Skip to content

Commit 1ffea97

Browse files
preamespreames
authored
[indvars] Support known positive extends in getExtendedOperandRecurrence (#70990)
IndVars has the existing notion of a narrow definition which is known to positive and thus both sign and zero extension kinds are actually the same operations. There's existing logic for forming a SCEV based on the extension kind and the no-wrap flags. This change extends that logic to form the opposite extension kind for a positive def if doing so is allowed by the flags. Note that we already do something analogous for the getWideRecurrence case as well.
1 parent 7ca0f44 commit 1ffea97

File tree

4 files changed

+53
-38
lines changed

4 files changed

+53
-38
lines changed

‎llvm/lib/Transforms/Utils/SimplifyIndVar.cpp

Lines changed: 16 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1393,7 +1393,22 @@ WidenIV::getExtendedOperandRecurrence(WidenIV::NarrowIVDefUse DU) {
13931393
else if (ExtKind == ExtendKind::Zero && OBO->hasNoUnsignedWrap())
13941394
ExtendOperExpr = SE->getZeroExtendExpr(
13951395
SE->getSCEV(DU.NarrowUse->getOperand(ExtendOperIdx)), WideType);
1396-
else
1396+
else if (DU.NeverNegative) {
1397+
// For a non-negative NarrowDef, we can choose either type of
1398+
// extension. We want to use the current extend kind if legal
1399+
// (see above), and we only hit this code if we need to check
1400+
// the opposite case.
1401+
if (OBO->hasNoSignedWrap()) {
1402+
ExtKind = ExtendKind::Sign;
1403+
ExtendOperExpr = SE->getSignExtendExpr(
1404+
SE->getSCEV(DU.NarrowUse->getOperand(ExtendOperIdx)), WideType);
1405+
} else if (OBO->hasNoUnsignedWrap()) {
1406+
ExtKind = ExtendKind::Zero;
1407+
ExtendOperExpr = SE->getZeroExtendExpr(
1408+
SE->getSCEV(DU.NarrowUse->getOperand(ExtendOperIdx)), WideType);
1409+
} else
1410+
return {nullptr, ExtendKind::Unknown};
1411+
} else
13971412
return {nullptr, ExtendKind::Unknown};
13981413

13991414
// When creating this SCEV expr, don't apply the current operations NSW or NUW

‎llvm/test/Transforms/IndVarSimplify/promote-iv-to-eliminate-casts.ll

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -196,8 +196,8 @@ define void @promote_latch_condition_decrementing_loop_01(ptr %p, ptr %a) {
196196
; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[LOOP]] ], [ [[TMP0]], [[PREHEADER]] ]
197197
; CHECK-NEXT: [[EL:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDVARS_IV]]
198198
; CHECK-NEXT: store atomic i32 0, ptr [[EL]] unordered, align 4
199-
; CHECK-NEXT: [[LOOPCOND:%.*]] = icmp slt i64 [[INDVARS_IV]], 1
200199
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nsw i64 [[INDVARS_IV]], -1
200+
; CHECK-NEXT: [[LOOPCOND:%.*]] = icmp slt i64 [[INDVARS_IV]], 1
201201
; CHECK-NEXT: br i1 [[LOOPCOND]], label [[LOOPEXIT_LOOPEXIT:%.*]], label [[LOOP]]
202202
;
203203

@@ -336,8 +336,8 @@ define void @promote_latch_condition_decrementing_loop_04(ptr %p, ptr %a, i1 %co
336336
; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[LOOP]] ], [ [[TMP0]], [[PREHEADER]] ]
337337
; CHECK-NEXT: [[EL:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDVARS_IV]]
338338
; CHECK-NEXT: store atomic i32 0, ptr [[EL]] unordered, align 4
339-
; CHECK-NEXT: [[LOOPCOND:%.*]] = icmp slt i64 [[INDVARS_IV]], 1
340339
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nsw i64 [[INDVARS_IV]], -1
340+
; CHECK-NEXT: [[LOOPCOND:%.*]] = icmp slt i64 [[INDVARS_IV]], 1
341341
; CHECK-NEXT: br i1 [[LOOPCOND]], label [[LOOPEXIT_LOOPEXIT:%.*]], label [[LOOP]]
342342
;
343343

@@ -398,8 +398,8 @@ define void @promote_latch_condition_decrementing_loop_05(ptr %p, ptr %a, i1 %co
398398
; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[LOOP]] ], [ [[TMP0]], [[PREHEADER]] ]
399399
; CHECK-NEXT: [[EL:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDVARS_IV]]
400400
; CHECK-NEXT: store atomic i32 0, ptr [[EL]] unordered, align 4
401-
; CHECK-NEXT: [[LOOPCOND:%.*]] = icmp slt i64 [[INDVARS_IV]], 1
402401
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nsw i64 [[INDVARS_IV]], -1
402+
; CHECK-NEXT: [[LOOPCOND:%.*]] = icmp slt i64 [[INDVARS_IV]], 1
403403
; CHECK-NEXT: br i1 [[LOOPCOND]], label [[LOOPEXIT_LOOPEXIT:%.*]], label [[LOOP]]
404404
;
405405

‎llvm/test/Transforms/IndVarSimplify/widen-nonnegative.ll

Lines changed: 24 additions & 18 deletions
Original file line numberDiff line numberDiff line change
@@ -143,14 +143,15 @@ exit:
143143
define void @sext_add_nuw(ptr %A, i32 %offset, i32 %M) {
144144
; CHECK-LABEL: @sext_add_nuw(
145145
; CHECK-NEXT: entry:
146+
; CHECK-NEXT: [[TMP0:%.*]] = zext i32 [[OFFSET:%.*]] to i64
146147
; CHECK-NEXT: [[SMAX:%.*]] = call i32 @llvm.smax.i32(i32 [[M:%.*]], i32 1)
147148
; CHECK-NEXT: [[WIDE_TRIP_COUNT:%.*]] = zext i32 [[SMAX]] to i64
148149
; CHECK-NEXT: br label [[FOR_BODY:%.*]]
149150
; CHECK: for.body:
150151
; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ 0, [[ENTRY:%.*]] ]
151-
; CHECK-NEXT: [[TMP0:%.*]] = trunc i64 [[INDVARS_IV]] to i32
152-
; CHECK-NEXT: [[ADD_US:%.*]] = add nuw i32 [[TMP0]], [[OFFSET:%.*]]
153-
; CHECK-NEXT: [[IDXPROM_US:%.*]] = sext i32 [[ADD_US]] to i64
152+
; CHECK-NEXT: [[TMP1:%.*]] = add nuw nsw i64 [[INDVARS_IV]], [[TMP0]]
153+
; CHECK-NEXT: [[TMP2:%.*]] = trunc i64 [[TMP1]] to i32
154+
; CHECK-NEXT: [[IDXPROM_US:%.*]] = sext i32 [[TMP2]] to i64
154155
; CHECK-NEXT: [[ARRAYIDX_US:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[IDXPROM_US]]
155156
; CHECK-NEXT: tail call void @use_ptr(ptr [[ARRAYIDX_US]])
156157
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
@@ -215,14 +216,15 @@ exit:
215216
define void @zext_add_nsw(ptr %A, i32 %offset, i32 %M) {
216217
; CHECK-LABEL: @zext_add_nsw(
217218
; CHECK-NEXT: entry:
219+
; CHECK-NEXT: [[TMP0:%.*]] = sext i32 [[OFFSET:%.*]] to i64
218220
; CHECK-NEXT: [[SMAX:%.*]] = call i32 @llvm.smax.i32(i32 [[M:%.*]], i32 1)
219221
; CHECK-NEXT: [[WIDE_TRIP_COUNT:%.*]] = zext i32 [[SMAX]] to i64
220222
; CHECK-NEXT: br label [[FOR_BODY:%.*]]
221223
; CHECK: for.body:
222224
; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ 0, [[ENTRY:%.*]] ]
223-
; CHECK-NEXT: [[TMP0:%.*]] = trunc i64 [[INDVARS_IV]] to i32
224-
; CHECK-NEXT: [[ADD_US:%.*]] = add nsw i32 [[TMP0]], [[OFFSET:%.*]]
225-
; CHECK-NEXT: [[IDXPROM_US:%.*]] = zext i32 [[ADD_US]] to i64
225+
; CHECK-NEXT: [[TMP1:%.*]] = add nsw i64 [[INDVARS_IV]], [[TMP0]]
226+
; CHECK-NEXT: [[TMP2:%.*]] = trunc i64 [[TMP1]] to i32
227+
; CHECK-NEXT: [[IDXPROM_US:%.*]] = zext i32 [[TMP2]] to i64
226228
; CHECK-NEXT: [[ARRAYIDX_US:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[IDXPROM_US]]
227229
; CHECK-NEXT: tail call void @use_ptr(ptr [[ARRAYIDX_US]])
228230
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
@@ -322,14 +324,15 @@ exit:
322324
define void @zext_nneg_add_nsw(ptr %A, i32 %offset, i32 %M) {
323325
; CHECK-LABEL: @zext_nneg_add_nsw(
324326
; CHECK-NEXT: entry:
327+
; CHECK-NEXT: [[TMP0:%.*]] = sext i32 [[OFFSET:%.*]] to i64
325328
; CHECK-NEXT: [[SMAX:%.*]] = call i32 @llvm.smax.i32(i32 [[M:%.*]], i32 1)
326329
; CHECK-NEXT: [[WIDE_TRIP_COUNT:%.*]] = zext i32 [[SMAX]] to i64
327330
; CHECK-NEXT: br label [[FOR_BODY:%.*]]
328331
; CHECK: for.body:
329332
; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ 0, [[ENTRY:%.*]] ]
330-
; CHECK-NEXT: [[TMP0:%.*]] = trunc i64 [[INDVARS_IV]] to i32
331-
; CHECK-NEXT: [[ADD_US:%.*]] = add nsw i32 [[TMP0]], [[OFFSET:%.*]]
332-
; CHECK-NEXT: [[IDXPROM_US:%.*]] = zext nneg i32 [[ADD_US]] to i64
333+
; CHECK-NEXT: [[TMP1:%.*]] = add nsw i64 [[INDVARS_IV]], [[TMP0]]
334+
; CHECK-NEXT: [[TMP2:%.*]] = trunc i64 [[TMP1]] to i32
335+
; CHECK-NEXT: [[IDXPROM_US:%.*]] = zext nneg i32 [[TMP2]] to i64
333336
; CHECK-NEXT: [[ARRAYIDX_US:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[IDXPROM_US]]
334337
; CHECK-NEXT: tail call void @use_ptr(ptr [[ARRAYIDX_US]])
335338
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
@@ -467,14 +470,15 @@ exit:
467470
define void @sext_mul_nuw(ptr %A, i32 %multiple, i32 %M) {
468471
; CHECK-LABEL: @sext_mul_nuw(
469472
; CHECK-NEXT: entry:
473+
; CHECK-NEXT: [[TMP0:%.*]] = zext i32 [[MULTIPLE:%.*]] to i64
470474
; CHECK-NEXT: [[SMAX:%.*]] = call i32 @llvm.smax.i32(i32 [[M:%.*]], i32 1)
471475
; CHECK-NEXT: [[WIDE_TRIP_COUNT:%.*]] = zext i32 [[SMAX]] to i64
472476
; CHECK-NEXT: br label [[FOR_BODY:%.*]]
473477
; CHECK: for.body:
474478
; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ 0, [[ENTRY:%.*]] ]
475-
; CHECK-NEXT: [[TMP0:%.*]] = trunc i64 [[INDVARS_IV]] to i32
476-
; CHECK-NEXT: [[MUL_US:%.*]] = mul nuw i32 [[TMP0]], [[MULTIPLE:%.*]]
477-
; CHECK-NEXT: [[IDXPROM_US:%.*]] = sext i32 [[MUL_US]] to i64
479+
; CHECK-NEXT: [[TMP1:%.*]] = mul nuw nsw i64 [[INDVARS_IV]], [[TMP0]]
480+
; CHECK-NEXT: [[TMP2:%.*]] = trunc i64 [[TMP1]] to i32
481+
; CHECK-NEXT: [[IDXPROM_US:%.*]] = sext i32 [[TMP2]] to i64
478482
; CHECK-NEXT: [[ARRAYIDX_US:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[IDXPROM_US]]
479483
; CHECK-NEXT: tail call void @use_ptr(ptr [[ARRAYIDX_US]])
480484
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
@@ -539,14 +543,15 @@ exit:
539543
define void @zext_mul_nsw(ptr %A, i32 %multiple, i32 %M) {
540544
; CHECK-LABEL: @zext_mul_nsw(
541545
; CHECK-NEXT: entry:
546+
; CHECK-NEXT: [[TMP0:%.*]] = sext i32 [[MULTIPLE:%.*]] to i64
542547
; CHECK-NEXT: [[SMAX:%.*]] = call i32 @llvm.smax.i32(i32 [[M:%.*]], i32 1)
543548
; CHECK-NEXT: [[WIDE_TRIP_COUNT:%.*]] = zext i32 [[SMAX]] to i64
544549
; CHECK-NEXT: br label [[FOR_BODY:%.*]]
545550
; CHECK: for.body:
546551
; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ 0, [[ENTRY:%.*]] ]
547-
; CHECK-NEXT: [[TMP0:%.*]] = trunc i64 [[INDVARS_IV]] to i32
548-
; CHECK-NEXT: [[MUL_US:%.*]] = mul nsw i32 [[TMP0]], [[MULTIPLE:%.*]]
549-
; CHECK-NEXT: [[IDXPROM_US:%.*]] = zext i32 [[MUL_US]] to i64
552+
; CHECK-NEXT: [[TMP1:%.*]] = mul nsw i64 [[INDVARS_IV]], [[TMP0]]
553+
; CHECK-NEXT: [[TMP2:%.*]] = trunc i64 [[TMP1]] to i32
554+
; CHECK-NEXT: [[IDXPROM_US:%.*]] = zext i32 [[TMP2]] to i64
550555
; CHECK-NEXT: [[ARRAYIDX_US:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[IDXPROM_US]]
551556
; CHECK-NEXT: tail call void @use_ptr(ptr [[ARRAYIDX_US]])
552557
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
@@ -646,14 +651,15 @@ exit:
646651
define void @zext_nneg_mul_nsw(ptr %A, i32 %multiple, i32 %M) {
647652
; CHECK-LABEL: @zext_nneg_mul_nsw(
648653
; CHECK-NEXT: entry:
654+
; CHECK-NEXT: [[TMP0:%.*]] = sext i32 [[MULTIPLE:%.*]] to i64
649655
; CHECK-NEXT: [[SMAX:%.*]] = call i32 @llvm.smax.i32(i32 [[M:%.*]], i32 1)
650656
; CHECK-NEXT: [[WIDE_TRIP_COUNT:%.*]] = zext i32 [[SMAX]] to i64
651657
; CHECK-NEXT: br label [[FOR_BODY:%.*]]
652658
; CHECK: for.body:
653659
; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ 0, [[ENTRY:%.*]] ]
654-
; CHECK-NEXT: [[TMP0:%.*]] = trunc i64 [[INDVARS_IV]] to i32
655-
; CHECK-NEXT: [[MUL_US:%.*]] = mul nsw i32 [[TMP0]], [[MULTIPLE:%.*]]
656-
; CHECK-NEXT: [[IDXPROM_US:%.*]] = zext nneg i32 [[MUL_US]] to i64
660+
; CHECK-NEXT: [[TMP1:%.*]] = mul nsw i64 [[INDVARS_IV]], [[TMP0]]
661+
; CHECK-NEXT: [[TMP2:%.*]] = trunc i64 [[TMP1]] to i32
662+
; CHECK-NEXT: [[IDXPROM_US:%.*]] = zext nneg i32 [[TMP2]] to i64
657663
; CHECK-NEXT: [[ARRAYIDX_US:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[IDXPROM_US]]
658664
; CHECK-NEXT: tail call void @use_ptr(ptr [[ARRAYIDX_US]])
659665
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1

‎llvm/test/Transforms/LoopFlatten/widen-iv.ll

Lines changed: 10 additions & 16 deletions
Original file line numberDiff line numberDiff line change
@@ -1,11 +1,11 @@
11
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
22

3-
; RUN: opt < %s -S -passes='loop-simplify,loop(loop-flatten),dce,verify' -loop-flatten-widen-iv=true \
3+
; RUN: opt < %s -S -passes='loop-simplify,loop(loop-flatten),adce,verify' -loop-flatten-widen-iv=true \
44
; RUN: -verify-loop-info -verify-dom-info -verify-scev \
55
; RUN: -loop-flatten-cost-threshold=6 | \
66
; RUN: FileCheck %s --check-prefix=CHECK
77

8-
; RUN: opt < %s -S -passes='loop-simplify,loop(loop-flatten),dce,verify' -loop-flatten-widen-iv=false \
8+
; RUN: opt < %s -S -passes='loop-simplify,loop(loop-flatten),adce,verify' -loop-flatten-widen-iv=false \
99
; RUN: -verify-loop-info -verify-dom-info -verify-scev | \
1010
; RUN: FileCheck %s --check-prefix=DONTWIDEN
1111

@@ -29,17 +29,15 @@ define void @foo(ptr %A, i32 %N, i32 %M) {
2929
; CHECK-NEXT: [[FLATTEN_TRIPCOUNT:%.*]] = mul i64 [[TMP0]], [[TMP1]]
3030
; CHECK-NEXT: br label [[FOR_COND1_PREHEADER_US:%.*]]
3131
; CHECK: for.cond1.preheader.us:
32-
; CHECK-NEXT: [[INDVAR1:%.*]] = phi i64 [ [[INDVAR_NEXT2:%.*]], [[FOR_COND1_FOR_COND_CLEANUP3_CRIT_EDGE_US:%.*]] ], [ 0, [[FOR_COND1_PREHEADER_US_PREHEADER]] ]
33-
; CHECK-NEXT: [[FLATTEN_TRUNCIV:%.*]] = trunc i64 [[INDVAR1]] to i32
32+
; CHECK-NEXT: [[INDVAR2:%.*]] = phi i64 [ [[INDVAR_NEXT3:%.*]], [[FOR_COND1_FOR_COND_CLEANUP3_CRIT_EDGE_US:%.*]] ], [ 0, [[FOR_COND1_PREHEADER_US_PREHEADER]] ]
3433
; CHECK-NEXT: br label [[FOR_BODY4_US:%.*]]
3534
; CHECK: for.body4.us:
36-
; CHECK-NEXT: [[IDXPROM_US:%.*]] = sext i32 [[FLATTEN_TRUNCIV]] to i64
37-
; CHECK-NEXT: [[ARRAYIDX_US:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[IDXPROM_US]]
35+
; CHECK-NEXT: [[ARRAYIDX_US:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDVAR2]]
3836
; CHECK-NEXT: tail call void @f(ptr [[ARRAYIDX_US]])
3937
; CHECK-NEXT: br label [[FOR_COND1_FOR_COND_CLEANUP3_CRIT_EDGE_US]]
4038
; CHECK: for.cond1.for.cond.cleanup3_crit_edge.us:
41-
; CHECK-NEXT: [[INDVAR_NEXT2]] = add i64 [[INDVAR1]], 1
42-
; CHECK-NEXT: [[CMP_US:%.*]] = icmp slt i64 [[INDVAR_NEXT2]], [[FLATTEN_TRIPCOUNT]]
39+
; CHECK-NEXT: [[INDVAR_NEXT3]] = add i64 [[INDVAR2]], 1
40+
; CHECK-NEXT: [[CMP_US:%.*]] = icmp slt i64 [[INDVAR_NEXT3]], [[FLATTEN_TRIPCOUNT]]
4341
; CHECK-NEXT: br i1 [[CMP_US]], label [[FOR_COND1_PREHEADER_US]], label [[FOR_COND_CLEANUP_LOOPEXIT:%.*]]
4442
; CHECK: for.cond.cleanup.loopexit:
4543
; CHECK-NEXT: br label [[FOR_COND_CLEANUP]]
@@ -138,7 +136,6 @@ define void @foo2_sext(i32* nocapture readonly %A, i32 %N, i32 %M) {
138136
; CHECK-NEXT: br label [[FOR_COND1_PREHEADER_US:%.*]]
139137
; CHECK: for.cond1.preheader.us:
140138
; CHECK-NEXT: [[INDVAR2:%.*]] = phi i64 [ [[INDVAR_NEXT3:%.*]], [[FOR_COND1_FOR_COND_CLEANUP3_CRIT_EDGE_US:%.*]] ], [ 0, [[FOR_COND1_PREHEADER_US_PREHEADER]] ]
141-
; CHECK-NEXT: [[I_018_US:%.*]] = phi i32 [ [[INC6_US:%.*]], [[FOR_COND1_FOR_COND_CLEANUP3_CRIT_EDGE_US]] ], [ 0, [[FOR_COND1_PREHEADER_US_PREHEADER]] ]
142139
; CHECK-NEXT: br label [[FOR_BODY4_US:%.*]]
143140
; CHECK: for.body4.us:
144141
; CHECK-NEXT: [[ARRAYIDX_US:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDVAR2]]
@@ -147,7 +144,6 @@ define void @foo2_sext(i32* nocapture readonly %A, i32 %N, i32 %M) {
147144
; CHECK-NEXT: br label [[FOR_COND1_FOR_COND_CLEANUP3_CRIT_EDGE_US]]
148145
; CHECK: for.cond1.for.cond.cleanup3_crit_edge.us:
149146
; CHECK-NEXT: [[INDVAR_NEXT3]] = add i64 [[INDVAR2]], 1
150-
; CHECK-NEXT: [[INC6_US]] = add nuw nsw i32 [[I_018_US]], 1
151147
; CHECK-NEXT: [[CMP_US:%.*]] = icmp slt i64 [[INDVAR_NEXT3]], [[FLATTEN_TRIPCOUNT]]
152148
; CHECK-NEXT: br i1 [[CMP_US]], label [[FOR_COND1_PREHEADER_US]], label [[FOR_COND_CLEANUP_LOOPEXIT:%.*]]
153149
; CHECK: for.cond1.preheader:
@@ -1002,17 +998,15 @@ define void @foo_M_sext(ptr %A, i32 %N, i16 %M) {
1002998
; CHECK-NEXT: [[FLATTEN_TRIPCOUNT:%.*]] = mul i64 [[TMP0]], [[TMP1]]
1003999
; CHECK-NEXT: br label [[FOR_COND1_PREHEADER_US:%.*]]
10041000
; CHECK: for.cond1.preheader.us:
1005-
; CHECK-NEXT: [[INDVAR1:%.*]] = phi i64 [ [[INDVAR_NEXT2:%.*]], [[FOR_COND1_FOR_COND_CLEANUP3_CRIT_EDGE_US:%.*]] ], [ 0, [[FOR_COND1_PREHEADER_US_PREHEADER]] ]
1006-
; CHECK-NEXT: [[FLATTEN_TRUNCIV:%.*]] = trunc i64 [[INDVAR1]] to i32
1001+
; CHECK-NEXT: [[INDVAR2:%.*]] = phi i64 [ [[INDVAR_NEXT3:%.*]], [[FOR_COND1_FOR_COND_CLEANUP3_CRIT_EDGE_US:%.*]] ], [ 0, [[FOR_COND1_PREHEADER_US_PREHEADER]] ]
10071002
; CHECK-NEXT: br label [[FOR_BODY4_US:%.*]]
10081003
; CHECK: for.body4.us:
1009-
; CHECK-NEXT: [[IDXPROM_US:%.*]] = sext i32 [[FLATTEN_TRUNCIV]] to i64
1010-
; CHECK-NEXT: [[ARRAYIDX_US:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[IDXPROM_US]]
1004+
; CHECK-NEXT: [[ARRAYIDX_US:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDVAR2]]
10111005
; CHECK-NEXT: tail call void @f(ptr [[ARRAYIDX_US]])
10121006
; CHECK-NEXT: br label [[FOR_COND1_FOR_COND_CLEANUP3_CRIT_EDGE_US]]
10131007
; CHECK: for.cond1.for.cond.cleanup3_crit_edge.us:
1014-
; CHECK-NEXT: [[INDVAR_NEXT2]] = add i64 [[INDVAR1]], 1
1015-
; CHECK-NEXT: [[CMP_US:%.*]] = icmp slt i64 [[INDVAR_NEXT2]], [[FLATTEN_TRIPCOUNT]]
1008+
; CHECK-NEXT: [[INDVAR_NEXT3]] = add i64 [[INDVAR2]], 1
1009+
; CHECK-NEXT: [[CMP_US:%.*]] = icmp slt i64 [[INDVAR_NEXT3]], [[FLATTEN_TRIPCOUNT]]
10161010
; CHECK-NEXT: br i1 [[CMP_US]], label [[FOR_COND1_PREHEADER_US]], label [[FOR_COND_CLEANUP_LOOPEXIT:%.*]]
10171011
; CHECK: for.cond.cleanup.loopexit:
10181012
; CHECK-NEXT: br label [[FOR_COND_CLEANUP]]

0 commit comments

Comments
 (0)