1
0
Fork 0
mirror of https://github.com/VSadov/Satori.git synced 2025-06-10 18:11:04 +09:00

Fix GatherPrefetch and Prefetch APIs (#104494)

This commit is contained in:
Kunal Pathak 2024-07-08 20:33:47 -07:00 committed by GitHub
parent e125386a5a
commit 9882188fbe
Signed by: github
GPG key ID: B5690EEEBB952194
4 changed files with 24 additions and 22 deletions

View file

@ -2784,7 +2784,7 @@ GenTree* Compiler::impSpecialIntrinsic(NamedIntrinsic intrinsic,
op4 = getArgForHWIntrinsic(argType, argClass);
assert(HWIntrinsicInfo::isImmOp(intrinsic, op4));
op3 = addRangeCheckIfNeeded(intrinsic, op4, mustExpand, immLowerBound, immUpperBound);
op4 = addRangeCheckIfNeeded(intrinsic, op4, mustExpand, immLowerBound, immUpperBound);
argType = JITtype2varType(strip(info.compCompHnd->getArgType(sig, arg3, &argClass)));
op3 = getArgForHWIntrinsic(argType, argClass);

View file

@ -308,6 +308,11 @@ void CodeGen::genHWIntrinsic(GenTreeHWIntrinsic* node)
emitSize = emitActualTypeSize(intrin.baseType);
opt = INS_OPTS_NONE;
}
else if (HWIntrinsicInfo::IsScalable(intrin.id))
{
emitSize = EA_SCALABLE;
opt = emitter::optGetSveInsOpt(emitTypeSize(intrin.baseType));
}
else if (intrin.category == HW_Category_Special)
{
assert(intrin.id == NI_ArmBase_Yield);
@ -315,11 +320,6 @@ void CodeGen::genHWIntrinsic(GenTreeHWIntrinsic* node)
emitSize = EA_UNKNOWN;
opt = INS_OPTS_NONE;
}
else if (HWIntrinsicInfo::IsScalable(intrin.id))
{
emitSize = EA_SCALABLE;
opt = emitter::optGetSveInsOpt(emitTypeSize(intrin.baseType));
}
else
{
emitSize = emitActualTypeSize(Compiler::getSIMDTypeForSize(node->GetSimdSize()));

View file

@ -99,10 +99,10 @@ HARDWARE_INTRINSIC(Sve, FusedMultiplyAddNegated,
HARDWARE_INTRINSIC(Sve, FusedMultiplySubtract, -1, -1, false, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_sve_fmls, INS_sve_fmls}, HW_Category_SIMD, HW_Flag_Scalable|HW_Flag_EmbeddedMaskedOperation|HW_Flag_HasRMWSemantics|HW_Flag_LowMaskedOperation|HW_Flag_FmaIntrinsic|HW_Flag_SpecialCodeGen)
HARDWARE_INTRINSIC(Sve, FusedMultiplySubtractBySelectedScalar, -1, 4, true, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_sve_fmls, INS_sve_fmls}, HW_Category_SIMDByIndexedElement, HW_Flag_Scalable|HW_Flag_HasImmediateOperand|HW_Flag_HasRMWSemantics|HW_Flag_FmaIntrinsic|HW_Flag_LowVectorOperation)
HARDWARE_INTRINSIC(Sve, FusedMultiplySubtractNegated, -1, -1, false, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_sve_fnmls, INS_sve_fnmls}, HW_Category_SIMD, HW_Flag_Scalable|HW_Flag_EmbeddedMaskedOperation|HW_Flag_HasRMWSemantics|HW_Flag_LowMaskedOperation|HW_Flag_FmaIntrinsic|HW_Flag_SpecialCodeGen)
HARDWARE_INTRINSIC(Sve, GatherPrefetch16Bit, -1, -1, false, {INS_invalid, INS_invalid, INS_sve_prfh, INS_sve_prfh, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_Special, HW_Flag_Scalable|HW_Flag_BaseTypeFromFirstArg|HW_Flag_SpecialCodeGen|HW_Flag_ExplicitMaskedOperation|HW_Flag_LowMaskedOperation|HW_Flag_HasImmediateOperand|HW_Flag_HasEnumOperand)
HARDWARE_INTRINSIC(Sve, GatherPrefetch32Bit, -1, -1, false, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_sve_prfw, INS_sve_prfw, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_Special, HW_Flag_Scalable|HW_Flag_BaseTypeFromFirstArg|HW_Flag_SpecialCodeGen|HW_Flag_ExplicitMaskedOperation|HW_Flag_LowMaskedOperation|HW_Flag_HasImmediateOperand|HW_Flag_HasEnumOperand)
HARDWARE_INTRINSIC(Sve, GatherPrefetch64Bit, -1, -1, false, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_sve_prfd, INS_sve_prfd, INS_invalid, INS_invalid}, HW_Category_Special, HW_Flag_Scalable|HW_Flag_BaseTypeFromFirstArg|HW_Flag_SpecialCodeGen|HW_Flag_ExplicitMaskedOperation|HW_Flag_LowMaskedOperation|HW_Flag_HasImmediateOperand|HW_Flag_HasEnumOperand)
HARDWARE_INTRINSIC(Sve, GatherPrefetch8Bit, -1, -1, false, {INS_sve_prfb, INS_sve_prfb, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_Special, HW_Flag_Scalable|HW_Flag_BaseTypeFromFirstArg|HW_Flag_SpecialCodeGen|HW_Flag_ExplicitMaskedOperation|HW_Flag_LowMaskedOperation|HW_Flag_HasImmediateOperand|HW_Flag_HasEnumOperand)
HARDWARE_INTRINSIC(Sve, GatherPrefetch16Bit, -1, -1, false, {INS_invalid, INS_invalid, INS_sve_prfh, INS_sve_prfh, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_Special, HW_Flag_Scalable|HW_Flag_BaseTypeFromFirstArg|HW_Flag_SpecialCodeGen|HW_Flag_ExplicitMaskedOperation|HW_Flag_LowMaskedOperation|HW_Flag_HasImmediateOperand|HW_Flag_HasEnumOperand|HW_Flag_SpecialSideEffect_Other)
HARDWARE_INTRINSIC(Sve, GatherPrefetch32Bit, -1, -1, false, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_sve_prfw, INS_sve_prfw, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_Special, HW_Flag_Scalable|HW_Flag_BaseTypeFromFirstArg|HW_Flag_SpecialCodeGen|HW_Flag_ExplicitMaskedOperation|HW_Flag_LowMaskedOperation|HW_Flag_HasImmediateOperand|HW_Flag_HasEnumOperand|HW_Flag_SpecialSideEffect_Other)
HARDWARE_INTRINSIC(Sve, GatherPrefetch64Bit, -1, -1, false, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_sve_prfd, INS_sve_prfd, INS_invalid, INS_invalid}, HW_Category_Special, HW_Flag_Scalable|HW_Flag_BaseTypeFromFirstArg|HW_Flag_SpecialCodeGen|HW_Flag_ExplicitMaskedOperation|HW_Flag_LowMaskedOperation|HW_Flag_HasImmediateOperand|HW_Flag_HasEnumOperand|HW_Flag_SpecialSideEffect_Other)
HARDWARE_INTRINSIC(Sve, GatherPrefetch8Bit, -1, -1, false, {INS_sve_prfb, INS_sve_prfb, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_Special, HW_Flag_Scalable|HW_Flag_BaseTypeFromFirstArg|HW_Flag_SpecialCodeGen|HW_Flag_ExplicitMaskedOperation|HW_Flag_LowMaskedOperation|HW_Flag_HasImmediateOperand|HW_Flag_HasEnumOperand|HW_Flag_SpecialSideEffect_Other)
HARDWARE_INTRINSIC(Sve, GatherVector, -1, -1, false, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_sve_ld1w, INS_sve_ld1w, INS_sve_ld1d, INS_sve_ld1d, INS_sve_ld1w, INS_sve_ld1d}, HW_Category_MemoryLoad, HW_Flag_Scalable|HW_Flag_SpecialCodeGen|HW_Flag_ExplicitMaskedOperation|HW_Flag_LowMaskedOperation)
HARDWARE_INTRINSIC(Sve, GatherVectorByteZeroExtend, -1, -1, false, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_sve_ld1b, INS_sve_ld1b, INS_sve_ld1b, INS_sve_ld1b, INS_invalid, INS_invalid}, HW_Category_MemoryLoad, HW_Flag_Scalable|HW_Flag_SpecialCodeGen|HW_Flag_ExplicitMaskedOperation|HW_Flag_LowMaskedOperation)
HARDWARE_INTRINSIC(Sve, GatherVectorInt16SignExtend, -1, -1, false, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_sve_ld1sh, INS_sve_ld1sh, INS_sve_ld1sh, INS_sve_ld1sh, INS_invalid, INS_invalid}, HW_Category_MemoryLoad, HW_Flag_Scalable|HW_Flag_SpecialCodeGen|HW_Flag_ExplicitMaskedOperation|HW_Flag_LowMaskedOperation)
@ -192,10 +192,10 @@ HARDWARE_INTRINSIC(Sve, Not,
HARDWARE_INTRINSIC(Sve, Or, -1, -1, false, {INS_sve_orr, INS_sve_orr, INS_sve_orr, INS_sve_orr, INS_sve_orr, INS_sve_orr, INS_sve_orr, INS_sve_orr, INS_invalid, INS_invalid}, HW_Category_SIMD, HW_Flag_Scalable|HW_Flag_OptionalEmbeddedMaskedOperation|HW_Flag_HasRMWSemantics|HW_Flag_LowMaskedOperation)
HARDWARE_INTRINSIC(Sve, OrAcross, -1, -1, false, {INS_sve_orv, INS_sve_orv, INS_sve_orv, INS_sve_orv, INS_sve_orv, INS_sve_orv, INS_sve_orv, INS_sve_orv, INS_invalid, INS_invalid}, HW_Category_SIMD, HW_Flag_Scalable|HW_Flag_EmbeddedMaskedOperation|HW_Flag_LowMaskedOperation)
HARDWARE_INTRINSIC(Sve, PopCount, -1, -1, false, {INS_sve_cnt, INS_sve_cnt, INS_sve_cnt, INS_sve_cnt, INS_sve_cnt, INS_sve_cnt, INS_sve_cnt, INS_sve_cnt, INS_sve_cnt, INS_sve_cnt}, HW_Category_SIMD, HW_Flag_Scalable|HW_Flag_BaseTypeFromFirstArg|HW_Flag_EmbeddedMaskedOperation|HW_Flag_LowMaskedOperation)
HARDWARE_INTRINSIC(Sve, PrefetchBytes, -1, 3, false, {INS_invalid, INS_sve_prfb, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_Special, HW_Flag_Scalable|HW_Flag_SpecialCodeGen|HW_Flag_ExplicitMaskedOperation|HW_Flag_LowMaskedOperation|HW_Flag_BaseTypeFromFirstArg|HW_Flag_HasImmediateOperand|HW_Flag_HasEnumOperand)
HARDWARE_INTRINSIC(Sve, PrefetchInt16, -1, 3, false, {INS_invalid, INS_invalid, INS_invalid, INS_sve_prfh, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_Special, HW_Flag_Scalable|HW_Flag_SpecialCodeGen|HW_Flag_ExplicitMaskedOperation|HW_Flag_LowMaskedOperation|HW_Flag_BaseTypeFromFirstArg|HW_Flag_HasImmediateOperand|HW_Flag_HasEnumOperand)
HARDWARE_INTRINSIC(Sve, PrefetchInt32, -1, 3, false, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_sve_prfw, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_Special, HW_Flag_Scalable|HW_Flag_SpecialCodeGen|HW_Flag_ExplicitMaskedOperation|HW_Flag_LowMaskedOperation|HW_Flag_BaseTypeFromFirstArg|HW_Flag_HasImmediateOperand|HW_Flag_HasEnumOperand)
HARDWARE_INTRINSIC(Sve, PrefetchInt64, -1, 3, false, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_sve_prfd, INS_invalid, INS_invalid}, HW_Category_Special, HW_Flag_Scalable|HW_Flag_SpecialCodeGen|HW_Flag_ExplicitMaskedOperation|HW_Flag_LowMaskedOperation|HW_Flag_BaseTypeFromFirstArg|HW_Flag_HasImmediateOperand|HW_Flag_HasEnumOperand)
HARDWARE_INTRINSIC(Sve, PrefetchBytes, -1, 3, false, {INS_invalid, INS_sve_prfb, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_Special, HW_Flag_Scalable|HW_Flag_SpecialCodeGen|HW_Flag_ExplicitMaskedOperation|HW_Flag_LowMaskedOperation|HW_Flag_BaseTypeFromFirstArg|HW_Flag_HasImmediateOperand|HW_Flag_HasEnumOperand|HW_Flag_SpecialSideEffect_Other)
HARDWARE_INTRINSIC(Sve, PrefetchInt16, -1, 3, false, {INS_invalid, INS_invalid, INS_invalid, INS_sve_prfh, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_Special, HW_Flag_Scalable|HW_Flag_SpecialCodeGen|HW_Flag_ExplicitMaskedOperation|HW_Flag_LowMaskedOperation|HW_Flag_BaseTypeFromFirstArg|HW_Flag_HasImmediateOperand|HW_Flag_HasEnumOperand|HW_Flag_SpecialSideEffect_Other)
HARDWARE_INTRINSIC(Sve, PrefetchInt32, -1, 3, false, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_sve_prfw, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_Special, HW_Flag_Scalable|HW_Flag_SpecialCodeGen|HW_Flag_ExplicitMaskedOperation|HW_Flag_LowMaskedOperation|HW_Flag_BaseTypeFromFirstArg|HW_Flag_HasImmediateOperand|HW_Flag_HasEnumOperand|HW_Flag_SpecialSideEffect_Other)
HARDWARE_INTRINSIC(Sve, PrefetchInt64, -1, 3, false, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_sve_prfd, INS_invalid, INS_invalid}, HW_Category_Special, HW_Flag_Scalable|HW_Flag_SpecialCodeGen|HW_Flag_ExplicitMaskedOperation|HW_Flag_LowMaskedOperation|HW_Flag_BaseTypeFromFirstArg|HW_Flag_HasImmediateOperand|HW_Flag_HasEnumOperand|HW_Flag_SpecialSideEffect_Other)
HARDWARE_INTRINSIC(Sve, ReciprocalEstimate, -1, 1, true, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_sve_frecpe, INS_sve_frecpe}, HW_Category_SIMD, HW_Flag_Scalable)
HARDWARE_INTRINSIC(Sve, ReciprocalExponent, -1, -1, false, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_sve_frecpx, INS_sve_frecpx}, HW_Category_SIMD, HW_Flag_Scalable|HW_Flag_EmbeddedMaskedOperation|HW_Flag_LowMaskedOperation)
HARDWARE_INTRINSIC(Sve, ReciprocalSqrtEstimate, -1, 1, true, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_sve_frsqrte, INS_sve_frsqrte}, HW_Category_SIMD, HW_Flag_Scalable)

View file

@ -1980,18 +1980,10 @@ int LinearScan::BuildHWIntrinsic(GenTreeHWIntrinsic* intrinsicTree, int* pDstCou
case NI_Sve_LoadVectorNonTemporal:
case NI_Sve_LoadVector128AndReplicateToVector:
case NI_Sve_StoreAndZip:
case NI_Sve_PrefetchBytes:
case NI_Sve_PrefetchInt16:
case NI_Sve_PrefetchInt32:
case NI_Sve_PrefetchInt64:
assert(intrinsicTree->OperIsMemoryLoadOrStore());
srcCount += BuildAddrUses(intrin.op2);
break;
case NI_Sve_GatherPrefetch8Bit:
case NI_Sve_GatherPrefetch16Bit:
case NI_Sve_GatherPrefetch32Bit:
case NI_Sve_GatherPrefetch64Bit:
case NI_Sve_GatherVector:
case NI_Sve_GatherVectorByteZeroExtend:
case NI_Sve_GatherVectorInt16SignExtend:
@ -2004,6 +1996,16 @@ int LinearScan::BuildHWIntrinsic(GenTreeHWIntrinsic* intrinsicTree, int* pDstCou
case NI_Sve_GatherVectorUInt32WithByteOffsetsZeroExtend:
case NI_Sve_GatherVectorUInt32ZeroExtend:
assert(intrinsicTree->OperIsMemoryLoadOrStore());
FALLTHROUGH;
case NI_Sve_PrefetchBytes:
case NI_Sve_PrefetchInt16:
case NI_Sve_PrefetchInt32:
case NI_Sve_PrefetchInt64:
case NI_Sve_GatherPrefetch8Bit:
case NI_Sve_GatherPrefetch16Bit:
case NI_Sve_GatherPrefetch32Bit:
case NI_Sve_GatherPrefetch64Bit:
if (!varTypeIsSIMD(intrin.op2->gtType))
{
srcCount += BuildAddrUses(intrin.op2);