-
Notifications
You must be signed in to change notification settings - Fork 13.6k
[LoopPeel] Use loop guards when checking if last iter can be peeled. #142605
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
base: main
Are you sure you want to change the base?
Conversation
@llvm/pr-subscribers-llvm-transforms Author: Florian Hahn (fhahn) ChangesApply loop guards to BTC before checking if the last iteration should be peeled off. This also adds an assert to make sure applying the guards does not pessimize the results. I checked on a large test set and it did not trigger there, but it adds an additional guard to catch potential cases where loop-guards pessimize results. Peels ~15% more loops. Full diff: https://github.com/llvm/llvm-project/pull/142605.diff 2 Files Affected:
diff --git a/llvm/lib/Transforms/Utils/LoopPeel.cpp b/llvm/lib/Transforms/Utils/LoopPeel.cpp
index bd025fddd0cf7..564c2298f6f2f 100644
--- a/llvm/lib/Transforms/Utils/LoopPeel.cpp
+++ b/llvm/lib/Transforms/Utils/LoopPeel.cpp
@@ -357,17 +357,14 @@ bool llvm::canPeelLastIteration(const Loop &L, ScalarEvolution &SE) {
m_scev_AffineAddRec(m_SCEV(), m_scev_One(), m_SpecificLoop(&L)));
}
-/// Returns true if the last iteration can be peeled off and the condition (Pred
-/// LeftAR, RightSCEV) is known at the last iteration and the inverse condition
-/// is known at the second-to-last.
+/// Returns true if the last iteration should be peeled off, i.e. the condition
+/// (Pred LeftAR, RightSCEV) is known at the last iteration and the inverse
+/// condition is known at the second-to-last.
static bool shouldPeelLastIteration(Loop &L, CmpPredicate Pred,
const SCEVAddRecExpr *LeftAR,
- const SCEV *RightSCEV, ScalarEvolution &SE,
+ const SCEV *RightSCEV, const SCEV *BTC,
+ ScalarEvolution &SE,
const TargetTransformInfo &TTI) {
- if (!canPeelLastIteration(L, SE))
- return false;
-
- const SCEV *BTC = SE.getBackedgeTakenCount(&L);
SCEVExpander Expander(SE, L.getHeader()->getDataLayout(), "loop-peel");
if (!SE.isKnownNonZero(BTC) &&
Expander.isHighCostExpansion(BTC, &L, SCEVCheapExpansionBudget, &TTI,
@@ -377,7 +374,6 @@ static bool shouldPeelLastIteration(Loop &L, CmpPredicate Pred,
const SCEV *ValAtLastIter = LeftAR->evaluateAtIteration(BTC, SE);
const SCEV *ValAtSecondToLastIter = LeftAR->evaluateAtIteration(
SE.getMinusSCEV(BTC, SE.getOne(BTC->getType())), SE);
-
return SE.isKnownPredicate(ICmpInst::getInversePredicate(Pred), ValAtLastIter,
RightSCEV) &&
SE.isKnownPredicate(Pred, ValAtSecondToLastIter, RightSCEV);
@@ -484,8 +480,19 @@ countToEliminateCompares(Loop &L, unsigned MaxPeelCount, ScalarEvolution &SE,
const SCEV *Step = LeftAR->getStepRecurrence(SE);
if (!PeelWhilePredicateIsKnown(NewPeelCount, IterVal, RightSCEV, Step,
Pred)) {
- if (shouldPeelLastIteration(L, Pred, LeftAR, RightSCEV, SE, TTI))
+ if (!canPeelLastIteration(L, SE))
+ return;
+
+ const SCEV *BTC = SE.getBackedgeTakenCount(&L);
+ auto Guards = ScalarEvolution::LoopGuards::collect(&L, SE);
+ if (shouldPeelLastIteration(L, Pred, LeftAR,
+ SE.applyLoopGuards(RightSCEV, Guards),
+ SE.applyLoopGuards(BTC, Guards), SE, TTI))
DesiredPeelCountLast = 1;
+ else
+ assert(!shouldPeelLastIteration(L, Pred, LeftAR, RightSCEV, BTC, SE,
+ TTI) &&
+ "loop guards pessimized result");
return;
}
diff --git a/llvm/test/Transforms/LoopUnroll/peel-last-iteration-with-guards.ll b/llvm/test/Transforms/LoopUnroll/peel-last-iteration-with-guards.ll
index af07a97131322..824e23fcf3e6e 100644
--- a/llvm/test/Transforms/LoopUnroll/peel-last-iteration-with-guards.ll
+++ b/llvm/test/Transforms/LoopUnroll/peel-last-iteration-with-guards.ll
@@ -13,14 +13,33 @@ define void @peel_with_guard_known_nonnegative_1(i32 %n) {
; CHECK-NEXT: [[N_EXT:%.*]] = zext i32 [[N]] to i64
; CHECK-NEXT: [[N_1:%.*]] = add i32 [[N]], 1
; CHECK-NEXT: [[WIDE_TRIP_COUNT:%.*]] = zext i32 [[N_1]] to i64
+; CHECK-NEXT: [[TMP0:%.*]] = add nsw i64 [[WIDE_TRIP_COUNT]], -1
+; CHECK-NEXT: [[TMP1:%.*]] = icmp ne i64 [[TMP0]], 0
+; CHECK-NEXT: br i1 [[TMP1]], label %[[PH_SPLIT:.*]], label %[[EXIT_LOOPEXIT_PEEL_BEGIN:.*]]
+; CHECK: [[PH_SPLIT]]:
; CHECK-NEXT: br label %[[LOOP:.*]]
; CHECK: [[LOOP]]:
-; CHECK-NEXT: [[IV:%.*]] = phi i64 [ 0, %[[PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ]
+; CHECK-NEXT: [[IV1:%.*]] = phi i64 [ 0, %[[PH_SPLIT]] ], [ [[IV_NEXT1:%.*]], %[[LOOP]] ]
+; CHECK-NEXT: [[IV_NEXT1]] = add nuw nsw i64 [[IV1]], 1
+; CHECK-NEXT: [[TMP2:%.*]] = sub i64 [[WIDE_TRIP_COUNT]], 1
+; CHECK-NEXT: [[EC1:%.*]] = icmp eq i64 [[IV_NEXT1]], [[TMP2]]
+; CHECK-NEXT: br i1 [[EC1]], label %[[EXIT_LOOPEXIT_PEEL_BEGIN_LOOPEXIT:.*]], label %[[LOOP]], !llvm.loop [[LOOP0:![0-9]+]]
+; CHECK: [[EXIT_LOOPEXIT_PEEL_BEGIN_LOOPEXIT]]:
+; CHECK-NEXT: [[DOTPH:%.*]] = phi i64 [ [[IV_NEXT1]], %[[LOOP]] ]
+; CHECK-NEXT: br label %[[EXIT_LOOPEXIT_PEEL_BEGIN]]
+; CHECK: [[EXIT_LOOPEXIT_PEEL_BEGIN]]:
+; CHECK-NEXT: [[IV:%.*]] = phi i64 [ 0, %[[PH]] ], [ [[DOTPH]], %[[EXIT_LOOPEXIT_PEEL_BEGIN_LOOPEXIT]] ]
+; CHECK-NEXT: br label %[[LOOP_PEEL:.*]]
+; CHECK: [[LOOP_PEEL]]:
; CHECK-NEXT: [[C:%.*]] = icmp eq i64 [[IV]], [[N_EXT]]
; CHECK-NEXT: [[SPEC_SELECT:%.*]] = select i1 [[C]], i32 10, i32 20
-; CHECK-NEXT: [[IV_NEXT]] = add i64 [[IV]], 1
+; CHECK-NEXT: [[IV_NEXT:%.*]] = add i64 [[IV]], 1
; CHECK-NEXT: [[EC:%.*]] = icmp eq i64 [[IV_NEXT]], [[WIDE_TRIP_COUNT]]
-; CHECK-NEXT: br i1 [[EC]], label %[[EXIT_LOOPEXIT:.*]], label %[[LOOP]]
+; CHECK-NEXT: br i1 [[EC]], label %[[EXIT_LOOPEXIT_PEEL_NEXT:.*]], label %[[EXIT_LOOPEXIT_PEEL_NEXT]]
+; CHECK: [[EXIT_LOOPEXIT_PEEL_NEXT]]:
+; CHECK-NEXT: br label %[[LOOP_PEEL_NEXT:.*]]
+; CHECK: [[LOOP_PEEL_NEXT]]:
+; CHECK-NEXT: br label %[[EXIT_LOOPEXIT:.*]]
; CHECK: [[EXIT_LOOPEXIT]]:
; CHECK-NEXT: br label %[[EXIT]]
; CHECK: [[EXIT]]:
@@ -137,7 +156,7 @@ define void @peel_with_guard2(i32 %n) {
; CHECK-NEXT: [[IV_NEXT]] = add nuw i32 [[IV]], 1
; CHECK-NEXT: [[TMP2:%.*]] = sub i32 [[N]], 1
; CHECK-NEXT: [[EC:%.*]] = icmp eq i32 [[IV_NEXT]], [[TMP2]]
-; CHECK-NEXT: br i1 [[EC]], label %[[EXIT_LOOPEXIT_PEEL_BEGIN_LOOPEXIT:.*]], label %[[LOOP_HEADER]], !llvm.loop [[LOOP0:![0-9]+]]
+; CHECK-NEXT: br i1 [[EC]], label %[[EXIT_LOOPEXIT_PEEL_BEGIN_LOOPEXIT:.*]], label %[[LOOP_HEADER]], !llvm.loop [[LOOP2:![0-9]+]]
; CHECK: [[EXIT_LOOPEXIT_PEEL_BEGIN_LOOPEXIT]]:
; CHECK-NEXT: [[DOTPH:%.*]] = phi i32 [ [[IV_NEXT]], %[[LOOP_LATCH]] ]
; CHECK-NEXT: br label %[[EXIT_LOOPEXIT_PEEL_BEGIN]]
@@ -188,4 +207,5 @@ exit:
;.
; CHECK: [[LOOP0]] = distinct !{[[LOOP0]], [[META1:![0-9]+]]}
; CHECK: [[META1]] = !{!"llvm.loop.peeled.count", i32 1}
+; CHECK: [[LOOP2]] = distinct !{[[LOOP2]], [[META1]]}
;.
|
else | ||
assert(!shouldPeelLastIteration(L, Pred, LeftAR, RightSCEV, BTC, SE, | ||
TTI) && | ||
"loop guards pessimized result"); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I don't think this assertion should exist. Based on past experience, you can always construct edge cases where this happens.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Yeah fair enough, I removed it. Still wondering if there's a nice way to have some kind of verification as opt-in?
if (!canPeelLastIteration(L, SE)) | ||
return false; | ||
|
||
const SCEV *BTC = SE.getBackedgeTakenCount(&L); | ||
SCEVExpander Expander(SE, L.getHeader()->getDataLayout(), "loop-peel"); | ||
if (!SE.isKnownNonZero(BTC) && | ||
Expander.isHighCostExpansion(BTC, &L, SCEVCheapExpansionBudget, &TTI, |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The high cost expansion check should probably not use the BTC with loop guards? We're not going to actually expand it with the loop guards, right?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Yep loop-guards aren't used during expansion, removed for now. Some of the info may be used by later passes to simplifiy the expansion.
Apply loop guards to BTC before checking if the last iteration should be peeled off. This also adds an assert to make sure applying the guards does not pessimize the results. I checked on a large test set and it did not trigger there, but it adds an additional guard to catch potential cases where loop-guards pessimize results. Peels ~15% more loops.
8e0bea9
to
4b4a394
Compare
4b4a394
to
c7deb11
Compare
Apply loop guards to BTC before checking if the last iteration should be peeled off. This also adds an assert to make sure applying the guards does not pessimize the results. I checked on a large test set and it did not trigger there, but it adds an additional guard to catch potential cases where loop-guards pessimize results.
Peels ~15% more loops.