1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s  -O3 -simplifycfg -keep-loops=false -mcpu=core-avx2 -mtriple=x86_64-unknown-linux-gnu -S | FileCheck --check-prefix AUTO_VEC %s
3
4target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
5
6; This test checks auto-vectorization with FP induction variable.
7; The FP operation is not "fast" and requires "fast-math" function attribute.
8
9;void fp_iv_loop1(float * __restrict__ A, int N) {
10;  float x = 1.0;
11;  for (int i=0; i < N; ++i) {
12;    A[i] = x;
13;    x += 0.5;
14;  }
15;}
16
17define void @fp_iv_loop1(float* noalias nocapture %A, i32 %N) #0 {
18; AUTO_VEC-LABEL: @fp_iv_loop1(
19; AUTO_VEC-NEXT:  entry:
20; AUTO_VEC-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[N:%.*]], 0
21; AUTO_VEC-NEXT:    br i1 [[CMP4]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_END:%.*]]
22; AUTO_VEC:       for.body.preheader:
23; AUTO_VEC-NEXT:    [[ZEXT:%.*]] = zext i32 [[N]] to i64
24; AUTO_VEC-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[N]], 32
25; AUTO_VEC-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[FOR_BODY:%.*]], label [[VECTOR_PH:%.*]]
26; AUTO_VEC:       vector.ph:
27; AUTO_VEC-NEXT:    [[N_VEC:%.*]] = and i64 [[ZEXT]], 4294967264
28; AUTO_VEC-NEXT:    [[CAST_CRD:%.*]] = sitofp i64 [[N_VEC]] to float
29; AUTO_VEC-NEXT:    [[TMP0:%.*]] = fmul fast float [[CAST_CRD]], 5.000000e-01
30; AUTO_VEC-NEXT:    [[IND_END:%.*]] = fadd fast float [[TMP0]], 1.000000e+00
31; AUTO_VEC-NEXT:    [[TMP1:%.*]] = add nsw i64 [[N_VEC]], -32
32; AUTO_VEC-NEXT:    [[TMP2:%.*]] = lshr exact i64 [[TMP1]], 5
33; AUTO_VEC-NEXT:    [[TMP3:%.*]] = add nuw nsw i64 [[TMP2]], 1
34; AUTO_VEC-NEXT:    [[XTRAITER:%.*]] = and i64 [[TMP3]], 3
35; AUTO_VEC-NEXT:    [[TMP4:%.*]] = icmp ult i64 [[TMP1]], 96
36; AUTO_VEC-NEXT:    br i1 [[TMP4]], label [[MIDDLE_BLOCK_UNR_LCSSA:%.*]], label [[VECTOR_PH_NEW:%.*]]
37; AUTO_VEC:       vector.ph.new:
38; AUTO_VEC-NEXT:    [[UNROLL_ITER:%.*]] = and i64 [[TMP3]], 1152921504606846972
39; AUTO_VEC-NEXT:    br label [[VECTOR_BODY:%.*]]
40; AUTO_VEC:       vector.body:
41; AUTO_VEC-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH_NEW]] ], [ [[INDEX_NEXT_3:%.*]], [[VECTOR_BODY]] ]
42; AUTO_VEC-NEXT:    [[VEC_IND:%.*]] = phi <8 x float> [ <float 1.000000e+00, float 1.500000e+00, float 2.000000e+00, float 2.500000e+00, float 3.000000e+00, float 3.500000e+00, float 4.000000e+00, float 4.500000e+00>, [[VECTOR_PH_NEW]] ], [ [[VEC_IND_NEXT_3:%.*]], [[VECTOR_BODY]] ]
43; AUTO_VEC-NEXT:    [[NITER:%.*]] = phi i64 [ [[UNROLL_ITER]], [[VECTOR_PH_NEW]] ], [ [[NITER_NSUB_3:%.*]], [[VECTOR_BODY]] ]
44; AUTO_VEC-NEXT:    [[STEP_ADD:%.*]] = fadd fast <8 x float> [[VEC_IND]], <float 4.000000e+00, float 4.000000e+00, float 4.000000e+00, float 4.000000e+00, float 4.000000e+00, float 4.000000e+00, float 4.000000e+00, float 4.000000e+00>
45; AUTO_VEC-NEXT:    [[STEP_ADD2:%.*]] = fadd fast <8 x float> [[VEC_IND]], <float 8.000000e+00, float 8.000000e+00, float 8.000000e+00, float 8.000000e+00, float 8.000000e+00, float 8.000000e+00, float 8.000000e+00, float 8.000000e+00>
46; AUTO_VEC-NEXT:    [[STEP_ADD3:%.*]] = fadd fast <8 x float> [[VEC_IND]], <float 1.200000e+01, float 1.200000e+01, float 1.200000e+01, float 1.200000e+01, float 1.200000e+01, float 1.200000e+01, float 1.200000e+01, float 1.200000e+01>
47; AUTO_VEC-NEXT:    [[TMP5:%.*]] = getelementptr inbounds float, float* [[A:%.*]], i64 [[INDEX]]
48; AUTO_VEC-NEXT:    [[TMP6:%.*]] = bitcast float* [[TMP5]] to <8 x float>*
49; AUTO_VEC-NEXT:    store <8 x float> [[VEC_IND]], <8 x float>* [[TMP6]], align 4
50; AUTO_VEC-NEXT:    [[TMP7:%.*]] = getelementptr inbounds float, float* [[TMP5]], i64 8
51; AUTO_VEC-NEXT:    [[TMP8:%.*]] = bitcast float* [[TMP7]] to <8 x float>*
52; AUTO_VEC-NEXT:    store <8 x float> [[STEP_ADD]], <8 x float>* [[TMP8]], align 4
53; AUTO_VEC-NEXT:    [[TMP9:%.*]] = getelementptr inbounds float, float* [[TMP5]], i64 16
54; AUTO_VEC-NEXT:    [[TMP10:%.*]] = bitcast float* [[TMP9]] to <8 x float>*
55; AUTO_VEC-NEXT:    store <8 x float> [[STEP_ADD2]], <8 x float>* [[TMP10]], align 4
56; AUTO_VEC-NEXT:    [[TMP11:%.*]] = getelementptr inbounds float, float* [[TMP5]], i64 24
57; AUTO_VEC-NEXT:    [[TMP12:%.*]] = bitcast float* [[TMP11]] to <8 x float>*
58; AUTO_VEC-NEXT:    store <8 x float> [[STEP_ADD3]], <8 x float>* [[TMP12]], align 4
59; AUTO_VEC-NEXT:    [[INDEX_NEXT:%.*]] = or i64 [[INDEX]], 32
60; AUTO_VEC-NEXT:    [[VEC_IND_NEXT:%.*]] = fadd fast <8 x float> [[VEC_IND]], <float 1.600000e+01, float 1.600000e+01, float 1.600000e+01, float 1.600000e+01, float 1.600000e+01, float 1.600000e+01, float 1.600000e+01, float 1.600000e+01>
61; AUTO_VEC-NEXT:    [[STEP_ADD_1:%.*]] = fadd fast <8 x float> [[VEC_IND]], <float 2.000000e+01, float 2.000000e+01, float 2.000000e+01, float 2.000000e+01, float 2.000000e+01, float 2.000000e+01, float 2.000000e+01, float 2.000000e+01>
62; AUTO_VEC-NEXT:    [[STEP_ADD2_1:%.*]] = fadd fast <8 x float> [[VEC_IND]], <float 2.400000e+01, float 2.400000e+01, float 2.400000e+01, float 2.400000e+01, float 2.400000e+01, float 2.400000e+01, float 2.400000e+01, float 2.400000e+01>
63; AUTO_VEC-NEXT:    [[STEP_ADD3_1:%.*]] = fadd fast <8 x float> [[VEC_IND]], <float 2.800000e+01, float 2.800000e+01, float 2.800000e+01, float 2.800000e+01, float 2.800000e+01, float 2.800000e+01, float 2.800000e+01, float 2.800000e+01>
64; AUTO_VEC-NEXT:    [[TMP13:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[INDEX_NEXT]]
65; AUTO_VEC-NEXT:    [[TMP14:%.*]] = bitcast float* [[TMP13]] to <8 x float>*
66; AUTO_VEC-NEXT:    store <8 x float> [[VEC_IND_NEXT]], <8 x float>* [[TMP14]], align 4
67; AUTO_VEC-NEXT:    [[TMP15:%.*]] = getelementptr inbounds float, float* [[TMP13]], i64 8
68; AUTO_VEC-NEXT:    [[TMP16:%.*]] = bitcast float* [[TMP15]] to <8 x float>*
69; AUTO_VEC-NEXT:    store <8 x float> [[STEP_ADD_1]], <8 x float>* [[TMP16]], align 4
70; AUTO_VEC-NEXT:    [[TMP17:%.*]] = getelementptr inbounds float, float* [[TMP13]], i64 16
71; AUTO_VEC-NEXT:    [[TMP18:%.*]] = bitcast float* [[TMP17]] to <8 x float>*
72; AUTO_VEC-NEXT:    store <8 x float> [[STEP_ADD2_1]], <8 x float>* [[TMP18]], align 4
73; AUTO_VEC-NEXT:    [[TMP19:%.*]] = getelementptr inbounds float, float* [[TMP13]], i64 24
74; AUTO_VEC-NEXT:    [[TMP20:%.*]] = bitcast float* [[TMP19]] to <8 x float>*
75; AUTO_VEC-NEXT:    store <8 x float> [[STEP_ADD3_1]], <8 x float>* [[TMP20]], align 4
76; AUTO_VEC-NEXT:    [[INDEX_NEXT_1:%.*]] = or i64 [[INDEX]], 64
77; AUTO_VEC-NEXT:    [[VEC_IND_NEXT_1:%.*]] = fadd fast <8 x float> [[VEC_IND]], <float 3.200000e+01, float 3.200000e+01, float 3.200000e+01, float 3.200000e+01, float 3.200000e+01, float 3.200000e+01, float 3.200000e+01, float 3.200000e+01>
78; AUTO_VEC-NEXT:    [[STEP_ADD_2:%.*]] = fadd fast <8 x float> [[VEC_IND]], <float 3.600000e+01, float 3.600000e+01, float 3.600000e+01, float 3.600000e+01, float 3.600000e+01, float 3.600000e+01, float 3.600000e+01, float 3.600000e+01>
79; AUTO_VEC-NEXT:    [[STEP_ADD2_2:%.*]] = fadd fast <8 x float> [[VEC_IND]], <float 4.000000e+01, float 4.000000e+01, float 4.000000e+01, float 4.000000e+01, float 4.000000e+01, float 4.000000e+01, float 4.000000e+01, float 4.000000e+01>
80; AUTO_VEC-NEXT:    [[STEP_ADD3_2:%.*]] = fadd fast <8 x float> [[VEC_IND]], <float 4.400000e+01, float 4.400000e+01, float 4.400000e+01, float 4.400000e+01, float 4.400000e+01, float 4.400000e+01, float 4.400000e+01, float 4.400000e+01>
81; AUTO_VEC-NEXT:    [[TMP21:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[INDEX_NEXT_1]]
82; AUTO_VEC-NEXT:    [[TMP22:%.*]] = bitcast float* [[TMP21]] to <8 x float>*
83; AUTO_VEC-NEXT:    store <8 x float> [[VEC_IND_NEXT_1]], <8 x float>* [[TMP22]], align 4
84; AUTO_VEC-NEXT:    [[TMP23:%.*]] = getelementptr inbounds float, float* [[TMP21]], i64 8
85; AUTO_VEC-NEXT:    [[TMP24:%.*]] = bitcast float* [[TMP23]] to <8 x float>*
86; AUTO_VEC-NEXT:    store <8 x float> [[STEP_ADD_2]], <8 x float>* [[TMP24]], align 4
87; AUTO_VEC-NEXT:    [[TMP25:%.*]] = getelementptr inbounds float, float* [[TMP21]], i64 16
88; AUTO_VEC-NEXT:    [[TMP26:%.*]] = bitcast float* [[TMP25]] to <8 x float>*
89; AUTO_VEC-NEXT:    store <8 x float> [[STEP_ADD2_2]], <8 x float>* [[TMP26]], align 4
90; AUTO_VEC-NEXT:    [[TMP27:%.*]] = getelementptr inbounds float, float* [[TMP21]], i64 24
91; AUTO_VEC-NEXT:    [[TMP28:%.*]] = bitcast float* [[TMP27]] to <8 x float>*
92; AUTO_VEC-NEXT:    store <8 x float> [[STEP_ADD3_2]], <8 x float>* [[TMP28]], align 4
93; AUTO_VEC-NEXT:    [[INDEX_NEXT_2:%.*]] = or i64 [[INDEX]], 96
94; AUTO_VEC-NEXT:    [[VEC_IND_NEXT_2:%.*]] = fadd fast <8 x float> [[VEC_IND]], <float 4.800000e+01, float 4.800000e+01, float 4.800000e+01, float 4.800000e+01, float 4.800000e+01, float 4.800000e+01, float 4.800000e+01, float 4.800000e+01>
95; AUTO_VEC-NEXT:    [[STEP_ADD_3:%.*]] = fadd fast <8 x float> [[VEC_IND]], <float 5.200000e+01, float 5.200000e+01, float 5.200000e+01, float 5.200000e+01, float 5.200000e+01, float 5.200000e+01, float 5.200000e+01, float 5.200000e+01>
96; AUTO_VEC-NEXT:    [[STEP_ADD2_3:%.*]] = fadd fast <8 x float> [[VEC_IND]], <float 5.600000e+01, float 5.600000e+01, float 5.600000e+01, float 5.600000e+01, float 5.600000e+01, float 5.600000e+01, float 5.600000e+01, float 5.600000e+01>
97; AUTO_VEC-NEXT:    [[STEP_ADD3_3:%.*]] = fadd fast <8 x float> [[VEC_IND]], <float 6.000000e+01, float 6.000000e+01, float 6.000000e+01, float 6.000000e+01, float 6.000000e+01, float 6.000000e+01, float 6.000000e+01, float 6.000000e+01>
98; AUTO_VEC-NEXT:    [[TMP29:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[INDEX_NEXT_2]]
99; AUTO_VEC-NEXT:    [[TMP30:%.*]] = bitcast float* [[TMP29]] to <8 x float>*
100; AUTO_VEC-NEXT:    store <8 x float> [[VEC_IND_NEXT_2]], <8 x float>* [[TMP30]], align 4
101; AUTO_VEC-NEXT:    [[TMP31:%.*]] = getelementptr inbounds float, float* [[TMP29]], i64 8
102; AUTO_VEC-NEXT:    [[TMP32:%.*]] = bitcast float* [[TMP31]] to <8 x float>*
103; AUTO_VEC-NEXT:    store <8 x float> [[STEP_ADD_3]], <8 x float>* [[TMP32]], align 4
104; AUTO_VEC-NEXT:    [[TMP33:%.*]] = getelementptr inbounds float, float* [[TMP29]], i64 16
105; AUTO_VEC-NEXT:    [[TMP34:%.*]] = bitcast float* [[TMP33]] to <8 x float>*
106; AUTO_VEC-NEXT:    store <8 x float> [[STEP_ADD2_3]], <8 x float>* [[TMP34]], align 4
107; AUTO_VEC-NEXT:    [[TMP35:%.*]] = getelementptr inbounds float, float* [[TMP29]], i64 24
108; AUTO_VEC-NEXT:    [[TMP36:%.*]] = bitcast float* [[TMP35]] to <8 x float>*
109; AUTO_VEC-NEXT:    store <8 x float> [[STEP_ADD3_3]], <8 x float>* [[TMP36]], align 4
110; AUTO_VEC-NEXT:    [[INDEX_NEXT_3]] = add i64 [[INDEX]], 128
111; AUTO_VEC-NEXT:    [[VEC_IND_NEXT_3]] = fadd fast <8 x float> [[VEC_IND]], <float 6.400000e+01, float 6.400000e+01, float 6.400000e+01, float 6.400000e+01, float 6.400000e+01, float 6.400000e+01, float 6.400000e+01, float 6.400000e+01>
112; AUTO_VEC-NEXT:    [[NITER_NSUB_3]] = add i64 [[NITER]], -4
113; AUTO_VEC-NEXT:    [[NITER_NCMP_3:%.*]] = icmp eq i64 [[NITER_NSUB_3]], 0
114; AUTO_VEC-NEXT:    br i1 [[NITER_NCMP_3]], label [[MIDDLE_BLOCK_UNR_LCSSA]], label [[VECTOR_BODY]], [[LOOP0:!llvm.loop !.*]]
115; AUTO_VEC:       middle.block.unr-lcssa:
116; AUTO_VEC-NEXT:    [[INDEX_UNR:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT_3]], [[VECTOR_BODY]] ]
117; AUTO_VEC-NEXT:    [[VEC_IND_UNR:%.*]] = phi <8 x float> [ <float 1.000000e+00, float 1.500000e+00, float 2.000000e+00, float 2.500000e+00, float 3.000000e+00, float 3.500000e+00, float 4.000000e+00, float 4.500000e+00>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT_3]], [[VECTOR_BODY]] ]
118; AUTO_VEC-NEXT:    [[LCMP_MOD_NOT:%.*]] = icmp eq i64 [[XTRAITER]], 0
119; AUTO_VEC-NEXT:    br i1 [[LCMP_MOD_NOT]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY_EPIL:%.*]]
120; AUTO_VEC:       vector.body.epil:
121; AUTO_VEC-NEXT:    [[INDEX_EPIL:%.*]] = phi i64 [ [[INDEX_NEXT_EPIL:%.*]], [[VECTOR_BODY_EPIL]] ], [ [[INDEX_UNR]], [[MIDDLE_BLOCK_UNR_LCSSA]] ]
122; AUTO_VEC-NEXT:    [[VEC_IND_EPIL:%.*]] = phi <8 x float> [ [[VEC_IND_NEXT_EPIL:%.*]], [[VECTOR_BODY_EPIL]] ], [ [[VEC_IND_UNR]], [[MIDDLE_BLOCK_UNR_LCSSA]] ]
123; AUTO_VEC-NEXT:    [[EPIL_ITER:%.*]] = phi i64 [ [[EPIL_ITER_SUB:%.*]], [[VECTOR_BODY_EPIL]] ], [ [[XTRAITER]], [[MIDDLE_BLOCK_UNR_LCSSA]] ]
124; AUTO_VEC-NEXT:    [[STEP_ADD_EPIL:%.*]] = fadd fast <8 x float> [[VEC_IND_EPIL]], <float 4.000000e+00, float 4.000000e+00, float 4.000000e+00, float 4.000000e+00, float 4.000000e+00, float 4.000000e+00, float 4.000000e+00, float 4.000000e+00>
125; AUTO_VEC-NEXT:    [[STEP_ADD2_EPIL:%.*]] = fadd fast <8 x float> [[VEC_IND_EPIL]], <float 8.000000e+00, float 8.000000e+00, float 8.000000e+00, float 8.000000e+00, float 8.000000e+00, float 8.000000e+00, float 8.000000e+00, float 8.000000e+00>
126; AUTO_VEC-NEXT:    [[STEP_ADD3_EPIL:%.*]] = fadd fast <8 x float> [[VEC_IND_EPIL]], <float 1.200000e+01, float 1.200000e+01, float 1.200000e+01, float 1.200000e+01, float 1.200000e+01, float 1.200000e+01, float 1.200000e+01, float 1.200000e+01>
127; AUTO_VEC-NEXT:    [[TMP37:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[INDEX_EPIL]]
128; AUTO_VEC-NEXT:    [[TMP38:%.*]] = bitcast float* [[TMP37]] to <8 x float>*
129; AUTO_VEC-NEXT:    store <8 x float> [[VEC_IND_EPIL]], <8 x float>* [[TMP38]], align 4
130; AUTO_VEC-NEXT:    [[TMP39:%.*]] = getelementptr inbounds float, float* [[TMP37]], i64 8
131; AUTO_VEC-NEXT:    [[TMP40:%.*]] = bitcast float* [[TMP39]] to <8 x float>*
132; AUTO_VEC-NEXT:    store <8 x float> [[STEP_ADD_EPIL]], <8 x float>* [[TMP40]], align 4
133; AUTO_VEC-NEXT:    [[TMP41:%.*]] = getelementptr inbounds float, float* [[TMP37]], i64 16
134; AUTO_VEC-NEXT:    [[TMP42:%.*]] = bitcast float* [[TMP41]] to <8 x float>*
135; AUTO_VEC-NEXT:    store <8 x float> [[STEP_ADD2_EPIL]], <8 x float>* [[TMP42]], align 4
136; AUTO_VEC-NEXT:    [[TMP43:%.*]] = getelementptr inbounds float, float* [[TMP37]], i64 24
137; AUTO_VEC-NEXT:    [[TMP44:%.*]] = bitcast float* [[TMP43]] to <8 x float>*
138; AUTO_VEC-NEXT:    store <8 x float> [[STEP_ADD3_EPIL]], <8 x float>* [[TMP44]], align 4
139; AUTO_VEC-NEXT:    [[INDEX_NEXT_EPIL]] = add i64 [[INDEX_EPIL]], 32
140; AUTO_VEC-NEXT:    [[VEC_IND_NEXT_EPIL]] = fadd fast <8 x float> [[VEC_IND_EPIL]], <float 1.600000e+01, float 1.600000e+01, float 1.600000e+01, float 1.600000e+01, float 1.600000e+01, float 1.600000e+01, float 1.600000e+01, float 1.600000e+01>
141; AUTO_VEC-NEXT:    [[EPIL_ITER_SUB]] = add i64 [[EPIL_ITER]], -1
142; AUTO_VEC-NEXT:    [[EPIL_ITER_CMP_NOT:%.*]] = icmp eq i64 [[EPIL_ITER_SUB]], 0
143; AUTO_VEC-NEXT:    br i1 [[EPIL_ITER_CMP_NOT]], label [[MIDDLE_BLOCK]], label [[VECTOR_BODY_EPIL]], [[LOOP2:!llvm.loop !.*]]
144; AUTO_VEC:       middle.block:
145; AUTO_VEC-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N_VEC]], [[ZEXT]]
146; AUTO_VEC-NEXT:    br i1 [[CMP_N]], label [[FOR_END]], label [[FOR_BODY]]
147; AUTO_VEC:       for.body:
148; AUTO_VEC-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ 0, [[FOR_BODY_PREHEADER]] ], [ [[N_VEC]], [[MIDDLE_BLOCK]] ]
149; AUTO_VEC-NEXT:    [[X_06:%.*]] = phi float [ [[CONV1:%.*]], [[FOR_BODY]] ], [ 1.000000e+00, [[FOR_BODY_PREHEADER]] ], [ [[IND_END]], [[MIDDLE_BLOCK]] ]
150; AUTO_VEC-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[INDVARS_IV]]
151; AUTO_VEC-NEXT:    store float [[X_06]], float* [[ARRAYIDX]], align 4
152; AUTO_VEC-NEXT:    [[CONV1]] = fadd float [[X_06]], 5.000000e-01
153; AUTO_VEC-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
154; AUTO_VEC-NEXT:    [[TMP45:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], [[ZEXT]]
155; AUTO_VEC-NEXT:    br i1 [[TMP45]], label [[FOR_END]], label [[FOR_BODY]], [[LOOP4:!llvm.loop !.*]]
156; AUTO_VEC:       for.end:
157; AUTO_VEC-NEXT:    ret void
158;
159entry:
160  %cmp4 = icmp sgt i32 %N, 0
161  br i1 %cmp4, label %for.body.preheader, label %for.end
162
163for.body.preheader:                               ; preds = %entry
164  br label %for.body
165
166for.body:                                         ; preds = %for.body.preheader, %for.body
167  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %for.body.preheader ]
168  %x.06 = phi float [ %conv1, %for.body ], [ 1.000000e+00, %for.body.preheader ]
169  %arrayidx = getelementptr inbounds float, float* %A, i64 %indvars.iv
170  store float %x.06, float* %arrayidx, align 4
171  %conv1 = fadd float %x.06, 5.000000e-01
172  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
173  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
174  %exitcond = icmp eq i32 %lftr.wideiv, %N
175  br i1 %exitcond, label %for.end.loopexit, label %for.body
176
177for.end.loopexit:                                 ; preds = %for.body
178  br label %for.end
179
180for.end:                                          ; preds = %for.end.loopexit, %entry
181  ret void
182}
183
184; The same as the previous, FP operation is not fast, different function attribute
185; Vectorization should be rejected.
186;void fp_iv_loop2(float * __restrict__ A, int N) {
187;  float x = 1.0;
188;  for (int i=0; i < N; ++i) {
189;    A[i] = x;
190;    x += 0.5;
191;  }
192;}
193
194define void @fp_iv_loop2(float* noalias nocapture %A, i32 %N) #1 {
195; AUTO_VEC-LABEL: @fp_iv_loop2(
196; AUTO_VEC-NEXT:  entry:
197; AUTO_VEC-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[N:%.*]], 0
198; AUTO_VEC-NEXT:    br i1 [[CMP4]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_END:%.*]]
199; AUTO_VEC:       for.body.preheader:
200; AUTO_VEC-NEXT:    [[ZEXT:%.*]] = zext i32 [[N]] to i64
201; AUTO_VEC-NEXT:    [[TMP0:%.*]] = add nsw i64 [[ZEXT]], -1
202; AUTO_VEC-NEXT:    [[XTRAITER:%.*]] = and i64 [[ZEXT]], 7
203; AUTO_VEC-NEXT:    [[TMP1:%.*]] = icmp ult i64 [[TMP0]], 7
204; AUTO_VEC-NEXT:    br i1 [[TMP1]], label [[FOR_END_LOOPEXIT_UNR_LCSSA:%.*]], label [[FOR_BODY_PREHEADER_NEW:%.*]]
205; AUTO_VEC:       for.body.preheader.new:
206; AUTO_VEC-NEXT:    [[UNROLL_ITER:%.*]] = and i64 [[ZEXT]], 4294967288
207; AUTO_VEC-NEXT:    br label [[FOR_BODY:%.*]]
208; AUTO_VEC:       for.body:
209; AUTO_VEC-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ 0, [[FOR_BODY_PREHEADER_NEW]] ], [ [[INDVARS_IV_NEXT_7:%.*]], [[FOR_BODY]] ]
210; AUTO_VEC-NEXT:    [[X_06:%.*]] = phi float [ 1.000000e+00, [[FOR_BODY_PREHEADER_NEW]] ], [ [[CONV1_7:%.*]], [[FOR_BODY]] ]
211; AUTO_VEC-NEXT:    [[NITER:%.*]] = phi i64 [ [[UNROLL_ITER]], [[FOR_BODY_PREHEADER_NEW]] ], [ [[NITER_NSUB_7:%.*]], [[FOR_BODY]] ]
212; AUTO_VEC-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[A:%.*]], i64 [[INDVARS_IV]]
213; AUTO_VEC-NEXT:    store float [[X_06]], float* [[ARRAYIDX]], align 4
214; AUTO_VEC-NEXT:    [[CONV1:%.*]] = fadd float [[X_06]], 5.000000e-01
215; AUTO_VEC-NEXT:    [[INDVARS_IV_NEXT:%.*]] = or i64 [[INDVARS_IV]], 1
216; AUTO_VEC-NEXT:    [[ARRAYIDX_1:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[INDVARS_IV_NEXT]]
217; AUTO_VEC-NEXT:    store float [[CONV1]], float* [[ARRAYIDX_1]], align 4
218; AUTO_VEC-NEXT:    [[CONV1_1:%.*]] = fadd float [[CONV1]], 5.000000e-01
219; AUTO_VEC-NEXT:    [[INDVARS_IV_NEXT_1:%.*]] = or i64 [[INDVARS_IV]], 2
220; AUTO_VEC-NEXT:    [[ARRAYIDX_2:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[INDVARS_IV_NEXT_1]]
221; AUTO_VEC-NEXT:    store float [[CONV1_1]], float* [[ARRAYIDX_2]], align 4
222; AUTO_VEC-NEXT:    [[CONV1_2:%.*]] = fadd float [[CONV1_1]], 5.000000e-01
223; AUTO_VEC-NEXT:    [[INDVARS_IV_NEXT_2:%.*]] = or i64 [[INDVARS_IV]], 3
224; AUTO_VEC-NEXT:    [[ARRAYIDX_3:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[INDVARS_IV_NEXT_2]]
225; AUTO_VEC-NEXT:    store float [[CONV1_2]], float* [[ARRAYIDX_3]], align 4
226; AUTO_VEC-NEXT:    [[CONV1_3:%.*]] = fadd float [[CONV1_2]], 5.000000e-01
227; AUTO_VEC-NEXT:    [[INDVARS_IV_NEXT_3:%.*]] = or i64 [[INDVARS_IV]], 4
228; AUTO_VEC-NEXT:    [[ARRAYIDX_4:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[INDVARS_IV_NEXT_3]]
229; AUTO_VEC-NEXT:    store float [[CONV1_3]], float* [[ARRAYIDX_4]], align 4
230; AUTO_VEC-NEXT:    [[CONV1_4:%.*]] = fadd float [[CONV1_3]], 5.000000e-01
231; AUTO_VEC-NEXT:    [[INDVARS_IV_NEXT_4:%.*]] = or i64 [[INDVARS_IV]], 5
232; AUTO_VEC-NEXT:    [[ARRAYIDX_5:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[INDVARS_IV_NEXT_4]]
233; AUTO_VEC-NEXT:    store float [[CONV1_4]], float* [[ARRAYIDX_5]], align 4
234; AUTO_VEC-NEXT:    [[CONV1_5:%.*]] = fadd float [[CONV1_4]], 5.000000e-01
235; AUTO_VEC-NEXT:    [[INDVARS_IV_NEXT_5:%.*]] = or i64 [[INDVARS_IV]], 6
236; AUTO_VEC-NEXT:    [[ARRAYIDX_6:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[INDVARS_IV_NEXT_5]]
237; AUTO_VEC-NEXT:    store float [[CONV1_5]], float* [[ARRAYIDX_6]], align 4
238; AUTO_VEC-NEXT:    [[CONV1_6:%.*]] = fadd float [[CONV1_5]], 5.000000e-01
239; AUTO_VEC-NEXT:    [[INDVARS_IV_NEXT_6:%.*]] = or i64 [[INDVARS_IV]], 7
240; AUTO_VEC-NEXT:    [[ARRAYIDX_7:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[INDVARS_IV_NEXT_6]]
241; AUTO_VEC-NEXT:    store float [[CONV1_6]], float* [[ARRAYIDX_7]], align 4
242; AUTO_VEC-NEXT:    [[CONV1_7]] = fadd float [[CONV1_6]], 5.000000e-01
243; AUTO_VEC-NEXT:    [[INDVARS_IV_NEXT_7]] = add nuw nsw i64 [[INDVARS_IV]], 8
244; AUTO_VEC-NEXT:    [[NITER_NSUB_7]] = add i64 [[NITER]], -8
245; AUTO_VEC-NEXT:    [[NITER_NCMP_7:%.*]] = icmp eq i64 [[NITER_NSUB_7]], 0
246; AUTO_VEC-NEXT:    br i1 [[NITER_NCMP_7]], label [[FOR_END_LOOPEXIT_UNR_LCSSA]], label [[FOR_BODY]]
247; AUTO_VEC:       for.end.loopexit.unr-lcssa:
248; AUTO_VEC-NEXT:    [[INDVARS_IV_UNR:%.*]] = phi i64 [ 0, [[FOR_BODY_PREHEADER]] ], [ [[INDVARS_IV_NEXT_7]], [[FOR_BODY]] ]
249; AUTO_VEC-NEXT:    [[X_06_UNR:%.*]] = phi float [ 1.000000e+00, [[FOR_BODY_PREHEADER]] ], [ [[CONV1_7]], [[FOR_BODY]] ]
250; AUTO_VEC-NEXT:    [[LCMP_MOD_NOT:%.*]] = icmp eq i64 [[XTRAITER]], 0
251; AUTO_VEC-NEXT:    br i1 [[LCMP_MOD_NOT]], label [[FOR_END]], label [[FOR_BODY_EPIL:%.*]]
252; AUTO_VEC:       for.body.epil:
253; AUTO_VEC-NEXT:    [[INDVARS_IV_EPIL:%.*]] = phi i64 [ [[INDVARS_IV_NEXT_EPIL:%.*]], [[FOR_BODY_EPIL]] ], [ [[INDVARS_IV_UNR]], [[FOR_END_LOOPEXIT_UNR_LCSSA]] ]
254; AUTO_VEC-NEXT:    [[X_06_EPIL:%.*]] = phi float [ [[CONV1_EPIL:%.*]], [[FOR_BODY_EPIL]] ], [ [[X_06_UNR]], [[FOR_END_LOOPEXIT_UNR_LCSSA]] ]
255; AUTO_VEC-NEXT:    [[EPIL_ITER:%.*]] = phi i64 [ [[EPIL_ITER_SUB:%.*]], [[FOR_BODY_EPIL]] ], [ [[XTRAITER]], [[FOR_END_LOOPEXIT_UNR_LCSSA]] ]
256; AUTO_VEC-NEXT:    [[ARRAYIDX_EPIL:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[INDVARS_IV_EPIL]]
257; AUTO_VEC-NEXT:    store float [[X_06_EPIL]], float* [[ARRAYIDX_EPIL]], align 4
258; AUTO_VEC-NEXT:    [[CONV1_EPIL]] = fadd float [[X_06_EPIL]], 5.000000e-01
259; AUTO_VEC-NEXT:    [[INDVARS_IV_NEXT_EPIL]] = add nuw nsw i64 [[INDVARS_IV_EPIL]], 1
260; AUTO_VEC-NEXT:    [[EPIL_ITER_SUB]] = add i64 [[EPIL_ITER]], -1
261; AUTO_VEC-NEXT:    [[EPIL_ITER_CMP_NOT:%.*]] = icmp eq i64 [[EPIL_ITER_SUB]], 0
262; AUTO_VEC-NEXT:    br i1 [[EPIL_ITER_CMP_NOT]], label [[FOR_END]], label [[FOR_BODY_EPIL]], [[LOOP6:!llvm.loop !.*]]
263; AUTO_VEC:       for.end:
264; AUTO_VEC-NEXT:    ret void
265;
266entry:
267  %cmp4 = icmp sgt i32 %N, 0
268  br i1 %cmp4, label %for.body.preheader, label %for.end
269
270for.body.preheader:                               ; preds = %entry
271  br label %for.body
272
273for.body:                                         ; preds = %for.body.preheader, %for.body
274  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %for.body.preheader ]
275  %x.06 = phi float [ %conv1, %for.body ], [ 1.000000e+00, %for.body.preheader ]
276  %arrayidx = getelementptr inbounds float, float* %A, i64 %indvars.iv
277  store float %x.06, float* %arrayidx, align 4
278  %conv1 = fadd float %x.06, 5.000000e-01
279  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
280  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
281  %exitcond = icmp eq i32 %lftr.wideiv, %N
282  br i1 %exitcond, label %for.end.loopexit, label %for.body
283
284for.end.loopexit:                                 ; preds = %for.body
285  br label %for.end
286
287for.end:                                          ; preds = %for.end.loopexit, %entry
288  ret void
289}
290
291define double @external_use_with_fast_math(double* %a, i64 %n) {
292; AUTO_VEC-LABEL: @external_use_with_fast_math(
293; AUTO_VEC-NEXT:  entry:
294; AUTO_VEC-NEXT:    [[TMP0:%.*]] = icmp sgt i64 [[N:%.*]], 1
295; AUTO_VEC-NEXT:    [[SMAX:%.*]] = select i1 [[TMP0]], i64 [[N]], i64 1
296; AUTO_VEC-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[SMAX]], 16
297; AUTO_VEC-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[FOR_BODY:%.*]], label [[VECTOR_PH:%.*]]
298; AUTO_VEC:       vector.ph:
299; AUTO_VEC-NEXT:    [[N_VEC:%.*]] = and i64 [[SMAX]], 9223372036854775792
300; AUTO_VEC-NEXT:    [[CAST_CRD:%.*]] = sitofp i64 [[N_VEC]] to double
301; AUTO_VEC-NEXT:    [[TMP1:%.*]] = fmul fast double [[CAST_CRD]], 3.000000e+00
302; AUTO_VEC-NEXT:    [[TMP2:%.*]] = add nsw i64 [[N_VEC]], -16
303; AUTO_VEC-NEXT:    [[TMP3:%.*]] = lshr exact i64 [[TMP2]], 4
304; AUTO_VEC-NEXT:    [[TMP4:%.*]] = add nuw nsw i64 [[TMP3]], 1
305; AUTO_VEC-NEXT:    [[XTRAITER:%.*]] = and i64 [[TMP4]], 3
306; AUTO_VEC-NEXT:    [[TMP5:%.*]] = icmp ult i64 [[TMP2]], 48
307; AUTO_VEC-NEXT:    br i1 [[TMP5]], label [[MIDDLE_BLOCK_UNR_LCSSA:%.*]], label [[VECTOR_PH_NEW:%.*]]
308; AUTO_VEC:       vector.ph.new:
309; AUTO_VEC-NEXT:    [[UNROLL_ITER:%.*]] = and i64 [[TMP4]], 2305843009213693948
310; AUTO_VEC-NEXT:    br label [[VECTOR_BODY:%.*]]
311; AUTO_VEC:       vector.body:
312; AUTO_VEC-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH_NEW]] ], [ [[INDEX_NEXT_3:%.*]], [[VECTOR_BODY]] ]
313; AUTO_VEC-NEXT:    [[VEC_IND:%.*]] = phi <4 x double> [ <double 0.000000e+00, double 3.000000e+00, double 6.000000e+00, double 9.000000e+00>, [[VECTOR_PH_NEW]] ], [ [[VEC_IND_NEXT_3:%.*]], [[VECTOR_BODY]] ]
314; AUTO_VEC-NEXT:    [[NITER:%.*]] = phi i64 [ [[UNROLL_ITER]], [[VECTOR_PH_NEW]] ], [ [[NITER_NSUB_3:%.*]], [[VECTOR_BODY]] ]
315; AUTO_VEC-NEXT:    [[STEP_ADD:%.*]] = fadd fast <4 x double> [[VEC_IND]], <double 1.200000e+01, double 1.200000e+01, double 1.200000e+01, double 1.200000e+01>
316; AUTO_VEC-NEXT:    [[STEP_ADD2:%.*]] = fadd fast <4 x double> [[VEC_IND]], <double 2.400000e+01, double 2.400000e+01, double 2.400000e+01, double 2.400000e+01>
317; AUTO_VEC-NEXT:    [[STEP_ADD3:%.*]] = fadd fast <4 x double> [[VEC_IND]], <double 3.600000e+01, double 3.600000e+01, double 3.600000e+01, double 3.600000e+01>
318; AUTO_VEC-NEXT:    [[TMP6:%.*]] = getelementptr double, double* [[A:%.*]], i64 [[INDEX]]
319; AUTO_VEC-NEXT:    [[TMP7:%.*]] = bitcast double* [[TMP6]] to <4 x double>*
320; AUTO_VEC-NEXT:    store <4 x double> [[VEC_IND]], <4 x double>* [[TMP7]], align 8
321; AUTO_VEC-NEXT:    [[TMP8:%.*]] = getelementptr double, double* [[TMP6]], i64 4
322; AUTO_VEC-NEXT:    [[TMP9:%.*]] = bitcast double* [[TMP8]] to <4 x double>*
323; AUTO_VEC-NEXT:    store <4 x double> [[STEP_ADD]], <4 x double>* [[TMP9]], align 8
324; AUTO_VEC-NEXT:    [[TMP10:%.*]] = getelementptr double, double* [[TMP6]], i64 8
325; AUTO_VEC-NEXT:    [[TMP11:%.*]] = bitcast double* [[TMP10]] to <4 x double>*
326; AUTO_VEC-NEXT:    store <4 x double> [[STEP_ADD2]], <4 x double>* [[TMP11]], align 8
327; AUTO_VEC-NEXT:    [[TMP12:%.*]] = getelementptr double, double* [[TMP6]], i64 12
328; AUTO_VEC-NEXT:    [[TMP13:%.*]] = bitcast double* [[TMP12]] to <4 x double>*
329; AUTO_VEC-NEXT:    store <4 x double> [[STEP_ADD3]], <4 x double>* [[TMP13]], align 8
330; AUTO_VEC-NEXT:    [[INDEX_NEXT:%.*]] = or i64 [[INDEX]], 16
331; AUTO_VEC-NEXT:    [[VEC_IND_NEXT:%.*]] = fadd fast <4 x double> [[VEC_IND]], <double 4.800000e+01, double 4.800000e+01, double 4.800000e+01, double 4.800000e+01>
332; AUTO_VEC-NEXT:    [[STEP_ADD_1:%.*]] = fadd fast <4 x double> [[VEC_IND]], <double 6.000000e+01, double 6.000000e+01, double 6.000000e+01, double 6.000000e+01>
333; AUTO_VEC-NEXT:    [[STEP_ADD2_1:%.*]] = fadd fast <4 x double> [[VEC_IND]], <double 7.200000e+01, double 7.200000e+01, double 7.200000e+01, double 7.200000e+01>
334; AUTO_VEC-NEXT:    [[STEP_ADD3_1:%.*]] = fadd fast <4 x double> [[VEC_IND]], <double 8.400000e+01, double 8.400000e+01, double 8.400000e+01, double 8.400000e+01>
335; AUTO_VEC-NEXT:    [[TMP14:%.*]] = getelementptr double, double* [[A]], i64 [[INDEX_NEXT]]
336; AUTO_VEC-NEXT:    [[TMP15:%.*]] = bitcast double* [[TMP14]] to <4 x double>*
337; AUTO_VEC-NEXT:    store <4 x double> [[VEC_IND_NEXT]], <4 x double>* [[TMP15]], align 8
338; AUTO_VEC-NEXT:    [[TMP16:%.*]] = getelementptr double, double* [[TMP14]], i64 4
339; AUTO_VEC-NEXT:    [[TMP17:%.*]] = bitcast double* [[TMP16]] to <4 x double>*
340; AUTO_VEC-NEXT:    store <4 x double> [[STEP_ADD_1]], <4 x double>* [[TMP17]], align 8
341; AUTO_VEC-NEXT:    [[TMP18:%.*]] = getelementptr double, double* [[TMP14]], i64 8
342; AUTO_VEC-NEXT:    [[TMP19:%.*]] = bitcast double* [[TMP18]] to <4 x double>*
343; AUTO_VEC-NEXT:    store <4 x double> [[STEP_ADD2_1]], <4 x double>* [[TMP19]], align 8
344; AUTO_VEC-NEXT:    [[TMP20:%.*]] = getelementptr double, double* [[TMP14]], i64 12
345; AUTO_VEC-NEXT:    [[TMP21:%.*]] = bitcast double* [[TMP20]] to <4 x double>*
346; AUTO_VEC-NEXT:    store <4 x double> [[STEP_ADD3_1]], <4 x double>* [[TMP21]], align 8
347; AUTO_VEC-NEXT:    [[INDEX_NEXT_1:%.*]] = or i64 [[INDEX]], 32
348; AUTO_VEC-NEXT:    [[VEC_IND_NEXT_1:%.*]] = fadd fast <4 x double> [[VEC_IND]], <double 9.600000e+01, double 9.600000e+01, double 9.600000e+01, double 9.600000e+01>
349; AUTO_VEC-NEXT:    [[STEP_ADD_2:%.*]] = fadd fast <4 x double> [[VEC_IND]], <double 1.080000e+02, double 1.080000e+02, double 1.080000e+02, double 1.080000e+02>
350; AUTO_VEC-NEXT:    [[STEP_ADD2_2:%.*]] = fadd fast <4 x double> [[VEC_IND]], <double 1.200000e+02, double 1.200000e+02, double 1.200000e+02, double 1.200000e+02>
351; AUTO_VEC-NEXT:    [[STEP_ADD3_2:%.*]] = fadd fast <4 x double> [[VEC_IND]], <double 1.320000e+02, double 1.320000e+02, double 1.320000e+02, double 1.320000e+02>
352; AUTO_VEC-NEXT:    [[TMP22:%.*]] = getelementptr double, double* [[A]], i64 [[INDEX_NEXT_1]]
353; AUTO_VEC-NEXT:    [[TMP23:%.*]] = bitcast double* [[TMP22]] to <4 x double>*
354; AUTO_VEC-NEXT:    store <4 x double> [[VEC_IND_NEXT_1]], <4 x double>* [[TMP23]], align 8
355; AUTO_VEC-NEXT:    [[TMP24:%.*]] = getelementptr double, double* [[TMP22]], i64 4
356; AUTO_VEC-NEXT:    [[TMP25:%.*]] = bitcast double* [[TMP24]] to <4 x double>*
357; AUTO_VEC-NEXT:    store <4 x double> [[STEP_ADD_2]], <4 x double>* [[TMP25]], align 8
358; AUTO_VEC-NEXT:    [[TMP26:%.*]] = getelementptr double, double* [[TMP22]], i64 8
359; AUTO_VEC-NEXT:    [[TMP27:%.*]] = bitcast double* [[TMP26]] to <4 x double>*
360; AUTO_VEC-NEXT:    store <4 x double> [[STEP_ADD2_2]], <4 x double>* [[TMP27]], align 8
361; AUTO_VEC-NEXT:    [[TMP28:%.*]] = getelementptr double, double* [[TMP22]], i64 12
362; AUTO_VEC-NEXT:    [[TMP29:%.*]] = bitcast double* [[TMP28]] to <4 x double>*
363; AUTO_VEC-NEXT:    store <4 x double> [[STEP_ADD3_2]], <4 x double>* [[TMP29]], align 8
364; AUTO_VEC-NEXT:    [[INDEX_NEXT_2:%.*]] = or i64 [[INDEX]], 48
365; AUTO_VEC-NEXT:    [[VEC_IND_NEXT_2:%.*]] = fadd fast <4 x double> [[VEC_IND]], <double 1.440000e+02, double 1.440000e+02, double 1.440000e+02, double 1.440000e+02>
366; AUTO_VEC-NEXT:    [[STEP_ADD_3:%.*]] = fadd fast <4 x double> [[VEC_IND]], <double 1.560000e+02, double 1.560000e+02, double 1.560000e+02, double 1.560000e+02>
367; AUTO_VEC-NEXT:    [[STEP_ADD2_3:%.*]] = fadd fast <4 x double> [[VEC_IND]], <double 1.680000e+02, double 1.680000e+02, double 1.680000e+02, double 1.680000e+02>
368; AUTO_VEC-NEXT:    [[STEP_ADD3_3:%.*]] = fadd fast <4 x double> [[VEC_IND]], <double 1.800000e+02, double 1.800000e+02, double 1.800000e+02, double 1.800000e+02>
369; AUTO_VEC-NEXT:    [[TMP30:%.*]] = getelementptr double, double* [[A]], i64 [[INDEX_NEXT_2]]
370; AUTO_VEC-NEXT:    [[TMP31:%.*]] = bitcast double* [[TMP30]] to <4 x double>*
371; AUTO_VEC-NEXT:    store <4 x double> [[VEC_IND_NEXT_2]], <4 x double>* [[TMP31]], align 8
372; AUTO_VEC-NEXT:    [[TMP32:%.*]] = getelementptr double, double* [[TMP30]], i64 4
373; AUTO_VEC-NEXT:    [[TMP33:%.*]] = bitcast double* [[TMP32]] to <4 x double>*
374; AUTO_VEC-NEXT:    store <4 x double> [[STEP_ADD_3]], <4 x double>* [[TMP33]], align 8
375; AUTO_VEC-NEXT:    [[TMP34:%.*]] = getelementptr double, double* [[TMP30]], i64 8
376; AUTO_VEC-NEXT:    [[TMP35:%.*]] = bitcast double* [[TMP34]] to <4 x double>*
377; AUTO_VEC-NEXT:    store <4 x double> [[STEP_ADD2_3]], <4 x double>* [[TMP35]], align 8
378; AUTO_VEC-NEXT:    [[TMP36:%.*]] = getelementptr double, double* [[TMP30]], i64 12
379; AUTO_VEC-NEXT:    [[TMP37:%.*]] = bitcast double* [[TMP36]] to <4 x double>*
380; AUTO_VEC-NEXT:    store <4 x double> [[STEP_ADD3_3]], <4 x double>* [[TMP37]], align 8
381; AUTO_VEC-NEXT:    [[INDEX_NEXT_3]] = add i64 [[INDEX]], 64
382; AUTO_VEC-NEXT:    [[VEC_IND_NEXT_3]] = fadd fast <4 x double> [[VEC_IND]], <double 1.920000e+02, double 1.920000e+02, double 1.920000e+02, double 1.920000e+02>
383; AUTO_VEC-NEXT:    [[NITER_NSUB_3]] = add i64 [[NITER]], -4
384; AUTO_VEC-NEXT:    [[NITER_NCMP_3:%.*]] = icmp eq i64 [[NITER_NSUB_3]], 0
385; AUTO_VEC-NEXT:    br i1 [[NITER_NCMP_3]], label [[MIDDLE_BLOCK_UNR_LCSSA]], label [[VECTOR_BODY]], [[LOOP7:!llvm.loop !.*]]
386; AUTO_VEC:       middle.block.unr-lcssa:
387; AUTO_VEC-NEXT:    [[INDEX_UNR:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT_3]], [[VECTOR_BODY]] ]
388; AUTO_VEC-NEXT:    [[VEC_IND_UNR:%.*]] = phi <4 x double> [ <double 0.000000e+00, double 3.000000e+00, double 6.000000e+00, double 9.000000e+00>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT_3]], [[VECTOR_BODY]] ]
389; AUTO_VEC-NEXT:    [[LCMP_MOD_NOT:%.*]] = icmp eq i64 [[XTRAITER]], 0
390; AUTO_VEC-NEXT:    br i1 [[LCMP_MOD_NOT]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY_EPIL:%.*]]
391; AUTO_VEC:       vector.body.epil:
392; AUTO_VEC-NEXT:    [[INDEX_EPIL:%.*]] = phi i64 [ [[INDEX_NEXT_EPIL:%.*]], [[VECTOR_BODY_EPIL]] ], [ [[INDEX_UNR]], [[MIDDLE_BLOCK_UNR_LCSSA]] ]
393; AUTO_VEC-NEXT:    [[VEC_IND_EPIL:%.*]] = phi <4 x double> [ [[VEC_IND_NEXT_EPIL:%.*]], [[VECTOR_BODY_EPIL]] ], [ [[VEC_IND_UNR]], [[MIDDLE_BLOCK_UNR_LCSSA]] ]
394; AUTO_VEC-NEXT:    [[EPIL_ITER:%.*]] = phi i64 [ [[EPIL_ITER_SUB:%.*]], [[VECTOR_BODY_EPIL]] ], [ [[XTRAITER]], [[MIDDLE_BLOCK_UNR_LCSSA]] ]
395; AUTO_VEC-NEXT:    [[STEP_ADD_EPIL:%.*]] = fadd fast <4 x double> [[VEC_IND_EPIL]], <double 1.200000e+01, double 1.200000e+01, double 1.200000e+01, double 1.200000e+01>
396; AUTO_VEC-NEXT:    [[STEP_ADD2_EPIL:%.*]] = fadd fast <4 x double> [[VEC_IND_EPIL]], <double 2.400000e+01, double 2.400000e+01, double 2.400000e+01, double 2.400000e+01>
397; AUTO_VEC-NEXT:    [[STEP_ADD3_EPIL:%.*]] = fadd fast <4 x double> [[VEC_IND_EPIL]], <double 3.600000e+01, double 3.600000e+01, double 3.600000e+01, double 3.600000e+01>
398; AUTO_VEC-NEXT:    [[TMP38:%.*]] = getelementptr double, double* [[A]], i64 [[INDEX_EPIL]]
399; AUTO_VEC-NEXT:    [[TMP39:%.*]] = bitcast double* [[TMP38]] to <4 x double>*
400; AUTO_VEC-NEXT:    store <4 x double> [[VEC_IND_EPIL]], <4 x double>* [[TMP39]], align 8
401; AUTO_VEC-NEXT:    [[TMP40:%.*]] = getelementptr double, double* [[TMP38]], i64 4
402; AUTO_VEC-NEXT:    [[TMP41:%.*]] = bitcast double* [[TMP40]] to <4 x double>*
403; AUTO_VEC-NEXT:    store <4 x double> [[STEP_ADD_EPIL]], <4 x double>* [[TMP41]], align 8
404; AUTO_VEC-NEXT:    [[TMP42:%.*]] = getelementptr double, double* [[TMP38]], i64 8
405; AUTO_VEC-NEXT:    [[TMP43:%.*]] = bitcast double* [[TMP42]] to <4 x double>*
406; AUTO_VEC-NEXT:    store <4 x double> [[STEP_ADD2_EPIL]], <4 x double>* [[TMP43]], align 8
407; AUTO_VEC-NEXT:    [[TMP44:%.*]] = getelementptr double, double* [[TMP38]], i64 12
408; AUTO_VEC-NEXT:    [[TMP45:%.*]] = bitcast double* [[TMP44]] to <4 x double>*
409; AUTO_VEC-NEXT:    store <4 x double> [[STEP_ADD3_EPIL]], <4 x double>* [[TMP45]], align 8
410; AUTO_VEC-NEXT:    [[INDEX_NEXT_EPIL]] = add i64 [[INDEX_EPIL]], 16
411; AUTO_VEC-NEXT:    [[VEC_IND_NEXT_EPIL]] = fadd fast <4 x double> [[VEC_IND_EPIL]], <double 4.800000e+01, double 4.800000e+01, double 4.800000e+01, double 4.800000e+01>
412; AUTO_VEC-NEXT:    [[EPIL_ITER_SUB]] = add i64 [[EPIL_ITER]], -1
413; AUTO_VEC-NEXT:    [[EPIL_ITER_CMP_NOT:%.*]] = icmp eq i64 [[EPIL_ITER_SUB]], 0
414; AUTO_VEC-NEXT:    br i1 [[EPIL_ITER_CMP_NOT]], label [[MIDDLE_BLOCK]], label [[VECTOR_BODY_EPIL]], [[LOOP8:!llvm.loop !.*]]
415; AUTO_VEC:       middle.block:
416; AUTO_VEC-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[SMAX]], [[N_VEC]]
417; AUTO_VEC-NEXT:    [[TMP46:%.*]] = add nsw i64 [[N_VEC]], -1
418; AUTO_VEC-NEXT:    [[CAST_CMO:%.*]] = sitofp i64 [[TMP46]] to double
419; AUTO_VEC-NEXT:    [[TMP47:%.*]] = fmul fast double [[CAST_CMO]], 3.000000e+00
420; AUTO_VEC-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[FOR_BODY]]
421; AUTO_VEC:       for.body:
422; AUTO_VEC-NEXT:    [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ 0, [[ENTRY:%.*]] ], [ [[N_VEC]], [[MIDDLE_BLOCK]] ]
423; AUTO_VEC-NEXT:    [[J:%.*]] = phi double [ [[J_NEXT:%.*]], [[FOR_BODY]] ], [ 0.000000e+00, [[ENTRY]] ], [ [[TMP1]], [[MIDDLE_BLOCK]] ]
424; AUTO_VEC-NEXT:    [[T0:%.*]] = getelementptr double, double* [[A]], i64 [[I]]
425; AUTO_VEC-NEXT:    store double [[J]], double* [[T0]], align 8
426; AUTO_VEC-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 1
427; AUTO_VEC-NEXT:    [[J_NEXT]] = fadd fast double [[J]], 3.000000e+00
428; AUTO_VEC-NEXT:    [[EXITCOND_NOT:%.*]] = icmp eq i64 [[I_NEXT]], [[SMAX]]
429; AUTO_VEC-NEXT:    br i1 [[EXITCOND_NOT]], label [[FOR_END]], label [[FOR_BODY]], [[LOOP9:!llvm.loop !.*]]
430; AUTO_VEC:       for.end:
431; AUTO_VEC-NEXT:    [[J_LCSSA:%.*]] = phi double [ [[TMP47]], [[MIDDLE_BLOCK]] ], [ [[J]], [[FOR_BODY]] ]
432; AUTO_VEC-NEXT:    ret double [[J_LCSSA]]
433;
434entry:
435  br label %for.body
436
437for.body:
438  %i = phi i64 [ 0, %entry ], [%i.next, %for.body]
439  %j = phi double [ 0.0, %entry ], [ %j.next, %for.body ]
440  %t0 = getelementptr double, double* %a, i64 %i
441  store double %j, double* %t0
442  %i.next = add i64 %i, 1
443  %j.next = fadd fast double %j, 3.0
444  %cond = icmp slt i64 %i.next, %n
445  br i1 %cond, label %for.body, label %for.end
446
447for.end:
448  %t1 = phi double [ %j, %for.body ]
449  ret double %t1
450}
451
452define double @external_use_without_fast_math(double* %a, i64 %n) {
453; AUTO_VEC-LABEL: @external_use_without_fast_math(
454; AUTO_VEC-NEXT:  entry:
455; AUTO_VEC-NEXT:    [[TMP0:%.*]] = icmp sgt i64 [[N:%.*]], 1
456; AUTO_VEC-NEXT:    [[SMAX:%.*]] = select i1 [[TMP0]], i64 [[N]], i64 1
457; AUTO_VEC-NEXT:    [[TMP1:%.*]] = add nsw i64 [[SMAX]], -1
458; AUTO_VEC-NEXT:    [[XTRAITER:%.*]] = and i64 [[SMAX]], 7
459; AUTO_VEC-NEXT:    [[TMP2:%.*]] = icmp ult i64 [[TMP1]], 7
460; AUTO_VEC-NEXT:    br i1 [[TMP2]], label [[FOR_END_UNR_LCSSA:%.*]], label [[ENTRY_NEW:%.*]]
461; AUTO_VEC:       entry.new:
462; AUTO_VEC-NEXT:    [[UNROLL_ITER:%.*]] = and i64 [[SMAX]], 9223372036854775800
463; AUTO_VEC-NEXT:    br label [[FOR_BODY:%.*]]
464; AUTO_VEC:       for.body:
465; AUTO_VEC-NEXT:    [[I:%.*]] = phi i64 [ 0, [[ENTRY_NEW]] ], [ [[I_NEXT_7:%.*]], [[FOR_BODY]] ]
466; AUTO_VEC-NEXT:    [[J:%.*]] = phi double [ 0.000000e+00, [[ENTRY_NEW]] ], [ [[J_NEXT_7:%.*]], [[FOR_BODY]] ]
467; AUTO_VEC-NEXT:    [[NITER:%.*]] = phi i64 [ [[UNROLL_ITER]], [[ENTRY_NEW]] ], [ [[NITER_NSUB_7:%.*]], [[FOR_BODY]] ]
468; AUTO_VEC-NEXT:    [[T0:%.*]] = getelementptr double, double* [[A:%.*]], i64 [[I]]
469; AUTO_VEC-NEXT:    store double [[J]], double* [[T0]], align 8
470; AUTO_VEC-NEXT:    [[I_NEXT:%.*]] = or i64 [[I]], 1
471; AUTO_VEC-NEXT:    [[J_NEXT:%.*]] = fadd double [[J]], 3.000000e+00
472; AUTO_VEC-NEXT:    [[T0_1:%.*]] = getelementptr double, double* [[A]], i64 [[I_NEXT]]
473; AUTO_VEC-NEXT:    store double [[J_NEXT]], double* [[T0_1]], align 8
474; AUTO_VEC-NEXT:    [[I_NEXT_1:%.*]] = or i64 [[I]], 2
475; AUTO_VEC-NEXT:    [[J_NEXT_1:%.*]] = fadd double [[J_NEXT]], 3.000000e+00
476; AUTO_VEC-NEXT:    [[T0_2:%.*]] = getelementptr double, double* [[A]], i64 [[I_NEXT_1]]
477; AUTO_VEC-NEXT:    store double [[J_NEXT_1]], double* [[T0_2]], align 8
478; AUTO_VEC-NEXT:    [[I_NEXT_2:%.*]] = or i64 [[I]], 3
479; AUTO_VEC-NEXT:    [[J_NEXT_2:%.*]] = fadd double [[J_NEXT_1]], 3.000000e+00
480; AUTO_VEC-NEXT:    [[T0_3:%.*]] = getelementptr double, double* [[A]], i64 [[I_NEXT_2]]
481; AUTO_VEC-NEXT:    store double [[J_NEXT_2]], double* [[T0_3]], align 8
482; AUTO_VEC-NEXT:    [[I_NEXT_3:%.*]] = or i64 [[I]], 4
483; AUTO_VEC-NEXT:    [[J_NEXT_3:%.*]] = fadd double [[J_NEXT_2]], 3.000000e+00
484; AUTO_VEC-NEXT:    [[T0_4:%.*]] = getelementptr double, double* [[A]], i64 [[I_NEXT_3]]
485; AUTO_VEC-NEXT:    store double [[J_NEXT_3]], double* [[T0_4]], align 8
486; AUTO_VEC-NEXT:    [[I_NEXT_4:%.*]] = or i64 [[I]], 5
487; AUTO_VEC-NEXT:    [[J_NEXT_4:%.*]] = fadd double [[J_NEXT_3]], 3.000000e+00
488; AUTO_VEC-NEXT:    [[T0_5:%.*]] = getelementptr double, double* [[A]], i64 [[I_NEXT_4]]
489; AUTO_VEC-NEXT:    store double [[J_NEXT_4]], double* [[T0_5]], align 8
490; AUTO_VEC-NEXT:    [[I_NEXT_5:%.*]] = or i64 [[I]], 6
491; AUTO_VEC-NEXT:    [[J_NEXT_5:%.*]] = fadd double [[J_NEXT_4]], 3.000000e+00
492; AUTO_VEC-NEXT:    [[T0_6:%.*]] = getelementptr double, double* [[A]], i64 [[I_NEXT_5]]
493; AUTO_VEC-NEXT:    store double [[J_NEXT_5]], double* [[T0_6]], align 8
494; AUTO_VEC-NEXT:    [[I_NEXT_6:%.*]] = or i64 [[I]], 7
495; AUTO_VEC-NEXT:    [[J_NEXT_6:%.*]] = fadd double [[J_NEXT_5]], 3.000000e+00
496; AUTO_VEC-NEXT:    [[T0_7:%.*]] = getelementptr double, double* [[A]], i64 [[I_NEXT_6]]
497; AUTO_VEC-NEXT:    store double [[J_NEXT_6]], double* [[T0_7]], align 8
498; AUTO_VEC-NEXT:    [[I_NEXT_7]] = add nuw nsw i64 [[I]], 8
499; AUTO_VEC-NEXT:    [[J_NEXT_7]] = fadd double [[J_NEXT_6]], 3.000000e+00
500; AUTO_VEC-NEXT:    [[NITER_NSUB_7]] = add i64 [[NITER]], -8
501; AUTO_VEC-NEXT:    [[NITER_NCMP_7:%.*]] = icmp eq i64 [[NITER_NSUB_7]], 0
502; AUTO_VEC-NEXT:    br i1 [[NITER_NCMP_7]], label [[FOR_END_UNR_LCSSA]], label [[FOR_BODY]]
503; AUTO_VEC:       for.end.unr-lcssa:
504; AUTO_VEC-NEXT:    [[J_LCSSA_PH:%.*]] = phi double [ undef, [[ENTRY:%.*]] ], [ [[J_NEXT_6]], [[FOR_BODY]] ]
505; AUTO_VEC-NEXT:    [[I_UNR:%.*]] = phi i64 [ 0, [[ENTRY]] ], [ [[I_NEXT_7]], [[FOR_BODY]] ]
506; AUTO_VEC-NEXT:    [[J_UNR:%.*]] = phi double [ 0.000000e+00, [[ENTRY]] ], [ [[J_NEXT_7]], [[FOR_BODY]] ]
507; AUTO_VEC-NEXT:    [[LCMP_MOD_NOT:%.*]] = icmp eq i64 [[XTRAITER]], 0
508; AUTO_VEC-NEXT:    br i1 [[LCMP_MOD_NOT]], label [[FOR_END:%.*]], label [[FOR_BODY_EPIL:%.*]]
509; AUTO_VEC:       for.body.epil:
510; AUTO_VEC-NEXT:    [[I_EPIL:%.*]] = phi i64 [ [[I_NEXT_EPIL:%.*]], [[FOR_BODY_EPIL]] ], [ [[I_UNR]], [[FOR_END_UNR_LCSSA]] ]
511; AUTO_VEC-NEXT:    [[J_EPIL:%.*]] = phi double [ [[J_NEXT_EPIL:%.*]], [[FOR_BODY_EPIL]] ], [ [[J_UNR]], [[FOR_END_UNR_LCSSA]] ]
512; AUTO_VEC-NEXT:    [[EPIL_ITER:%.*]] = phi i64 [ [[EPIL_ITER_SUB:%.*]], [[FOR_BODY_EPIL]] ], [ [[XTRAITER]], [[FOR_END_UNR_LCSSA]] ]
513; AUTO_VEC-NEXT:    [[T0_EPIL:%.*]] = getelementptr double, double* [[A]], i64 [[I_EPIL]]
514; AUTO_VEC-NEXT:    store double [[J_EPIL]], double* [[T0_EPIL]], align 8
515; AUTO_VEC-NEXT:    [[I_NEXT_EPIL]] = add nuw nsw i64 [[I_EPIL]], 1
516; AUTO_VEC-NEXT:    [[J_NEXT_EPIL]] = fadd double [[J_EPIL]], 3.000000e+00
517; AUTO_VEC-NEXT:    [[EPIL_ITER_SUB]] = add i64 [[EPIL_ITER]], -1
518; AUTO_VEC-NEXT:    [[EPIL_ITER_CMP_NOT:%.*]] = icmp eq i64 [[EPIL_ITER_SUB]], 0
519; AUTO_VEC-NEXT:    br i1 [[EPIL_ITER_CMP_NOT]], label [[FOR_END]], label [[FOR_BODY_EPIL]], [[LOOP10:!llvm.loop !.*]]
520; AUTO_VEC:       for.end:
521; AUTO_VEC-NEXT:    [[J_LCSSA:%.*]] = phi double [ [[J_LCSSA_PH]], [[FOR_END_UNR_LCSSA]] ], [ [[J_EPIL]], [[FOR_BODY_EPIL]] ]
522; AUTO_VEC-NEXT:    ret double [[J_LCSSA]]
523;
524entry:
525  br label %for.body
526
527for.body:
528  %i = phi i64 [ 0, %entry ], [%i.next, %for.body]
529  %j = phi double [ 0.0, %entry ], [ %j.next, %for.body ]
530  %t0 = getelementptr double, double* %a, i64 %i
531  store double %j, double* %t0
532  %i.next = add i64 %i, 1
533  %j.next = fadd double %j, 3.0
534  %cond = icmp slt i64 %i.next, %n
535  br i1 %cond, label %for.body, label %for.end
536
537for.end:
538  %t1 = phi double [ %j, %for.body ]
539  ret double %t1
540}
541
542attributes #0 = { "no-nans-fp-math"="true" }
543attributes #1 = { "no-nans-fp-math"="false" }
544