Matt Arsenault | 3f71c0e | 2017-11-29 00:55:57 +0000 | [diff] [blame^] | 1 | ; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,SICIVI,FUNC %s |
| 2 | ; RUN: llc -mtriple=amdgcn--amdhsa -mcpu=kaveri -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,SICIVI,FUNC %s |
| 3 | ; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,SICIVI,FUNC %s |
| 4 | ; RUN: llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX9,FUNC %s |
| 5 | ; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck -check-prefixes=EG,FUNC %s |
Matt Arsenault | d1097a3 | 2016-06-02 19:54:26 +0000 | [diff] [blame] | 6 | |
| 7 | ; FUNC-LABEL: {{^}}local_load_f64: |
Matt Arsenault | 3f71c0e | 2017-11-29 00:55:57 +0000 | [diff] [blame^] | 8 | ; SICIV: s_mov_b32 m0 |
| 9 | ; GFX9-NOT: m0 |
| 10 | |
Matt Arsenault | d1097a3 | 2016-06-02 19:54:26 +0000 | [diff] [blame] | 11 | ; GCN: ds_read_b64 [[VAL:v\[[0-9]+:[0-9]+\]]], v{{[0-9]+}}{{$}} |
| 12 | ; GCN: ds_write_b64 v{{[0-9]+}}, [[VAL]] |
| 13 | |
| 14 | ; EG: LDS_READ_RET |
| 15 | ; EG: LDS_READ_RET |
Matt Arsenault | 3dbeefa | 2017-03-21 21:39:51 +0000 | [diff] [blame] | 16 | define amdgpu_kernel void @local_load_f64(double addrspace(3)* %out, double addrspace(3)* %in) #0 { |
Matt Arsenault | d1097a3 | 2016-06-02 19:54:26 +0000 | [diff] [blame] | 17 | %ld = load double, double addrspace(3)* %in |
| 18 | store double %ld, double addrspace(3)* %out |
| 19 | ret void |
| 20 | } |
| 21 | |
| 22 | ; FUNC-LABEL: {{^}}local_load_v2f64: |
Matt Arsenault | 3f71c0e | 2017-11-29 00:55:57 +0000 | [diff] [blame^] | 23 | ; SICIV: s_mov_b32 m0 |
| 24 | ; GFX9-NOT: m0 |
| 25 | |
Matt Arsenault | d1097a3 | 2016-06-02 19:54:26 +0000 | [diff] [blame] | 26 | ; GCN: ds_read2_b64 |
| 27 | |
| 28 | ; EG: LDS_READ_RET |
| 29 | ; EG: LDS_READ_RET |
| 30 | ; EG: LDS_READ_RET |
| 31 | ; EG: LDS_READ_RET |
Matt Arsenault | 3dbeefa | 2017-03-21 21:39:51 +0000 | [diff] [blame] | 32 | define amdgpu_kernel void @local_load_v2f64(<2 x double> addrspace(3)* %out, <2 x double> addrspace(3)* %in) #0 { |
Matt Arsenault | d1097a3 | 2016-06-02 19:54:26 +0000 | [diff] [blame] | 33 | entry: |
| 34 | %ld = load <2 x double>, <2 x double> addrspace(3)* %in |
| 35 | store <2 x double> %ld, <2 x double> addrspace(3)* %out |
| 36 | ret void |
| 37 | } |
| 38 | |
| 39 | ; FUNC-LABEL: {{^}}local_load_v3f64: |
Matt Arsenault | 3f71c0e | 2017-11-29 00:55:57 +0000 | [diff] [blame^] | 40 | ; SICIV: s_mov_b32 m0 |
| 41 | ; GFX9-NOT: m0 |
| 42 | |
Matt Arsenault | d1097a3 | 2016-06-02 19:54:26 +0000 | [diff] [blame] | 43 | ; GCN-DAG: ds_read2_b64 |
| 44 | ; GCN-DAG: ds_read_b64 |
| 45 | |
| 46 | ; EG: LDS_READ_RET |
| 47 | ; EG: LDS_READ_RET |
| 48 | ; EG: LDS_READ_RET |
| 49 | ; EG: LDS_READ_RET |
| 50 | ; EG: LDS_READ_RET |
| 51 | ; EG: LDS_READ_RET |
Matt Arsenault | 3dbeefa | 2017-03-21 21:39:51 +0000 | [diff] [blame] | 52 | define amdgpu_kernel void @local_load_v3f64(<3 x double> addrspace(3)* %out, <3 x double> addrspace(3)* %in) #0 { |
Matt Arsenault | d1097a3 | 2016-06-02 19:54:26 +0000 | [diff] [blame] | 53 | entry: |
| 54 | %ld = load <3 x double>, <3 x double> addrspace(3)* %in |
| 55 | store <3 x double> %ld, <3 x double> addrspace(3)* %out |
| 56 | ret void |
| 57 | } |
| 58 | |
| 59 | ; FUNC-LABEL: {{^}}local_load_v4f64: |
Matt Arsenault | 3f71c0e | 2017-11-29 00:55:57 +0000 | [diff] [blame^] | 60 | ; SICIV: s_mov_b32 m0 |
| 61 | ; GFX9-NOT: m0 |
| 62 | |
Matt Arsenault | d1097a3 | 2016-06-02 19:54:26 +0000 | [diff] [blame] | 63 | ; GCN: ds_read2_b64 |
| 64 | ; GCN: ds_read2_b64 |
| 65 | |
| 66 | ; EG: LDS_READ_RET |
| 67 | ; EG: LDS_READ_RET |
| 68 | ; EG: LDS_READ_RET |
| 69 | ; EG: LDS_READ_RET |
| 70 | |
| 71 | ; EG: LDS_READ_RET |
| 72 | ; EG: LDS_READ_RET |
| 73 | ; EG: LDS_READ_RET |
| 74 | ; EG: LDS_READ_RET |
Matt Arsenault | 3dbeefa | 2017-03-21 21:39:51 +0000 | [diff] [blame] | 75 | define amdgpu_kernel void @local_load_v4f64(<4 x double> addrspace(3)* %out, <4 x double> addrspace(3)* %in) #0 { |
Matt Arsenault | d1097a3 | 2016-06-02 19:54:26 +0000 | [diff] [blame] | 76 | entry: |
| 77 | %ld = load <4 x double>, <4 x double> addrspace(3)* %in |
| 78 | store <4 x double> %ld, <4 x double> addrspace(3)* %out |
| 79 | ret void |
| 80 | } |
| 81 | |
| 82 | ; FUNC-LABEL: {{^}}local_load_v8f64: |
Matt Arsenault | 3f71c0e | 2017-11-29 00:55:57 +0000 | [diff] [blame^] | 83 | ; SICIV: s_mov_b32 m0 |
| 84 | ; GFX9-NOT: m0 |
| 85 | |
Matt Arsenault | d1097a3 | 2016-06-02 19:54:26 +0000 | [diff] [blame] | 86 | ; GCN: ds_read2_b64 |
| 87 | ; GCN: ds_read2_b64 |
| 88 | ; GCN: ds_read2_b64 |
| 89 | ; GCN: ds_read2_b64 |
| 90 | |
| 91 | ; EG: LDS_READ_RET |
| 92 | ; EG: LDS_READ_RET |
| 93 | ; EG: LDS_READ_RET |
| 94 | ; EG: LDS_READ_RET |
| 95 | ; EG: LDS_READ_RET |
| 96 | ; EG: LDS_READ_RET |
| 97 | ; EG: LDS_READ_RET |
| 98 | ; EG: LDS_READ_RET |
| 99 | ; EG: LDS_READ_RET |
| 100 | ; EG: LDS_READ_RET |
| 101 | ; EG: LDS_READ_RET |
| 102 | ; EG: LDS_READ_RET |
| 103 | ; EG: LDS_READ_RET |
| 104 | ; EG: LDS_READ_RET |
| 105 | ; EG: LDS_READ_RET |
| 106 | ; EG: LDS_READ_RET |
Matt Arsenault | 3dbeefa | 2017-03-21 21:39:51 +0000 | [diff] [blame] | 107 | define amdgpu_kernel void @local_load_v8f64(<8 x double> addrspace(3)* %out, <8 x double> addrspace(3)* %in) #0 { |
Matt Arsenault | d1097a3 | 2016-06-02 19:54:26 +0000 | [diff] [blame] | 108 | entry: |
| 109 | %ld = load <8 x double>, <8 x double> addrspace(3)* %in |
| 110 | store <8 x double> %ld, <8 x double> addrspace(3)* %out |
| 111 | ret void |
| 112 | } |
| 113 | |
| 114 | ; FUNC-LABEL: {{^}}local_load_v16f64: |
Matt Arsenault | 3f71c0e | 2017-11-29 00:55:57 +0000 | [diff] [blame^] | 115 | ; SICIV: s_mov_b32 m0 |
| 116 | ; GFX9-NOT: m0 |
| 117 | |
Matt Arsenault | d1097a3 | 2016-06-02 19:54:26 +0000 | [diff] [blame] | 118 | ; GCN: ds_read2_b64 |
| 119 | ; GCN: ds_read2_b64 |
| 120 | ; GCN: ds_read2_b64 |
| 121 | ; GCN: ds_read2_b64 |
| 122 | ; GCN: ds_read2_b64 |
| 123 | ; GCN: ds_read2_b64 |
| 124 | ; GCN: ds_read2_b64 |
| 125 | ; GCN: ds_read2_b64 |
| 126 | |
| 127 | ; EG: LDS_READ_RET |
| 128 | ; EG: LDS_READ_RET |
| 129 | ; EG: LDS_READ_RET |
| 130 | ; EG: LDS_READ_RET |
| 131 | |
| 132 | ; EG: LDS_READ_RET |
| 133 | ; EG: LDS_READ_RET |
| 134 | ; EG: LDS_READ_RET |
| 135 | ; EG: LDS_READ_RET |
| 136 | |
| 137 | ; EG: LDS_READ_RET |
| 138 | ; EG: LDS_READ_RET |
| 139 | ; EG: LDS_READ_RET |
| 140 | ; EG: LDS_READ_RET |
| 141 | |
| 142 | ; EG: LDS_READ_RET |
| 143 | ; EG: LDS_READ_RET |
| 144 | ; EG: LDS_READ_RET |
| 145 | ; EG: LDS_READ_RET |
| 146 | |
| 147 | ; EG: LDS_READ_RET |
| 148 | ; EG: LDS_READ_RET |
| 149 | ; EG: LDS_READ_RET |
| 150 | ; EG: LDS_READ_RET |
| 151 | |
| 152 | ; EG: LDS_READ_RET |
| 153 | ; EG: LDS_READ_RET |
| 154 | ; EG: LDS_READ_RET |
| 155 | ; EG: LDS_READ_RET |
| 156 | |
| 157 | ; EG: LDS_READ_RET |
| 158 | ; EG: LDS_READ_RET |
| 159 | ; EG: LDS_READ_RET |
| 160 | ; EG: LDS_READ_RET |
| 161 | |
| 162 | ; EG: LDS_READ_RET |
| 163 | ; EG: LDS_READ_RET |
| 164 | ; EG: LDS_READ_RET |
| 165 | ; EG: LDS_READ_RET |
Matt Arsenault | 3dbeefa | 2017-03-21 21:39:51 +0000 | [diff] [blame] | 166 | define amdgpu_kernel void @local_load_v16f64(<16 x double> addrspace(3)* %out, <16 x double> addrspace(3)* %in) #0 { |
Matt Arsenault | d1097a3 | 2016-06-02 19:54:26 +0000 | [diff] [blame] | 167 | entry: |
| 168 | %ld = load <16 x double>, <16 x double> addrspace(3)* %in |
| 169 | store <16 x double> %ld, <16 x double> addrspace(3)* %out |
| 170 | ret void |
| 171 | } |
| 172 | |
| 173 | attributes #0 = { nounwind } |