Skip to content

Commit 7d544c6

Browse files
arsenmarsenm
authored
AMDGPU: Add v_smfmac_f32_32x32x64_fp8_fp8 for gfx950 (#117259)
1 parent 90dc644 commit 7d544c6

File tree

12 files changed

+505
-1
lines changed

12 files changed

+505
-1
lines changed

‎clang/include/clang/Basic/BuiltinsAMDGPU.def

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -457,6 +457,7 @@ TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_16x16x128_fp8_fp8, "V4fV4iV8iV4fiIiIi
457457
TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_32x32x64_bf8_bf8, "V16fV4iV8iV16fiIiIi", "nc", "gfx950-insts")
458458
TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_32x32x64_bf8_fp8, "V16fV4iV8iV16fiIiIi", "nc", "gfx950-insts")
459459
TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_32x32x64_fp8_bf8, "V16fV4iV8iV16fiIiIi", "nc", "gfx950-insts")
460+
TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_32x32x64_fp8_fp8, "V16fV4iV8iV16fiIiIi", "nc", "gfx950-insts")
460461

461462
//===----------------------------------------------------------------------===//
462463
// GFX12+ only builtins.

‎clang/test/CodeGenOpenCL/builtins-amdgcn-mfma.cl

Lines changed: 7 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -559,4 +559,11 @@ void test_smfmac_f32_32x32x64_fp8_bf8(global v16f* out, v4i a, v8i b, v16f c, in
559559
*out = __builtin_amdgcn_smfmac_f32_32x32x64_fp8_bf8(a, b, c, idx, 0, 0);
560560
}
561561

562+
// CHECK-GFX950-LABEL: @test_smfmac_f32_32x32x64_fp8_fp8
563+
// CHECK-GFX950: call <16 x float> @llvm.amdgcn.smfmac.f32.32x32x64.fp8.fp8(<4 x i32> %a, <8 x i32> %b, <16 x float> %c, i32 %idx, i32 0, i32 0)
564+
void test_smfmac_f32_32x32x64_fp8_fp8(global v16f* out, v4i a, v8i b, v16f c, int idx)
565+
{
566+
*out = __builtin_amdgcn_smfmac_f32_32x32x64_fp8_fp8(a, b, c, idx, 0, 0);
567+
}
568+
562569
#endif

‎clang/test/SemaOpenCL/builtins-amdgcn-error-gfx950-param.cl

Lines changed: 6 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -142,3 +142,9 @@ void test_smfmac_f32_32x32x64_fp8_bf8(global float16* out, int4 a, int8 b, float
142142
*out = __builtin_amdgcn_smfmac_f32_32x32x64_fp8_bf8(a, b, c, idx, d, 0); // expected-error{{argument to '__builtin_amdgcn_smfmac_f32_32x32x64_fp8_bf8' must be a constant integer}}
143143
*out = __builtin_amdgcn_smfmac_f32_32x32x64_fp8_bf8(a, b, c, idx, 0, d); // expected-error{{argument to '__builtin_amdgcn_smfmac_f32_32x32x64_fp8_bf8' must be a constant integer}}
144144
}
145+
146+
void test_smfmac_f32_32x32x64_fp8_fp8(global float16* out, int4 a, int8 b, float16 c, int idx, int d)
147+
{
148+
*out = __builtin_amdgcn_smfmac_f32_32x32x64_fp8_fp8(a, b, c, idx, d, 0); // expected-error{{argument to '__builtin_amdgcn_smfmac_f32_32x32x64_fp8_fp8' must be a constant integer}}
149+
*out = __builtin_amdgcn_smfmac_f32_32x32x64_fp8_fp8(a, b, c, idx, 0, d); // expected-error{{argument to '__builtin_amdgcn_smfmac_f32_32x32x64_fp8_fp8' must be a constant integer}}
150+
}

‎clang/test/SemaOpenCL/builtins-amdgcn-error-gfx950.cl

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -47,6 +47,7 @@ void test(__global float4* out0, half8 a0, half8 b0, float4 c0,
4747
*out13 = __builtin_amdgcn_smfmac_f32_32x32x64_bf8_bf8(a13, b13, c13, 0, 0, 0); // expected-error{{'__builtin_amdgcn_smfmac_f32_32x32x64_bf8_bf8' needs target feature gfx950-insts}}
4848
*out13 = __builtin_amdgcn_smfmac_f32_32x32x64_bf8_fp8(a13, b13, c13, 0, 0, 0); // expected-error{{'__builtin_amdgcn_smfmac_f32_32x32x64_bf8_fp8' needs target feature gfx950-insts}}
4949
*out13 = __builtin_amdgcn_smfmac_f32_32x32x64_fp8_bf8(a13, b13, c13, 0, 0, 0); // expected-error{{'__builtin_amdgcn_smfmac_f32_32x32x64_fp8_bf8' needs target feature gfx950-insts}}
50+
*out13 = __builtin_amdgcn_smfmac_f32_32x32x64_fp8_fp8(a13, b13, c13, 0, 0, 0); // expected-error{{'__builtin_amdgcn_smfmac_f32_32x32x64_fp8_fp8' needs target feature gfx950-insts}}
5051
*out14 = __builtin_amdgcn_mfma_scale_f32_16x16x128_f8f6f4(a14, b14, c14, 0, 0, 0, d14, 0, e14); // expected-error{{'__builtin_amdgcn_mfma_scale_f32_16x16x128_f8f6f4' needs target feature gfx950-insts}}
5152
*out15 = __builtin_amdgcn_mfma_scale_f32_32x32x64_f8f6f4(a15, b15, c15, 0, 0, 0, d15, 0, e15); // expected-error{{'__builtin_amdgcn_mfma_scale_f32_32x32x64_f8f6f4' needs target feature gfx950-insts}}
5253
}

‎llvm/include/llvm/IR/IntrinsicsAMDGPU.td

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -3165,6 +3165,7 @@ def int_amdgcn_smfmac_f32_16x16x128_fp8_fp8 : AMDGPUMSmfmacIntrinsic<llvm_v4f32_
31653165
def int_amdgcn_smfmac_f32_32x32x64_bf8_bf8 : AMDGPUMSmfmacIntrinsic<llvm_v16f32_ty, llvm_v4i32_ty, llvm_v8i32_ty>;
31663166
def int_amdgcn_smfmac_f32_32x32x64_bf8_fp8 : AMDGPUMSmfmacIntrinsic<llvm_v16f32_ty, llvm_v4i32_ty, llvm_v8i32_ty>;
31673167
def int_amdgcn_smfmac_f32_32x32x64_fp8_bf8 : AMDGPUMSmfmacIntrinsic<llvm_v16f32_ty, llvm_v4i32_ty, llvm_v8i32_ty>;
3168+
def int_amdgcn_smfmac_f32_32x32x64_fp8_fp8 : AMDGPUMSmfmacIntrinsic<llvm_v16f32_ty, llvm_v4i32_ty, llvm_v8i32_ty>;
31683169
}
31693170

31703171
//===----------------------------------------------------------------------===//

‎llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1103,6 +1103,7 @@ bool AMDGPUInstructionSelector::selectG_INTRINSIC(MachineInstr &I) const {
11031103
case Intrinsic::amdgcn_smfmac_f32_32x32x64_bf8_bf8:
11041104
case Intrinsic::amdgcn_smfmac_f32_32x32x64_bf8_fp8:
11051105
case Intrinsic::amdgcn_smfmac_f32_32x32x64_fp8_bf8:
1106+
case Intrinsic::amdgcn_smfmac_f32_32x32x64_fp8_fp8:
11061107
return selectSMFMACIntrin(I);
11071108
default:
11081109
return selectImpl(I, *CoverageInfo);
@@ -3563,6 +3564,9 @@ bool AMDGPUInstructionSelector::selectSMFMACIntrin(MachineInstr &MI) const {
35633564
case Intrinsic::amdgcn_smfmac_f32_32x32x64_fp8_bf8:
35643565
Opc = AMDGPU::V_SMFMAC_F32_32X32X64_FP8_BF8_e64;
35653566
break;
3567+
case Intrinsic::amdgcn_smfmac_f32_32x32x64_fp8_fp8:
3568+
Opc = AMDGPU::V_SMFMAC_F32_32X32X64_FP8_FP8_e64;
3569+
break;
35663570
default:
35673571
llvm_unreachable("unhandled smfmac intrinsic");
35683572
}

‎llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp

Lines changed: 2 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -4817,7 +4817,8 @@ AMDGPURegisterBankInfo::getInstrMapping(const MachineInstr &MI) const {
48174817
case Intrinsic::amdgcn_smfmac_f32_16x16x128_fp8_fp8:
48184818
case Intrinsic::amdgcn_smfmac_f32_32x32x64_bf8_bf8:
48194819
case Intrinsic::amdgcn_smfmac_f32_32x32x64_bf8_fp8:
4820-
case Intrinsic::amdgcn_smfmac_f32_32x32x64_fp8_bf8: {
4820+
case Intrinsic::amdgcn_smfmac_f32_32x32x64_fp8_bf8:
4821+
case Intrinsic::amdgcn_smfmac_f32_32x32x64_fp8_fp8: {
48214822
// vdst, srcA, srcB, srcC, idx
48224823
OpdsMapping[0] = getAGPROpMapping(MI.getOperand(0).getReg(), MRI, *TRI);
48234824
OpdsMapping[2] = getVGPROpMapping(MI.getOperand(2).getReg(), MRI, *TRI);

‎llvm/lib/Target/AMDGPU/VOP3PInstructions.td

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1065,6 +1065,7 @@ defm V_SMFMAC_F32_16X16X128_FP8_FP8 : SMFMACInst<"v_smfmac_f32_16x16x128_fp8_fp8
10651065
defm V_SMFMAC_F32_32X32X64_BF8_BF8 : SMFMACInst<"v_smfmac_f32_32x32x64_bf8_bf8", "F32_32X32X64_F8", int_amdgcn_smfmac_f32_32x32x64_bf8_bf8>;
10661066
defm V_SMFMAC_F32_32X32X64_BF8_FP8 : SMFMACInst<"v_smfmac_f32_32x32x64_bf8_fp8", "F32_32X32X64_F8", int_amdgcn_smfmac_f32_32x32x64_bf8_fp8>;
10671067
defm V_SMFMAC_F32_32X32X64_FP8_BF8 : SMFMACInst<"v_smfmac_f32_32x32x64_fp8_bf8", "F32_32X32X64_F8", int_amdgcn_smfmac_f32_32x32x64_fp8_bf8>;
1068+
defm V_SMFMAC_F32_32X32X64_FP8_FP8 : SMFMACInst<"v_smfmac_f32_32x32x64_fp8_fp8", "F32_32X32X64_F8", int_amdgcn_smfmac_f32_32x32x64_fp8_fp8>;
10681069
}
10691070

10701071
def MAIInstInfoTable : GenericTable {
@@ -2170,6 +2171,7 @@ defm V_SMFMAC_F32_16X16X128_FP8_FP8 : VOP3P_Real_SMFMAC <0x43, "v_smfmac_f32_16x
21702171
defm V_SMFMAC_F32_32X32X64_BF8_BF8 : VOP3P_Real_SMFMAC <0x4b, "v_smfmac_f32_32x32x64bf8bf8">;
21712172
defm V_SMFMAC_F32_32X32X64_BF8_FP8 : VOP3P_Real_SMFMAC <0x4e, "v_smfmac_f32_32x32x64bf8fp8">;
21722173
defm V_SMFMAC_F32_32X32X64_FP8_BF8 : VOP3P_Real_SMFMAC <0x4f, "v_smfmac_f32_32x32x64fp8bf8">;
2174+
defm V_SMFMAC_F32_32X32X64_FP8_FP8 : VOP3P_Real_SMFMAC <0x53, "v_smfmac_f32_32x32x64fp8fp8">;
21732175

21742176
defm V_PK_FMA_F32 : VOP3P_Real_vi <0x30>;
21752177
defm V_PK_MUL_F32 : VOP3P_Real_vi <0x31>;

‎llvm/test/Analysis/UniformityAnalysis/AMDGPU/intrinsics.ll

Lines changed: 9 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -431,6 +431,15 @@ define amdgpu_kernel void @smfmac_f32_32x32x64_fp8_bf8(<4 x i32> %arg0, <8 x i32
431431
ret void
432432
}
433433

434+
declare <16 x float> @llvm.amdgcn.smfmac.f32.32x32x64.fp8.fp8(<4 x i32>, <8 x i32>, <16 x float>, i32, i32, i32)
435+
436+
; CHECK: DIVERGENT: %result = call <16 x float> @llvm.amdgcn.smfmac.f32.32x32x64.fp8.fp8(<4 x i32> %arg0, <8 x i32> %arg1, <16 x float> %arg2, i32 %arg3, i32 1, i32 2)
437+
define amdgpu_kernel void @smfmac_f32_32x32x64_fp8_fp8(<4 x i32> %arg0, <8 x i32> %arg1, <16 x float> %arg2, i32 %arg3, ptr addrspace(1) %out) {
438+
%result = call <16 x float> @llvm.amdgcn.smfmac.f32.32x32x64.fp8.fp8(<4 x i32> %arg0, <8 x i32> %arg1, <16 x float> %arg2, i32 %arg3, i32 1, i32 2)
439+
store <16 x float> %result, ptr addrspace(1) %out
440+
ret void
441+
}
442+
434443
declare i32 @llvm.amdgcn.ds.swizzle(i32, i32) #1
435444
declare i32 @llvm.amdgcn.permlane16.i32(i32, i32, i32, i32, i1, i1) #1
436445
declare i32 @llvm.amdgcn.permlanex16.i32(i32, i32, i32, i32, i1, i1) #1

0 commit comments

Comments
 (0)