Skip to content

Add AvxIfma + AvxIfma.V512 hardware intrinsics #130079

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
jamesburton wants to merge 7 commits into
base: main
Choose a base branch
from
Open

Add AvxIfma + AvxIfma.V512 hardware intrinsics #130079

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
a584a6f
Add AvxIfma + AvxIfma.V512 hardware intrinsics (structural)
jamesburton Jul 1, 2026
e99f064
Add AvxIfma/AvxIfma.V512 tests + NativeAOT smoke wiring
jamesburton Jul 1, 2026
add8927
Fix hwintrinsiclistxarch.h ordering — binarySearchId requires alphabe…
jamesburton Jul 1, 2026
fc7299f
Add AvxIfma type lookups for X64_AVXIFMA / X86_AVXIFMA cases
jamesburton Jul 1, 2026
ff40b90
Set 'AvxIfma' as the ManagedName so ILC can look it up
jamesburton Jul 1, 2026
6896fd6
Merge remote-tracking branch 'upstream/main' into feature/avxifma
jamesburton Jul 3, 2026
52cbd74
Fix simdSize for AVX512v2 MultiplyAdd52High/Low fallback entries
jamesburton Jul 4, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 4 additions & 0 deletions src/coreclr/jit/hwintrinsiccodegenxarch.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -927,6 +927,10 @@ void CodeGen::genHWIntrinsic(GenTreeHWIntrinsic* node)
case NI_AVXVNNI_MultiplyWideningAndAddSaturate:
case NI_AVX512v3_MultiplyWideningAndAdd:
case NI_AVX512v3_MultiplyWideningAndAddSaturate:
case NI_AVXIFMA_MultiplyAdd52Low:
case NI_AVXIFMA_MultiplyAdd52High:
case NI_AVX512v2_MultiplyAdd52Low:
case NI_AVX512v2_MultiplyAdd52High:
case NI_AVX512BMM_BitMultiplyMatrix16x16WithOrReduction:
case NI_AVX512BMM_BitMultiplyMatrix16x16WithXorReduction:
{
Expand Down
12 changes: 12 additions & 0 deletions src/coreclr/jit/hwintrinsiclistxarch.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -634,6 +634,8 @@ HARDWARE_INTRINSIC(AVX512_X64, ConvertToUInt64WithTruncation,
// Intrinsics for AVX512-IFMA, AVX512-VBMI
#define FIRST_NI_AVX512v2 NI_AVX512v2_MultiShift
HARDWARE_INTRINSIC(AVX512v2, MultiShift, -1, 2, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_vpmultishiftqb, INS_vpmultishiftqb, INS_invalid, INS_invalid, 3, -1, HW_Category_SimpleSIMD, HW_Flag_BaseTypeFromSecondArg)
HARDWARE_INTRINSIC(AVX512v2, MultiplyAdd52High, -1, 3, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_vpmadd52huq, INS_invalid, INS_invalid, 5, -1, HW_Category_SimpleSIMD, HW_Flag_NoFlag)
HARDWARE_INTRINSIC(AVX512v2, MultiplyAdd52Low, -1, 3, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_vpmadd52luq, INS_invalid, INS_invalid, 5, -1, HW_Category_SimpleSIMD, HW_Flag_NoFlag)
HARDWARE_INTRINSIC(AVX512v2, PermuteVar16x8, 16, 2, INS_vpermb, INS_vpermb, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, 3, -1, HW_Category_SimpleSIMD, HW_Flag_SpecialImport)
HARDWARE_INTRINSIC(AVX512v2, PermuteVar16x8x2, 16, 3, INS_vpermt2b, INS_vpermt2b, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, 4, -1, HW_Category_SimpleSIMD, HW_Flag_SpecialCodeGen|HW_Flag_PermuteVar2x|HW_Flag_RmwIntrinsic)
HARDWARE_INTRINSIC(AVX512v2, PermuteVar32x8, 32, 2, INS_vpermb, INS_vpermb, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, 3, -1, HW_Category_SimpleSIMD, HW_Flag_SpecialImport)
Expand Down Expand Up @@ -709,6 +711,16 @@ HARDWARE_INTRINSIC(AVXVNNI, MultiplyWideningAndAddSaturate,
// ISA Function name SIMD size NumArg Instructions IntCost FltCost Category Flags
// TYP_BYTE, TYP_UBYTE, TYP_SHORT, TYP_USHORT, TYP_INT, TYP_UINT, TYP_LONG, TYP_ULONG, TYP_FLOAT, TYP_DOUBLE
// ***************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************
// Intrinsics for AVX-IFMA (VEX-encoded VPMADD52*UQ on V128 / V256)
#define FIRST_NI_AVXIFMA NI_AVXIFMA_MultiplyAdd52High
HARDWARE_INTRINSIC(AVXIFMA, MultiplyAdd52High, -1, 3, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_vpmadd52huq, INS_invalid, INS_invalid, 5, -1, HW_Category_SimpleSIMD, HW_Flag_NoEvexSemantics)
HARDWARE_INTRINSIC(AVXIFMA, MultiplyAdd52Low, -1, 3, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_vpmadd52luq, INS_invalid, INS_invalid, 5, -1, HW_Category_SimpleSIMD, HW_Flag_NoEvexSemantics)
#define LAST_NI_AVXIFMA NI_AVXIFMA_MultiplyAdd52Low

// ***************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************
// ISA Function name SIMD size NumArg Instructions IntCost FltCost Category Flags
// {TYP_BYTE, TYP_UBYTE, TYP_SHORT, TYP_USHORT, TYP_INT, TYP_UINT, TYP_LONG, TYP_ULONG, TYP_FLOAT, TYP_DOUBLE}
// ***************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************
// AVXVNNIINT Intrinsics
#define FIRST_NI_AVXVNNIINT NI_AVXVNNIINT_MultiplyWideningAndAdd
HARDWARE_INTRINSIC(AVXVNNIINT, MultiplyWideningAndAdd, -1, 3, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, 5, -1, HW_Category_SimpleSIMD, HW_Flag_BaseTypeFromSecondArg|HW_Flag_SpecialCodeGen|HW_Flag_SpecialImport)
Expand Down
17 changes: 16 additions & 1 deletion src/coreclr/jit/hwintrinsicxarch.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -166,6 +166,14 @@ static CORINFO_InstructionSet V512VersionOfIsa(CORINFO_InstructionSet isa)
return InstructionSet_AVX512v3;
}

case InstructionSet_AVXIFMA:
case InstructionSet_AVX512v2:
{
// AvxIfma.V512 lifts under AVX512v2, which carries the EVEX-encoded
// VPMADD52*UQ on ZMM. Same rationale as AvxVnni.V512 above.
return InstructionSet_AVX512v2;
}

default:
{
return InstructionSet_NONE;
Expand Down Expand Up @@ -276,7 +284,14 @@ CORINFO_InstructionSet Compiler::lookupInstructionSet(const char* className)
}
else if (strcmp(className + 3, "Ifma") == 0)
{
return InstructionSet_AVXIFMA;
if (compSupportsHWIntrinsic(InstructionSet_AVXIFMA))
{
return InstructionSet_AVXIFMA;
}
else
{
return InstructionSet_AVX512v2;
}
}
else if (strncmp(className + 3, "Vnni", 4) == 0)
{
Expand Down
3 changes: 3 additions & 0 deletions src/coreclr/jit/lowerxarch.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10175,6 +10175,9 @@ void Lowering::ContainCheckHWIntrinsic(GenTreeHWIntrinsic* node)
assert((intrinsicId == NI_X86Base_DivRem) || (intrinsicId == NI_X86Base_X64_DivRem) ||
(intrinsicId == NI_AVX512v3_MultiplyWideningAndAdd) ||
(intrinsicId == NI_AVX512v3_MultiplyWideningAndAddSaturate) ||
(intrinsicId == NI_AVX512v2_MultiplyAdd52Low) ||
(intrinsicId == NI_AVX512v2_MultiplyAdd52High) ||
(intrinsicId >= FIRST_NI_AVXIFMA && intrinsicId <= LAST_NI_AVXIFMA) ||
(intrinsicId >= FIRST_NI_AVXVNNI && intrinsicId <= LAST_NI_AVXVNNIINT_V512));
TryMakeSrcContainedOrRegOptional(node, op3);
break;
Expand Down
4 changes: 4 additions & 0 deletions src/coreclr/jit/lsraxarch.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2734,6 +2734,10 @@ int LinearScan::BuildHWIntrinsic(GenTreeHWIntrinsic* intrinsicTree, int* pDstCou
case NI_AVXVNNI_MultiplyWideningAndAddSaturate:
case NI_AVX512v3_MultiplyWideningAndAdd:
case NI_AVX512v3_MultiplyWideningAndAddSaturate:
case NI_AVXIFMA_MultiplyAdd52Low:
case NI_AVXIFMA_MultiplyAdd52High:
case NI_AVX512v2_MultiplyAdd52Low:
case NI_AVX512v2_MultiplyAdd52High:
case NI_AVXVNNIINT_MultiplyWideningAndAdd:
case NI_AVXVNNIINT_MultiplyWideningAndAddSaturate:
case NI_AVXVNNIINT_V512_MultiplyWideningAndAdd:
Expand Down
1 change: 1 addition & 0 deletions src/coreclr/tools/Common/InstructionSetHelpers.cs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -291,6 +291,7 @@ public static InstructionSetSupport ConfigureInstructionSetSupport(string instru
optimisticInstructionSetSupportBuilder.AddSupportedInstructionSet("avx10v2");
optimisticInstructionSetSupportBuilder.AddSupportedInstructionSet("avxvnniint_v512");
optimisticInstructionSetSupportBuilder.AddSupportedInstructionSet("avxvnni_v512");
optimisticInstructionSetSupportBuilder.AddSupportedInstructionSet("avxifma_v512");
optimisticInstructionSetSupportBuilder.AddSupportedInstructionSet("avx512vp2intersect");
optimisticInstructionSetSupportBuilder.AddSupportedInstructionSet("aes_v512");
optimisticInstructionSetSupportBuilder.AddSupportedInstructionSet("gfni_v512");
Expand Down
Loading
Loading