1; RUN: opt < %s -S -loop-unroll -unroll-max-iteration-count-to-analyze=100 -unroll-dynamic-cost-savings-discount=1000 -unroll-threshold=10 -unroll-percent-dynamic-cost-saved-threshold=50 | FileCheck %s
2target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
3
4@known_constant = internal unnamed_addr constant [10 x i32] [i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1], align 16
5
6; We should be able to propagate constant data through different types of
7; casts. For example, in this test we have a load, which becomes constant after
8; unrolling, which then is truncated to i8. Obviously, truncated value is also a
9; constant, which can be used in the further simplifications.
10;
11; We expect this loop to be unrolled, because in this case load would become
12; constant, which is 0 in many cases, and which, in its turn, helps to simplify
13; following multiplication and addition. In total, unrolling should help to
14; optimize  ~60% of all instructions in this case.
15;
16; CHECK-LABEL: @const_load_trunc
17; CHECK-NOT: br i1
18; CHECK: ret i8 %
19define i8 @const_load_trunc(i32* noalias nocapture readonly %src) {
20entry:
21  br label %loop
22
23loop:                                                ; preds = %loop, %entry
24  %iv = phi i64 [ 0, %entry ], [ %inc, %loop ]
25  %r  = phi i8 [ 0, %entry ], [ %add, %loop ]
26  %arrayidx = getelementptr inbounds i32, i32* %src, i64 %iv
27  %src_element = load i32, i32* %arrayidx, align 4
28  %array_const_idx = getelementptr inbounds [10 x i32], [10 x i32]* @known_constant, i64 0, i64 %iv
29  %const_array_element = load i32, i32* %array_const_idx, align 4
30  %x = trunc i32 %src_element to i8
31  %y = trunc i32 %const_array_element to i8
32  %mul = mul nsw i8 %x, %y
33  %add = add nsw i8 %mul, %r
34  %inc = add nuw nsw i64 %iv, 1
35  %exitcond86.i = icmp eq i64 %inc, 10
36  br i1 %exitcond86.i, label %loop.end, label %loop
37
38loop.end:                                            ; preds = %loop
39  %r.lcssa = phi i8 [ %r, %loop ]
40  ret i8 %r.lcssa
41}
42
43; The same test as before, but with ZEXT instead of TRUNC.
44; CHECK-LABEL: @const_load_zext
45; CHECK-NOT: br i1
46; CHECK: ret i64 %
47define i64 @const_load_zext(i32* noalias nocapture readonly %src) {
48entry:
49  br label %loop
50
51loop:                                                ; preds = %loop, %entry
52  %iv = phi i64 [ 0, %entry ], [ %inc, %loop ]
53  %r  = phi i64 [ 0, %entry ], [ %add, %loop ]
54  %arrayidx = getelementptr inbounds i32, i32* %src, i64 %iv
55  %src_element = load i32, i32* %arrayidx, align 4
56  %array_const_idx = getelementptr inbounds [10 x i32], [10 x i32]* @known_constant, i64 0, i64 %iv
57  %const_array_element = load i32, i32* %array_const_idx, align 4
58  %x = zext i32 %src_element to i64
59  %y = zext i32 %const_array_element to i64
60  %mul = mul nsw i64 %x, %y
61  %add = add nsw i64 %mul, %r
62  %inc = add nuw nsw i64 %iv, 1
63  %exitcond86.i = icmp eq i64 %inc, 10
64  br i1 %exitcond86.i, label %loop.end, label %loop
65
66loop.end:                                            ; preds = %loop
67  %r.lcssa = phi i64 [ %r, %loop ]
68  ret i64 %r.lcssa
69}
70
71; The same test as the first one, but with SEXT instead of TRUNC.
72; CHECK-LABEL: @const_load_sext
73; CHECK-NOT: br i1
74; CHECK: ret i64 %
75define i64 @const_load_sext(i32* noalias nocapture readonly %src) {
76entry:
77  br label %loop
78
79loop:                                                ; preds = %loop, %entry
80  %iv = phi i64 [ 0, %entry ], [ %inc, %loop ]
81  %r  = phi i64 [ 0, %entry ], [ %add, %loop ]
82  %arrayidx = getelementptr inbounds i32, i32* %src, i64 %iv
83  %src_element = load i32, i32* %arrayidx, align 4
84  %array_const_idx = getelementptr inbounds [10 x i32], [10 x i32]* @known_constant, i64 0, i64 %iv
85  %const_array_element = load i32, i32* %array_const_idx, align 4
86  %x = sext i32 %src_element to i64
87  %y = sext i32 %const_array_element to i64
88  %mul = mul nsw i64 %x, %y
89  %add = add nsw i64 %mul, %r
90  %inc = add nuw nsw i64 %iv, 1
91  %exitcond86.i = icmp eq i64 %inc, 10
92  br i1 %exitcond86.i, label %loop.end, label %loop
93
94loop.end:                                            ; preds = %loop
95  %r.lcssa = phi i64 [ %r, %loop ]
96  ret i64 %r.lcssa
97}
98