xref: /llvm-project/llvm/test/CodeGen/AMDGPU/image-schedule.ll (revision bdf2fbba9cee60b4b260ff17e4f44c475c11e715)
1; RUN: llc -mtriple=amdgcn--amdpal -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN %s
2
3; The first image store and the second image load use the same descriptor and
4; the same coordinate. Check that they do not get swapped by the machine
5; instruction scheduler.
6
7; GCN-LABEL: {{^}}_amdgpu_cs_main:
8; GCN: image_load
9; GCN: image_store
10; GCN: image_load
11; GCN: image_store
12
13define dllexport amdgpu_cs void @_amdgpu_cs_main(i32 inreg %arg, i32 inreg %arg1, i32 inreg %arg2, <3 x i32> inreg %arg3, i32 inreg %arg4, <3 x i32> %arg5) local_unnamed_addr #0 {
14.entry:
15  %tmp = call i64 @llvm.amdgcn.s.getpc() #1
16  %tmp6 = bitcast i64 %tmp to <2 x i32>
17  %.0.vec.insert = insertelement <2 x i32> undef, i32 %arg2, i32 0
18  %.4.vec.insert = shufflevector <2 x i32> %.0.vec.insert, <2 x i32> %tmp6, <2 x i32> <i32 0, i32 3>
19  %tmp7 = bitcast <2 x i32> %.4.vec.insert to i64
20  %tmp8 = inttoptr i64 %tmp7 to ptr addrspace(4)
21  %tmp9 = add <3 x i32> %arg3, %arg5
22  %tmp10 = getelementptr [4294967295 x i8], ptr addrspace(4) %tmp8, i64 0, i64 32
23  %tmp12 = load <8 x i32>, ptr addrspace(4) %tmp10, align 16
24  %tmp13.0 = extractelement <3 x i32> %tmp9, i32 0
25  %tmp13.1 = extractelement <3 x i32> %tmp9, i32 1
26  %tmp14 = call <4 x float> @llvm.amdgcn.image.load.2d.v4f32.i32(i32 15, i32 %tmp13.0, i32 %tmp13.1, <8 x i32> %tmp12, i32 0, i32 0) #0
27  %tmp15 = inttoptr i64 %tmp7 to ptr addrspace(4)
28  %tmp16 = load <8 x i32>, ptr addrspace(4) %tmp15, align 16
29  call void @llvm.amdgcn.image.store.2d.v4f32.i32(<4 x float> %tmp14, i32 15, i32 %tmp13.0, i32 %tmp13.1, <8 x i32> %tmp16, i32 0, i32 0) #0
30  %tmp17 = load <8 x i32>, ptr addrspace(4) %tmp15, align 16
31  %tmp18 = call <4 x float> @llvm.amdgcn.image.load.2d.v4f32.i32(i32 165, i32 %tmp13.0, i32 %tmp13.1, <8 x i32> %tmp17, i32 0, i32 0) #0
32  %tmp19 = getelementptr [4294967295 x i8], ptr addrspace(4) %tmp8, i64 0, i64 64
33  %tmp21 = load <8 x i32>, ptr addrspace(4) %tmp19, align 16
34  call void @llvm.amdgcn.image.store.2d.v4f32.i32(<4 x float> %tmp18, i32 15, i32 %tmp13.0, i32 %tmp13.1, <8 x i32> %tmp21, i32 0, i32 0) #0
35  ret void
36}
37
38; Function Attrs: nounwind readnone speculatable
39declare i64 @llvm.amdgcn.s.getpc() #1
40
41; Function Attrs: nounwind readonly
42declare <4 x float> @llvm.amdgcn.image.load.2d.v4f32.i32(i32, i32, i32, <8 x i32>, i32, i32) #2
43
44; Function Attrs: nounwind writeonly
45declare void @llvm.amdgcn.image.store.2d.v4f32.i32(<4 x float>, i32, i32, i32, <8 x i32>, i32, i32) #3
46
47attributes #0 = { nounwind }
48attributes #1 = { nounwind readnone speculatable }
49attributes #2 = { nounwind readonly }
50attributes #3 = { nounwind writeonly }
51
52!0 = !{}
53