1; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s
2; RUN: llc -march=amdgcn -mcpu=verde -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
3; RUN: llc -march=amdgcn -mcpu=bonaire -verify-machineinstrs < %s | FileCheck -check-prefix=CI -check-prefix=FUNC %s
4
5@local_memory.local_mem = internal unnamed_addr addrspace(3) global [128 x i32] undef, align 4
6
7
8; Check that the LDS size emitted correctly
9; EG: .long 166120
10; EG-NEXT: .long 128
11; SI: .long 47180
12; SI-NEXT: .long 71560
13; CI: .long 47180
14; CI-NEXT: .long 38792
15
16; FUNC-LABEL: {{^}}local_memory:
17
18; EG: LDS_WRITE
19; SI-NOT: s_wqm_b64
20; SI: ds_write_b32
21
22; GROUP_BARRIER must be the last instruction in a clause
23; EG: GROUP_BARRIER
24; EG-NEXT: ALU clause
25; SI: s_barrier
26
27; EG: LDS_READ_RET
28; SI: ds_read_b32 {{v[0-9]+}},
29
30define void @local_memory(i32 addrspace(1)* %out) {
31entry:
32  %y.i = call i32 @llvm.r600.read.tidig.x() #0
33  %arrayidx = getelementptr inbounds [128 x i32], [128 x i32] addrspace(3)* @local_memory.local_mem, i32 0, i32 %y.i
34  store i32 %y.i, i32 addrspace(3)* %arrayidx, align 4
35  %add = add nsw i32 %y.i, 1
36  %cmp = icmp eq i32 %add, 16
37  %.add = select i1 %cmp, i32 0, i32 %add
38  call void @llvm.AMDGPU.barrier.local()
39  %arrayidx1 = getelementptr inbounds [128 x i32], [128 x i32] addrspace(3)* @local_memory.local_mem, i32 0, i32 %.add
40  %0 = load i32, i32 addrspace(3)* %arrayidx1, align 4
41  %arrayidx2 = getelementptr inbounds i32, i32 addrspace(1)* %out, i32 %y.i
42  store i32 %0, i32 addrspace(1)* %arrayidx2, align 4
43  ret void
44}
45
46declare i32 @llvm.r600.read.tidig.x() #0
47declare void @llvm.AMDGPU.barrier.local()
48
49attributes #0 = { readnone }
50