| blob | f8a60d37801ebec808d96e92cccbfab5b6f10d7a |
1 // RUN: mlir-opt %s -convert-amdgpu-to-rocdl=chipset=gfx940 -cse | FileCheck %s
2 func.func @mfma_to_rocdl(%arg0 : f32, %arg1 : vector<32xf32>,
3 %arg2 : vector<16xf32>, %arg3 : vector<4xf32>,
4 %arg4 : vector<4xf16>, %arg5 : vector<4xi8>,
5 %arg6 : vector<32xi32>, %arg7 : vector<16xi32>,
6 %arg8 : vector<4xi32>, %arg9 : vector<2xbf16>,
7 %arg10 : vector<4xbf16>, %arg11 : f64,
8 %arg12 : vector<4xf64>, %arg13 : vector<8xi8>,
9 %arg14 : vector<2xf32>, %arg15 : vector<8xf8E5M2FNUZ>,
10 %arg16 : vector<8xf8E4M3FNUZ>) {
11 // CHECK: rocdl.mfma.f32.32x32x1f32{{.*}}: (f32, f32, vector<32xf32>, i32, i32, i32) -> vector<32xf32>
12 amdgpu.mfma %arg0 * %arg0 + %arg1 { abid = 0 : i32, cbsz = 0 : i32, k = 1 : i32, m = 32 : i32, n = 32 : i32, blocks = 2 : i32 } blgp = none : f32, f32, vector<32xf32>
13 // CHECK: rocdl.mfma.f32.16x16x1f32{{.*}}: (f32, f32, vector<16xf32>, i32, i32, i32) -> vector<16xf32>
14 amdgpu.mfma %arg0 * %arg0 + %arg2 { abid = 0 : i32, cbsz = 0 : i32, k = 1 : i32, m = 16 : i32, n = 16 : i32, blocks = 4 : i32 } blgp = none : f32, f32, vector<16xf32>
15 // CHECK: rocdl.mfma.f32.4x4x1f32{{.*}}: (f32, f32, vector<4xf32>, i32, i32, i32) -> vector<4xf32>
16 amdgpu.mfma %arg0 * %arg0 + %arg3 { abid = 0 : i32, cbsz = 0 : i32, k = 1 : i32, m = 4 : i32, n = 4 : i32, blocks = 16 : i32 } blgp = none : f32, f32, vector<4xf32>
17 // CHECK: rocdl.mfma.f32.32x32x2f32{{.*}}: (f32, f32, vector<16xf32>, i32, i32, i32) -> vector<16xf32>
18 amdgpu.mfma %arg0 * %arg0 + %arg2 { abid = 0 : i32, cbsz = 0 : i32, k = 2 : i32, m = 32 : i32, n = 32 : i32, blocks = 1 : i32 } blgp = none : f32, f32, vector<16xf32>
19 // CHECK: rocdl.mfma.f32.16x16x4f32{{.*}}: (f32, f32, vector<4xf32>, i32, i32, i32) -> vector<4xf32>
20 amdgpu.mfma %arg0 * %arg0 + %arg3 { abid = 0 : i32, cbsz = 0 : i32, k = 4 : i32, m = 16 : i32, n = 16 : i32, blocks = 1 : i32 } blgp = none : f32, f32, vector<4xf32>
21 // CHECK: rocdl.mfma.f32.32x32x4f16{{.*}}: (vector<4xf16>, vector<4xf16>, vector<32xf32>, i32, i32, i32) -> vector<32xf32>
22 amdgpu.mfma %arg4 * %arg4 + %arg1 { abid = 0 : i32, cbsz = 0 : i32, k = 4 : i32, m = 32 : i32, n = 32 : i32, blocks = 2 : i32 } blgp = none : vector<4xf16>, vector<4xf16>, vector<32xf32>
23 // CHECK: rocdl.mfma.f32.16x16x4f16{{.*}}: (vector<4xf16>, vector<4xf16>, vector<16xf32>, i32, i32, i32) -> vector<16xf32>
24 amdgpu.mfma %arg4 * %arg4 + %arg2 { abid = 0 : i32, cbsz = 0 : i32, k = 4 : i32, m = 16 : i32, n = 16 : i32, blocks = 4 : i32 } blgp = none : vector<4xf16>, vector<4xf16>, vector<16xf32>
25 // CHECK: rocdl.mfma.f32.4x4x4f16{{.*}}: (vector<4xf16>, vector<4xf16>, vector<4xf32>, i32, i32, i32) -> vector<4xf32>
26 amdgpu.mfma %arg4 * %arg4 + %arg3 { abid = 0 : i32, cbsz = 0 : i32, k = 4 : i32, m = 4 : i32, n = 4 : i32, blocks = 16 : i32 } blgp = none : vector<4xf16>, vector<4xf16>, vector<4xf32>
27 // CHECK: rocdl.mfma.f32.32x32x8f16{{.*}}: (vector<4xf16>, vector<4xf16>, vector<16xf32>, i32, i32, i32) -> vector<16xf32>
28 amdgpu.mfma %arg4 * %arg4 + %arg2 { abid = 0 : i32, cbsz = 0 : i32, k = 8 : i32, m = 32 : i32, n = 32 : i32, blocks = 1 : i32 } blgp = none : vector<4xf16>, vector<4xf16>, vector<16xf32>
29 // CHECK: rocdl.mfma.f32.16x16x16f16{{.*}}: (vector<4xf16>, vector<4xf16>, vector<4xf32>, i32, i32, i32) -> vector<4xf32>
30 amdgpu.mfma %arg4 * %arg4 + %arg3 { abid = 0 : i32, cbsz = 0 : i32, k = 16 : i32, m = 16 : i32, n = 16 : i32, blocks = 1 : i32 } blgp = none : vector<4xf16>, vector<4xf16>, vector<4xf32>
31 // CHECK: %[[BITCAST_4xi8_i32:.+]] = llvm.bitcast {{.*}} : vector<4xi8> to i32
32 // CHECK: rocdl.mfma.i32.32x32x4i8 %[[BITCAST_4xi8_i32]], %[[BITCAST_4xi8_i32]], {{.*}}: (i32, i32, vector<32xi32>, i32, i32, i32) -> vector<32xi32>
33 amdgpu.mfma %arg5 * %arg5 + %arg6 { abid = 0 : i32, cbsz = 0 : i32, k = 4 : i32, m = 32 : i32, n = 32 : i32, blocks = 2 : i32 } blgp = none : vector<4xi8>, vector<4xi8>, vector<32xi32>
34 // CHECK: rocdl.mfma.i32.16x16x4i8{{.*}}: (i32, i32, vector<16xi32>, i32, i32, i32) -> vector<16xi32>
35 amdgpu.mfma %arg5 * %arg5 + %arg7 { abid = 0 : i32, cbsz = 0 : i32, k = 4 : i32, m = 16 : i32, n = 16 : i32, blocks = 4 : i32 } blgp = none : vector<4xi8>, vector<4xi8>, vector<16xi32>
36 // CHECK: rocdl.mfma.i32.4x4x4i8{{.*}}: (i32, i32, vector<4xi32>, i32, i32, i32) -> vector<4xi32>
37 amdgpu.mfma %arg5 * %arg5 + %arg8 { abid = 0 : i32, cbsz = 0 : i32, k = 4 : i32, m = 4 : i32, n = 4 : i32, blocks = 16 : i32 } blgp = none : vector<4xi8>, vector<4xi8>, vector<4xi32>
38 // CHECK: rocdl.mfma.i32.32x32x8i8{{.*}}: (i32, i32, vector<16xi32>, i32, i32, i32) -> vector<16xi32>
39 amdgpu.mfma %arg5 * %arg5 + %arg7 { abid = 0 : i32, cbsz = 0 : i32, k = 8 : i32, m = 32 : i32, n = 32 : i32, blocks = 1 : i32 } blgp = none : vector<4xi8>, vector<4xi8>, vector<16xi32>
40 // CHECK: rocdl.mfma.i32.16x16x16i8{{.*}}: (i32, i32, vector<4xi32>, i32, i32, i32) -> vector<4xi32>
41 amdgpu.mfma %arg5 * %arg5 + %arg8 { abid = 0 : i32, cbsz = 0 : i32, k = 16 : i32, m = 16 : i32, n = 16 : i32, blocks = 1 : i32 } blgp = none : vector<4xi8>, vector<4xi8>, vector<4xi32>
42 // CHECK: %[[BITCAST_2xbf16_2xi16:.+]] = llvm.bitcast {{.*}} : vector<2xbf16> to vector<2xi16>
43 // CHECK: rocdl.mfma.f32.32x32x2bf16 %[[BITCAST_2xbf16_2xi16]], %[[BITCAST_2xbf16_2xi16]], %{{.*}}: (vector<2xi16>, vector<2xi16>, vector<32xf32>, i32, i32, i32) -> vector<32xf32>
44 amdgpu.mfma %arg9 * %arg9 + %arg1 { abid = 0 : i32, cbsz = 0 : i32, k = 2 : i32, m = 32 : i32, n = 32 : i32, blocks = 2 : i32 } blgp = none : vector<2xbf16>, vector<2xbf16>, vector<32xf32>
45 // CHECK: rocdl.mfma.f32.16x16x2bf16{{.*}}: (vector<2xi16>, vector<2xi16>, vector<16xf32>, i32, i32, i32) -> vector<16xf32>
46 amdgpu.mfma %arg9 * %arg9 + %arg2 { abid = 0 : i32, cbsz = 0 : i32, k = 2 : i32, m = 16 : i32, n = 16 : i32, blocks = 4 : i32 } blgp = none : vector<2xbf16>, vector<2xbf16>, vector<16xf32>
47 // CHECK: rocdl.mfma.f32.4x4x2bf16{{.*}}: (vector<2xi16>, vector<2xi16>, vector<4xf32>, i32, i32, i32) -> vector<4xf32>
48 amdgpu.mfma %arg9 * %arg9 + %arg3 { abid = 0 : i32, cbsz = 0 : i32, k = 2 : i32, m = 4 : i32, n = 4 : i32, blocks = 16 : i32 } blgp = none : vector<2xbf16>, vector<2xbf16>, vector<4xf32>
49 // CHECK: rocdl.mfma.f32.32x32x4bf16{{.*}}: (vector<2xi16>, vector<2xi16>, vector<16xf32>, i32, i32, i32) -> vector<16xf32>
50 amdgpu.mfma %arg9 * %arg9 + %arg2 { abid = 0 : i32, cbsz = 0 : i32, k = 4 : i32, m = 32 : i32, n = 32 : i32, blocks = 1 : i32 } blgp = none : vector<2xbf16>, vector<2xbf16>, vector<16xf32>
51 // CHECK: rocdl.mfma.f32.16x16x8bf16{{.*}}: (vector<2xi16>, vector<2xi16>, vector<4xf32>, i32, i32, i32) -> vector<4xf32>
52 amdgpu.mfma %arg9 * %arg9 + %arg3 { abid = 0 : i32, cbsz = 0 : i32, k = 8 : i32, m = 16 : i32, n = 16 : i32, blocks = 1 : i32 } blgp = none : vector<2xbf16>, vector<2xbf16>, vector<4xf32>
53 // CHECK: %[[BITCAST_4xbf16_4xi16:.+]] = llvm.bitcast {{.*}} : vector<4xbf16> to vector<4xi16>
54 // CHECK: rocdl.mfma.f32.32x32x4bf16.1k %[[BITCAST_4xbf16_4xi16]], %[[BITCAST_4xbf16_4xi16]], {{.*}}: (vector<4xi16>, vector<4xi16>, vector<32xf32>, i32, i32, i32) -> vector<32xf32>
55 amdgpu.mfma %arg10 * %arg10 + %arg1 { abid = 0 : i32, cbsz = 0 : i32, k = 4 : i32, m = 32 : i32, n = 32 : i32, blocks = 2 : i32 } blgp = none : vector<4xbf16>, vector<4xbf16>, vector<32xf32>
56 // CHECK: rocdl.mfma.f32.16x16x4bf16.1k{{.*}}: (vector<4xi16>, vector<4xi16>, vector<16xf32>, i32, i32, i32) -> vector<16xf32>
57 amdgpu.mfma %arg10 * %arg10 + %arg2 { abid = 0 : i32, cbsz = 0 : i32, k = 4 : i32, m = 16 : i32, n = 16 : i32, blocks = 4 : i32 } blgp = none : vector<4xbf16>, vector<4xbf16>, vector<16xf32>
58 // CHECK: rocdl.mfma.f32.4x4x4bf16.1k{{.*}}: (vector<4xi16>, vector<4xi16>, vector<4xf32>, i32, i32, i32) -> vector<4xf32>
59 amdgpu.mfma %arg10 * %arg10 + %arg3 { abid = 0 : i32, cbsz = 0 : i32, k = 4 : i32, m = 4 : i32, n = 4 : i32, blocks = 16 : i32 } blgp = none : vector<4xbf16>, vector<4xbf16>, vector<4xf32>
60 // CHECK: rocdl.mfma.f32.32x32x8bf16.1k{{.*}}: (vector<4xi16>, vector<4xi16>, vector<16xf32>, i32, i32, i32) -> vector<16xf32>
61 amdgpu.mfma %arg10 * %arg10 + %arg2 { abid = 0 : i32, cbsz = 0 : i32, k = 8 : i32, m = 32 : i32, n = 32 : i32, blocks = 1 : i32 } blgp = none : vector<4xbf16>, vector<4xbf16>, vector<16xf32>
62 // CHECK: rocdl.mfma.f32.16x16x16bf16.1k{{.*}}: (vector<4xi16>, vector<4xi16>, vector<4xf32>, i32, i32, i32) -> vector<4xf32>
63 amdgpu.mfma %arg10 * %arg10 + %arg3 { abid = 0 : i32, cbsz = 0 : i32, k = 16 : i32, m = 16 : i32, n = 16 : i32, blocks = 1 : i32 } blgp = none : vector<4xbf16>, vector<4xbf16>, vector<4xf32>
64 // CHECK: rocdl.mfma.f64.16x16x4f64{{.*}}: (f64, f64, vector<4xf64>, i32, i32, i32) -> vector<4xf64>
65 amdgpu.mfma %arg11 * %arg11 + %arg12 { abid = 0 : i32, cbsz = 0 : i32, k = 4 : i32, m = 16 : i32, n = 16 : i32, blocks = 1 : i32 } blgp = none : f64, f64, vector<4xf64>
66 // CHECK: rocdl.mfma.f64.4x4x4f64{{.*}}: (f64, f64, f64, i32, i32, i32) -> f64
67 amdgpu.mfma %arg11 * %arg11 + %arg11 { abid = 0 : i32, cbsz = 0 : i32, k = 4 : i32, m = 4 : i32, n = 4 : i32, blocks = 4 : i32 } blgp = none : f64, f64, f64
68 // CHECK: %[[BITCAST_8xi8_i64:.+]] = llvm.bitcast {{.*}} : vector<8xi8> to i64
69 // CHECK: rocdl.mfma.i32.16x16x32.i8 %[[BITCAST_8xi8_i64]], %[[BITCAST_8xi8_i64]], {{.*}}: (i64, i64, vector<4xi32>, i32, i32, i32) -> vector<4xi32>
70 amdgpu.mfma %arg13 * %arg13 + %arg8 { abid = 0 : i32, cbsz = 0 : i32, k = 32 : i32, m = 16 : i32, n = 16 : i32, blocks = 1 : i32 } blgp = none : vector<8xi8>, vector<8xi8>, vector<4xi32>
71 // CHECK: rocdl.mfma.i32.32x32x16.i8{{.*}}: (i64, i64, vector<16xi32>, i32, i32, i32) -> vector<16xi32>
72 amdgpu.mfma %arg13 * %arg13 + %arg7 { abid = 0 : i32, cbsz = 0 : i32, k = 16 : i32, m = 32 : i32, n = 32 : i32, blocks = 1 : i32 } blgp = none : vector<8xi8>, vector<8xi8>, vector<16xi32>
73 // CHECK: rocdl.mfma.f32.16x16x8.xf32{{.*}}: (vector<2xf32>, vector<2xf32>, vector<4xf32>, i32, i32, i32) -> vector<4xf32>
74 amdgpu.mfma %arg14 * %arg14 + %arg3 { abid = 0 : i32, cbsz = 0 : i32, k = 8 : i32, m = 16 : i32, n = 16 : i32, blocks = 1 : i32, reducePrecision } blgp = none : vector<2xf32>, vector<2xf32>, vector<4xf32>
75 // CHECK: rocdl.mfma.f32.32x32x4.xf32{{.*}}: (vector<2xf32>, vector<2xf32>, vector<16xf32>, i32, i32, i32) -> vector<16xf32>
76 amdgpu.mfma %arg14 * %arg14 + %arg2 { abid = 0 : i32, cbsz = 0 : i32, k = 4 : i32, m = 32 : i32, n = 32 : i32, blocks = 1 : i32, reducePrecision } blgp = none : vector<2xf32>, vector<2xf32>, vector<16xf32>
77 // CHECK: %[[BITCAST_8xi8_i64_1:.+]] = llvm.bitcast {{.*}} : vector<8xi8> to i64
78 // CHECK: rocdl.mfma.f32.16x16x32.bf8.bf8 %[[BITCAST_8xi8_i64_1]], %[[BITCAST_8xi8_i64_1]], {{.*}}: (i64, i64, vector<4xf32>, i32, i32, i32) -> vector<4xf32>
79 amdgpu.mfma %arg15 * %arg15 + %arg3 { abid = 0 : i32, cbsz = 0 : i32, k = 32 : i32, m = 16 : i32, n = 16 : i32, blocks = 1 : i32 } blgp = none : vector<8xf8E5M2FNUZ>, vector<8xf8E5M2FNUZ>, vector<4xf32>
80 // CHECK: %[[BITCAST_8xi8_i64_2:.+]] = llvm.bitcast {{.*}} : vector<8xi8> to i64
81 // CHECK: rocdl.mfma.f32.16x16x32.bf8.fp8 %[[BITCAST_8xi8_i64_1]], %[[BITCAST_8xi8_i64_2]], {{.*}}: (i64, i64, vector<4xf32>, i32, i32, i32) -> vector<4xf32>
82 amdgpu.mfma %arg15 * %arg16 + %arg3 { abid = 0 : i32, cbsz = 0 : i32, k = 32 : i32, m = 16 : i32, n = 16 : i32, blocks = 1 : i32 } blgp = none : vector<8xf8E5M2FNUZ>, vector<8xf8E4M3FNUZ>, vector<4xf32>
83 // CHECK: rocdl.mfma.f32.16x16x32.fp8.bf8{{.*}}: (i64, i64, vector<4xf32>, i32, i32, i32) -> vector<4xf32>
84 amdgpu.mfma %arg16 * %arg15 + %arg3 { abid = 0 : i32, cbsz = 0 : i32, k = 32 : i32, m = 16 : i32, n = 16 : i32, blocks = 1 : i32 } blgp = none : vector<8xf8E4M3FNUZ>, vector<8xf8E5M2FNUZ>, vector<4xf32>
85 // CHECK: rocdl.mfma.f32.16x16x32.fp8.fp8{{.*}}: (i64, i64, vector<4xf32>, i32, i32, i32) -> vector<4xf32>
86 amdgpu.mfma %arg16 * %arg16 + %arg3 { abid = 0 : i32, cbsz = 0 : i32, k = 32 : i32, m = 16 : i32, n = 16 : i32, blocks = 1 : i32 } blgp = none : vector<8xf8E4M3FNUZ>, vector<8xf8E4M3FNUZ>, vector<4xf32>
87 // CHECK: rocdl.mfma.f32.32x32x16.bf8.bf8{{.*}}: (i64, i64, vector<16xf32>, i32, i32, i32) -> vector<16xf32>
88 amdgpu.mfma %arg15 * %arg15 + %arg2 { abid = 0 : i32, cbsz = 0 : i32, k = 16 : i32, m = 32 : i32, n = 32 : i32, blocks = 1 : i32 } blgp = none : vector<8xf8E5M2FNUZ>, vector<8xf8E5M2FNUZ>, vector<16xf32>
89 // CHECK: rocdl.mfma.f32.32x32x16.bf8.fp8{{.*}}: (i64, i64, vector<16xf32>, i32, i32, i32) -> vector<16xf32>
90 amdgpu.mfma %arg15 * %arg16 + %arg2 { abid = 0 : i32, cbsz = 0 : i32, k = 16 : i32, m = 32 : i32, n = 32 : i32, blocks = 1 : i32 } blgp = none : vector<8xf8E5M2FNUZ>, vector<8xf8E4M3FNUZ>, vector<16xf32>
91 // CHECK: rocdl.mfma.f32.32x32x16.fp8.bf8{{.*}}: (i64, i64, vector<16xf32>, i32, i32, i32) -> vector<16xf32>
92 amdgpu.mfma %arg16 * %arg15 + %arg2 { abid = 0 : i32, cbsz = 0 : i32, k = 16 : i32, m = 32 : i32, n = 32 : i32, blocks = 1 : i32 } blgp = none : vector<8xf8E4M3FNUZ>, vector<8xf8E5M2FNUZ>, vector<16xf32>
93 // CHECK: rocdl.mfma.f32.32x32x16.fp8.fp8{{.*}}: (i64, i64, vector<16xf32>, i32, i32, i32) -> vector<16xf32>
94 amdgpu.mfma %arg16 * %arg16 + %arg2 { abid = 0 : i32, cbsz = 0 : i32, k = 16 : i32, m = 32 : i32, n = 32 : i32, blocks = 1 : i32 } blgp = none : vector<8xf8E4M3FNUZ>, vector<8xf8E4M3FNUZ>, vector<16xf32>
96 func.return
97 }