Skip to content

Commit 9882188

Browse files
authored
Fix GatherPrefetch and Prefetch APIs (#104494)
1 parent e125386 commit 9882188

File tree

4 files changed

+24
-22
lines changed

4 files changed

+24
-22
lines changed

src/coreclr/jit/hwintrinsicarm64.cpp

+1-1
Original file line numberDiff line numberDiff line change
@@ -2784,7 +2784,7 @@ GenTree* Compiler::impSpecialIntrinsic(NamedIntrinsic intrinsic,
27842784
op4 = getArgForHWIntrinsic(argType, argClass);
27852785

27862786
assert(HWIntrinsicInfo::isImmOp(intrinsic, op4));
2787-
op3 = addRangeCheckIfNeeded(intrinsic, op4, mustExpand, immLowerBound, immUpperBound);
2787+
op4 = addRangeCheckIfNeeded(intrinsic, op4, mustExpand, immLowerBound, immUpperBound);
27882788

27892789
argType = JITtype2varType(strip(info.compCompHnd->getArgType(sig, arg3, &argClass)));
27902790
op3 = getArgForHWIntrinsic(argType, argClass);

src/coreclr/jit/hwintrinsiccodegenarm64.cpp

+5-5
Original file line numberDiff line numberDiff line change
@@ -308,18 +308,18 @@ void CodeGen::genHWIntrinsic(GenTreeHWIntrinsic* node)
308308
emitSize = emitActualTypeSize(intrin.baseType);
309309
opt = INS_OPTS_NONE;
310310
}
311+
else if (HWIntrinsicInfo::IsScalable(intrin.id))
312+
{
313+
emitSize = EA_SCALABLE;
314+
opt = emitter::optGetSveInsOpt(emitTypeSize(intrin.baseType));
315+
}
311316
else if (intrin.category == HW_Category_Special)
312317
{
313318
assert(intrin.id == NI_ArmBase_Yield);
314319

315320
emitSize = EA_UNKNOWN;
316321
opt = INS_OPTS_NONE;
317322
}
318-
else if (HWIntrinsicInfo::IsScalable(intrin.id))
319-
{
320-
emitSize = EA_SCALABLE;
321-
opt = emitter::optGetSveInsOpt(emitTypeSize(intrin.baseType));
322-
}
323323
else
324324
{
325325
emitSize = emitActualTypeSize(Compiler::getSIMDTypeForSize(node->GetSimdSize()));

src/coreclr/jit/hwintrinsiclistarm64sve.h

+8-8
Original file line numberDiff line numberDiff line change
@@ -99,10 +99,10 @@ HARDWARE_INTRINSIC(Sve, FusedMultiplyAddNegated,
9999
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)
100100
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)
101101
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)
102-
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)
103-
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)
104-
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)
105-
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)
102+
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)
103+
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)
104+
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)
105+
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)
106106
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)
107107
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)
108108
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,
192192
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)
193193
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)
194194
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)
195-
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)
196-
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)
197-
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)
198-
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)
195+
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)
196+
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)
197+
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)
198+
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)
199199
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)
200200
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)
201201
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)

src/coreclr/jit/lsraarm64.cpp

+10-8
Original file line numberDiff line numberDiff line change
@@ -1980,18 +1980,10 @@ int LinearScan::BuildHWIntrinsic(GenTreeHWIntrinsic* intrinsicTree, int* pDstCou
19801980
case NI_Sve_LoadVectorNonTemporal:
19811981
case NI_Sve_LoadVector128AndReplicateToVector:
19821982
case NI_Sve_StoreAndZip:
1983-
case NI_Sve_PrefetchBytes:
1984-
case NI_Sve_PrefetchInt16:
1985-
case NI_Sve_PrefetchInt32:
1986-
case NI_Sve_PrefetchInt64:
19871983
assert(intrinsicTree->OperIsMemoryLoadOrStore());
19881984
srcCount += BuildAddrUses(intrin.op2);
19891985
break;
19901986

1991-
case NI_Sve_GatherPrefetch8Bit:
1992-
case NI_Sve_GatherPrefetch16Bit:
1993-
case NI_Sve_GatherPrefetch32Bit:
1994-
case NI_Sve_GatherPrefetch64Bit:
19951987
case NI_Sve_GatherVector:
19961988
case NI_Sve_GatherVectorByteZeroExtend:
19971989
case NI_Sve_GatherVectorInt16SignExtend:
@@ -2004,6 +1996,16 @@ int LinearScan::BuildHWIntrinsic(GenTreeHWIntrinsic* intrinsicTree, int* pDstCou
20041996
case NI_Sve_GatherVectorUInt32WithByteOffsetsZeroExtend:
20051997
case NI_Sve_GatherVectorUInt32ZeroExtend:
20061998
assert(intrinsicTree->OperIsMemoryLoadOrStore());
1999+
FALLTHROUGH;
2000+
2001+
case NI_Sve_PrefetchBytes:
2002+
case NI_Sve_PrefetchInt16:
2003+
case NI_Sve_PrefetchInt32:
2004+
case NI_Sve_PrefetchInt64:
2005+
case NI_Sve_GatherPrefetch8Bit:
2006+
case NI_Sve_GatherPrefetch16Bit:
2007+
case NI_Sve_GatherPrefetch32Bit:
2008+
case NI_Sve_GatherPrefetch64Bit:
20072009
if (!varTypeIsSIMD(intrin.op2->gtType))
20082010
{
20092011
srcCount += BuildAddrUses(intrin.op2);

0 commit comments

Comments
 (0)