can remove use_first_nan, because there is no longer any code that reads it. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-27-peter.maydell@linaro.org --- include/fpu/softfloat-helpers.h | 5 ----- include/fpu/softfloat-types.h | 1 - target/xtensa/fpu_helper.c | 1 - 3 files changed, 7 deletions(-) diff --git a/include/fpu/softfloat-helpers.h b/include/fpu/softfloat-helpers.h index XXXXXXX..XXXXXXX 100644 --- a/include/fpu/softfloat-helpers.h +++ b/include/fpu/softfloat-helpers.h @@ -XXX,XX +XXX,XX @@ static inline void set_snan_bit_is_one(bool val, float_status *status) status->snan_bit_is_one = val; } -static inline void set_use_first_nan(bool val, float_status *status) -{ - status->use_first_nan = val; -} - static inline void set_no_signaling_nans(bool val, float_status *status) { status->no_signaling_nans = val; diff --git a/include/fpu/softfloat-types.h b/include/fpu/softfloat-types.h index XXXXXXX..XXXXXXX 100644 --- a/include/fpu/softfloat-types.h +++ b/include/fpu/softfloat-types.h @@ -XXX,XX +XXX,XX @@ typedef struct float_status { * softfloat-specialize.inc.c) */ bool snan_bit_is_one; - bool use_first_nan; bool no_signaling_nans; /* should overflowed results subtract re_bias to its exponent? */ bool rebias_overflow; diff --git a/target/xtensa/fpu_helper.c b/target/xtensa/fpu_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/xtensa/fpu_helper.c +++ b/target/xtensa/fpu_helper.c @@ -XXX,XX +XXX,XX @@ static const struct { void xtensa_use_first_nan(CPUXtensaState *env, bool use_first) { - set_use_first_nan(use_first, &env->fp_status); set_float_2nan_prop_rule(use_first ? float_2nan_prop_ab : float_2nan_prop_ba, &env->fp_status); set_float_3nan_prop_rule(use_first ? float_3nan_prop_abc : float_3nan_prop_cba, -- 2.34.1 Currently m68k_cpu_reset_hold() calls floatx80_default_nan(NULL) to get the NaN bit pattern to reset the FPU registers. This works because it happens that our implementation of floatx80_default_nan() doesn't actually look at the float_status pointer except for TARGET_MIPS. However, this isn't guaranteed, and to be able to remove the ifdef in floatx80_default_nan() we're going to need a real float_status here. Rearrange m68k_cpu_reset_hold() so that we initialize env->fp_status earlier, and thus can pass it to floatx80_default_nan(). Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-28-peter.maydell@linaro.org --- target/m68k/cpu.c | 12 +++++++----- 1 file changed, 7 insertions(+), 5 deletions(-) diff --git a/target/m68k/cpu.c b/target/m68k/cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/m68k/cpu.c +++ b/target/m68k/cpu.c @@ -XXX,XX +XXX,XX @@ static void m68k_cpu_reset_hold(Object *obj, ResetType type) CPUState *cs = CPU(obj); M68kCPUClass *mcc = M68K_CPU_GET_CLASS(obj); CPUM68KState *env = cpu_env(cs); - floatx80 nan = floatx80_default_nan(NULL); + floatx80 nan; int i; if (mcc->parent_phases.hold) { @@ -XXX,XX +XXX,XX @@ static void m68k_cpu_reset_hold(Object *obj, ResetType type) #else cpu_m68k_set_sr(env, SR_S | SR_I); #endif - for (i = 0; i < 8; i++) { - env->fregs[i].d = nan; - } - cpu_m68k_set_fpcr(env, 0); /* * M68000 FAMILY PROGRAMMER'S REFERENCE MANUAL * 3.4 FLOATING-POINT INSTRUCTION DETAILS @@ -XXX,XX +XXX,XX @@ static void m68k_cpu_reset_hold(Object *obj, ResetType type) * preceding paragraph for nonsignaling NaNs. */ set_float_2nan_prop_rule(float_2nan_prop_ab, &env->fp_status); + + nan = floatx80_default_nan(&env->fp_status); + for (i = 0; i < 8; i++) { + env->fregs[i].d = nan; + } + cpu_m68k_set_fpcr(env, 0); env->fpsr = 0; /* TODO: We should set PC from the interrupt vector. */ -- 2.34.1 We create our 128-bit default NaN by calling parts64_default_nan() and then adjusting the result. We can do the same trick for creating the floatx80 default NaN, which lets us drop a target ifdef. floatx80 is used only by: i386 m68k arm nwfpe old floating-point emulation emulation support (which is essentially dead, especially the parts involving floatx80) PPC (only in the xsrqpxp instruction, which just rounds an input value by converting to floatx80 and back, so will never generate the default NaN) The floatx80 default NaN as currently implemented is: m68k: sign = 0, exp = 1...1, int = 1, frac = 1....1 i386: sign = 1, exp = 1...1, int = 1, frac = 10...0 These are the same as the parts64_default_nan for these architectures. This is technically a possible behaviour change for arm linux-user nwfpe emulation emulation, because the default NaN will now have the sign bit clear. But we were already generating a different floatx80 default NaN from the real kernel emulation we are supposedly following, which appears to use an all-bits-1 value: https://elixir.bootlin.com/linux/v6.12/source/arch/arm/nwfpe/softfloat-specialize#L267 This won't affect the only "real" use of the nwfpe emulation, which is ancient binaries that used it as part of the old floating point calling convention; that only uses loads and stores of 32 and 64 bit floats, not any of the floatx80 behaviour the original hardware had. We also get the nwfpe float64 default NaN value wrong: https://elixir.bootlin.com/linux/v6.12/source/arch/arm/nwfpe/softfloat-specialize#L166 so if we ever cared about this obscure corner the right fix would be to correct that so nwfpe used its own default-NaN setting rather than the Arm VFP one. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-29-peter.maydell@linaro.org --- fpu/softfloat-specialize.c.inc | 20 ++++++++++---------- 1 file changed, 10 insertions(+), 10 deletions(-) diff --git a/fpu/softfloat-specialize.c.inc b/fpu/softfloat-specialize.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-specialize.c.inc +++ b/fpu/softfloat-specialize.c.inc @@ -XXX,XX +XXX,XX @@ static void parts128_silence_nan(FloatParts128 *p, float_status *status) floatx80 floatx80_default_nan(float_status *status) { floatx80 r; + /* + * Extrapolate from the choices made by parts64_default_nan to fill + * in the floatx80 format. We assume that floatx80's explicit + * integer bit is always set (this is true for i386 and m68k, + * which are the only real users of this format). + */ + FloatParts64 p64; + parts64_default_nan(&p64, status); - /* None of the targets that have snan_bit_is_one use floatx80. */ - assert(!snan_bit_is_one(status)); -#if defined(TARGET_M68K) - r.low = UINT64_C(0xFFFFFFFFFFFFFFFF); - r.high = 0x7FFF; -#else - /* X86 */ - r.low = UINT64_C(0xC000000000000000); - r.high = 0xFFFF; -#endif + r.high = 0x7FFF | (p64.sign << 15); + r.low = (1ULL << DECOMPOSED_BINARY_POINT) | p64.frac; return r; } -- 2.34.1 In target/loongarch's helper_fclass_s() and helper_fclass_d() we pass a zero-initialized float_status struct to float32_is_quiet_nan() and float64_is_quiet_nan(), with the cryptic comment "for snan_bit_is_one". This pattern appears to have been copied from target/riscv, where it is used because the functions there do not have ready access to the CPU state struct. The comment presumably refers to the fact that the main reason the is_quiet_nan() functions want the float_state is because they want to know about the snan_bit_is_one config. In the loongarch helpers, though, we have the CPU state struct to hand. Use the usual env->fp_status here. This avoids our needing to track that we need to update the initializer of the local float_status structs when the core softfloat code adds new options for targets to configure their behaviour. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-30-peter.maydell@linaro.org --- target/loongarch/tcg/fpu_helper.c | 6 ++---- 1 file changed, 2 insertions(+), 4 deletions(-) diff --git a/target/loongarch/tcg/fpu_helper.c b/target/loongarch/tcg/fpu_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/loongarch/tcg/fpu_helper.c +++ b/target/loongarch/tcg/fpu_helper.c @@ -XXX,XX +XXX,XX @@ uint64_t helper_fclass_s(CPULoongArchState *env, uint64_t fj) } else if (float32_is_zero_or_denormal(f)) { return sign ? 1 << 4 : 1 << 8; } else if (float32_is_any_nan(f)) { - float_status s = { }; /* for snan_bit_is_one */ - return float32_is_quiet_nan(f, &s) ? 1 << 1 : 1 << 0; + return float32_is_quiet_nan(f, &env->fp_status) ? 1 << 1 : 1 << 0; } else { return sign ? 1 << 3 : 1 << 7; } @@ -XXX,XX +XXX,XX @@ uint64_t helper_fclass_d(CPULoongArchState *env, uint64_t fj) } else if (float64_is_zero_or_denormal(f)) { return sign ? 1 << 4 : 1 << 8; } else if (float64_is_any_nan(f)) { - float_status s = { }; /* for snan_bit_is_one */ - return float64_is_quiet_nan(f, &s) ? 1 << 1 : 1 << 0; + return float64_is_quiet_nan(f, &env->fp_status) ? 1 << 1 : 1 << 0; } else { return sign ? 1 << 3 : 1 << 7; } -- 2.34.1 In the frem helper, we have a local float_status because we want to execute the floatx80_div() with a custom rounding mode. Instead of zero-initializing the local float_status and then having to set it up with the m68k standard behaviour (including the NaN propagation rule and copying the rounding precision from env->fp_status), initialize it as a complete copy of env->fp_status. This will avoid our having to add new code in this function for every new config knob we add to fp_status. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-31-peter.maydell@linaro.org --- target/m68k/fpu_helper.c | 6 ++---- 1 file changed, 2 insertions(+), 4 deletions(-) diff --git a/target/m68k/fpu_helper.c b/target/m68k/fpu_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/m68k/fpu_helper.c +++ b/target/m68k/fpu_helper.c @@ -XXX,XX +XXX,XX @@ void HELPER(frem)(CPUM68KState *env, FPReg *res, FPReg *val0, FPReg *val1) fp_rem = floatx80_rem(val1->d, val0->d, &env->fp_status); if (!floatx80_is_any_nan(fp_rem)) { - float_status fp_status = { }; + /* Use local temporary fp_status to set different rounding mode */ + float_status fp_status = env->fp_status; uint32_t quotient; int sign; /* Calculate quotient directly using round to nearest mode */ - set_float_2nan_prop_rule(float_2nan_prop_ab, &fp_status); set_float_rounding_mode(float_round_nearest_even, &fp_status); - set_floatx80_rounding_precision( - get_floatx80_rounding_precision(&env->fp_status), &fp_status); fp_quot.d = floatx80_div(val1->d, val0->d, &fp_status); sign = extractFloatx80Sign(fp_quot.d); -- 2.34.1 In cf_fpu_gdb_get_reg() and cf_fpu_gdb_set_reg() we do the conversion from float64 to floatx80 using a scratch float_status, because we don't want the conversion to affect the CPU's floating point exception status. Currently we use a zero-initialized float_status. This will get steadily more awkward as we add config knobs to float_status that the target must initialize. Avoid having to add any of that configuration here by instead initializing our local float_status from the env->fp_status. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-32-peter.maydell@linaro.org --- target/m68k/helper.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) diff --git a/target/m68k/helper.c b/target/m68k/helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/m68k/helper.c +++ b/target/m68k/helper.c @@ -XXX,XX +XXX,XX @@ static int cf_fpu_gdb_get_reg(CPUState *cs, GByteArray *mem_buf, int n) CPUM68KState *env = &cpu->env; if (n < 8) { - float_status s = {}; + /* Use scratch float_status so any exceptions don't change CPU state */ + float_status s = env->fp_status; return gdb_get_reg64(mem_buf, floatx80_to_float64(env->fregs[n].d, &s)); } switch (n) { @@ -XXX,XX +XXX,XX @@ static int cf_fpu_gdb_set_reg(CPUState *cs, uint8_t *mem_buf, int n) CPUM68KState *env = &cpu->env; if (n < 8) { - float_status s = {}; + /* Use scratch float_status so any exceptions don't change CPU state */ + float_status s = env->fp_status; env->fregs[n].d = float64_to_floatx80(ldq_be_p(mem_buf), &s); return 8; } -- 2.34.1 In the helper functions flcmps and flcmpd we use a scratch float_status so that we don't change the CPU state if the comparison raises any floating point exception flags. Instead of zero-initializing this scratch float_status, initialize it as a copy of env->fp_status. This avoids the need to explicitly initialize settings like the NaN propagation rule or others we might add to softfloat in future. To do this we need to pass the CPU env pointer in to the helper. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-33-peter.maydell@linaro.org --- target/sparc/helper.h | 4 ++-- target/sparc/fop_helper.c | 8 ++++---- target/sparc/translate.c | 4 ++-- 3 files changed, 8 insertions(+), 8 deletions(-) diff --git a/target/sparc/helper.h b/target/sparc/helper.h index XXXXXXX..XXXXXXX 100644 --- a/target/sparc/helper.h +++ b/target/sparc/helper.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_3(fcmpd, TCG_CALL_NO_WG, i32, env, f64, f64) DEF_HELPER_FLAGS_3(fcmped, TCG_CALL_NO_WG, i32, env, f64, f64) DEF_HELPER_FLAGS_3(fcmpq, TCG_CALL_NO_WG, i32, env, i128, i128) DEF_HELPER_FLAGS_3(fcmpeq, TCG_CALL_NO_WG, i32, env, i128, i128) -DEF_HELPER_FLAGS_2(flcmps, TCG_CALL_NO_RWG_SE, i32, f32, f32) -DEF_HELPER_FLAGS_2(flcmpd, TCG_CALL_NO_RWG_SE, i32, f64, f64) +DEF_HELPER_FLAGS_3(flcmps, TCG_CALL_NO_RWG_SE, i32, env, f32, f32) +DEF_HELPER_FLAGS_3(flcmpd, TCG_CALL_NO_RWG_SE, i32, env, f64, f64) DEF_HELPER_2(raise_exception, noreturn, env, int) DEF_HELPER_FLAGS_3(faddd, TCG_CALL_NO_WG, f64, env, f64, f64) diff --git a/target/sparc/fop_helper.c b/target/sparc/fop_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/sparc/fop_helper.c +++ b/target/sparc/fop_helper.c @@ -XXX,XX +XXX,XX @@ uint32_t helper_fcmpeq(CPUSPARCState *env, Int128 src1, Int128 src2) return finish_fcmp(env, r, GETPC()); } -uint32_t helper_flcmps(float32 src1, float32 src2) +uint32_t helper_flcmps(CPUSPARCState *env, float32 src1, float32 src2) { /* * FLCMP never raises an exception nor modifies any FSR fields. * Perform the comparison with a dummy fp environment. */ - float_status discard = { }; + float_status discard = env->fp_status; FloatRelation r; set_float_2nan_prop_rule(float_2nan_prop_s_ba, &discard); @@ -XXX,XX +XXX,XX @@ uint32_t helper_flcmps(float32 src1, float32 src2) g_assert_not_reached(); } -uint32_t helper_flcmpd(float64 src1, float64 src2) +uint32_t helper_flcmpd(CPUSPARCState *env, float64 src1, float64 src2) { - float_status discard = { }; + float_status discard = env->fp_status; FloatRelation r; set_float_2nan_prop_rule(float_2nan_prop_s_ba, &discard); diff --git a/target/sparc/translate.c b/target/sparc/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/sparc/translate.c +++ b/target/sparc/translate.c @@ -XXX,XX +XXX,XX @@ static bool trans_FLCMPs(DisasContext *dc, arg_FLCMPs *a) src1 = gen_load_fpr_F(dc, a->rs1); src2 = gen_load_fpr_F(dc, a->rs2); - gen_helper_flcmps(cpu_fcc[a->cc], src1, src2); + gen_helper_flcmps(cpu_fcc[a->cc], tcg_env, src1, src2); return advance_pc(dc); } @@ -XXX,XX +XXX,XX @@ static bool trans_FLCMPd(DisasContext *dc, arg_FLCMPd *a) src1 = gen_load_fpr_D(dc, a->rs1); src2 = gen_load_fpr_D(dc, a->rs2); - gen_helper_flcmpd(cpu_fcc[a->cc], src1, src2); + gen_helper_flcmpd(cpu_fcc[a->cc], tcg_env, src1, src2); return advance_pc(dc); } -- 2.34.1 In the helper_compute_fprf functions, we pass a dummy float_status in to the is_signaling_nan() function. This is unnecessary, because we have convenient access to the CPU env pointer here and that is already set up with the correct values for the snan_bit_is_one and no_signaling_nans config settings. is_signaling_nan() doesn't ever update the fp_status with any exception flags, so there is no reason not to use env->fp_status here. Use env->fp_status instead of the dummy fp_status. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-34-peter.maydell@linaro.org --- target/ppc/fpu_helper.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) diff --git a/target/ppc/fpu_helper.c b/target/ppc/fpu_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/ppc/fpu_helper.c +++ b/target/ppc/fpu_helper.c @@ -XXX,XX +XXX,XX @@ void helper_compute_fprf_##tp(CPUPPCState *env, tp arg) \ } else if (tp##_is_infinity(arg)) { \ fprf = neg ? 0x09 << FPSCR_FPRF : 0x05 << FPSCR_FPRF; \ } else { \ - float_status dummy = { }; /* snan_bit_is_one = 0 */ \ - if (tp##_is_signaling_nan(arg, &dummy)) { \ + if (tp##_is_signaling_nan(arg, &env->fp_status)) { \ fprf = 0x00 << FPSCR_FPRF; \ } else { \ fprf = 0x11 << FPSCR_FPRF; \ -- 2.34.1 From: Richard Henderson <richard.henderson@linaro.org> Now that float_status has a bunch of fp parameters, it is easier to copy an existing structure than create one from scratch. Begin by copying the structure that corresponds to the FPSR and make only the adjustments required for BFloat16 semantics. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Message-id: 20241203203949.483774-2-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/arm/tcg/vec_helper.c | 20 +++++++------------- 1 file changed, 7 insertions(+), 13 deletions(-) diff --git a/target/arm/tcg/vec_helper.c b/target/arm/tcg/vec_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/vec_helper.c +++ b/target/arm/tcg/vec_helper.c @@ -XXX,XX +XXX,XX @@ bool is_ebf(CPUARMState *env, float_status *statusp, float_status *oddstatusp) * no effect on AArch32 instructions. */ bool ebf = is_a64(env) && env->vfp.fpcr & FPCR_EBF; - *statusp = (float_status){ - .tininess_before_rounding = float_tininess_before_rounding, - .float_rounding_mode = float_round_to_odd_inf, - .flush_to_zero = true, - .flush_inputs_to_zero = true, - .default_nan_mode = true, - }; + + *statusp = env->vfp.fp_status; + set_default_nan_mode(true, statusp); if (ebf) { - float_status *fpst = &env->vfp.fp_status; - set_flush_to_zero(get_flush_to_zero(fpst), statusp); - set_flush_inputs_to_zero(get_flush_inputs_to_zero(fpst), statusp); - set_float_rounding_mode(get_float_rounding_mode(fpst), statusp); - /* EBF=1 needs to do a step with round-to-odd semantics */ *oddstatusp = *statusp; set_float_rounding_mode(float_round_to_odd, oddstatusp); + } else { + set_flush_to_zero(true, statusp); + set_flush_inputs_to_zero(true, statusp); + set_float_rounding_mode(float_round_to_odd_inf, statusp); } - return ebf; } -- 2.34.1 Currently we hardcode the default NaN value in parts64_default_nan() using a compile-time ifdef ladder. This is awkward for two cases: * for single-QEMU-binary we can't hard-code target-specifics like this * for Arm FEAT_AFP the default NaN value depends on FPCR.AH (specifically the sign bit is different) Add a field to float_status to specify the default NaN value; fall back to the old ifdef behaviour if these are not set. The default NaN value is specified by setting a uint8_t to a pattern corresponding to the sign and upper fraction parts of the NaN; the lower bits of the fraction are set from bit 0 of the pattern. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-35-peter.maydell@linaro.org --- include/fpu/softfloat-helpers.h | 11 +++++++ include/fpu/softfloat-types.h | 10 ++++++ fpu/softfloat-specialize.c.inc | 55 ++++++++++++++++++++------------- 3 files changed, 54 insertions(+), 22 deletions(-) diff --git a/include/fpu/softfloat-helpers.h b/include/fpu/softfloat-helpers.h index XXXXXXX..XXXXXXX 100644 --- a/include/fpu/softfloat-helpers.h +++ b/include/fpu/softfloat-helpers.h @@ -XXX,XX +XXX,XX @@ static inline void set_float_infzeronan_rule(FloatInfZeroNaNRule rule, status->float_infzeronan_rule = rule; } +static inline void set_float_default_nan_pattern(uint8_t dnan_pattern, + float_status *status) +{ + status->default_nan_pattern = dnan_pattern; +} + static inline void set_flush_to_zero(bool val, float_status *status) { status->flush_to_zero = val; @@ -XXX,XX +XXX,XX @@ static inline FloatInfZeroNaNRule get_float_infzeronan_rule(float_status *status return status->float_infzeronan_rule; } +static inline uint8_t get_float_default_nan_pattern(float_status *status) +{ + return status->default_nan_pattern; +} + static inline bool get_flush_to_zero(float_status *status) { return status->flush_to_zero; diff --git a/include/fpu/softfloat-types.h b/include/fpu/softfloat-types.h index XXXXXXX..XXXXXXX 100644 --- a/include/fpu/softfloat-types.h +++ b/include/fpu/softfloat-types.h @@ -XXX,XX +XXX,XX @@ typedef struct float_status { /* should denormalised inputs go to zero and set the input_denormal flag? */ bool flush_inputs_to_zero; bool default_nan_mode; + /* + * The pattern to use for the default NaN. Here the high bit specifies + * the default NaN's sign bit, and bits 6..0 specify the high bits of the + * fractional part. The low bits of the fractional part are copies of bit 0. + * The exponent of the default NaN is (as for any NaN) always all 1s. + * Note that a value of 0 here is not a valid NaN. The target must set + * this to the correct non-zero value, or we will assert when trying to + * create a default NaN. + */ + uint8_t default_nan_pattern; /* * The flags below are not used on all specializations and may * constant fold away (see snan_bit_is_one()/no_signalling_nans() in diff --git a/fpu/softfloat-specialize.c.inc b/fpu/softfloat-specialize.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-specialize.c.inc +++ b/fpu/softfloat-specialize.c.inc @@ -XXX,XX +XXX,XX @@ static void parts64_default_nan(FloatParts64 *p, float_status *status) { bool sign = 0; uint64_t frac; + uint8_t dnan_pattern = status->default_nan_pattern; + if (dnan_pattern == 0) { #if defined(TARGET_SPARC) || defined(TARGET_M68K) - /* !snan_bit_is_one, set all bits */ - frac = (1ULL << DECOMPOSED_BINARY_POINT) - 1; -#elif defined(TARGET_I386) || defined(TARGET_X86_64) \ + /* Sign bit clear, all frac bits set */ + dnan_pattern = 0b01111111; +#elif defined(TARGET_I386) || defined(TARGET_X86_64) \ || defined(TARGET_MICROBLAZE) - /* !snan_bit_is_one, set sign and msb */ - frac = 1ULL << (DECOMPOSED_BINARY_POINT - 1); - sign = 1; + /* Sign bit set, most significant frac bit set */ + dnan_pattern = 0b11000000; #elif defined(TARGET_HPPA) - /* snan_bit_is_one, set msb-1. */ - frac = 1ULL << (DECOMPOSED_BINARY_POINT - 2); + /* Sign bit clear, msb-1 frac bit set */ + dnan_pattern = 0b00100000; #elif defined(TARGET_HEXAGON) - sign = 1; - frac = ~0ULL; + /* Sign bit set, all frac bits set. */ + dnan_pattern = 0b11111111; #else - /* - * This case is true for Alpha, ARM, MIPS, OpenRISC, PPC, RISC-V, - * S390, SH4, TriCore, and Xtensa. Our other supported targets - * do not have floating-point. - */ - if (snan_bit_is_one(status)) { - /* set all bits other than msb */ - frac = (1ULL << (DECOMPOSED_BINARY_POINT - 1)) - 1; - } else { - /* set msb */ - frac = 1ULL << (DECOMPOSED_BINARY_POINT - 1); - } + /* + * This case is true for Alpha, ARM, MIPS, OpenRISC, PPC, RISC-V, + * S390, SH4, TriCore, and Xtensa. Our other supported targets + * do not have floating-point. + */ + if (snan_bit_is_one(status)) { + /* sign bit clear, set all frac bits other than msb */ + dnan_pattern = 0b00111111; + } else { + /* sign bit clear, set frac msb */ + dnan_pattern = 0b01000000; + } #endif + } + assert(dnan_pattern != 0); + + sign = dnan_pattern >> 7; + /* + * Place default_nan_pattern [6:0] into bits [62:56], + * and replecate bit [0] down into [55:0] + */ + frac = deposit64(0, DECOMPOSED_BINARY_POINT - 7, 7, dnan_pattern); + frac = deposit64(frac, 0, DECOMPOSED_BINARY_POINT - 7, -(dnan_pattern & 1)); *p = (FloatParts64) { .cls = float_class_qnan, -- 2.34.1 Set the default NaN pattern explicitly for the tests/fp code. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-36-peter.maydell@linaro.org --- tests/fp/fp-bench.c | 1 + tests/fp/fp-test-log2.c | 1 + tests/fp/fp-test.c | 1 + 3 files changed, 3 insertions(+) diff --git a/tests/fp/fp-bench.c b/tests/fp/fp-bench.c index XXXXXXX..XXXXXXX 100644 --- a/tests/fp/fp-bench.c +++ b/tests/fp/fp-bench.c @@ -XXX,XX +XXX,XX @@ static void run_bench(void) set_float_2nan_prop_rule(float_2nan_prop_s_ab, &soft_status); set_float_3nan_prop_rule(float_3nan_prop_s_cab, &soft_status); set_float_infzeronan_rule(float_infzeronan_dnan_if_qnan, &soft_status); + set_float_default_nan_pattern(0b01000000, &soft_status); f = bench_funcs[operation][precision]; g_assert(f); diff --git a/tests/fp/fp-test-log2.c b/tests/fp/fp-test-log2.c index XXXXXXX..XXXXXXX 100644 --- a/tests/fp/fp-test-log2.c +++ b/tests/fp/fp-test-log2.c @@ -XXX,XX +XXX,XX @@ int main(int ac, char **av) int i; set_float_2nan_prop_rule(float_2nan_prop_s_ab, &qsf); + set_float_default_nan_pattern(0b01000000, &qsf); set_float_rounding_mode(float_round_nearest_even, &qsf); test.d = 0.0; diff --git a/tests/fp/fp-test.c b/tests/fp/fp-test.c index XXXXXXX..XXXXXXX 100644 --- a/tests/fp/fp-test.c +++ b/tests/fp/fp-test.c @@ -XXX,XX +XXX,XX @@ void run_test(void) */ set_float_2nan_prop_rule(float_2nan_prop_s_ab, &qsf); set_float_3nan_prop_rule(float_3nan_prop_s_cab, &qsf); + set_float_default_nan_pattern(0b01000000, &qsf); set_float_infzeronan_rule(float_infzeronan_dnan_if_qnan, &qsf); genCases_setLevel(test_level); -- 2.34.1 Set the default NaN pattern explicitly, and remove the ifdef from parts64_default_nan(). Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-37-peter.maydell@linaro.org --- target/microblaze/cpu.c | 2 ++ fpu/softfloat-specialize.c.inc | 3 +-- 2 files changed, 3 insertions(+), 2 deletions(-) diff --git a/target/microblaze/cpu.c b/target/microblaze/cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/microblaze/cpu.c +++ b/target/microblaze/cpu.c @@ -XXX,XX +XXX,XX @@ static void mb_cpu_reset_hold(Object *obj, ResetType type) * this architecture. */ set_float_2nan_prop_rule(float_2nan_prop_x87, &env->fp_status); + /* Default NaN: sign bit set, most significant frac bit set */ + set_float_default_nan_pattern(0b11000000, &env->fp_status); #if defined(CONFIG_USER_ONLY) /* start in user mode with interrupts enabled. */ diff --git a/fpu/softfloat-specialize.c.inc b/fpu/softfloat-specialize.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-specialize.c.inc +++ b/fpu/softfloat-specialize.c.inc @@ -XXX,XX +XXX,XX @@ static void parts64_default_nan(FloatParts64 *p, float_status *status) #if defined(TARGET_SPARC) || defined(TARGET_M68K) /* Sign bit clear, all frac bits set */ dnan_pattern = 0b01111111; -#elif defined(TARGET_I386) || defined(TARGET_X86_64) \ - || defined(TARGET_MICROBLAZE) +#elif defined(TARGET_I386) || defined(TARGET_X86_64) /* Sign bit set, most significant frac bit set */ dnan_pattern = 0b11000000; #elif defined(TARGET_HPPA) -- 2.34.1 Set the default NaN pattern explicitly, and remove the ifdef from parts64_default_nan(). Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-38-peter.maydell@linaro.org --- target/i386/tcg/fpu_helper.c | 4 ++++ fpu/softfloat-specialize.c.inc | 3 --- 2 files changed, 4 insertions(+), 3 deletions(-) diff --git a/target/i386/tcg/fpu_helper.c b/target/i386/tcg/fpu_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/i386/tcg/fpu_helper.c +++ b/target/i386/tcg/fpu_helper.c @@ -XXX,XX +XXX,XX @@ void cpu_init_fp_statuses(CPUX86State *env) */ set_float_infzeronan_rule(float_infzeronan_dnan_never, &env->sse_status); set_float_3nan_prop_rule(float_3nan_prop_abc, &env->sse_status); + /* Default NaN: sign bit set, most significant frac bit set */ + set_float_default_nan_pattern(0b11000000, &env->fp_status); + set_float_default_nan_pattern(0b11000000, &env->mmx_status); + set_float_default_nan_pattern(0b11000000, &env->sse_status); } static inline uint8_t save_exception_flags(CPUX86State *env) diff --git a/fpu/softfloat-specialize.c.inc b/fpu/softfloat-specialize.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-specialize.c.inc +++ b/fpu/softfloat-specialize.c.inc @@ -XXX,XX +XXX,XX @@ static void parts64_default_nan(FloatParts64 *p, float_status *status) #if defined(TARGET_SPARC) || defined(TARGET_M68K) /* Sign bit clear, all frac bits set */ dnan_pattern = 0b01111111; -#elif defined(TARGET_I386) || defined(TARGET_X86_64) - /* Sign bit set, most significant frac bit set */ - dnan_pattern = 0b11000000; #elif defined(TARGET_HPPA) /* Sign bit clear, msb-1 frac bit set */ dnan_pattern = 0b00100000; -- 2.34.1 Set the default NaN pattern explicitly, and remove the ifdef from parts64_default_nan(). Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-39-peter.maydell@linaro.org --- target/hppa/fpu_helper.c | 2 ++ fpu/softfloat-specialize.c.inc | 3 --- 2 files changed, 2 insertions(+), 3 deletions(-) diff --git a/target/hppa/fpu_helper.c b/target/hppa/fpu_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/hppa/fpu_helper.c +++ b/target/hppa/fpu_helper.c @@ -XXX,XX +XXX,XX @@ void HELPER(loaded_fr0)(CPUHPPAState *env) set_float_3nan_prop_rule(float_3nan_prop_abc, &env->fp_status); /* For inf * 0 + NaN, return the input NaN */ set_float_infzeronan_rule(float_infzeronan_dnan_never, &env->fp_status); + /* Default NaN: sign bit clear, msb-1 frac bit set */ + set_float_default_nan_pattern(0b00100000, &env->fp_status); } void cpu_hppa_loaded_fr0(CPUHPPAState *env) diff --git a/fpu/softfloat-specialize.c.inc b/fpu/softfloat-specialize.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-specialize.c.inc +++ b/fpu/softfloat-specialize.c.inc @@ -XXX,XX +XXX,XX @@ static void parts64_default_nan(FloatParts64 *p, float_status *status) #if defined(TARGET_SPARC) || defined(TARGET_M68K) /* Sign bit clear, all frac bits set */ dnan_pattern = 0b01111111; -#elif defined(TARGET_HPPA) - /* Sign bit clear, msb-1 frac bit set */ - dnan_pattern = 0b00100000; #elif defined(TARGET_HEXAGON) /* Sign bit set, all frac bits set. */ dnan_pattern = 0b11111111; -- 2.34.1 Set the default NaN pattern explicitly for the alpha target. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-40-peter.maydell@linaro.org --- target/alpha/cpu.c | 2 ++ 1 file changed, 2 insertions(+) diff --git a/target/alpha/cpu.c b/target/alpha/cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/alpha/cpu.c +++ b/target/alpha/cpu.c @@ -XXX,XX +XXX,XX @@ static void alpha_cpu_initfn(Object *obj) * operand in Fa. That is float_2nan_prop_ba. */ set_float_2nan_prop_rule(float_2nan_prop_x87, &env->fp_status); + /* Default NaN: sign bit clear, msb frac bit set */ + set_float_default_nan_pattern(0b01000000, &env->fp_status); #if defined(CONFIG_USER_ONLY) env->flags = ENV_FLAG_PS_USER | ENV_FLAG_FEN; cpu_alpha_store_fpcr(env, (uint64_t)(FPCR_INVD | FPCR_DZED | FPCR_OVFD -- 2.34.1 Set the default NaN pattern explicitly for the arm target. This includes setting it for the old linux-user nwfpe emulation. For nwfpe, our default doesn't match the real kernel, but we avoid making a behaviour change in this commit. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-41-peter.maydell@linaro.org --- linux-user/arm/nwfpe/fpa11.c | 5 +++++ target/arm/cpu.c | 2 ++ 2 files changed, 7 insertions(+) diff --git a/linux-user/arm/nwfpe/fpa11.c b/linux-user/arm/nwfpe/fpa11.c index XXXXXXX..XXXXXXX 100644 --- a/linux-user/arm/nwfpe/fpa11.c +++ b/linux-user/arm/nwfpe/fpa11.c @@ -XXX,XX +XXX,XX @@ void resetFPA11(void) * this late date. */ set_float_2nan_prop_rule(float_2nan_prop_s_ab, &fpa11->fp_status); + /* + * Use the same default NaN value as Arm VFP. This doesn't match + * the Linux kernel's nwfpe emulation, which uses an all-1s value. + */ + set_float_default_nan_pattern(0b01000000, &fpa11->fp_status); } void SetRoundingMode(const unsigned int opcode) 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_register_el_change_hook(ARMCPU *cpu, ARMELChangeHookFn *hook, * the pseudocode function the arguments are in the order c, a, b. * * 0 * Inf + NaN returns the default NaN if the input NaN is quiet, * and the input NaN if it is signalling + * * Default NaN has sign bit clear, msb frac bit set */ static void arm_set_default_fp_behaviours(float_status *s) { @@ -XXX,XX +XXX,XX @@ static void arm_set_default_fp_behaviours(float_status *s) set_float_2nan_prop_rule(float_2nan_prop_s_ab, s); set_float_3nan_prop_rule(float_3nan_prop_s_cab, s); set_float_infzeronan_rule(float_infzeronan_dnan_if_qnan, s); + set_float_default_nan_pattern(0b01000000, s); } static void cp_reg_reset(gpointer key, gpointer value, gpointer opaque) -- 2.34.1 Set the default NaN pattern explicitly for loongarch. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-42-peter.maydell@linaro.org --- target/loongarch/tcg/fpu_helper.c | 2 ++ 1 file changed, 2 insertions(+) diff --git a/target/loongarch/tcg/fpu_helper.c b/target/loongarch/tcg/fpu_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/loongarch/tcg/fpu_helper.c +++ b/target/loongarch/tcg/fpu_helper.c @@ -XXX,XX +XXX,XX @@ void restore_fp_status(CPULoongArchState *env) */ set_float_infzeronan_rule(float_infzeronan_dnan_never, &env->fp_status); set_float_3nan_prop_rule(float_3nan_prop_s_cab, &env->fp_status); + /* Default NaN: sign bit clear, msb frac bit set */ + set_float_default_nan_pattern(0b01000000, &env->fp_status); } int ieee_ex_to_loongarch(int xcpt) -- 2.34.1 Set the default NaN pattern explicitly for m68k. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-43-peter.maydell@linaro.org --- target/m68k/cpu.c | 2 ++ fpu/softfloat-specialize.c.inc | 2 +- 2 files changed, 3 insertions(+), 1 deletion(-) diff --git a/target/m68k/cpu.c b/target/m68k/cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/m68k/cpu.c +++ b/target/m68k/cpu.c @@ -XXX,XX +XXX,XX @@ static void m68k_cpu_reset_hold(Object *obj, ResetType type) * preceding paragraph for nonsignaling NaNs. */ set_float_2nan_prop_rule(float_2nan_prop_ab, &env->fp_status); + /* Default NaN: sign bit clear, all frac bits set */ + set_float_default_nan_pattern(0b01111111, &env->fp_status); nan = floatx80_default_nan(&env->fp_status); for (i = 0; i < 8; i++) { diff --git a/fpu/softfloat-specialize.c.inc b/fpu/softfloat-specialize.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-specialize.c.inc +++ b/fpu/softfloat-specialize.c.inc @@ -XXX,XX +XXX,XX @@ static void parts64_default_nan(FloatParts64 *p, float_status *status) uint8_t dnan_pattern = status->default_nan_pattern; if (dnan_pattern == 0) { -#if defined(TARGET_SPARC) || defined(TARGET_M68K) +#if defined(TARGET_SPARC) /* Sign bit clear, all frac bits set */ dnan_pattern = 0b01111111; #elif defined(TARGET_HEXAGON) -- 2.34.1 Set the default NaN pattern explicitly for MIPS. Note that this is our only target which currently changes the default NaN at runtime (which it was previously doing indirectly when it changed the snan_bit_is_one setting). Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-44-peter.maydell@linaro.org --- target/mips/fpu_helper.h | 7 +++++++ target/mips/msa.c | 3 +++ 2 files changed, 10 insertions(+) diff --git a/target/mips/fpu_helper.h b/target/mips/fpu_helper.h index XXXXXXX..XXXXXXX 100644 --- a/target/mips/fpu_helper.h +++ b/target/mips/fpu_helper.h @@ -XXX,XX +XXX,XX @@ static inline void restore_snan_bit_mode(CPUMIPSState *env) set_float_infzeronan_rule(izn_rule, &env->active_fpu.fp_status); nan3_rule = nan2008 ? float_3nan_prop_s_cab : float_3nan_prop_s_abc; set_float_3nan_prop_rule(nan3_rule, &env->active_fpu.fp_status); + /* + * With nan2008, the default NaN value has the sign bit clear and the + * frac msb set; with the older mode, the sign bit is clear, and all + * frac bits except the msb are set. + */ + set_float_default_nan_pattern(nan2008 ? 0b01000000 : 0b00111111, + &env->active_fpu.fp_status); } diff --git a/target/mips/msa.c b/target/mips/msa.c index XXXXXXX..XXXXXXX 100644 --- a/target/mips/msa.c +++ b/target/mips/msa.c @@ -XXX,XX +XXX,XX @@ void msa_reset(CPUMIPSState *env) /* Inf * 0 + NaN returns the input NaN */ set_float_infzeronan_rule(float_infzeronan_dnan_never, &env->active_tc.msa_fp_status); + /* Default NaN: sign bit clear, frac msb set */ + set_float_default_nan_pattern(0b01000000, + &env->active_tc.msa_fp_status); } -- 2.34.1 Set the default NaN pattern explicitly for openrisc. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-45-peter.maydell@linaro.org --- target/openrisc/cpu.c | 2 ++ 1 file changed, 2 insertions(+) diff --git a/target/openrisc/cpu.c b/target/openrisc/cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/openrisc/cpu.c +++ b/target/openrisc/cpu.c @@ -XXX,XX +XXX,XX @@ static void openrisc_cpu_reset_hold(Object *obj, ResetType type) */ set_float_2nan_prop_rule(float_2nan_prop_x87, &cpu->env.fp_status); + /* Default NaN: sign bit clear, frac msb set */ + set_float_default_nan_pattern(0b01000000, &cpu->env.fp_status); #ifndef CONFIG_USER_ONLY cpu->env.picmr = 0x00000000; -- 2.34.1 Set the default NaN pattern explicitly for ppc. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-46-peter.maydell@linaro.org --- target/ppc/cpu_init.c | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/target/ppc/cpu_init.c b/target/ppc/cpu_init.c index XXXXXXX..XXXXXXX 100644 --- a/target/ppc/cpu_init.c +++ b/target/ppc/cpu_init.c @@ -XXX,XX +XXX,XX @@ static void ppc_cpu_reset_hold(Object *obj, ResetType type) set_float_infzeronan_rule(float_infzeronan_dnan_never, &env->fp_status); set_float_infzeronan_rule(float_infzeronan_dnan_never, &env->vec_status); + /* Default NaN: sign bit clear, set frac msb */ + set_float_default_nan_pattern(0b01000000, &env->fp_status); + set_float_default_nan_pattern(0b01000000, &env->vec_status); + for (i = 0; i < ARRAY_SIZE(env->spr_cb); i++) { ppc_spr_t *spr = &env->spr_cb[i]; -- 2.34.1 Set the default NaN pattern explicitly for sh4. Note that sh4 is one of the only three targets (the others being HPPA and sometimes MIPS) that has snan_bit_is_one set. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-47-peter.maydell@linaro.org --- target/sh4/cpu.c | 2 ++ 1 file changed, 2 insertions(+) diff --git a/target/sh4/cpu.c b/target/sh4/cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/sh4/cpu.c +++ b/target/sh4/cpu.c @@ -XXX,XX +XXX,XX @@ static void superh_cpu_reset_hold(Object *obj, ResetType type) set_flush_to_zero(1, &env->fp_status); #endif set_default_nan_mode(1, &env->fp_status); + /* sign bit clear, set all frac bits other than msb */ + set_float_default_nan_pattern(0b00111111, &env->fp_status); } static void superh_cpu_disas_set_info(CPUState *cpu, disassemble_info *info) -- 2.34.1 Set the default NaN pattern explicitly for rx. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-48-peter.maydell@linaro.org --- target/rx/cpu.c | 2 ++ 1 file changed, 2 insertions(+) diff --git a/target/rx/cpu.c b/target/rx/cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/rx/cpu.c +++ b/target/rx/cpu.c @@ -XXX,XX +XXX,XX @@ static void rx_cpu_reset_hold(Object *obj, ResetType type) * then prefer dest over source", which is float_2nan_prop_s_ab. */ set_float_2nan_prop_rule(float_2nan_prop_x87, &env->fp_status); + /* Default NaN value: sign bit clear, set frac msb */ + set_float_default_nan_pattern(0b01000000, &env->fp_status); } static ObjectClass *rx_cpu_class_by_name(const char *cpu_model) -- 2.34.1 Set the default NaN pattern explicitly for s390x. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-49-peter.maydell@linaro.org --- target/s390x/cpu.c | 2 ++ 1 file changed, 2 insertions(+) diff --git a/target/s390x/cpu.c b/target/s390x/cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/s390x/cpu.c +++ b/target/s390x/cpu.c @@ -XXX,XX +XXX,XX @@ static void s390_cpu_reset_hold(Object *obj, ResetType type) set_float_3nan_prop_rule(float_3nan_prop_s_abc, &env->fpu_status); set_float_infzeronan_rule(float_infzeronan_dnan_always, &env->fpu_status); + /* Default NaN value: sign bit clear, frac msb set */ + set_float_default_nan_pattern(0b01000000, &env->fpu_status); /* fall through */ case RESET_TYPE_S390_CPU_NORMAL: env->psw.mask &= ~PSW_MASK_RI; -- 2.34.1 Set the default NaN pattern explicitly for SPARC, and remove the ifdef from parts64_default_nan. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-50-peter.maydell@linaro.org --- target/sparc/cpu.c | 2 ++ fpu/softfloat-specialize.c.inc | 5 +---- 2 files changed, 3 insertions(+), 4 deletions(-) diff --git a/target/sparc/cpu.c b/target/sparc/cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/sparc/cpu.c +++ b/target/sparc/cpu.c @@ -XXX,XX +XXX,XX @@ static void sparc_cpu_realizefn(DeviceState *dev, Error **errp) set_float_3nan_prop_rule(float_3nan_prop_s_cba, &env->fp_status); /* For inf * 0 + NaN, return the input NaN */ set_float_infzeronan_rule(float_infzeronan_dnan_never, &env->fp_status); + /* Default NaN value: sign bit clear, all frac bits set */ + set_float_default_nan_pattern(0b01111111, &env->fp_status); cpu_exec_realizefn(cs, &local_err); if (local_err != NULL) { diff --git a/fpu/softfloat-specialize.c.inc b/fpu/softfloat-specialize.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-specialize.c.inc +++ b/fpu/softfloat-specialize.c.inc @@ -XXX,XX +XXX,XX @@ static void parts64_default_nan(FloatParts64 *p, float_status *status) uint8_t dnan_pattern = status->default_nan_pattern; if (dnan_pattern == 0) { -#if defined(TARGET_SPARC) - /* Sign bit clear, all frac bits set */ - dnan_pattern = 0b01111111; -#elif defined(TARGET_HEXAGON) +#if defined(TARGET_HEXAGON) /* Sign bit set, all frac bits set. */ dnan_pattern = 0b11111111; #else -- 2.34.1 Set the default NaN pattern explicitly for xtensa. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-51-peter.maydell@linaro.org --- target/xtensa/cpu.c | 2 ++ 1 file changed, 2 insertions(+) diff --git a/target/xtensa/cpu.c b/target/xtensa/cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/xtensa/cpu.c +++ b/target/xtensa/cpu.c @@ -XXX,XX +XXX,XX @@ static void xtensa_cpu_reset_hold(Object *obj, ResetType type) /* For inf * 0 + NaN, return the input NaN */ set_float_infzeronan_rule(float_infzeronan_dnan_never, &env->fp_status); set_no_signaling_nans(!dfpu, &env->fp_status); + /* Default NaN value: sign bit clear, set frac msb */ + set_float_default_nan_pattern(0b01000000, &env->fp_status); xtensa_use_first_nan(env, !dfpu); } -- 2.34.1 Set the default NaN pattern explicitly for hexagon. Remove the ifdef from parts64_default_nan(); the only remaining unconverted targets all use the default case. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-52-peter.maydell@linaro.org --- target/hexagon/cpu.c | 2 ++ fpu/softfloat-specialize.c.inc | 5 ----- 2 files changed, 2 insertions(+), 5 deletions(-) diff --git a/target/hexagon/cpu.c b/target/hexagon/cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/hexagon/cpu.c +++ b/target/hexagon/cpu.c @@ -XXX,XX +XXX,XX @@ static void hexagon_cpu_reset_hold(Object *obj, ResetType type) set_default_nan_mode(1, &env->fp_status); set_float_detect_tininess(float_tininess_before_rounding, &env->fp_status); + /* Default NaN value: sign bit set, all frac bits set */ + set_float_default_nan_pattern(0b11111111, &env->fp_status); } static void hexagon_cpu_disas_set_info(CPUState *s, disassemble_info *info) diff --git a/fpu/softfloat-specialize.c.inc b/fpu/softfloat-specialize.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-specialize.c.inc +++ b/fpu/softfloat-specialize.c.inc @@ -XXX,XX +XXX,XX @@ static void parts64_default_nan(FloatParts64 *p, float_status *status) uint8_t dnan_pattern = status->default_nan_pattern; if (dnan_pattern == 0) { -#if defined(TARGET_HEXAGON) - /* Sign bit set, all frac bits set. */ - dnan_pattern = 0b11111111; -#else /* * This case is true for Alpha, ARM, MIPS, OpenRISC, PPC, RISC-V, * S390, SH4, TriCore, and Xtensa. Our other supported targets @@ -XXX,XX +XXX,XX @@ static void parts64_default_nan(FloatParts64 *p, float_status *status) /* sign bit clear, set frac msb */ dnan_pattern = 0b01000000; } -#endif } assert(dnan_pattern != 0); -- 2.34.1 Set the default NaN pattern explicitly for riscv. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-53-peter.maydell@linaro.org --- target/riscv/cpu.c | 2 ++ 1 file changed, 2 insertions(+) diff --git a/target/riscv/cpu.c b/target/riscv/cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/riscv/cpu.c +++ b/target/riscv/cpu.c @@ -XXX,XX +XXX,XX @@ static void riscv_cpu_reset_hold(Object *obj, ResetType type) cs->exception_index = RISCV_EXCP_NONE; env->load_res = -1; set_default_nan_mode(1, &env->fp_status); + /* Default NaN value: sign bit clear, frac msb set */ + set_float_default_nan_pattern(0b01000000, &env->fp_status); env->vill = true; #ifndef CONFIG_USER_ONLY -- 2.34.1 Set the default NaN pattern explicitly for tricore. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-54-peter.maydell@linaro.org --- target/tricore/helper.c | 2 ++ 1 file changed, 2 insertions(+) diff --git a/target/tricore/helper.c b/target/tricore/helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/tricore/helper.c +++ b/target/tricore/helper.c @@ -XXX,XX +XXX,XX @@ void fpu_set_state(CPUTriCoreState *env) set_flush_to_zero(1, &env->fp_status); set_float_detect_tininess(float_tininess_before_rounding, &env->fp_status); set_default_nan_mode(1, &env->fp_status); + /* Default NaN pattern: sign bit clear, frac msb set */ + set_float_default_nan_pattern(0b01000000, &env->fp_status); } uint32_t psw_read(CPUTriCoreState *env) -- 2.34.1 Now that all our targets have bene converted to explicitly specify their pattern for the default NaN value we can remove the remaining fallback code in parts64_default_nan(). Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241202131347.498124-55-peter.maydell@linaro.org --- fpu/softfloat-specialize.c.inc | 14 -------------- 1 file changed, 14 deletions(-) diff --git a/fpu/softfloat-specialize.c.inc b/fpu/softfloat-specialize.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-specialize.c.inc +++ b/fpu/softfloat-specialize.c.inc @@ -XXX,XX +XXX,XX @@ static void parts64_default_nan(FloatParts64 *p, float_status *status) uint64_t frac; uint8_t dnan_pattern = status->default_nan_pattern; - if (dnan_pattern == 0) { - /* - * This case is true for Alpha, ARM, MIPS, OpenRISC, PPC, RISC-V, - * S390, SH4, TriCore, and Xtensa. Our other supported targets - * do not have floating-point. - */ - if (snan_bit_is_one(status)) { - /* sign bit clear, set all frac bits other than msb */ - dnan_pattern = 0b00111111; - } else { - /* sign bit clear, set frac msb */ - dnan_pattern = 0b01000000; - } - } assert(dnan_pattern != 0); sign = dnan_pattern >> 7; -- 2.34.1 From: Richard Henderson <richard.henderson@linaro.org> Inline pickNaNMulAdd into its only caller. This makes one assert redundant with the immediately preceding IF. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org> Message-id: 20241203203949.483774-3-richard.henderson@linaro.org [PMM: keep comment from old code in new location] Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- fpu/softfloat-parts.c.inc | 41 +++++++++++++++++++++++++- fpu/softfloat-specialize.c.inc | 54 ---------------------------------- 2 files changed, 40 insertions(+), 55 deletions(-) diff --git a/fpu/softfloat-parts.c.inc b/fpu/softfloat-parts.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-parts.c.inc +++ b/fpu/softfloat-parts.c.inc @@ -XXX,XX +XXX,XX @@ static FloatPartsN *partsN(pick_nan_muladd)(FloatPartsN *a, FloatPartsN *b, } if (s->default_nan_mode) { + /* + * We guarantee not to require the target to tell us how to + * pick a NaN if we're always returning the default NaN. + * But if we're not in default-NaN mode then the target must + * specify. + */ which = 3; + } else if (infzero) { + /* + * Inf * 0 + NaN -- some implementations return the + * default NaN here, and some return the input NaN. + */ + switch (s->float_infzeronan_rule) { + case float_infzeronan_dnan_never: + which = 2; + break; + case float_infzeronan_dnan_always: + which = 3; + break; + case float_infzeronan_dnan_if_qnan: + which = is_qnan(c->cls) ? 3 : 2; + break; + default: + g_assert_not_reached(); + } } else { - which = pickNaNMulAdd(a->cls, b->cls, c->cls, infzero, have_snan, s); + FloatClass cls[3] = { a->cls, b->cls, c->cls }; + Float3NaNPropRule rule = s->float_3nan_prop_rule; + + assert(rule != float_3nan_prop_none); + if (have_snan && (rule & R_3NAN_SNAN_MASK)) { + /* We have at least one SNaN input and should prefer it */ + do { + which = rule & R_3NAN_1ST_MASK; + rule >>= R_3NAN_1ST_LENGTH; + } while (!is_snan(cls[which])); + } else { + do { + which = rule & R_3NAN_1ST_MASK; + rule >>= R_3NAN_1ST_LENGTH; + } while (!is_nan(cls[which])); + } } if (which == 3) { diff --git a/fpu/softfloat-specialize.c.inc b/fpu/softfloat-specialize.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-specialize.c.inc +++ b/fpu/softfloat-specialize.c.inc @@ -XXX,XX +XXX,XX @@ static int pickNaN(FloatClass a_cls, FloatClass b_cls, } } -/*---------------------------------------------------------------------------- -| Select which NaN to propagate for a three-input operation. -| For the moment we assume that no CPU needs the 'larger significand' -| information. -| Return values : 0 : a; 1 : b; 2 : c; 3 : default-NaN -*----------------------------------------------------------------------------*/ -static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls, - bool infzero, bool have_snan, float_status *status) -{ - FloatClass cls[3] = { a_cls, b_cls, c_cls }; - Float3NaNPropRule rule = status->float_3nan_prop_rule; - int which; - - /* - * We guarantee not to require the target to tell us how to - * pick a NaN if we're always returning the default NaN. - * But if we're not in default-NaN mode then the target must - * specify. - */ - assert(!status->default_nan_mode); - - if (infzero) { - /* - * Inf * 0 + NaN -- some implementations return the default NaN here, - * and some return the input NaN. - */ - switch (status->float_infzeronan_rule) { - case float_infzeronan_dnan_never: - return 2; - case float_infzeronan_dnan_always: - return 3; - case float_infzeronan_dnan_if_qnan: - return is_qnan(c_cls) ? 3 : 2; - default: - g_assert_not_reached(); - } - } - - assert(rule != float_3nan_prop_none); - if (have_snan && (rule & R_3NAN_SNAN_MASK)) { - /* We have at least one SNaN input and should prefer it */ - do { - which = rule & R_3NAN_1ST_MASK; - rule >>= R_3NAN_1ST_LENGTH; - } while (!is_snan(cls[which])); - } else { - do { - which = rule & R_3NAN_1ST_MASK; - rule >>= R_3NAN_1ST_LENGTH; - } while (!is_nan(cls[which])); - } - return which; -} - /*---------------------------------------------------------------------------- | Returns 1 if the double-precision floating-point value `a' is a quiet | NaN; otherwise returns 0. -- 2.34.1 From: Richard Henderson <richard.henderson@linaro.org> Remove "3" as a special case for which and simply branch to return the desired value. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org> Message-id: 20241203203949.483774-4-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- fpu/softfloat-parts.c.inc | 20 ++++++++++---------- 1 file changed, 10 insertions(+), 10 deletions(-) diff --git a/fpu/softfloat-parts.c.inc b/fpu/softfloat-parts.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-parts.c.inc +++ b/fpu/softfloat-parts.c.inc @@ -XXX,XX +XXX,XX @@ static FloatPartsN *partsN(pick_nan_muladd)(FloatPartsN *a, FloatPartsN *b, * But if we're not in default-NaN mode then the target must * specify. */ - which = 3; + goto default_nan; } else if (infzero) { /* * Inf * 0 + NaN -- some implementations return the @@ -XXX,XX +XXX,XX @@ static FloatPartsN *partsN(pick_nan_muladd)(FloatPartsN *a, FloatPartsN *b, */ switch (s->float_infzeronan_rule) { case float_infzeronan_dnan_never: - which = 2; break; case float_infzeronan_dnan_always: - which = 3; - break; + goto default_nan; case float_infzeronan_dnan_if_qnan: - which = is_qnan(c->cls) ? 3 : 2; + if (is_qnan(c->cls)) { + goto default_nan; + } break; default: g_assert_not_reached(); } + which = 2; } else { FloatClass cls[3] = { a->cls, b->cls, c->cls }; Float3NaNPropRule rule = s->float_3nan_prop_rule; @@ -XXX,XX +XXX,XX @@ static FloatPartsN *partsN(pick_nan_muladd)(FloatPartsN *a, FloatPartsN *b, } } - if (which == 3) { - parts_default_nan(a, s); - return a; - } - switch (which) { case 0: break; @@ -XXX,XX +XXX,XX @@ static FloatPartsN *partsN(pick_nan_muladd)(FloatPartsN *a, FloatPartsN *b, parts_silence_nan(a, s); } return a; + + default_nan: + parts_default_nan(a, s); + return a; } /* -- 2.34.1 From: Richard Henderson <richard.henderson@linaro.org> Assign the pointer return value to 'a' directly, rather than going through an intermediary index. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org> Message-id: 20241203203949.483774-5-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- fpu/softfloat-parts.c.inc | 32 ++++++++++---------------------- 1 file changed, 10 insertions(+), 22 deletions(-) diff --git a/fpu/softfloat-parts.c.inc b/fpu/softfloat-parts.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-parts.c.inc +++ b/fpu/softfloat-parts.c.inc @@ -XXX,XX +XXX,XX @@ static FloatPartsN *partsN(pick_nan_muladd)(FloatPartsN *a, FloatPartsN *b, FloatPartsN *c, float_status *s, int ab_mask, int abc_mask) { - int which; bool infzero = (ab_mask == float_cmask_infzero); bool have_snan = (abc_mask & float_cmask_snan); + FloatPartsN *ret; if (unlikely(have_snan)) { float_raise(float_flag_invalid | float_flag_invalid_snan, s); @@ -XXX,XX +XXX,XX @@ static FloatPartsN *partsN(pick_nan_muladd)(FloatPartsN *a, FloatPartsN *b, default: g_assert_not_reached(); } - which = 2; + ret = c; } else { - FloatClass cls[3] = { a->cls, b->cls, c->cls }; + FloatPartsN *val[3] = { a, b, c }; Float3NaNPropRule rule = s->float_3nan_prop_rule; assert(rule != float_3nan_prop_none); if (have_snan && (rule & R_3NAN_SNAN_MASK)) { /* We have at least one SNaN input and should prefer it */ do { - which = rule & R_3NAN_1ST_MASK; + ret = val[rule & R_3NAN_1ST_MASK]; rule >>= R_3NAN_1ST_LENGTH; - } while (!is_snan(cls[which])); + } while (!is_snan(ret->cls)); } else { do { - which = rule & R_3NAN_1ST_MASK; + ret = val[rule & R_3NAN_1ST_MASK]; rule >>= R_3NAN_1ST_LENGTH; - } while (!is_nan(cls[which])); + } while (!is_nan(ret->cls)); } } - switch (which) { - case 0: - break; - case 1: - a = b; - break; - case 2: - a = c; - break; - default: - g_assert_not_reached(); + if (is_snan(ret->cls)) { + parts_silence_nan(ret, s); } - if (is_snan(a->cls)) { - parts_silence_nan(a, s); - } - return a; + return ret; default_nan: parts_default_nan(a, s); -- 2.34.1 From: Richard Henderson <richard.henderson@linaro.org> While all indices into val[] should be in [0-2], the mask applied is two bits. To help static analysis see there is no possibility of read beyond the end of the array, pad the array to 4 entries, with the final being (implicitly) NULL. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org> Message-id: 20241203203949.483774-6-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- fpu/softfloat-parts.c.inc | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/fpu/softfloat-parts.c.inc b/fpu/softfloat-parts.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-parts.c.inc +++ b/fpu/softfloat-parts.c.inc @@ -XXX,XX +XXX,XX @@ static FloatPartsN *partsN(pick_nan_muladd)(FloatPartsN *a, FloatPartsN *b, } ret = c; } else { - FloatPartsN *val[3] = { a, b, c }; + FloatPartsN *val[R_3NAN_1ST_MASK + 1] = { a, b, c }; Float3NaNPropRule rule = s->float_3nan_prop_rule; assert(rule != float_3nan_prop_none); -- 2.34.1 From: Richard Henderson <richard.henderson@linaro.org> This function is part of the public interface and is not "specialized" to any target in any way. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Message-id: 20241203203949.483774-7-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- fpu/softfloat.c | 52 ++++++++++++++++++++++++++++++++++ fpu/softfloat-specialize.c.inc | 52 ---------------------------------- 2 files changed, 52 insertions(+), 52 deletions(-) diff --git a/fpu/softfloat.c b/fpu/softfloat.c index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat.c +++ b/fpu/softfloat.c @@ -XXX,XX +XXX,XX @@ void normalizeFloatx80Subnormal(uint64_t aSig, int32_t *zExpPtr, *zExpPtr = 1 - shiftCount; } +/*---------------------------------------------------------------------------- +| Takes two extended double-precision floating-point values `a' and `b', one +| of which is a NaN, and returns the appropriate NaN result. If either `a' or +| `b' is a signaling NaN, the invalid exception is raised. +*----------------------------------------------------------------------------*/ + +floatx80 propagateFloatx80NaN(floatx80 a, floatx80 b, float_status *status) +{ + bool aIsLargerSignificand; + FloatClass a_cls, b_cls; + + /* This is not complete, but is good enough for pickNaN. */ + a_cls = (!floatx80_is_any_nan(a) + ? float_class_normal + : floatx80_is_signaling_nan(a, status) + ? float_class_snan + : float_class_qnan); + b_cls = (!floatx80_is_any_nan(b) + ? float_class_normal + : floatx80_is_signaling_nan(b, status) + ? float_class_snan + : float_class_qnan); + + if (is_snan(a_cls) || is_snan(b_cls)) { + float_raise(float_flag_invalid, status); + } + + if (status->default_nan_mode) { + return floatx80_default_nan(status); + } + + if (a.low < b.low) { + aIsLargerSignificand = 0; + } else if (b.low < a.low) { + aIsLargerSignificand = 1; + } else { + aIsLargerSignificand = (a.high < b.high) ? 1 : 0; + } + + if (pickNaN(a_cls, b_cls, aIsLargerSignificand, status)) { + if (is_snan(b_cls)) { + return floatx80_silence_nan(b, status); + } + return b; + } else { + if (is_snan(a_cls)) { + return floatx80_silence_nan(a, status); + } + return a; + } +} + /*---------------------------------------------------------------------------- | Takes an abstract floating-point value having sign `zSign', exponent `zExp', | and extended significand formed by the concatenation of `zSig0' and `zSig1', diff --git a/fpu/softfloat-specialize.c.inc b/fpu/softfloat-specialize.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-specialize.c.inc +++ b/fpu/softfloat-specialize.c.inc @@ -XXX,XX +XXX,XX @@ floatx80 floatx80_silence_nan(floatx80 a, float_status *status) return a; } -/*---------------------------------------------------------------------------- -| Takes two extended double-precision floating-point values `a' and `b', one -| of which is a NaN, and returns the appropriate NaN result. If either `a' or -| `b' is a signaling NaN, the invalid exception is raised. -*----------------------------------------------------------------------------*/ - -floatx80 propagateFloatx80NaN(floatx80 a, floatx80 b, float_status *status) -{ - bool aIsLargerSignificand; - FloatClass a_cls, b_cls; - - /* This is not complete, but is good enough for pickNaN. */ - a_cls = (!floatx80_is_any_nan(a) - ? float_class_normal - : floatx80_is_signaling_nan(a, status) - ? float_class_snan - : float_class_qnan); - b_cls = (!floatx80_is_any_nan(b) - ? float_class_normal - : floatx80_is_signaling_nan(b, status) - ? float_class_snan - : float_class_qnan); - - if (is_snan(a_cls) || is_snan(b_cls)) { - float_raise(float_flag_invalid, status); - } - - if (status->default_nan_mode) { - return floatx80_default_nan(status); - } - - if (a.low < b.low) { - aIsLargerSignificand = 0; - } else if (b.low < a.low) { - aIsLargerSignificand = 1; - } else { - aIsLargerSignificand = (a.high < b.high) ? 1 : 0; - } - - if (pickNaN(a_cls, b_cls, aIsLargerSignificand, status)) { - if (is_snan(b_cls)) { - return floatx80_silence_nan(b, status); - } - return b; - } else { - if (is_snan(a_cls)) { - return floatx80_silence_nan(a, status); - } - return a; - } -} - /*---------------------------------------------------------------------------- | Returns 1 if the quadruple-precision floating-point value `a' is a quiet | NaN; otherwise returns 0. -- 2.34.1 From: Richard Henderson <richard.henderson@linaro.org> Unpacking and repacking the parts may be slightly more work than we did before, but we get to reuse more code. For a code path handling exceptional values, this is an improvement. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20241203203949.483774-8-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> --- fpu/softfloat.c | 43 +++++-------------------------------------- 1 file changed, 5 insertions(+), 38 deletions(-) diff --git a/fpu/softfloat.c b/fpu/softfloat.c index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat.c +++ b/fpu/softfloat.c @@ -XXX,XX +XXX,XX @@ void normalizeFloatx80Subnormal(uint64_t aSig, int32_t *zExpPtr, floatx80 propagateFloatx80NaN(floatx80 a, floatx80 b, float_status *status) { - bool aIsLargerSignificand; - FloatClass a_cls, b_cls; + FloatParts128 pa, pb, *pr; - /* This is not complete, but is good enough for pickNaN. */ - a_cls = (!floatx80_is_any_nan(a) - ? float_class_normal - : floatx80_is_signaling_nan(a, status) - ? float_class_snan - : float_class_qnan); - b_cls = (!floatx80_is_any_nan(b) - ? float_class_normal - : floatx80_is_signaling_nan(b, status) - ? float_class_snan - : float_class_qnan); - - if (is_snan(a_cls) || is_snan(b_cls)) { - float_raise(float_flag_invalid, status); - } - - if (status->default_nan_mode) { + if (!floatx80_unpack_canonical(&pa, a, status) || + !floatx80_unpack_canonical(&pb, b, status)) { return floatx80_default_nan(status); } - if (a.low < b.low) { - aIsLargerSignificand = 0; - } else if (b.low < a.low) { - aIsLargerSignificand = 1; - } else { - aIsLargerSignificand = (a.high < b.high) ? 1 : 0; - } - - if (pickNaN(a_cls, b_cls, aIsLargerSignificand, status)) { - if (is_snan(b_cls)) { - return floatx80_silence_nan(b, status); - } - return b; - } else { - if (is_snan(a_cls)) { - return floatx80_silence_nan(a, status); - } - return a; - } + pr = parts_pick_nan(&pa, &pb, status); + return floatx80_round_pack_canonical(pr, status); } /*---------------------------------------------------------------------------- -- 2.34.1 From: Richard Henderson <richard.henderson@linaro.org> Inline pickNaN into its only caller. This makes one assert redundant with the immediately preceding IF. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org> Message-id: 20241203203949.483774-9-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- fpu/softfloat-parts.c.inc | 82 +++++++++++++++++++++++++---- fpu/softfloat-specialize.c.inc | 96 ---------------------------------- 2 files changed, 73 insertions(+), 105 deletions(-) diff --git a/fpu/softfloat-parts.c.inc b/fpu/softfloat-parts.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-parts.c.inc +++ b/fpu/softfloat-parts.c.inc @@ -XXX,XX +XXX,XX @@ static void partsN(return_nan)(FloatPartsN *a, float_status *s) static FloatPartsN *partsN(pick_nan)(FloatPartsN *a, FloatPartsN *b, float_status *s) { + int cmp, which; + if (is_snan(a->cls) || is_snan(b->cls)) { float_raise(float_flag_invalid | float_flag_invalid_snan, s); } if (s->default_nan_mode) { parts_default_nan(a, s); - } else { - int cmp = frac_cmp(a, b); - if (cmp == 0) { - cmp = a->sign < b->sign; - } + return a; + } - if (pickNaN(a->cls, b->cls, cmp > 0, s)) { - a = b; - } + cmp = frac_cmp(a, b); + if (cmp == 0) { + cmp = a->sign < b->sign; + } + + switch (s->float_2nan_prop_rule) { + case float_2nan_prop_s_ab: if (is_snan(a->cls)) { - parts_silence_nan(a, s); + which = 0; + } else if (is_snan(b->cls)) { + which = 1; + } else if (is_qnan(a->cls)) { + which = 0; + } else { + which = 1; } + break; + case float_2nan_prop_s_ba: + if (is_snan(b->cls)) { + which = 1; + } else if (is_snan(a->cls)) { + which = 0; + } else if (is_qnan(b->cls)) { + which = 1; + } else { + which = 0; + } + break; + case float_2nan_prop_ab: + which = is_nan(a->cls) ? 0 : 1; + break; + case float_2nan_prop_ba: + which = is_nan(b->cls) ? 1 : 0; + break; + case float_2nan_prop_x87: + /* + * This implements x87 NaN propagation rules: + * SNaN + QNaN => return the QNaN + * two SNaNs => return the one with the larger significand, silenced + * two QNaNs => return the one with the larger significand + * SNaN and a non-NaN => return the SNaN, silenced + * QNaN and a non-NaN => return the QNaN + * + * If we get down to comparing significands and they are the same, + * return the NaN with the positive sign bit (if any). + */ + if (is_snan(a->cls)) { + if (is_snan(b->cls)) { + which = cmp > 0 ? 0 : 1; + } else { + which = is_qnan(b->cls) ? 1 : 0; + } + } else if (is_qnan(a->cls)) { + if (is_snan(b->cls) || !is_qnan(b->cls)) { + which = 0; + } else { + which = cmp > 0 ? 0 : 1; + } + } else { + which = 1; + } + break; + default: + g_assert_not_reached(); + } + + if (which) { + a = b; + } + if (is_snan(a->cls)) { + parts_silence_nan(a, s); } return a; } diff --git a/fpu/softfloat-specialize.c.inc b/fpu/softfloat-specialize.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-specialize.c.inc +++ b/fpu/softfloat-specialize.c.inc @@ -XXX,XX +XXX,XX @@ bool float32_is_signaling_nan(float32 a_, float_status *status) } } -/*---------------------------------------------------------------------------- -| Select which NaN to propagate for a two-input operation. -| IEEE754 doesn't specify all the details of this, so the -| algorithm is target-specific. -| The routine is passed various bits of information about the -| two NaNs and should return 0 to select NaN a and 1 for NaN b. -| Note that signalling NaNs are always squashed to quiet NaNs -| by the caller, by calling floatXX_silence_nan() before -| returning them. -| -| aIsLargerSignificand is only valid if both a and b are NaNs -| of some kind, and is true if a has the larger significand, -| or if both a and b have the same significand but a is -| positive but b is negative. It is only needed for the x87 -| tie-break rule. -*----------------------------------------------------------------------------*/ - -static int pickNaN(FloatClass a_cls, FloatClass b_cls, - bool aIsLargerSignificand, float_status *status) -{ - /* - * We guarantee not to require the target to tell us how to - * pick a NaN if we're always returning the default NaN. - * But if we're not in default-NaN mode then the target must - * specify via set_float_2nan_prop_rule(). - */ - assert(!status->default_nan_mode); - - switch (status->float_2nan_prop_rule) { - case float_2nan_prop_s_ab: - if (is_snan(a_cls)) { - return 0; - } else if (is_snan(b_cls)) { - return 1; - } else if (is_qnan(a_cls)) { - return 0; - } else { - return 1; - } - break; - case float_2nan_prop_s_ba: - if (is_snan(b_cls)) { - return 1; - } else if (is_snan(a_cls)) { - return 0; - } else if (is_qnan(b_cls)) { - return 1; - } else { - return 0; - } - break; - case float_2nan_prop_ab: - if (is_nan(a_cls)) { - return 0; - } else { - return 1; - } - break; - case float_2nan_prop_ba: - if (is_nan(b_cls)) { - return 1; - } else { - return 0; - } - break; - case float_2nan_prop_x87: - /* - * This implements x87 NaN propagation rules: - * SNaN + QNaN => return the QNaN - * two SNaNs => return the one with the larger significand, silenced - * two QNaNs => return the one with the larger significand - * SNaN and a non-NaN => return the SNaN, silenced - * QNaN and a non-NaN => return the QNaN - * - * If we get down to comparing significands and they are the same, - * return the NaN with the positive sign bit (if any). - */ - if (is_snan(a_cls)) { - if (is_snan(b_cls)) { - return aIsLargerSignificand ? 0 : 1; - } - return is_qnan(b_cls) ? 1 : 0; - } else if (is_qnan(a_cls)) { - if (is_snan(b_cls) || !is_qnan(b_cls)) { - return 0; - } else { - return aIsLargerSignificand ? 0 : 1; - } - } else { - return 1; - } - default: - g_assert_not_reached(); - } -} - /*---------------------------------------------------------------------------- | Returns 1 if the double-precision floating-point value `a' is a quiet | NaN; otherwise returns 0. -- 2.34.1 From: Richard Henderson <richard.henderson@linaro.org> Remember if there was an SNaN, and use that to simplify float_2nan_prop_s_{ab,ba} to only the snan component. Then, fall through to the corresponding float_2nan_prop_{ab,ba} case to handle any remaining nans, which must be quiet. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Message-id: 20241203203949.483774-10-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- fpu/softfloat-parts.c.inc | 32 ++++++++++++-------------------- 1 file changed, 12 insertions(+), 20 deletions(-) diff --git a/fpu/softfloat-parts.c.inc b/fpu/softfloat-parts.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-parts.c.inc +++ b/fpu/softfloat-parts.c.inc @@ -XXX,XX +XXX,XX @@ static void partsN(return_nan)(FloatPartsN *a, float_status *s) static FloatPartsN *partsN(pick_nan)(FloatPartsN *a, FloatPartsN *b, float_status *s) { + bool have_snan = false; int cmp, which; if (is_snan(a->cls) || is_snan(b->cls)) { float_raise(float_flag_invalid | float_flag_invalid_snan, s); + have_snan = true; } if (s->default_nan_mode) { @@ -XXX,XX +XXX,XX @@ static FloatPartsN *partsN(pick_nan)(FloatPartsN *a, FloatPartsN *b, switch (s->float_2nan_prop_rule) { case float_2nan_prop_s_ab: - if (is_snan(a->cls)) { - which = 0; - } else if (is_snan(b->cls)) { - which = 1; - } else if (is_qnan(a->cls)) { - which = 0; - } else { - which = 1; + if (have_snan) { + which = is_snan(a->cls) ? 0 : 1; + break; } - break; - case float_2nan_prop_s_ba: - if (is_snan(b->cls)) { - which = 1; - } else if (is_snan(a->cls)) { - which = 0; - } else if (is_qnan(b->cls)) { - which = 1; - } else { - which = 0; - } - break; + /* fall through */ case float_2nan_prop_ab: which = is_nan(a->cls) ? 0 : 1; break; + case float_2nan_prop_s_ba: + if (have_snan) { + which = is_snan(b->cls) ? 1 : 0; + break; + } + /* fall through */ case float_2nan_prop_ba: which = is_nan(b->cls) ? 1 : 0; break; -- 2.34.1 From: Richard Henderson <richard.henderson@linaro.org> Move the fractional comparison to the end of the float_2nan_prop_x87 case. This is not required for any other 2nan propagation rule. Reorganize the x87 case itself to break out of the switch when the fractional comparison is not required. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Message-id: 20241203203949.483774-11-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- fpu/softfloat-parts.c.inc | 19 +++++++++---------- 1 file changed, 9 insertions(+), 10 deletions(-) diff --git a/fpu/softfloat-parts.c.inc b/fpu/softfloat-parts.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-parts.c.inc +++ b/fpu/softfloat-parts.c.inc @@ -XXX,XX +XXX,XX @@ static FloatPartsN *partsN(pick_nan)(FloatPartsN *a, FloatPartsN *b, return a; } - cmp = frac_cmp(a, b); - if (cmp == 0) { - cmp = a->sign < b->sign; - } - switch (s->float_2nan_prop_rule) { case float_2nan_prop_s_ab: if (have_snan) { @@ -XXX,XX +XXX,XX @@ static FloatPartsN *partsN(pick_nan)(FloatPartsN *a, FloatPartsN *b, * return the NaN with the positive sign bit (if any). */ if (is_snan(a->cls)) { - if (is_snan(b->cls)) { - which = cmp > 0 ? 0 : 1; - } else { + if (!is_snan(b->cls)) { which = is_qnan(b->cls) ? 1 : 0; + break; } } else if (is_qnan(a->cls)) { if (is_snan(b->cls) || !is_qnan(b->cls)) { which = 0; - } else { - which = cmp > 0 ? 0 : 1; + break; } } else { which = 1; + break; } + cmp = frac_cmp(a, b); + if (cmp == 0) { + cmp = a->sign < b->sign; + } + which = cmp > 0 ? 0 : 1; break; default: g_assert_not_reached(); -- 2.34.1 From: Richard Henderson <richard.henderson@linaro.org> Replace the "index" selecting between A and B with a result variable of the proper type. This improves clarity within the function. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org> Message-id: 20241203203949.483774-12-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- fpu/softfloat-parts.c.inc | 28 +++++++++++++--------------- 1 file changed, 13 insertions(+), 15 deletions(-) diff --git a/fpu/softfloat-parts.c.inc b/fpu/softfloat-parts.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-parts.c.inc +++ b/fpu/softfloat-parts.c.inc @@ -XXX,XX +XXX,XX @@ static FloatPartsN *partsN(pick_nan)(FloatPartsN *a, FloatPartsN *b, float_status *s) { bool have_snan = false; - int cmp, which; + FloatPartsN *ret; + int cmp; if (is_snan(a->cls) || is_snan(b->cls)) { float_raise(float_flag_invalid | float_flag_invalid_snan, s); @@ -XXX,XX +XXX,XX @@ static FloatPartsN *partsN(pick_nan)(FloatPartsN *a, FloatPartsN *b, switch (s->float_2nan_prop_rule) { case float_2nan_prop_s_ab: if (have_snan) { - which = is_snan(a->cls) ? 0 : 1; + ret = is_snan(a->cls) ? a : b; break; } /* fall through */ case float_2nan_prop_ab: - which = is_nan(a->cls) ? 0 : 1; + ret = is_nan(a->cls) ? a : b; break; case float_2nan_prop_s_ba: if (have_snan) { - which = is_snan(b->cls) ? 1 : 0; + ret = is_snan(b->cls) ? b : a; break; } /* fall through */ case float_2nan_prop_ba: - which = is_nan(b->cls) ? 1 : 0; + ret = is_nan(b->cls) ? b : a; break; case float_2nan_prop_x87: /* @@ -XXX,XX +XXX,XX @@ static FloatPartsN *partsN(pick_nan)(FloatPartsN *a, FloatPartsN *b, */ if (is_snan(a->cls)) { if (!is_snan(b->cls)) { - which = is_qnan(b->cls) ? 1 : 0; + ret = is_qnan(b->cls) ? b : a; break; } } else if (is_qnan(a->cls)) { if (is_snan(b->cls) || !is_qnan(b->cls)) { - which = 0; + ret = a; break; } } else { - which = 1; + ret = b; break; } cmp = frac_cmp(a, b); if (cmp == 0) { cmp = a->sign < b->sign; } - which = cmp > 0 ? 0 : 1; + ret = cmp > 0 ? a : b; break; default: g_assert_not_reached(); } - if (which) { - a = b; + if (is_snan(ret->cls)) { + parts_silence_nan(ret, s); } - if (is_snan(a->cls)) { - parts_silence_nan(a, s); - } - return a; + return ret; } static FloatPartsN *partsN(pick_nan_muladd)(FloatPartsN *a, FloatPartsN *b, -- 2.34.1 From: Leif Lindholm <quic_llindhol@quicinc.com> I'm migrating to Qualcomm's new open source email infrastructure, so update my email address, and update the mailmap to match. Signed-off-by: Leif Lindholm <leif.lindholm@oss.qualcomm.com> Reviewed-by: Leif Lindholm <quic_llindhol@quicinc.com> Reviewed-by: Brian Cain <brian.cain@oss.qualcomm.com> Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org> Tested-by: Philippe Mathieu-Daudé <philmd@linaro.org> Message-id: 20241205114047.1125842-1-leif.lindholm@oss.qualcomm.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- MAINTAINERS | 2 +- .mailmap | 5 +++-- 2 files changed, 4 insertions(+), 3 deletions(-) diff --git a/MAINTAINERS b/MAINTAINERS index XXXXXXX..XXXXXXX 100644 --- a/MAINTAINERS +++ b/MAINTAINERS @@ -XXX,XX +XXX,XX @@ F: include/hw/ssi/imx_spi.h SBSA-REF M: Radoslaw Biernacki <rad@semihalf.com> M: Peter Maydell <peter.maydell@linaro.org> -R: Leif Lindholm <quic_llindhol@quicinc.com> +R: Leif Lindholm <leif.lindholm@oss.qualcomm.com> R: Marcin Juszkiewicz <marcin.juszkiewicz@linaro.org> L: qemu-arm@nongnu.org S: Maintained diff --git a/.mailmap b/.mailmap index XXXXXXX..XXXXXXX 100644 --- a/.mailmap +++ b/.mailmap @@ -XXX,XX +XXX,XX @@ Huacai Chen <chenhuacai@kernel.org> <chenhc@lemote.com> Huacai Chen <chenhuacai@kernel.org> <chenhuacai@loongson.cn> James Hogan <jhogan@kernel.org> <james.hogan@imgtec.com> Juan Quintela <quintela@trasno.org> <quintela@redhat.com> -Leif Lindholm <quic_llindhol@quicinc.com> <leif.lindholm@linaro.org> -Leif Lindholm <quic_llindhol@quicinc.com> <leif@nuviainc.com> +Leif Lindholm <leif.lindholm@oss.qualcomm.com> <quic_llindhol@quicinc.com> +Leif Lindholm <leif.lindholm@oss.qualcomm.com> <leif.lindholm@linaro.org> +Leif Lindholm <leif.lindholm@oss.qualcomm.com> <leif@nuviainc.com> Luc Michel <luc@lmichel.fr> <luc.michel@git.antfield.fr> Luc Michel <luc@lmichel.fr> <luc.michel@greensocs.com> Luc Michel <luc@lmichel.fr> <lmichel@kalray.eu> -- 2.34.1 From: Vikram Garhwal <vikram.garhwal@bytedance.com> Previously, maintainer role was paused due to inactive email id. Commit id: c009d715721861984c4987bcc78b7ee183e86d75. Signed-off-by: Vikram Garhwal <vikram.garhwal@bytedance.com> Reviewed-by: Francisco Iglesias <francisco.iglesias@amd.com> Message-id: 20241204184205.12952-1-vikram.garhwal@bytedance.com Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- MAINTAINERS | 2 ++ 1 file changed, 2 insertions(+) diff --git a/MAINTAINERS b/MAINTAINERS index XXXXXXX..XXXXXXX 100644 --- a/MAINTAINERS +++ b/MAINTAINERS @@ -XXX,XX +XXX,XX @@ F: tests/qtest/fuzz-sb16-test.c Xilinx CAN M: Francisco Iglesias <francisco.iglesias@amd.com> +M: Vikram Garhwal <vikram.garhwal@bytedance.com> S: Maintained F: hw/net/can/xlnx-* F: include/hw/net/xlnx-* @@ -XXX,XX +XXX,XX @@ F: include/hw/rx/ CAN bus subsystem and hardware M: Pavel Pisa <pisa@cmp.felk.cvut.cz> M: Francisco Iglesias <francisco.iglesias@amd.com> +M: Vikram Garhwal <vikram.garhwal@bytedance.com> S: Maintained W: https://canbus.pages.fel.cvut.cz/ F: net/can/* -- 2.34.1