• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1; RUN: opt %loadPolly -analyze -polly-scops < %s | FileCheck %s --check-prefix=SCOP
2; RUN: opt %loadPolly -analyze -polly-codegen-ppcg -polly-acc-dump-kernel-ir < %s | FileCheck %s --check-prefix=KERNEL-IR
3; RUN: opt %loadPolly -S -polly-codegen-ppcg  < %s | FileCheck %s --check-prefix=HOST-IR
4
5; Test that we do recognise and codegen a kernel that has intrinsics.
6
7; REQUIRES: pollyacc
8
9; Check that we model the kernel as a scop.
10; SCOP:      Function: f
11; SCOP-NEXT:       Region: %entry.split---%for.end
12
13; Check that the intrinsic call is present in the kernel IR.
14; KERNEL-IR:   %p_sqrt = tail call float @llvm.sqrt.f32(float %A.arr.i.val_p_scalar_)
15; KERNEL-IR:   declare float @llvm.sqrt.f32(float)
16; KERNEL-IR:   declare float @llvm.fabs.f32(float)
17
18
19; Check that kernel launch is generated in host IR.
20; the declare would not be generated unless a call to a kernel exists.
21; HOST-IR: declare void @polly_launchKernel(i8*, i32, i32, i32, i32, i32, i8*)
22
23
24; void f(float *A, float *B, int N) {
25;   for(int i = 0; i < N; i++) {
26;       float tmp0 = A[i];
27;       float tmp1 = sqrt(tmp1);
28;       float tmp2 = fabs(tmp2);
29;       float tmp3 = copysignf(tmp1, tmp2);
30;       B[i] = tmp4;
31;   }
32; }
33
34target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
35
36define void @f(float* %A, float* %B, i32 %N) {
37entry:
38  br label %entry.split
39
40entry.split:                                      ; preds = %entry
41  %cmp1 = icmp sgt i32 %N, 0
42  br i1 %cmp1, label %for.body.lr.ph, label %for.end
43
44for.body.lr.ph:                                   ; preds = %entry.split
45  br label %for.body
46
47for.body:                                         ; preds = %for.body.lr.ph, %for.body
48  %indvars.iv = phi i64 [ 0, %for.body.lr.ph ], [ %indvars.iv.next, %for.body ]
49  %A.arr.i = getelementptr inbounds float, float* %A, i64 %indvars.iv
50  %A.arr.i.val = load float, float* %A.arr.i, align 4
51  ; Call to intrinsics that should be part of the kernel.
52  %sqrt = tail call float @llvm.sqrt.f32(float %A.arr.i.val)
53  %fabs = tail call float @llvm.fabs.f32(float %sqrt);
54  %copysign = tail call float @llvm.copysign.f32(float %sqrt, float %fabs);
55  %B.arr.i = getelementptr inbounds float, float* %B, i64 %indvars.iv
56  store float %copysign, float* %B.arr.i, align 4
57
58  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
59  %wide.trip.count = zext i32 %N to i64
60  %exitcond = icmp ne i64 %indvars.iv.next, %wide.trip.count
61  br i1 %exitcond, label %for.body, label %for.cond.for.end_crit_edge
62
63for.cond.for.end_crit_edge:                       ; preds = %for.body
64  br label %for.end
65
66for.end:                                          ; preds = %for.cond.for.end_crit_edge, %entry.split
67  ret void
68}
69
70; Function Attrs: nounwind readnone
71declare float @llvm.sqrt.f32(float) #0
72declare float @llvm.fabs.f32(float) #0
73declare float @llvm.copysign.f32(float, float) #0
74
75attributes #0 = { nounwind readnone }
76
77