xref: /llvm-project/llvm/test/Transforms/LoopVectorize/X86/float-induction-x86.ll (revision 7f3428d3ed71d87a2088b77b6cab9f3d86544234) 
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s -passes="default<O3>,simplifycfg" -simplifycfg-require-and-preserve-domtree=1 -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; FMF is required on the IR instructions.
8
9;void fp_iv_loop1(ptr __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(ptr 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 [[ITER_CHECK:%.*]], label [[FOR_END:%.*]]
22; AUTO_VEC:       iter.check:
23; AUTO_VEC-NEXT:    [[ZEXT:%.*]] = zext nneg i32 [[N]] to i64
24; AUTO_VEC-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[N]], 4
25; AUTO_VEC-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[FOR_BODY:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]]
26; AUTO_VEC:       vector.main.loop.iter.check:
27; AUTO_VEC-NEXT:    [[MIN_ITERS_CHECK1:%.*]] = icmp ult i32 [[N]], 32
28; AUTO_VEC-NEXT:    br i1 [[MIN_ITERS_CHECK1]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]]
29; AUTO_VEC:       vector.ph:
30; AUTO_VEC-NEXT:    [[N_VEC:%.*]] = and i64 [[ZEXT]], 2147483616
31; AUTO_VEC-NEXT:    [[DOTCAST:%.*]] = uitofp nneg i64 [[N_VEC]] to float
32; AUTO_VEC-NEXT:    [[TMP0:%.*]] = fmul fast float [[DOTCAST]], 5.000000e-01
33; AUTO_VEC-NEXT:    [[IND_END:%.*]] = fadd fast float [[TMP0]], 1.000000e+00
34; AUTO_VEC-NEXT:    br label [[VECTOR_BODY:%.*]]
35; AUTO_VEC:       vector.body:
36; AUTO_VEC-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
37; 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]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
38; AUTO_VEC-NEXT:    [[STEP_ADD:%.*]] = fadd fast <8 x float> [[VEC_IND]], splat (float 4.000000e+00)
39; AUTO_VEC-NEXT:    [[STEP_ADD2:%.*]] = fadd fast <8 x float> [[VEC_IND]], splat (float 8.000000e+00)
40; AUTO_VEC-NEXT:    [[STEP_ADD3:%.*]] = fadd fast <8 x float> [[VEC_IND]], splat (float 1.200000e+01)
41; AUTO_VEC-NEXT:    [[TMP1:%.*]] = getelementptr inbounds nuw float, ptr [[A:%.*]], i64 [[INDEX]]
42; AUTO_VEC-NEXT:    [[TMP2:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP1]], i64 32
43; AUTO_VEC-NEXT:    [[TMP3:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP1]], i64 64
44; AUTO_VEC-NEXT:    [[TMP4:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP1]], i64 96
45; AUTO_VEC-NEXT:    store <8 x float> [[VEC_IND]], ptr [[TMP1]], align 4
46; AUTO_VEC-NEXT:    store <8 x float> [[STEP_ADD]], ptr [[TMP2]], align 4
47; AUTO_VEC-NEXT:    store <8 x float> [[STEP_ADD2]], ptr [[TMP3]], align 4
48; AUTO_VEC-NEXT:    store <8 x float> [[STEP_ADD3]], ptr [[TMP4]], align 4
49; AUTO_VEC-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32
50; AUTO_VEC-NEXT:    [[VEC_IND_NEXT]] = fadd fast <8 x float> [[VEC_IND]], splat (float 1.600000e+01)
51; AUTO_VEC-NEXT:    [[TMP5:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
52; AUTO_VEC-NEXT:    br i1 [[TMP5]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
53; AUTO_VEC:       middle.block:
54; AUTO_VEC-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N_VEC]], [[ZEXT]]
55; AUTO_VEC-NEXT:    br i1 [[CMP_N]], label [[FOR_END]], label [[VEC_EPILOG_ITER_CHECK:%.*]]
56; AUTO_VEC:       vec.epilog.iter.check:
57; AUTO_VEC-NEXT:    [[DOTCAST7:%.*]] = uitofp nneg i64 [[N_VEC]] to float
58; AUTO_VEC-NEXT:    [[TMP6:%.*]] = fmul fast float [[DOTCAST7]], 5.000000e-01
59; AUTO_VEC-NEXT:    [[IND_END8:%.*]] = fadd fast float [[TMP6]], 1.000000e+00
60; AUTO_VEC-NEXT:    [[N_VEC_REMAINING:%.*]] = and i64 [[ZEXT]], 28
61; AUTO_VEC-NEXT:    [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp eq i64 [[N_VEC_REMAINING]], 0
62; AUTO_VEC-NEXT:    br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[FOR_BODY]], label [[VEC_EPILOG_PH]]
63; AUTO_VEC:       vec.epilog.ph:
64; AUTO_VEC-NEXT:    [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
65; AUTO_VEC-NEXT:    [[BC_RESUME_VAL:%.*]] = phi float [ [[IND_END]], [[VEC_EPILOG_ITER_CHECK]] ], [ 1.000000e+00, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
66; AUTO_VEC-NEXT:    [[N_VEC3:%.*]] = and i64 [[ZEXT]], 2147483644
67; AUTO_VEC-NEXT:    [[DOTCAST5:%.*]] = uitofp nneg i64 [[N_VEC3]] to float
68; AUTO_VEC-NEXT:    [[TMP7:%.*]] = fmul fast float [[DOTCAST5]], 5.000000e-01
69; AUTO_VEC-NEXT:    [[IND_END6:%.*]] = fadd fast float [[TMP7]], 1.000000e+00
70; AUTO_VEC-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <4 x float> poison, float [[BC_RESUME_VAL]], i64 0
71; AUTO_VEC-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <4 x float> [[DOTSPLATINSERT]], <4 x float> poison, <4 x i32> zeroinitializer
72; AUTO_VEC-NEXT:    [[INDUCTION:%.*]] = fadd fast <4 x float> [[DOTSPLAT]], <float 0.000000e+00, float 5.000000e-01, float 1.000000e+00, float 1.500000e+00>
73; AUTO_VEC-NEXT:    br label [[VEC_EPILOG_VECTOR_BODY:%.*]]
74; AUTO_VEC:       vec.epilog.vector.body:
75; AUTO_VEC-NEXT:    [[INDEX10:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT13:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]
76; AUTO_VEC-NEXT:    [[VEC_IND11:%.*]] = phi <4 x float> [ [[INDUCTION]], [[VEC_EPILOG_PH]] ], [ [[VEC_IND_NEXT12:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]
77; AUTO_VEC-NEXT:    [[TMP8:%.*]] = getelementptr inbounds nuw float, ptr [[A]], i64 [[INDEX10]]
78; AUTO_VEC-NEXT:    store <4 x float> [[VEC_IND11]], ptr [[TMP8]], align 4
79; AUTO_VEC-NEXT:    [[INDEX_NEXT13]] = add nuw i64 [[INDEX10]], 4
80; AUTO_VEC-NEXT:    [[VEC_IND_NEXT12]] = fadd fast <4 x float> [[VEC_IND11]], splat (float 2.000000e+00)
81; AUTO_VEC-NEXT:    [[TMP9:%.*]] = icmp eq i64 [[INDEX_NEXT13]], [[N_VEC3]]
82; AUTO_VEC-NEXT:    br i1 [[TMP9]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]
83; AUTO_VEC:       vec.epilog.middle.block:
84; AUTO_VEC-NEXT:    [[CMP_N14:%.*]] = icmp eq i64 [[N_VEC3]], [[ZEXT]]
85; AUTO_VEC-NEXT:    br i1 [[CMP_N14]], label [[FOR_END]], label [[FOR_BODY]]
86; AUTO_VEC:       for.body:
87; AUTO_VEC-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[ITER_CHECK]] ], [ [[N_VEC3]], [[VEC_EPILOG_MIDDLE_BLOCK]] ]
88; AUTO_VEC-NEXT:    [[X_06:%.*]] = phi float [ [[CONV1:%.*]], [[FOR_BODY]] ], [ [[IND_END8]], [[VEC_EPILOG_ITER_CHECK]] ], [ 1.000000e+00, [[ITER_CHECK]] ], [ [[IND_END6]], [[VEC_EPILOG_MIDDLE_BLOCK]] ]
89; AUTO_VEC-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds nuw float, ptr [[A]], i64 [[INDVARS_IV]]
90; AUTO_VEC-NEXT:    store float [[X_06]], ptr [[ARRAYIDX]], align 4
91; AUTO_VEC-NEXT:    [[CONV1]] = fadd fast float [[X_06]], 5.000000e-01
92; AUTO_VEC-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
93; AUTO_VEC-NEXT:    [[TMP10:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], [[ZEXT]]
94; AUTO_VEC-NEXT:    br i1 [[TMP10]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
95; AUTO_VEC:       for.end:
96; AUTO_VEC-NEXT:    ret void
97;
98entry:
99  %cmp4 = icmp sgt i32 %N, 0
100  br i1 %cmp4, label %for.body.preheader, label %for.end
101
102for.body.preheader:                               ; preds = %entry
103  br label %for.body
104
105for.body:                                         ; preds = %for.body.preheader, %for.body
106  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %for.body.preheader ]
107  %x.06 = phi float [ %conv1, %for.body ], [ 1.000000e+00, %for.body.preheader ]
108  %arrayidx = getelementptr inbounds float, ptr %A, i64 %indvars.iv
109  store float %x.06, ptr %arrayidx, align 4
110  %conv1 = fadd fast float %x.06, 5.000000e-01
111  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
112  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
113  %exitcond = icmp eq i32 %lftr.wideiv, %N
114  br i1 %exitcond, label %for.end.loopexit, label %for.body
115
116for.end.loopexit:                                 ; preds = %for.body
117  br label %for.end
118
119for.end:                                          ; preds = %for.end.loopexit, %entry
120  ret void
121}
122
123; The same as the previous, but FP operation has no FMF.
124; Vectorization should be rejected.
125;void fp_iv_loop2(ptr __restrict__ A, int N) {
126;  float x = 1.0;
127;  for (int i=0; i < N; ++i) {
128;    A[i] = x;
129;    x += 0.5;
130;  }
131;}
132
133define void @fp_iv_loop2(ptr noalias nocapture %A, i32 %N) {
134; AUTO_VEC-LABEL: @fp_iv_loop2(
135; AUTO_VEC-NEXT:  entry:
136; AUTO_VEC-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[N:%.*]], 0
137; AUTO_VEC-NEXT:    br i1 [[CMP4]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_END:%.*]]
138; AUTO_VEC:       for.body.preheader:
139; AUTO_VEC-NEXT:    [[ZEXT:%.*]] = zext nneg i32 [[N]] to i64
140; AUTO_VEC-NEXT:    [[XTRAITER:%.*]] = and i64 [[ZEXT]], 7
141; AUTO_VEC-NEXT:    [[TMP0:%.*]] = icmp ult i32 [[N]], 8
142; AUTO_VEC-NEXT:    br i1 [[TMP0]], label [[FOR_END_LOOPEXIT_UNR_LCSSA:%.*]], label [[FOR_BODY_PREHEADER_NEW:%.*]]
143; AUTO_VEC:       for.body.preheader.new:
144; AUTO_VEC-NEXT:    [[UNROLL_ITER:%.*]] = and i64 [[ZEXT]], 2147483640
145; AUTO_VEC-NEXT:    br label [[FOR_BODY:%.*]]
146; AUTO_VEC:       for.body:
147; AUTO_VEC-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ 0, [[FOR_BODY_PREHEADER_NEW]] ], [ [[INDVARS_IV_NEXT_7:%.*]], [[FOR_BODY]] ]
148; AUTO_VEC-NEXT:    [[X_06:%.*]] = phi float [ 1.000000e+00, [[FOR_BODY_PREHEADER_NEW]] ], [ [[CONV1_7:%.*]], [[FOR_BODY]] ]
149; AUTO_VEC-NEXT:    [[NITER:%.*]] = phi i64 [ 0, [[FOR_BODY_PREHEADER_NEW]] ], [ [[NITER_NEXT_7:%.*]], [[FOR_BODY]] ]
150; AUTO_VEC-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds nuw float, ptr [[A:%.*]], i64 [[INDVARS_IV]]
151; AUTO_VEC-NEXT:    store float [[X_06]], ptr [[ARRAYIDX]], align 4
152; AUTO_VEC-NEXT:    [[CONV1:%.*]] = fadd float [[X_06]], 5.000000e-01
153; AUTO_VEC-NEXT:    [[INDVARS_IV_NEXT:%.*]] = or disjoint i64 [[INDVARS_IV]], 1
154; AUTO_VEC-NEXT:    [[ARRAYIDX_1:%.*]] = getelementptr inbounds nuw float, ptr [[A]], i64 [[INDVARS_IV_NEXT]]
155; AUTO_VEC-NEXT:    store float [[CONV1]], ptr [[ARRAYIDX_1]], align 4
156; AUTO_VEC-NEXT:    [[CONV1_1:%.*]] = fadd float [[CONV1]], 5.000000e-01
157; AUTO_VEC-NEXT:    [[INDVARS_IV_NEXT_1:%.*]] = or disjoint i64 [[INDVARS_IV]], 2
158; AUTO_VEC-NEXT:    [[ARRAYIDX_2:%.*]] = getelementptr inbounds nuw float, ptr [[A]], i64 [[INDVARS_IV_NEXT_1]]
159; AUTO_VEC-NEXT:    store float [[CONV1_1]], ptr [[ARRAYIDX_2]], align 4
160; AUTO_VEC-NEXT:    [[CONV1_2:%.*]] = fadd float [[CONV1_1]], 5.000000e-01
161; AUTO_VEC-NEXT:    [[INDVARS_IV_NEXT_2:%.*]] = or disjoint i64 [[INDVARS_IV]], 3
162; AUTO_VEC-NEXT:    [[ARRAYIDX_3:%.*]] = getelementptr inbounds nuw float, ptr [[A]], i64 [[INDVARS_IV_NEXT_2]]
163; AUTO_VEC-NEXT:    store float [[CONV1_2]], ptr [[ARRAYIDX_3]], align 4
164; AUTO_VEC-NEXT:    [[CONV1_3:%.*]] = fadd float [[CONV1_2]], 5.000000e-01
165; AUTO_VEC-NEXT:    [[INDVARS_IV_NEXT_3:%.*]] = or disjoint i64 [[INDVARS_IV]], 4
166; AUTO_VEC-NEXT:    [[ARRAYIDX_4:%.*]] = getelementptr inbounds nuw float, ptr [[A]], i64 [[INDVARS_IV_NEXT_3]]
167; AUTO_VEC-NEXT:    store float [[CONV1_3]], ptr [[ARRAYIDX_4]], align 4
168; AUTO_VEC-NEXT:    [[CONV1_4:%.*]] = fadd float [[CONV1_3]], 5.000000e-01
169; AUTO_VEC-NEXT:    [[INDVARS_IV_NEXT_4:%.*]] = or disjoint i64 [[INDVARS_IV]], 5
170; AUTO_VEC-NEXT:    [[ARRAYIDX_5:%.*]] = getelementptr inbounds nuw float, ptr [[A]], i64 [[INDVARS_IV_NEXT_4]]
171; AUTO_VEC-NEXT:    store float [[CONV1_4]], ptr [[ARRAYIDX_5]], align 4
172; AUTO_VEC-NEXT:    [[CONV1_5:%.*]] = fadd float [[CONV1_4]], 5.000000e-01
173; AUTO_VEC-NEXT:    [[INDVARS_IV_NEXT_5:%.*]] = or disjoint i64 [[INDVARS_IV]], 6
174; AUTO_VEC-NEXT:    [[ARRAYIDX_6:%.*]] = getelementptr inbounds nuw float, ptr [[A]], i64 [[INDVARS_IV_NEXT_5]]
175; AUTO_VEC-NEXT:    store float [[CONV1_5]], ptr [[ARRAYIDX_6]], align 4
176; AUTO_VEC-NEXT:    [[CONV1_6:%.*]] = fadd float [[CONV1_5]], 5.000000e-01
177; AUTO_VEC-NEXT:    [[INDVARS_IV_NEXT_6:%.*]] = or disjoint i64 [[INDVARS_IV]], 7
178; AUTO_VEC-NEXT:    [[ARRAYIDX_7:%.*]] = getelementptr inbounds nuw float, ptr [[A]], i64 [[INDVARS_IV_NEXT_6]]
179; AUTO_VEC-NEXT:    store float [[CONV1_6]], ptr [[ARRAYIDX_7]], align 4
180; AUTO_VEC-NEXT:    [[CONV1_7]] = fadd float [[CONV1_6]], 5.000000e-01
181; AUTO_VEC-NEXT:    [[INDVARS_IV_NEXT_7]] = add nuw nsw i64 [[INDVARS_IV]], 8
182; AUTO_VEC-NEXT:    [[NITER_NEXT_7]] = add i64 [[NITER]], 8
183; AUTO_VEC-NEXT:    [[NITER_NCMP_7:%.*]] = icmp eq i64 [[NITER_NEXT_7]], [[UNROLL_ITER]]
184; AUTO_VEC-NEXT:    br i1 [[NITER_NCMP_7]], label [[FOR_END_LOOPEXIT_UNR_LCSSA]], label [[FOR_BODY]]
185; AUTO_VEC:       for.end.loopexit.unr-lcssa:
186; AUTO_VEC-NEXT:    [[INDVARS_IV_UNR:%.*]] = phi i64 [ 0, [[FOR_BODY_PREHEADER]] ], [ [[INDVARS_IV_NEXT_7]], [[FOR_BODY]] ]
187; AUTO_VEC-NEXT:    [[X_06_UNR:%.*]] = phi float [ 1.000000e+00, [[FOR_BODY_PREHEADER]] ], [ [[CONV1_7]], [[FOR_BODY]] ]
188; AUTO_VEC-NEXT:    [[LCMP_MOD_NOT:%.*]] = icmp eq i64 [[XTRAITER]], 0
189; AUTO_VEC-NEXT:    br i1 [[LCMP_MOD_NOT]], label [[FOR_END]], label [[FOR_BODY_EPIL:%.*]]
190; AUTO_VEC:       for.body.epil:
191; AUTO_VEC-NEXT:    [[INDVARS_IV_EPIL:%.*]] = phi i64 [ [[INDVARS_IV_NEXT_EPIL:%.*]], [[FOR_BODY_EPIL]] ], [ [[INDVARS_IV_UNR]], [[FOR_END_LOOPEXIT_UNR_LCSSA]] ]
192; AUTO_VEC-NEXT:    [[X_06_EPIL:%.*]] = phi float [ [[CONV1_EPIL:%.*]], [[FOR_BODY_EPIL]] ], [ [[X_06_UNR]], [[FOR_END_LOOPEXIT_UNR_LCSSA]] ]
193; AUTO_VEC-NEXT:    [[EPIL_ITER:%.*]] = phi i64 [ [[EPIL_ITER_NEXT:%.*]], [[FOR_BODY_EPIL]] ], [ 0, [[FOR_END_LOOPEXIT_UNR_LCSSA]] ]
194; AUTO_VEC-NEXT:    [[ARRAYIDX_EPIL:%.*]] = getelementptr inbounds nuw float, ptr [[A]], i64 [[INDVARS_IV_EPIL]]
195; AUTO_VEC-NEXT:    store float [[X_06_EPIL]], ptr [[ARRAYIDX_EPIL]], align 4
196; AUTO_VEC-NEXT:    [[CONV1_EPIL]] = fadd float [[X_06_EPIL]], 5.000000e-01
197; AUTO_VEC-NEXT:    [[INDVARS_IV_NEXT_EPIL]] = add nuw nsw i64 [[INDVARS_IV_EPIL]], 1
198; AUTO_VEC-NEXT:    [[EPIL_ITER_NEXT]] = add i64 [[EPIL_ITER]], 1
199; AUTO_VEC-NEXT:    [[EPIL_ITER_CMP_NOT:%.*]] = icmp eq i64 [[EPIL_ITER_NEXT]], [[XTRAITER]]
200; AUTO_VEC-NEXT:    br i1 [[EPIL_ITER_CMP_NOT]], label [[FOR_END]], label [[FOR_BODY_EPIL]], !llvm.loop [[LOOP5:![0-9]+]]
201; AUTO_VEC:       for.end:
202; AUTO_VEC-NEXT:    ret void
203;
204entry:
205  %cmp4 = icmp sgt i32 %N, 0
206  br i1 %cmp4, label %for.body.preheader, label %for.end
207
208for.body.preheader:                               ; preds = %entry
209  br label %for.body
210
211for.body:                                         ; preds = %for.body.preheader, %for.body
212  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %for.body.preheader ]
213  %x.06 = phi float [ %conv1, %for.body ], [ 1.000000e+00, %for.body.preheader ]
214  %arrayidx = getelementptr inbounds float, ptr %A, i64 %indvars.iv
215  store float %x.06, ptr %arrayidx, align 4
216  %conv1 = fadd float %x.06, 5.000000e-01
217  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
218  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
219  %exitcond = icmp eq i32 %lftr.wideiv, %N
220  br i1 %exitcond, label %for.end.loopexit, label %for.body
221
222for.end.loopexit:                                 ; preds = %for.body
223  br label %for.end
224
225for.end:                                          ; preds = %for.end.loopexit, %entry
226  ret void
227}
228
229define double @external_use_with_fast_math(ptr %a, i64 %n) {
230; AUTO_VEC-LABEL: @external_use_with_fast_math(
231; AUTO_VEC-NEXT:  entry:
232; AUTO_VEC-NEXT:    [[SMAX:%.*]] = tail call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 1)
233; AUTO_VEC-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp slt i64 [[N]], 16
234; AUTO_VEC-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[FOR_BODY:%.*]], label [[VECTOR_PH:%.*]]
235; AUTO_VEC:       vector.ph:
236; AUTO_VEC-NEXT:    [[N_VEC:%.*]] = and i64 [[SMAX]], 9223372036854775792
237; AUTO_VEC-NEXT:    [[DOTCAST:%.*]] = uitofp nneg i64 [[N_VEC]] to double
238; AUTO_VEC-NEXT:    [[TMP0:%.*]] = fmul fast double [[DOTCAST]], 3.000000e+00
239; AUTO_VEC-NEXT:    br label [[VECTOR_BODY:%.*]]
240; AUTO_VEC:       vector.body:
241; AUTO_VEC-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
242; 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]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
243; AUTO_VEC-NEXT:    [[STEP_ADD:%.*]] = fadd fast <4 x double> [[VEC_IND]], splat (double 1.200000e+01)
244; AUTO_VEC-NEXT:    [[STEP_ADD_2:%.*]] = fadd fast <4 x double> [[VEC_IND]], splat (double 2.400000e+01)
245; AUTO_VEC-NEXT:    [[STEP_ADD_3:%.*]] = fadd fast <4 x double> [[VEC_IND]], splat (double 3.600000e+01)
246; AUTO_VEC-NEXT:    [[TMP1:%.*]] = getelementptr double, ptr [[A:%.*]], i64 [[INDEX]]
247; AUTO_VEC-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr [[TMP1]], i64 32
248; AUTO_VEC-NEXT:    [[TMP3:%.*]] = getelementptr i8, ptr [[TMP1]], i64 64
249; AUTO_VEC-NEXT:    [[TMP4:%.*]] = getelementptr i8, ptr [[TMP1]], i64 96
250; AUTO_VEC-NEXT:    store <4 x double> [[VEC_IND]], ptr [[TMP1]], align 8
251; AUTO_VEC-NEXT:    store <4 x double> [[STEP_ADD]], ptr [[TMP2]], align 8
252; AUTO_VEC-NEXT:    store <4 x double> [[STEP_ADD_2]], ptr [[TMP3]], align 8
253; AUTO_VEC-NEXT:    store <4 x double> [[STEP_ADD_3]], ptr [[TMP4]], align 8
254; AUTO_VEC-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
255; AUTO_VEC-NEXT:    [[VEC_IND_NEXT]] = fadd fast <4 x double> [[VEC_IND]], splat (double 4.800000e+01)
256; AUTO_VEC-NEXT:    [[TMP5:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
257; AUTO_VEC-NEXT:    br i1 [[TMP5]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP7:![0-9]+]]
258; AUTO_VEC:       middle.block:
259; AUTO_VEC-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[SMAX]], [[N_VEC]]
260; AUTO_VEC-NEXT:    [[IND_ESCAPE:%.*]] = fadd fast double [[TMP0]], -3.000000e+00
261; AUTO_VEC-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[FOR_BODY]]
262; AUTO_VEC:       for.body:
263; AUTO_VEC-NEXT:    [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ 0, [[ENTRY:%.*]] ], [ [[N_VEC]], [[MIDDLE_BLOCK]] ]
264; AUTO_VEC-NEXT:    [[J:%.*]] = phi double [ [[J_NEXT:%.*]], [[FOR_BODY]] ], [ 0.000000e+00, [[ENTRY]] ], [ [[TMP0]], [[MIDDLE_BLOCK]] ]
265; AUTO_VEC-NEXT:    [[T0:%.*]] = getelementptr double, ptr [[A]], i64 [[I]]
266; AUTO_VEC-NEXT:    store double [[J]], ptr [[T0]], align 8
267; AUTO_VEC-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 1
268; AUTO_VEC-NEXT:    [[J_NEXT]] = fadd fast double [[J]], 3.000000e+00
269; AUTO_VEC-NEXT:    [[EXITCOND_NOT:%.*]] = icmp eq i64 [[I_NEXT]], [[SMAX]]
270; AUTO_VEC-NEXT:    br i1 [[EXITCOND_NOT]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]]
271; AUTO_VEC:       for.end:
272; AUTO_VEC-NEXT:    [[J_LCSSA:%.*]] = phi double [ [[IND_ESCAPE]], [[MIDDLE_BLOCK]] ], [ [[J]], [[FOR_BODY]] ]
273; AUTO_VEC-NEXT:    ret double [[J_LCSSA]]
274;
275entry:
276  br label %for.body
277
278for.body:
279  %i = phi i64 [ 0, %entry ], [%i.next, %for.body]
280  %j = phi double [ 0.0, %entry ], [ %j.next, %for.body ]
281  %t0 = getelementptr double, ptr %a, i64 %i
282  store double %j, ptr %t0
283  %i.next = add i64 %i, 1
284  %j.next = fadd fast double %j, 3.0
285  %cond = icmp slt i64 %i.next, %n
286  br i1 %cond, label %for.body, label %for.end
287
288for.end:
289  %t1 = phi double [ %j, %for.body ]
290  ret double %t1
291}
292
293define double @external_use_without_fast_math(ptr %a, i64 %n) {
294; AUTO_VEC-LABEL: @external_use_without_fast_math(
295; AUTO_VEC-NEXT:  entry:
296; AUTO_VEC-NEXT:    [[SMAX:%.*]] = tail call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 1)
297; AUTO_VEC-NEXT:    [[XTRAITER:%.*]] = and i64 [[SMAX]], 7
298; AUTO_VEC-NEXT:    [[TMP0:%.*]] = icmp slt i64 [[N]], 8
299; AUTO_VEC-NEXT:    br i1 [[TMP0]], label [[FOR_END_UNR_LCSSA:%.*]], label [[ENTRY_NEW:%.*]]
300; AUTO_VEC:       entry.new:
301; AUTO_VEC-NEXT:    [[UNROLL_ITER:%.*]] = and i64 [[SMAX]], 9223372036854775800
302; AUTO_VEC-NEXT:    br label [[FOR_BODY:%.*]]
303; AUTO_VEC:       for.body:
304; AUTO_VEC-NEXT:    [[I:%.*]] = phi i64 [ 0, [[ENTRY_NEW]] ], [ [[I_NEXT_7:%.*]], [[FOR_BODY]] ]
305; AUTO_VEC-NEXT:    [[J:%.*]] = phi double [ 0.000000e+00, [[ENTRY_NEW]] ], [ [[J_NEXT_7:%.*]], [[FOR_BODY]] ]
306; AUTO_VEC-NEXT:    [[NITER:%.*]] = phi i64 [ 0, [[ENTRY_NEW]] ], [ [[NITER_NEXT_7:%.*]], [[FOR_BODY]] ]
307; AUTO_VEC-NEXT:    [[T0:%.*]] = getelementptr double, ptr [[A:%.*]], i64 [[I]]
308; AUTO_VEC-NEXT:    store double [[J]], ptr [[T0]], align 8
309; AUTO_VEC-NEXT:    [[I_NEXT:%.*]] = or disjoint i64 [[I]], 1
310; AUTO_VEC-NEXT:    [[J_NEXT:%.*]] = fadd double [[J]], 3.000000e+00
311; AUTO_VEC-NEXT:    [[T0_1:%.*]] = getelementptr double, ptr [[A]], i64 [[I_NEXT]]
312; AUTO_VEC-NEXT:    store double [[J_NEXT]], ptr [[T0_1]], align 8
313; AUTO_VEC-NEXT:    [[I_NEXT_1:%.*]] = or disjoint i64 [[I]], 2
314; AUTO_VEC-NEXT:    [[J_NEXT_1:%.*]] = fadd double [[J_NEXT]], 3.000000e+00
315; AUTO_VEC-NEXT:    [[T0_2:%.*]] = getelementptr double, ptr [[A]], i64 [[I_NEXT_1]]
316; AUTO_VEC-NEXT:    store double [[J_NEXT_1]], ptr [[T0_2]], align 8
317; AUTO_VEC-NEXT:    [[I_NEXT_2:%.*]] = or disjoint i64 [[I]], 3
318; AUTO_VEC-NEXT:    [[J_NEXT_2:%.*]] = fadd double [[J_NEXT_1]], 3.000000e+00
319; AUTO_VEC-NEXT:    [[T0_3:%.*]] = getelementptr double, ptr [[A]], i64 [[I_NEXT_2]]
320; AUTO_VEC-NEXT:    store double [[J_NEXT_2]], ptr [[T0_3]], align 8
321; AUTO_VEC-NEXT:    [[I_NEXT_3:%.*]] = or disjoint i64 [[I]], 4
322; AUTO_VEC-NEXT:    [[J_NEXT_3:%.*]] = fadd double [[J_NEXT_2]], 3.000000e+00
323; AUTO_VEC-NEXT:    [[T0_4:%.*]] = getelementptr double, ptr [[A]], i64 [[I_NEXT_3]]
324; AUTO_VEC-NEXT:    store double [[J_NEXT_3]], ptr [[T0_4]], align 8
325; AUTO_VEC-NEXT:    [[I_NEXT_4:%.*]] = or disjoint i64 [[I]], 5
326; AUTO_VEC-NEXT:    [[J_NEXT_4:%.*]] = fadd double [[J_NEXT_3]], 3.000000e+00
327; AUTO_VEC-NEXT:    [[T0_5:%.*]] = getelementptr double, ptr [[A]], i64 [[I_NEXT_4]]
328; AUTO_VEC-NEXT:    store double [[J_NEXT_4]], ptr [[T0_5]], align 8
329; AUTO_VEC-NEXT:    [[I_NEXT_5:%.*]] = or disjoint i64 [[I]], 6
330; AUTO_VEC-NEXT:    [[J_NEXT_5:%.*]] = fadd double [[J_NEXT_4]], 3.000000e+00
331; AUTO_VEC-NEXT:    [[T0_6:%.*]] = getelementptr double, ptr [[A]], i64 [[I_NEXT_5]]
332; AUTO_VEC-NEXT:    store double [[J_NEXT_5]], ptr [[T0_6]], align 8
333; AUTO_VEC-NEXT:    [[I_NEXT_6:%.*]] = or disjoint i64 [[I]], 7
334; AUTO_VEC-NEXT:    [[J_NEXT_6:%.*]] = fadd double [[J_NEXT_5]], 3.000000e+00
335; AUTO_VEC-NEXT:    [[T0_7:%.*]] = getelementptr double, ptr [[A]], i64 [[I_NEXT_6]]
336; AUTO_VEC-NEXT:    store double [[J_NEXT_6]], ptr [[T0_7]], align 8
337; AUTO_VEC-NEXT:    [[I_NEXT_7]] = add nuw nsw i64 [[I]], 8
338; AUTO_VEC-NEXT:    [[J_NEXT_7]] = fadd double [[J_NEXT_6]], 3.000000e+00
339; AUTO_VEC-NEXT:    [[NITER_NEXT_7]] = add i64 [[NITER]], 8
340; AUTO_VEC-NEXT:    [[NITER_NCMP_7:%.*]] = icmp eq i64 [[NITER_NEXT_7]], [[UNROLL_ITER]]
341; AUTO_VEC-NEXT:    br i1 [[NITER_NCMP_7]], label [[FOR_END_UNR_LCSSA]], label [[FOR_BODY]]
342; AUTO_VEC:       for.end.unr-lcssa:
343; AUTO_VEC-NEXT:    [[J_LCSSA_PH:%.*]] = phi double [ poison, [[ENTRY:%.*]] ], [ [[J_NEXT_6]], [[FOR_BODY]] ]
344; AUTO_VEC-NEXT:    [[I_UNR:%.*]] = phi i64 [ 0, [[ENTRY]] ], [ [[I_NEXT_7]], [[FOR_BODY]] ]
345; AUTO_VEC-NEXT:    [[J_UNR:%.*]] = phi double [ 0.000000e+00, [[ENTRY]] ], [ [[J_NEXT_7]], [[FOR_BODY]] ]
346; AUTO_VEC-NEXT:    [[LCMP_MOD_NOT:%.*]] = icmp eq i64 [[XTRAITER]], 0
347; AUTO_VEC-NEXT:    br i1 [[LCMP_MOD_NOT]], label [[FOR_END:%.*]], label [[FOR_BODY_EPIL:%.*]]
348; AUTO_VEC:       for.body.epil:
349; AUTO_VEC-NEXT:    [[I_EPIL:%.*]] = phi i64 [ [[I_NEXT_EPIL:%.*]], [[FOR_BODY_EPIL]] ], [ [[I_UNR]], [[FOR_END_UNR_LCSSA]] ]
350; AUTO_VEC-NEXT:    [[J_EPIL:%.*]] = phi double [ [[J_NEXT_EPIL:%.*]], [[FOR_BODY_EPIL]] ], [ [[J_UNR]], [[FOR_END_UNR_LCSSA]] ]
351; AUTO_VEC-NEXT:    [[EPIL_ITER:%.*]] = phi i64 [ [[EPIL_ITER_NEXT:%.*]], [[FOR_BODY_EPIL]] ], [ 0, [[FOR_END_UNR_LCSSA]] ]
352; AUTO_VEC-NEXT:    [[T0_EPIL:%.*]] = getelementptr double, ptr [[A]], i64 [[I_EPIL]]
353; AUTO_VEC-NEXT:    store double [[J_EPIL]], ptr [[T0_EPIL]], align 8
354; AUTO_VEC-NEXT:    [[I_NEXT_EPIL]] = add nuw nsw i64 [[I_EPIL]], 1
355; AUTO_VEC-NEXT:    [[J_NEXT_EPIL]] = fadd double [[J_EPIL]], 3.000000e+00
356; AUTO_VEC-NEXT:    [[EPIL_ITER_NEXT]] = add i64 [[EPIL_ITER]], 1
357; AUTO_VEC-NEXT:    [[EPIL_ITER_CMP_NOT:%.*]] = icmp eq i64 [[EPIL_ITER_NEXT]], [[XTRAITER]]
358; AUTO_VEC-NEXT:    br i1 [[EPIL_ITER_CMP_NOT]], label [[FOR_END]], label [[FOR_BODY_EPIL]], !llvm.loop [[LOOP9:![0-9]+]]
359; AUTO_VEC:       for.end:
360; AUTO_VEC-NEXT:    [[J_LCSSA:%.*]] = phi double [ [[J_LCSSA_PH]], [[FOR_END_UNR_LCSSA]] ], [ [[J_EPIL]], [[FOR_BODY_EPIL]] ]
361; AUTO_VEC-NEXT:    ret double [[J_LCSSA]]
362;
363entry:
364  br label %for.body
365
366for.body:
367  %i = phi i64 [ 0, %entry ], [%i.next, %for.body]
368  %j = phi double [ 0.0, %entry ], [ %j.next, %for.body ]
369  %t0 = getelementptr double, ptr %a, i64 %i
370  store double %j, ptr %t0
371  %i.next = add i64 %i, 1
372  %j.next = fadd double %j, 3.0
373  %cond = icmp slt i64 %i.next, %n
374  br i1 %cond, label %for.body, label %for.end
375
376for.end:
377  %t1 = phi double [ %j, %for.body ]
378  ret double %t1
379}
380
381;;  void fadd_induction(ptr p, unsigned N) {
382;;    float x = 1.0f;
383;;    for (unsigned i=0; i!=N; ++i) {
384;;      p[i] = p[i] + x;
385;;      x += 42.0f;
386;;    }
387;;  }
388
389define void @fadd_reassoc_FMF(ptr nocapture %p, i32 %N) {
390; AUTO_VEC-LABEL: @fadd_reassoc_FMF(
391; AUTO_VEC-NEXT:  entry:
392; AUTO_VEC-NEXT:    [[CMP_NOT11:%.*]] = icmp eq i32 [[N:%.*]], 0
393; AUTO_VEC-NEXT:    br i1 [[CMP_NOT11]], label [[FOR_COND_CLEANUP:%.*]], label [[ITER_CHECK:%.*]]
394; AUTO_VEC:       iter.check:
395; AUTO_VEC-NEXT:    [[TMP0:%.*]] = zext i32 [[N]] to i64
396; AUTO_VEC-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[N]], 4
397; AUTO_VEC-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[FOR_BODY:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]]
398; AUTO_VEC:       vector.main.loop.iter.check:
399; AUTO_VEC-NEXT:    [[MIN_ITERS_CHECK1:%.*]] = icmp ult i32 [[N]], 32
400; AUTO_VEC-NEXT:    br i1 [[MIN_ITERS_CHECK1]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]]
401; AUTO_VEC:       vector.ph:
402; AUTO_VEC-NEXT:    [[N_VEC:%.*]] = and i64 [[TMP0]], 4294967264
403; AUTO_VEC-NEXT:    [[DOTCAST:%.*]] = uitofp nneg i64 [[N_VEC]] to float
404; AUTO_VEC-NEXT:    [[TMP1:%.*]] = fmul reassoc float [[DOTCAST]], 4.200000e+01
405; AUTO_VEC-NEXT:    [[IND_END:%.*]] = fadd reassoc float [[TMP1]], 1.000000e+00
406; AUTO_VEC-NEXT:    br label [[VECTOR_BODY:%.*]]
407; AUTO_VEC:       vector.body:
408; AUTO_VEC-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
409; AUTO_VEC-NEXT:    [[VEC_IND:%.*]] = phi <8 x float> [ <float 1.000000e+00, float 4.300000e+01, float 8.500000e+01, float 1.270000e+02, float 1.690000e+02, float 2.110000e+02, float 2.530000e+02, float 2.950000e+02>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
410; AUTO_VEC-NEXT:    [[STEP_ADD:%.*]] = fadd reassoc <8 x float> [[VEC_IND]], splat (float 3.360000e+02)
411; AUTO_VEC-NEXT:    [[STEP_ADD2:%.*]] = fadd reassoc <8 x float> [[STEP_ADD]], splat (float 3.360000e+02)
412; AUTO_VEC-NEXT:    [[STEP_ADD3:%.*]] = fadd reassoc <8 x float> [[STEP_ADD2]], splat (float 3.360000e+02)
413; AUTO_VEC-NEXT:    [[TMP2:%.*]] = getelementptr inbounds nuw float, ptr [[P:%.*]], i64 [[INDEX]]
414; AUTO_VEC-NEXT:    [[TMP3:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP2]], i64 32
415; AUTO_VEC-NEXT:    [[TMP4:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP2]], i64 64
416; AUTO_VEC-NEXT:    [[TMP5:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP2]], i64 96
417; AUTO_VEC-NEXT:    [[WIDE_LOAD:%.*]] = load <8 x float>, ptr [[TMP2]], align 4
418; AUTO_VEC-NEXT:    [[WIDE_LOAD2:%.*]] = load <8 x float>, ptr [[TMP3]], align 4
419; AUTO_VEC-NEXT:    [[WIDE_LOAD3:%.*]] = load <8 x float>, ptr [[TMP4]], align 4
420; AUTO_VEC-NEXT:    [[WIDE_LOAD4:%.*]] = load <8 x float>, ptr [[TMP5]], align 4
421; AUTO_VEC-NEXT:    [[TMP6:%.*]] = fadd reassoc <8 x float> [[VEC_IND]], [[WIDE_LOAD]]
422; AUTO_VEC-NEXT:    [[TMP7:%.*]] = fadd reassoc <8 x float> [[STEP_ADD]], [[WIDE_LOAD2]]
423; AUTO_VEC-NEXT:    [[TMP8:%.*]] = fadd reassoc <8 x float> [[STEP_ADD2]], [[WIDE_LOAD3]]
424; AUTO_VEC-NEXT:    [[TMP9:%.*]] = fadd reassoc <8 x float> [[STEP_ADD3]], [[WIDE_LOAD4]]
425; AUTO_VEC-NEXT:    store <8 x float> [[TMP6]], ptr [[TMP2]], align 4
426; AUTO_VEC-NEXT:    store <8 x float> [[TMP7]], ptr [[TMP3]], align 4
427; AUTO_VEC-NEXT:    store <8 x float> [[TMP8]], ptr [[TMP4]], align 4
428; AUTO_VEC-NEXT:    store <8 x float> [[TMP9]], ptr [[TMP5]], align 4
429; AUTO_VEC-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32
430; AUTO_VEC-NEXT:    [[VEC_IND_NEXT]] = fadd reassoc <8 x float> [[STEP_ADD3]], splat (float 3.360000e+02)
431; AUTO_VEC-NEXT:    [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
432; AUTO_VEC-NEXT:    br i1 [[TMP10]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]]
433; AUTO_VEC:       middle.block:
434; AUTO_VEC-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N_VEC]], [[TMP0]]
435; AUTO_VEC-NEXT:    br i1 [[CMP_N]], label [[FOR_COND_CLEANUP]], label [[VEC_EPILOG_ITER_CHECK:%.*]]
436; AUTO_VEC:       vec.epilog.iter.check:
437; AUTO_VEC-NEXT:    [[DOTCAST10:%.*]] = uitofp nneg i64 [[N_VEC]] to float
438; AUTO_VEC-NEXT:    [[TMP11:%.*]] = fmul reassoc float [[DOTCAST10]], 4.200000e+01
439; AUTO_VEC-NEXT:    [[IND_END11:%.*]] = fadd reassoc float [[TMP11]], 1.000000e+00
440; AUTO_VEC-NEXT:    [[N_VEC_REMAINING:%.*]] = and i64 [[TMP0]], 28
441; AUTO_VEC-NEXT:    [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp eq i64 [[N_VEC_REMAINING]], 0
442; AUTO_VEC-NEXT:    br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[FOR_BODY]], label [[VEC_EPILOG_PH]]
443; AUTO_VEC:       vec.epilog.ph:
444; AUTO_VEC-NEXT:    [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
445; AUTO_VEC-NEXT:    [[BC_RESUME_VAL:%.*]] = phi float [ [[IND_END]], [[VEC_EPILOG_ITER_CHECK]] ], [ 1.000000e+00, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
446; AUTO_VEC-NEXT:    [[N_VEC6:%.*]] = and i64 [[TMP0]], 4294967292
447; AUTO_VEC-NEXT:    [[DOTCAST8:%.*]] = uitofp nneg i64 [[N_VEC6]] to float
448; AUTO_VEC-NEXT:    [[TMP12:%.*]] = fmul reassoc float [[DOTCAST8]], 4.200000e+01
449; AUTO_VEC-NEXT:    [[IND_END9:%.*]] = fadd reassoc float [[TMP12]], 1.000000e+00
450; AUTO_VEC-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <4 x float> poison, float [[BC_RESUME_VAL]], i64 0
451; AUTO_VEC-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <4 x float> [[DOTSPLATINSERT]], <4 x float> poison, <4 x i32> zeroinitializer
452; AUTO_VEC-NEXT:    [[INDUCTION:%.*]] = fadd reassoc <4 x float> [[DOTSPLAT]], <float 0.000000e+00, float 4.200000e+01, float 8.400000e+01, float 1.260000e+02>
453; AUTO_VEC-NEXT:    br label [[VEC_EPILOG_VECTOR_BODY:%.*]]
454; AUTO_VEC:       vec.epilog.vector.body:
455; AUTO_VEC-NEXT:    [[INDEX13:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT17:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]
456; AUTO_VEC-NEXT:    [[VEC_IND14:%.*]] = phi <4 x float> [ [[INDUCTION]], [[VEC_EPILOG_PH]] ], [ [[VEC_IND_NEXT15:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]
457; AUTO_VEC-NEXT:    [[TMP13:%.*]] = getelementptr inbounds nuw float, ptr [[P]], i64 [[INDEX13]]
458; AUTO_VEC-NEXT:    [[WIDE_LOAD16:%.*]] = load <4 x float>, ptr [[TMP13]], align 4
459; AUTO_VEC-NEXT:    [[TMP14:%.*]] = fadd reassoc <4 x float> [[VEC_IND14]], [[WIDE_LOAD16]]
460; AUTO_VEC-NEXT:    store <4 x float> [[TMP14]], ptr [[TMP13]], align 4
461; AUTO_VEC-NEXT:    [[INDEX_NEXT17]] = add nuw i64 [[INDEX13]], 4
462; AUTO_VEC-NEXT:    [[VEC_IND_NEXT15]] = fadd reassoc <4 x float> [[VEC_IND14]], splat (float 1.680000e+02)
463; AUTO_VEC-NEXT:    [[TMP15:%.*]] = icmp eq i64 [[INDEX_NEXT17]], [[N_VEC6]]
464; AUTO_VEC-NEXT:    br i1 [[TMP15]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP11:![0-9]+]]
465; AUTO_VEC:       vec.epilog.middle.block:
466; AUTO_VEC-NEXT:    [[CMP_N18:%.*]] = icmp eq i64 [[N_VEC6]], [[TMP0]]
467; AUTO_VEC-NEXT:    br i1 [[CMP_N18]], label [[FOR_COND_CLEANUP]], label [[FOR_BODY]]
468; AUTO_VEC:       for.cond.cleanup:
469; AUTO_VEC-NEXT:    ret void
470; AUTO_VEC:       for.body:
471; AUTO_VEC-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[ITER_CHECK]] ], [ [[N_VEC6]], [[VEC_EPILOG_MIDDLE_BLOCK]] ]
472; AUTO_VEC-NEXT:    [[X_012:%.*]] = phi float [ [[ADD3:%.*]], [[FOR_BODY]] ], [ [[IND_END11]], [[VEC_EPILOG_ITER_CHECK]] ], [ 1.000000e+00, [[ITER_CHECK]] ], [ [[IND_END9]], [[VEC_EPILOG_MIDDLE_BLOCK]] ]
473; AUTO_VEC-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds nuw float, ptr [[P]], i64 [[INDVARS_IV]]
474; AUTO_VEC-NEXT:    [[TMP16:%.*]] = load float, ptr [[ARRAYIDX]], align 4
475; AUTO_VEC-NEXT:    [[ADD:%.*]] = fadd reassoc float [[X_012]], [[TMP16]]
476; AUTO_VEC-NEXT:    store float [[ADD]], ptr [[ARRAYIDX]], align 4
477; AUTO_VEC-NEXT:    [[ADD3]] = fadd reassoc float [[X_012]], 4.200000e+01
478; AUTO_VEC-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
479; AUTO_VEC-NEXT:    [[CMP_NOT:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], [[TMP0]]
480; AUTO_VEC-NEXT:    br i1 [[CMP_NOT]], label [[FOR_COND_CLEANUP]], label [[FOR_BODY]], !llvm.loop [[LOOP12:![0-9]+]]
481;
482entry:
483  %cmp.not11 = icmp eq i32 %N, 0
484  br i1 %cmp.not11, label %for.cond.cleanup, label %for.body.preheader
485
486for.body.preheader:
487  %0 = zext i32 %N to i64
488  br label %for.body
489
490for.cond.cleanup:
491  ret void
492
493for.body:
494  %indvars.iv = phi i64 [ 0, %for.body.preheader ], [ %indvars.iv.next, %for.body ]
495  %x.012 = phi float [ 1.000000e+00, %for.body.preheader ], [ %add3, %for.body ]
496  %arrayidx = getelementptr inbounds float, ptr %p, i64 %indvars.iv
497  %1 = load float, ptr %arrayidx, align 4
498  %add = fadd reassoc float %x.012, %1
499  store float %add, ptr %arrayidx, align 4
500  %add3 = fadd reassoc float %x.012, 4.200000e+01
501  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
502  %cmp.not = icmp eq i64 %indvars.iv.next, %0
503  br i1 %cmp.not, label %for.cond.cleanup, label %for.body
504}
505