Skip to content

Commit 29e4e6a

Browse files
fhahnfhahn
committed
[LAA] Support assumptions with non-constant deref sizes. (llvm#156758)
Update evaluatePtrAddrecAtMaxBTCWillNotWrap to support non-constant sizes in dereferenceable assumptions. Apply loop-guards in a few places needed to reason about expressions involving trip counts of the from (BTC - 1). PR: llvm#156758 (cherry picked from commit b400fd1)
1 parent 57bd0c8 commit 29e4e6a

File tree

4 files changed

+174
-22
lines changed

4 files changed

+174
-22
lines changed

‎llvm/lib/Analysis/Loads.cpp‎

Lines changed: 2 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -395,7 +395,8 @@ bool llvm::isDereferenceableAndAlignedInLoop(
395395
Base, Alignment,
396396
[&SE, AccessSizeSCEV, &LoopGuards](const RetainedKnowledge &RK) {
397397
return SE.isKnownPredicate(
398-
CmpInst::ICMP_ULE, AccessSizeSCEV,
398+
CmpInst::ICMP_ULE,
399+
SE.applyLoopGuards(AccessSizeSCEV, *LoopGuards),
399400
SE.applyLoopGuards(SE.getSCEV(RK.IRArgValue), *LoopGuards));
400401
},
401402
DL, HeaderFirstNonPHI, AC, &DT) ||

‎llvm/lib/Analysis/LoopAccessAnalysis.cpp‎

Lines changed: 7 additions & 7 deletions
Original file line numberDiff line numberDiff line change
@@ -238,8 +238,8 @@ static bool evaluatePtrAddRecAtMaxBTCWillNotWrap(
238238
StartPtrV, {Attribute::Dereferenceable}, *AC,
239239
L->getLoopPredecessor()->getTerminator(), DT);
240240
if (DerefRK) {
241-
DerefBytesSCEV = SE.getUMaxExpr(
242-
DerefBytesSCEV, SE.getConstant(WiderTy, DerefRK.ArgValue));
241+
DerefBytesSCEV =
242+
SE.getUMaxExpr(DerefBytesSCEV, SE.getSCEV(DerefRK.IRArgValue));
243243
}
244244
}
245245

@@ -259,6 +259,10 @@ static bool evaluatePtrAddRecAtMaxBTCWillNotWrap(
259259
const SCEV *StartOffset = SE.getNoopOrZeroExtend(
260260
SE.getMinusSCEV(AR->getStart(), StartPtr), WiderTy);
261261

262+
if (!LoopGuards)
263+
LoopGuards.emplace(ScalarEvolution::LoopGuards::collect(AR->getLoop(), SE));
264+
MaxBTC = SE.applyLoopGuards(MaxBTC, *LoopGuards);
265+
262266
const SCEV *OffsetAtLastIter =
263267
mulSCEVOverflow(MaxBTC, SE.getAbsExpr(Step, /*IsNSW=*/false), SE);
264268
if (!OffsetAtLastIter) {
@@ -288,11 +292,7 @@ static bool evaluatePtrAddRecAtMaxBTCWillNotWrap(
288292
if (!EndBytes)
289293
return false;
290294

291-
if (!LoopGuards)
292-
LoopGuards.emplace(
293-
ScalarEvolution::LoopGuards::collect(AR->getLoop(), SE));
294-
295-
EndBytes = SE.applyLoopGuards(EndBytes, *LoopGuards);
295+
DerefBytesSCEV = SE.applyLoopGuards(DerefBytesSCEV, *LoopGuards);
296296
return SE.isKnownPredicate(CmpInst::ICMP_ULE, EndBytes, DerefBytesSCEV);
297297
}
298298

‎llvm/test/Transforms/LoopVectorize/single-early-exit-deref-assumptions.ll‎

Lines changed: 155 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -13,9 +13,11 @@ define i64 @early_exit_alignment_and_deref_known_via_assumption_with_constant_si
1313
; CHECK: vector.body:
1414
; CHECK-NEXT: [[INDEX1:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT3:%.*]], [[LOOP]] ]
1515
; CHECK-NEXT: [[TMP0:%.*]] = getelementptr inbounds i8, ptr [[P1]], i64 [[INDEX1]]
16-
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i8>, ptr [[TMP0]], align 1
16+
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i8, ptr [[TMP0]], i32 0
17+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i8>, ptr [[TMP1]], align 1
1718
; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i8, ptr [[P2]], i64 [[INDEX1]]
18-
; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <4 x i8>, ptr [[TMP2]], align 1
19+
; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i8, ptr [[TMP2]], i32 0
20+
; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <4 x i8>, ptr [[TMP3]], align 1
1921
; CHECK-NEXT: [[TMP4:%.*]] = icmp ne <4 x i8> [[WIDE_LOAD]], [[WIDE_LOAD2]]
2022
; CHECK-NEXT: [[INDEX_NEXT3]] = add nuw i64 [[INDEX1]], 4
2123
; CHECK-NEXT: [[TMP5:%.*]] = call i1 @llvm.vector.reduce.or.v4i1(<4 x i1> [[TMP4]])
@@ -25,13 +27,13 @@ define i64 @early_exit_alignment_and_deref_known_via_assumption_with_constant_si
2527
; CHECK: middle.split:
2628
; CHECK-NEXT: br i1 [[TMP5]], label [[VECTOR_EARLY_EXIT:%.*]], label [[MIDDLE_BLOCK:%.*]]
2729
; CHECK: middle.block:
28-
; CHECK-NEXT: br label [[LOOP_END:%.*]]
30+
; CHECK-NEXT: br i1 true, label [[LOOP_END:%.*]], label [[SCALAR_PH]]
2931
; CHECK: vector.early.exit:
3032
; CHECK-NEXT: [[TMP8:%.*]] = call i64 @llvm.experimental.cttz.elts.i64.v4i1(<4 x i1> [[TMP4]], i1 true)
3133
; CHECK-NEXT: [[TMP9:%.*]] = add i64 [[INDEX1]], [[TMP8]]
3234
; CHECK-NEXT: br label [[LOOP_END]]
3335
; CHECK: scalar.ph:
34-
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ]
36+
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 1024, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
3537
; CHECK-NEXT: br label [[LOOP1:%.*]]
3638
; CHECK: loop:
3739
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ [[INDEX_NEXT:%.*]], [[LOOP_INC:%.*]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
@@ -129,21 +131,52 @@ define i64 @early_exit_alignment_and_deref_known_via_assumption_n_not_zero(ptr n
129131
; CHECK-NEXT: [[C:%.*]] = icmp ne i64 [[N]], 0
130132
; CHECK-NEXT: br i1 [[C]], label [[LOOP_PREHEADER:%.*]], label [[LOOP_END:%.*]]
131133
; CHECK: loop.preheader:
134+
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 4
135+
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
136+
; CHECK: vector.ph:
137+
; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 4
138+
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]
139+
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
140+
; CHECK: vector.body:
141+
; CHECK-NEXT: [[INDEX1:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT3:%.*]], [[VECTOR_BODY]] ]
142+
; CHECK-NEXT: [[TMP0:%.*]] = getelementptr inbounds i8, ptr [[P1]], i64 [[INDEX1]]
143+
; CHECK-NEXT: [[TMP9:%.*]] = getelementptr inbounds i8, ptr [[TMP0]], i32 0
144+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i8>, ptr [[TMP9]], align 1
145+
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i8, ptr [[P2]], i64 [[INDEX1]]
146+
; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i8, ptr [[TMP1]], i32 0
147+
; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <4 x i8>, ptr [[TMP3]], align 1
148+
; CHECK-NEXT: [[TMP2:%.*]] = icmp ne <4 x i8> [[WIDE_LOAD]], [[WIDE_LOAD2]]
149+
; CHECK-NEXT: [[INDEX_NEXT3]] = add nuw i64 [[INDEX1]], 4
150+
; CHECK-NEXT: [[TMP4:%.*]] = call i1 @llvm.vector.reduce.or.v4i1(<4 x i1> [[TMP2]])
151+
; CHECK-NEXT: [[TMP5:%.*]] = icmp eq i64 [[INDEX_NEXT3]], [[N_VEC]]
152+
; CHECK-NEXT: [[TMP6:%.*]] = or i1 [[TMP4]], [[TMP5]]
153+
; CHECK-NEXT: br i1 [[TMP6]], label [[MIDDLE_SPLIT:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
154+
; CHECK: middle.split:
155+
; CHECK-NEXT: br i1 [[TMP4]], label [[VECTOR_EARLY_EXIT:%.*]], label [[MIDDLE_BLOCK:%.*]]
156+
; CHECK: middle.block:
157+
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
158+
; CHECK-NEXT: br i1 [[CMP_N]], label [[LOOP_END_LOOPEXIT:%.*]], label [[SCALAR_PH]]
159+
; CHECK: vector.early.exit:
160+
; CHECK-NEXT: [[TMP7:%.*]] = call i64 @llvm.experimental.cttz.elts.i64.v4i1(<4 x i1> [[TMP2]], i1 true)
161+
; CHECK-NEXT: [[TMP8:%.*]] = add i64 [[INDEX1]], [[TMP7]]
162+
; CHECK-NEXT: br label [[LOOP_END_LOOPEXIT]]
163+
; CHECK: scalar.ph:
164+
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[LOOP_PREHEADER]] ]
132165
; CHECK-NEXT: br label [[LOOP:%.*]]
133166
; CHECK: loop:
134-
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ [[INDEX_NEXT:%.*]], [[LOOP_INC:%.*]] ], [ 0, [[LOOP_PREHEADER]] ]
167+
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ [[INDEX_NEXT:%.*]], [[LOOP_INC:%.*]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
135168
; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i8, ptr [[P1]], i64 [[INDEX]]
136169
; CHECK-NEXT: [[LD1:%.*]] = load i8, ptr [[ARRAYIDX]], align 1
137170
; CHECK-NEXT: [[ARRAYIDX1:%.*]] = getelementptr inbounds i8, ptr [[P2]], i64 [[INDEX]]
138171
; CHECK-NEXT: [[LD2:%.*]] = load i8, ptr [[ARRAYIDX1]], align 1
139172
; CHECK-NEXT: [[CMP3:%.*]] = icmp eq i8 [[LD1]], [[LD2]]
140-
; CHECK-NEXT: br i1 [[CMP3]], label [[LOOP_INC]], label [[LOOP_END_LOOPEXIT:%.*]]
173+
; CHECK-NEXT: br i1 [[CMP3]], label [[LOOP_INC]], label [[LOOP_END_LOOPEXIT]]
141174
; CHECK: loop.inc:
142175
; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 1
143176
; CHECK-NEXT: [[EXITCOND:%.*]] = icmp ne i64 [[INDEX_NEXT]], [[N]]
144-
; CHECK-NEXT: br i1 [[EXITCOND]], label [[LOOP]], label [[LOOP_END_LOOPEXIT]]
177+
; CHECK-NEXT: br i1 [[EXITCOND]], label [[LOOP]], label [[LOOP_END_LOOPEXIT]], !llvm.loop [[LOOP5:![0-9]+]]
145178
; CHECK: loop.end.loopexit:
146-
; CHECK-NEXT: [[RETVAL_PH:%.*]] = phi i64 [ -1, [[LOOP_INC]] ], [ [[INDEX]], [[LOOP]] ]
179+
; CHECK-NEXT: [[RETVAL_PH:%.*]] = phi i64 [ -1, [[LOOP_INC]] ], [ [[INDEX]], [[LOOP]] ], [ -1, [[MIDDLE_BLOCK]] ], [ [[TMP8]], [[VECTOR_EARLY_EXIT]] ]
147180
; CHECK-NEXT: br label [[LOOP_END]]
148181
; CHECK: loop.end:
149182
; CHECK-NEXT: [[RETVAL:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[RETVAL_PH]], [[LOOP_END_LOOPEXIT]] ]
@@ -220,3 +253,117 @@ loop.end:
220253
%retval = phi i64 [ %index, %loop ], [ -1, %loop.inc ]
221254
ret i64 %retval
222255
}
256+
257+
define i64 @early_exit_alignment_and_deref_known_via_assumption_n_not_zero_i16(ptr noalias %p1, ptr noalias %p2, i32 %n) nofree nosync {
258+
; CHECK-LABEL: define i64 @early_exit_alignment_and_deref_known_via_assumption_n_not_zero_i16(
259+
; CHECK-SAME: ptr noalias [[P1:%.*]], ptr noalias [[P2:%.*]], i32 [[N:%.*]]) #[[ATTR0]] {
260+
; CHECK-NEXT: entry:
261+
; CHECK-NEXT: [[N_EXT:%.*]] = zext i32 [[N]] to i64
262+
; CHECK-NEXT: [[N_SCALED:%.*]] = shl nuw nsw i64 [[N_EXT]], 1
263+
; CHECK-NEXT: call void @llvm.assume(i1 true) [ "align"(ptr [[P1]], i64 4), "dereferenceable"(ptr [[P1]], i64 [[N_SCALED]]) ]
264+
; CHECK-NEXT: call void @llvm.assume(i1 true) [ "align"(ptr [[P2]], i64 4), "dereferenceable"(ptr [[P2]], i64 [[N_SCALED]]) ]
265+
; CHECK-NEXT: [[C:%.*]] = icmp ne i64 [[N_EXT]], 0
266+
; CHECK-NEXT: br i1 [[C]], label [[LOOP_PREHEADER:%.*]], label [[LOOP_END:%.*]]
267+
; CHECK: loop.preheader:
268+
; CHECK-NEXT: br label [[LOOP:%.*]]
269+
; CHECK: loop:
270+
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ [[INDEX_NEXT:%.*]], [[LOOP_INC:%.*]] ], [ 0, [[LOOP_PREHEADER]] ]
271+
; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i8, ptr [[P1]], i64 [[INDEX]]
272+
; CHECK-NEXT: [[LD1:%.*]] = load i8, ptr [[ARRAYIDX]], align 1
273+
; CHECK-NEXT: [[ARRAYIDX1:%.*]] = getelementptr inbounds i8, ptr [[P2]], i64 [[INDEX]]
274+
; CHECK-NEXT: [[LD2:%.*]] = load i8, ptr [[ARRAYIDX1]], align 1
275+
; CHECK-NEXT: [[CMP3:%.*]] = icmp eq i8 [[LD1]], [[LD2]]
276+
; CHECK-NEXT: br i1 [[CMP3]], label [[LOOP_INC]], label [[LOOP_END_LOOPEXIT:%.*]]
277+
; CHECK: loop.inc:
278+
; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 1
279+
; CHECK-NEXT: [[EXITCOND:%.*]] = icmp ne i64 [[INDEX_NEXT]], [[N_EXT]]
280+
; CHECK-NEXT: br i1 [[EXITCOND]], label [[LOOP]], label [[LOOP_END_LOOPEXIT]]
281+
; CHECK: loop.end.loopexit:
282+
; CHECK-NEXT: [[RETVAL_PH:%.*]] = phi i64 [ -1, [[LOOP_INC]] ], [ [[INDEX]], [[LOOP]] ]
283+
; CHECK-NEXT: br label [[LOOP_END]]
284+
; CHECK: loop.end:
285+
; CHECK-NEXT: [[RETVAL:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[RETVAL_PH]], [[LOOP_END_LOOPEXIT]] ]
286+
; CHECK-NEXT: ret i64 [[RETVAL]]
287+
;
288+
entry:
289+
%n.ext = zext i32 %n to i64
290+
%n.scaled = shl nuw nsw i64 %n.ext, 1
291+
call void @llvm.assume(i1 true) [ "align"(ptr %p1, i64 4), "dereferenceable"(ptr %p1, i64 %n.scaled) ]
292+
call void @llvm.assume(i1 true) [ "align"(ptr %p2, i64 4), "dereferenceable"(ptr %p2, i64 %n.scaled) ]
293+
%c = icmp ne i64 %n.ext, 0
294+
br i1 %c, label %loop, label %loop.end
295+
296+
loop:
297+
%index = phi i64 [ 0, %entry ], [ %index.next, %loop.inc ]
298+
%gep.p1 = getelementptr inbounds i8, ptr %p1, i64 %index
299+
%ld1 = load i8, ptr %gep.p1, align 1
300+
%gep.p2 = getelementptr inbounds i8, ptr %p2, i64 %index
301+
%ld2 = load i8, ptr %gep.p2, align 1
302+
%c.0 = icmp eq i8 %ld1, %ld2
303+
br i1 %c.0, label %loop.inc, label %loop.end
304+
305+
loop.inc:
306+
%index.next = add i64 %index, 1
307+
%exitcond = icmp ne i64 %index.next, %n.ext
308+
br i1 %exitcond, label %loop, label %loop.end
309+
310+
loop.end:
311+
%retval = phi i64 [ 0, %entry ], [ %index, %loop ], [ -1, %loop.inc ]
312+
ret i64 %retval
313+
}
314+
315+
define i64 @early_exit_alignment_and_deref_known_via_assumption_n_not_zero_i16_ptr_iv(ptr %A, i32 noundef %n) nofree nosync {
316+
; CHECK-LABEL: define i64 @early_exit_alignment_and_deref_known_via_assumption_n_not_zero_i16_ptr_iv(
317+
; CHECK-SAME: ptr [[A:%.*]], i32 noundef [[N:%.*]]) #[[ATTR0]] {
318+
; CHECK-NEXT: entry:
319+
; CHECK-NEXT: call void @llvm.assume(i1 true) [ "align"(ptr [[A]], i64 2) ]
320+
; CHECK-NEXT: [[N_EXT:%.*]] = zext i32 [[N]] to i64
321+
; CHECK-NEXT: [[MUL:%.*]] = shl nuw nsw i64 [[N_EXT]], 1
322+
; CHECK-NEXT: call void @llvm.assume(i1 true) [ "dereferenceable"(ptr [[A]], i64 [[MUL]]) ]
323+
; CHECK-NEXT: [[A_END:%.*]] = getelementptr i8, ptr [[A]], i64 [[MUL]]
324+
; CHECK-NEXT: [[PRE:%.*]] = icmp eq i32 [[N]], 0
325+
; CHECK-NEXT: br i1 [[PRE]], label [[EXIT:%.*]], label [[LOOP_HEADER_PREHEADER:%.*]]
326+
; CHECK: loop.header.preheader:
327+
; CHECK-NEXT: br label [[LOOP_HEADER1:%.*]]
328+
; CHECK: loop.header:
329+
; CHECK-NEXT: [[IV1:%.*]] = phi ptr [ [[IV_NEXT1:%.*]], [[LOOP_LATCH1:%.*]] ], [ [[A]], [[LOOP_HEADER_PREHEADER]] ]
330+
; CHECK-NEXT: [[L:%.*]] = load i16, ptr [[IV1]], align 2
331+
; CHECK-NEXT: [[C_0:%.*]] = icmp eq i16 [[L]], 0
332+
; CHECK-NEXT: br i1 [[C_0]], label [[EXIT_LOOPEXIT:%.*]], label [[LOOP_LATCH1]]
333+
; CHECK: loop.latch:
334+
; CHECK-NEXT: [[IV_NEXT1]] = getelementptr inbounds nuw i8, ptr [[IV1]], i64 2
335+
; CHECK-NEXT: [[EC:%.*]] = icmp eq ptr [[IV_NEXT1]], [[A_END]]
336+
; CHECK-NEXT: br i1 [[EC]], label [[EXIT_LOOPEXIT]], label [[LOOP_HEADER1]]
337+
; CHECK: exit.loopexit:
338+
; CHECK-NEXT: [[P_PH:%.*]] = phi ptr [ [[A_END]], [[LOOP_LATCH1]] ], [ [[IV1]], [[LOOP_HEADER1]] ]
339+
; CHECK-NEXT: br label [[EXIT]]
340+
; CHECK: exit:
341+
; CHECK-NEXT: [[P:%.*]] = phi ptr [ [[A]], [[ENTRY:%.*]] ], [ [[P_PH]], [[EXIT_LOOPEXIT]] ]
342+
; CHECK-NEXT: [[RES:%.*]] = ptrtoint ptr [[P]] to i64
343+
; CHECK-NEXT: ret i64 [[RES]]
344+
;
345+
entry:
346+
call void @llvm.assume(i1 true) [ "align"(ptr %A, i64 2) ]
347+
%n.ext = zext i32 %n to i64
348+
%mul = shl nuw nsw i64 %n.ext, 1
349+
call void @llvm.assume(i1 true) [ "dereferenceable"(ptr %A, i64 %mul) ]
350+
%A.end = getelementptr i8, ptr %A, i64 %mul
351+
%pre = icmp eq i32 %n, 0
352+
br i1 %pre, label %exit, label %loop.header
353+
354+
loop.header:
355+
%iv = phi ptr [ %iv.next, %loop.latch ], [ %A, %entry ]
356+
%l = load i16, ptr %iv, align 2
357+
%c.0 = icmp eq i16 %l, 0
358+
br i1 %c.0, label %exit, label %loop.latch
359+
360+
loop.latch:
361+
%iv.next = getelementptr inbounds nuw i8, ptr %iv, i64 2
362+
%ec = icmp eq ptr %iv.next, %A.end
363+
br i1 %ec, label %exit, label %loop.header
364+
365+
exit:
366+
%p = phi ptr [ %A, %entry ], [ %iv, %loop.header ], [ %A.end, %loop.latch ]
367+
%res = ptrtoint ptr %p to i64
368+
ret i64 %res
369+
}

‎llvm/test/Transforms/LoopVectorize/single_early_exit.ll‎

Lines changed: 10 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -355,11 +355,11 @@ define void @inner_loop_trip_count_depends_on_outer_iv(ptr align 8 dereferenceab
355355
; CHECK: vector.body:
356356
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
357357
; CHECK-NEXT: [[TMP0:%.*]] = getelementptr ptr, ptr [[GEP_SRC]], i64 [[INDEX]]
358-
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x ptr>, ptr [[TMP0]], align 8
358+
; CHECK-NEXT: [[TMP2:%.*]] = getelementptr ptr, ptr [[TMP0]], i32 0
359+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x ptr>, ptr [[TMP2]], align 8
359360
; CHECK-NEXT: [[TMP1:%.*]] = icmp eq <4 x ptr> [[WIDE_LOAD]], zeroinitializer
360361
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
361-
; CHECK-NEXT: [[TMP2:%.*]] = freeze <4 x i1> [[TMP1]]
362-
; CHECK-NEXT: [[TMP3:%.*]] = call i1 @llvm.vector.reduce.or.v4i1(<4 x i1> [[TMP2]])
362+
; CHECK-NEXT: [[TMP3:%.*]] = call i1 @llvm.vector.reduce.or.v4i1(<4 x i1> [[TMP1]])
363363
; CHECK-NEXT: [[TMP4:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
364364
; CHECK-NEXT: [[TMP5:%.*]] = or i1 [[TMP3]], [[TMP4]]
365365
; CHECK-NEXT: br i1 [[TMP5]], label [[MIDDLE_SPLIT:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]]
@@ -452,11 +452,11 @@ define i64 @loop_guard_needed_to_prove_dereferenceable(i32 %x, i1 %cmp2) {
452452
; CHECK: vector.body:
453453
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
454454
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr [32 x i32], ptr [[A]], i64 0, i64 [[INDEX]]
455-
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i32>, ptr [[TMP1]], align 4
455+
; CHECK-NEXT: [[TMP3:%.*]] = getelementptr i32, ptr [[TMP1]], i32 0
456+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i32>, ptr [[TMP3]], align 4
456457
; CHECK-NEXT: [[TMP2:%.*]] = icmp eq <4 x i32> [[WIDE_LOAD]], zeroinitializer
457458
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
458-
; CHECK-NEXT: [[TMP3:%.*]] = freeze <4 x i1> [[TMP2]]
459-
; CHECK-NEXT: [[TMP4:%.*]] = call i1 @llvm.vector.reduce.or.v4i1(<4 x i1> [[TMP3]])
459+
; CHECK-NEXT: [[TMP4:%.*]] = call i1 @llvm.vector.reduce.or.v4i1(<4 x i1> [[TMP2]])
460460
; CHECK-NEXT: [[TMP5:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
461461
; CHECK-NEXT: [[TMP6:%.*]] = or i1 [[TMP4]], [[TMP5]]
462462
; CHECK-NEXT: br i1 [[TMP6]], label [[MIDDLE_SPLIT:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP12:![0-9]+]]
@@ -583,4 +583,8 @@ exit:
583583
; CHECK: [[LOOP7]] = distinct !{[[LOOP7]], [[META1]]}
584584
; CHECK: [[LOOP8]] = distinct !{[[LOOP8]], [[META1]], [[META2]]}
585585
; CHECK: [[LOOP9]] = distinct !{[[LOOP9]], [[META2]], [[META1]]}
586+
; CHECK: [[LOOP10]] = distinct !{[[LOOP10]], [[META1]], [[META2]]}
587+
; CHECK: [[LOOP11]] = distinct !{[[LOOP11]], [[META2]], [[META1]]}
588+
; CHECK: [[LOOP12]] = distinct !{[[LOOP12]], [[META1]], [[META2]]}
589+
; CHECK: [[LOOP13]] = distinct !{[[LOOP13]], [[META2]], [[META1]]}
586590
;.

0 commit comments

Comments
 (0)