1; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,SICIVI,FUNC %s 2; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,SICIVI,FUNC %s 3; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck -check-prefixes=EG,FUNC %s 4 5; Testing for ds_read/write_128 6; RUN: llc -march=amdgcn -mcpu=tahiti -mattr=+enable-ds128 < %s | FileCheck -check-prefixes=SI,FUNC %s 7; RUN: llc -march=amdgcn -mcpu=tonga -mattr=+enable-ds128 < %s | FileCheck -check-prefixes=CIVI,FUNC %s 8; RUN: llc -march=amdgcn -mcpu=gfx900 -mattr=+enable-ds128 < %s | FileCheck -check-prefixes=CIVI,FUNC %s 9 10; FUNC-LABEL: {{^}}load_f32_local: 11; SICIVI: s_mov_b32 m0 12; GFX9-NOT: m0 13; GCN: ds_read_b32 14 15; EG: LDS_READ_RET 16define amdgpu_kernel void @load_f32_local(float addrspace(1)* %out, float addrspace(3)* %in) #0 { 17entry: 18 %tmp0 = load float, float addrspace(3)* %in 19 store float %tmp0, float addrspace(1)* %out 20 ret void 21} 22 23; FUNC-LABEL: {{^}}load_v2f32_local: 24; SICIVI: s_mov_b32 m0 25; GFX9-NOT: m0 26 27; GCN: ds_read_b64 28 29; EG: LDS_READ_RET 30; EG: LDS_READ_RET 31define amdgpu_kernel void @load_v2f32_local(<2 x float> addrspace(1)* %out, <2 x float> addrspace(3)* %in) #0 { 32entry: 33 %tmp0 = load <2 x float>, <2 x float> addrspace(3)* %in 34 store <2 x float> %tmp0, <2 x float> addrspace(1)* %out 35 ret void 36} 37 38; FIXME: should this do a read2_b64? 39; FUNC-LABEL: {{^}}local_load_v3f32: 40; SICIVI: s_mov_b32 m0 41; GFX9-NOT: m0 42 43; GCN-DAG: ds_read_b32 v{{[0-9]+}}, v{{[0-9]+}} offset:8 44; GCN-DAG: ds_read_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+$}} 45; GCN: s_waitcnt 46; GCN-DAG: ds_write_b64 47; GCN-DAG: ds_write_b32 v{{[0-9]+}}, v{{[0-9]+}} offset:8{{$}} 48 49; EG: LDS_READ_RET 50; EG: LDS_READ_RET 51; EG: LDS_READ_RET 52define amdgpu_kernel void @local_load_v3f32(<3 x float> addrspace(3)* %out, <3 x float> addrspace(3)* %in) #0 { 53entry: 54 %tmp0 = load <3 x float>, <3 x float> addrspace(3)* %in 55 store <3 x float> %tmp0, <3 x float> addrspace(3)* %out 56 ret void 57} 58 59; FUNC-LABEL: {{^}}local_load_v4f32: 60; SICIVI: s_mov_b32 m0 61; GFX9-NOT: m0 62 63; GCN: ds_read2_b64 64 65; EG: LDS_READ_RET 66; EG: LDS_READ_RET 67; EG: LDS_READ_RET 68; EG: LDS_READ_RET 69define amdgpu_kernel void @local_load_v4f32(<4 x float> addrspace(3)* %out, <4 x float> addrspace(3)* %in) #0 { 70entry: 71 %tmp0 = load <4 x float>, <4 x float> addrspace(3)* %in 72 store <4 x float> %tmp0, <4 x float> addrspace(3)* %out 73 ret void 74} 75 76; FUNC-LABEL: {{^}}local_load_v8f32: 77; SICIVI: s_mov_b32 m0 78; GFX9-NOT: m0 79 80; GCN: ds_read2_b64 81; GCN: ds_read2_b64 82 83; EG: LDS_READ_RET 84; EG: LDS_READ_RET 85; EG: LDS_READ_RET 86; EG: LDS_READ_RET 87; EG: LDS_READ_RET 88; EG: LDS_READ_RET 89; EG: LDS_READ_RET 90; EG: LDS_READ_RET 91define amdgpu_kernel void @local_load_v8f32(<8 x float> addrspace(3)* %out, <8 x float> addrspace(3)* %in) #0 { 92entry: 93 %tmp0 = load <8 x float>, <8 x float> addrspace(3)* %in 94 store <8 x float> %tmp0, <8 x float> addrspace(3)* %out 95 ret void 96} 97 98; FUNC-LABEL: {{^}}local_load_v16f32: 99; SICIVI: s_mov_b32 m0 100; GFX9-NOT: m0 101 102; GCN: ds_read2_b64 103; GCN: ds_read2_b64 104; GCN: ds_read2_b64 105; GCN: ds_read2_b64 106 107; EG: LDS_READ_RET 108; EG: LDS_READ_RET 109; EG: LDS_READ_RET 110; EG: LDS_READ_RET 111; EG: LDS_READ_RET 112; EG: LDS_READ_RET 113; EG: LDS_READ_RET 114; EG: LDS_READ_RET 115; EG: LDS_READ_RET 116; EG: LDS_READ_RET 117; EG: LDS_READ_RET 118; EG: LDS_READ_RET 119; EG: LDS_READ_RET 120; EG: LDS_READ_RET 121; EG: LDS_READ_RET 122; EG: LDS_READ_RET 123define amdgpu_kernel void @local_load_v16f32(<16 x float> addrspace(3)* %out, <16 x float> addrspace(3)* %in) #0 { 124entry: 125 %tmp0 = load <16 x float>, <16 x float> addrspace(3)* %in 126 store <16 x float> %tmp0, <16 x float> addrspace(3)* %out 127 ret void 128} 129 130; Tests if ds_read/write_b128 gets generated for the 16 byte aligned load. 131; FUNC-LABEL: {{^}}local_v4f32_to_128: 132 133; SI-NOT: ds_read_b128 134; SI-NOT: ds_write_b128 135 136; CIVI: ds_read_b128 137; CIVI: ds_write_b128 138 139; EG: LDS_READ_RET 140; EG: LDS_READ_RET 141; EG: LDS_READ_RET 142; EG: LDS_READ_RET 143define amdgpu_kernel void @local_v4f32_to_128(<4 x float> addrspace(3)* %out, <4 x float> addrspace(3)* %in) { 144 %ld = load <4 x float>, <4 x float> addrspace(3)* %in, align 16 145 store <4 x float> %ld, <4 x float> addrspace(3)* %out, align 16 146 ret void 147} 148 149attributes #0 = { nounwind } 150