1 ; RUN: opt < %s -loop-vectorize -force-vector-interleave=1 -force-vector-width=4 -dce -instcombine -S | FileCheck --check-prefix VEC4_INTERL1 %s
2 ; RUN: opt < %s -loop-vectorize -force-vector-interleave=2 -force-vector-width=4 -dce -instcombine -S | FileCheck --check-prefix VEC4_INTERL2 %s
3 ; RUN: opt < %s -loop-vectorize -force-vector-interleave=2 -force-vector-width=1 -dce -instcombine -S | FileCheck --check-prefix VEC1_INTERL2 %s
4 ; RUN: opt < %s -loop-vectorize -force-vector-interleave=1 -force-vector-width=2 -dce -simplifycfg -simplifycfg-require-and-preserve-domtree=1 -instcombine -simplifycfg -simplifycfg-require-and-preserve-domtree=1 -keep-loops=false -S | FileCheck --check-prefix VEC2_INTERL1_PRED_STORE %s
6 @fp_inc = common global float 0.000000e+00, align 4
8 ;void fp_iv_loop1(float init, float * __restrict__ A, int N) {
10 ; for (int i=0; i < N; ++i) {
16 ; VEC4_INTERL1-LABEL: @fp_iv_loop1_fast_FMF(
17 ; VEC4_INTERL1: vector.ph:
18 ; VEC4_INTERL1: [[DOTSPLATINSERT:%.*]] = insertelement <4 x float> poison, float %init, i32 0
19 ; VEC4_INTERL1-NEXT: [[DOTSPLAT:%.*]] = shufflevector <4 x float> [[DOTSPLATINSERT]], <4 x float> poison, <4 x i32> zeroinitializer
20 ; VEC4_INTERL1-NEXT: [[DOTSPLATINSERT2:%.*]] = insertelement <4 x float> poison, float %fpinc, i32 0
21 ; VEC4_INTERL1-NEXT: [[DOTSPLAT3:%.*]] = shufflevector <4 x float> [[DOTSPLATINSERT2]], <4 x float> poison, <4 x i32> zeroinitializer
22 ; VEC4_INTERL1-NEXT: [[TMP5:%.*]] = fmul fast <4 x float> [[DOTSPLAT3]], <float 0.000000e+00, float 1.000000e+00, float 2.000000e+00, float 3.000000e+00>
23 ; VEC4_INTERL1-NEXT: [[INDUCTION4:%.*]] = fsub fast <4 x float> [[DOTSPLAT]], [[TMP5]]
24 ; VEC4_INTERL1-NEXT: [[TMP6:%.*]] = fmul fast float %fpinc, 4.000000e+00
25 ; VEC4_INTERL1-NEXT: [[DOTSPLATINSERT5:%.*]] = insertelement <4 x float> poison, float [[TMP6]], i32 0
26 ; VEC4_INTERL1-NEXT: [[DOTSPLAT6:%.*]] = shufflevector <4 x float> [[DOTSPLATINSERT5]], <4 x float> poison, <4 x i32> zeroinitializer
27 ; VEC4_INTERL1-NEXT: br label %vector.body
28 ; VEC4_INTERL1: vector.body:
29 ; VEC4_INTERL1-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %vector.ph ], [ [[INDEX_NEXT:%.*]], %vector.body ]
30 ; VEC4_INTERL1-NEXT: [[VEC_IND:%.*]] = phi <4 x float> [ [[INDUCTION4]], %vector.ph ], [ [[VEC_IND_NEXT:%.*]], %vector.body ]
31 ; VEC4_INTERL1-NEXT: [[TMP8:%.*]] = getelementptr inbounds float, float* %A, i64 [[INDEX]]
32 ; VEC4_INTERL1-NEXT: [[TMP9:%.*]] = bitcast float* [[TMP8]] to <4 x float>*
33 ; VEC4_INTERL1-NEXT: store <4 x float> [[VEC_IND]], <4 x float>* [[TMP9]], align 4
34 ; VEC4_INTERL1-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
35 ; VEC4_INTERL1-NEXT: [[VEC_IND_NEXT]] = fsub fast <4 x float> [[VEC_IND]], [[DOTSPLAT6]]
36 ; VEC4_INTERL1: br i1 {{.*}}, label %middle.block, label %vector.body
38 ; VEC4_INTERL2-LABEL: @fp_iv_loop1_fast_FMF(
39 ; VEC4_INTERL2: vector.ph:
40 ; VEC4_INTERL2: [[DOTSPLATINSERT:%.*]] = insertelement <4 x float> poison, float %init, i32 0
41 ; VEC4_INTERL2-NEXT: [[DOTSPLAT:%.*]] = shufflevector <4 x float> [[DOTSPLATINSERT]], <4 x float> poison, <4 x i32> zeroinitializer
42 ; VEC4_INTERL2-NEXT: [[DOTSPLATINSERT3:%.*]] = insertelement <4 x float> poison, float %fpinc, i32 0
43 ; VEC4_INTERL2-NEXT: [[DOTSPLAT4:%.*]] = shufflevector <4 x float> [[DOTSPLATINSERT3]], <4 x float> poison, <4 x i32> zeroinitializer
44 ; VEC4_INTERL2-NEXT: [[TMP5:%.*]] = fmul fast <4 x float> [[DOTSPLAT4]], <float 0.000000e+00, float 1.000000e+00, float 2.000000e+00, float 3.000000e+00>
45 ; VEC4_INTERL2-NEXT: [[INDUCTION5:%.*]] = fsub fast <4 x float> [[DOTSPLAT]], [[TMP5]]
46 ; VEC4_INTERL2-NEXT: [[TMP6:%.*]] = fmul fast float %fpinc, 4.000000e+00
47 ; VEC4_INTERL2-NEXT: [[DOTSPLATINSERT6:%.*]] = insertelement <4 x float> poison, float [[TMP6]], i32 0
48 ; VEC4_INTERL2-NEXT: [[DOTSPLAT7:%.*]] = shufflevector <4 x float> [[DOTSPLATINSERT6]], <4 x float> poison, <4 x i32> zeroinitializer
49 ; VEC4_INTERL2-NEXT: br label %vector.body
50 ; VEC4_INTERL2: vector.body:
51 ; VEC4_INTERL2-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %vector.ph ], [ [[INDEX_NEXT:%.*]], %vector.body ]
52 ; VEC4_INTERL2-NEXT: [[VEC_IND:%.*]] = phi <4 x float> [ [[INDUCTION5]], %vector.ph ], [ [[VEC_IND_NEXT:%.*]], %vector.body ]
53 ; VEC4_INTERL2-NEXT: [[STEP_ADD:%.*]] = fsub fast <4 x float> [[VEC_IND]], [[DOTSPLAT7]]
54 ; VEC4_INTERL2-NEXT: [[TMP9:%.*]] = getelementptr inbounds float, float* %A, i64 [[INDEX]]
55 ; VEC4_INTERL2-NEXT: [[TMP10:%.*]] = bitcast float* [[TMP9]] to <4 x float>*
56 ; VEC4_INTERL2-NEXT: store <4 x float> [[VEC_IND]], <4 x float>* [[TMP10]], align 4
57 ; VEC4_INTERL2-NEXT: [[TMP11:%.*]] = getelementptr inbounds float, float* [[TMP9]], i64 4
58 ; VEC4_INTERL2-NEXT: [[TMP12:%.*]] = bitcast float* [[TMP11]] to <4 x float>*
59 ; VEC4_INTERL2-NEXT: store <4 x float> [[STEP_ADD]], <4 x float>* [[TMP12]], align 4
60 ; VEC4_INTERL2-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8
61 ; VEC4_INTERL2-NEXT: [[VEC_IND_NEXT]] = fsub fast <4 x float> [[STEP_ADD]], [[DOTSPLAT7]]
62 ; VEC4_INTERL2: br i1 {{.*}}, label %middle.block, label %vector.body
64 ; VEC1_INTERL2-LABEL: @fp_iv_loop1_fast_FMF(
65 ; VEC1_INTERL2: vector.ph:
66 ; VEC1_INTERL2: br label %vector.body
67 ; VEC1_INTERL2: vector.body:
68 ; VEC1_INTERL2-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %vector.ph ], [ [[INDEX_NEXT:%.*]], %vector.body ]
69 ; VEC1_INTERL2-NEXT: [[INDUCTION2:%.*]] = or i64 [[INDEX]], 1
70 ; VEC1_INTERL2-NEXT: [[TMP6:%.*]] = sitofp i64 [[INDEX]] to float
71 ; VEC1_INTERL2-NEXT: [[TMP7:%.*]] = fmul fast float %fpinc, [[TMP6]]
72 ; VEC1_INTERL2-NEXT: [[FP_OFFSET_IDX:%.*]] = fsub fast float %init, [[TMP7]]
73 ; VEC1_INTERL2-NEXT: [[TMP8:%.*]] = fsub fast float [[FP_OFFSET_IDX]], %fpinc
74 ; VEC1_INTERL2-NEXT: [[TMP9:%.*]] = getelementptr inbounds float, float* %A, i64 [[INDEX]]
75 ; VEC1_INTERL2-NEXT: [[TMP10:%.*]] = getelementptr inbounds float, float* %A, i64 [[INDUCTION2]]
76 ; VEC1_INTERL2-NEXT: store float [[FP_OFFSET_IDX]], float* [[TMP9]], align 4
77 ; VEC1_INTERL2-NEXT: store float [[TMP8]], float* [[TMP10]], align 4
78 ; VEC1_INTERL2-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
79 ; VEC1_INTERL2: br i1 {{.*}}, label %middle.block, label %vector.body
81 define void @fp_iv_loop1_fast_FMF(float %init, float* noalias nocapture %A, i32 %N) {
83 %cmp4 = icmp sgt i32 %N, 0
84 br i1 %cmp4, label %for.body.lr.ph, label %for.end
86 for.body.lr.ph: ; preds = %entry
87 %fpinc = load float, float* @fp_inc, align 4
90 for.body: ; preds = %for.body, %for.body.lr.ph
91 %indvars.iv = phi i64 [ 0, %for.body.lr.ph ], [ %indvars.iv.next, %for.body ]
92 %x.05 = phi float [ %init, %for.body.lr.ph ], [ %add, %for.body ]
93 %arrayidx = getelementptr inbounds float, float* %A, i64 %indvars.iv
94 store float %x.05, float* %arrayidx, align 4
95 %add = fsub fast float %x.05, %fpinc
96 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
97 %lftr.wideiv = trunc i64 %indvars.iv.next to i32
98 %exitcond = icmp eq i32 %lftr.wideiv, %N
99 br i1 %exitcond, label %for.end.loopexit, label %for.body
101 for.end.loopexit: ; preds = %for.body
104 for.end: ; preds = %for.end.loopexit, %entry
108 ; We do not need the full 'fast' FMF to vectorize the loop, but the code can't become
109 ; 'fast' spontaneously - FMF should propagate from the original IR.
111 define void @fp_iv_loop1_reassoc_FMF(float %init, float* noalias nocapture %A, i32 %N) {
112 ; VEC4_INTERL1-LABEL: @fp_iv_loop1_reassoc_FMF(
113 ; VEC4_INTERL1-NEXT: entry:
114 ; VEC4_INTERL1-NEXT: [[CMP4:%.*]] = icmp sgt i32 [[N:%.*]], 0
115 ; VEC4_INTERL1-NEXT: br i1 [[CMP4]], label [[FOR_BODY_LR_PH:%.*]], label [[FOR_END:%.*]]
116 ; VEC4_INTERL1: for.body.lr.ph:
117 ; VEC4_INTERL1-NEXT: [[FPINC:%.*]] = load float, float* @fp_inc, align 4
118 ; VEC4_INTERL1-NEXT: [[TMP0:%.*]] = add i32 [[N]], -1
119 ; VEC4_INTERL1-NEXT: [[TMP1:%.*]] = zext i32 [[TMP0]] to i64
120 ; VEC4_INTERL1-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1
121 ; VEC4_INTERL1-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 3
122 ; VEC4_INTERL1-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
123 ; VEC4_INTERL1: vector.ph:
124 ; VEC4_INTERL1-NEXT: [[N_VEC:%.*]] = and i64 [[TMP2]], 8589934588
125 ; VEC4_INTERL1-NEXT: [[CAST_CRD:%.*]] = sitofp i64 [[N_VEC]] to float
126 ; VEC4_INTERL1-NEXT: [[TMP3:%.*]] = fmul reassoc float [[FPINC]], [[CAST_CRD]]
127 ; VEC4_INTERL1-NEXT: [[IND_END:%.*]] = fsub reassoc float [[INIT:%.*]], [[TMP3]]
128 ; VEC4_INTERL1-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <4 x float> poison, float [[INIT]], i32 0
129 ; VEC4_INTERL1-NEXT: [[DOTSPLAT:%.*]] = shufflevector <4 x float> [[DOTSPLATINSERT]], <4 x float> poison, <4 x i32> zeroinitializer
130 ; VEC4_INTERL1-NEXT: [[DOTSPLATINSERT2:%.*]] = insertelement <4 x float> poison, float [[FPINC]], i32 0
131 ; VEC4_INTERL1-NEXT: [[DOTSPLAT3:%.*]] = shufflevector <4 x float> [[DOTSPLATINSERT2]], <4 x float> poison, <4 x i32> zeroinitializer
132 ; VEC4_INTERL1-NEXT: [[TMP4:%.*]] = fmul reassoc <4 x float> [[DOTSPLAT3]], <float 0.000000e+00, float 1.000000e+00, float 2.000000e+00, float 3.000000e+00>
133 ; VEC4_INTERL1-NEXT: [[INDUCTION:%.*]] = fsub reassoc <4 x float> [[DOTSPLAT]], [[TMP4]]
134 ; VEC4_INTERL1-NEXT: [[TMP5:%.*]] = fmul reassoc float [[FPINC]], 4.000000e+00
135 ; VEC4_INTERL1-NEXT: [[DOTSPLATINSERT4:%.*]] = insertelement <4 x float> poison, float [[TMP5]], i32 0
136 ; VEC4_INTERL1-NEXT: [[DOTSPLAT5:%.*]] = shufflevector <4 x float> [[DOTSPLATINSERT4]], <4 x float> poison, <4 x i32> zeroinitializer
137 ; VEC4_INTERL1-NEXT: br label [[VECTOR_BODY:%.*]]
138 ; VEC4_INTERL1: vector.body:
139 ; VEC4_INTERL1-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
140 ; VEC4_INTERL1-NEXT: [[VEC_IND:%.*]] = phi <4 x float> [ [[INDUCTION]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
141 ; VEC4_INTERL1-NEXT: [[TMP6:%.*]] = getelementptr inbounds float, float* [[A:%.*]], i64 [[INDEX]]
142 ; VEC4_INTERL1-NEXT: [[TMP7:%.*]] = bitcast float* [[TMP6]] to <4 x float>*
143 ; VEC4_INTERL1-NEXT: store <4 x float> [[VEC_IND]], <4 x float>* [[TMP7]], align 4
144 ; VEC4_INTERL1-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
145 ; VEC4_INTERL1-NEXT: [[VEC_IND_NEXT]] = fsub reassoc <4 x float> [[VEC_IND]], [[DOTSPLAT5]]
146 ; VEC4_INTERL1-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
147 ; VEC4_INTERL1-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], [[LOOP4:!llvm.loop !.*]]
148 ; VEC4_INTERL1: middle.block:
149 ; VEC4_INTERL1-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]]
150 ; VEC4_INTERL1-NEXT: br i1 [[CMP_N]], label [[FOR_END_LOOPEXIT:%.*]], label [[SCALAR_PH]]
151 ; VEC4_INTERL1: scalar.ph:
152 ; VEC4_INTERL1-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_LR_PH]] ]
153 ; VEC4_INTERL1-NEXT: [[BC_RESUME_VAL1:%.*]] = phi float [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[INIT]], [[FOR_BODY_LR_PH]] ]
154 ; VEC4_INTERL1-NEXT: br label [[FOR_BODY:%.*]]
155 ; VEC4_INTERL1: for.body:
156 ; VEC4_INTERL1-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]
157 ; VEC4_INTERL1-NEXT: [[X_05:%.*]] = phi float [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ], [ [[ADD:%.*]], [[FOR_BODY]] ]
158 ; VEC4_INTERL1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[INDVARS_IV]]
159 ; VEC4_INTERL1-NEXT: store float [[X_05]], float* [[ARRAYIDX]], align 4
160 ; VEC4_INTERL1-NEXT: [[ADD]] = fsub reassoc float [[X_05]], [[FPINC]]
161 ; VEC4_INTERL1-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
162 ; VEC4_INTERL1-NEXT: [[LFTR_WIDEIV:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i32
163 ; VEC4_INTERL1-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[LFTR_WIDEIV]], [[N]]
164 ; VEC4_INTERL1-NEXT: br i1 [[EXITCOND]], label [[FOR_END_LOOPEXIT]], label [[FOR_BODY]], [[LOOP5:!llvm.loop !.*]]
165 ; VEC4_INTERL1: for.end.loopexit:
166 ; VEC4_INTERL1-NEXT: br label [[FOR_END]]
167 ; VEC4_INTERL1: for.end:
168 ; VEC4_INTERL1-NEXT: ret void
170 ; VEC4_INTERL2-LABEL: @fp_iv_loop1_reassoc_FMF(
171 ; VEC4_INTERL2-NEXT: entry:
172 ; VEC4_INTERL2-NEXT: [[CMP4:%.*]] = icmp sgt i32 [[N:%.*]], 0
173 ; VEC4_INTERL2-NEXT: br i1 [[CMP4]], label [[FOR_BODY_LR_PH:%.*]], label [[FOR_END:%.*]]
174 ; VEC4_INTERL2: for.body.lr.ph:
175 ; VEC4_INTERL2-NEXT: [[FPINC:%.*]] = load float, float* @fp_inc, align 4
176 ; VEC4_INTERL2-NEXT: [[TMP0:%.*]] = add i32 [[N]], -1
177 ; VEC4_INTERL2-NEXT: [[TMP1:%.*]] = zext i32 [[TMP0]] to i64
178 ; VEC4_INTERL2-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1
179 ; VEC4_INTERL2-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 7
180 ; VEC4_INTERL2-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
181 ; VEC4_INTERL2: vector.ph:
182 ; VEC4_INTERL2-NEXT: [[N_VEC:%.*]] = and i64 [[TMP2]], 8589934584
183 ; VEC4_INTERL2-NEXT: [[CAST_CRD:%.*]] = sitofp i64 [[N_VEC]] to float
184 ; VEC4_INTERL2-NEXT: [[TMP3:%.*]] = fmul reassoc float [[FPINC]], [[CAST_CRD]]
185 ; VEC4_INTERL2-NEXT: [[IND_END:%.*]] = fsub reassoc float [[INIT:%.*]], [[TMP3]]
186 ; VEC4_INTERL2-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <4 x float> poison, float [[INIT]], i32 0
187 ; VEC4_INTERL2-NEXT: [[DOTSPLAT:%.*]] = shufflevector <4 x float> [[DOTSPLATINSERT]], <4 x float> poison, <4 x i32> zeroinitializer
188 ; VEC4_INTERL2-NEXT: [[DOTSPLATINSERT2:%.*]] = insertelement <4 x float> poison, float [[FPINC]], i32 0
189 ; VEC4_INTERL2-NEXT: [[DOTSPLAT3:%.*]] = shufflevector <4 x float> [[DOTSPLATINSERT2]], <4 x float> poison, <4 x i32> zeroinitializer
190 ; VEC4_INTERL2-NEXT: [[TMP4:%.*]] = fmul reassoc <4 x float> [[DOTSPLAT3]], <float 0.000000e+00, float 1.000000e+00, float 2.000000e+00, float 3.000000e+00>
191 ; VEC4_INTERL2-NEXT: [[INDUCTION:%.*]] = fsub reassoc <4 x float> [[DOTSPLAT]], [[TMP4]]
192 ; VEC4_INTERL2-NEXT: [[TMP5:%.*]] = fmul reassoc float [[FPINC]], 4.000000e+00
193 ; VEC4_INTERL2-NEXT: [[DOTSPLATINSERT4:%.*]] = insertelement <4 x float> poison, float [[TMP5]], i32 0
194 ; VEC4_INTERL2-NEXT: [[DOTSPLAT5:%.*]] = shufflevector <4 x float> [[DOTSPLATINSERT4]], <4 x float> poison, <4 x i32> zeroinitializer
195 ; VEC4_INTERL2-NEXT: br label [[VECTOR_BODY:%.*]]
196 ; VEC4_INTERL2: vector.body:
197 ; VEC4_INTERL2-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
198 ; VEC4_INTERL2-NEXT: [[VEC_IND:%.*]] = phi <4 x float> [ [[INDUCTION]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
199 ; VEC4_INTERL2-NEXT: [[STEP_ADD:%.*]] = fsub reassoc <4 x float> [[VEC_IND]], [[DOTSPLAT5]]
200 ; VEC4_INTERL2-NEXT: [[TMP6:%.*]] = getelementptr inbounds float, float* [[A:%.*]], i64 [[INDEX]]
201 ; VEC4_INTERL2-NEXT: [[TMP7:%.*]] = bitcast float* [[TMP6]] to <4 x float>*
202 ; VEC4_INTERL2-NEXT: store <4 x float> [[VEC_IND]], <4 x float>* [[TMP7]], align 4
203 ; VEC4_INTERL2-NEXT: [[TMP8:%.*]] = getelementptr inbounds float, float* [[TMP6]], i64 4
204 ; VEC4_INTERL2-NEXT: [[TMP9:%.*]] = bitcast float* [[TMP8]] to <4 x float>*
205 ; VEC4_INTERL2-NEXT: store <4 x float> [[STEP_ADD]], <4 x float>* [[TMP9]], align 4
206 ; VEC4_INTERL2-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8
207 ; VEC4_INTERL2-NEXT: [[VEC_IND_NEXT]] = fsub reassoc <4 x float> [[STEP_ADD]], [[DOTSPLAT5]]
208 ; VEC4_INTERL2-NEXT: [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
209 ; VEC4_INTERL2-NEXT: br i1 [[TMP10]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], [[LOOP4:!llvm.loop !.*]]
210 ; VEC4_INTERL2: middle.block:
211 ; VEC4_INTERL2-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]]
212 ; VEC4_INTERL2-NEXT: br i1 [[CMP_N]], label [[FOR_END_LOOPEXIT:%.*]], label [[SCALAR_PH]]
213 ; VEC4_INTERL2: scalar.ph:
214 ; VEC4_INTERL2-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_LR_PH]] ]
215 ; VEC4_INTERL2-NEXT: [[BC_RESUME_VAL1:%.*]] = phi float [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[INIT]], [[FOR_BODY_LR_PH]] ]
216 ; VEC4_INTERL2-NEXT: br label [[FOR_BODY:%.*]]
217 ; VEC4_INTERL2: for.body:
218 ; VEC4_INTERL2-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]
219 ; VEC4_INTERL2-NEXT: [[X_05:%.*]] = phi float [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ], [ [[ADD:%.*]], [[FOR_BODY]] ]
220 ; VEC4_INTERL2-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[INDVARS_IV]]
221 ; VEC4_INTERL2-NEXT: store float [[X_05]], float* [[ARRAYIDX]], align 4
222 ; VEC4_INTERL2-NEXT: [[ADD]] = fsub reassoc float [[X_05]], [[FPINC]]
223 ; VEC4_INTERL2-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
224 ; VEC4_INTERL2-NEXT: [[LFTR_WIDEIV:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i32
225 ; VEC4_INTERL2-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[LFTR_WIDEIV]], [[N]]
226 ; VEC4_INTERL2-NEXT: br i1 [[EXITCOND]], label [[FOR_END_LOOPEXIT]], label [[FOR_BODY]], [[LOOP5:!llvm.loop !.*]]
227 ; VEC4_INTERL2: for.end.loopexit:
228 ; VEC4_INTERL2-NEXT: br label [[FOR_END]]
229 ; VEC4_INTERL2: for.end:
230 ; VEC4_INTERL2-NEXT: ret void
232 ; VEC1_INTERL2-LABEL: @fp_iv_loop1_reassoc_FMF(
233 ; VEC1_INTERL2-NEXT: entry:
234 ; VEC1_INTERL2-NEXT: [[CMP4:%.*]] = icmp sgt i32 [[N:%.*]], 0
235 ; VEC1_INTERL2-NEXT: br i1 [[CMP4]], label [[FOR_BODY_LR_PH:%.*]], label [[FOR_END:%.*]]
236 ; VEC1_INTERL2: for.body.lr.ph:
237 ; VEC1_INTERL2-NEXT: [[FPINC:%.*]] = load float, float* @fp_inc, align 4
238 ; VEC1_INTERL2-NEXT: [[TMP0:%.*]] = add i32 [[N]], -1
239 ; VEC1_INTERL2-NEXT: [[TMP1:%.*]] = zext i32 [[TMP0]] to i64
240 ; VEC1_INTERL2-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1
241 ; VEC1_INTERL2-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp eq i32 [[TMP0]], 0
242 ; VEC1_INTERL2-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
243 ; VEC1_INTERL2: vector.ph:
244 ; VEC1_INTERL2-NEXT: [[N_VEC:%.*]] = and i64 [[TMP2]], 8589934590
245 ; VEC1_INTERL2-NEXT: [[CAST_CRD:%.*]] = sitofp i64 [[N_VEC]] to float
246 ; VEC1_INTERL2-NEXT: [[TMP3:%.*]] = fmul reassoc float [[FPINC]], [[CAST_CRD]]
247 ; VEC1_INTERL2-NEXT: [[IND_END:%.*]] = fsub reassoc float [[INIT:%.*]], [[TMP3]]
248 ; VEC1_INTERL2-NEXT: br label [[VECTOR_BODY:%.*]]
249 ; VEC1_INTERL2: vector.body:
250 ; VEC1_INTERL2-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
251 ; VEC1_INTERL2-NEXT: [[INDUCTION2:%.*]] = or i64 [[INDEX]], 1
252 ; VEC1_INTERL2-NEXT: [[TMP4:%.*]] = sitofp i64 [[INDEX]] to float
253 ; VEC1_INTERL2-NEXT: [[TMP5:%.*]] = fmul reassoc float [[FPINC]], [[TMP4]]
254 ; VEC1_INTERL2-NEXT: [[OFFSET_IDX:%.*]] = fsub reassoc float [[INIT]], [[TMP5]]
255 ; VEC1_INTERL2-NEXT: [[TMP6:%.*]] = fmul reassoc float [[FPINC]], 0.000000e+00
256 ; VEC1_INTERL2-NEXT: [[TMP7:%.*]] = fsub reassoc float [[OFFSET_IDX]], [[TMP6]]
257 ; VEC1_INTERL2-NEXT: [[TMP8:%.*]] = fsub reassoc float [[OFFSET_IDX]], [[FPINC]]
258 ; VEC1_INTERL2-NEXT: [[TMP9:%.*]] = getelementptr inbounds float, float* [[A:%.*]], i64 [[INDEX]]
259 ; VEC1_INTERL2-NEXT: [[TMP10:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[INDUCTION2]]
260 ; VEC1_INTERL2-NEXT: store float [[TMP7]], float* [[TMP9]], align 4
261 ; VEC1_INTERL2-NEXT: store float [[TMP8]], float* [[TMP10]], align 4
262 ; VEC1_INTERL2-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
263 ; VEC1_INTERL2-NEXT: [[TMP11:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
264 ; VEC1_INTERL2-NEXT: br i1 [[TMP11]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
265 ; VEC1_INTERL2: middle.block:
266 ; VEC1_INTERL2-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]]
267 ; VEC1_INTERL2-NEXT: br i1 [[CMP_N]], label [[FOR_END_LOOPEXIT:%.*]], label [[SCALAR_PH]]
268 ; VEC1_INTERL2: scalar.ph:
269 ; VEC1_INTERL2-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_LR_PH]] ]
270 ; VEC1_INTERL2-NEXT: [[BC_RESUME_VAL1:%.*]] = phi float [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[INIT]], [[FOR_BODY_LR_PH]] ]
271 ; VEC1_INTERL2-NEXT: br label [[FOR_BODY:%.*]]
272 ; VEC1_INTERL2: for.body:
273 ; VEC1_INTERL2-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]
274 ; VEC1_INTERL2-NEXT: [[X_05:%.*]] = phi float [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ], [ [[ADD:%.*]], [[FOR_BODY]] ]
275 ; VEC1_INTERL2-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[A:%.*]], i64 [[INDVARS_IV]]
276 ; VEC1_INTERL2-NEXT: store float [[X_05]], float* [[ARRAYIDX]], align 4
277 ; VEC1_INTERL2-NEXT: [[ADD]] = fsub reassoc float [[X_05]], [[FPINC]]
278 ; VEC1_INTERL2-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
279 ; VEC1_INTERL2-NEXT: [[LFTR_WIDEIV:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i32
280 ; VEC1_INTERL2-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[LFTR_WIDEIV]], [[N]]
281 ; VEC1_INTERL2-NEXT: br i1 [[EXITCOND]], label [[FOR_END_LOOPEXIT:%.*]], label [[FOR_BODY]]
282 ; VEC1_INTERL2: for.end.loopexit:
283 ; VEC1_INTERL2-NEXT: br label [[FOR_END]]
284 ; VEC1_INTERL2: for.end:
285 ; VEC1_INTERL2-NEXT: ret void
287 ; VEC2_INTERL1_PRED_STORE-LABEL: @fp_iv_loop1_reassoc_FMF(
288 ; VEC2_INTERL1_PRED_STORE-NEXT: entry:
289 ; VEC2_INTERL1_PRED_STORE-NEXT: [[CMP4:%.*]] = icmp sgt i32 [[N:%.*]], 0
290 ; VEC2_INTERL1_PRED_STORE-NEXT: br i1 [[CMP4]], label [[FOR_BODY_LR_PH:%.*]], label [[FOR_END:%.*]]
291 ; VEC2_INTERL1_PRED_STORE: for.body.lr.ph:
292 ; VEC2_INTERL1_PRED_STORE-NEXT: [[FPINC:%.*]] = load float, float* @fp_inc, align 4
293 ; VEC2_INTERL1_PRED_STORE-NEXT: [[TMP0:%.*]] = add i32 [[N]], -1
294 ; VEC2_INTERL1_PRED_STORE-NEXT: [[TMP1:%.*]] = zext i32 [[TMP0]] to i64
295 ; VEC2_INTERL1_PRED_STORE-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1
296 ; VEC2_INTERL1_PRED_STORE-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp eq i32 [[TMP0]], 0
297 ; VEC2_INTERL1_PRED_STORE-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[FOR_BODY:%.*]], label [[VECTOR_PH:%.*]]
298 ; VEC2_INTERL1_PRED_STORE: vector.ph:
299 ; VEC2_INTERL1_PRED_STORE-NEXT: [[N_VEC:%.*]] = and i64 [[TMP2]], 8589934590
300 ; VEC2_INTERL1_PRED_STORE-NEXT: [[CAST_CRD:%.*]] = sitofp i64 [[N_VEC]] to float
301 ; VEC2_INTERL1_PRED_STORE-NEXT: [[TMP3:%.*]] = fmul reassoc float [[FPINC]], [[CAST_CRD]]
302 ; VEC2_INTERL1_PRED_STORE-NEXT: [[IND_END:%.*]] = fsub reassoc float [[INIT:%.*]], [[TMP3]]
303 ; VEC2_INTERL1_PRED_STORE-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <2 x float> poison, float [[INIT]], i32 0
304 ; VEC2_INTERL1_PRED_STORE-NEXT: [[DOTSPLAT:%.*]] = shufflevector <2 x float> [[DOTSPLATINSERT]], <2 x float> poison, <2 x i32> zeroinitializer
305 ; VEC2_INTERL1_PRED_STORE-NEXT: [[DOTSPLATINSERT2:%.*]] = insertelement <2 x float> poison, float [[FPINC]], i32 0
306 ; VEC2_INTERL1_PRED_STORE-NEXT: [[DOTSPLAT3:%.*]] = shufflevector <2 x float> [[DOTSPLATINSERT2]], <2 x float> poison, <2 x i32> zeroinitializer
307 ; VEC2_INTERL1_PRED_STORE-NEXT: [[TMP4:%.*]] = fmul reassoc <2 x float> [[DOTSPLAT3]], <float 0.000000e+00, float 1.000000e+00>
308 ; VEC2_INTERL1_PRED_STORE-NEXT: [[INDUCTION:%.*]] = fsub reassoc <2 x float> [[DOTSPLAT]], [[TMP4]]
309 ; VEC2_INTERL1_PRED_STORE-NEXT: [[TMP5:%.*]] = fmul reassoc float [[FPINC]], 2.000000e+00
310 ; VEC2_INTERL1_PRED_STORE-NEXT: [[DOTSPLATINSERT4:%.*]] = insertelement <2 x float> poison, float [[TMP5]], i32 0
311 ; VEC2_INTERL1_PRED_STORE-NEXT: [[DOTSPLAT5:%.*]] = shufflevector <2 x float> [[DOTSPLATINSERT4]], <2 x float> poison, <2 x i32> zeroinitializer
312 ; VEC2_INTERL1_PRED_STORE-NEXT: br label [[VECTOR_BODY:%.*]]
313 ; VEC2_INTERL1_PRED_STORE: vector.body:
314 ; VEC2_INTERL1_PRED_STORE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
315 ; VEC2_INTERL1_PRED_STORE-NEXT: [[VEC_IND:%.*]] = phi <2 x float> [ [[INDUCTION]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
316 ; VEC2_INTERL1_PRED_STORE-NEXT: [[TMP6:%.*]] = getelementptr inbounds float, float* [[A:%.*]], i64 [[INDEX]]
317 ; VEC2_INTERL1_PRED_STORE-NEXT: [[TMP7:%.*]] = bitcast float* [[TMP6]] to <2 x float>*
318 ; VEC2_INTERL1_PRED_STORE-NEXT: store <2 x float> [[VEC_IND]], <2 x float>* [[TMP7]], align 4
319 ; VEC2_INTERL1_PRED_STORE-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
320 ; VEC2_INTERL1_PRED_STORE-NEXT: [[VEC_IND_NEXT]] = fsub reassoc <2 x float> [[VEC_IND]], [[DOTSPLAT5]]
321 ; VEC2_INTERL1_PRED_STORE-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
322 ; VEC2_INTERL1_PRED_STORE-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], [[LOOP4:!llvm.loop !.*]]
323 ; VEC2_INTERL1_PRED_STORE: middle.block:
324 ; VEC2_INTERL1_PRED_STORE-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]]
325 ; VEC2_INTERL1_PRED_STORE-NEXT: br i1 [[CMP_N]], label [[FOR_END]], label [[FOR_BODY]]
326 ; VEC2_INTERL1_PRED_STORE: for.body:
327 ; VEC2_INTERL1_PRED_STORE-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_LR_PH]] ]
328 ; VEC2_INTERL1_PRED_STORE-NEXT: [[X_05:%.*]] = phi float [ [[ADD:%.*]], [[FOR_BODY]] ], [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[INIT]], [[FOR_BODY_LR_PH]] ]
329 ; VEC2_INTERL1_PRED_STORE-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[INDVARS_IV]]
330 ; VEC2_INTERL1_PRED_STORE-NEXT: store float [[X_05]], float* [[ARRAYIDX]], align 4
331 ; VEC2_INTERL1_PRED_STORE-NEXT: [[ADD]] = fsub reassoc float [[X_05]], [[FPINC]]
332 ; VEC2_INTERL1_PRED_STORE-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
333 ; VEC2_INTERL1_PRED_STORE-NEXT: [[LFTR_WIDEIV:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i32
334 ; VEC2_INTERL1_PRED_STORE-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[LFTR_WIDEIV]], [[N]]
335 ; VEC2_INTERL1_PRED_STORE-NEXT: br i1 [[EXITCOND]], label [[FOR_END]], label [[FOR_BODY]], [[LOOP5:!llvm.loop !.*]]
336 ; VEC2_INTERL1_PRED_STORE: for.end:
337 ; VEC2_INTERL1_PRED_STORE-NEXT: ret void
340 %cmp4 = icmp sgt i32 %N, 0
341 br i1 %cmp4, label %for.body.lr.ph, label %for.end
343 for.body.lr.ph: ; preds = %entry
344 %fpinc = load float, float* @fp_inc, align 4
347 for.body: ; preds = %for.body, %for.body.lr.ph
348 %indvars.iv = phi i64 [ 0, %for.body.lr.ph ], [ %indvars.iv.next, %for.body ]
349 %x.05 = phi float [ %init, %for.body.lr.ph ], [ %add, %for.body ]
350 %arrayidx = getelementptr inbounds float, float* %A, i64 %indvars.iv
351 store float %x.05, float* %arrayidx, align 4
352 %add = fsub reassoc float %x.05, %fpinc
353 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
354 %lftr.wideiv = trunc i64 %indvars.iv.next to i32
355 %exitcond = icmp eq i32 %lftr.wideiv, %N
356 br i1 %exitcond, label %for.end.loopexit, label %for.body
358 for.end.loopexit: ; preds = %for.body
361 for.end: ; preds = %for.end.loopexit, %entry
365 ;void fp_iv_loop2(float init, float * __restrict__ A, int N) {
367 ; for (int i=0; i < N; ++i) {
373 ; VEC4_INTERL1-LABEL: @fp_iv_loop2(
374 ; VEC4_INTERL1: vector.ph:
375 ; VEC4_INTERL1: [[DOTSPLATINSERT:%.*]] = insertelement <4 x float> poison, float %init, i32 0
376 ; VEC4_INTERL1-NEXT: [[DOTSPLAT:%.*]] = shufflevector <4 x float> [[DOTSPLATINSERT]], <4 x float> poison, <4 x i32> zeroinitializer
377 ; VEC4_INTERL1-NEXT: [[INDUCTION2:%.*]] = fadd fast <4 x float> [[DOTSPLAT]], <float 0.000000e+00, float 5.000000e-01, float 1.000000e+00, float 1.500000e+00>
378 ; VEC4_INTERL1-NEXT: br label %vector.body
379 ; VEC4_INTERL1: vector.body:
380 ; VEC4_INTERL1-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %vector.ph ], [ [[INDEX_NEXT:%.*]], %vector.body ]
381 ; VEC4_INTERL1-NEXT: [[VEC_IND:%.*]] = phi <4 x float> [ [[INDUCTION2]], %vector.ph ], [ [[VEC_IND_NEXT:%.*]], %vector.body ]
382 ; VEC4_INTERL1-NEXT: [[TMP7:%.*]] = getelementptr inbounds float, float* %A, i64 [[INDEX]]
383 ; VEC4_INTERL1-NEXT: [[TMP8:%.*]] = bitcast float* [[TMP7]] to <4 x float>*
384 ; VEC4_INTERL1-NEXT: store <4 x float> [[VEC_IND]], <4 x float>* [[TMP8]], align 4
385 ; VEC4_INTERL1-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
386 ; VEC4_INTERL1-NEXT: [[VEC_IND_NEXT]] = fadd fast <4 x float> [[VEC_IND]], <float 2.000000e+00, float 2.000000e+00, float 2.000000e+00, float 2.000000e+00>
387 ; VEC4_INTERL1: br i1 {{.*}}, label %middle.block, label %vector.body
389 define void @fp_iv_loop2(float %init, float* noalias nocapture %A, i32 %N) #0 {
391 %cmp4 = icmp sgt i32 %N, 0
392 br i1 %cmp4, label %for.body.preheader, label %for.end
394 for.body.preheader: ; preds = %entry
397 for.body: ; preds = %for.body.preheader, %for.body
398 %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %for.body.preheader ]
399 %x.06 = phi float [ %conv1, %for.body ], [ %init, %for.body.preheader ]
400 %arrayidx = getelementptr inbounds float, float* %A, i64 %indvars.iv
401 store float %x.06, float* %arrayidx, align 4
402 %conv1 = fadd fast float %x.06, 5.000000e-01
403 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
404 %lftr.wideiv = trunc i64 %indvars.iv.next to i32
405 %exitcond = icmp eq i32 %lftr.wideiv, %N
406 br i1 %exitcond, label %for.end.loopexit, label %for.body
408 for.end.loopexit: ; preds = %for.body
411 for.end: ; preds = %for.end.loopexit, %entry
415 ;void fp_iv_loop3(float init, float * __restrict__ A, float * __restrict__ B, float * __restrict__ C, int N) {
419 ; for (; i < N; ++i) {
428 ; VEC4_INTERL1-LABEL: @fp_iv_loop3(
429 ; VEC4_INTERL1: for.body.lr.ph:
430 ; VEC4_INTERL1: [[TMP0:%.*]] = load float, float* @fp_inc, align 4
431 ; VEC4_INTERL1: vector.ph:
432 ; VEC4_INTERL1: [[DOTSPLATINSERT:%.*]] = insertelement <4 x float> poison, float %init, i32 0
433 ; VEC4_INTERL1-NEXT: [[DOTSPLAT:%.*]] = shufflevector <4 x float> [[DOTSPLATINSERT]], <4 x float> poison, <4 x i32> zeroinitializer
434 ; VEC4_INTERL1-NEXT: [[DOTSPLATINSERT5:%.*]] = insertelement <4 x float> poison, float [[TMP0]], i32 0
435 ; VEC4_INTERL1-NEXT: [[DOTSPLAT6:%.*]] = shufflevector <4 x float> [[DOTSPLATINSERT5]], <4 x float> poison, <4 x i32> zeroinitializer
436 ; VEC4_INTERL1-NEXT: [[TMP7:%.*]] = fmul fast <4 x float> [[DOTSPLAT6]], <float 0.000000e+00, float 1.000000e+00, float 2.000000e+00, float 3.000000e+00>
437 ; VEC4_INTERL1-NEXT: [[INDUCTION7:%.*]] = fadd fast <4 x float> [[DOTSPLAT]], [[TMP7]]
438 ; VEC4_INTERL1-NEXT: [[TMP8:%.*]] = fmul fast float [[TMP0]], 4.000000e+00
439 ; VEC4_INTERL1-NEXT: [[DOTSPLATINSERT8:%.*]] = insertelement <4 x float> poison, float [[TMP8]], i32 0
440 ; VEC4_INTERL1-NEXT: [[DOTSPLAT9:%.*]] = shufflevector <4 x float> [[DOTSPLATINSERT8]], <4 x float> poison, <4 x i32> zeroinitializer
441 ; VEC4_INTERL1-NEXT: [[BROADCAST_SPLATINSERT12:%.*]] = insertelement <4 x float> poison, float [[TMP0]], i32 0
442 ; VEC4_INTERL1-NEXT: [[BROADCAST_SPLAT13:%.*]] = shufflevector <4 x float> [[BROADCAST_SPLATINSERT12]], <4 x float> poison, <4 x i32> zeroinitializer
443 ; VEC4_INTERL1-NEXT: br label [[VECTOR_BODY:%.*]]
444 ; VEC4_INTERL1: vector.body:
445 ; VEC4_INTERL1-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %vector.ph ], [ [[INDEX_NEXT:%.*]], %vector.body ]
446 ; VEC4_INTERL1-NEXT: [[VEC_IND:%.*]] = phi <4 x float> [ <float 0x3FB99999A0000000, float 0xBFD99999A0000000, float 0xBFECCCCCC0000000, float 0xBFF6666660000000>, %vector.ph ], [ [[VEC_IND_NEXT:%.*]], %vector.body ]
447 ; VEC4_INTERL1-NEXT: [[VEC_IND10:%.*]] = phi <4 x float> [ [[INDUCTION7]], %vector.ph ], [ [[VEC_IND_NEXT11:%.*]], %vector.body ]
448 ; VEC4_INTERL1-NEXT: [[TMP12:%.*]] = getelementptr inbounds float, float* %A, i64 [[INDEX]]
449 ; VEC4_INTERL1-NEXT: [[TMP13:%.*]] = bitcast float* [[TMP12]] to <4 x float>*
450 ; VEC4_INTERL1-NEXT: store <4 x float> [[VEC_IND10]], <4 x float>* [[TMP13]], align 4
451 ; VEC4_INTERL1-NEXT: [[TMP14:%.*]] = fadd fast <4 x float> [[VEC_IND10]], [[BROADCAST_SPLAT13]]
452 ; VEC4_INTERL1-NEXT: [[TMP15:%.*]] = fadd fast <4 x float> [[VEC_IND]], <float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01>
453 ; VEC4_INTERL1-NEXT: [[TMP16:%.*]] = fadd fast <4 x float> [[TMP15]], [[TMP14]]
454 ; VEC4_INTERL1-NEXT: [[TMP17:%.*]] = getelementptr inbounds float, float* %B, i64 [[INDEX]]
455 ; VEC4_INTERL1-NEXT: [[TMP18:%.*]] = bitcast float* [[TMP17]] to <4 x float>*
456 ; VEC4_INTERL1-NEXT: store <4 x float> [[TMP16]], <4 x float>* [[TMP18]], align 4
457 ; VEC4_INTERL1-NEXT: [[TMP19:%.*]] = getelementptr inbounds float, float* %C, i64 [[INDEX]]
458 ; VEC4_INTERL1-NEXT: [[TMP20:%.*]] = bitcast float* [[TMP19]] to <4 x float>*
459 ; VEC4_INTERL1-NEXT: store <4 x float> [[TMP15]], <4 x float>* [[TMP20]], align 4
460 ; VEC4_INTERL1-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
461 ; VEC4_INTERL1-NEXT: [[VEC_IND_NEXT]] = fadd fast <4 x float> [[VEC_IND]], <float -2.000000e+00, float -2.000000e+00, float -2.000000e+00, float -2.000000e+00>
462 ; VEC4_INTERL1-NEXT: [[VEC_IND_NEXT11]] = fadd fast <4 x float> [[VEC_IND10]], [[DOTSPLAT9]]
463 ; VEC4_INTERL1: br i1 {{.*}}, label %middle.block, label %vector.body
465 define void @fp_iv_loop3(float %init, float* noalias nocapture %A, float* noalias nocapture %B, float* noalias nocapture %C, i32 %N) {
467 %cmp9 = icmp sgt i32 %N, 0
468 br i1 %cmp9, label %for.body.lr.ph, label %for.end
470 for.body.lr.ph: ; preds = %entry
471 %0 = load float, float* @fp_inc, align 4
474 for.body: ; preds = %for.body, %for.body.lr.ph
475 %indvars.iv = phi i64 [ 0, %for.body.lr.ph ], [ %indvars.iv.next, %for.body ]
476 %y.012 = phi float [ 0x3FB99999A0000000, %for.body.lr.ph ], [ %conv1, %for.body ]
477 %x.011 = phi float [ %init, %for.body.lr.ph ], [ %add, %for.body ]
478 %arrayidx = getelementptr inbounds float, float* %A, i64 %indvars.iv
479 store float %x.011, float* %arrayidx, align 4
480 %add = fadd fast float %x.011, %0
481 %conv1 = fadd fast float %y.012, -5.000000e-01
482 %add2 = fadd fast float %conv1, %add
483 %arrayidx4 = getelementptr inbounds float, float* %B, i64 %indvars.iv
484 store float %add2, float* %arrayidx4, align 4
485 %arrayidx6 = getelementptr inbounds float, float* %C, i64 %indvars.iv
486 store float %conv1, float* %arrayidx6, align 4
487 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
488 %lftr.wideiv = trunc i64 %indvars.iv.next to i32
489 %exitcond = icmp eq i32 %lftr.wideiv, %N
490 br i1 %exitcond, label %for.end.loopexit, label %for.body
499 ; Start and step values are constants. There is no 'fmul' operation in this case
500 ;void fp_iv_loop4(float * __restrict__ A, int N) {
502 ; for (int i=0; i < N; ++i) {
508 ; VEC4_INTERL1-LABEL: @fp_iv_loop4(
509 ; VEC4_INTERL1: vector.ph:
510 ; VEC4_INTERL1: br label %vector.body
511 ; VEC4_INTERL1: vector.body:
512 ; VEC4_INTERL1-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %vector.ph ], [ [[INDEX_NEXT:%.*]], %vector.body ]
513 ; VEC4_INTERL1-NEXT: [[VEC_IND:%.*]] = phi <4 x float> [ <float 1.000000e+00, float 1.500000e+00, float 2.000000e+00, float 2.500000e+00>, %vector.ph ], [ [[VEC_IND_NEXT:%.*]], %vector.body ]
514 ; VEC4_INTERL1-NEXT: [[TMP7:%.*]] = getelementptr inbounds float, float* %A, i64 [[INDEX]]
515 ; VEC4_INTERL1-NEXT: [[TMP8:%.*]] = bitcast float* [[TMP7]] to <4 x float>*
516 ; VEC4_INTERL1-NEXT: store <4 x float> [[VEC_IND]], <4 x float>* [[TMP8]], align 4
517 ; VEC4_INTERL1-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
518 ; VEC4_INTERL1-NEXT: [[VEC_IND_NEXT]] = fadd fast <4 x float> [[VEC_IND]], <float 2.000000e+00, float 2.000000e+00, float 2.000000e+00, float 2.000000e+00>
519 ; VEC4_INTERL1: br i1 {{.*}}, label %middle.block, label %vector.body
521 define void @fp_iv_loop4(float* noalias nocapture %A, i32 %N) {
523 %cmp4 = icmp sgt i32 %N, 0
524 br i1 %cmp4, label %for.body.preheader, label %for.end
526 for.body.preheader: ; preds = %entry
529 for.body: ; preds = %for.body.preheader, %for.body
530 %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %for.body.preheader ]
531 %x.06 = phi float [ %conv1, %for.body ], [ 1.000000e+00, %for.body.preheader ]
532 %arrayidx = getelementptr inbounds float, float* %A, i64 %indvars.iv
533 store float %x.06, float* %arrayidx, align 4
534 %conv1 = fadd fast float %x.06, 5.000000e-01
535 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
536 %lftr.wideiv = trunc i64 %indvars.iv.next to i32
537 %exitcond = icmp eq i32 %lftr.wideiv, %N
538 br i1 %exitcond, label %for.end.loopexit, label %for.body
540 for.end.loopexit: ; preds = %for.body
543 for.end: ; preds = %for.end.loopexit, %entry
547 ; VEC2_INTERL1_PRED_STORE-LABEL: @non_primary_iv_float_scalar(
548 ; VEC2_INTERL1_PRED_STORE: vector.body:
549 ; VEC2_INTERL1_PRED_STORE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %vector.ph ], [ [[INDEX_NEXT:%.*]], %[[PRED_STORE_CONTINUE7:.*]] ]
550 ; VEC2_INTERL1_PRED_STORE-NEXT: [[TMP1:%.*]] = sitofp i64 [[INDEX]] to float
551 ; VEC2_INTERL1_PRED_STORE-NEXT: [[TMP2:%.*]] = getelementptr inbounds float, float* %A, i64 [[INDEX]]
552 ; VEC2_INTERL1_PRED_STORE-NEXT: [[TMP3:%.*]] = bitcast float* [[TMP2]] to <2 x float>*
553 ; VEC2_INTERL1_PRED_STORE-NEXT: [[WIDE_LOAD:%.*]] = load <2 x float>, <2 x float>* [[TMP3]], align 4
554 ; VEC2_INTERL1_PRED_STORE-NEXT: [[TMP4:%.*]] = fcmp fast oeq <2 x float> [[WIDE_LOAD]], zeroinitializer
555 ; VEC2_INTERL1_PRED_STORE-NEXT: [[TMP5:%.*]] = extractelement <2 x i1> [[TMP4]], i32 0
556 ; VEC2_INTERL1_PRED_STORE-NEXT: br i1 [[TMP5]], label %[[PRED_STORE_IF:.*]], label %[[PRED_STORE_CONTINUE:.*]]
557 ; VEC2_INTERL1_PRED_STORE: [[PRED_STORE_IF]]:
558 ; VEC2_INTERL1_PRED_STORE-NEXT: store float [[TMP1]], float* [[TMP2]], align 4
559 ; VEC2_INTERL1_PRED_STORE-NEXT: br label %[[PRED_STORE_CONTINUE]]
560 ; VEC2_INTERL1_PRED_STORE: [[PRED_STORE_CONTINUE]]:
561 ; VEC2_INTERL1_PRED_STORE-NEXT: [[TMP8:%.*]] = extractelement <2 x i1> [[TMP4]], i32 1
562 ; VEC2_INTERL1_PRED_STORE-NEXT: br i1 [[TMP8]], label %[[PRED_STORE_IF6:.*]], label %[[PRED_STORE_CONTINUE7]]
563 ; VEC2_INTERL1_PRED_STORE: [[PRED_STORE_IF6]]:
564 ; VEC2_INTERL1_PRED_STORE-NEXT: [[TMP9:%.*]] = fadd fast float [[TMP1]], 1.000000e+00
565 ; VEC2_INTERL1_PRED_STORE-NEXT: [[TMP10:%.*]] = or i64 [[INDEX]], 1
566 ; VEC2_INTERL1_PRED_STORE-NEXT: [[TMP11:%.*]] = getelementptr inbounds float, float* %A, i64 [[TMP10]]
567 ; VEC2_INTERL1_PRED_STORE-NEXT: store float [[TMP9]], float* [[TMP11]], align 4
568 ; VEC2_INTERL1_PRED_STORE-NEXT: br label %[[PRED_STORE_CONTINUE7]]
569 ; VEC2_INTERL1_PRED_STORE: [[PRED_STORE_CONTINUE7]]:
570 ; VEC2_INTERL1_PRED_STORE-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
571 ; VEC2_INTERL1_PRED_STORE: br i1 {{.*}}, label %middle.block, label %vector.body
573 define void @non_primary_iv_float_scalar(float* %A, i64 %N) {
578 %i = phi i64 [ %i.next, %for.inc ], [ 0, %entry ]
579 %j = phi float [ %j.next, %for.inc ], [ 0.0, %entry ]
580 %tmp0 = getelementptr inbounds float, float* %A, i64 %i
581 %tmp1 = load float, float* %tmp0, align 4
582 %tmp2 = fcmp fast oeq float %tmp1, 0.0
583 br i1 %tmp2, label %if.pred, label %for.inc
586 store float %j, float* %tmp0, align 4
590 %i.next = add nuw nsw i64 %i, 1
591 %j.next = fadd fast float %j, 1.0
592 %cond = icmp slt i64 %i.next, %N
593 br i1 %cond, label %for.body, label %for.end