Skip to content

Commit

Permalink
Avx512 arithmetic (#84429)
Browse files Browse the repository at this point in the history
* Add and Subtract support for AVX512

* Removing absovled condition

* fixing formatting issues

* Addressing review comments,
Using non saturation version for add.

* Update src/coreclr/jit/hwintrinsiclistxarch.h

Co-authored-by: Tanner Gooding <tagoo@outlook.com>

---------

Co-authored-by: Tanner Gooding <tagoo@outlook.com>
  • Loading branch information
jkrishnavs and tannergooding committed Apr 9, 2023
1 parent 0fc78e6 commit 1411364
Show file tree
Hide file tree
Showing 3 changed files with 39 additions and 1 deletion.
26 changes: 26 additions & 0 deletions src/coreclr/jit/gentree.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -19487,6 +19487,19 @@ GenTree* Compiler::gtNewSimdBinOpNode(genTreeOps op,
intrinsic = NI_AVX2_Add;
}
}
else if (simdSize == 64)
{
if (varTypeIsSmall(simdBaseType))
{
assert(compIsaSupportedDebugOnly(InstructionSet_AVX512BW));
intrinsic = NI_AVX512BW_Add;
}
else
{
assert(compIsaSupportedDebugOnly(InstructionSet_AVX512F));
intrinsic = NI_AVX512F_Add;
}
}
else if (simdBaseType == TYP_FLOAT)
{
intrinsic = NI_SSE_Add;
Expand Down Expand Up @@ -19862,6 +19875,19 @@ GenTree* Compiler::gtNewSimdBinOpNode(genTreeOps op,
intrinsic = NI_AVX2_Subtract;
}
}
else if (simdSize == 64)
{
if (varTypeIsSmall(simdBaseType))
{
assert(compIsaSupportedDebugOnly(InstructionSet_AVX512BW));
intrinsic = NI_AVX512BW_Subtract;
}
else
{
assert(compIsaSupportedDebugOnly(InstructionSet_AVX512F));
intrinsic = NI_AVX512F_Subtract;
}
}
else if (simdBaseType == TYP_FLOAT)
{
intrinsic = NI_SSE_Subtract;
Expand Down
10 changes: 9 additions & 1 deletion src/coreclr/jit/hwintrinsiclistxarch.h
Original file line number Diff line number Diff line change
Expand Up @@ -241,6 +241,7 @@ HARDWARE_INTRINSIC(Vector256, Xor,
// {TYP_BYTE, TYP_UBYTE, TYP_SHORT, TYP_USHORT, TYP_INT, TYP_UINT, TYP_LONG, TYP_ULONG, TYP_FLOAT, TYP_DOUBLE}
// ***************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************
// Vector512 Intrinsics
HARDWARE_INTRINSIC(Vector512, Add, 64, 2, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_Helper, HW_Flag_SpecialImport|HW_Flag_NoCodeGen)
HARDWARE_INTRINSIC(Vector512, AndNot, 64, 2, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_Helper, HW_Flag_SpecialImport|HW_Flag_NoCodeGen)
HARDWARE_INTRINSIC(Vector512, BitwiseAnd, 64, 2, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_Helper, HW_Flag_SpecialImport|HW_Flag_NoCodeGen)
HARDWARE_INTRINSIC(Vector512, BitwiseOr, 64, 2, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_Helper, HW_Flag_SpecialImport|HW_Flag_NoCodeGen)
Expand Down Expand Up @@ -274,14 +275,17 @@ HARDWARE_INTRINSIC(Vector512, LoadAligned,
HARDWARE_INTRINSIC(Vector512, LoadAlignedNonTemporal, 64, 1, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_Helper, HW_Flag_SpecialImport|HW_Flag_NoCodeGen)
HARDWARE_INTRINSIC(Vector512, LoadUnsafe, 64, -1, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_Helper, HW_Flag_SpecialImport|HW_Flag_NoCodeGen)
HARDWARE_INTRINSIC(Vector512, OnesComplement, 64, 1, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_Helper, HW_Flag_SpecialImport|HW_Flag_NoCodeGen)
HARDWARE_INTRINSIC(Vector512, op_Addition, 64, 2, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_Helper, HW_Flag_SpecialImport|HW_Flag_NoCodeGen)
HARDWARE_INTRINSIC(Vector512, op_BitwiseAnd, 64, 2, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_Helper, HW_Flag_SpecialImport|HW_Flag_NoCodeGen|HW_Flag_Commutative)
HARDWARE_INTRINSIC(Vector512, op_BitwiseOr, 64, 2, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_Helper, HW_Flag_SpecialImport|HW_Flag_NoCodeGen|HW_Flag_Commutative)
HARDWARE_INTRINSIC(Vector512, op_ExclusiveOr, 64, 2, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_Helper, HW_Flag_SpecialImport|HW_Flag_NoCodeGen)
HARDWARE_INTRINSIC(Vector512, op_OnesComplement, 64, 1, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_Helper, HW_Flag_SpecialImport|HW_Flag_NoCodeGen)
HARDWARE_INTRINSIC(Vector512, op_Subtraction, 64, 2, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_Helper, HW_Flag_SpecialImport|HW_Flag_NoCodeGen)
HARDWARE_INTRINSIC(Vector512, Store, 64, 2, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_Helper, HW_Flag_SpecialImport|HW_Flag_BaseTypeFromFirstArg|HW_Flag_NoCodeGen)
HARDWARE_INTRINSIC(Vector512, StoreAligned, 64, 2, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_Helper, HW_Flag_SpecialImport|HW_Flag_BaseTypeFromFirstArg|HW_Flag_NoCodeGen)
HARDWARE_INTRINSIC(Vector512, StoreAlignedNonTemporal, 64, 2, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_Helper, HW_Flag_SpecialImport|HW_Flag_BaseTypeFromFirstArg|HW_Flag_NoCodeGen)
HARDWARE_INTRINSIC(Vector512, StoreUnsafe, 64, -1, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_Helper, HW_Flag_SpecialImport|HW_Flag_BaseTypeFromFirstArg|HW_Flag_NoCodeGen)
HARDWARE_INTRINSIC(Vector512, Subtract, 64, 2, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_Helper, HW_Flag_SpecialImport|HW_Flag_NoCodeGen)
HARDWARE_INTRINSIC(Vector512, WithLower, 64, 2, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_Helper, HW_Flag_SpecialImport|HW_Flag_NoCodeGen)
HARDWARE_INTRINSIC(Vector512, WithUpper, 64, 2, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_Helper, HW_Flag_SpecialImport|HW_Flag_NoCodeGen)
HARDWARE_INTRINSIC(Vector512, Xor, 64, 2, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_Helper, HW_Flag_SpecialImport|HW_Flag_NoCodeGen)
Expand Down Expand Up @@ -775,6 +779,7 @@ HARDWARE_INTRINSIC(AVX2, Xor,
// {TYP_BYTE, TYP_UBYTE, TYP_SHORT, TYP_USHORT, TYP_INT, TYP_UINT, TYP_LONG, TYP_ULONG, TYP_FLOAT, TYP_DOUBLE}
// ***************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************
// AVX512F Intrinsics
HARDWARE_INTRINSIC(AVX512F, Add, 64, 2, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_paddd, INS_paddd, INS_paddq, INS_paddq, INS_addps, INS_addpd}, HW_Category_SimpleSIMD, HW_Flag_Commutative)
HARDWARE_INTRINSIC(AVX512F, And, 64, 2, {INS_pand, INS_pand, INS_pand, INS_pand, INS_pand, INS_pand, INS_vpandq, INS_vpandq, INS_andps, INS_andpd}, HW_Category_SimpleSIMD, HW_Flag_Commutative)
HARDWARE_INTRINSIC(AVX512F, AndNot, 64, 2, {INS_pandn, INS_pandn, INS_pandn, INS_pandn, INS_pandn, INS_pandn, INS_vpandnq, INS_vpandnq, INS_andnps, INS_andnpd}, HW_Category_SimpleSIMD, HW_Flag_NoFlag)
HARDWARE_INTRINSIC(AVX512F, BroadcastScalarToVector512, 64, 1, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_vpbroadcastd, INS_vpbroadcastd, INS_vpbroadcastq, INS_vpbroadcastq, INS_vbroadcastss, INS_vbroadcastsd}, HW_Category_SIMDScalar, HW_Flag_MaybeMemoryLoad)
Expand All @@ -787,14 +792,17 @@ HARDWARE_INTRINSIC(AVX512F, Or,
HARDWARE_INTRINSIC(AVX512F, Store, 64, 2, {INS_movdqu, INS_movdqu, INS_movdqu, INS_movdqu, INS_movdqu, INS_movdqu, INS_vmovdqu64, INS_vmovdqu64, INS_movups, INS_movupd}, HW_Category_Helper, HW_Flag_SpecialImport|HW_Flag_BaseTypeFromSecondArg|HW_Flag_NoCodeGen)
HARDWARE_INTRINSIC(AVX512F, StoreAligned, 64, 2, {INS_movdqa, INS_movdqa, INS_movdqa, INS_movdqa, INS_movdqa, INS_movdqa, INS_vmovdqa64, INS_vmovdqa64, INS_movaps, INS_movapd}, HW_Category_MemoryStore, HW_Flag_NoRMWSemantics|HW_Flag_BaseTypeFromSecondArg)
HARDWARE_INTRINSIC(AVX512F, StoreAlignedNonTemporal, 64, 2, {INS_movntdq, INS_movntdq, INS_movntdq, INS_movntdq, INS_movntdq, INS_movntdq, INS_movntdq, INS_movntdq, INS_movntps, INS_movntpd}, HW_Category_MemoryStore, HW_Flag_NoRMWSemantics|HW_Flag_BaseTypeFromSecondArg)
HARDWARE_INTRINSIC(AVX512F, Subtract, 64, 2, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_psubd, INS_psubd, INS_psubq, INS_psubq, INS_subps, INS_subpd}, HW_Category_SimpleSIMD, HW_Flag_NoFlag)
HARDWARE_INTRINSIC(AVX512F, Xor, 64, 2, {INS_pxor, INS_pxor, INS_pxor, INS_pxor, INS_pxor, INS_pxor, INS_vpxorq, INS_vpxorq, INS_xorps, INS_xorpd}, HW_Category_SimpleSIMD, HW_Flag_Commutative)

// ***************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************
// ISA Function name SIMD size NumArg Instructions Category Flags
// {TYP_BYTE, TYP_UBYTE, TYP_SHORT, TYP_USHORT, TYP_INT, TYP_UINT, TYP_LONG, TYP_ULONG, TYP_FLOAT, TYP_DOUBLE}
// ***************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************
// AVX512F Intrinsics
// AVX512BW Intrinsics
HARDWARE_INTRINSIC(AVX512BW, Add, 64, 2, {INS_paddb, INS_paddb, INS_paddw, INS_paddw, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_SimpleSIMD, HW_Flag_Commutative)
HARDWARE_INTRINSIC(AVX512BW, BroadcastScalarToVector512, 64, 1, {INS_vpbroadcastb, INS_vpbroadcastb, INS_vpbroadcastw, INS_vpbroadcastw, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_SIMDScalar, HW_Flag_MaybeMemoryLoad)
HARDWARE_INTRINSIC(AVX512BW, Subtract, 64, 2, {INS_psubb, INS_psubb, INS_psubw, INS_psubw, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_SimpleSIMD, HW_Flag_NoFlag)

// ***************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************
// ISA Function name SIMD size NumArg Instructions Category Flags
Expand Down
4 changes: 4 additions & 0 deletions src/coreclr/jit/hwintrinsicxarch.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -619,8 +619,10 @@ GenTree* Compiler::impSpecialIntrinsic(NamedIntrinsic intrinsic,

case NI_Vector128_Add:
case NI_Vector256_Add:
case NI_Vector512_Add:
case NI_Vector128_op_Addition:
case NI_Vector256_op_Addition:
case NI_Vector512_op_Addition:
{
assert(sig->numArgs == 2);

Expand Down Expand Up @@ -2054,8 +2056,10 @@ GenTree* Compiler::impSpecialIntrinsic(NamedIntrinsic intrinsic,

case NI_Vector128_Subtract:
case NI_Vector256_Subtract:
case NI_Vector512_Subtract:
case NI_Vector128_op_Subtraction:
case NI_Vector256_op_Subtraction:
case NI_Vector512_op_Subtraction:
{
assert(sig->numArgs == 2);

Expand Down

0 comments on commit 1411364

Please sign in to comment.