|
83
|
1 ; RUN: llc < %s -march=nvptx -mcpu=sm_20 | FileCheck %s
|
|
|
2
|
|
|
3 target datalayout = "e-i64:64-v16:16-v32:32-n16:32:64"
|
|
|
4 target triple = "nvptx64-nvidia-cuda"
|
|
|
5
|
|
|
6 %class.float3 = type { float, float, float }
|
|
|
7
|
|
|
8 ; Function Attrs: nounwind
|
|
|
9 ; CHECK-LABEL: some_kernel
|
|
|
10 define void @some_kernel(%class.float3* nocapture %dst) #0 {
|
|
|
11 _ZL11compute_vecRK6float3jb.exit:
|
|
|
12 %ret_vec.sroa.8.i = alloca float, align 4
|
|
|
13 %0 = tail call i32 @llvm.ptx.read.ctaid.x()
|
|
|
14 %1 = tail call i32 @llvm.ptx.read.ntid.x()
|
|
|
15 %2 = mul nsw i32 %1, %0
|
|
|
16 %3 = tail call i32 @llvm.ptx.read.tid.x()
|
|
|
17 %4 = add nsw i32 %2, %3
|
|
|
18 %5 = zext i32 %4 to i64
|
|
|
19 %6 = bitcast float* %ret_vec.sroa.8.i to i8*
|
|
|
20 call void @llvm.lifetime.start(i64 4, i8* %6)
|
|
|
21 %7 = and i32 %4, 15
|
|
|
22 %8 = icmp eq i32 %7, 0
|
|
|
23 %9 = select i1 %8, float 0.000000e+00, float -1.000000e+00
|
|
|
24 store float %9, float* %ret_vec.sroa.8.i, align 4
|
|
|
25 ; CHECK: setp.lt.f32 %p{{[0-9]+}}, %f{{[0-9]+}}, 0f00000000
|
|
|
26 %10 = fcmp olt float %9, 0.000000e+00
|
|
95
|
27 %ret_vec.sroa.8.i.val = load float, float* %ret_vec.sroa.8.i, align 4
|
|
83
|
28 %11 = select i1 %10, float 0.000000e+00, float %ret_vec.sroa.8.i.val
|
|
|
29 call void @llvm.lifetime.end(i64 4, i8* %6)
|
|
95
|
30 %12 = getelementptr inbounds %class.float3, %class.float3* %dst, i64 %5, i32 0
|
|
83
|
31 store float 0.000000e+00, float* %12, align 4
|
|
95
|
32 %13 = getelementptr inbounds %class.float3, %class.float3* %dst, i64 %5, i32 1
|
|
83
|
33 store float %11, float* %13, align 4
|
|
95
|
34 %14 = getelementptr inbounds %class.float3, %class.float3* %dst, i64 %5, i32 2
|
|
83
|
35 store float 0.000000e+00, float* %14, align 4
|
|
|
36 ret void
|
|
|
37 }
|
|
|
38
|
|
|
39 ; Function Attrs: nounwind readnone
|
|
|
40 declare i32 @llvm.ptx.read.ctaid.x() #1
|
|
|
41
|
|
|
42 ; Function Attrs: nounwind readnone
|
|
|
43 declare i32 @llvm.ptx.read.ntid.x() #1
|
|
|
44
|
|
|
45 ; Function Attrs: nounwind readnone
|
|
|
46 declare i32 @llvm.ptx.read.tid.x() #1
|
|
|
47
|
|
|
48 ; Function Attrs: nounwind
|
|
|
49 declare void @llvm.lifetime.start(i64, i8* nocapture) #2
|
|
|
50
|
|
|
51 ; Function Attrs: nounwind
|
|
|
52 declare void @llvm.lifetime.end(i64, i8* nocapture) #2
|
|
|
53
|
|
|
54 attributes #0 = { nounwind "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "stack-protector-buffer-size"="8" "unsafe-fp-math"="true" "use-soft-float"="false" }
|
|
|
55 attributes #1 = { nounwind readnone }
|
|
|
56 attributes #2 = { nounwind }
|
|
|
57
|
|
|
58 !nvvm.annotations = !{!0}
|
|
|
59 !llvm.ident = !{!1}
|
|
|
60
|
|
|
61 !0 = !{void (%class.float3*)* @some_kernel, !"kernel", i32 1}
|
|
|
62 !1 = !{!"clang version 3.5.1 (tags/RELEASE_351/final)"}
|
