1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s  -loop-vectorize -force-vector-interleave=1 -force-vector-width=4 -dce -instcombine -S | FileCheck %s
3
4target 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-n8:16:32:64-S128"
5
6; Make sure we vectorize this loop:
7; int foo(float *a, float *b, int n) {
8;   for (int i=0; i<n; ++i)
9;     a[i] = b[i] * 3;
10; }
11
12define i32 @foo(float* nocapture %a, float* nocapture %b, i32 %n) nounwind uwtable ssp {
13; CHECK-LABEL: @foo(
14; CHECK-NEXT:  entry:
15; CHECK-NEXT:    [[CMP6:%.*]] = icmp sgt i32 [[N:%.*]], 0, !dbg !4
16; CHECK-NEXT:    br i1 [[CMP6]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_END:%.*]], !dbg !4
17; CHECK:       for.body.preheader:
18; CHECK-NEXT:    [[TMP0:%.*]] = add i32 [[N]], -1, !dbg !9
19; CHECK-NEXT:    [[TMP1:%.*]] = zext i32 [[TMP0]] to i64, !dbg !9
20; CHECK-NEXT:    [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1, !dbg !9
21; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP2]], 4, !dbg !9
22; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]], !dbg !9
23; CHECK:       vector.memcheck:
24; CHECK-NEXT:    [[TMP3:%.*]] = add i32 [[N]], -1, !dbg !9
25; CHECK-NEXT:    [[TMP4:%.*]] = zext i32 [[TMP3]] to i64, !dbg !9
26; CHECK-NEXT:    [[TMP5:%.*]] = add nuw nsw i64 [[TMP4]], 1, !dbg !9
27; CHECK-NEXT:    [[SCEVGEP:%.*]] = getelementptr float, float* [[A:%.*]], i64 [[TMP5]], !dbg !9
28; CHECK-NEXT:    [[SCEVGEP4:%.*]] = getelementptr float, float* [[B:%.*]], i64 [[TMP5]], !dbg !9
29; CHECK-NEXT:    [[BOUND0:%.*]] = icmp ugt float* [[SCEVGEP4]], [[A]], !dbg !9
30; CHECK-NEXT:    [[BOUND1:%.*]] = icmp ugt float* [[SCEVGEP]], [[B]], !dbg !9
31; CHECK-NEXT:    [[MEMCHECK_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]], !dbg !9
32; CHECK-NEXT:    br i1 [[MEMCHECK_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]], !dbg !9
33; CHECK:       vector.ph:
34; CHECK-NEXT:    [[N_VEC:%.*]] = and i64 [[TMP2]], 8589934588, !dbg !9
35; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]], !dbg !9
36; CHECK:       vector.body:
37; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ], !dbg !9
38; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds float, float* [[B]], i64 [[INDEX]], !dbg !9
39; CHECK-NEXT:    [[TMP8:%.*]] = bitcast float* [[TMP7]] to <4 x float>*, !dbg !9
40; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <4 x float>, <4 x float>* [[TMP8]], align 4, !dbg !9, !alias.scope !10
41; CHECK-NEXT:    [[TMP9:%.*]] = fmul <4 x float> [[WIDE_LOAD]], <float 3.000000e+00, float 3.000000e+00, float 3.000000e+00, float 3.000000e+00>, !dbg !9
42; CHECK-NEXT:    [[TMP10:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[INDEX]], !dbg !9
43; CHECK-NEXT:    [[TMP11:%.*]] = bitcast float* [[TMP10]] to <4 x float>*, !dbg !9
44; CHECK-NEXT:    store <4 x float> [[TMP9]], <4 x float>* [[TMP11]], align 4, !dbg !9, !alias.scope !13, !noalias !10
45; CHECK-NEXT:    [[INDEX_NEXT]] = add i64 [[INDEX]], 4, !dbg !9
46; CHECK-NEXT:    [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]], !dbg !9
47; CHECK-NEXT:    br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !dbg !9, !llvm.loop !15
48; CHECK:       middle.block:
49; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]]
50; CHECK-NEXT:    br i1 [[CMP_N]], label [[FOR_END_LOOPEXIT:%.*]], label [[SCALAR_PH]], !dbg !9
51; CHECK:       scalar.ph:
52; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_PREHEADER]] ], [ 0, [[VECTOR_MEMCHECK]] ]
53; CHECK-NEXT:    br label [[FOR_BODY:%.*]], !dbg !9
54; CHECK:       for.body:
55; CHECK-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], !dbg !9
56; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[B]], i64 [[INDVARS_IV]], !dbg !9
57; CHECK-NEXT:    [[TMP13:%.*]] = load float, float* [[ARRAYIDX]], align 4, !dbg !9
58; CHECK-NEXT:    [[MUL:%.*]] = fmul float [[TMP13]], 3.000000e+00, !dbg !9
59; CHECK-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[INDVARS_IV]], !dbg !9
60; CHECK-NEXT:    store float [[MUL]], float* [[ARRAYIDX2]], align 4, !dbg !9
61; CHECK-NEXT:    [[INDVARS_IV_NEXT]] = add i64 [[INDVARS_IV]], 1, !dbg !9
62; CHECK-NEXT:    [[LFTR_WIDEIV:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i32, !dbg !9
63; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i32 [[LFTR_WIDEIV]], [[N]], !dbg !9
64; CHECK-NEXT:    br i1 [[EXITCOND]], label [[FOR_END_LOOPEXIT]], label [[FOR_BODY]], !dbg !9, !llvm.loop !17
65; CHECK:       for.end.loopexit:
66; CHECK-NEXT:    br label [[FOR_END]], !dbg !18
67; CHECK:       for.end:
68; CHECK-NEXT:    ret i32 undef, !dbg !18
69;
70entry:
71  %cmp6 = icmp sgt i32 %n, 0, !dbg !6
72  br i1 %cmp6, label %for.body, label %for.end, !dbg !6
73
74for.body:                                         ; preds = %entry, %for.body
75  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ], !dbg !7
76  %arrayidx = getelementptr inbounds float, float* %b, i64 %indvars.iv, !dbg !7
77  %0 = load float, float* %arrayidx, align 4, !dbg !7
78  %mul = fmul float %0, 3.000000e+00, !dbg !7
79  %arrayidx2 = getelementptr inbounds float, float* %a, i64 %indvars.iv, !dbg !7
80  store float %mul, float* %arrayidx2, align 4, !dbg !7
81  %indvars.iv.next = add i64 %indvars.iv, 1, !dbg !7
82  %lftr.wideiv = trunc i64 %indvars.iv.next to i32, !dbg !7
83  %exitcond = icmp eq i32 %lftr.wideiv, %n, !dbg !7
84  br i1 %exitcond, label %for.end, label %for.body, !dbg !7
85
86for.end:                                          ; preds = %for.body, %entry
87  ret i32 undef, !dbg !8
88}
89
90; Make sure that we try to vectorize loops with a runtime check if the
91; dependency check fails.
92
93; CHECK-LABEL: test_runtime_check
94; CHECK:      <4 x float>
95define void @test_runtime_check(float* %a, float %b, i64 %offset, i64 %offset2, i64 %n) {
96entry:
97  br label %for.body
98
99for.body:
100  %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
101  %ind.sum = add i64 %iv, %offset
102  %arr.idx = getelementptr inbounds float, float* %a, i64 %ind.sum
103  %l1 = load float, float* %arr.idx, align 4
104  %ind.sum2 = add i64 %iv, %offset2
105  %arr.idx2 = getelementptr inbounds float, float* %a, i64 %ind.sum2
106  %l2 = load float, float* %arr.idx2, align 4
107  %m = fmul fast float %b, %l2
108  %ad = fadd fast float %l1, %m
109  store float %ad, float* %arr.idx, align 4
110  %iv.next = add nuw nsw i64 %iv, 1
111  %exitcond = icmp eq i64 %iv.next, %n
112  br i1 %exitcond, label %loopexit, label %for.body
113
114loopexit:
115  ret void
116}
117
118; Check we do not generate runtime checks if we found a known dependence preventing
119; vectorization. In this case, it is a read of c[i-1] followed by a write of c[i].
120; The runtime checks would always fail.
121
122; void test_runtime_check2(float *a, float b, unsigned offset, unsigned offset2, unsigned n, float *c) {
123;   for (unsigned i = 1; i < n; i++) {
124;     a[i+o1] += a[i+o2] + b;
125;     c[i] = c[i-1] + b;
126;   }
127; }
128;
129; CHECK-LABEL: test_runtime_check2
130; CHECK-NOT:      <4 x float>
131define void @test_runtime_check2(float* %a, float %b, i64 %offset, i64 %offset2, i64 %n, float* %c) {
132entry:
133  br label %for.body
134
135for.body:
136  %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
137  %ind.sum = add i64 %iv, %offset
138  %arr.idx = getelementptr inbounds float, float* %a, i64 %ind.sum
139  %l1 = load float, float* %arr.idx, align 4
140  %ind.sum2 = add i64 %iv, %offset2
141  %arr.idx2 = getelementptr inbounds float, float* %a, i64 %ind.sum2
142  %l2 = load float, float* %arr.idx2, align 4
143  %m = fmul fast float %b, %l2
144  %ad = fadd fast float %l1, %m
145  store float %ad, float* %arr.idx, align 4
146  %c.ind = add i64 %iv, -1
147  %c.idx = getelementptr inbounds float, float* %c, i64 %c.ind
148  %lc = load float, float* %c.idx, align 4
149  %vc = fadd float %lc, 1.0
150  %c.idx2 = getelementptr inbounds float, float* %c, i64 %iv
151  store float %vc, float* %c.idx2
152  %iv.next = add nuw nsw i64 %iv, 1
153  %exitcond = icmp eq i64 %iv.next, %n
154  br i1 %exitcond, label %loopexit, label %for.body
155
156loopexit:
157  ret void
158}
159
160; CHECK: !9 = !DILocation(line: 101, column: 1, scope: !{{.*}})
161
162!llvm.module.flags = !{!0, !1}
163!llvm.dbg.cu = !{!9}
164!0 = !{i32 2, !"Dwarf Version", i32 4}
165!1 = !{i32 2, !"Debug Info Version", i32 3}
166
167!2 = !{}
168!3 = !DISubroutineType(types: !2)
169!4 = !DIFile(filename: "test.cpp", directory: "/tmp")
170!5 = distinct !DISubprogram(name: "foo", scope: !4, file: !4, line: 99, type: !3, isLocal: false, isDefinition: true, scopeLine: 100, flags: DIFlagPrototyped, isOptimized: false, unit: !9, retainedNodes: !2)
171!6 = !DILocation(line: 100, column: 1, scope: !5)
172!7 = !DILocation(line: 101, column: 1, scope: !5)
173!8 = !DILocation(line: 102, column: 1, scope: !5)
174!9 = distinct !DICompileUnit(language: DW_LANG_C99, producer: "clang",
175                             file: !10,
176                             isOptimized: true, flags: "-O2",
177                             splitDebugFilename: "abc.debug", emissionKind: 2)
178!10 = !DIFile(filename: "path/to/file", directory: "/path/to/dir")
179!11 = !{i32 2, !"Debug Info Version", i32 3}
180