Skip to content

Commit cd20600

Browse files
aleks-tmbAleksander Popov
and
Aleksander Popov
authored
[LoopConstrainer] Apply loop gurads to check that loop bounds are safe (#71531)
Loop guards that apply to loop SCEV bounds allow IRCE for cases with compound loop bounds such as: if (K > 0 && M > 0) for (i = 0; i < min(K, M); i++) {...} if (K > 0 && M > 0) for (i = min(K, M); i >= 0; i--) {...} Otherwise SCEV couldn't prove that loops have safe bounds in these cases. Co-authored-by: Aleksander Popov <[email protected]>
1 parent 13ccaf9 commit cd20600

File tree

2 files changed

+90
-17
lines changed

2 files changed

+90
-17
lines changed

llvm/lib/Transforms/Utils/LoopConstrainer.cpp

Lines changed: 14 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -42,8 +42,11 @@ static bool isSafeDecreasingBound(const SCEV *Start, const SCEV *BoundSCEV,
4242
ICmpInst::Predicate BoundPred =
4343
IsSigned ? CmpInst::ICMP_SGT : CmpInst::ICMP_UGT;
4444

45+
auto StartLG = SE.applyLoopGuards(Start, L);
46+
auto BoundLG = SE.applyLoopGuards(BoundSCEV, L);
47+
4548
if (LatchBrExitIdx == 1)
46-
return SE.isLoopEntryGuardedByCond(L, BoundPred, Start, BoundSCEV);
49+
return SE.isLoopEntryGuardedByCond(L, BoundPred, StartLG, BoundLG);
4750

4851
assert(LatchBrExitIdx == 0 && "LatchBrExitIdx should be either 0 or 1");
4952

@@ -54,10 +57,10 @@ static bool isSafeDecreasingBound(const SCEV *Start, const SCEV *BoundSCEV,
5457
const SCEV *Limit = SE.getMinusSCEV(SE.getConstant(Min), StepPlusOne);
5558

5659
const SCEV *MinusOne =
57-
SE.getMinusSCEV(BoundSCEV, SE.getOne(BoundSCEV->getType()));
60+
SE.getMinusSCEV(BoundLG, SE.getOne(BoundLG->getType()));
5861

59-
return SE.isLoopEntryGuardedByCond(L, BoundPred, Start, MinusOne) &&
60-
SE.isLoopEntryGuardedByCond(L, BoundPred, BoundSCEV, Limit);
62+
return SE.isLoopEntryGuardedByCond(L, BoundPred, StartLG, MinusOne) &&
63+
SE.isLoopEntryGuardedByCond(L, BoundPred, BoundLG, Limit);
6164
}
6265

6366
/// Given a loop with an increasing induction variable, is it possible to
@@ -86,8 +89,11 @@ static bool isSafeIncreasingBound(const SCEV *Start, const SCEV *BoundSCEV,
8689
ICmpInst::Predicate BoundPred =
8790
IsSigned ? CmpInst::ICMP_SLT : CmpInst::ICMP_ULT;
8891

92+
auto StartLG = SE.applyLoopGuards(Start, L);
93+
auto BoundLG = SE.applyLoopGuards(BoundSCEV, L);
94+
8995
if (LatchBrExitIdx == 1)
90-
return SE.isLoopEntryGuardedByCond(L, BoundPred, Start, BoundSCEV);
96+
return SE.isLoopEntryGuardedByCond(L, BoundPred, StartLG, BoundLG);
9197

9298
assert(LatchBrExitIdx == 0 && "LatchBrExitIdx should be 0 or 1");
9399

@@ -97,9 +103,9 @@ static bool isSafeIncreasingBound(const SCEV *Start, const SCEV *BoundSCEV,
97103
: APInt::getMaxValue(BitWidth);
98104
const SCEV *Limit = SE.getMinusSCEV(SE.getConstant(Max), StepMinusOne);
99105

100-
return (SE.isLoopEntryGuardedByCond(L, BoundPred, Start,
101-
SE.getAddExpr(BoundSCEV, Step)) &&
102-
SE.isLoopEntryGuardedByCond(L, BoundPred, BoundSCEV, Limit));
106+
return (SE.isLoopEntryGuardedByCond(L, BoundPred, StartLG,
107+
SE.getAddExpr(BoundLG, Step)) &&
108+
SE.isLoopEntryGuardedByCond(L, BoundPred, BoundLG, Limit));
103109
}
104110

105111
/// Returns estimate for max latch taken count of the loop of the narrowest

llvm/test/Transforms/IRCE/compound-loop-bound.ll

Lines changed: 76 additions & 9 deletions
Original file line numberDiff line numberDiff line change
@@ -16,23 +16,56 @@ define void @incrementing_loop(ptr %arr, ptr %len_ptr, i32 %K, i32 %M) {
1616
; CHECK-NEXT: br i1 [[AND]], label [[PREHEADER:%.*]], label [[EXIT:%.*]]
1717
; CHECK: preheader:
1818
; CHECK-NEXT: [[SMIN:%.*]] = call i32 @llvm.smin.i32(i32 [[K]], i32 [[M]])
19+
; CHECK-NEXT: [[SMIN1:%.*]] = call i32 @llvm.smin.i32(i32 [[LEN]], i32 [[M]])
20+
; CHECK-NEXT: [[SMIN2:%.*]] = call i32 @llvm.smin.i32(i32 [[SMIN1]], i32 [[K]])
21+
; CHECK-NEXT: [[EXIT_MAINLOOP_AT:%.*]] = call i32 @llvm.smax.i32(i32 [[SMIN2]], i32 0)
22+
; CHECK-NEXT: [[TMP0:%.*]] = icmp slt i32 0, [[EXIT_MAINLOOP_AT]]
23+
; CHECK-NEXT: br i1 [[TMP0]], label [[LOOP_PREHEADER:%.*]], label [[MAIN_PSEUDO_EXIT:%.*]]
24+
; CHECK: loop.preheader:
1925
; CHECK-NEXT: br label [[LOOP:%.*]]
2026
; CHECK: loop:
21-
; CHECK-NEXT: [[IDX:%.*]] = phi i32 [ 0, [[PREHEADER]] ], [ [[IDX_NEXT:%.*]], [[IN_BOUNDS:%.*]] ]
22-
; CHECK-NEXT: [[IDX_NEXT]] = add i32 [[IDX]], 1
27+
; CHECK-NEXT: [[IDX:%.*]] = phi i32 [ [[IDX_NEXT:%.*]], [[IN_BOUNDS:%.*]] ], [ 0, [[LOOP_PREHEADER]] ]
28+
; CHECK-NEXT: [[IDX_NEXT]] = add nsw i32 [[IDX]], 1
2329
; CHECK-NEXT: [[GUARD:%.*]] = icmp slt i32 [[IDX]], [[LEN]]
24-
; CHECK-NEXT: br i1 [[GUARD]], label [[IN_BOUNDS]], label [[OUT_OF_BOUNDS:%.*]]
30+
; CHECK-NEXT: br i1 true, label [[IN_BOUNDS]], label [[OUT_OF_BOUNDS_LOOPEXIT3:%.*]]
2531
; CHECK: in.bounds:
2632
; CHECK-NEXT: [[ADDR:%.*]] = getelementptr i32, ptr [[ARR]], i32 [[IDX]]
2733
; CHECK-NEXT: store i32 0, ptr [[ADDR]], align 4
2834
; CHECK-NEXT: [[NEXT:%.*]] = icmp slt i32 [[IDX_NEXT]], [[SMIN]]
29-
; CHECK-NEXT: br i1 [[NEXT]], label [[LOOP]], label [[EXIT_LOOPEXIT:%.*]]
35+
; CHECK-NEXT: [[TMP1:%.*]] = icmp slt i32 [[IDX_NEXT]], [[EXIT_MAINLOOP_AT]]
36+
; CHECK-NEXT: br i1 [[TMP1]], label [[LOOP]], label [[MAIN_EXIT_SELECTOR:%.*]]
37+
; CHECK: main.exit.selector:
38+
; CHECK-NEXT: [[IDX_NEXT_LCSSA:%.*]] = phi i32 [ [[IDX_NEXT]], [[IN_BOUNDS]] ]
39+
; CHECK-NEXT: [[TMP2:%.*]] = icmp slt i32 [[IDX_NEXT_LCSSA]], [[SMIN]]
40+
; CHECK-NEXT: br i1 [[TMP2]], label [[MAIN_PSEUDO_EXIT]], label [[EXIT_LOOPEXIT:%.*]]
41+
; CHECK: main.pseudo.exit:
42+
; CHECK-NEXT: [[IDX_COPY:%.*]] = phi i32 [ 0, [[PREHEADER]] ], [ [[IDX_NEXT_LCSSA]], [[MAIN_EXIT_SELECTOR]] ]
43+
; CHECK-NEXT: [[INDVAR_END:%.*]] = phi i32 [ 0, [[PREHEADER]] ], [ [[IDX_NEXT_LCSSA]], [[MAIN_EXIT_SELECTOR]] ]
44+
; CHECK-NEXT: br label [[POSTLOOP:%.*]]
45+
; CHECK: out.of.bounds.loopexit:
46+
; CHECK-NEXT: br label [[OUT_OF_BOUNDS:%.*]]
47+
; CHECK: out.of.bounds.loopexit3:
48+
; CHECK-NEXT: br label [[OUT_OF_BOUNDS]]
3049
; CHECK: out.of.bounds:
3150
; CHECK-NEXT: ret void
51+
; CHECK: exit.loopexit.loopexit:
52+
; CHECK-NEXT: br label [[EXIT_LOOPEXIT]]
3253
; CHECK: exit.loopexit:
3354
; CHECK-NEXT: br label [[EXIT]]
3455
; CHECK: exit:
3556
; CHECK-NEXT: ret void
57+
; CHECK: postloop:
58+
; CHECK-NEXT: br label [[LOOP_POSTLOOP:%.*]]
59+
; CHECK: loop.postloop:
60+
; CHECK-NEXT: [[IDX_POSTLOOP:%.*]] = phi i32 [ [[IDX_COPY]], [[POSTLOOP]] ], [ [[IDX_NEXT_POSTLOOP:%.*]], [[IN_BOUNDS_POSTLOOP:%.*]] ]
61+
; CHECK-NEXT: [[IDX_NEXT_POSTLOOP]] = add i32 [[IDX_POSTLOOP]], 1
62+
; CHECK-NEXT: [[GUARD_POSTLOOP:%.*]] = icmp slt i32 [[IDX_POSTLOOP]], [[LEN]]
63+
; CHECK-NEXT: br i1 [[GUARD_POSTLOOP]], label [[IN_BOUNDS_POSTLOOP]], label [[OUT_OF_BOUNDS_LOOPEXIT:%.*]]
64+
; CHECK: in.bounds.postloop:
65+
; CHECK-NEXT: [[ADDR_POSTLOOP:%.*]] = getelementptr i32, ptr [[ARR]], i32 [[IDX_POSTLOOP]]
66+
; CHECK-NEXT: store i32 0, ptr [[ADDR_POSTLOOP]], align 4
67+
; CHECK-NEXT: [[NEXT_POSTLOOP:%.*]] = icmp slt i32 [[IDX_NEXT_POSTLOOP]], [[SMIN]]
68+
; CHECK-NEXT: br i1 [[NEXT_POSTLOOP]], label [[LOOP_POSTLOOP]], label [[EXIT_LOOPEXIT_LOOPEXIT:%.*]], !llvm.loop [[LOOP1:![0-9]+]], !loop_constrainer.loop.clone !6
3669
;
3770
entry:
3871
%len = load i32, ptr %len_ptr, !range !0
@@ -78,24 +111,58 @@ define void @decrementing_loop(ptr %arr, ptr %len_ptr, i32 %K, i32 %M) {
78111
; CHECK-NEXT: [[AND:%.*]] = and i1 [[CHECK0]], [[CHECK1]]
79112
; CHECK-NEXT: br i1 [[AND]], label [[PREHEADER:%.*]], label [[EXIT:%.*]]
80113
; CHECK: preheader:
81-
; CHECK-NEXT: [[SMIN:%.*]] = call i32 @llvm.smin.i32(i32 [[K]], i32 [[M]])
114+
; CHECK-NEXT: [[INDVAR_START:%.*]] = call i32 @llvm.smin.i32(i32 [[K]], i32 [[M]])
115+
; CHECK-NEXT: [[TMP0:%.*]] = add i32 [[INDVAR_START]], 1
116+
; CHECK-NEXT: [[SMIN:%.*]] = call i32 @llvm.smin.i32(i32 [[LEN]], i32 [[TMP0]])
117+
; CHECK-NEXT: [[SMAX:%.*]] = call i32 @llvm.smax.i32(i32 [[SMIN]], i32 0)
118+
; CHECK-NEXT: [[EXIT_PRELOOP_AT:%.*]] = add nsw i32 [[SMAX]], -1
119+
; CHECK-NEXT: [[TMP1:%.*]] = icmp sgt i32 [[INDVAR_START]], [[EXIT_PRELOOP_AT]]
120+
; CHECK-NEXT: br i1 [[TMP1]], label [[LOOP_PRELOOP_PREHEADER:%.*]], label [[PRELOOP_PSEUDO_EXIT:%.*]]
121+
; CHECK: loop.preloop.preheader:
122+
; CHECK-NEXT: br label [[LOOP_PRELOOP:%.*]]
123+
; CHECK: mainloop:
82124
; CHECK-NEXT: br label [[LOOP:%.*]]
83125
; CHECK: loop:
84-
; CHECK-NEXT: [[IDX:%.*]] = phi i32 [ [[SMIN]], [[PREHEADER]] ], [ [[IDX_DEC:%.*]], [[IN_BOUNDS:%.*]] ]
85-
; CHECK-NEXT: [[IDX_DEC]] = sub i32 [[IDX]], 1
126+
; CHECK-NEXT: [[IDX:%.*]] = phi i32 [ [[IDX_PRELOOP_COPY:%.*]], [[MAINLOOP:%.*]] ], [ [[IDX_DEC:%.*]], [[IN_BOUNDS:%.*]] ]
127+
; CHECK-NEXT: [[IDX_DEC]] = sub nsw i32 [[IDX]], 1
86128
; CHECK-NEXT: [[GUARD:%.*]] = icmp slt i32 [[IDX]], [[LEN]]
87-
; CHECK-NEXT: br i1 [[GUARD]], label [[IN_BOUNDS]], label [[OUT_OF_BOUNDS:%.*]]
129+
; CHECK-NEXT: br i1 true, label [[IN_BOUNDS]], label [[OUT_OF_BOUNDS_LOOPEXIT1:%.*]]
88130
; CHECK: in.bounds:
89131
; CHECK-NEXT: [[ADDR:%.*]] = getelementptr i32, ptr [[ARR]], i32 [[IDX]]
90132
; CHECK-NEXT: store i32 0, ptr [[ADDR]], align 4
91133
; CHECK-NEXT: [[NEXT:%.*]] = icmp sgt i32 [[IDX_DEC]], -1
92-
; CHECK-NEXT: br i1 [[NEXT]], label [[LOOP]], label [[EXIT_LOOPEXIT:%.*]]
134+
; CHECK-NEXT: br i1 [[NEXT]], label [[LOOP]], label [[EXIT_LOOPEXIT_LOOPEXIT:%.*]]
135+
; CHECK: out.of.bounds.loopexit:
136+
; CHECK-NEXT: br label [[OUT_OF_BOUNDS:%.*]]
137+
; CHECK: out.of.bounds.loopexit1:
138+
; CHECK-NEXT: br label [[OUT_OF_BOUNDS]]
93139
; CHECK: out.of.bounds:
94140
; CHECK-NEXT: ret void
141+
; CHECK: exit.loopexit.loopexit:
142+
; CHECK-NEXT: br label [[EXIT_LOOPEXIT:%.*]]
95143
; CHECK: exit.loopexit:
96144
; CHECK-NEXT: br label [[EXIT]]
97145
; CHECK: exit:
98146
; CHECK-NEXT: ret void
147+
; CHECK: loop.preloop:
148+
; CHECK-NEXT: [[IDX_PRELOOP:%.*]] = phi i32 [ [[IDX_DEC_PRELOOP:%.*]], [[IN_BOUNDS_PRELOOP:%.*]] ], [ [[INDVAR_START]], [[LOOP_PRELOOP_PREHEADER]] ]
149+
; CHECK-NEXT: [[IDX_DEC_PRELOOP]] = sub i32 [[IDX_PRELOOP]], 1
150+
; CHECK-NEXT: [[GUARD_PRELOOP:%.*]] = icmp slt i32 [[IDX_PRELOOP]], [[LEN]]
151+
; CHECK-NEXT: br i1 [[GUARD_PRELOOP]], label [[IN_BOUNDS_PRELOOP]], label [[OUT_OF_BOUNDS_LOOPEXIT:%.*]]
152+
; CHECK: in.bounds.preloop:
153+
; CHECK-NEXT: [[ADDR_PRELOOP:%.*]] = getelementptr i32, ptr [[ARR]], i32 [[IDX_PRELOOP]]
154+
; CHECK-NEXT: store i32 0, ptr [[ADDR_PRELOOP]], align 4
155+
; CHECK-NEXT: [[NEXT_PRELOOP:%.*]] = icmp sgt i32 [[IDX_DEC_PRELOOP]], -1
156+
; CHECK-NEXT: [[TMP2:%.*]] = icmp sgt i32 [[IDX_DEC_PRELOOP]], [[EXIT_PRELOOP_AT]]
157+
; CHECK-NEXT: br i1 [[TMP2]], label [[LOOP_PRELOOP]], label [[PRELOOP_EXIT_SELECTOR:%.*]], !llvm.loop [[LOOP7:![0-9]+]], !loop_constrainer.loop.clone !6
158+
; CHECK: preloop.exit.selector:
159+
; CHECK-NEXT: [[IDX_DEC_PRELOOP_LCSSA:%.*]] = phi i32 [ [[IDX_DEC_PRELOOP]], [[IN_BOUNDS_PRELOOP]] ]
160+
; CHECK-NEXT: [[TMP3:%.*]] = icmp sgt i32 [[IDX_DEC_PRELOOP_LCSSA]], -1
161+
; CHECK-NEXT: br i1 [[TMP3]], label [[PRELOOP_PSEUDO_EXIT]], label [[EXIT_LOOPEXIT]]
162+
; CHECK: preloop.pseudo.exit:
163+
; CHECK-NEXT: [[IDX_PRELOOP_COPY]] = phi i32 [ [[INDVAR_START]], [[PREHEADER]] ], [ [[IDX_DEC_PRELOOP_LCSSA]], [[PRELOOP_EXIT_SELECTOR]] ]
164+
; CHECK-NEXT: [[INDVAR_END:%.*]] = phi i32 [ [[INDVAR_START]], [[PREHEADER]] ], [ [[IDX_DEC_PRELOOP_LCSSA]], [[PRELOOP_EXIT_SELECTOR]] ]
165+
; CHECK-NEXT: br label [[MAINLOOP]]
99166
;
100167
entry:
101168
%len = load i32, ptr %len_ptr, !range !0

0 commit comments

Comments
 (0)