1 ;; This test is really dependent on propagating a lot of memory info around, but in the end, not
2 ;; screwing up a single add.
3 ; RUN: opt < %s -basicaa -newgvn -S | FileCheck %s
4 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
6 %struct.Letter = type { i32, i32, i32, i32 }
8 @alPhrase = external local_unnamed_addr global [26 x %struct.Letter], align 16
9 @aqMainMask = external local_unnamed_addr global [2 x i64], align 16
10 @aqMainSign = external local_unnamed_addr global [2 x i64], align 16
11 @cchPhraseLength = external local_unnamed_addr global i32, align 4
12 @auGlobalFrequency = external local_unnamed_addr global [26 x i32], align 16
13 @.str.7 = external hidden unnamed_addr constant [28 x i8], align 1
15 ; Function Attrs: nounwind uwtable
16 declare void @Fatal(i8*, i32) local_unnamed_addr #0
18 ; Function Attrs: nounwind readnone
19 declare i16** @__ctype_b_loc() local_unnamed_addr #1
21 ; Function Attrs: nounwind uwtable
22 define void @BuildMask(i8* nocapture readonly) local_unnamed_addr #0 {
23 tail call void @llvm.memset.p0i8.i64(i8* align 16 bitcast ([26 x %struct.Letter]* @alPhrase to i8*), i8 0, i64 416, i1 false)
24 tail call void @llvm.memset.p0i8.i64(i8* align 16 bitcast ([2 x i64]* @aqMainMask to i8*), i8 0, i64 16, i1 false)
25 tail call void @llvm.memset.p0i8.i64(i8* align 16 bitcast ([2 x i64]* @aqMainSign to i8*), i8 0, i64 16, i1 false)
28 .sink.split: ; preds = %14, %1
29 %.0 = phi i8* [ %0, %1 ], [ %.lcssa67, %14 ]
30 %.sink = phi i32 [ 0, %1 ], [ %23, %14 ]
31 store i32 %.sink, i32* @cchPhraseLength, align 4, !tbaa !1
34 ; <label>:2: ; preds = %6, %.sink.split
35 %.1 = phi i8* [ %.0, %.sink.split ], [ %3, %6 ]
36 %3 = getelementptr inbounds i8, i8* %.1, i64 1
37 %4 = load i8, i8* %.1, align 1, !tbaa !5
39 br i1 %5, label %.preheader.preheader, label %6
41 .preheader.preheader: ; preds = %2
44 ; <label>:6: ; preds = %2
45 %7 = tail call i16** @__ctype_b_loc() #4
46 %8 = load i16*, i16** %7, align 8, !tbaa !6
47 %9 = sext i8 %4 to i64
48 %10 = getelementptr inbounds i16, i16* %8, i64 %9
49 %11 = load i16, i16* %10, align 2, !tbaa !8
50 %12 = and i16 %11, 1024
51 %13 = icmp eq i16 %12, 0
52 br i1 %13, label %2, label %14
54 ; <label>:14: ; preds = %6
55 %.lcssa67 = phi i8* [ %3, %6 ]
56 %.lcssa65 = phi i8 [ %4, %6 ]
57 %15 = sext i8 %.lcssa65 to i32
58 %16 = tail call i32 @tolower(i32 %15) #5
59 %17 = add nsw i32 %16, -97
60 %18 = sext i32 %17 to i64
61 %19 = getelementptr inbounds [26 x %struct.Letter], [26 x %struct.Letter]* @alPhrase, i64 0, i64 %18, i32 0
62 %20 = load i32, i32* %19, align 16, !tbaa !10
64 store i32 %21, i32* %19, align 16, !tbaa !10
65 %22 = load i32, i32* @cchPhraseLength, align 4, !tbaa !1
66 %23 = add nsw i32 %22, 1
69 .preheader: ; preds = %58, %.preheader.preheader
70 %indvars.iv = phi i64 [ 0, %.preheader.preheader ], [ %indvars.iv.next, %58 ]
71 %.04961 = phi i32 [ %.2, %58 ], [ 0, %.preheader.preheader ]
72 %.05160 = phi i32 [ %.253, %58 ], [ 0, %.preheader.preheader ]
73 %24 = getelementptr inbounds [26 x %struct.Letter], [26 x %struct.Letter]* @alPhrase, i64 0, i64 %indvars.iv, i32 0
74 %25 = load i32, i32* %24, align 16, !tbaa !10
75 %26 = icmp eq i32 %25, 0
76 %27 = getelementptr inbounds [26 x i32], [26 x i32]* @auGlobalFrequency, i64 0, i64 %indvars.iv
77 br i1 %26, label %28, label %29
79 ; <label>:28: ; preds = %.preheader
80 store i32 -1, i32* %27, align 4, !tbaa !1
83 ; <label>:29: ; preds = %.preheader
84 store i32 0, i32* %27, align 4, !tbaa !1
85 %30 = zext i32 %25 to i64
86 br i1 false, label %._crit_edge, label %.lr.ph.preheader
88 .lr.ph.preheader: ; preds = %29
91 .lr.ph: ; preds = %.lr.ph, %.lr.ph.preheader
92 %.04658 = phi i64 [ %32, %.lr.ph ], [ 1, %.lr.ph.preheader ]
93 %.04857 = phi i32 [ %31, %.lr.ph ], [ 1, %.lr.ph.preheader ]
94 %31 = add nuw nsw i32 %.04857, 1
95 %32 = shl i64 %.04658, 1
96 %33 = icmp ult i64 %30, %32
97 br i1 %33, label %._crit_edge.loopexit, label %.lr.ph
99 ._crit_edge.loopexit: ; preds = %.lr.ph
100 %.lcssa63 = phi i32 [ %31, %.lr.ph ]
101 %.lcssa = phi i64 [ %32, %.lr.ph ]
102 br label %._crit_edge
104 ._crit_edge: ; preds = %._crit_edge.loopexit, %29
105 %.048.lcssa = phi i32 [ 1, %29 ], [ %.lcssa63, %._crit_edge.loopexit ]
106 %.046.lcssa = phi i64 [ 1, %29 ], [ %.lcssa, %._crit_edge.loopexit ]
107 %34 = add nsw i32 %.048.lcssa, %.04961
108 %35 = icmp ugt i32 %34, 64
109 br i1 %35, label %36, label %40
111 ; <label>:36: ; preds = %._crit_edge
112 ; This testcase essentially comes down to this little add.
113 ; If we screw up the revisitation of the users of store of %sink above
114 ; we will end up propagating and simplifying this to 1 in the final output
115 ; because we keep an optimistic assumption we should not.
116 ; CHECK: add i32 %.05160, 1
117 %37 = add i32 %.05160, 1
118 %38 = icmp ugt i32 %37, 1
119 br i1 %38, label %39, label %40
121 ; <label>:39: ; preds = %36
122 tail call void @Fatal(i8* getelementptr inbounds ([28 x i8], [28 x i8]* @.str.7, i64 0, i64 0), i32 0)
125 ; <label>:40: ; preds = %39, %36, %._crit_edge
126 %.152 = phi i32 [ %.05160, %._crit_edge ], [ %37, %39 ], [ %37, %36 ]
127 %.150 = phi i32 [ %.04961, %._crit_edge ], [ 0, %39 ], [ 0, %36 ]
128 %41 = add i64 %.046.lcssa, 4294967295
129 %42 = trunc i64 %41 to i32
130 %43 = getelementptr inbounds [26 x %struct.Letter], [26 x %struct.Letter]* @alPhrase, i64 0, i64 %indvars.iv, i32 2
131 store i32 %42, i32* %43, align 8, !tbaa !12
132 %44 = zext i32 %.150 to i64
133 %.046. = shl i64 %.046.lcssa, %44
134 %45 = zext i32 %.152 to i64
135 %46 = getelementptr inbounds [2 x i64], [2 x i64]* @aqMainSign, i64 0, i64 %45
136 %47 = load i64, i64* %46, align 8, !tbaa !13
137 %48 = or i64 %47, %.046.
138 store i64 %48, i64* %46, align 8, !tbaa !13
139 %49 = load i32, i32* %24, align 16, !tbaa !10
140 %50 = zext i32 %49 to i64
141 %51 = shl i64 %50, %44
142 %52 = getelementptr inbounds [2 x i64], [2 x i64]* @aqMainMask, i64 0, i64 %45
143 %53 = load i64, i64* %52, align 8, !tbaa !13
144 %54 = or i64 %51, %53
145 store i64 %54, i64* %52, align 8, !tbaa !13
146 %55 = getelementptr inbounds [26 x %struct.Letter], [26 x %struct.Letter]* @alPhrase, i64 0, i64 %indvars.iv, i32 1
147 store i32 %.150, i32* %55, align 4, !tbaa !15
148 %56 = getelementptr inbounds [26 x %struct.Letter], [26 x %struct.Letter]* @alPhrase, i64 0, i64 %indvars.iv, i32 3
149 store i32 %.152, i32* %56, align 4, !tbaa !16
150 %57 = add nsw i32 %.150, %.048.lcssa
153 ; <label>:58: ; preds = %40, %28
154 %.253 = phi i32 [ %.05160, %28 ], [ %.152, %40 ]
155 %.2 = phi i32 [ %.04961, %28 ], [ %57, %40 ]
156 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
157 %exitcond = icmp ne i64 %indvars.iv.next, 26
158 br i1 %exitcond, label %.preheader, label %59
160 ; <label>:59: ; preds = %58
164 ; Function Attrs: argmemonly nounwind
165 declare void @llvm.memset.p0i8.i64(i8* nocapture writeonly, i8, i64, i1) #2
167 ; Function Attrs: inlinehint nounwind readonly uwtable
168 declare i32 @tolower(i32) local_unnamed_addr #3
170 attributes #0 = { nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="false" "use-soft-float"="false" }
171 attributes #1 = { nounwind readnone "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="false" "use-soft-float"="false" }
172 attributes #2 = { argmemonly nounwind }
173 attributes #3 = { inlinehint nounwind readonly uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="false" "use-soft-float"="false" }
174 attributes #4 = { nounwind readnone }
175 attributes #5 = { nounwind readonly }
179 !0 = !{!"clang version 4.0.0"}
180 !1 = !{!2, !2, i64 0}
181 !2 = !{!"int", !3, i64 0}
182 !3 = !{!"omnipotent char", !4, i64 0}
183 !4 = !{!"Simple C/C++ TBAA"}
184 !5 = !{!3, !3, i64 0}
185 !6 = !{!7, !7, i64 0}
186 !7 = !{!"any pointer", !3, i64 0}
187 !8 = !{!9, !9, i64 0}
188 !9 = !{!"short", !3, i64 0}
189 !10 = !{!11, !2, i64 0}
190 !11 = !{!"", !2, i64 0, !2, i64 4, !2, i64 8, !2, i64 12}
191 !12 = !{!11, !2, i64 8}
192 !13 = !{!14, !14, i64 0}
193 !14 = !{!"long", !3, i64 0}
194 !15 = !{!11, !2, i64 4}
195 !16 = !{!11, !2, i64 12}