brintos

brintos / llvm-project-archived public Read only

0
0
Text · 4.2 KiB · ad01450 Raw
105 lines · plain
1// RUN: %clang_cc1 -triple amdgcn-unknown-unknown -target-cpu gfx1200 -target-feature +wavefrontsize64 -S -o - %s | FileCheck %s --check-prefix=CHECK-GFX12002 3typedef float  v4f   __attribute__((ext_vector_type(4)));4typedef half   v4h   __attribute__((ext_vector_type(4)));5typedef short  v4s   __attribute__((ext_vector_type(4)));6typedef int    v4i   __attribute__((ext_vector_type(4)));7 8// Wave649 10 11// CHECK-GFX1200-LABEL: test_amdgcn_wmma_f32_16x16x16_f16_w64:12// CHECK-GFX1200: v_wmma_f32_16x16x16_f16 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}]13//14void test_amdgcn_wmma_f32_16x16x16_f16_w64(global v4f* out, v4h a, v4h b, v4f c)15{16  *out = __builtin_amdgcn_wmma_f32_16x16x16_f16_w64_gfx12(a, b, c);17}18 19 20// CHECK-GFX1200-LABEL: test_amdgcn_wmma_f32_16x16x16_bf16_w64:21// CHECK-GFX1200: v_wmma_f32_16x16x16_bf16 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}]22//23void test_amdgcn_wmma_f32_16x16x16_bf16_w64(global v4f* out, v4s a, v4s b, v4f c)24{25  *out = __builtin_amdgcn_wmma_f32_16x16x16_bf16_w64_gfx12(a, b, c);26}27 28 29// CHECK-GFX1200-LABEL: test_amdgcn_wmma_f16_16x16x16_f16_w64:30// CHECK-GFX1200: v_wmma_f16_16x16x16_f16 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}]31//32void test_amdgcn_wmma_f16_16x16x16_f16_w64(global v4h* out, v4h a, v4h b, v4h c)33{34  *out = __builtin_amdgcn_wmma_f16_16x16x16_f16_w64_gfx12(a, b, c);35}36 37 38// CHECK-GFX1200-LABEL: test_amdgcn_wmma_bf16_16x16x16_bf16_w64:39// CHECK-GFX1200: v_wmma_bf16_16x16x16_bf16 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}]40//41void test_amdgcn_wmma_bf16_16x16x16_bf16_w64(global v4s* out, v4s a, v4s b, v4s c)42{43  *out = __builtin_amdgcn_wmma_bf16_16x16x16_bf16_w64_gfx12(a, b, c);44}45 46 47// CHECK-GFX1200-LABEL: test_amdgcn_wmma_i32_16x16x16_iu8_w64:48// CHECK-GFX1200: v_wmma_i32_16x16x16_iu8 v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, v{{[0-9]+}}, v[{{[0-9]+:[0-9]+}}] neg_lo:[1,1,0]49//50void test_amdgcn_wmma_i32_16x16x16_iu8_w64(global v4i* out, int a, int b, v4i c)51{52  *out = __builtin_amdgcn_wmma_i32_16x16x16_iu8_w64_gfx12(true, a, true, b, c, false);53}54 55 56// CHECK-GFX1200-LABEL: test_amdgcn_wmma_i32_16x16x16_iu4_w64:57// CHECK-GFX1200: v_wmma_i32_16x16x16_iu4 v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, v{{[0-9]+}}, v[{{[0-9]+:[0-9]+}}] neg_lo:[1,1,0]58//59void test_amdgcn_wmma_i32_16x16x16_iu4_w64(global v4i* out, int a, int b, v4i c)60{61  *out = __builtin_amdgcn_wmma_i32_16x16x16_iu4_w64_gfx12(true, a, true, b, c, false);62}63 64 65// CHECK-GFX1200-LABEL: test_amdgcn_wmma_f32_16x16x16_fp8_fp8_w32:66// CHECK-GFX1200: v_wmma_f32_16x16x16_fp8_fp8 v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, v{{[0-9]+}}, v[{{[0-9]+:[0-9]+}}]67//68void test_amdgcn_wmma_f32_16x16x16_fp8_fp8_w32(global v4f* out, int a, int b, v4f c)69{70  *out = __builtin_amdgcn_wmma_f32_16x16x16_fp8_fp8_w64_gfx12(a, b, c);71}72 73 74// CHECK-GFX1200-LABEL: test_amdgcn_wmma_f32_16x16x16_fp8_bf8_w32:75// CHECK-GFX1200: v_wmma_f32_16x16x16_fp8_bf8 v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, v{{[0-9]+}}, v[{{[0-9]+:[0-9]+}}]76//77void test_amdgcn_wmma_f32_16x16x16_fp8_bf8_w32(global v4f* out, int a, int b, v4f c)78{79  *out = __builtin_amdgcn_wmma_f32_16x16x16_fp8_bf8_w64_gfx12(a, b, c);80}81 82// CHECK-GFX1200-LABEL: test_amdgcn_wmma_f32_16x16x16_bf8_fp8_w32:83// CHECK-GFX1200: v_wmma_f32_16x16x16_bf8_fp8 v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, v{{[0-9]+}}, v[{{[0-9]+:[0-9]+}}]84//85void test_amdgcn_wmma_f32_16x16x16_bf8_fp8_w32(global v4f* out, int a, int b, v4f c)86{87  *out = __builtin_amdgcn_wmma_f32_16x16x16_bf8_fp8_w64_gfx12(a, b, c);88}89 90// CHECK-GFX1200-LABEL: test_amdgcn_wmma_f32_16x16x16_bf8_bf8_w32:91// CHECK-GFX1200: v_wmma_f32_16x16x16_bf8_bf8 v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, v{{[0-9]+}}, v[{{[0-9]+:[0-9]+}}]92//93void test_amdgcn_wmma_f32_16x16x16_bf8_bf8_w32(global v4f* out, int a, int b, v4f c)94{95  *out = __builtin_amdgcn_wmma_f32_16x16x16_bf8_bf8_w64_gfx12(a, b, c);96}97 98// CHECK-GFX1200-LABEL: test_amdgcn_wmma_i32_16x16x32_iu4_w32:99// CHECK-GFX1200: v_wmma_i32_16x16x32_iu4 v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, v{{[0-9]+}}, v[{{[0-9]+:[0-9]+}}] neg_lo:[1,1,0]100//101void test_amdgcn_wmma_i32_16x16x32_iu4_w32(global v4i* out, int a, int b, v4i c)102{103  *out = __builtin_amdgcn_wmma_i32_16x16x32_iu4_w64_gfx12(true, a, true, b, c, false);104}105