1 ; RUN: opt < %s -O3 -S | FileCheck %s
3 ; Make sure the call to syncthreads is not duplicate here by the LLVM
4 ; optimizations, because it has the noduplicate attribute set.
6 ; CHECK: call void @llvm.cuda.syncthreads
7 ; CHECK-NOT: call void @llvm.cuda.syncthreads
9 ; Function Attrs: nounwind
10 define void @foo(float* %output) #1 {
12 %output.addr = alloca float*, align 8
13 store float* %output, float** %output.addr, align 8
14 %0 = load float*, float** %output.addr, align 8
15 %arrayidx = getelementptr inbounds float, float* %0, i64 0
16 %1 = load float, float* %arrayidx, align 4
17 %conv = fpext float %1 to double
18 %cmp = fcmp olt double %conv, 1.000000e+01
19 br i1 %cmp, label %if.then, label %if.else
21 if.then: ; preds = %entry
22 %2 = load float*, float** %output.addr, align 8
23 %3 = load float, float* %2, align 4
24 %conv1 = fpext float %3 to double
25 %add = fadd double %conv1, 1.000000e+00
26 %conv2 = fptrunc double %add to float
27 store float %conv2, float* %2, align 4
30 if.else: ; preds = %entry
31 %4 = load float*, float** %output.addr, align 8
32 %5 = load float, float* %4, align 4
33 %conv3 = fpext float %5 to double
34 %add4 = fadd double %conv3, 2.000000e+00
35 %conv5 = fptrunc double %add4 to float
36 store float %conv5, float* %4, align 4
39 if.end: ; preds = %if.else, %if.then
40 call void @llvm.cuda.syncthreads()
41 %6 = load float*, float** %output.addr, align 8
42 %arrayidx6 = getelementptr inbounds float, float* %6, i64 0
43 %7 = load float, float* %arrayidx6, align 4
44 %conv7 = fpext float %7 to double
45 %cmp8 = fcmp olt double %conv7, 1.000000e+01
46 br i1 %cmp8, label %if.then9, label %if.else13
48 if.then9: ; preds = %if.end
49 %8 = load float*, float** %output.addr, align 8
50 %9 = load float, float* %8, align 4
51 %conv10 = fpext float %9 to double
52 %add11 = fadd double %conv10, 3.000000e+00
53 %conv12 = fptrunc double %add11 to float
54 store float %conv12, float* %8, align 4
57 if.else13: ; preds = %if.end
58 %10 = load float*, float** %output.addr, align 8
59 %11 = load float, float* %10, align 4
60 %conv14 = fpext float %11 to double
61 %add15 = fadd double %conv14, 4.000000e+00
62 %conv16 = fptrunc double %add15 to float
63 store float %conv16, float* %10, align 4
66 if.end17: ; preds = %if.else13, %if.then9
70 ; Function Attrs: noduplicate nounwind
71 declare void @llvm.cuda.syncthreads() #2
73 !0 = !{void (float*)* @foo, !"kernel", i32 1}
74 !1 = !{null, !"align", i32 8}