|
150
|
1 ; RUN: opt < %s -S -loop-unroll -unroll-max-iteration-count-to-analyze=100 -unroll-threshold=10 -unroll-max-percent-threshold-boost=200 | FileCheck %s
|
|
|
2 ; RUN: opt < %s -S -passes='require<opt-remark-emit>,loop(unroll-full)' -unroll-max-iteration-count-to-analyze=100 -unroll-threshold=10 -unroll-max-percent-threshold-boost=200 | FileCheck %s
|
|
|
3 target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
|
|
|
4
|
|
|
5 @known_constant = internal unnamed_addr constant [10 x i32] [i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1], align 16
|
|
|
6
|
|
|
7 ; If we can figure out result of comparison on each iteration, we can resolve
|
|
|
8 ; the depending branch. That means, that the unrolled version of the loop would
|
|
|
9 ; have less code, because we don't need not-taken basic blocks there.
|
|
|
10 ; This test checks that this is taken into consideration.
|
|
|
11 ; We expect this loop to be unrolled, because the most complicated part of its
|
|
|
12 ; body (if.then block) is never actually executed.
|
|
|
13 ; CHECK-LABEL: @branch_folded
|
|
|
14 ; CHECK-NOT: br i1 %
|
|
|
15 ; CHECK: ret i32
|
|
|
16 define i32 @branch_folded(i32* noalias nocapture readonly %b) {
|
|
|
17 entry:
|
|
|
18 br label %for.body
|
|
|
19
|
|
|
20 for.body: ; preds = %for.inc, %entry
|
|
|
21 %iv.0 = phi i64 [ 0, %entry ], [ %iv.1, %for.inc ]
|
|
|
22 %r.0 = phi i32 [ 0, %entry ], [ %r.1, %for.inc ]
|
|
|
23 %arrayidx1 = getelementptr inbounds [10 x i32], [10 x i32]* @known_constant, i64 0, i64 %iv.0
|
|
|
24 %x1 = load i32, i32* %arrayidx1, align 4
|
|
|
25 %cmp = icmp eq i32 %x1, 0
|
|
|
26 %iv.1 = add nuw nsw i64 %iv.0, 1
|
|
|
27 br i1 %cmp, label %if.then, label %for.inc
|
|
|
28
|
|
|
29 if.then: ; preds = %for.body
|
|
|
30 %arrayidx2 = getelementptr inbounds i32, i32* %b, i64 %iv.0
|
|
|
31 %x2 = load i32, i32* %arrayidx2, align 4
|
|
|
32 %add = add nsw i32 %x2, %r.0
|
|
|
33 br label %for.inc
|
|
|
34
|
|
|
35 for.inc: ; preds = %for.body, %if.then
|
|
|
36 %r.1 = phi i32 [ %add, %if.then ], [ %x1, %for.body ]
|
|
|
37 %exitcond = icmp eq i64 %iv.1, 10
|
|
|
38 br i1 %exitcond, label %for.end, label %for.body
|
|
|
39
|
|
|
40 for.end: ; preds = %for.inc
|
|
|
41 ret i32 %r.1
|
|
|
42 }
|
|
|
43
|
|
|
44 ; Check that we don't crash when we analyze icmp with pointer-typed IV and a
|
|
|
45 ; pointer.
|
|
|
46 ; CHECK-LABEL: @ptr_cmp_crash
|
|
|
47 ; CHECK: ret void
|
|
|
48 define void @ptr_cmp_crash() {
|
|
|
49 entry:
|
|
|
50 br label %while.body
|
|
|
51
|
|
|
52 while.body:
|
|
|
53 %iv.0 = phi i32* [ getelementptr inbounds ([10 x i32], [10 x i32]* @known_constant, i64 0, i64 0), %entry ], [ %iv.1, %while.body ]
|
|
|
54 %iv.1 = getelementptr inbounds i32, i32* %iv.0, i64 1
|
|
|
55 %exitcond = icmp eq i32* %iv.1, getelementptr inbounds ([10 x i32], [10 x i32]* @known_constant, i64 0, i64 9)
|
|
|
56 br i1 %exitcond, label %loop.exit, label %while.body
|
|
|
57
|
|
|
58 loop.exit:
|
|
|
59 ret void
|
|
|
60 }
|
|
|
61
|
|
|
62 ; Check that we don't crash when we analyze ptrtoint cast.
|
|
|
63 ; CHECK-LABEL: @ptrtoint_cast_crash
|
|
|
64 ; CHECK: ret void
|
|
|
65 define void @ptrtoint_cast_crash(i8 * %a) {
|
|
|
66 entry:
|
|
|
67 %limit = getelementptr i8, i8* %a, i64 512
|
|
|
68 br label %loop.body
|
|
|
69
|
|
|
70 loop.body:
|
|
|
71 %iv.0 = phi i8* [ %a, %entry ], [ %iv.1, %loop.body ]
|
|
|
72 %cast = ptrtoint i8* %iv.0 to i64
|
|
|
73 %iv.1 = getelementptr inbounds i8, i8* %iv.0, i64 1
|
|
|
74 %exitcond = icmp ne i8* %iv.1, %limit
|
|
|
75 br i1 %exitcond, label %loop.body, label %loop.exit
|
|
|
76
|
|
|
77 loop.exit:
|
|
|
78 ret void
|
|
|
79 }
|
