1; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=GCN-NOHSA -check-prefix=FUNC %s 2; RUN: llc -mtriple=amdgcn--amdhsa -mcpu=kaveri -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=GCN-HSA -check-prefix=FUNC %s 3; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=GCN-NOHSA -check-prefix=FUNC %s 4; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s 5 6; FUNC-LABEL: {{^}}constant_load_i32: 7; GCN: s_load_dword s{{[0-9]+}} 8 9; EG: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0 10define void @constant_load_i32(i32 addrspace(1)* %out, i32 addrspace(2)* %in) #0 { 11entry: 12 %ld = load i32, i32 addrspace(2)* %in 13 store i32 %ld, i32 addrspace(1)* %out 14 ret void 15} 16 17; FUNC-LABEL: {{^}}constant_load_v2i32: 18; GCN: s_load_dwordx2 19 20; EG: VTX_READ_64 21define void @constant_load_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> addrspace(2)* %in) #0 { 22entry: 23 %ld = load <2 x i32>, <2 x i32> addrspace(2)* %in 24 store <2 x i32> %ld, <2 x i32> addrspace(1)* %out 25 ret void 26} 27 28; FUNC-LABEL: {{^}}constant_load_v3i32: 29; GCN: s_load_dwordx4 30 31; EG: VTX_READ_128 32define void @constant_load_v3i32(<3 x i32> addrspace(1)* %out, <3 x i32> addrspace(2)* %in) #0 { 33entry: 34 %ld = load <3 x i32>, <3 x i32> addrspace(2)* %in 35 store <3 x i32> %ld, <3 x i32> addrspace(1)* %out 36 ret void 37} 38 39; FUNC-LABEL: {{^}}constant_load_v4i32: 40; GCN: s_load_dwordx4 41 42; EG: VTX_READ_128 43define void @constant_load_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(2)* %in) #0 { 44entry: 45 %ld = load <4 x i32>, <4 x i32> addrspace(2)* %in 46 store <4 x i32> %ld, <4 x i32> addrspace(1)* %out 47 ret void 48} 49 50; FUNC-LABEL: {{^}}constant_load_v8i32: 51; GCN: s_load_dwordx8 52 53; EG: VTX_READ_128 54; EG: VTX_READ_128 55define void @constant_load_v8i32(<8 x i32> addrspace(1)* %out, <8 x i32> addrspace(2)* %in) #0 { 56entry: 57 %ld = load <8 x i32>, <8 x i32> addrspace(2)* %in 58 store <8 x i32> %ld, <8 x i32> addrspace(1)* %out 59 ret void 60} 61 62; FUNC-LABEL: {{^}}constant_load_v16i32: 63; GCN: s_load_dwordx16 64 65; EG: VTX_READ_128 66; EG: VTX_READ_128 67; EG: VTX_READ_128 68; EG: VTX_READ_128 69define void @constant_load_v16i32(<16 x i32> addrspace(1)* %out, <16 x i32> addrspace(2)* %in) #0 { 70entry: 71 %ld = load <16 x i32>, <16 x i32> addrspace(2)* %in 72 store <16 x i32> %ld, <16 x i32> addrspace(1)* %out 73 ret void 74} 75 76; FUNC-LABEL: {{^}}constant_zextload_i32_to_i64: 77; GCN-DAG: s_load_dword s[[SLO:[0-9]+]], 78; GCN-DAG: v_mov_b32_e32 v[[SHI:[0-9]+]], 0{{$}} 79; GCN: store_dwordx2 80 81; EG: MEM_RAT 82; EG: MEM_RAT 83define void @constant_zextload_i32_to_i64(i64 addrspace(1)* %out, i32 addrspace(2)* %in) #0 { 84 %ld = load i32, i32 addrspace(2)* %in 85 %ext = zext i32 %ld to i64 86 store i64 %ext, i64 addrspace(1)* %out 87 ret void 88} 89 90; FUNC-LABEL: {{^}}constant_sextload_i32_to_i64: 91; GCN: s_load_dword s[[SLO:[0-9]+]] 92; GCN: s_ashr_i32 s[[HI:[0-9]+]], s[[SLO]], 31 93; GCN: store_dwordx2 94 95; EG: MEM_RAT 96; EG: MEM_RAT 97; EG: ASHR {{[* ]*}}T{{[0-9]\.[XYZW]}}, T{{[0-9]\.[XYZW]}}, literal.x 98; EG: 31 99define void @constant_sextload_i32_to_i64(i64 addrspace(1)* %out, i32 addrspace(2)* %in) #0 { 100 %ld = load i32, i32 addrspace(2)* %in 101 %ext = sext i32 %ld to i64 102 store i64 %ext, i64 addrspace(1)* %out 103 ret void 104} 105 106; FUNC-LABEL: {{^}}constant_zextload_v1i32_to_v1i64: 107; GCN: s_load_dword 108; GCN: store_dwordx2 109define void @constant_zextload_v1i32_to_v1i64(<1 x i64> addrspace(1)* %out, <1 x i32> addrspace(2)* %in) #0 { 110 %ld = load <1 x i32>, <1 x i32> addrspace(2)* %in 111 %ext = zext <1 x i32> %ld to <1 x i64> 112 store <1 x i64> %ext, <1 x i64> addrspace(1)* %out 113 ret void 114} 115 116; FUNC-LABEL: {{^}}constant_sextload_v1i32_to_v1i64: 117; GCN: s_load_dword s[[LO:[0-9]+]] 118; GCN: s_ashr_i32 s[[HI:[0-9]+]], s[[LO]], 31 119; GCN: store_dwordx2 120define void @constant_sextload_v1i32_to_v1i64(<1 x i64> addrspace(1)* %out, <1 x i32> addrspace(2)* %in) #0 { 121 %ld = load <1 x i32>, <1 x i32> addrspace(2)* %in 122 %ext = sext <1 x i32> %ld to <1 x i64> 123 store <1 x i64> %ext, <1 x i64> addrspace(1)* %out 124 ret void 125} 126 127; FUNC-LABEL: {{^}}constant_zextload_v2i32_to_v2i64: 128; GCN: s_load_dwordx2 s{{\[[0-9]+:[0-9]+\]}}, s{{\[[0-9]+:[0-9]+\]}}, 0x0{{$}} 129; GCN: store_dwordx4 130define void @constant_zextload_v2i32_to_v2i64(<2 x i64> addrspace(1)* %out, <2 x i32> addrspace(2)* %in) #0 { 131 %ld = load <2 x i32>, <2 x i32> addrspace(2)* %in 132 %ext = zext <2 x i32> %ld to <2 x i64> 133 store <2 x i64> %ext, <2 x i64> addrspace(1)* %out 134 ret void 135} 136 137; FUNC-LABEL: {{^}}constant_sextload_v2i32_to_v2i64: 138; GCN: s_load_dwordx2 s{{\[[0-9]+:[0-9]+\]}}, s{{\[[0-9]+:[0-9]+\]}}, 0x0{{$}} 139 140; GCN-DAG: s_ashr_i32 141; GCN-DAG: s_ashr_i32 142 143; GCN: store_dwordx4 144define void @constant_sextload_v2i32_to_v2i64(<2 x i64> addrspace(1)* %out, <2 x i32> addrspace(2)* %in) #0 { 145 %ld = load <2 x i32>, <2 x i32> addrspace(2)* %in 146 %ext = sext <2 x i32> %ld to <2 x i64> 147 store <2 x i64> %ext, <2 x i64> addrspace(1)* %out 148 ret void 149} 150 151; FUNC-LABEL: {{^}}constant_zextload_v4i32_to_v4i64: 152; GCN: s_load_dwordx4 153 154; GCN: store_dwordx4 155; GCN: store_dwordx4 156define void @constant_zextload_v4i32_to_v4i64(<4 x i64> addrspace(1)* %out, <4 x i32> addrspace(2)* %in) #0 { 157 %ld = load <4 x i32>, <4 x i32> addrspace(2)* %in 158 %ext = zext <4 x i32> %ld to <4 x i64> 159 store <4 x i64> %ext, <4 x i64> addrspace(1)* %out 160 ret void 161} 162 163; FUNC-LABEL: {{^}}constant_sextload_v4i32_to_v4i64: 164; GCN: s_load_dwordx4 165 166; GCN: s_ashr_i32 167; GCN: s_ashr_i32 168; GCN: s_ashr_i32 169; GCN: s_ashr_i32 170 171; GCN: store_dwordx4 172; GCN: store_dwordx4 173define void @constant_sextload_v4i32_to_v4i64(<4 x i64> addrspace(1)* %out, <4 x i32> addrspace(2)* %in) #0 { 174 %ld = load <4 x i32>, <4 x i32> addrspace(2)* %in 175 %ext = sext <4 x i32> %ld to <4 x i64> 176 store <4 x i64> %ext, <4 x i64> addrspace(1)* %out 177 ret void 178} 179 180; FUNC-LABEL: {{^}}constant_zextload_v8i32_to_v8i64: 181; GCN: s_load_dwordx8 182 183; GCN-NOHSA-DAG: buffer_store_dwordx4 184; GCN-NOHSA-DAG: buffer_store_dwordx4 185; GCN-NOHSA-DAG: buffer_store_dwordx4 186; GCN-NOHSA-DAG: buffer_store_dwordx4 187 188; GCN-HSA-DAG: flat_store_dwordx4 189; GCN-HSA-DAG: flat_store_dwordx4 190; GCN-SA-DAG: flat_store_dwordx4 191; GCN-HSA-DAG: flat_store_dwordx4 192define void @constant_zextload_v8i32_to_v8i64(<8 x i64> addrspace(1)* %out, <8 x i32> addrspace(2)* %in) #0 { 193 %ld = load <8 x i32>, <8 x i32> addrspace(2)* %in 194 %ext = zext <8 x i32> %ld to <8 x i64> 195 store <8 x i64> %ext, <8 x i64> addrspace(1)* %out 196 ret void 197} 198 199; FUNC-LABEL: {{^}}constant_sextload_v8i32_to_v8i64: 200; GCN: s_load_dwordx8 201 202; GCN: s_ashr_i32 203; GCN: s_ashr_i32 204; GCN: s_ashr_i32 205; GCN: s_ashr_i32 206; GCN: s_ashr_i32 207; GCN: s_ashr_i32 208; GCN: s_ashr_i32 209; GCN: s_ashr_i32 210 211; GCN-NOHSA-DAG: buffer_store_dwordx4 212; GCN-NOHSA-DAG: buffer_store_dwordx4 213; GCN-NOHSA-DAG: buffer_store_dwordx4 214; GCN-NOHSA-DAG: buffer_store_dwordx4 215 216; GCN-HSA-DAG: flat_store_dwordx4 217; GCN-HSA-DAG: flat_store_dwordx4 218; GCN-HSA-DAG: flat_store_dwordx4 219; GCN-HSA-DAG: flat_store_dwordx4 220define void @constant_sextload_v8i32_to_v8i64(<8 x i64> addrspace(1)* %out, <8 x i32> addrspace(2)* %in) #0 { 221 %ld = load <8 x i32>, <8 x i32> addrspace(2)* %in 222 %ext = sext <8 x i32> %ld to <8 x i64> 223 store <8 x i64> %ext, <8 x i64> addrspace(1)* %out 224 ret void 225} 226 227; FUNC-LABEL: {{^}}constant_sextload_v16i32_to_v16i64: 228; GCN: s_load_dwordx16 229 230 231; GCN-DAG: s_ashr_i32 232 233; GCN: store_dwordx4 234; GCN: store_dwordx4 235; GCN: store_dwordx4 236; GCN: store_dwordx4 237; GCN: store_dwordx4 238; GCN: store_dwordx4 239; GCN: store_dwordx4 240; GCN: store_dwordx4 241define void @constant_sextload_v16i32_to_v16i64(<16 x i64> addrspace(1)* %out, <16 x i32> addrspace(2)* %in) #0 { 242 %ld = load <16 x i32>, <16 x i32> addrspace(2)* %in 243 %ext = sext <16 x i32> %ld to <16 x i64> 244 store <16 x i64> %ext, <16 x i64> addrspace(1)* %out 245 ret void 246} 247 248; FUNC-LABEL: {{^}}constant_zextload_v16i32_to_v16i64 249; GCN: s_load_dwordx16 250 251; GCN-NOHSA: buffer_store_dwordx4 252; GCN-NOHSA: buffer_store_dwordx4 253; GCN-NOHSA: buffer_store_dwordx4 254; GCN-NOHSA: buffer_store_dwordx4 255; GCN-NOHSA: buffer_store_dwordx4 256; GCN-NOHSA: buffer_store_dwordx4 257; GCN-NOHSA: buffer_store_dwordx4 258; GCN-NOHSA: buffer_store_dwordx4 259 260; GCN-HSA: flat_store_dwordx4 261; GCN-HSA: flat_store_dwordx4 262; GCN-HSA: flat_store_dwordx4 263; GCN-HSA: flat_store_dwordx4 264; GCN-HSA: flat_store_dwordx4 265; GCN-HSA: flat_store_dwordx4 266; GCN-HSA: flat_store_dwordx4 267; GCN-HSA: flat_store_dwordx4 268define void @constant_zextload_v16i32_to_v16i64(<16 x i64> addrspace(1)* %out, <16 x i32> addrspace(2)* %in) #0 { 269 %ld = load <16 x i32>, <16 x i32> addrspace(2)* %in 270 %ext = zext <16 x i32> %ld to <16 x i64> 271 store <16 x i64> %ext, <16 x i64> addrspace(1)* %out 272 ret void 273} 274 275; FUNC-LABEL: {{^}}constant_sextload_v32i32_to_v32i64: 276 277; GCN: s_load_dwordx16 278; GCN: s_load_dwordx16 279 280; GCN-NOHSA: buffer_store_dwordx4 281; GCN-NOHSA: buffer_store_dwordx4 282; GCN-NOHSA: buffer_store_dwordx4 283; GCN-NOHSA: buffer_store_dwordx4 284 285; GCN-NOHSA: buffer_store_dwordx4 286; GCN-NOHSA: buffer_store_dwordx4 287; GCN-NOHSA: buffer_store_dwordx4 288; GCN-NOHSA: buffer_store_dwordx4 289 290; GCN-NOHSA: buffer_store_dwordx4 291; GCN-NOHSA: buffer_store_dwordx4 292; GCN-NOHSA: buffer_store_dwordx4 293; GCN-NOHSA: buffer_store_dwordx4 294 295; GCN-NOHSA: buffer_store_dwordx4 296; GCN-NOHSA: buffer_store_dwordx4 297; GCN-NOHSA: buffer_store_dwordx4 298; GCN-NOHSA: buffer_store_dwordx4 299 300; GCN-HSA: flat_store_dwordx4 301; GCN-HSA: flat_store_dwordx4 302; GCN-HSA: flat_store_dwordx4 303; GCN-HSA: flat_store_dwordx4 304 305; GCN-HSA: flat_store_dwordx4 306; GCN-HSA: flat_store_dwordx4 307; GCN-HSA: flat_store_dwordx4 308; GCN-HSA: flat_store_dwordx4 309 310; GCN-HSA: flat_store_dwordx4 311; GCN-HSA: flat_store_dwordx4 312; GCN-HSA: flat_store_dwordx4 313; GCN-HSA: flat_store_dwordx4 314 315; GCN-HSA: flat_store_dwordx4 316; GCN-HSA: flat_store_dwordx4 317; GCN-HSA: flat_store_dwordx4 318; GCN-HSA: flat_store_dwordx4 319 320define void @constant_sextload_v32i32_to_v32i64(<32 x i64> addrspace(1)* %out, <32 x i32> addrspace(2)* %in) #0 { 321 %ld = load <32 x i32>, <32 x i32> addrspace(2)* %in 322 %ext = sext <32 x i32> %ld to <32 x i64> 323 store <32 x i64> %ext, <32 x i64> addrspace(1)* %out 324 ret void 325} 326 327; FUNC-LABEL: {{^}}constant_zextload_v32i32_to_v32i64: 328; GCN: s_load_dwordx16 329; GCN: s_load_dwordx16 330 331; GCN-NOHSA-DAG: buffer_store_dwordx4 332; GCN-NOHSA-DAG: buffer_store_dwordx4 333; GCN-NOHSA-DAG: buffer_store_dwordx4 334; GCN-NOHSA-DAG: buffer_store_dwordx4 335 336; GCN-NOHSA-DAG: buffer_store_dwordx4 337; GCN-NOHSA-DAG: buffer_store_dwordx4 338; GCN-NOHSA-DAG: buffer_store_dwordx4 339; GCN-NOHSA-DAG: buffer_store_dwordx4 340 341; GCN-NOHSA-DAG: buffer_store_dwordx4 342; GCN-NOHSA-DAG: buffer_store_dwordx4 343; GCN-NOHSA-DAG: buffer_store_dwordx4 344; GCN-NOHSA-DAG: buffer_store_dwordx4 345 346; GCN-NOHSA-DAG: buffer_store_dwordx4 347; GCN-NOHSA-DAG: buffer_store_dwordx4 348; GCN-NOHSA-DAG: buffer_store_dwordx4 349; GCN-NOHSA-DAG: buffer_store_dwordx4 350 351 352; GCN-HSA-DAG: flat_store_dwordx4 353; GCN-HSA-DAG: flat_store_dwordx4 354; GCN-HSA-DAG: flat_store_dwordx4 355; GCN-HSA-DAG: flat_store_dwordx4 356 357; GCN-HSA-DAG: flat_store_dwordx4 358; GCN-HSA-DAG: flat_store_dwordx4 359; GCN-HSA-DAG: flat_store_dwordx4 360; GCN-HSA-DAG: flat_store_dwordx4 361 362; GCN-HSA-DAG: flat_store_dwordx4 363; GCN-HSA-DAG: flat_store_dwordx4 364; GCN-HSA-DAG: flat_store_dwordx4 365; GCN-HSA-DAG: flat_store_dwordx4 366 367; GCN-HSA-DAG: flat_store_dwordx4 368; GCN-HSA-DAG: flat_store_dwordx4 369; GCN-HSA-DAG: flat_store_dwordx4 370; GCN-HSA-DAG: flat_store_dwordx4 371define void @constant_zextload_v32i32_to_v32i64(<32 x i64> addrspace(1)* %out, <32 x i32> addrspace(2)* %in) #0 { 372 %ld = load <32 x i32>, <32 x i32> addrspace(2)* %in 373 %ext = zext <32 x i32> %ld to <32 x i64> 374 store <32 x i64> %ext, <32 x i64> addrspace(1)* %out 375 ret void 376} 377 378attributes #0 = { nounwind } 379