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 -instcombine -simplifycfg -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(
17 ; VEC4_INTERL1: vector.ph:
18 ; VEC4_INTERL1: [[DOTSPLATINSERT:%.*]] = insertelement <4 x float> undef, float %init, i32 0
19 ; VEC4_INTERL1-NEXT: [[DOTSPLAT:%.*]] = shufflevector <4 x float> [[DOTSPLATINSERT]], <4 x float> undef, <4 x i32> zeroinitializer
20 ; VEC4_INTERL1-NEXT: [[DOTSPLATINSERT2:%.*]] = insertelement <4 x float> undef, float %fpinc, i32 0
21 ; VEC4_INTERL1-NEXT: [[DOTSPLAT3:%.*]] = shufflevector <4 x float> [[DOTSPLATINSERT2]], <4 x float> undef, <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> undef, float [[TMP6]], i32 0
26 ; VEC4_INTERL1-NEXT: [[DOTSPLAT6:%.*]] = shufflevector <4 x float> [[DOTSPLATINSERT5]], <4 x float> undef, <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 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(
39 ; VEC4_INTERL2: vector.ph:
40 ; VEC4_INTERL2: [[DOTSPLATINSERT:%.*]] = insertelement <4 x float> undef, float %init, i32 0
41 ; VEC4_INTERL2-NEXT: [[DOTSPLAT:%.*]] = shufflevector <4 x float> [[DOTSPLATINSERT]], <4 x float> undef, <4 x i32> zeroinitializer
42 ; VEC4_INTERL2-NEXT: [[DOTSPLATINSERT3:%.*]] = insertelement <4 x float> undef, float %fpinc, i32 0
43 ; VEC4_INTERL2-NEXT: [[DOTSPLAT4:%.*]] = shufflevector <4 x float> [[DOTSPLATINSERT3]], <4 x float> undef, <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> undef, float [[TMP6]], i32 0
48 ; VEC4_INTERL2-NEXT: [[DOTSPLAT7:%.*]] = shufflevector <4 x float> [[DOTSPLATINSERT6]], <4 x float> undef, <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 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(
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 i64 [[INDEX]], 2
79 ; VEC1_INTERL2: br i1 {{.*}}, label %middle.block, label %vector.body
81 define void @fp_iv_loop1(float %init, float* noalias nocapture %A, i32 %N) #1 {
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 ;void fp_iv_loop2(float init, float * __restrict__ A, int N) {
110 ; for (int i=0; i < N; ++i) {
116 ; VEC4_INTERL1-LABEL: @fp_iv_loop2(
117 ; VEC4_INTERL1: vector.ph:
118 ; VEC4_INTERL1: [[DOTSPLATINSERT:%.*]] = insertelement <4 x float> undef, float %init, i32 0
119 ; VEC4_INTERL1-NEXT: [[DOTSPLAT:%.*]] = shufflevector <4 x float> [[DOTSPLATINSERT]], <4 x float> undef, <4 x i32> zeroinitializer
120 ; 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>
121 ; VEC4_INTERL1-NEXT: br label %vector.body
122 ; VEC4_INTERL1: vector.body:
123 ; VEC4_INTERL1-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %vector.ph ], [ [[INDEX_NEXT:%.*]], %vector.body ]
124 ; VEC4_INTERL1-NEXT: [[VEC_IND:%.*]] = phi <4 x float> [ [[INDUCTION2]], %vector.ph ], [ [[VEC_IND_NEXT:%.*]], %vector.body ]
125 ; VEC4_INTERL1-NEXT: [[TMP7:%.*]] = getelementptr inbounds float, float* %A, i64 [[INDEX]]
126 ; VEC4_INTERL1-NEXT: [[TMP8:%.*]] = bitcast float* [[TMP7]] to <4 x float>*
127 ; VEC4_INTERL1-NEXT: store <4 x float> [[VEC_IND]], <4 x float>* [[TMP8]], align 4
128 ; VEC4_INTERL1-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 4
129 ; 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>
130 ; VEC4_INTERL1: br i1 {{.*}}, label %middle.block, label %vector.body
132 define void @fp_iv_loop2(float %init, float* noalias nocapture %A, i32 %N) #0 {
134 %cmp4 = icmp sgt i32 %N, 0
135 br i1 %cmp4, label %for.body.preheader, label %for.end
137 for.body.preheader: ; preds = %entry
140 for.body: ; preds = %for.body.preheader, %for.body
141 %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %for.body.preheader ]
142 %x.06 = phi float [ %conv1, %for.body ], [ %init, %for.body.preheader ]
143 %arrayidx = getelementptr inbounds float, float* %A, i64 %indvars.iv
144 store float %x.06, float* %arrayidx, align 4
145 %conv1 = fadd fast float %x.06, 5.000000e-01
146 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
147 %lftr.wideiv = trunc i64 %indvars.iv.next to i32
148 %exitcond = icmp eq i32 %lftr.wideiv, %N
149 br i1 %exitcond, label %for.end.loopexit, label %for.body
151 for.end.loopexit: ; preds = %for.body
154 for.end: ; preds = %for.end.loopexit, %entry
158 ;void fp_iv_loop3(float init, float * __restrict__ A, float * __restrict__ B, float * __restrict__ C, int N) {
162 ; for (; i < N; ++i) {
171 ; VEC4_INTERL1-LABEL: @fp_iv_loop3(
172 ; VEC4_INTERL1: for.body.lr.ph:
173 ; VEC4_INTERL1: [[TMP0:%.*]] = load float, float* @fp_inc, align 4
174 ; VEC4_INTERL1: vector.ph:
175 ; VEC4_INTERL1: [[DOTSPLATINSERT:%.*]] = insertelement <4 x float> undef, float %init, i32 0
176 ; VEC4_INTERL1-NEXT: [[DOTSPLAT:%.*]] = shufflevector <4 x float> [[DOTSPLATINSERT]], <4 x float> undef, <4 x i32> zeroinitializer
177 ; VEC4_INTERL1-NEXT: [[DOTSPLATINSERT5:%.*]] = insertelement <4 x float> undef, float [[TMP0]], i32 0
178 ; VEC4_INTERL1-NEXT: [[DOTSPLAT6:%.*]] = shufflevector <4 x float> [[DOTSPLATINSERT5]], <4 x float> undef, <4 x i32> zeroinitializer
179 ; 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>
180 ; VEC4_INTERL1-NEXT: [[INDUCTION7:%.*]] = fadd fast <4 x float> [[DOTSPLAT]], [[TMP7]]
181 ; VEC4_INTERL1-NEXT: [[TMP8:%.*]] = fmul fast float [[TMP0]], 4.000000e+00
182 ; VEC4_INTERL1-NEXT: [[DOTSPLATINSERT8:%.*]] = insertelement <4 x float> undef, float [[TMP8]], i32 0
183 ; VEC4_INTERL1-NEXT: [[DOTSPLAT9:%.*]] = shufflevector <4 x float> [[DOTSPLATINSERT8]], <4 x float> undef, <4 x i32> zeroinitializer
184 ; VEC4_INTERL1-NEXT: [[BROADCAST_SPLATINSERT12:%.*]] = insertelement <4 x float> undef, float [[TMP0]], i32 0
185 ; VEC4_INTERL1-NEXT: [[BROADCAST_SPLAT13:%.*]] = shufflevector <4 x float> [[BROADCAST_SPLATINSERT12]], <4 x float> undef, <4 x i32> zeroinitializer
186 ; VEC4_INTERL1-NEXT: br label [[VECTOR_BODY:%.*]]
187 ; VEC4_INTERL1: vector.body:
188 ; VEC4_INTERL1-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %vector.ph ], [ [[INDEX_NEXT:%.*]], %vector.body ]
189 ; VEC4_INTERL1-NEXT: [[VEC_IND:%.*]] = phi <4 x float> [ <float 0x3FB99999A0000000, float 0xBFD99999A0000000, float 0xBFECCCCCC0000000, float 0xBFF6666660000000>, %vector.ph ], [ [[VEC_IND_NEXT:%.*]], %vector.body ]
190 ; VEC4_INTERL1-NEXT: [[VEC_IND10:%.*]] = phi <4 x float> [ [[INDUCTION7]], %vector.ph ], [ [[VEC_IND_NEXT11:%.*]], %vector.body ]
191 ; VEC4_INTERL1-NEXT: [[TMP12:%.*]] = getelementptr inbounds float, float* %A, i64 [[INDEX]]
192 ; VEC4_INTERL1-NEXT: [[TMP13:%.*]] = bitcast float* [[TMP12]] to <4 x float>*
193 ; VEC4_INTERL1-NEXT: store <4 x float> [[VEC_IND10]], <4 x float>* [[TMP13]], align 4
194 ; VEC4_INTERL1-NEXT: [[TMP14:%.*]] = fadd fast <4 x float> [[VEC_IND10]], [[BROADCAST_SPLAT13]]
195 ; 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>
196 ; VEC4_INTERL1-NEXT: [[TMP16:%.*]] = fadd fast <4 x float> [[TMP15]], [[TMP14]]
197 ; VEC4_INTERL1-NEXT: [[TMP17:%.*]] = getelementptr inbounds float, float* %B, i64 [[INDEX]]
198 ; VEC4_INTERL1-NEXT: [[TMP18:%.*]] = bitcast float* [[TMP17]] to <4 x float>*
199 ; VEC4_INTERL1-NEXT: store <4 x float> [[TMP16]], <4 x float>* [[TMP18]], align 4
200 ; VEC4_INTERL1-NEXT: [[TMP19:%.*]] = getelementptr inbounds float, float* %C, i64 [[INDEX]]
201 ; VEC4_INTERL1-NEXT: [[TMP20:%.*]] = bitcast float* [[TMP19]] to <4 x float>*
202 ; VEC4_INTERL1-NEXT: store <4 x float> [[TMP15]], <4 x float>* [[TMP20]], align 4
203 ; VEC4_INTERL1-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 4
204 ; 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>
205 ; VEC4_INTERL1-NEXT: [[VEC_IND_NEXT11]] = fadd fast <4 x float> [[VEC_IND10]], [[DOTSPLAT9]]
206 ; VEC4_INTERL1: br i1 {{.*}}, label %middle.block, label %vector.body
208 define void @fp_iv_loop3(float %init, float* noalias nocapture %A, float* noalias nocapture %B, float* noalias nocapture %C, i32 %N) #1 {
210 %cmp9 = icmp sgt i32 %N, 0
211 br i1 %cmp9, label %for.body.lr.ph, label %for.end
213 for.body.lr.ph: ; preds = %entry
214 %0 = load float, float* @fp_inc, align 4
217 for.body: ; preds = %for.body, %for.body.lr.ph
218 %indvars.iv = phi i64 [ 0, %for.body.lr.ph ], [ %indvars.iv.next, %for.body ]
219 %y.012 = phi float [ 0x3FB99999A0000000, %for.body.lr.ph ], [ %conv1, %for.body ]
220 %x.011 = phi float [ %init, %for.body.lr.ph ], [ %add, %for.body ]
221 %arrayidx = getelementptr inbounds float, float* %A, i64 %indvars.iv
222 store float %x.011, float* %arrayidx, align 4
223 %add = fadd fast float %x.011, %0
224 %conv1 = fadd fast float %y.012, -5.000000e-01
225 %add2 = fadd fast float %conv1, %add
226 %arrayidx4 = getelementptr inbounds float, float* %B, i64 %indvars.iv
227 store float %add2, float* %arrayidx4, align 4
228 %arrayidx6 = getelementptr inbounds float, float* %C, i64 %indvars.iv
229 store float %conv1, float* %arrayidx6, align 4
230 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
231 %lftr.wideiv = trunc i64 %indvars.iv.next to i32
232 %exitcond = icmp eq i32 %lftr.wideiv, %N
233 br i1 %exitcond, label %for.end.loopexit, label %for.body
242 ; Start and step values are constants. There is no 'fmul' operation in this case
243 ;void fp_iv_loop4(float * __restrict__ A, int N) {
245 ; for (int i=0; i < N; ++i) {
251 ; VEC4_INTERL1-LABEL: @fp_iv_loop4(
252 ; VEC4_INTERL1: vector.ph:
253 ; VEC4_INTERL1: br label %vector.body
254 ; VEC4_INTERL1: vector.body:
255 ; VEC4_INTERL1-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %vector.ph ], [ [[INDEX_NEXT:%.*]], %vector.body ]
256 ; 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 ]
257 ; VEC4_INTERL1-NEXT: [[TMP7:%.*]] = getelementptr inbounds float, float* %A, i64 [[INDEX]]
258 ; VEC4_INTERL1-NEXT: [[TMP8:%.*]] = bitcast float* [[TMP7]] to <4 x float>*
259 ; VEC4_INTERL1-NEXT: store <4 x float> [[VEC_IND]], <4 x float>* [[TMP8]], align 4
260 ; VEC4_INTERL1-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 4
261 ; 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>
262 ; VEC4_INTERL1: br i1 {{.*}}, label %middle.block, label %vector.body
264 define void @fp_iv_loop4(float* noalias nocapture %A, i32 %N) {
266 %cmp4 = icmp sgt i32 %N, 0
267 br i1 %cmp4, label %for.body.preheader, label %for.end
269 for.body.preheader: ; preds = %entry
272 for.body: ; preds = %for.body.preheader, %for.body
273 %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %for.body.preheader ]
274 %x.06 = phi float [ %conv1, %for.body ], [ 1.000000e+00, %for.body.preheader ]
275 %arrayidx = getelementptr inbounds float, float* %A, i64 %indvars.iv
276 store float %x.06, float* %arrayidx, align 4
277 %conv1 = fadd fast float %x.06, 5.000000e-01
278 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
279 %lftr.wideiv = trunc i64 %indvars.iv.next to i32
280 %exitcond = icmp eq i32 %lftr.wideiv, %N
281 br i1 %exitcond, label %for.end.loopexit, label %for.body
283 for.end.loopexit: ; preds = %for.body
286 for.end: ; preds = %for.end.loopexit, %entry
290 ; VEC2_INTERL1_PRED_STORE-LABEL: @non_primary_iv_float_scalar(
291 ; VEC2_INTERL1_PRED_STORE: vector.body:
292 ; VEC2_INTERL1_PRED_STORE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %vector.ph ], [ [[INDEX_NEXT:%.*]], %[[PRED_STORE_CONTINUE7:.*]] ]
293 ; VEC2_INTERL1_PRED_STORE-NEXT: [[TMP1:%.*]] = sitofp i64 [[INDEX]] to float
294 ; VEC2_INTERL1_PRED_STORE-NEXT: [[TMP2:%.*]] = getelementptr inbounds float, float* %A, i64 [[INDEX]]
295 ; VEC2_INTERL1_PRED_STORE-NEXT: [[TMP3:%.*]] = bitcast float* [[TMP2]] to <2 x float>*
296 ; VEC2_INTERL1_PRED_STORE-NEXT: [[WIDE_LOAD:%.*]] = load <2 x float>, <2 x float>* [[TMP3]], align 4
297 ; VEC2_INTERL1_PRED_STORE-NEXT: [[TMP4:%.*]] = fcmp fast oeq <2 x float> [[WIDE_LOAD]], zeroinitializer
298 ; VEC2_INTERL1_PRED_STORE-NEXT: [[TMP5:%.*]] = extractelement <2 x i1> [[TMP4]], i32 0
299 ; VEC2_INTERL1_PRED_STORE-NEXT: br i1 [[TMP5]], label %[[PRED_STORE_IF:.*]], label %[[PRED_STORE_CONTINUE:.*]]
300 ; VEC2_INTERL1_PRED_STORE: [[PRED_STORE_IF]]:
301 ; VEC2_INTERL1_PRED_STORE-NEXT: store float [[TMP1]], float* [[TMP2]], align 4
302 ; VEC2_INTERL1_PRED_STORE-NEXT: br label %[[PRED_STORE_CONTINUE]]
303 ; VEC2_INTERL1_PRED_STORE: [[PRED_STORE_CONTINUE]]:
304 ; VEC2_INTERL1_PRED_STORE-NEXT: [[TMP8:%.*]] = extractelement <2 x i1> [[TMP4]], i32 1
305 ; VEC2_INTERL1_PRED_STORE-NEXT: br i1 [[TMP8]], label %[[PRED_STORE_IF6:.*]], label %[[PRED_STORE_CONTINUE7]]
306 ; VEC2_INTERL1_PRED_STORE: [[PRED_STORE_IF6]]:
307 ; VEC2_INTERL1_PRED_STORE-NEXT: [[TMP9:%.*]] = fadd fast float [[TMP1]], 1.000000e+00
308 ; VEC2_INTERL1_PRED_STORE-NEXT: [[TMP10:%.*]] = or i64 [[INDEX]], 1
309 ; VEC2_INTERL1_PRED_STORE-NEXT: [[TMP11:%.*]] = getelementptr inbounds float, float* %A, i64 [[TMP10]]
310 ; VEC2_INTERL1_PRED_STORE-NEXT: store float [[TMP9]], float* [[TMP11]], align 4
311 ; VEC2_INTERL1_PRED_STORE-NEXT: br label %[[PRED_STORE_CONTINUE7]]
312 ; VEC2_INTERL1_PRED_STORE: [[PRED_STORE_CONTINUE7]]:
313 ; VEC2_INTERL1_PRED_STORE-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 2
314 ; VEC2_INTERL1_PRED_STORE: br i1 {{.*}}, label %middle.block, label %vector.body
316 define void @non_primary_iv_float_scalar(float* %A, i64 %N) {
321 %i = phi i64 [ %i.next, %for.inc ], [ 0, %entry ]
322 %j = phi float [ %j.next, %for.inc ], [ 0.0, %entry ]
323 %tmp0 = getelementptr inbounds float, float* %A, i64 %i
324 %tmp1 = load float, float* %tmp0, align 4
325 %tmp2 = fcmp fast oeq float %tmp1, 0.0
326 br i1 %tmp2, label %if.pred, label %for.inc
329 store float %j, float* %tmp0, align 4
333 %i.next = add nuw nsw i64 %i, 1
334 %j.next = fadd fast float %j, 1.0
335 %cond = icmp slt i64 %i.next, %N
336 br i1 %cond, label %for.body, label %for.end