1; RUN: opt < %s -S -analyze -scalar-evolution | FileCheck %s
2
3; ScalarEvolution should be able to fold away the sign-extensions
4; on this loop with a primary induction variable incremented with
5; a nsw add of 2 (this test is derived from the nsw-offset.ll test, but uses an
6; assume instead of a preheader conditional branch to guard the loop).
7
8target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128"
9
10define void @foo(i32 %no, double* nocapture %d, double* nocapture %q) nounwind {
11entry:
12  %n = and i32 %no, 4294967294
13  %0 = icmp sgt i32 %n, 0                         ; <i1> [#uses=1]
14  tail call void @llvm.assume(i1 %0)
15  br label %bb.nph
16
17bb.nph:                                           ; preds = %entry
18  br label %bb
19
20bb:                                               ; preds = %bb.nph, %bb1
21  %i.01 = phi i32 [ %16, %bb1 ], [ 0, %bb.nph ]   ; <i32> [#uses=5]
22
23; CHECK: %1 = sext i32 %i.01 to i64
24; CHECK: -->  {0,+,2}<nuw><nsw><%bb>
25  %1 = sext i32 %i.01 to i64                      ; <i64> [#uses=1]
26
27; CHECK: %2 = getelementptr inbounds double, double* %d, i64 %1
28; CHECK: -->  {%d,+,16}<nsw><%bb>
29  %2 = getelementptr inbounds double, double* %d, i64 %1  ; <double*> [#uses=1]
30
31  %3 = load double, double* %2, align 8                   ; <double> [#uses=1]
32  %4 = sext i32 %i.01 to i64                      ; <i64> [#uses=1]
33  %5 = getelementptr inbounds double, double* %q, i64 %4  ; <double*> [#uses=1]
34  %6 = load double, double* %5, align 8                   ; <double> [#uses=1]
35  %7 = or i32 %i.01, 1                            ; <i32> [#uses=1]
36
37; CHECK: %8 = sext i32 %7 to i64
38; CHECK: -->  {1,+,2}<nuw><nsw><%bb>
39  %8 = sext i32 %7 to i64                         ; <i64> [#uses=1]
40
41; CHECK: %9 = getelementptr inbounds double, double* %q, i64 %8
42; CHECK: {(8 + %q)<nsw>,+,16}<nsw><%bb>
43  %9 = getelementptr inbounds double, double* %q, i64 %8  ; <double*> [#uses=1]
44
45; Artificially repeat the above three instructions, this time using
46; add nsw instead of or.
47  %t7 = add nsw i32 %i.01, 1                            ; <i32> [#uses=1]
48
49; CHECK: %t8 = sext i32 %t7 to i64
50; CHECK: -->  {1,+,2}<nuw><nsw><%bb>
51  %t8 = sext i32 %t7 to i64                         ; <i64> [#uses=1]
52
53; CHECK: %t9 = getelementptr inbounds double, double* %q, i64 %t8
54; CHECK: {(8 + %q)<nsw>,+,16}<nsw><%bb>
55  %t9 = getelementptr inbounds double, double* %q, i64 %t8  ; <double*> [#uses=1]
56
57  %10 = load double, double* %9, align 8                  ; <double> [#uses=1]
58  %11 = fadd double %6, %10                       ; <double> [#uses=1]
59  %12 = fadd double %11, 3.200000e+00             ; <double> [#uses=1]
60  %13 = fmul double %3, %12                       ; <double> [#uses=1]
61  %14 = sext i32 %i.01 to i64                     ; <i64> [#uses=1]
62  %15 = getelementptr inbounds double, double* %d, i64 %14 ; <double*> [#uses=1]
63  store double %13, double* %15, align 8
64  %16 = add nsw i32 %i.01, 2                      ; <i32> [#uses=2]
65  br label %bb1
66
67bb1:                                              ; preds = %bb
68  %17 = icmp slt i32 %16, %n                      ; <i1> [#uses=1]
69  br i1 %17, label %bb, label %bb1.return_crit_edge
70
71bb1.return_crit_edge:                             ; preds = %bb1
72  br label %return
73
74return:                                           ; preds = %bb1.return_crit_edge, %entry
75  ret void
76}
77
78declare void @llvm.assume(i1) nounwind
79
80; Note: Without the preheader assume, there is an 'smax' in the
81; backedge-taken count expression:
82; CHECK: Loop %bb: backedge-taken count is ((-1 + (2 * (%no /u 2))) /u 2)
83; CHECK: Loop %bb: max backedge-taken count is 1073741822
84