diff options
Diffstat (limited to 'test/CodeGen/R600/local-memory.ll')
| -rw-r--r-- | test/CodeGen/R600/local-memory.ll | 11 |
1 files changed, 6 insertions, 5 deletions
diff --git a/test/CodeGen/R600/local-memory.ll b/test/CodeGen/R600/local-memory.ll index 68e72c5..9494ed7 100644 --- a/test/CodeGen/R600/local-memory.ll +++ b/test/CodeGen/R600/local-memory.ll @@ -4,7 +4,6 @@ @local_memory.local_mem = internal unnamed_addr addrspace(3) global [128 x i32] undef, align 4 -; FUNC-LABEL: {{^}}local_memory: ; Check that the LDS size emitted correctly ; EG: .long 166120 @@ -14,6 +13,8 @@ ; CI: .long 47180 ; CI-NEXT: .long 38792 +; FUNC-LABEL: {{^}}local_memory: + ; EG: LDS_WRITE ; SI-NOT: s_wqm_b64 ; SI: ds_write_b32 @@ -29,15 +30,15 @@ define void @local_memory(i32 addrspace(1)* %out) { entry: %y.i = call i32 @llvm.r600.read.tidig.x() #0 - %arrayidx = getelementptr inbounds [128 x i32] addrspace(3)* @local_memory.local_mem, i32 0, i32 %y.i + %arrayidx = getelementptr inbounds [128 x i32], [128 x i32] addrspace(3)* @local_memory.local_mem, i32 0, i32 %y.i store i32 %y.i, i32 addrspace(3)* %arrayidx, align 4 %add = add nsw i32 %y.i, 1 %cmp = icmp eq i32 %add, 16 %.add = select i1 %cmp, i32 0, i32 %add call void @llvm.AMDGPU.barrier.local() - %arrayidx1 = getelementptr inbounds [128 x i32] addrspace(3)* @local_memory.local_mem, i32 0, i32 %.add - %0 = load i32 addrspace(3)* %arrayidx1, align 4 - %arrayidx2 = getelementptr inbounds i32 addrspace(1)* %out, i32 %y.i + %arrayidx1 = getelementptr inbounds [128 x i32], [128 x i32] addrspace(3)* @local_memory.local_mem, i32 0, i32 %.add + %0 = load i32, i32 addrspace(3)* %arrayidx1, align 4 + %arrayidx2 = getelementptr inbounds i32, i32 addrspace(1)* %out, i32 %y.i store i32 %0, i32 addrspace(1)* %arrayidx2, align 4 ret void } |