1 ; RUN: llc -march=r600 -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI %s
3 declare i32 @llvm.r600.read.tidig.x() #0
4 declare void @llvm.AMDGPU.barrier.local() #1
6 ; Function Attrs: nounwind
7 ; SI-LABEL: @signed_ds_offset_addressing_loop
9 ; SI: V_ADD_I32_e32 [[VADDR:v[0-9]+]],
10 ; SI-DAG: DS_READ_B32 v{{[0-9]+}}, [[VADDR]], 0x0
11 ; SI-DAG: DS_READ_B32 v{{[0-9]+}}, [[VADDR]], 0x4
12 ; SI-DAG: DS_READ_B32 v{{[0-9]+}}, [[VADDR]], 0x80
13 ; SI-DAG: DS_READ_B32 v{{[0-9]+}}, [[VADDR]], 0x84
14 ; SI-DAG: DS_READ_B32 v{{[0-9]+}}, [[VADDR]], 0x100
16 define void @signed_ds_offset_addressing_loop(float addrspace(1)* noalias nocapture %out, float addrspace(3)* noalias nocapture readonly %lptr, i32 %n) #2 {
18 %x.i = tail call i32 @llvm.r600.read.tidig.x() #0
19 %mul = shl nsw i32 %x.i, 1
22 for.body: ; preds = %for.body, %entry
23 %sum.03 = phi float [ 0.000000e+00, %entry ], [ %add13, %for.body ]
24 %offset.02 = phi i32 [ %mul, %entry ], [ %add14, %for.body ]
25 %k.01 = phi i32 [ 0, %entry ], [ %inc, %for.body ]
26 tail call void @llvm.AMDGPU.barrier.local() #1
27 %arrayidx = getelementptr inbounds float addrspace(3)* %lptr, i32 %offset.02
28 %tmp = load float addrspace(3)* %arrayidx, align 4
29 %add1 = add nsw i32 %offset.02, 1
30 %arrayidx2 = getelementptr inbounds float addrspace(3)* %lptr, i32 %add1
31 %tmp1 = load float addrspace(3)* %arrayidx2, align 4
32 %add3 = add nsw i32 %offset.02, 32
33 %arrayidx4 = getelementptr inbounds float addrspace(3)* %lptr, i32 %add3
34 %tmp2 = load float addrspace(3)* %arrayidx4, align 4
35 %add5 = add nsw i32 %offset.02, 33
36 %arrayidx6 = getelementptr inbounds float addrspace(3)* %lptr, i32 %add5
37 %tmp3 = load float addrspace(3)* %arrayidx6, align 4
38 %add7 = add nsw i32 %offset.02, 64
39 %arrayidx8 = getelementptr inbounds float addrspace(3)* %lptr, i32 %add7
40 %tmp4 = load float addrspace(3)* %arrayidx8, align 4
41 %add9 = fadd float %tmp, %tmp1
42 %add10 = fadd float %add9, %tmp2
43 %add11 = fadd float %add10, %tmp3
44 %add12 = fadd float %add11, %tmp4
45 %add13 = fadd float %sum.03, %add12
46 %inc = add nsw i32 %k.01, 1
47 %add14 = add nsw i32 %offset.02, 97
48 %exitcond = icmp eq i32 %inc, 8
49 br i1 %exitcond, label %for.end, label %for.body
51 for.end: ; preds = %for.body
52 %tmp5 = sext i32 %x.i to i64
53 %arrayidx15 = getelementptr inbounds float addrspace(1)* %out, i64 %tmp5
54 store float %add13, float addrspace(1)* %arrayidx15, align 4
58 attributes #0 = { nounwind readnone }
59 attributes #1 = { noduplicate nounwind }
60 attributes #2 = { nounwind "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }