37 lines · plain
1; RUN: llc -mtriple=amdgcn-amd-amdhsa -stress-regalloc=1 < %s | FileCheck -check-prefix=GCN %s2 3; GCN-LABEL: {{^}}spill_csr_s5_copy:4; GCN: s_mov_b32 [[FP_SCRATCH_COPY:s[0-9]+]], s335; GCN: s_or_saveexec_b646; GCN-NEXT: buffer_store_dword v40, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill7; GCN-NEXT: buffer_store_dword v41, off, s[0:3], s33 offset:8 ; 4-byte Folded Spill8; GCN-NEXT: s_mov_b64 exec9; GCN: v_writelane_b32 v40, [[FP_SCRATCH_COPY]], 210; GCN: s_swappc_b6411 12; GCN: v_mov_b32_e32 [[K:v[0-9]+]], 913; GCN: buffer_store_dword [[K]], off, s[0:3], s33{{$}}14 15; GCN: v_readlane_b32 [[FP_SCRATCH_COPY:s[0-9]+]], v40, 216; GCN: s_or_saveexec_b6417; GCN-NEXT: buffer_load_dword v40, off, s[0:3], s33 offset:4 ; 4-byte Folded Reload18; GCN-NEXT: buffer_load_dword v41, off, s[0:3], s33 offset:8 ; 4-byte Folded Reload19; GCN: s_mov_b64 exec20; GCN: s_mov_b32 s33, [[FP_SCRATCH_COPY]]21; GCN: s_setpc_b6422define void @spill_csr_s5_copy() #0 {23bb:24 %alloca = alloca i32, addrspace(5)25 %tmp = tail call i64 @func() #126 %tmp1 = getelementptr inbounds i32, ptr addrspace(1) null, i64 %tmp27 %tmp2 = load i32, ptr addrspace(1) %tmp1, align 428 %tmp3 = zext i32 %tmp2 to i6429 store volatile i32 9, ptr addrspace(5) %alloca30 ret void31}32 33declare i64 @func() #034 35attributes #0 = { nounwind "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" }36attributes #1 = { nounwind readnone }37