Skip to content

[X86] Remove AMX-TRANSPOSE #165556

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

Merged
merged 9 commits into from
Oct 31, 2025
Merged

[X86] Remove AMX-TRANSPOSE #165556

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
12269e7
Remove amx_transpose
mikolaj-pirog Oct 29, 2025
4e86b45
Delete unused variable
mikolaj-pirog Oct 29, 2025
ef75547
Formatting
mikolaj-pirog Oct 29, 2025
6510899
Remove unused variable
mikolaj-pirog Oct 30, 2025
88d58b2
Reviewer suggestions
mikolaj-pirog Oct 30, 2025
cbede57
Formatting
mikolaj-pirog Oct 30, 2025
502ad98
Reviewer suggestions
mikolaj-pirog Oct 30, 2025
d3b0663
Fix typo
mikolaj-pirog Oct 30, 2025
e21b7a7
Remove unnecessary variable
mikolaj-pirog Oct 31, 2025
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
89 changes: 0 additions & 89 deletions clang/include/clang/Basic/BuiltinsX86_64.td
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -239,57 +239,6 @@ let Features = "amx-complex", Attributes = [NoThrow] in {
def tcmmrlfp16ps_internal : X86Builtin<"_Vector<256, int>(unsigned short, unsigned short, unsigned short, _Vector<256, int>, _Vector<256, int>, _Vector<256, int>)">;
}

let Features = "amx-transpose", Attributes = [NoThrow] in {
def t2rpntlvwz0_internal : X86Builtin<"void(unsigned short, unsigned short, unsigned short, _Vector<256, int *>, _Vector<256, int *>, void const *, size_t)">;
}

let Features = "amx-movrs,amx-transpose", Attributes = [NoThrow] in {
def t2rpntlvwz0rs_internal : X86Builtin<"void(unsigned short, unsigned short, unsigned short, _Vector<256, int *>, _Vector<256, int *>, void const *, size_t)">;
}

let Features = "amx-transpose", Attributes = [NoThrow] in {
def t2rpntlvwz0t1_internal : X86Builtin<"void(unsigned short, unsigned short, unsigned short, _Vector<256, int *>, _Vector<256, int *>, void const *, size_t)">;
}

let Features = "amx-movrs,amx-transpose", Attributes = [NoThrow] in {
def t2rpntlvwz0rst1_internal : X86Builtin<"void(unsigned short, unsigned short, unsigned short, _Vector<256, int *>, _Vector<256, int *>, void const *, size_t)">;
}

let Features = "amx-transpose", Attributes = [NoThrow] in {
def t2rpntlvwz1_internal : X86Builtin<"void(unsigned short, unsigned short, unsigned short, _Vector<256, int *>, _Vector<256, int *>, void const *, size_t)">;
}

let Features = "amx-movrs,amx-transpose", Attributes = [NoThrow] in {
def t2rpntlvwz1rs_internal : X86Builtin<"void(unsigned short, unsigned short, unsigned short, _Vector<256, int *>, _Vector<256, int *>, void const *, size_t)">;
}

let Features = "amx-transpose", Attributes = [NoThrow] in {
def t2rpntlvwz1t1_internal : X86Builtin<"void(unsigned short, unsigned short, unsigned short, _Vector<256, int *>, _Vector<256, int *>, void const *, size_t)">;
}

let Features = "amx-movrs,amx-transpose", Attributes = [NoThrow] in {
def t2rpntlvwz1rst1_internal : X86Builtin<"void(unsigned short, unsigned short, unsigned short, _Vector<256, int *>, _Vector<256, int *>, void const *, size_t)">;
}

let Features = "amx-transpose", Attributes = [NoThrow] in {
def ttransposed_internal : X86Builtin<"_Vector<256, int>(unsigned short, unsigned short, _Vector<256, int>)">;
}

let Features = "amx-bf16,amx-transpose", Attributes = [NoThrow] in {
def ttdpbf16ps_internal : X86Builtin<"_Vector<256, int>(unsigned short, unsigned short, unsigned short, _Vector<256, int>, _Vector<256, int>, _Vector<256, int>)">;
}

let Features = "amx-fp16,amx-transpose", Attributes = [NoThrow] in {
def ttdpfp16ps_internal : X86Builtin<"_Vector<256, int>(unsigned short, unsigned short, unsigned short, _Vector<256, int>, _Vector<256, int>, _Vector<256, int>)">;
}

let Features = "amx-complex,amx-transpose", Attributes = [NoThrow] in {
def ttcmmimfp16ps_internal : X86Builtin<"_Vector<256, int>(unsigned short, unsigned short, unsigned short, _Vector<256, int>, _Vector<256, int>, _Vector<256, int>)">;
def ttcmmrlfp16ps_internal : X86Builtin<"_Vector<256, int>(unsigned short, unsigned short, unsigned short, _Vector<256, int>, _Vector<256, int>, _Vector<256, int>)">;
def tconjtcmmimfp16ps_internal : X86Builtin<"_Vector<256, int>(unsigned short, unsigned short, unsigned short, _Vector<256, int>, _Vector<256, int>, _Vector<256, int>)">;
def tconjtfp16_internal : X86Builtin<"_Vector<256, int>(unsigned short, unsigned short, _Vector<256, int>)">;
}

let Features = "amx-avx512,avx10.2", Attributes = [NoThrow] in {
def tcvtrowd2ps_internal : X86Builtin<"_Vector<16, float>(unsigned short, unsigned short, _Vector<256, int>, unsigned int)">;
def tcvtrowps2bf16h_internal : X86Builtin<"_Vector<32, __bf16>(unsigned short, unsigned short, _Vector<256, int>, unsigned int)">;
Expand All @@ -303,10 +252,6 @@ let Features = "amx-tf32", Attributes = [NoThrow] in {
def tmmultf32ps_internal : X86Builtin<"_Vector<256, int>(unsigned short, unsigned short, unsigned short, _Vector<256, int>, _Vector<256, int>, _Vector<256, int>)">;
}

let Features = "amx-tf32,amx-transpose", Attributes = [NoThrow] in {
def ttmmultf32ps_internal : X86Builtin<"_Vector<256, int>(unsigned short, unsigned short, unsigned short, _Vector<256, int>, _Vector<256, int>, _Vector<256, int>)">;
}

let Features = "amx-fp8", Attributes = [NoThrow] in {
def tdpbf8ps_internal : X86Builtin<"_Vector<256, int>(unsigned short, unsigned short, unsigned short, _Vector<256, int>, _Vector<256, int>, _Vector<256, int>)">;
def tdpbhf8ps_internal : X86Builtin<"_Vector<256, int>(unsigned short, unsigned short, unsigned short, _Vector<256, int>, _Vector<256, int>, _Vector<256, int>)">;
Expand All @@ -321,13 +266,6 @@ let Features = "amx-tile", Attributes = [NoThrow] in {
def tilezero : X86Builtin<"void(unsigned char)">;
}

let Features = "amx-movrs,amx-transpose", Attributes = [NoThrow] in {
def t2rpntlvwz0rs : X86Builtin<"void(_Constant unsigned char, void const *, size_t)">;
def t2rpntlvwz0rst1 : X86Builtin<"void(_Constant unsigned char, void const *, size_t)">;
def t2rpntlvwz1rs : X86Builtin<"void(_Constant unsigned char, void const *, size_t)">;
def t2rpntlvwz1rst1 : X86Builtin<"void(_Constant unsigned char, void const *, size_t)">;
}

let Features = "amx-movrs", Attributes = [NoThrow] in {
def tileloaddrs64 : X86Builtin<"void(_Constant unsigned char, void const *, size_t)">;
def tileloaddrst164 : X86Builtin<"void(_Constant unsigned char, void const *, size_t)">;
Expand Down Expand Up @@ -359,29 +297,6 @@ let Features = "amx-complex", Attributes = [NoThrow] in {
def tcmmrlfp16ps : X86Builtin<"void(_Constant unsigned char, _Constant unsigned char, _Constant unsigned char)">;
}

let Features = "amx-transpose", Attributes = [NoThrow] in {
def t2rpntlvwz0 : X86Builtin<"void(_Constant unsigned char, void const *, size_t)">;
def t2rpntlvwz0t1 : X86Builtin<"void(_Constant unsigned char, void const *, size_t)">;
def t2rpntlvwz1 : X86Builtin<"void(_Constant unsigned char, void const *, size_t)">;
def t2rpntlvwz1t1 : X86Builtin<"void(_Constant unsigned char, void const *, size_t)">;
def ttransposed : X86Builtin<"void(_Constant unsigned char, _Constant unsigned char)">;
}

let Features = "amx-bf16,amx-transpose", Attributes = [NoThrow] in {
def ttdpbf16ps : X86Builtin<"void(_Constant unsigned char, _Constant unsigned char, _Constant unsigned char)">;
}

let Features = "amx-fp16,amx-transpose", Attributes = [NoThrow] in {
def ttdpfp16ps : X86Builtin<"void(_Constant unsigned char, _Constant unsigned char, _Constant unsigned char)">;
}

let Features = "amx-complex,amx-transpose", Attributes = [NoThrow] in {
def ttcmmimfp16ps : X86Builtin<"void(_Constant unsigned char, _Constant unsigned char, _Constant unsigned char)">;
def ttcmmrlfp16ps : X86Builtin<"void(_Constant unsigned char, _Constant unsigned char, _Constant unsigned char)">;
def tconjtcmmimfp16ps : X86Builtin<"void(_Constant unsigned char, _Constant unsigned char, _Constant unsigned char)">;
def tconjtfp16 : X86Builtin<"void(_Constant unsigned char, _Constant unsigned char)">;
}

let Features = "amx-avx512,avx10.2", Attributes = [NoThrow] in {
def tcvtrowd2ps : X86Builtin<"_Vector<16, float>(_Constant unsigned char, unsigned int)">;
def tcvtrowps2bf16h : X86Builtin<"_Vector<32, __bf16>(_Constant unsigned char, unsigned int)">;
Expand All @@ -406,10 +321,6 @@ let Features = "amx-tf32", Attributes = [NoThrow] in {
def tmmultf32ps : X86Builtin<"void(_Constant unsigned char, _Constant unsigned char, _Constant unsigned char)">;
}

let Features = "amx-tf32,amx-transpose", Attributes = [NoThrow] in {
def ttmmultf32ps : X86Builtin<"void(_Constant unsigned char, _Constant unsigned char, _Constant unsigned char)">;
}

let Features = "prefetchi", Attributes = [NoThrow, Const] in {
def prefetchi : X86Builtin<"void(void const *, unsigned int)">;
}
Expand Down
2 changes: 0 additions & 2 deletions clang/include/clang/Driver/Options.td
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6695,8 +6695,6 @@ def mamx_tf32 : Flag<["-"], "mamx-tf32">, Group<m_x86_Features_Group>;
def mno_amx_tf32 : Flag<["-"], "mno-amx-tf32">, Group<m_x86_Features_Group>;
def mamx_tile : Flag<["-"], "mamx-tile">, Group<m_x86_Features_Group>;
def mno_amx_tile : Flag<["-"], "mno-amx-tile">, Group<m_x86_Features_Group>;
def mamx_transpose : Flag<["-"], "mamx-transpose">, Group<m_x86_Features_Group>;
def mno_amx_transpose : Flag<["-"], "mno-amx-transpose">, Group<m_x86_Features_Group>;
def mamx_movrs: Flag<["-"], "mamx-movrs">, Group<m_x86_Features_Group>;
def mno_amx_movrs: Flag<["-"], "mno-amx-movrs">, Group<m_x86_Features_Group>;
def mcmpccxadd : Flag<["-"], "mcmpccxadd">, Group<m_x86_Features_Group>;
Expand Down
6 changes: 0 additions & 6 deletions clang/lib/Basic/Targets/X86.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -396,8 +396,6 @@ bool X86TargetInfo::handleTargetFeatures(std::vector<std::string> &Features,
HasAMXFP8 = true;
} else if (Feature == "+amx-movrs") {
HasAMXMOVRS = true;
} else if (Feature == "+amx-transpose") {
HasAMXTRANSPOSE = true;
} else if (Feature == "+amx-avx512") {
HasAMXAVX512 = true;
} else if (Feature == "+amx-tf32") {
Expand Down Expand Up @@ -925,8 +923,6 @@ void X86TargetInfo::getTargetDefines(const LangOptions &Opts,
Builder.defineMacro("__AMX_FP8__");
if (HasAMXMOVRS)
Builder.defineMacro("__AMX_MOVRS__");
if (HasAMXTRANSPOSE)
Builder.defineMacro("__AMX_TRANSPOSE__");
if (HasAMXAVX512)
Builder.defineMacro("__AMX_AVX512__");
if (HasAMXTF32)
Expand Down Expand Up @@ -1068,7 +1064,6 @@ bool X86TargetInfo::isValidFeatureName(StringRef Name) const {
.Case("amx-movrs", true)
.Case("amx-tf32", true)
.Case("amx-tile", true)
.Case("amx-transpose", true)
.Case("avx", true)
.Case("avx10.1", true)
.Case("avx10.2", true)
Expand Down Expand Up @@ -1189,7 +1184,6 @@ bool X86TargetInfo::hasFeature(StringRef Feature) const {
.Case("amx-movrs", HasAMXMOVRS)
.Case("amx-tf32", HasAMXTF32)
.Case("amx-tile", HasAMXTILE)
.Case("amx-transpose", HasAMXTRANSPOSE)
.Case("avx", SSELevel >= AVX)
.Case("avx10.1", HasAVX10_1)
.Case("avx10.2", HasAVX10_2)
Expand Down
1 change: 0 additions & 1 deletion clang/lib/Basic/Targets/X86.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -160,7 +160,6 @@ class LLVM_LIBRARY_VISIBILITY X86TargetInfo : public TargetInfo {
bool HasAMXCOMPLEX = false;
bool HasAMXFP8 = false;
bool HasAMXMOVRS = false;
bool HasAMXTRANSPOSE = false;
bool HasAMXAVX512 = false;
bool HasAMXTF32 = false;
bool HasSERIALIZE = false;
Expand Down
68 changes: 0 additions & 68 deletions clang/lib/CodeGen/TargetBuiltins/X86.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2931,74 +2931,6 @@ Value *CodeGenFunction::EmitX86BuiltinExpr(unsigned BuiltinID,
// instruction, but it will create a memset that won't be optimized away.
return Builder.CreateMemSet(Ops[0], Ops[1], Ops[2], Align(1), true);
}
// Corresponding to intrisics which will return 2 tiles (tile0_tile1).
case X86::BI__builtin_ia32_t2rpntlvwz0_internal:
case X86::BI__builtin_ia32_t2rpntlvwz0rs_internal:
case X86::BI__builtin_ia32_t2rpntlvwz0t1_internal:
case X86::BI__builtin_ia32_t2rpntlvwz0rst1_internal:
case X86::BI__builtin_ia32_t2rpntlvwz1_internal:
case X86::BI__builtin_ia32_t2rpntlvwz1rs_internal:
case X86::BI__builtin_ia32_t2rpntlvwz1t1_internal:
case X86::BI__builtin_ia32_t2rpntlvwz1rst1_internal: {
Intrinsic::ID IID;
switch (BuiltinID) {
default:
llvm_unreachable("Unsupported intrinsic!");
case X86::BI__builtin_ia32_t2rpntlvwz0_internal:
IID = Intrinsic::x86_t2rpntlvwz0_internal;
break;
case X86::BI__builtin_ia32_t2rpntlvwz0rs_internal:
IID = Intrinsic::x86_t2rpntlvwz0rs_internal;
break;
case X86::BI__builtin_ia32_t2rpntlvwz0t1_internal:
IID = Intrinsic::x86_t2rpntlvwz0t1_internal;
break;
case X86::BI__builtin_ia32_t2rpntlvwz0rst1_internal:
IID = Intrinsic::x86_t2rpntlvwz0rst1_internal;
break;
case X86::BI__builtin_ia32_t2rpntlvwz1_internal:
IID = Intrinsic::x86_t2rpntlvwz1_internal;
break;
case X86::BI__builtin_ia32_t2rpntlvwz1rs_internal:
IID = Intrinsic::x86_t2rpntlvwz1rs_internal;
break;
case X86::BI__builtin_ia32_t2rpntlvwz1t1_internal:
IID = Intrinsic::x86_t2rpntlvwz1t1_internal;
break;
case X86::BI__builtin_ia32_t2rpntlvwz1rst1_internal:
IID = Intrinsic::x86_t2rpntlvwz1rst1_internal;
break;
}

// Ops = (Row0, Col0, Col1, DstPtr0, DstPtr1, SrcPtr, Stride)
Value *Call = Builder.CreateCall(CGM.getIntrinsic(IID),
{Ops[0], Ops[1], Ops[2], Ops[5], Ops[6]});

auto *PtrTy = E->getArg(3)->getType()->getAs<PointerType>();
assert(PtrTy && "arg3 must be of pointer type");
QualType PtreeTy = PtrTy->getPointeeType();
llvm::Type *TyPtee = ConvertType(PtreeTy);

// Bitcast amx type (x86_amx) to vector type (256 x i32)
// Then store tile0 into DstPtr0
Value *T0 = Builder.CreateExtractValue(Call, 0);
Value *VecT0 = Builder.CreateIntrinsic(Intrinsic::x86_cast_tile_to_vector,
{TyPtee}, {T0});
Builder.CreateDefaultAlignedStore(VecT0, Ops[3]);

// Then store tile1 into DstPtr1
Value *T1 = Builder.CreateExtractValue(Call, 1);
Value *VecT1 = Builder.CreateIntrinsic(Intrinsic::x86_cast_tile_to_vector,
{TyPtee}, {T1});
Value *Store = Builder.CreateDefaultAlignedStore(VecT1, Ops[4]);

// Note: Here we escape directly use x86_tilestored64_internal to store
// the results due to it can't make sure the Mem written scope. This may
// cause shapes reloads after first amx intrinsic, which current amx reg-
// ister allocation has no ability to handle it.

return Store;
}
case X86::BI__ud2:
// llvm.trap makes a ud2a instruction on x86.
return EmitTrapCall(Intrinsic::trap);
Expand Down
6 changes: 0 additions & 6 deletions clang/lib/Headers/CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -162,18 +162,12 @@ set(x86_files
adxintrin.h
ammintrin.h
amxavx512intrin.h
amxbf16transposeintrin.h
amxcomplexintrin.h
amxcomplextransposeintrin.h
amxfp16intrin.h
amxfp16transposeintrin.h
amxfp8intrin.h
amxintrin.h
amxmovrsintrin.h
amxmovrstransposeintrin.h
amxtf32intrin.h
amxtf32transposeintrin.h
amxtransposeintrin.h
avx10_2_512bf16intrin.h
avx10_2_512convertintrin.h
avx10_2_512minmaxintrin.h
Expand Down
94 changes: 0 additions & 94 deletions clang/lib/Headers/amxbf16transposeintrin.h
View file
Open in desktop

This file was deleted.

Loading
Loading