• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1; RUN: opt %loadPolly -analyze -polly-scops < %s | FileCheck %s -check-prefix=SCOP
2; RUN: opt %loadPolly -S -polly-codegen-ppcg < %s | FileCheck %s -check-prefix=HOST-IR
3
4; REQUIRES: pollyacc
5
6; Check that we detect a scop.
7; SCOP:       Function: checkScalarKill
8; SCOP-NEXT: Region: %XLoopInit---%for.end
9; SCOP-NEXT: Max Loop Depth:  1
10
11; Check that we have a scalar that is not a phi node in the scop.
12; SCOP: i32 MemRef_x_0; // Element size 4
13
14; Check that kernel launch is generated in host IR.
15; the declare would not be generated unless a call to a kernel exists.
16; HOST-IR: declare void @polly_launchKernel(i8*, i32, i32, i32, i32, i32, i8*)
17
18; Check that we add variables that are local to a scop into the kills that we
19; pass to PPCG. This should enable PPCG to codegen this example.
20; void checkScalarKill(int A[], int B[], int C[], const int control1, int control2) {
21; int x;
22; #pragma scop
23;     for(int i = 0; i < 1000; i++) {
24; XLoopInit:        x = 0;
25;
26;         if (control1 > 2)
27;             C1Add: x += 10;
28;         if (control2 > 3)
29;             C2Add: x += A[i];
30;
31; BLoopAccumX:        B[i] += x;
32;     }
33;
34; #pragma endscop
35; }
36; ModuleID = 'test.ll'
37target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
38
39define void @checkScalarKill(i32* %A, i32* %B, i32* %C, i32 %control1, i32 %control2) {
40entry:
41  br label %entry.split
42
43entry.split:                                      ; preds = %entry
44  br label %XLoopInit
45
46XLoopInit:                                        ; preds = %entry.split, %BLoopAccumX
47  %indvars.iv = phi i64 [ 0, %entry.split ], [ %indvars.iv.next, %BLoopAccumX ]
48  %cmp1 = icmp sgt i32 %control1, 2
49  %x.0 = select i1 %cmp1, i32 10, i32 0
50  %cmp2 = icmp sgt i32 %control2, 3
51  br i1 %cmp2, label %C2Add, label %BLoopAccumX
52
53C2Add:                                            ; preds = %XLoopInit
54  %arrayidx = getelementptr inbounds i32, i32* %A, i64 %indvars.iv
55  %tmp6 = load i32, i32* %arrayidx, align 4
56  %add4 = add nsw i32 %tmp6, %x.0
57  br label %BLoopAccumX
58
59BLoopAccumX:                                      ; preds = %XLoopInit, %C2Add
60  %x.1 = phi i32 [ %add4, %C2Add ], [ %x.0, %XLoopInit ]
61  %arrayidx7 = getelementptr inbounds i32, i32* %B, i64 %indvars.iv
62  %tmp11 = load i32, i32* %arrayidx7, align 4
63  %add8 = add nsw i32 %tmp11, %x.1
64  store i32 %add8, i32* %arrayidx7, align 4
65  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
66  %exitcond = icmp ne i64 %indvars.iv.next, 1000
67  br i1 %exitcond, label %XLoopInit, label %for.end
68
69for.end:                                          ; preds = %BLoopAccumX
70  ret void
71}
72