[clang][modules] Don't prevent translation of FW_Private includes when explicitly...
[llvm-project.git] / clang / test / CodeGenOpenCL / builtins-amdgcn-wmma-w32.cl
blob07e4f71f6dbd32a8c05afbc31f86e491f99543c2
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 -S -emit-llvm -o - %s | FileCheck %s --check-prefix=CHECK-GFX1100
5 typedef float v4f __attribute__((ext_vector_type(4)));
6 typedef float v8f __attribute__((ext_vector_type(8)));
7 typedef half v16h __attribute__((ext_vector_type(16)));
8 typedef int v2i __attribute__((ext_vector_type(2)));
9 typedef int v4i __attribute__((ext_vector_type(4)));
10 typedef int v8i __attribute__((ext_vector_type(8)));
11 typedef short v16s __attribute__((ext_vector_type(16)));
13 #ifdef WMMA_GFX1100_TESTS
15 // Wave32
18 // amdgcn_wmma_f32_16x16x16_f16
21 // CHECK-GFX1100-LABEL: @test_amdgcn_wmma_f32_16x16x16_f16_w32(
22 // CHECK-GFX1100-NEXT: entry:
23 // CHECK-GFX1100-NEXT: [[TMP0:%.*]] = tail call <8 x float> @llvm.amdgcn.wmma.f32.16x16x16.f16.v8f32(<16 x half> [[A:%.*]], <16 x half> [[B:%.*]], <8 x float> [[C:%.*]])
24 // CHECK-GFX1100-NEXT: store <8 x float> [[TMP0]], ptr addrspace(1) [[OUT:%.*]], align 32, !tbaa [[TBAA4:![0-9]+]]
25 // CHECK-GFX1100-NEXT: ret void
27 void test_amdgcn_wmma_f32_16x16x16_f16_w32(global v8f* out, v16h a, v16h b, v8f c)
29 *out = __builtin_amdgcn_wmma_f32_16x16x16_f16_w32(a, b, c);
33 // amdgcn_wmma_f32_16x16x16_bf16
36 // CHECK-GFX1100-LABEL: @test_amdgcn_wmma_f32_16x16x16_bf16_w32(
37 // CHECK-GFX1100-NEXT: entry:
38 // CHECK-GFX1100-NEXT: [[TMP0:%.*]] = tail call <8 x float> @llvm.amdgcn.wmma.f32.16x16x16.bf16.v8f32(<16 x i16> [[A:%.*]], <16 x i16> [[B:%.*]], <8 x float> [[C:%.*]])
39 // CHECK-GFX1100-NEXT: store <8 x float> [[TMP0]], ptr addrspace(1) [[OUT:%.*]], align 32, !tbaa [[TBAA4]]
40 // CHECK-GFX1100-NEXT: ret void
42 void test_amdgcn_wmma_f32_16x16x16_bf16_w32(global v8f* out, v16s a, v16s b, v8f c)
44 *out = __builtin_amdgcn_wmma_f32_16x16x16_bf16_w32(a, b, c);
48 // amdgcn_wmma_f16_16x16x16_f16
51 // CHECK-GFX1100-LABEL: @test_amdgcn_wmma_f16_16x16x16_f16_w32(
52 // CHECK-GFX1100-NEXT: entry:
53 // CHECK-GFX1100-NEXT: [[TMP0:%.*]] = tail call <16 x half> @llvm.amdgcn.wmma.f16.16x16x16.f16.v16f16(<16 x half> [[A:%.*]], <16 x half> [[B:%.*]], <16 x half> [[C:%.*]], i1 true)
54 // CHECK-GFX1100-NEXT: store <16 x half> [[TMP0]], ptr addrspace(1) [[OUT:%.*]], align 32, !tbaa [[TBAA4]]
55 // CHECK-GFX1100-NEXT: ret void
57 void test_amdgcn_wmma_f16_16x16x16_f16_w32(global v16h* out, v16h a, v16h b, v16h c)
59 *out = __builtin_amdgcn_wmma_f16_16x16x16_f16_w32(a, b, c, true);
63 // amdgcn_wmma_bf16_16x16x16_bf16
66 // CHECK-GFX1100-LABEL: @test_amdgcn_wmma_bf16_16x16x16_bf16_w32(
67 // CHECK-GFX1100-NEXT: entry:
68 // CHECK-GFX1100-NEXT: [[TMP0:%.*]] = tail call <16 x i16> @llvm.amdgcn.wmma.bf16.16x16x16.bf16.v16i16(<16 x i16> [[A:%.*]], <16 x i16> [[B:%.*]], <16 x i16> [[C:%.*]], i1 true)
69 // CHECK-GFX1100-NEXT: store <16 x i16> [[TMP0]], ptr addrspace(1) [[OUT:%.*]], align 32, !tbaa [[TBAA4]]
70 // CHECK-GFX1100-NEXT: ret void
72 void test_amdgcn_wmma_bf16_16x16x16_bf16_w32(global v16s* out, v16s a, v16s b, v16s c)
74 *out = __builtin_amdgcn_wmma_bf16_16x16x16_bf16_w32(a, b, c, true);
78 // amdgcn_wmma_i32_16x16x16_iu8
81 // CHECK-GFX1100-LABEL: @test_amdgcn_wmma_i32_16x16x16_iu8_w32(
82 // CHECK-GFX1100-NEXT: entry:
83 // CHECK-GFX1100-NEXT: [[TMP0:%.*]] = tail call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8.v8i32(i1 true, <4 x i32> [[A:%.*]], i1 true, <4 x i32> [[B:%.*]], <8 x i32> [[C:%.*]], i1 false)
84 // CHECK-GFX1100-NEXT: store <8 x i32> [[TMP0]], ptr addrspace(1) [[OUT:%.*]], align 32, !tbaa [[TBAA4]]
85 // CHECK-GFX1100-NEXT: ret void
87 void test_amdgcn_wmma_i32_16x16x16_iu8_w32(global v8i* out, v4i a, v4i b, v8i c)
89 *out = __builtin_amdgcn_wmma_i32_16x16x16_iu8_w32(true, a, true, b, c, false);
93 // amdgcn_wmma_i32_16x16x16_iu4
96 // CHECK-GFX1100-LABEL: @test_amdgcn_wmma_i32_16x16x16_iu4_w32(
97 // CHECK-GFX1100-NEXT: entry:
98 // CHECK-GFX1100-NEXT: [[TMP0:%.*]] = tail call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4.v8i32(i1 true, <2 x i32> [[A:%.*]], i1 true, <2 x i32> [[B:%.*]], <8 x i32> [[C:%.*]], i1 false)
99 // CHECK-GFX1100-NEXT: store <8 x i32> [[TMP0]], ptr addrspace(1) [[OUT:%.*]], align 32, !tbaa [[TBAA4]]
100 // CHECK-GFX1100-NEXT: ret void
102 void test_amdgcn_wmma_i32_16x16x16_iu4_w32(global v8i* out, v2i a, v2i b, v8i c)
104 *out = __builtin_amdgcn_wmma_i32_16x16x16_iu4_w32(true, a, true, b, c, false);
107 #endif