1 ; RUN: opt -S -loop-vectorize -force-vector-interleave=1 -force-vector-width=2 < %s 2>&1 | FileCheck %s
3 target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:32:32-n8:16:32-S128"
5 @f = common global i32 0, align 4
6 @.str = private unnamed_addr constant [4 x i8] c"%d\0A\00", align 1
7 @c = common global i32 0, align 4
8 @a = common global i32 0, align 4
9 @b = common global i32 0, align 4
10 @e = common global i32 0, align 4
12 ; It has a value that is used outside of the loop
13 ; and is not a recognized reduction variable "tmp17".
14 ; However, tmp17 is a non-header phi which is an allowed exit.
16 ; CHECK-LABEL: @test1(
17 ; CHECK: %vec.ind = phi <2 x i32>
18 ; CHECK: [[CMP:%[a-zA-Z0-9.]+]] = icmp sgt <2 x i32> %vec.ind, <i32 10, i32 10>
19 ; CHECK: %predphi = select <2 x i1> [[CMP]], <2 x i32> <i32 1, i32 1>, <2 x i32> zeroinitializer
21 ; CHECK-LABEL: middle.block:
22 ; CHECK: [[E1:%[a-zA-Z0-9.]+]] = extractelement <2 x i32> %predphi, i32 1
24 ; CHECK-LABEL: f1.exit.loopexit:
25 ; CHECK: %.lcssa = phi i32 [ %tmp17, %bb16 ], [ [[E1]], %middle.block ]
29 %b.promoted = load i32, i32* @b, align 4
33 %tmp8 = phi i32 [ %tmp18, %bb16 ], [ %b.promoted, %bb ]
34 %tmp2 = icmp sgt i32 %tmp8, 10
35 br i1 %tmp2, label %bb16, label %bb10
41 %tmp17 = phi i32 [ 0, %bb10 ], [ 1, %.lr.ph.i ]
42 %tmp18 = add nsw i32 %tmp8, 1
43 %tmp19 = icmp slt i32 %tmp18, 4
44 br i1 %tmp19, label %.lr.ph.i, label %f1.exit.loopexit
47 %.lcssa = phi i32 [ %tmp17, %bb16 ]
51 ; non-hdr phi depends on header phi.
52 ; CHECK-LABEL: @test2(
53 ; CHECK: %vec.ind = phi <2 x i32>
54 ; CHECK: [[CMP:%[a-zA-Z0-9.]+]] = icmp sgt <2 x i32> %vec.ind, <i32 10, i32 10>
55 ; CHECK: %predphi = select <2 x i1> [[CMP]], <2 x i32> <i32 1, i32 1>, <2 x i32> %vec.ind
57 ; CHECK-LABEL: middle.block:
58 ; CHECK: [[E1:%[a-zA-Z0-9.]+]] = extractelement <2 x i32> %predphi, i32 1
60 ; CHECK-LABEL: f1.exit.loopexit:
61 ; CHECK: %.lcssa = phi i32 [ %tmp17, %bb16 ], [ [[E1]], %middle.block ]
64 %b.promoted = load i32, i32* @b, align 4
68 %tmp8 = phi i32 [ %tmp18, %bb16 ], [ %b.promoted, %bb ]
69 %tmp2 = icmp sgt i32 %tmp8, 10
70 br i1 %tmp2, label %bb16, label %bb10
76 %tmp17 = phi i32 [ %tmp8, %bb10 ], [ 1, %.lr.ph.i ]
77 %tmp18 = add nsw i32 %tmp8, 1
78 %tmp19 = icmp slt i32 %tmp18, 4
79 br i1 %tmp19, label %.lr.ph.i, label %f1.exit.loopexit
82 %.lcssa = phi i32 [ %tmp17, %bb16 ]
86 ; more than 2 incoming values for tmp17 phi that is used outside loop.
88 ; CHECK-LABEL: vector.body:
89 ; CHECK: %predphi = select <2 x i1> %{{.*}}, <2 x i32> <i32 1, i32 1>, <2 x i32> zeroinitializer
90 ; CHECK: %predphi1 = select <2 x i1> %{{.*}}, <2 x i32> <i32 2, i32 2>, <2 x i32> %predphi
92 ; CHECK-LABEL: middle.block:
93 ; CHECK: [[E1:%[a-zA-Z0-9.]+]] = extractelement <2 x i32> %predphi1, i32 1
95 ; CHECK-LABEL: f1.exit.loopexit:
96 ; CHECK: phi i32 [ %tmp17, %bb16 ], [ [[E1]], %middle.block ]
97 define i32 @test3(i32 %N) {
99 %b.promoted = load i32, i32* @b, align 4
103 %tmp8 = phi i32 [ %tmp18, %bb16 ], [ %b.promoted, %bb ]
104 %tmp2 = icmp sgt i32 %tmp8, 10
105 br i1 %tmp2, label %bb16, label %bb10
108 %cmp = icmp sgt i32 %tmp8, %N
109 br i1 %cmp, label %bb12, label %bb16
115 %tmp17 = phi i32 [ 0, %bb10 ], [ 1, %.lr.ph.i ], [ 2, %bb12 ]
116 %tmp18 = add nsw i32 %tmp8, 1
117 %tmp19 = icmp slt i32 %tmp18, 4
118 br i1 %tmp19, label %.lr.ph.i, label %f1.exit.loopexit
121 %.lcssa = phi i32 [ %tmp17, %bb16 ]
125 ; more than one incoming value for outside user: %.lcssa
126 ; CHECK-LABEL: test4(
127 ; CHECK-LABEL: vector.body:
128 ; CHECK: %predphi = select <2 x i1>
130 ; CHECK-LABEL: middle.block:
131 ; CHECK: [[E1:%[a-zA-Z0-9.]+]] = extractelement <2 x i32> %predphi, i32 1
133 ; CHECK-LABEL: f1.exit.loopexit.loopexit:
134 ; CHECK: %tmp17.lcssa = phi i32 [ %tmp17, %bb16 ], [ [[E1]], %middle.block ]
135 ; CHECK-NEXT: br label %f1.exit.loopexit
137 ; CHECK-LABEL: f1.exit.loopexit:
138 ; CHECK: %.lcssa = phi i32 [ 2, %bb ], [ %tmp17.lcssa, %f1.exit.loopexit.loopexit ]
139 define i32 @test4(i32 %N) {
141 %b.promoted = load i32, i32* @b, align 4
142 %icmp = icmp slt i32 %b.promoted, %N
143 br i1 %icmp, label %f1.exit.loopexit, label %.lr.ph.i
146 %tmp8 = phi i32 [ %tmp18, %bb16 ], [ %b.promoted, %bb ]
147 %tmp2 = icmp sgt i32 %tmp8, 10
148 br i1 %tmp2, label %bb16, label %bb10
154 %tmp17 = phi i32 [ 0, %bb10 ], [ 1, %.lr.ph.i ]
155 %tmp18 = add nsw i32 %tmp8, 1
156 %tmp19 = icmp slt i32 %tmp18, 4
157 br i1 %tmp19, label %.lr.ph.i, label %f1.exit.loopexit
160 %.lcssa = phi i32 [ %tmp17, %bb16 ], [ 2, %bb ]
164 ; non hdr phi that depends on reduction and is used outside the loop.
165 ; reduction phis are only allowed to have bump or reduction operations as the inside user, so we should
166 ; not vectorize this.
167 ; CHECK-LABEL: reduction_sum(
168 ; CHECK-NOT: <2 x i32>
169 define i32 @reduction_sum(i32 %n, i32* noalias nocapture %A, i32* noalias nocapture %B) nounwind uwtable readonly noinline ssp {
171 %c1 = icmp sgt i32 %n, 0
172 br i1 %c1, label %header, label %._crit_edge
174 header: ; preds = %0, %.lr.ph
175 %indvars.iv = phi i64 [ %indvars.iv.next, %bb16 ], [ 0, %entry ]
176 %sum.02 = phi i32 [ %c9, %bb16 ], [ 0, %entry ]
177 %c2 = getelementptr inbounds i32, i32* %A, i64 %indvars.iv
178 %c3 = load i32, i32* %c2, align 4
179 %c4 = getelementptr inbounds i32, i32* %B, i64 %indvars.iv
180 %c5 = load i32, i32* %c4, align 4
181 %tmp2 = icmp sgt i32 %sum.02, 10
182 br i1 %tmp2, label %bb16, label %bb10
188 %tmp17 = phi i32 [ %sum.02, %bb10 ], [ 1, %header ]
189 %c6 = trunc i64 %indvars.iv to i32
190 %c7 = add i32 %sum.02, %c6
191 %c8 = add i32 %c7, %c3
192 %c9 = add i32 %c8, %c5
193 %indvars.iv.next = add i64 %indvars.iv, 1
194 %lftr.wideiv = trunc i64 %indvars.iv.next to i32
195 %exitcond = icmp eq i32 %lftr.wideiv, %n
196 br i1 %exitcond, label %._crit_edge, label %header
198 ._crit_edge: ; preds = %.lr.ph, %0
199 %sum.0.lcssa = phi i32 [ 0, %entry ], [ %c9, %bb16 ]
200 %nonhdr.lcssa = phi i32 [ 1, %entry], [ %tmp17, %bb16 ]
204 ; invalid cyclic dependency with header phi iv, which prevents iv from being
205 ; recognized as induction var.
207 ; CHECK-LABEL: cyclic_dep_with_indvar(
208 ; CHECK-NOT: <2 x i32>
209 define i32 @cyclic_dep_with_indvar() {
211 %b.promoted = load i32, i32* @b, align 4
215 %iv = phi i32 [ %ivnext, %bb16 ], [ %b.promoted, %bb ]
216 %tmp2 = icmp sgt i32 %iv, 10
217 br i1 %tmp2, label %bb16, label %bb10
223 %tmp17 = phi i32 [ 0, %bb10 ], [ %iv, %.lr.ph.i ]
224 %ivnext = add nsw i32 %tmp17, 1
225 %tmp19 = icmp slt i32 %ivnext, 4
226 br i1 %tmp19, label %.lr.ph.i, label %f1.exit.loopexit
229 %.lcssa = phi i32 [ %tmp17, %bb16 ]
233 ; non-reduction phi 'tmp17' used outside loop has cyclic dependence with %x.05 phi
235 ; CHECK-LABEL: not_valid_reduction(
236 ; CHECK-NOT: <2 x i32>
237 define i32 @not_valid_reduction(i32 %n, i32* noalias nocapture %A) nounwind uwtable readonly {
239 %cmp4 = icmp sgt i32 %n, 0
240 br i1 %cmp4, label %for.body, label %for.end
242 for.body: ; preds = %entry, %for.body
243 %indvars.iv = phi i64 [ %indvars.iv.next, %latch ], [ 0, %entry ]
244 %x.05 = phi i32 [ %tmp17, %latch ], [ 0, %entry ]
245 %arrayidx = getelementptr inbounds i32, i32* %A, i64 %indvars.iv
246 %tmp0 = load i32, i32* %arrayidx, align 4
247 %tmp2 = icmp sgt i64 %indvars.iv, 10
248 %sub = sub nsw i32 %x.05, %tmp0
249 br i1 %tmp2, label %bb16, label %bb10
255 %tmp17 = phi i32 [ 1, %bb10 ], [ %sub, %for.body ]
259 %indvars.iv.next = add i64 %indvars.iv, 1
260 %lftr.wideiv = trunc i64 %indvars.iv.next to i32
261 %exitcond = icmp eq i32 %lftr.wideiv, %n
262 br i1 %exitcond, label %for.end, label %for.body
264 for.end: ; preds = %for.body, %entry
265 %x.0.lcssa = phi i32 [ 0, %entry ], [ %tmp17 , %latch ]
270 ; CHECK-LABEL: @outside_user_non_phi(
271 ; CHECK: %vec.ind = phi <2 x i32>
272 ; CHECK: [[CMP:%[a-zA-Z0-9.]+]] = icmp sgt <2 x i32> %vec.ind, <i32 10, i32 10>
273 ; CHECK: %predphi = select <2 x i1> [[CMP]], <2 x i32> <i32 1, i32 1>, <2 x i32> zeroinitializer
274 ; CHECK: [[TRUNC:%[a-zA-Z0-9.]+]] = trunc <2 x i32> %predphi to <2 x i8>
276 ; CHECK-LABEL: middle.block:
277 ; CHECK: [[E1:%[a-zA-Z0-9.]+]] = extractelement <2 x i8> [[TRUNC]], i32 1
279 ; CHECK-LABEL: f1.exit.loopexit:
280 ; CHECK: %.lcssa = phi i8 [ %tmp17.trunc, %bb16 ], [ [[E1]], %middle.block ]
281 define i8 @outside_user_non_phi() {
283 %b.promoted = load i32, i32* @b, align 4
287 %tmp8 = phi i32 [ %tmp18, %bb16 ], [ %b.promoted, %bb ]
288 %tmp2 = icmp sgt i32 %tmp8, 10
289 br i1 %tmp2, label %bb16, label %bb10
295 %tmp17 = phi i32 [ 0, %bb10 ], [ 1, %.lr.ph.i ]
296 %tmp17.trunc = trunc i32 %tmp17 to i8
297 %tmp18 = add nsw i32 %tmp8, 1
298 %tmp19 = icmp slt i32 %tmp18, 4
299 br i1 %tmp19, label %.lr.ph.i, label %f1.exit.loopexit
302 %.lcssa = phi i8 [ %tmp17.trunc, %bb16 ]
306 ; CHECK-LABEL: no_vectorize_reduction_with_outside_use(
307 ; CHECK-NOT: <2 x i32>
308 define i32 @no_vectorize_reduction_with_outside_use(i32 %n, i32* nocapture %A, i32* nocapture %B) nounwind uwtable readonly {
310 %cmp7 = icmp sgt i32 %n, 0
311 br i1 %cmp7, label %for.body, label %for.end
313 for.body: ; preds = %entry, %for.body
314 %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
315 %result.08 = phi i32 [ %or, %for.body ], [ 0, %entry ]
316 %arrayidx = getelementptr inbounds i32, i32* %A, i64 %indvars.iv
317 %0 = load i32, i32* %arrayidx, align 4
318 %arrayidx2 = getelementptr inbounds i32, i32* %B, i64 %indvars.iv
319 %1 = load i32, i32* %arrayidx2, align 4
320 %add = add nsw i32 %1, %0
321 %or = or i32 %add, %result.08
322 %indvars.iv.next = add i64 %indvars.iv, 1
323 %lftr.wideiv = trunc i64 %indvars.iv.next to i32
324 %exitcond = icmp eq i32 %lftr.wideiv, %n
325 br i1 %exitcond, label %for.end, label %for.body
327 for.end: ; preds = %for.body, %entry
328 %result.0.lcssa = phi i32 [ 0, %entry ], [ %1, %for.body ]
329 ret i32 %result.0.lcssa
333 ; vectorize c[i] = a[i] + b[i] loop where result of c[i] is used outside the
335 ; CHECK-LABEL: sum_arrays_outside_use(
336 ; CHECK-LABEL: vector.memcheck:
337 ; CHECK: br i1 %memcheck.conflict, label %scalar.ph, label %vector.ph
339 ; CHECK-LABEL: vector.body:
340 ; CHECK: %wide.load = load <2 x i32>, <2 x i32>*
341 ; CHECK: %wide.load16 = load <2 x i32>, <2 x i32>*
342 ; CHECK: [[ADD:%[a-zA-Z0-9.]+]] = add nsw <2 x i32> %wide.load, %wide.load16
343 ; CHECK: store <2 x i32>
345 ; CHECK-LABEL: middle.block:
346 ; CHECK: [[E1:%[a-zA-Z0-9.]+]] = extractelement <2 x i32> [[ADD]], i32 1
348 ; CHECK-LABEL: f1.exit.loopexit:
349 ; CHECK: %.lcssa = phi i32 [ %sum, %.lr.ph.i ], [ [[E1]], %middle.block ]
350 define i32 @sum_arrays_outside_use(i32* %B, i32* %A, i32* %C, i32 %N) {
352 %b.promoted = load i32, i32* @b, align 4
356 %iv = phi i32 [ %ivnext, %.lr.ph.i ], [ %b.promoted, %bb ]
357 %indvars.iv = sext i32 %iv to i64
358 %arrayidx2 = getelementptr inbounds i32, i32* %B, i64 %indvars.iv
359 %Bload = load i32, i32* %arrayidx2, align 4
360 %arrayidx = getelementptr inbounds i32, i32* %A, i64 %indvars.iv
361 %Aload = load i32, i32* %arrayidx, align 4
362 %sum = add nsw i32 %Bload, %Aload
363 %arrayidx3 = getelementptr inbounds i32, i32* %C, i64 %indvars.iv
364 store i32 %sum, i32* %arrayidx3, align 4
365 %ivnext = add nsw i32 %iv, 1
366 %tmp19 = icmp slt i32 %ivnext, %N
367 br i1 %tmp19, label %.lr.ph.i, label %f1.exit.loopexit
370 %.lcssa = phi i32 [ %sum, %.lr.ph.i ]
374 @tab = common global [32 x i8] zeroinitializer, align 1
376 ; CHECK-LABEL: non_uniform_live_out()
377 ; CHECK-LABEL: vector.body:
378 ; CHECK: %vec.ind = phi <2 x i32> [ <i32 0, i32 1>, %vector.ph ], [ %vec.ind.next, %vector.body ]
379 ; CHECK: [[ADD:%[a-zA-Z0-9.]+]] = add <2 x i32> %vec.ind, <i32 7, i32 7>
380 ; CHECK: [[EE:%[a-zA-Z0-9.]+]] = extractelement <2 x i32> [[ADD]], i32 0
381 ; CHECK: [[GEP:%[a-zA-Z0-9.]+]] = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 [[EE]]
382 ; CHECK-NEXT: [[GEP2:%[a-zA-Z0-9.]+]] = getelementptr inbounds i8, i8* [[GEP]], i32 0
383 ; CHECK-NEXT: [[BC:%[a-zA-Z0-9.]+]] = bitcast i8* [[GEP2]] to <2 x i8>*
384 ; CHECK-NEXT: %wide.load = load <2 x i8>, <2 x i8>* [[BC]]
385 ; CHECK-NEXT: [[ADD2:%[a-zA-Z0-9.]+]] = add <2 x i8> %wide.load, <i8 1, i8 1>
386 ; CHECK: store <2 x i8> [[ADD2]], <2 x i8>*
388 ; CHECK-LABEL: middle.block:
389 ; CHECK: [[ADDEE:%[a-zA-Z0-9.]+]] = extractelement <2 x i32> [[ADD]], i32 1
391 ; CHECK-LABEL: for.end:
392 ; CHECK: %lcssa = phi i32 [ %i.09, %for.body ], [ [[ADDEE]], %middle.block ]
393 ; CHECK: %arrayidx.out = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 %lcssa
394 define i32 @non_uniform_live_out() {
398 for.body: ; preds = %for.body, %entry
399 %i.08 = phi i32 [ 0, %entry ], [ %inc, %for.body ]
400 %i.09 = add i32 %i.08, 7
401 %arrayidx = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 %i.09
402 %0 = load i8, i8* %arrayidx, align 1
404 store i8 %bump, i8* %arrayidx, align 1
405 %inc = add nsw i32 %i.08, 1
406 %exitcond = icmp eq i32 %i.08, 20000
407 br i1 %exitcond, label %for.end, label %for.body
409 for.end: ; preds = %for.body
410 %lcssa = phi i32 [%i.09, %for.body]
411 %arrayidx.out = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 %lcssa
412 store i8 42, i8* %arrayidx.out, align 1