Revert "[llvm] Improve llvm.objectsize computation by computing GEP, alloca and mallo...
[llvm-project.git] / clang / test / CodeGenOpenCL / builtins-amdgcn-wmma-w32.cl
blob2f9a367ecab8a1236050fd32b317364e85ec1164
1 // NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
2 // REQUIRES: amdgpu-registered-target
3 // RUN: %clang_cc1 -triple amdgcn-unknown-unknown -target-cpu gfx1100 -DWMMA_GFX1100_TESTS -emit-llvm -o - %s | FileCheck %s --check-prefix=CHECK-GFX1100
5 typedef float v8f __attribute__((ext_vector_type(8)));
6 typedef half v16h __attribute__((ext_vector_type(16)));
7 typedef int v2i __attribute__((ext_vector_type(2)));
8 typedef int v4i __attribute__((ext_vector_type(4)));
9 typedef int v8i __attribute__((ext_vector_type(8)));
10 typedef short v16s __attribute__((ext_vector_type(16)));
12 #ifdef WMMA_GFX1100_TESTS
14 // Wave32
17 // amdgcn_wmma_f32_16x16x16_f16
20 // CHECK-GFX1100-LABEL: @test_amdgcn_wmma_f32_16x16x16_f16_w32(
21 // CHECK-GFX1100-NEXT: entry:
22 // CHECK-GFX1100-NEXT: [[TMP0:%.*]] = tail call <8 x float> @llvm.amdgcn.wmma.f32.16x16x16.f16.v8f32.v16f16(<16 x half> [[A:%.*]], <16 x half> [[B:%.*]], <8 x float> [[C:%.*]])
23 // CHECK-GFX1100-NEXT: store <8 x float> [[TMP0]], ptr addrspace(1) [[OUT:%.*]], align 32, !tbaa [[TBAA4:![0-9]+]]
24 // CHECK-GFX1100-NEXT: ret void
26 void test_amdgcn_wmma_f32_16x16x16_f16_w32(global v8f* out, v16h a, v16h b, v8f c)
28 *out = __builtin_amdgcn_wmma_f32_16x16x16_f16_w32(a, b, c);
32 // amdgcn_wmma_f32_16x16x16_bf16
35 // CHECK-GFX1100-LABEL: @test_amdgcn_wmma_f32_16x16x16_bf16_w32(
36 // CHECK-GFX1100-NEXT: entry:
37 // CHECK-GFX1100-NEXT: [[TMP0:%.*]] = tail call <8 x float> @llvm.amdgcn.wmma.f32.16x16x16.bf16.v8f32.v16i16(<16 x i16> [[A:%.*]], <16 x i16> [[B:%.*]], <8 x float> [[C:%.*]])
38 // CHECK-GFX1100-NEXT: store <8 x float> [[TMP0]], ptr addrspace(1) [[OUT:%.*]], align 32, !tbaa [[TBAA4]]
39 // CHECK-GFX1100-NEXT: ret void
41 void test_amdgcn_wmma_f32_16x16x16_bf16_w32(global v8f* out, v16s a, v16s b, v8f c)
43 *out = __builtin_amdgcn_wmma_f32_16x16x16_bf16_w32(a, b, c);
47 // amdgcn_wmma_f16_16x16x16_f16
50 // CHECK-GFX1100-LABEL: @test_amdgcn_wmma_f16_16x16x16_f16_w32(
51 // CHECK-GFX1100-NEXT: entry:
52 // CHECK-GFX1100-NEXT: [[TMP0:%.*]] = tail call <16 x half> @llvm.amdgcn.wmma.f16.16x16x16.f16.v16f16.v16f16(<16 x half> [[A:%.*]], <16 x half> [[B:%.*]], <16 x half> [[C:%.*]], i1 true)
53 // CHECK-GFX1100-NEXT: store <16 x half> [[TMP0]], ptr addrspace(1) [[OUT:%.*]], align 32, !tbaa [[TBAA4]]
54 // CHECK-GFX1100-NEXT: ret void
56 void test_amdgcn_wmma_f16_16x16x16_f16_w32(global v16h* out, v16h a, v16h b, v16h c)
58 *out = __builtin_amdgcn_wmma_f16_16x16x16_f16_w32(a, b, c, true);
62 // amdgcn_wmma_bf16_16x16x16_bf16
65 // CHECK-GFX1100-LABEL: @test_amdgcn_wmma_bf16_16x16x16_bf16_w32(
66 // CHECK-GFX1100-NEXT: entry:
67 // CHECK-GFX1100-NEXT: [[TMP0:%.*]] = tail call <16 x i16> @llvm.amdgcn.wmma.bf16.16x16x16.bf16.v16i16.v16i16(<16 x i16> [[A:%.*]], <16 x i16> [[B:%.*]], <16 x i16> [[C:%.*]], i1 true)
68 // CHECK-GFX1100-NEXT: store <16 x i16> [[TMP0]], ptr addrspace(1) [[OUT:%.*]], align 32, !tbaa [[TBAA4]]
69 // CHECK-GFX1100-NEXT: ret void
71 void test_amdgcn_wmma_bf16_16x16x16_bf16_w32(global v16s* out, v16s a, v16s b, v16s c)
73 *out = __builtin_amdgcn_wmma_bf16_16x16x16_bf16_w32(a, b, c, true);
77 // amdgcn_wmma_f16_16x16x16_f16_tied
80 // CHECK-GFX1100-LABEL: @test_amdgcn_wmma_f16_16x16x16_f16_tied_w32(
81 // CHECK-GFX1100-NEXT: entry:
82 // CHECK-GFX1100-NEXT: [[TMP0:%.*]] = tail call <16 x half> @llvm.amdgcn.wmma.f16.16x16x16.f16.tied.v16f16.v16f16(<16 x half> [[A:%.*]], <16 x half> [[B:%.*]], <16 x half> [[C:%.*]], i1 true)
83 // CHECK-GFX1100-NEXT: store <16 x half> [[TMP0]], ptr addrspace(1) [[OUT:%.*]], align 32, !tbaa [[TBAA4]]
84 // CHECK-GFX1100-NEXT: ret void
86 void test_amdgcn_wmma_f16_16x16x16_f16_tied_w32(global v16h* out, v16h a, v16h b, v16h c)
88 *out = __builtin_amdgcn_wmma_f16_16x16x16_f16_tied_w32(a, b, c, true);
92 // amdgcn_wmma_bf16_16x16x16_bf16_tied
95 // CHECK-GFX1100-LABEL: @test_amdgcn_wmma_bf16_16x16x16_bf16_tied_w32(
96 // CHECK-GFX1100-NEXT: entry:
97 // CHECK-GFX1100-NEXT: [[TMP0:%.*]] = tail call <16 x i16> @llvm.amdgcn.wmma.bf16.16x16x16.bf16.tied.v16i16.v16i16(<16 x i16> [[A:%.*]], <16 x i16> [[B:%.*]], <16 x i16> [[C:%.*]], i1 true)
98 // CHECK-GFX1100-NEXT: store <16 x i16> [[TMP0]], ptr addrspace(1) [[OUT:%.*]], align 32, !tbaa [[TBAA4]]
99 // CHECK-GFX1100-NEXT: ret void
101 void test_amdgcn_wmma_bf16_16x16x16_bf16_tied_w32(global v16s* out, v16s a, v16s b, v16s c)
103 *out = __builtin_amdgcn_wmma_bf16_16x16x16_bf16_tied_w32(a, b, c, true);
107 // amdgcn_wmma_i32_16x16x16_iu8
110 // CHECK-GFX1100-LABEL: @test_amdgcn_wmma_i32_16x16x16_iu8_w32(
111 // CHECK-GFX1100-NEXT: entry:
112 // CHECK-GFX1100-NEXT: [[TMP0:%.*]] = tail call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8.v8i32.v4i32(i1 true, <4 x i32> [[A:%.*]], i1 true, <4 x i32> [[B:%.*]], <8 x i32> [[C:%.*]], i1 false)
113 // CHECK-GFX1100-NEXT: store <8 x i32> [[TMP0]], ptr addrspace(1) [[OUT:%.*]], align 32, !tbaa [[TBAA4]]
114 // CHECK-GFX1100-NEXT: ret void
116 void test_amdgcn_wmma_i32_16x16x16_iu8_w32(global v8i* out, v4i a, v4i b, v8i c)
118 *out = __builtin_amdgcn_wmma_i32_16x16x16_iu8_w32(true, a, true, b, c, false);
122 // amdgcn_wmma_i32_16x16x16_iu4
125 // CHECK-GFX1100-LABEL: @test_amdgcn_wmma_i32_16x16x16_iu4_w32(
126 // CHECK-GFX1100-NEXT: entry:
127 // CHECK-GFX1100-NEXT: [[TMP0:%.*]] = tail call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4.v8i32.v2i32(i1 true, <2 x i32> [[A:%.*]], i1 true, <2 x i32> [[B:%.*]], <8 x i32> [[C:%.*]], i1 false)
128 // CHECK-GFX1100-NEXT: store <8 x i32> [[TMP0]], ptr addrspace(1) [[OUT:%.*]], align 32, !tbaa [[TBAA4]]
129 // CHECK-GFX1100-NEXT: ret void
131 void test_amdgcn_wmma_i32_16x16x16_iu4_w32(global v8i* out, v2i a, v2i b, v8i c)
133 *out = __builtin_amdgcn_wmma_i32_16x16x16_iu4_w32(true, a, true, b, c, false);
136 #endif