Skip to content

Commit 90dc644

Browse files
arsenmarsenm
authored
AMDGPU: Add v_smfmac_f32_32x32x32x64_fp8_bf8 for gfx950 (llvm#117258)
1 parent 8d3435f commit 90dc644

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
@@ -456,6 +456,7 @@ TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_16x16x128_fp8_bf8, "V4fV4iV8iV4fiIiIi
456456
TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_16x16x128_fp8_fp8, "V4fV4iV8iV4fiIiIi", "nc", "gfx950-insts")
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")
459+
TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_32x32x64_fp8_bf8, "V16fV4iV8iV16fiIiIi", "nc", "gfx950-insts")
459460

460461
//===----------------------------------------------------------------------===//
461462
// GFX12+ only builtins.

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

Lines changed: 7 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -552,4 +552,11 @@ void test_smfmac_f32_32x32x64_bf8_fp8(global v16f* out, v4i a, v8i b, v16f c, in
552552
*out = __builtin_amdgcn_smfmac_f32_32x32x64_bf8_fp8(a, b, c, idx, 0, 0);
553553
}
554554

555+
// CHECK-GFX950-LABEL: @test_smfmac_f32_32x32x64_fp8_bf8
556+
// CHECK-GFX950: call <16 x float> @llvm.amdgcn.smfmac.f32.32x32x64.fp8.bf8(<4 x i32> %a, <8 x i32> %b, <16 x float> %c, i32 %idx, i32 0, i32 0)
557+
void test_smfmac_f32_32x32x64_fp8_bf8(global v16f* out, v4i a, v8i b, v16f c, int idx)
558+
{
559+
*out = __builtin_amdgcn_smfmac_f32_32x32x64_fp8_bf8(a, b, c, idx, 0, 0);
560+
}
561+
555562
#endif

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

Lines changed: 6 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -136,3 +136,9 @@ void test_smfmac_f32_32x32x64_bf8_fp8(global float16* out, int4 a, int8 b, float
136136
*out = __builtin_amdgcn_smfmac_f32_32x32x64_bf8_fp8(a, b, c, idx, d, 0); // expected-error{{argument to '__builtin_amdgcn_smfmac_f32_32x32x64_bf8_fp8' must be a constant integer}}
137137
*out = __builtin_amdgcn_smfmac_f32_32x32x64_bf8_fp8(a, b, c, idx, 0, d); // expected-error{{argument to '__builtin_amdgcn_smfmac_f32_32x32x64_bf8_fp8' must be a constant integer}}
138138
}
139+
140+
void test_smfmac_f32_32x32x64_fp8_bf8(global float16* out, int4 a, int8 b, float16 c, int idx, int d)
141+
{
142+
*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}}
143+
*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}}
144+
}

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

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -46,6 +46,7 @@ void test(__global float4* out0, half8 a0, half8 b0, float4 c0,
4646
*out12 = __builtin_amdgcn_smfmac_f32_16x16x128_fp8_fp8(a12, b12, c12, 0, 0, 0); // expected-error{{'__builtin_amdgcn_smfmac_f32_16x16x128_fp8_fp8' needs target feature gfx950-insts}}
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}}
49+
*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}}
4950
*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}}
5051
*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}}
5152
}

‎llvm/include/llvm/IR/IntrinsicsAMDGPU.td‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -3164,6 +3164,7 @@ def int_amdgcn_smfmac_f32_16x16x128_fp8_bf8 : AMDGPUMSmfmacIntrinsic<llvm_v4f32_
31643164
def int_amdgcn_smfmac_f32_16x16x128_fp8_fp8 : AMDGPUMSmfmacIntrinsic<llvm_v4f32_ty, llvm_v4i32_ty, llvm_v8i32_ty>;
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>;
3167+
def int_amdgcn_smfmac_f32_32x32x64_fp8_bf8 : AMDGPUMSmfmacIntrinsic<llvm_v16f32_ty, llvm_v4i32_ty, llvm_v8i32_ty>;
31673168
}
31683169

31693170
//===----------------------------------------------------------------------===//

‎llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp‎

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1102,6 +1102,7 @@ bool AMDGPUInstructionSelector::selectG_INTRINSIC(MachineInstr &I) const {
11021102
case Intrinsic::amdgcn_smfmac_f32_16x16x128_fp8_fp8:
11031103
case Intrinsic::amdgcn_smfmac_f32_32x32x64_bf8_bf8:
11041104
case Intrinsic::amdgcn_smfmac_f32_32x32x64_bf8_fp8:
1105+
case Intrinsic::amdgcn_smfmac_f32_32x32x64_fp8_bf8:
11051106
return selectSMFMACIntrin(I);
11061107
default:
11071108
return selectImpl(I, *CoverageInfo);
@@ -3559,6 +3560,9 @@ bool AMDGPUInstructionSelector::selectSMFMACIntrin(MachineInstr &MI) const {
35593560
case Intrinsic::amdgcn_smfmac_f32_32x32x64_bf8_fp8:
35603561
Opc = AMDGPU::V_SMFMAC_F32_32X32X64_BF8_FP8_e64;
35613562
break;
3563+
case Intrinsic::amdgcn_smfmac_f32_32x32x64_fp8_bf8:
3564+
Opc = AMDGPU::V_SMFMAC_F32_32X32X64_FP8_BF8_e64;
3565+
break;
35623566
default:
35633567
llvm_unreachable("unhandled smfmac intrinsic");
35643568
}

‎llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp‎

Lines changed: 2 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -4816,7 +4816,8 @@ AMDGPURegisterBankInfo::getInstrMapping(const MachineInstr &MI) const {
48164816
case Intrinsic::amdgcn_smfmac_f32_16x16x128_fp8_bf8:
48174817
case Intrinsic::amdgcn_smfmac_f32_16x16x128_fp8_fp8:
48184818
case Intrinsic::amdgcn_smfmac_f32_32x32x64_bf8_bf8:
4819-
case Intrinsic::amdgcn_smfmac_f32_32x32x64_bf8_fp8: {
4819+
case Intrinsic::amdgcn_smfmac_f32_32x32x64_bf8_fp8:
4820+
case Intrinsic::amdgcn_smfmac_f32_32x32x64_fp8_bf8: {
48204821
// vdst, srcA, srcB, srcC, idx
48214822
OpdsMapping[0] = getAGPROpMapping(MI.getOperand(0).getReg(), MRI, *TRI);
48224823
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
@@ -1064,6 +1064,7 @@ defm V_SMFMAC_F32_16X16X128_FP8_BF8 : SMFMACInst<"v_smfmac_f32_16x16x128_fp8_bf8
10641064
defm V_SMFMAC_F32_16X16X128_FP8_FP8 : SMFMACInst<"v_smfmac_f32_16x16x128_fp8_fp8", "F32_16X16X128_F8", int_amdgcn_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>;
1067+
defm V_SMFMAC_F32_32X32X64_FP8_BF8 : SMFMACInst<"v_smfmac_f32_32x32x64_fp8_bf8", "F32_32X32X64_F8", int_amdgcn_smfmac_f32_32x32x64_fp8_bf8>;
10671068
}
10681069

10691070
def MAIInstInfoTable : GenericTable {
@@ -2168,6 +2169,7 @@ defm V_SMFMAC_F32_16X16X128_FP8_BF8 : VOP3P_Real_SMFMAC <0x3d, "v_smfmac_f32_16x
21682169
defm V_SMFMAC_F32_16X16X128_FP8_FP8 : VOP3P_Real_SMFMAC <0x43, "v_smfmac_f32_16x16x128fp8fp8">;
21692170
defm V_SMFMAC_F32_32X32X64_BF8_BF8 : VOP3P_Real_SMFMAC <0x4b, "v_smfmac_f32_32x32x64bf8bf8">;
21702171
defm V_SMFMAC_F32_32X32X64_BF8_FP8 : VOP3P_Real_SMFMAC <0x4e, "v_smfmac_f32_32x32x64bf8fp8">;
2172+
defm V_SMFMAC_F32_32X32X64_FP8_BF8 : VOP3P_Real_SMFMAC <0x4f, "v_smfmac_f32_32x32x64fp8bf8">;
21712173

21722174
defm V_PK_FMA_F32 : VOP3P_Real_vi <0x30>;
21732175
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
@@ -422,6 +422,15 @@ define amdgpu_kernel void @smfmac_f32_32x32x64_bf8_fp8(<4 x i32> %arg0, <8 x i32
422422
ret void
423423
}
424424

425+
declare <16 x float> @llvm.amdgcn.smfmac.f32.32x32x64.fp8.bf8(<4 x i32>, <8 x i32>, <16 x float>, i32, i32, i32)
426+
427+
; CHECK: DIVERGENT: %result = call <16 x float> @llvm.amdgcn.smfmac.f32.32x32x64.fp8.bf8(<4 x i32> %arg0, <8 x i32> %arg1, <16 x float> %arg2, i32 %arg3, i32 1, i32 2)
428+
define amdgpu_kernel void @smfmac_f32_32x32x64_fp8_bf8(<4 x i32> %arg0, <8 x i32> %arg1, <16 x float> %arg2, i32 %arg3, ptr addrspace(1) %out) {
429+
%result = call <16 x float> @llvm.amdgcn.smfmac.f32.32x32x64.fp8.bf8(<4 x i32> %arg0, <8 x i32> %arg1, <16 x float> %arg2, i32 %arg3, i32 1, i32 2)
430+
store <16 x float> %result, ptr addrspace(1) %out
431+
ret void
432+
}
433+
425434
declare i32 @llvm.amdgcn.ds.swizzle(i32, i32) #1
426435
declare i32 @llvm.amdgcn.permlane16.i32(i32, i32, i32, i32, i1, i1) #1
427436
declare i32 @llvm.amdgcn.permlanex16.i32(i32, i32, i32, i32, i1, i1) #1

0 commit comments

Comments
 (0)