1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
|
; RUN: opt %loadPolly -polly-process-unprofitable=false -polly-print-detect -disable-output < %s | FileCheck %s
; RUN: opt %loadPolly -polly-allow-nonaffine-loops -polly-print-detect -disable-output < %s | FileCheck %s --check-prefix=NALOOPS
; RUN: opt %loadPolly -polly-allow-nonaffine-loops -polly-process-unprofitable=false -polly-print-detect -disable-output < %s | FileCheck %s --check-prefix=PROFIT
; The latch conditions of the outer loop are not affine, thus the loop cannot
; handled by the domain generation and needs to be overapproximated.
; CHECK-NOT: Valid
; NALOOPS: Valid Region for Scop: for.body.6 => for.end.45
; PROFIT-NOT: Valid
; ModuleID = '/home/johannes/Downloads/bug.ll'
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
; Function Attrs: nounwind uwtable
define void @kernel_reg_detect(ptr %path) #0 {
entry:
br label %for.body.6
for.body.6: ; preds = %for.inc.43, %for.body.6, %entry
%indvars.iv9 = phi i64 [ %indvars.iv.next10, %for.body.6 ], [ 0, %for.inc.43 ], [ 0, %entry ]
%indvars.iv.next10 = add nuw nsw i64 %indvars.iv9, 1
%lftr.wideiv = trunc i64 %indvars.iv.next10 to i32
%exitcond = icmp ne i32 %lftr.wideiv, 6
br i1 %exitcond, label %for.body.6, label %for.inc.40
for.inc.40: ; preds = %for.inc.40, %for.body.6
%tmp = load i32, ptr %path, align 4
store i32 0, ptr %path, align 4
%exitcond22 = icmp ne i64 0, 6
br i1 %exitcond22, label %for.inc.40, label %for.inc.43
for.inc.43: ; preds = %for.inc.40
%exitcond23 = icmp ne i32 0, 10000
br i1 %exitcond23, label %for.body.6, label %for.end.45
for.end.45: ; preds = %for.inc.43
ret void
}
|
