1; RUN: llc -mtriple=amdgcn--amdhsa -mcpu=fiji -verify-machineinstrs < %s | FileCheck %s
2
3declare i32 @llvm.amdgcn.ds.permute(i32, i32) #0
4
5; CHECK-LABEL: {{^}}ds_permute:
6; CHECK: ds_permute_b32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
7define amdgpu_kernel void @ds_permute(i32 addrspace(1)* %out, i32 %index, i32 %src) nounwind {
8  %bpermute = call i32 @llvm.amdgcn.ds.permute(i32 %index, i32 %src) #0
9  store i32 %bpermute, i32 addrspace(1)* %out, align 4
10  ret void
11}
12
13; CHECK-LABEL: {{^}}ds_permute_imm_offset:
14; CHECK: ds_permute_b32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}} offset:4
15define amdgpu_kernel void @ds_permute_imm_offset(i32 addrspace(1)* %out, i32 %base_index, i32 %src) nounwind {
16  %index = add i32 %base_index, 4
17  %bpermute = call i32 @llvm.amdgcn.ds.permute(i32 %index, i32 %src) #0
18  store i32 %bpermute, i32 addrspace(1)* %out, align 4
19  ret void
20}
21
22attributes #0 = { nounwind readnone convergent }
23