1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt -loop-vectorize -instcombine -force-vector-width=4 -S < %s 2>&1 | FileCheck %s
3
4target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
5
6;;;; Derived from the following C code
7;; void forked_ptrs_different_base_same_offset(float *A, float *B, float *C, int *D) {
8;;   for (int i=0; i<100; i++) {
9;;     if (D[i] != 0) {
10;;       C[i] = A[i];
11;;     } else {
12;;       C[i] = B[i];
13;;     }
14;;   }
15;; }
16
17define dso_local void @forked_ptrs_different_base_same_offset(float* nocapture readonly %Base1, float* nocapture readonly %Base2, float* nocapture %Dest, i32* nocapture readonly %Preds) {
18; CHECK-LABEL: @forked_ptrs_different_base_same_offset(
19; CHECK-NEXT:  entry:
20; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
21; CHECK:       for.cond.cleanup:
22; CHECK-NEXT:    ret void
23; CHECK:       for.body:
24; CHECK-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]
25; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[PREDS:%.*]], i64 [[INDVARS_IV]]
26; CHECK-NEXT:    [[TMP0:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
27; CHECK-NEXT:    [[CMP1_NOT:%.*]] = icmp eq i32 [[TMP0]], 0
28; CHECK-NEXT:    [[SPEC_SELECT:%.*]] = select i1 [[CMP1_NOT]], float* [[BASE2:%.*]], float* [[BASE1:%.*]]
29; CHECK-NEXT:    [[DOTSINK_IN:%.*]] = getelementptr inbounds float, float* [[SPEC_SELECT]], i64 [[INDVARS_IV]]
30; CHECK-NEXT:    [[DOTSINK:%.*]] = load float, float* [[DOTSINK_IN]], align 4
31; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds float, float* [[DEST:%.*]], i64 [[INDVARS_IV]]
32; CHECK-NEXT:    store float [[DOTSINK]], float* [[TMP1]], align 4
33; CHECK-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
34; CHECK-NEXT:    [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], 100
35; CHECK-NEXT:    br i1 [[EXITCOND_NOT]], label [[FOR_COND_CLEANUP:%.*]], label [[FOR_BODY]]
36;
37entry:
38  br label %for.body
39
40for.cond.cleanup:
41  ret void
42
43for.body:
44  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
45  %arrayidx = getelementptr inbounds i32, i32* %Preds, i64 %indvars.iv
46  %0 = load i32, i32* %arrayidx, align 4
47  %cmp1.not = icmp eq i32 %0, 0
48  %spec.select = select i1 %cmp1.not, float* %Base2, float* %Base1
49  %.sink.in = getelementptr inbounds float, float* %spec.select, i64 %indvars.iv
50  %.sink = load float, float* %.sink.in, align 4
51  %1 = getelementptr inbounds float, float* %Dest, i64 %indvars.iv
52  store float %.sink, float* %1, align 4
53  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
54  %exitcond.not = icmp eq i64 %indvars.iv.next, 100
55  br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
56}
57
58;;;; Derived from the following C code
59;; void forked_ptrs_same_base_different_offset(float *A, float *B, int *C) {
60;;   int offset;
61;;   for (int i = 0; i < 100; i++) {
62;;     if (C[i] != 0)
63;;       offset = i;
64;;     else
65;;       offset = i+1;
66;;     B[i] = A[offset];
67;;   }
68;; }
69
70define dso_local void @forked_ptrs_same_base_different_offset(float* nocapture readonly %Base, float* nocapture %Dest, i32* nocapture readonly %Preds) {
71; CHECK-LABEL: @forked_ptrs_same_base_different_offset(
72; CHECK-NEXT:  entry:
73; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
74; CHECK:       for.cond.cleanup:
75; CHECK-NEXT:    ret void
76; CHECK:       for.body:
77; CHECK-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]
78; CHECK-NEXT:    [[I_014:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[ADD:%.*]], [[FOR_BODY]] ]
79; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[PREDS:%.*]], i64 [[INDVARS_IV]]
80; CHECK-NEXT:    [[TMP0:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
81; CHECK-NEXT:    [[CMP1_NOT:%.*]] = icmp eq i32 [[TMP0]], 0
82; CHECK-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
83; CHECK-NEXT:    [[ADD]] = add nuw nsw i32 [[I_014]], 1
84; CHECK-NEXT:    [[TMP1:%.*]] = trunc i64 [[INDVARS_IV]] to i32
85; CHECK-NEXT:    [[OFFSET_0:%.*]] = select i1 [[CMP1_NOT]], i32 [[ADD]], i32 [[TMP1]]
86; CHECK-NEXT:    [[IDXPROM213:%.*]] = zext i32 [[OFFSET_0]] to i64
87; CHECK-NEXT:    [[ARRAYIDX3:%.*]] = getelementptr inbounds float, float* [[BASE:%.*]], i64 [[IDXPROM213]]
88; CHECK-NEXT:    [[TMP2:%.*]] = load float, float* [[ARRAYIDX3]], align 4
89; CHECK-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds float, float* [[DEST:%.*]], i64 [[INDVARS_IV]]
90; CHECK-NEXT:    store float [[TMP2]], float* [[ARRAYIDX5]], align 4
91; CHECK-NEXT:    [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], 100
92; CHECK-NEXT:    br i1 [[EXITCOND_NOT]], label [[FOR_COND_CLEANUP:%.*]], label [[FOR_BODY]]
93;
94entry:
95  br label %for.body
96
97for.cond.cleanup:                                 ; preds = %for.body
98  ret void
99
100for.body:                                         ; preds = %entry, %for.body
101  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
102  %i.014 = phi i32 [ 0, %entry ], [ %add, %for.body ]
103  %arrayidx = getelementptr inbounds i32, i32* %Preds, i64 %indvars.iv
104  %0 = load i32, i32* %arrayidx, align 4
105  %cmp1.not = icmp eq i32 %0, 0
106  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
107  %add = add nuw nsw i32 %i.014, 1
108  %1 = trunc i64 %indvars.iv to i32
109  %offset.0 = select i1 %cmp1.not, i32 %add, i32 %1
110  %idxprom213 = zext i32 %offset.0 to i64
111  %arrayidx3 = getelementptr inbounds float, float* %Base, i64 %idxprom213
112  %2 = load float, float* %arrayidx3, align 4
113  %arrayidx5 = getelementptr inbounds float, float* %Dest, i64 %indvars.iv
114  store float %2, float* %arrayidx5, align 4
115  %exitcond.not = icmp eq i64 %indvars.iv.next, 100
116  br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
117}
118