3 ; RUN: opt -passes=loop-vectorize -debug-only=loop-vectorize -force-vector-interleave=1 -force-vector-width=4 -prefer-inloop-reductions -enable-interleaved-mem-accesses=true -enable-masked-interleaved-mem-accesses -force-widen-divrem-via-safe-divisor=0 -disable-output %s 2>&1 | FileCheck %s
5 target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
7 ; Tests for printing VPlans.
9 define void @print_call_and_memory(i64 %n, ptr noalias %y, ptr noalias %x) nounwind uwtable {
10 ; CHECK-LABEL: Checking a loop in 'print_call_and_memory'
11 ; CHECK: VPlan 'Initial VPlan for VF={4},UF>=1' {
12 ; CHECK-NEXT: Live-in vp<[[VEC_TC:%.+]]> = vector-trip-count
13 ; CHECK-NEXT: Live-in ir<%n> = original trip-count
15 ; CHECK-NEXT: vector.ph:
16 ; CHECK-NEXT: Successor(s): vector loop
18 ; CHECK-NEXT: <x1> vector loop: {
19 ; CHECK-NEXT: vector.body:
20 ; CHECK-NEXT: EMIT vp<[[CAN_IV:%.+]]> = CANONICAL-INDUCTION
21 ; CHECK-NEXT: vp<[[STEPS:%.+]]> = SCALAR-STEPS vp<[[CAN_IV]]>, ir<1>
22 ; CHECK-NEXT: CLONE ir<%arrayidx> = getelementptr inbounds ir<%y>, vp<[[STEPS]]>
23 ; CHECK-NEXT: WIDEN ir<%lv> = load ir<%arrayidx>
24 ; CHECK-NEXT: WIDEN-CALL ir<%call> = call @llvm.sqrt.f32(ir<%lv>)
25 ; CHECK-NEXT: CLONE ir<%arrayidx2> = getelementptr inbounds ir<%x>, vp<[[STEPS]]>
26 ; CHECK-NEXT: WIDEN store ir<%arrayidx2>, ir<%call>
27 ; CHECK-NEXT: EMIT vp<[[CAN_IV_NEXT:%.+]]> = VF * UF + nuw vp<[[CAN_IV]]>
28 ; CHECK-NEXT: EMIT branch-on-count vp<[[CAN_IV_NEXT]]>, vp<[[VEC_TC]]>
29 ; CHECK-NEXT: No successors
31 ; CHECK-NEXT: Successor(s): middle.block
33 ; CHECK-NEXT: middle.block:
34 ; CHECK-NEXT: No successors
38 %cmp6 = icmp sgt i64 %n, 0
39 br i1 %cmp6, label %for.body, label %for.end
41 for.body: ; preds = %entry, %for.body
42 %iv = phi i64 [ %iv.next, %for.body ], [ 0, %entry ]
43 %arrayidx = getelementptr inbounds float, ptr %y, i64 %iv
44 %lv = load float, ptr %arrayidx, align 4
45 %call = tail call float @llvm.sqrt.f32(float %lv) nounwind readnone
46 %arrayidx2 = getelementptr inbounds float, ptr %x, i64 %iv
47 store float %call, ptr %arrayidx2, align 4
48 %iv.next = add i64 %iv, 1
49 %exitcond = icmp eq i64 %iv.next, %n
50 br i1 %exitcond, label %for.end, label %for.body
52 for.end: ; preds = %for.body, %entry
56 define void @print_widen_gep_and_select(i64 %n, ptr noalias %y, ptr noalias %x, ptr %z) nounwind uwtable {
57 ; CHECK-LABEL: Checking a loop in 'print_widen_gep_and_select'
58 ; CHECK: VPlan 'Initial VPlan for VF={4},UF>=1' {
59 ; CHECK-NEXT: Live-in vp<[[VEC_TC:%.+]]> = vector-trip-count
60 ; CHECK-NEXT: Live-in ir<%n> = original trip-count
62 ; CHECK-NEXT: vector.ph:
63 ; CHECK-NEXT: Successor(s): vector loop
65 ; CHECK-NEXT: <x1> vector loop: {
66 ; CHECK-NEXT: vector.body:
67 ; CHECK-NEXT: EMIT vp<[[CAN_IV:%.+]]> = CANONICAL-INDUCTION
68 ; CHECK-NEXT: WIDEN-INDUCTION %iv = phi %iv.next, 0, ir<1>
69 ; CHECK-NEXT: vp<[[STEPS:%.+]]> = SCALAR-STEPS vp<[[CAN_IV]]>, ir<1>
70 ; CHECK-NEXT: WIDEN-GEP Inv[Var] ir<%arrayidx> = getelementptr inbounds ir<%y>, ir<%iv>
71 ; CHECK-NEXT: WIDEN ir<%lv> = load ir<%arrayidx>
72 ; CHECK-NEXT: WIDEN ir<%cmp> = icmp eq ir<%arrayidx>, ir<%z>
73 ; CHECK-NEXT: WIDEN-SELECT ir<%sel> = select ir<%cmp>, ir<1.000000e+01>, ir<2.000000e+01>
74 ; CHECK-NEXT: WIDEN ir<%add> = fadd ir<%lv>, ir<%sel>
75 ; CHECK-NEXT: CLONE ir<%arrayidx2> = getelementptr inbounds ir<%x>, vp<[[STEPS]]>
76 ; CHECK-NEXT: WIDEN store ir<%arrayidx2>, ir<%add>
77 ; CHECK-NEXT: EMIT vp<[[CAN_IV_NEXT:%.+]]> = VF * UF + nuw vp<[[CAN_IV]]>
78 ; CHECK-NEXT: EMIT branch-on-count vp<[[CAN_IV_NEXT]]>, vp<[[VEC_TC]]>
79 ; CHECK-NEXT: No successors
81 ; CHECK-NEXT: Successor(s): middle.block
83 ; CHECK-NEXT: middle.block:
84 ; CHECK-NEXT: No successors
88 %cmp6 = icmp sgt i64 %n, 0
89 br i1 %cmp6, label %for.body, label %for.end
91 for.body: ; preds = %entry, %for.body
92 %iv = phi i64 [ %iv.next, %for.body ], [ 0, %entry ]
93 %arrayidx = getelementptr inbounds float, ptr %y, i64 %iv
94 %lv = load float, ptr %arrayidx, align 4
95 %cmp = icmp eq ptr %arrayidx, %z
96 %sel = select i1 %cmp, float 10.0, float 20.0
97 %add = fadd float %lv, %sel
98 %arrayidx2 = getelementptr inbounds float, ptr %x, i64 %iv
99 store float %add, ptr %arrayidx2, align 4
100 %iv.next = add i64 %iv, 1
101 %exitcond = icmp eq i64 %iv.next, %n
102 br i1 %exitcond, label %for.end, label %for.body
104 for.end: ; preds = %for.body, %entry
108 define float @print_reduction(i64 %n, ptr noalias %y) {
109 ; CHECK-LABEL: Checking a loop in 'print_reduction'
110 ; CHECK: VPlan 'Initial VPlan for VF={4},UF>=1' {
111 ; CHECK-NEXT: Live-in vp<[[VEC_TC:%.+]]> = vector-trip-count
112 ; CHECK-NEXT: Live-in ir<%n> = original trip-count
114 ; CHECK-NEXT: vector.ph:
115 ; CHECK-NEXT: Successor(s): vector loop
117 ; CHECK-NEXT: <x1> vector loop: {
118 ; CHECK-NEXT: vector.body:
119 ; CHECK-NEXT: EMIT vp<[[CAN_IV:%.+]]> = CANONICAL-INDUCTION
120 ; CHECK-NEXT: WIDEN-REDUCTION-PHI ir<%red> = phi ir<0.000000e+00>, ir<%red.next>
121 ; CHECK-NEXT: vp<[[STEPS:%.+]]> = SCALAR-STEPS vp<[[CAN_IV]]>, ir<1>
122 ; CHECK-NEXT: CLONE ir<%arrayidx> = getelementptr inbounds ir<%y>, vp<[[STEPS]]>
123 ; CHECK-NEXT: WIDEN ir<%lv> = load ir<%arrayidx>
124 ; CHECK-NEXT: REDUCE ir<%red.next> = ir<%red> + fast reduce.fadd (ir<%lv>)
125 ; CHECK-NEXT: EMIT vp<[[CAN_IV_NEXT:%.+]]> = VF * UF + nuw vp<[[CAN_IV]]>
126 ; CHECK-NEXT: EMIT branch-on-count vp<[[CAN_IV_NEXT]]>, vp<[[VEC_TC]]>
127 ; CHECK-NEXT: No successors
129 ; CHECK-NEXT: Successor(s): middle.block
131 ; CHECK-NEXT: middle.block:
132 ; CHECK-NEXT: No successors
134 ; CHECK-NEXT: Live-out float %red.next.lcssa = ir<%red.next>
140 for.body: ; preds = %entry, %for.body
141 %iv = phi i64 [ %iv.next, %for.body ], [ 0, %entry ]
142 %red = phi float [ %red.next, %for.body ], [ 0.0, %entry ]
143 %arrayidx = getelementptr inbounds float, ptr %y, i64 %iv
144 %lv = load float, ptr %arrayidx, align 4
145 %red.next = fadd fast float %lv, %red
146 %iv.next = add i64 %iv, 1
147 %exitcond = icmp eq i64 %iv.next, %n
148 br i1 %exitcond, label %for.end, label %for.body
150 for.end: ; preds = %for.body, %entry
154 define void @print_reduction_with_invariant_store(i64 %n, ptr noalias %y, ptr noalias %dst) {
155 ; CHECK-LABEL: Checking a loop in 'print_reduction_with_invariant_store'
156 ; CHECK: VPlan 'Initial VPlan for VF={4},UF>=1' {
157 ; CHECK-NEXT: Live-in vp<[[VEC_TC:%.+]]> = vector-trip-count
158 ; CHECK-NEXT: Live-in ir<%n> = original trip-count
160 ; CHECK-NEXT: vector.ph:
161 ; CHECK-NEXT: Successor(s): vector loop
163 ; CHECK-NEXT: <x1> vector loop: {
164 ; CHECK-NEXT: vector.body:
165 ; CHECK-NEXT: EMIT vp<[[CAN_IV:%.+]]> = CANONICAL-INDUCTION
166 ; CHECK-NEXT: WIDEN-REDUCTION-PHI ir<%red> = phi ir<0.000000e+00>, ir<%red.next>
167 ; CHECK-NEXT: vp<[[IV:%.+]]> = SCALAR-STEPS vp<[[CAN_IV]]>, ir<1>
168 ; CHECK-NEXT: CLONE ir<%arrayidx> = getelementptr inbounds ir<%y>, vp<[[IV]]>
169 ; CHECK-NEXT: WIDEN ir<%lv> = load ir<%arrayidx>
170 ; CHECK-NEXT: REDUCE ir<%red.next> = ir<%red> + fast reduce.fadd (ir<%lv>) (with final reduction value stored in invariant address sank outside of loop)
171 ; CHECK-NEXT: EMIT vp<[[CAN_IV_NEXT:%.+]]> = VF * UF + nuw vp<[[CAN_IV]]>
172 ; CHECK-NEXT: EMIT branch-on-count vp<[[CAN_IV_NEXT]]>, vp<[[VEC_TC]]>
173 ; CHECK-NEXT: No successors
175 ; CHECK-NEXT: Successor(s): middle.block
177 ; CHECK-NEXT: middle.block:
178 ; CHECK-NEXT: No successors
184 for.body: ; preds = %entry, %for.body
185 %iv = phi i64 [ %iv.next, %for.body ], [ 0, %entry ]
186 %red = phi float [ %red.next, %for.body ], [ 0.0, %entry ]
187 %arrayidx = getelementptr inbounds float, ptr %y, i64 %iv
188 %lv = load float, ptr %arrayidx, align 4
189 %red.next = fadd fast float %lv, %red
190 store float %red.next, ptr %dst, align 4
191 %iv.next = add i64 %iv, 1
192 %exitcond = icmp eq i64 %iv.next, %n
193 br i1 %exitcond, label %for.end, label %for.body
195 for.end: ; preds = %for.body, %entry
199 define void @print_replicate_predicated_phi(i64 %n, ptr %x) {
200 ; CHECK-LABEL: Checking a loop in 'print_replicate_predicated_phi'
201 ; CHECK: VPlan 'Initial VPlan for VF={4},UF>=1' {
202 ; CHECK-NEXT: Live-in vp<[[VEC_TC:%.+]]> = vector-trip-count
203 ; CHECK-NEXT: vp<[[TC:%.+]]> = original trip-count
206 ; CHECK-NEXT: EMIT vp<[[TC]]> = EXPAND SCEV (1 smax %n)
207 ; CHECK-NEXT: No successors
209 ; CHECK-NEXT: vector.ph:
210 ; CHECK-NEXT: Successor(s): vector loop
212 ; CHECK-NEXT: <x1> vector loop: {
213 ; CHECK-NEXT: vector.body:
214 ; CHECK-NEXT: EMIT vp<[[CAN_IV:%.+]]> = CANONICAL-INDUCTION
215 ; CHECK-NEXT: WIDEN-INDUCTION %i = phi 0, %i.next, ir<1>
216 ; CHECK-NEXT: vp<[[STEPS:%.+]]> = SCALAR-STEPS vp<[[CAN_IV]]>, ir<1>
217 ; CHECK-NEXT: WIDEN ir<%cmp> = icmp ult ir<%i>, ir<5>
218 ; CHECK-NEXT: Successor(s): pred.udiv
220 ; CHECK-NEXT: <xVFxUF> pred.udiv: {
221 ; CHECK-NEXT: pred.udiv.entry:
222 ; CHECK-NEXT: BRANCH-ON-MASK ir<%cmp>
223 ; CHECK-NEXT: Successor(s): pred.udiv.if, pred.udiv.continue
225 ; CHECK-NEXT: pred.udiv.if:
226 ; CHECK-NEXT: REPLICATE ir<%tmp4> = udiv ir<%n>, vp<[[STEPS]]> (S->V)
227 ; CHECK-NEXT: Successor(s): pred.udiv.continue
229 ; CHECK-NEXT: pred.udiv.continue:
230 ; CHECK-NEXT: PHI-PREDICATED-INSTRUCTION vp<[[PRED:%.+]]> = ir<%tmp4>
231 ; CHECK-NEXT: No successors
233 ; CHECK-NEXT: Successor(s): if.then.0
235 ; CHECK-NEXT: if.then.0:
236 ; CHECK-NEXT: EMIT vp<[[NOT:%.+]]> = not ir<%cmp>
237 ; CHECK-NEXT: BLEND ir<%d> = ir<0>/vp<[[NOT]]> vp<[[PRED]]>/ir<%cmp>
238 ; CHECK-NEXT: CLONE ir<%idx> = getelementptr ir<%x>, vp<[[STEPS]]>
239 ; CHECK-NEXT: WIDEN store ir<%idx>, ir<%d>
240 ; CHECK-NEXT: EMIT vp<[[CAN_IV_NEXT:%.+]]> = VF * UF + nuw vp<[[CAN_IV]]>
241 ; CHECK-NEXT: EMIT branch-on-count vp<[[CAN_IV_NEXT]]>, vp<[[VEC_TC]]>
242 ; CHECK-NEXT: No successors
244 ; CHECK-NEXT: Successor(s): middle.block
246 ; CHECK-NEXT: middle.block:
247 ; CHECK-NEXT: No successors
253 for.body: ; preds = %for.inc, %entry
254 %i = phi i64 [ 0, %entry ], [ %i.next, %for.inc ]
255 %cmp = icmp ult i64 %i, 5
256 br i1 %cmp, label %if.then, label %for.inc
258 if.then: ; preds = %for.body
259 %tmp4 = udiv i64 %n, %i
262 for.inc: ; preds = %if.then, %for.body
263 %d = phi i64 [ 0, %for.body ], [ %tmp4, %if.then ]
264 %idx = getelementptr i64, ptr %x, i64 %i
265 store i64 %d, ptr %idx
266 %i.next = add nuw nsw i64 %i, 1
267 %cond = icmp slt i64 %i.next, %n
268 br i1 %cond, label %for.body, label %for.end
270 for.end: ; preds = %for.inc
274 @AB = common global [1024 x i32] zeroinitializer, align 4
275 @CD = common global [1024 x i32] zeroinitializer, align 4
277 define void @print_interleave_groups(i32 %C, i32 %D) {
278 ; CHECK-LABEL: Checking a loop in 'print_interleave_groups'
279 ; CHECK: VPlan 'Initial VPlan for VF={4},UF>=1' {
280 ; CHECK-NEXT: Live-in vp<[[VEC_TC:%.+]]> = vector-trip-count
281 ; CHECK-NEXT: Live-in ir<256> = original trip-count
283 ; CHECK-NEXT: vector.ph:
284 ; CHECK-NEXT: Successor(s): vector loop
286 ; CHECK-NEXT: <x1> vector loop: {
287 ; CHECK-NEXT: vector.body:
288 ; CHECK-NEXT: EMIT vp<[[CAN_IV:%.+]]> = CANONICAL-INDUCTION
289 ; CHECK-NEXT: vp<[[DERIVED_IV:%.+]]> = DERIVED-IV ir<0> + vp<[[CAN_IV]]> * ir<4>
290 ; CHECK-NEXT: vp<[[STEPS:%.+]]> = SCALAR-STEPS vp<[[DERIVED_IV]]>, ir<4>
291 ; CHECK-NEXT: CLONE ir<%gep.AB.0> = getelementptr inbounds ir<@AB>, ir<0>, vp<[[STEPS]]>
292 ; CHECK-NEXT: INTERLEAVE-GROUP with factor 4 at %AB.0, ir<%gep.AB.0>
293 ; CHECK-NEXT: ir<%AB.0> = load from index 0
294 ; CHECK-NEXT: ir<%AB.1> = load from index 1
295 ; CHECK-NEXT: ir<%AB.3> = load from index 3
296 ; CHECK-NEXT: CLONE ir<%iv.plus.3> = add vp<[[STEPS]]>, ir<3>
297 ; CHECK-NEXT: WIDEN ir<%add> = add nsw ir<%AB.0>, ir<%AB.1>
298 ; CHECK-NEXT: CLONE ir<%gep.CD.3> = getelementptr inbounds ir<@CD>, ir<0>, ir<%iv.plus.3>
299 ; CHECK-NEXT: INTERLEAVE-GROUP with factor 4 at <badref>, ir<%gep.CD.3>
300 ; CHECK-NEXT: store ir<%add> to index 0
301 ; CHECK-NEXT: store ir<1> to index 1
302 ; CHECK-NEXT: store ir<2> to index 2
303 ; CHECK-NEXT: store ir<%AB.3> to index 3
304 ; CHECK-NEXT: EMIT vp<[[CAN_IV_NEXT:%.+]]> = VF * UF + nuw vp<[[CAN_IV]]>
305 ; CHECK-NEXT: EMIT branch-on-count vp<[[CAN_IV_NEXT]]>, vp<[[VEC_TC]]>
306 ; CHECK-NEXT: No successors
308 ; CHECK-NEXT: Successor(s): middle.block
310 ; CHECK-NEXT: middle.block:
311 ; CHECK-NEXT: No successors
318 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
319 %gep.AB.0= getelementptr inbounds [1024 x i32], ptr @AB, i64 0, i64 %iv
320 %AB.0 = load i32, ptr %gep.AB.0, align 4
321 %iv.plus.1 = add i64 %iv, 1
322 %gep.AB.1 = getelementptr inbounds [1024 x i32], ptr @AB, i64 0, i64 %iv.plus.1
323 %AB.1 = load i32, ptr %gep.AB.1, align 4
324 %iv.plus.2 = add i64 %iv, 2
325 %iv.plus.3 = add i64 %iv, 3
326 %gep.AB.3 = getelementptr inbounds [1024 x i32], ptr @AB, i64 0, i64 %iv.plus.3
327 %AB.3 = load i32, ptr %gep.AB.3, align 4
328 %add = add nsw i32 %AB.0, %AB.1
329 %gep.CD.0 = getelementptr inbounds [1024 x i32], ptr @CD, i64 0, i64 %iv
330 store i32 %add, ptr %gep.CD.0, align 4
331 %gep.CD.1 = getelementptr inbounds [1024 x i32], ptr @CD, i64 0, i64 %iv.plus.1
332 store i32 1, ptr %gep.CD.1, align 4
333 %gep.CD.2 = getelementptr inbounds [1024 x i32], ptr @CD, i64 0, i64 %iv.plus.2
334 store i32 2, ptr %gep.CD.2, align 4
335 %gep.CD.3 = getelementptr inbounds [1024 x i32], ptr @CD, i64 0, i64 %iv.plus.3
336 store i32 %AB.3, ptr %gep.CD.3, align 4
337 %iv.next = add nuw nsw i64 %iv, 4
338 %cmp = icmp slt i64 %iv.next, 1024
339 br i1 %cmp, label %for.body, label %for.end
345 define float @print_fmuladd_strict(ptr %a, ptr %b, i64 %n) {
346 ; CHECK-LABEL: Checking a loop in 'print_fmuladd_strict'
347 ; CHECK: VPlan 'Initial VPlan for VF={4},UF>=1' {
348 ; CHECK-NEXT: Live-in vp<[[VEC_TC:%.+]]> = vector-trip-count
349 ; CHECK-NEXT: Live-in ir<%n> = original trip-count
351 ; CHECK-NEXT: vector.ph:
352 ; CHECK-NEXT: Successor(s): vector loop
354 ; CHECK-NEXT: <x1> vector loop: {
355 ; CHECK-NEXT: vector.body:
356 ; CHECK-NEXT: EMIT vp<[[CAN_IV:%.+]]> = CANONICAL-INDUCTION
357 ; CHECK-NEXT: WIDEN-REDUCTION-PHI ir<%sum.07> = phi ir<0.000000e+00>, ir<%muladd>
358 ; CHECK-NEXT: vp<[[STEPS:%.+]]> = SCALAR-STEPS vp<[[CAN_IV]]>, ir<1>
359 ; CHECK-NEXT: CLONE ir<%arrayidx> = getelementptr inbounds ir<%a>, vp<[[STEPS]]>
360 ; CHECK-NEXT: WIDEN ir<%l.a> = load ir<%arrayidx>
361 ; CHECK-NEXT: CLONE ir<%arrayidx2> = getelementptr inbounds ir<%b>, vp<[[STEPS]]>
362 ; CHECK-NEXT: WIDEN ir<%l.b> = load ir<%arrayidx2>
363 ; CHECK-NEXT: EMIT vp<[[FMUL:%.+]]> = fmul nnan ninf nsz ir<%l.a>, ir<%l.b>
364 ; CHECK-NEXT: REDUCE ir<[[MULADD:%.+]]> = ir<%sum.07> + nnan ninf nsz reduce.fadd (vp<[[FMUL]]>)
365 ; CHECK-NEXT: EMIT vp<[[CAN_IV_NEXT:%.+]]> = VF * UF + nuw vp<[[CAN_IV]]>
366 ; CHECK-NEXT: EMIT branch-on-count vp<[[CAN_IV_NEXT]]>, vp<[[VEC_TC]]>
367 ; CHECK-NEXT: No successors
369 ; CHECK-NEXT: Successor(s): middle.block
371 ; CHECK-NEXT: middle.block:
372 ; CHECK-NEXT: No successors
374 ; CHECK-NEXT: Live-out float %muladd.lcssa = ir<%muladd>
381 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
382 %sum.07 = phi float [ 0.000000e+00, %entry ], [ %muladd, %for.body ]
383 %arrayidx = getelementptr inbounds float, ptr %a, i64 %iv
384 %l.a = load float, ptr %arrayidx, align 4
385 %arrayidx2 = getelementptr inbounds float, ptr %b, i64 %iv
386 %l.b = load float, ptr %arrayidx2, align 4
387 %muladd = tail call nnan ninf nsz float @llvm.fmuladd.f32(float %l.a, float %l.b, float %sum.07)
388 %iv.next = add nuw nsw i64 %iv, 1
389 %exitcond.not = icmp eq i64 %iv.next, %n
390 br i1 %exitcond.not, label %for.end, label %for.body
396 define void @debug_loc_vpinstruction(ptr nocapture %asd, ptr nocapture %bsd) !dbg !5 {
397 ; CHECK-LABEL: Checking a loop in 'debug_loc_vpinstruction'
398 ; CHECK: VPlan 'Initial VPlan for VF={4},UF>=1' {
399 ; CHECK-NEXT: Live-in vp<[[VEC_TC:%.+]]> = vector-trip-count
400 ; CHECK-NEXT: Live-in ir<128> = original trip-count
402 ; CHECK-NEXT: vector.ph:
403 ; CHECK-NEXT: Successor(s): vector loop
405 ; CHECK-NEXT: <x1> vector loop: {
406 ; CHECK-NEXT: vector.body:
407 ; CHECK-NEXT: EMIT vp<[[CAN_IV:%.+]]> = CANONICAL-INDUCTION
408 ; CHECK-NEXT: vp<[[STEPS:%.+]]> = SCALAR-STEPS vp<[[CAN_IV]]>, ir<1>
409 ; CHECK-NEXT: CLONE ir<%isd> = getelementptr inbounds ir<%asd>, vp<[[STEPS]]>
410 ; CHECK-NEXT: WIDEN ir<%lsd> = load ir<%isd>
411 ; CHECK-NEXT: WIDEN ir<%psd> = add nuw nsw ir<%lsd>, ir<23>
412 ; CHECK-NEXT: WIDEN ir<%cmp1> = icmp slt ir<%lsd>, ir<100>
413 ; CHECK-NEXT: WIDEN ir<%cmp2> = icmp sge ir<%lsd>, ir<200>
414 ; CHECK-NEXT: EMIT vp<[[NOT1:%.+]]> = not ir<%cmp1>, !dbg /tmp/s.c:5:3
415 ; CHECK-NEXT: EMIT vp<[[SEL1:%.+]]> = select vp<[[NOT1]]>, ir<%cmp2>, ir<false>, !dbg /tmp/s.c:5:21
416 ; CHECK-NEXT: EMIT vp<[[OR1:%.+]]> = or vp<[[SEL1]]>, ir<%cmp1>
417 ; CHECK-NEXT: Successor(s): pred.sdiv
419 ; CHECK-NEXT: <xVFxUF> pred.sdiv: {
420 ; CHECK-NEXT: pred.sdiv.entry:
421 ; CHECK-NEXT: BRANCH-ON-MASK vp<[[OR1]]>
422 ; CHECK-NEXT: Successor(s): pred.sdiv.if, pred.sdiv.continue
424 ; CHECK-NEXT: pred.sdiv.if:
425 ; CHECK-NEXT: REPLICATE ir<%sd1> = sdiv ir<%psd>, ir<%lsd> (S->V)
426 ; CHECK-NEXT: Successor(s): pred.sdiv.continue
428 ; CHECK-NEXT: pred.sdiv.continue:
429 ; CHECK-NEXT: PHI-PREDICATED-INSTRUCTION vp<[[PHI:%.+]]> = ir<%sd1>
430 ; CHECK-NEXT: No successors
432 ; CHECK-NEXT: Successor(s): if.then.0
434 ; CHECK-NEXT: if.then.0:
435 ; CHECK-NEXT: EMIT vp<[[NOT2:%.+]]> = not ir<%cmp2>
436 ; CHECK-NEXT: EMIT vp<[[SEL2:%.+]]> = select vp<[[NOT1]]>, vp<[[NOT2]]>, ir<false>
437 ; CHECK-NEXT: BLEND ir<%ysd.0> = vp<[[PHI]]>/vp<[[OR1]]> ir<%psd>/vp<[[SEL2]]>
438 ; CHECK-NEXT: WIDEN store ir<%isd>, ir<%ysd.0>
439 ; CHECK-NEXT: EMIT vp<[[CAN_IV_NEXT:%.+]]> = VF * UF + nuw vp<[[CAN_IV]]>
440 ; CHECK-NEXT: EMIT branch-on-count vp<[[CAN_IV_NEXT]]>, vp<[[VEC_TC]]>
441 ; CHECK-NEXT: No successors
443 ; CHECK-NEXT: Successor(s): middle.block
445 ; CHECK-NEXT: middle.block:
446 ; CHECK-NEXT: No successors
453 %iv = phi i64 [ 0, %entry ], [ %iv.next, %if.end ]
454 %isd = getelementptr inbounds i32, ptr %asd, i64 %iv
455 %lsd = load i32, ptr %isd, align 4
456 %psd = add nuw nsw i32 %lsd, 23
457 %cmp1 = icmp slt i32 %lsd, 100
458 br i1 %cmp1, label %if.then, label %check, !dbg !7
461 %cmp2 = icmp sge i32 %lsd, 200
462 br i1 %cmp2, label %if.then, label %if.end, !dbg !8
465 %sd1 = sdiv i32 %psd, %lsd
469 %ysd.0 = phi i32 [ %sd1, %if.then ], [ %psd, %check ]
470 store i32 %ysd.0, ptr %isd, align 4
471 %iv.next = add nuw nsw i64 %iv, 1
472 %exitcond = icmp eq i64 %iv.next, 128
473 br i1 %exitcond, label %exit, label %loop
479 declare float @llvm.sqrt.f32(float) nounwind readnone
480 declare float @llvm.fmuladd.f32(float, float, float)
482 define void @print_expand_scev(i64 %y, ptr %ptr) {
483 ; CHECK-LABEL: Checking a loop in 'print_expand_scev'
484 ; CHECK: VPlan 'Initial VPlan for VF={4},UF>=1' {
485 ; CHECK-NEXT: Live-in vp<[[VTC:%.+]]> = vector-trip-count
486 ; CHECK-NEXT: vp<[[TC:%.+]]> = original trip-count
489 ; CHECK-NEXT: EMIT vp<[[TC]]> = EXPAND SCEV (1 + ((15 + (%y /u 492802768830814060))<nuw><nsw> /u (1 + (%y /u 492802768830814060))<nuw><nsw>))<nuw><nsw>
490 ; CHECK-NEXT: EMIT vp<[[EXP_SCEV:%.+]]> = EXPAND SCEV (1 + (%y /u 492802768830814060))<nuw><nsw>
491 ; CHECK-NEXT: No successors
493 ; CHECK-NEXT: vector.ph:
494 ; CHECK-NEXT: Successor(s): vector loop
496 ; CHECK-NEXT: <x1> vector loop: {
497 ; CHECK-NEXT: vector.body:
498 ; CHECK-NEXT: EMIT vp<[[CAN_IV:%.+]]> = CANONICAL-INDUCTION
499 ; CHECK-NEXT: WIDEN-INDUCTION\l" +
500 ; CHECK-NEXT: " %iv = phi %iv.next, 0\l" +
501 ; CHECK-NEXT: " ir<%v2>, vp<[[EXP_SCEV]]>
502 ; CHECK-NEXT: vp<[[DERIVED_IV:%.+]]> = DERIVED-IV ir<0> + vp<[[CAN_IV]]> * vp<[[EXP_SCEV]]>
503 ; CHECK-NEXT: vp<[[STEPS:%.+]]> = SCALAR-STEPS vp<[[DERIVED_IV]]>, vp<[[EXP_SCEV]]>
504 ; CHECK-NEXT: WIDEN ir<%v3> = add nuw ir<%v2>, ir<1>
505 ; CHECK-NEXT: REPLICATE ir<%gep> = getelementptr inbounds ir<%ptr>, vp<[[STEPS]]>
506 ; CHECK-NEXT: REPLICATE store ir<%v3>, ir<%gep>
507 ; CHECK-NEXT: EMIT vp<[[CAN_INC:%.+]]> = VF * UF + nuw vp<[[CAN_IV]]>
508 ; CHECK-NEXT: EMIT branch-on-count vp<[[CAN_INC]]>, vp<[[VTC]]>
509 ; CHECK-NEXT: No successors
511 ; CHECK-NEXT: Successor(s): middle.block
513 ; CHECK-NEXT: middle.block:
514 ; CHECK-NEXT: No successors
518 %div = udiv i64 %y, 492802768830814060
519 %inc = add i64 %div, 1
522 loop: ; preds = %loop, %entry
523 %iv = phi i64 [ %iv.next, %loop ], [ 0, %entry ]
524 %v2 = trunc i64 %iv to i8
525 %v3 = add nuw i8 %v2, 1
526 %gep = getelementptr inbounds i8, ptr %ptr, i64 %iv
527 store i8 %v3, ptr %gep
529 %cmp15 = icmp slt i8 %v3, 10000
530 %iv.next = add i64 %iv, %inc
531 br i1 %cmp15, label %loop, label %loop.exit
537 define i32 @print_exit_value(ptr %ptr, i32 %off) {
538 ; CHECK-LABEL: Checking a loop in 'print_exit_value'
539 ; CHECK: VPlan 'Initial VPlan for VF={4},UF>=1' {
540 ; CHECK-NEXT: Live-in vp<[[VEC_TC:%.+]]> = vector-trip-count
541 ; CHECK-NEXT: Live-in ir<1000> = original trip-count
543 ; CHECK-NEXT: vector.ph:
544 ; CHECK-NEXT: Successor(s): vector loop
546 ; CHECK-NEXT: <x1> vector loop: {
547 ; CHECK-NEXT: vector.body:
548 ; CHECK-NEXT: EMIT vp<[[CAN_IV:%.+]]> = CANONICAL-INDUCTION
549 ; CHECK-NEXT: WIDEN-INDUCTION %iv = phi 0, %iv.next, ir<1>
550 ; CHECK-NEXT: vp<[[STEPS:%.+]]> = SCALAR-STEPS vp<[[CAN_IV]]>, ir<1>
551 ; CHECK-NEXT: CLONE ir<%gep> = getelementptr inbounds ir<%ptr>, vp<[[STEPS]]>
552 ; CHECK-NEXT: WIDEN ir<%add> = add ir<%iv>, ir<%off>
553 ; CHECK-NEXT: WIDEN store ir<%gep>, ir<0>
554 ; CHECK-NEXT: EMIT vp<[[CAN_IV_NEXT:%.+]]> = VF * UF + nuw vp<[[CAN_IV]]>
555 ; CHECK-NEXT: EMIT branch-on-count vp<[[CAN_IV_NEXT]]>, vp<[[VEC_TC]]>
556 ; CHECK-NEXT: No successors
558 ; CHECK-NEXT: Successor(s): middle.block
560 ; CHECK-NEXT: middle.block:
561 ; CHECK-NEXT: No successors
563 ; CHECK-NEXT: Live-out i32 %lcssa = ir<%add>
570 %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
571 %gep = getelementptr inbounds i8, ptr %ptr, i32 %iv
572 %add = add i32 %iv, %off
574 %iv.next = add nsw i32 %iv, 1
575 %ec = icmp eq i32 %iv.next, 1000
576 br i1 %ec, label %exit, label %loop
579 %lcssa = phi i32 [ %add, %loop ]
583 define void @print_fast_math_flags(i64 %n, ptr noalias %y, ptr noalias %x, ptr %z) {
584 ; CHECK-LABEL: Checking a loop in 'print_fast_math_flags'
585 ; CHECK: VPlan 'Initial VPlan for VF={4},UF>=1' {
586 ; CHECK-NEXT: Live-in vp<[[VEC_TC:%.+]]> = vector-trip-count
587 ; CHECK-NEXT: Live-in ir<%n> = original trip-count
589 ; CHECK-NEXT: vector.ph:
590 ; CHECK-NEXT: Successor(s): vector loop
592 ; CHECK-NEXT: <x1> vector loop: {
593 ; CHECK-NEXT: vector.body:
594 ; CHECK-NEXT: EMIT vp<[[CAN_IV:%.+]]> = CANONICAL-INDUCTION
595 ; CHECK-NEXT: vp<[[STEPS:%.+]]> = SCALAR-STEPS vp<[[CAN_IV]]>, ir<1>
596 ; CHECK-NEXT: CLONE ir<%gep.y> = getelementptr inbounds ir<%y>, vp<[[STEPS]]>
597 ; CHECK-NEXT: WIDEN ir<%lv> = load ir<%gep.y>
598 ; CHECK-NEXT: WIDEN ir<%add> = fadd nnan ir<%lv>, ir<1.000000e+00>
599 ; CHECK-NEXT: WIDEN ir<%mul> = fmul reassoc nnan ninf nsz arcp contract afn ir<%add>, ir<2.000000e+00>
600 ; CHECK-NEXT: WIDEN ir<%div> = fdiv reassoc nsz contract ir<%mul>, ir<2.000000e+00>
601 ; CHECK-NEXT: CLONE ir<%gep.x> = getelementptr inbounds ir<%x>, vp<[[STEPS]]>
602 ; CHECK-NEXT: WIDEN store ir<%gep.x>, ir<%div>
603 ; CHECK-NEXT: EMIT vp<[[CAN_IV_NEXT:%.+]]> = VF * UF + nuw vp<[[CAN_IV]]>
604 ; CHECK-NEXT: EMIT branch-on-count vp<[[CAN_IV_NEXT]]>, vp<[[VEC_TC]]>
605 ; CHECK-NEXT: No successors
607 ; CHECK-NEXT: Successor(s): middle.block
609 ; CHECK-NEXT: middle.block:
610 ; CHECK-NEXT: No successors
617 %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
618 %gep.y = getelementptr inbounds float, ptr %y, i64 %iv
619 %lv = load float, ptr %gep.y, align 4
620 %add = fadd nnan float %lv, 1.0
621 %mul = fmul fast float %add, 2.0
622 %div = fdiv nsz reassoc contract float %mul, 2.0
623 %gep.x = getelementptr inbounds float, ptr %x, i64 %iv
624 store float %div, ptr %gep.x, align 4
625 %iv.next = add i64 %iv, 1
626 %exitcond = icmp eq i64 %iv.next, %n
627 br i1 %exitcond, label %exit, label %loop
633 define void @print_exact_flags(i64 %n, ptr noalias %x) {
634 ; CHECK-LABEL: Checking a loop in 'print_exact_flags'
635 ; CHECK: VPlan 'Initial VPlan for VF={4},UF>=1' {
636 ; CHECK-NEXT: Live-in vp<[[VEC_TC:%.+]]> = vector-trip-count
637 ; CHECK-NEXT: Live-in ir<%n> = original trip-count
639 ; CHECK-NEXT: vector.ph:
640 ; CHECK-NEXT: Successor(s): vector loop
642 ; CHECK-NEXT: <x1> vector loop: {
643 ; CHECK-NEXT: vector.body:
644 ; CHECK-NEXT: EMIT vp<[[CAN_IV:%.+]]> = CANONICAL-INDUCTION
645 ; CHECK-NEXT: vp<[[STEPS:%.+]]> = SCALAR-STEPS vp<[[CAN_IV]]>, ir<1>
646 ; CHECK-NEXT: CLONE ir<%gep.x> = getelementptr inbounds ir<%x>, vp<[[STEPS]]>
647 ; CHECK-NEXT: WIDEN ir<%lv> = load ir<%gep.x>
648 ; CHECK-NEXT: WIDEN ir<%div.1> = udiv exact ir<%lv>, ir<20>
649 ; CHECK-NEXT: WIDEN ir<%div.2> = udiv ir<%lv>, ir<60>
650 ; CHECK-NEXT: WIDEN ir<%add> = add nuw nsw ir<%div.1>, ir<%div.2>
651 ; CHECK-NEXT: WIDEN store ir<%gep.x>, ir<%add>
652 ; CHECK-NEXT: EMIT vp<[[CAN_IV_NEXT:%.+]]> = VF * UF + nuw vp<[[CAN_IV]]>
653 ; CHECK-NEXT: EMIT branch-on-count vp<[[CAN_IV_NEXT]]>, vp<[[VEC_TC]]>
654 ; CHECK-NEXT: No successors
656 ; CHECK-NEXT: Successor(s): middle.block
658 ; CHECK-NEXT: middle.block:
659 ; CHECK-NEXT: No successors
666 %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
667 %gep.x = getelementptr inbounds i32, ptr %x, i64 %iv
668 %lv = load i32, ptr %gep.x, align 4
669 %div.1 = udiv exact i32 %lv, 20
670 %div.2 = udiv i32 %lv, 60
671 %add = add nsw nuw i32 %div.1, %div.2
672 store i32 %add, ptr %gep.x, align 4
673 %iv.next = add i64 %iv, 1
674 %exitcond = icmp eq i64 %iv.next, %n
675 br i1 %exitcond, label %exit, label %loop
681 define void @print_call_flags(ptr readonly %src, ptr noalias %dest, i64 %n) {
682 ; CHECK-LABEL: Checking a loop in 'print_call_flags'
683 ; CHECK: VPlan 'Initial VPlan for VF={4},UF>=1' {
684 ; CHECK-NEXT: Live-in vp<[[VEC_TC:%.+]]> = vector-trip-count
685 ; CHECK-NEXT: Live-in ir<%n> = original trip-count
687 ; CHECK-NEXT: vector.ph:
688 ; CHECK-NEXT: Successor(s): vector loop
690 ; CHECK-NEXT: <x1> vector loop: {
691 ; CHECK-NEXT: vector.body:
692 ; CHECK-NEXT: EMIT vp<[[CAN_IV:%.+]]> = CANONICAL-INDUCTION
693 ; CHECK-NEXT: vp<[[STEPS:%.+]]> = SCALAR-STEPS vp<[[CAN_IV]]>, ir<1>
694 ; CHECK-NEXT: CLONE ir<%ld.addr> = getelementptr inbounds ir<%src>, vp<%2>
695 ; CHECK-NEXT: WIDEN ir<%ld.value> = load ir<%ld.addr>
696 ; CHECK-NEXT: WIDEN ir<%ifcond> = fcmp oeq ir<%ld.value>, ir<5.000000e+00>
697 ; CHECK-NEXT: Successor(s): pred.call
699 ; CHECK-NEXT: <xVFxUF> pred.call: {
700 ; CHECK-NEXT: pred.call.entry:
701 ; CHECK-NEXT: BRANCH-ON-MASK ir<%ifcond>
702 ; CHECK-NEXT: Successor(s): pred.call.if, pred.call.continue
704 ; CHECK-NEXT: pred.call.if:
705 ; CHECK-NEXT: REPLICATE ir<%foo.ret.1> = call nnan ninf nsz @foo(ir<%ld.value>) (S->V)
706 ; CHECK-NEXT: REPLICATE ir<%foo.ret.2> = call @foo(ir<%ld.value>) (S->V)
707 ; CHECK-NEXT: Successor(s): pred.call.continue
709 ; CHECK-NEXT: pred.call.continue:
710 ; CHECK-NEXT: PHI-PREDICATED-INSTRUCTION vp<%8> = ir<%foo.ret.1>
711 ; CHECK-NEXT: PHI-PREDICATED-INSTRUCTION vp<%9> = ir<%foo.ret.2>
712 ; CHECK-NEXT: No successors
714 ; CHECK-NEXT: Successor(s): if.then.1
716 ; CHECK-NEXT: if.then.1:
717 ; CHECK-NEXT: WIDEN ir<%fadd> = fadd vp<%8>, vp<%9>
718 ; CHECK-NEXT: EMIT vp<%11> = not ir<%ifcond>
719 ; CHECK-NEXT: BLEND ir<%st.value> = ir<%ld.value>/vp<%11> ir<%fadd>/ir<%ifcond>
720 ; CHECK-NEXT: CLONE ir<%st.addr> = getelementptr inbounds ir<%dest>, vp<%2>
721 ; CHECK-NEXT: WIDEN store ir<%st.addr>, ir<%st.value>
722 ; CHECK-NEXT: EMIT vp<[[CAN_IV_NEXT:%.+]]> = VF * UF + nuw vp<[[CAN_IV]]>
723 ; CHECK-NEXT: EMIT branch-on-count vp<[[CAN_IV_NEXT]]>, vp<[[VEC_TC]]>
724 ; CHECK-NEXT: No successors
726 ; CHECK-NEXT: Successor(s): middle.block
728 ; CHECK-NEXT: middle.block:
729 ; CHECK-NEXT: No successors
736 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.loop ]
737 %ld.addr = getelementptr inbounds float, ptr %src, i64 %iv
738 %ld.value = load float , ptr %ld.addr, align 8
739 %ifcond = fcmp oeq float %ld.value, 5.0
740 br i1 %ifcond, label %if.then, label %for.loop
743 %foo.ret.1 = call nnan nsz ninf float @foo(float %ld.value) #0
744 %foo.ret.2 = call float @foo(float %ld.value) #0
745 %fadd = fadd float %foo.ret.1, %foo.ret.2
749 %st.value = phi float [ %ld.value, %for.body ], [ %fadd, %if.then ]
750 %st.addr = getelementptr inbounds float, ptr %dest, i64 %iv
751 store float %st.value, ptr %st.addr, align 8
752 %iv.next = add nsw nuw i64 %iv, 1
753 %loopcond = icmp eq i64 %iv.next, %n
754 br i1 %loopcond, label %end, label %for.body
761 !llvm.module.flags = !{!3, !4}
763 declare float @foo(float) #0
764 declare <2 x float> @vector_foo(<2 x float>, <2 x i1>)
766 ; We need a vector variant in order to allow for vectorization at present, but
767 ; we want to test scalarization of conditional calls. If we provide a variant
768 ; with a different number of lanes than the VF we force via
769 ; "-force-vector-width=4", then it should pass the legality checks but
770 ; scalarize. TODO: Remove the requirement to have a variant.
771 attributes #0 = { readonly nounwind "vector-function-abi-variant"="_ZGV_LLVM_M2v_foo(vector_foo)" }
773 !0 = distinct !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang", isOptimized: true, runtimeVersion: 0, emissionKind: NoDebug, enums: !2)
774 !1 = !DIFile(filename: "/tmp/s.c", directory: "/tmp")
776 !3 = !{i32 2, !"Debug Info Version", i32 3}
777 !4 = !{i32 7, !"PIC Level", i32 2}
778 !5 = distinct !DISubprogram(name: "f", scope: !1, file: !1, line: 4, type: !6, scopeLine: 4, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0, retainedNodes: !2)
779 !6 = !DISubroutineType(types: !2)
780 !7 = !DILocation(line: 5, column: 3, scope: !5)
781 !8 = !DILocation(line: 5, column: 21, scope: !5)