1 ; RUN: llc < %s -march=amdgcn -verify-machineinstrs | FileCheck -check-prefix=GCN %s
3 ; GCN-LABEL: {{^}}checkTwoBlocksWithUniformBranch
6 define amdgpu_kernel void @checkTwoBlocksWithUniformBranch(i32 addrspace(1)* nocapture %out, i32 %width, float %xPos, float %yPos, float %xStep, float %yStep, i32 %maxIter) {
8 %conv = call i32 @llvm.amdgcn.workitem.id.x() #1
9 %rem = urem i32 %conv, %width
10 %div = udiv i32 %conv, %width
11 %conv1 = sitofp i32 %rem to float
12 %x = tail call float @llvm.fmuladd.f32(float %xStep, float %conv1, float %xPos)
13 %conv2 = sitofp i32 %div to float
14 %y = tail call float @llvm.fmuladd.f32(float %yStep, float %conv2, float %yPos)
15 %yy = fmul float %y, %y
16 %xy = tail call float @llvm.fmuladd.f32(float %x, float %x, float %yy)
17 %cmp01 = fcmp ole float %xy, 4.000000e+00
18 %cmp02 = icmp ne i32 %maxIter, 0
19 %cond01 = and i1 %cmp02, %cmp01
20 br i1 %cond01, label %for.body.preheader, label %for.end
22 for.body.preheader: ; preds = %entry
25 for.body: ; preds = %for.body.preheader, %for.body
26 %x_val = phi float [ %call8, %for.body ], [ %x, %for.body.preheader ]
27 %iter_val = phi i32 [ %inc, %for.body ], [ 0, %for.body.preheader ]
28 %y_val = phi float [ %call9, %for.body ], [ %y, %for.body.preheader ]
29 %sub = fsub float -0.000000e+00, %y_val
30 %call7 = tail call float @llvm.fmuladd.f32(float %x_val, float %x_val, float %x) #1
31 %call8 = tail call float @llvm.fmuladd.f32(float %sub, float %y_val, float %call7) #1
32 %mul = fmul float %x_val, 2.000000e+00
33 %call9 = tail call float @llvm.fmuladd.f32(float %mul, float %y_val, float %y) #1
34 %inc = add nuw i32 %iter_val, 1
35 %mul3 = fmul float %call9, %call9
36 %0 = tail call float @llvm.fmuladd.f32(float %call8, float %call8, float %mul3)
37 %cmp = fcmp ole float %0, 4.000000e+00
38 %cmp5 = icmp ult i32 %inc, %maxIter
39 %or.cond = and i1 %cmp5, %cmp
40 br i1 %or.cond, label %for.body, label %for.end.loopexit
42 for.end.loopexit: ; preds = %for.body
45 for.end: ; preds = %for.end.loopexit, %entry
46 %iter.0.lcssa = phi i32 [ 0, %entry ], [ %inc, %for.end.loopexit ]
47 %idxprom = ashr exact i32 %conv, 32
48 %arrayidx = getelementptr inbounds i32, i32 addrspace(1)* %out, i32 %idxprom
49 store i32 %iter.0.lcssa, i32 addrspace(1)* %arrayidx, align 4
53 ; Function Attrs: nounwind readnone
54 declare i32 @llvm.amdgcn.workitem.id.x() #0
55 declare float @llvm.fmuladd.f32(float, float, float) #1
57 attributes #0 = { nounwind readnone }
58 attributes #1 = { readnone }