llvm-project

1
// RUN: %clang_cc1 -triple amdgcn-unknown-unknown -target-cpu gfx1200 -target-feature +wavefrontsize32 -S -o - %s | FileCheck %s --check-prefix=CHECK-GFX1200
2

3
typedef int    v2i   __attribute__((ext_vector_type(2)));
4
typedef int    v4i   __attribute__((ext_vector_type(4)));
5
typedef float  v8f   __attribute__((ext_vector_type(8)));
6
typedef half   v8h   __attribute__((ext_vector_type(8)));
7
typedef short  v8s   __attribute__((ext_vector_type(8)));
8
typedef int    v8i   __attribute__((ext_vector_type(8)));
9
typedef half  v16h   __attribute__((ext_vector_type(16)));
10
typedef short v16s   __attribute__((ext_vector_type(16)));
11

12
// Wave32
13

14

15
// CHECK-GFX1200-LABEL: test_amdgcn_swmmac_f32_16x16x32_f16_w32:
16
// CHECK-GFX1200: v_swmmac_f32_16x16x32_f16 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}
17
//
18
void test_amdgcn_swmmac_f32_16x16x32_f16_w32(global v8f* out, v8h a, v16h b, v8f c, short index)
19
{
20
  *out = __builtin_amdgcn_swmmac_f32_16x16x32_f16_w32(a, b, c, index);
21
}
22

23

24
// CHECK-GFX1200-LABEL: test_amdgcn_swmmac_f32_16x16x32_bf16_w32:
25
// CHECK-GFX1200: v_swmmac_f32_16x16x32_bf16 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}
26
//
27
void test_amdgcn_swmmac_f32_16x16x32_bf16_w32(global v8f* out, v8s a, v16s b, v8f c, short index)
28
{
29
  *out = __builtin_amdgcn_swmmac_f32_16x16x32_bf16_w32(a, b, c, index);
30
}
31

32

33
// CHECK-GFX1200-LABEL: test_amdgcn_swmmac_f16_16x16x32_f16_w32:
34
// CHECK-GFX1200: v_swmmac_f16_16x16x32_f16 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}
35
//
36
void test_amdgcn_swmmac_f16_16x16x32_f16_w32(global v8h* out, v8h a, v16h b, v8h c, short index)
37
{
38
  *out = __builtin_amdgcn_swmmac_f16_16x16x32_f16_w32(a, b, c, index);
39
}
40

41

42
// CHECK-GFX1200-LABEL: test_amdgcn_swmmac_bf16_16x16x32_bf16_w32:
43
// CHECK-GFX1200: v_swmmac_bf16_16x16x32_bf16 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}
44
//
45
void test_amdgcn_swmmac_bf16_16x16x32_bf16_w32(global v8s* out, v8s a, v16s b, v8s c, short index)
46
{
47
  *out = __builtin_amdgcn_swmmac_bf16_16x16x32_bf16_w32(a, b, c, index);
48
}
49

50

51
// CHECK-GFX1200-LABEL: test_amdgcn_swmmac_i32_16x16x32_iu8_w32:
52
// CHECK-GFX1200: v_swmmac_i32_16x16x32_iu8 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} neg_lo:[1,1,0] clamp
53
//
54
void test_amdgcn_swmmac_i32_16x16x32_iu8_w32(global v8i* out, v2i a, v4i b, v8i c, short index)
55
{
56
  *out = __builtin_amdgcn_swmmac_i32_16x16x32_iu8_w32(true, a, true, b, c, index, true);
57
}
58

59

60
// CHECK-GFX1200-LABEL: test_amdgcn_swmmac_i32_16x16x32_iu4_w32:
61
// CHECK-GFX1200: v_swmmac_i32_16x16x32_iu4 v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} neg_lo:[1,1,0] clamp
62
//
63
void test_amdgcn_swmmac_i32_16x16x32_iu4_w32(global v8i* out, int a, v2i b, v8i c, short index)
64
{
65
  *out = __builtin_amdgcn_swmmac_i32_16x16x32_iu4_w32(true, a, true, b, c, index, true);
66
}
67

68

69
// CHECK-GFX1200-LABEL: test_amdgcn_swmmac_i32_16x16x64_iu4_w32:
70
// CHECK-GFX1200: v_swmmac_i32_16x16x64_iu4 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} neg_lo:[1,1,0] clamp
71
//
72
void test_amdgcn_swmmac_i32_16x16x64_iu4_w32(global v8i* out, v2i a, v4i b, v8i c, short index)
73
{
74
  *out = __builtin_amdgcn_swmmac_i32_16x16x64_iu4_w32(true, a, true, b, c, index, true);
75
}
76

77

78
// CHECK-GFX1200-LABEL: test_amdgcn_swmmac_f32_16x16x32_fp8_fp8_w32:
79
// CHECK-GFX1200: v_swmmac_f32_16x16x32_fp8_fp8 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}
80
//
81
void test_amdgcn_swmmac_f32_16x16x32_fp8_fp8_w32(global v8f* out, v2i a, v4i b, v8f c, short index)
82
{
83
  *out = __builtin_amdgcn_swmmac_f32_16x16x32_fp8_fp8_w32(a, b, c, index);
84
}
85

86

87
// CHECK-GFX1200-LABEL: test_amdgcn_swmmac_f32_16x16x32_fp8_bf8_w32:
88
// CHECK-GFX1200: v_swmmac_f32_16x16x32_fp8_bf8 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}
89
//
90
void test_amdgcn_swmmac_f32_16x16x32_fp8_bf8_w32(global v8f* out, v2i a, v4i b, v8f c, short index)
91
{
92
  *out = __builtin_amdgcn_swmmac_f32_16x16x32_fp8_bf8_w32(a, b, c, index);
93
}
94

95

96
// CHECK-GFX1200-LABEL: test_amdgcn_swmmac_f32_16x16x32_bf8_fp8_w32:
97
// CHECK-GFX1200: v_swmmac_f32_16x16x32_bf8_fp8 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}
98
//
99
void test_amdgcn_swmmac_f32_16x16x32_bf8_fp8_w32(global v8f* out, v2i a, v4i b, v8f c, short index)
100
{
101
  *out = __builtin_amdgcn_swmmac_f32_16x16x32_bf8_fp8_w32(a, b, c, index);
102
}
103

104
// CHECK-GFX1200-LABEL: test_amdgcn_swmmac_f32_16x16x32_bf8_bf8_w32:
105
// CHECK-GFX1200: v_swmmac_f32_16x16x32_bf8_bf8 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}
106
//
107
void test_amdgcn_swmmac_f32_16x16x32_bf8_bf8_w32(global v8f* out, v2i a, v4i b, v8f c, short index)
108
{
109
  *out = __builtin_amdgcn_swmmac_f32_16x16x32_bf8_bf8_w32(a, b, c, index);
110
}
111