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; RUN: opt < %s -loop-vectorize -disable-basic-aa -S -pass-remarks-analysis='loop-vectorize' 2>&1 | FileCheck %s -check-prefix=FORCED_OPTSIZE
4
5target 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"
6
7; Make sure we vectorize this loop:
8; int foo(float *a, float *b, int n) {
9;   for (int i=0; i<n; ++i)
10;     a[i] = b[i] * 3;
11; }
12
13define i32 @foo(float* nocapture %a, float* nocapture %b, i32 %n) nounwind uwtable ssp {
14; CHECK-LABEL: @foo(
15; CHECK-NEXT:  entry:
16; CHECK-NEXT:    [[B2:%.*]] = ptrtoint float* [[B:%.*]] to i64, [[DBG4:!dbg !.*]]
17; CHECK-NEXT:    [[A1:%.*]] = ptrtoint float* [[A:%.*]] to i64, [[DBG4]]
18; CHECK-NEXT:    [[CMP6:%.*]] = icmp sgt i32 [[N:%.*]], 0, [[DBG4]]
19; CHECK-NEXT:    br i1 [[CMP6]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_END:%.*]], [[DBG4]]
20; CHECK:       for.body.preheader:
21; CHECK-NEXT:    [[TMP0:%.*]] = add i32 [[N]], -1, [[DBG9:!dbg !.*]]
22; CHECK-NEXT:    [[TMP1:%.*]] = zext i32 [[TMP0]] to i64, [[DBG9]]
23; CHECK-NEXT:    [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1, [[DBG9]]
24; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 3, [[DBG9]]
25; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]], [[DBG9]]
26; CHECK:       vector.memcheck:
27; CHECK-NEXT:    [[TMP3:%.*]] = sub i64 [[A1]], [[B2]], [[DBG9]]
28; CHECK-NEXT:    [[DIFF_CHECK:%.*]] = icmp ult i64 [[TMP3]], 16, [[DBG9]]
29; CHECK-NEXT:    br i1 [[DIFF_CHECK]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]], [[DBG9]]
30; CHECK:       vector.ph:
31; CHECK-NEXT:    [[N_VEC:%.*]] = and i64 [[TMP2]], 8589934588, [[DBG9]]
32; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]], [[DBG9]]
33; CHECK:       vector.body:
34; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ], [[DBG9]]
35; CHECK-NEXT:    [[TMP6:%.*]] = getelementptr inbounds float, float* [[B]], i64 [[INDEX]], [[DBG9]]
36; CHECK-NEXT:    [[TMP7:%.*]] = bitcast float* [[TMP6]] to <4 x float>*, [[DBG9]]
37; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <4 x float>, <4 x float>* [[TMP7]], align 4, [[DBG9]]
38; CHECK-NEXT:    [[TMP8:%.*]] = fmul <4 x float> [[WIDE_LOAD]], <float 3.000000e+00, float 3.000000e+00, float 3.000000e+00, float 3.000000e+00>, [[DBG9]]
39; CHECK-NEXT:    [[TMP9:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[INDEX]], [[DBG9]]
40; CHECK-NEXT:    [[TMP10:%.*]] = bitcast float* [[TMP9]] to <4 x float>*, [[DBG9]]
41; CHECK-NEXT:    store <4 x float> [[TMP8]], <4 x float>* [[TMP10]], align 4, [[DBG9]]
42; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4, [[DBG9]]
43; CHECK-NEXT:    [[TMP11:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]], [[DBG9]]
44; CHECK-NEXT:    br i1 [[TMP11]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], [[DBG9]], [[LOOP15:!llvm.loop !.*]]
45; CHECK:       middle.block:
46; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]], [[DBG9]]
47; CHECK-NEXT:    br i1 [[CMP_N]], label [[FOR_END_LOOPEXIT:%.*]], label [[SCALAR_PH]], [[DBG9]]
48; CHECK:       scalar.ph:
49; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_PREHEADER]] ], [ 0, [[VECTOR_MEMCHECK]] ], [[DBG9]]
50; CHECK-NEXT:    br label [[FOR_BODY:%.*]], [[DBG9]]
51; CHECK:       for.body:
52; CHECK-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [[DBG9]]
53; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[B]], i64 [[INDVARS_IV]], [[DBG9]]
54; CHECK-NEXT:    [[TMP12:%.*]] = load float, float* [[ARRAYIDX]], align 4, [[DBG9]]
55; CHECK-NEXT:    [[MUL:%.*]] = fmul float [[TMP12]], 3.000000e+00, [[DBG9]]
56; CHECK-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[INDVARS_IV]], [[DBG9]]
57; CHECK-NEXT:    store float [[MUL]], float* [[ARRAYIDX2]], align 4, [[DBG9]]
58; CHECK-NEXT:    [[INDVARS_IV_NEXT]] = add i64 [[INDVARS_IV]], 1, [[DBG9]]
59; CHECK-NEXT:    [[LFTR_WIDEIV:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i32, [[DBG9]]
60; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i32 [[LFTR_WIDEIV]], [[N]], [[DBG9]]
61; CHECK-NEXT:    br i1 [[EXITCOND]], label [[FOR_END_LOOPEXIT]], label [[FOR_BODY]], [[DBG9]], [[LOOP17:!llvm.loop !.*]]
62; CHECK:       for.end.loopexit:
63; CHECK-NEXT:    br label [[FOR_END]], [[DBG18:!dbg !.*]]
64; CHECK:       for.end:
65; CHECK-NEXT:    ret i32 undef, [[DBG18]]
66;
67entry:
68  %cmp6 = icmp sgt i32 %n, 0, !dbg !6
69  br i1 %cmp6, label %for.body, label %for.end, !dbg !6
70
71for.body:                                         ; preds = %entry, %for.body
72  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ], !dbg !7
73  %arrayidx = getelementptr inbounds float, float* %b, i64 %indvars.iv, !dbg !7
74  %0 = load float, float* %arrayidx, align 4, !dbg !7
75  %mul = fmul float %0, 3.000000e+00, !dbg !7
76  %arrayidx2 = getelementptr inbounds float, float* %a, i64 %indvars.iv, !dbg !7
77  store float %mul, float* %arrayidx2, align 4, !dbg !7
78  %indvars.iv.next = add i64 %indvars.iv, 1, !dbg !7
79  %lftr.wideiv = trunc i64 %indvars.iv.next to i32, !dbg !7
80  %exitcond = icmp eq i32 %lftr.wideiv, %n, !dbg !7
81  br i1 %exitcond, label %for.end, label %for.body, !dbg !7
82
83for.end:                                          ; preds = %for.body, %entry
84  ret i32 undef, !dbg !8
85}
86
87; Make sure that we try to vectorize loops with a runtime check if the
88; dependency check fails.
89
90; CHECK-LABEL: test_runtime_check
91; CHECK:      <4 x float>
92define void @test_runtime_check(float* %a, float %b, i64 %offset, i64 %offset2, i64 %n) {
93entry:
94  br label %for.body
95
96for.body:
97  %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
98  %ind.sum = add i64 %iv, %offset
99  %arr.idx = getelementptr inbounds float, float* %a, i64 %ind.sum
100  %l1 = load float, float* %arr.idx, align 4
101  %ind.sum2 = add i64 %iv, %offset2
102  %arr.idx2 = getelementptr inbounds float, float* %a, i64 %ind.sum2
103  %l2 = load float, float* %arr.idx2, align 4
104  %m = fmul fast float %b, %l2
105  %ad = fadd fast float %l1, %m
106  store float %ad, float* %arr.idx, align 4
107  %iv.next = add nuw nsw i64 %iv, 1
108  %exitcond = icmp eq i64 %iv.next, %n
109  br i1 %exitcond, label %loopexit, label %for.body
110
111loopexit:
112  ret void
113}
114
115; Check we do not generate runtime checks if we found a known dependence preventing
116; vectorization. In this case, it is a read of c[i-1] followed by a write of c[i].
117; The runtime checks would always fail.
118
119; void test_runtime_check2(float *a, float b, unsigned offset, unsigned offset2, unsigned n, float *c) {
120;   for (unsigned i = 1; i < n; i++) {
121;     a[i+o1] += a[i+o2] + b;
122;     c[i] = c[i-1] + b;
123;   }
124; }
125;
126; CHECK-LABEL: test_runtime_check2
127; CHECK-NOT:      <4 x float>
128define void @test_runtime_check2(float* %a, float %b, i64 %offset, i64 %offset2, i64 %n, float* %c) {
129entry:
130  br label %for.body
131
132for.body:
133  %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
134  %ind.sum = add i64 %iv, %offset
135  %arr.idx = getelementptr inbounds float, float* %a, i64 %ind.sum
136  %l1 = load float, float* %arr.idx, align 4
137  %ind.sum2 = add i64 %iv, %offset2
138  %arr.idx2 = getelementptr inbounds float, float* %a, i64 %ind.sum2
139  %l2 = load float, float* %arr.idx2, align 4
140  %m = fmul fast float %b, %l2
141  %ad = fadd fast float %l1, %m
142  store float %ad, float* %arr.idx, align 4
143  %c.ind = add i64 %iv, -1
144  %c.idx = getelementptr inbounds float, float* %c, i64 %c.ind
145  %lc = load float, float* %c.idx, align 4
146  %vc = fadd float %lc, 1.0
147  %c.idx2 = getelementptr inbounds float, float* %c, i64 %iv
148  store float %vc, float* %c.idx2
149  %iv.next = add nuw nsw i64 %iv, 1
150  %exitcond = icmp eq i64 %iv.next, %n
151  br i1 %exitcond, label %loopexit, label %for.body
152
153loopexit:
154  ret void
155}
156
157; CHECK: !9 = !DILocation(line: 101, column: 1, scope: !{{.*}})
158
159define dso_local void @forced_optsize(i64* noalias nocapture readonly %x_p, i64* noalias nocapture readonly %y_p, i64* noalias nocapture %z_p) minsize optsize {
160;
161; FORCED_OPTSIZE: remark: <unknown>:0:0: loop not vectorized: runtime pointer checks needed. Enable vectorization of this loop with '#pragma clang loop vectorize(enable)' when compiling with -Os/-Oz
162; FORCED_OPTSIZE-LABEL: @forced_optsize(
163; FORCED_OPTSIZE-NOT:       vector.body:
164;
165entry:
166  br label %for.body
167
168for.cond.cleanup:
169  ret void
170
171for.body:
172  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
173  %arrayidx = getelementptr inbounds i64, i64* %x_p, i64 %indvars.iv
174  %0 = load i64, i64* %arrayidx, align 8
175  %arrayidx2 = getelementptr inbounds i64, i64* %y_p, i64 %indvars.iv
176  %1 = load i64, i64* %arrayidx2, align 8
177  %add = add nsw i64 %1, %0
178  %arrayidx4 = getelementptr inbounds i64, i64* %z_p, i64 %indvars.iv
179  store i64 %add, i64* %arrayidx4, align 8
180  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
181  %exitcond = icmp eq i64 %indvars.iv.next, 128
182  br i1 %exitcond, label %for.cond.cleanup, label %for.body, !llvm.loop !12
183}
184
185!llvm.module.flags = !{!0, !1}
186!llvm.dbg.cu = !{!9}
187!0 = !{i32 2, !"Dwarf Version", i32 4}
188!1 = !{i32 2, !"Debug Info Version", i32 3}
189
190!2 = !{}
191!3 = !DISubroutineType(types: !2)
192!4 = !DIFile(filename: "test.cpp", directory: "/tmp")
193!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)
194!6 = !DILocation(line: 100, column: 1, scope: !5)
195!7 = !DILocation(line: 101, column: 1, scope: !5)
196!8 = !DILocation(line: 102, column: 1, scope: !5)
197!9 = distinct !DICompileUnit(language: DW_LANG_C99, producer: "clang",
198                             file: !10,
199                             isOptimized: true, flags: "-O2",
200                             splitDebugFilename: "abc.debug", emissionKind: 2)
201!10 = !DIFile(filename: "path/to/file", directory: "/path/to/dir")
202!11 = !{i32 2, !"Debug Info Version", i32 3}
203!12 = distinct !{!12, !13, !14}
204!13 = !{!"llvm.loop.vectorize.width", i32 2}
205!14 = !{!"llvm.loop.vectorize.enable", i1 true}
206