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
7 ; Testing for ds_read/write_b128
8 ; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs -mattr=+enable-ds128 < %s | FileCheck -check-prefixes=CIVI,FUNC %s
9 ; RUN: llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs -mattr=+enable-ds128 < %s | FileCheck -check-prefixes=CIVI,FUNC %s
11 ; FUNC-LABEL: {{^}}local_load_i64:
12 ; SICIVI: s_mov_b32 m0
15 ; GCN: ds_read_b64 [[VAL:v\[[0-9]+:[0-9]+\]]], v{{[0-9]+}}{{$}}
16 ; GCN: ds_write_b64 v{{[0-9]+}}, [[VAL]]
20 define amdgpu_kernel void @local_load_i64(i64 addrspace(3)* %out, i64 addrspace(3)* %in) #0 {
21 %ld = load i64, i64 addrspace(3)* %in
22 store i64 %ld, i64 addrspace(3)* %out
26 ; FUNC-LABEL: {{^}}local_load_v2i64:
27 ; SICIVI: s_mov_b32 m0
36 define amdgpu_kernel void @local_load_v2i64(<2 x i64> addrspace(3)* %out, <2 x i64> addrspace(3)* %in) #0 {
38 %ld = load <2 x i64>, <2 x i64> addrspace(3)* %in
39 store <2 x i64> %ld, <2 x i64> addrspace(3)* %out
43 ; Tests if ds_read/write_b128 gets generated for the 16 byte aligned load.
44 ; FUNC-LABEL: {{^}}local_load_v2i64_to_128:
49 define amdgpu_kernel void @local_load_v2i64_to_128(<2 x i64> addrspace(3)* %out, <2 x i64> addrspace(3)* %in) {
51 %ld = load <2 x i64>, <2 x i64> addrspace(3)* %in, align 16
52 store <2 x i64> %ld, <2 x i64> addrspace(3)* %out, align 16
56 ; FUNC-LABEL: {{^}}local_load_v3i64:
57 ; SICIVI: s_mov_b32 m0
60 ; GCN-DAG: ds_read2_b64
61 ; GCN-DAG: ds_read_b64
69 define amdgpu_kernel void @local_load_v3i64(<3 x i64> addrspace(3)* %out, <3 x i64> addrspace(3)* %in) #0 {
71 %ld = load <3 x i64>, <3 x i64> addrspace(3)* %in
72 store <3 x i64> %ld, <3 x i64> addrspace(3)* %out
76 ; FUNC-LABEL: {{^}}local_load_v4i64:
77 ; SICIVI: s_mov_b32 m0
92 define amdgpu_kernel void @local_load_v4i64(<4 x i64> addrspace(3)* %out, <4 x i64> addrspace(3)* %in) #0 {
94 %ld = load <4 x i64>, <4 x i64> addrspace(3)* %in
95 store <4 x i64> %ld, <4 x i64> addrspace(3)* %out
99 ; FUNC-LABEL: {{^}}local_load_v8i64:
100 ; SICIVI: s_mov_b32 m0
124 define amdgpu_kernel void @local_load_v8i64(<8 x i64> addrspace(3)* %out, <8 x i64> addrspace(3)* %in) #0 {
126 %ld = load <8 x i64>, <8 x i64> addrspace(3)* %in
127 store <8 x i64> %ld, <8 x i64> addrspace(3)* %out
131 ; FUNC-LABEL: {{^}}local_load_v16i64:
132 ; SICIVI: s_mov_b32 m0
183 define amdgpu_kernel void @local_load_v16i64(<16 x i64> addrspace(3)* %out, <16 x i64> addrspace(3)* %in) #0 {
185 %ld = load <16 x i64>, <16 x i64> addrspace(3)* %in
186 store <16 x i64> %ld, <16 x i64> addrspace(3)* %out
190 attributes #0 = { nounwind }