[ConstraintElim] Handle switch cases with the same destination #76928
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
@llvm/pr-subscribers-llvm-transforms Author: Yingwei Zheng (dtcxzyw) ChangesThe original implementation only adds the fact The original motivation of this patch is to optimize a pattern in cvc5. For example:
Compile-time impact: http://llvm-compile-time-tracker.com/compare.php?from=8c72ff716b3e4b298695fa3faf6add860c6dbcb2&to=24a93e6c04b60b5d7d567379b8a0cd082c6554b0&stat=instructions%3Au
Full diff: https://github.com/llvm/llvm-project/pull/76928.diff 2 Files Affected:
diff --git a/llvm/lib/Transforms/Scalar/ConstraintElimination.cpp b/llvm/lib/Transforms/Scalar/ConstraintElimination.cpp
index cc93c8617c05e2..59c23ea5993f6f 100644
--- a/llvm/lib/Transforms/Scalar/ConstraintElimination.cpp
+++ b/llvm/lib/Transforms/Scalar/ConstraintElimination.cpp
@@ -1057,13 +1057,49 @@ void State::addInfoFor(BasicBlock &BB) {
}
if (auto *Switch = dyn_cast<SwitchInst>(BB.getTerminator())) {
- for (auto &Case : Switch->cases()) {
- BasicBlock *Succ = Case.getCaseSuccessor();
- Value *V = Case.getCaseValue();
- if (!canAddSuccessor(BB, Succ))
+ DenseMap<BasicBlock *, SmallVector<ConstantInt *>> BBMap;
+ for (auto &Case : Switch->cases())
+ BBMap[Case.getCaseSuccessor()].push_back(Case.getCaseValue());
+ // TODO: handle default dest
+ BBMap.erase(Switch->getDefaultDest());
+ Value *Cond = Switch->getCondition();
+ Type *Ty = Switch->getCondition()->getType();
+ unsigned BitWidth = Ty->getScalarSizeInBits();
+ for (auto &[Succ, Values] : BBMap) {
+ if (Succ->getUniquePredecessor() != &BB)
continue;
- WorkList.emplace_back(FactOrCheck::getConditionFact(
- DT.getNode(Succ), CmpInst::ICMP_EQ, Switch->getCondition(), V));
+ DomTreeNode *Node = DT.getNode(Succ);
+ if (Values.size() == 1)
+ WorkList.emplace_back(FactOrCheck::getConditionFact(
+ Node, CmpInst::ICMP_EQ, Cond, Values.back()));
+ else {
+ APInt SignedMin = APInt::getSignedMaxValue(BitWidth);
+ APInt SignedMax = APInt::getSignedMinValue(BitWidth);
+ APInt UnsignedMin = APInt::getMaxValue(BitWidth);
+ APInt UnsignedMax = APInt::getMinValue(BitWidth);
+
+ for (ConstantInt *V : Values) {
+ const APInt &C = V->getValue();
+ SignedMin = APIntOps::smin(SignedMin, C);
+ SignedMax = APIntOps::smax(SignedMax, C);
+ UnsignedMin = APIntOps::umin(UnsignedMin, C);
+ UnsignedMax = APIntOps::umax(UnsignedMax, C);
+ }
+ if (!SignedMin.isMinSignedValue())
+ WorkList.emplace_back(FactOrCheck::getConditionFact(
+ Node, CmpInst::ICMP_SGE, Cond, ConstantInt::get(Ty, SignedMin)));
+ if (!SignedMax.isMaxSignedValue())
+ WorkList.emplace_back(FactOrCheck::getConditionFact(
+ Node, CmpInst::ICMP_SLE, Cond, ConstantInt::get(Ty, SignedMax)));
+ if (!UnsignedMin.isMinValue())
+ WorkList.emplace_back(
+ FactOrCheck::getConditionFact(Node, CmpInst::ICMP_UGE, Cond,
+ ConstantInt::get(Ty, UnsignedMin)));
+ if (!UnsignedMax.isMaxValue())
+ WorkList.emplace_back(
+ FactOrCheck::getConditionFact(Node, CmpInst::ICMP_ULE, Cond,
+ ConstantInt::get(Ty, UnsignedMax)));
+ }
}
return;
}
diff --git a/llvm/test/Transforms/ConstraintElimination/switch.ll b/llvm/test/Transforms/ConstraintElimination/switch.ll
index b104b26e83cff2..fd581d081b4d3b 100644
--- a/llvm/test/Transforms/ConstraintElimination/switch.ll
+++ b/llvm/test/Transforms/ConstraintElimination/switch.ll
@@ -151,10 +151,8 @@ define i1 @switch_same_destination_for_different_cases(i8 %x) {
; CHECK: exit.2:
; CHECK-NEXT: [[C_3:%.*]] = icmp ult i8 [[X]], 7
; CHECK-NEXT: call void @use(i1 [[C_3]])
-; CHECK-NEXT: [[C_4:%.*]] = icmp ult i8 [[X]], 6
-; CHECK-NEXT: call void @use(i1 [[C_4]])
-; CHECK-NEXT: [[C_5:%.*]] = icmp ult i8 [[X]], 11
-; CHECK-NEXT: call void @use(i1 [[C_5]])
+; CHECK-NEXT: call void @use(i1 false)
+; CHECK-NEXT: call void @use(i1 true)
; CHECK-NEXT: [[C_6:%.*]] = icmp ult i8 [[X]], 10
; CHECK-NEXT: ret i1 [[C_6]]
;
@@ -181,4 +179,115 @@ exit.2:
ret i1 %c.6
}
+define i1 @test_switch_with_same_dest(i32 %a) {
+; CHECK-LABEL: @test_switch_with_same_dest(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: [[B:%.*]] = and i32 [[A:%.*]], 1023
+; CHECK-NEXT: switch i32 [[B]], label [[SW_DEFAULT:%.*]] [
+; CHECK-NEXT: i32 37, label [[SW_BB:%.*]]
+; CHECK-NEXT: i32 38, label [[SW_BB]]
+; CHECK-NEXT: ]
+; CHECK: sw.bb:
+; CHECK-NEXT: ret i1 false
+; CHECK: sw.default:
+; CHECK-NEXT: ret i1 false
+;
+entry:
+ %b = and i32 %a, 1023
+ switch i32 %b, label %sw.default [
+ i32 37, label %sw.bb
+ i32 38, label %sw.bb
+ ]
+
+sw.bb:
+ %cmp = icmp eq i32 %b, 1023
+ ret i1 %cmp
+sw.default:
+ ret i1 false
+}
+
+define i1 @test_switch_with_same_dest_zext(i16 %a) {
+; CHECK-LABEL: @test_switch_with_same_dest_zext(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: [[B:%.*]] = and i16 [[A:%.*]], 1023
+; CHECK-NEXT: [[B_EXT:%.*]] = zext nneg i16 [[B]] to i32
+; CHECK-NEXT: switch i32 [[B_EXT]], label [[SW_DEFAULT:%.*]] [
+; CHECK-NEXT: i32 37, label [[SW_BB:%.*]]
+; CHECK-NEXT: i32 38, label [[SW_BB]]
+; CHECK-NEXT: ]
+; CHECK: sw.bb:
+; CHECK-NEXT: ret i1 false
+; CHECK: sw.default:
+; CHECK-NEXT: ret i1 false
+;
+entry:
+ %b = and i16 %a, 1023
+ %b.ext = zext nneg i16 %b to i32
+ switch i32 %b.ext, label %sw.default [
+ i32 37, label %sw.bb
+ i32 38, label %sw.bb
+ ]
+
+sw.bb:
+ %cmp = icmp eq i16 %b, 1023
+ ret i1 %cmp
+sw.default:
+ ret i1 false
+}
+
+; Negative tests
+
+define i1 @test_switch_with_same_dest_fail1(i32 %a) {
+; CHECK-LABEL: @test_switch_with_same_dest_fail1(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: [[B:%.*]] = and i32 [[A:%.*]], 1023
+; CHECK-NEXT: switch i32 [[B]], label [[SW_BB:%.*]] [
+; CHECK-NEXT: i32 37, label [[SW_BB]]
+; CHECK-NEXT: i32 38, label [[SW_BB]]
+; CHECK-NEXT: ]
+; CHECK: sw.bb:
+; CHECK-NEXT: [[CMP:%.*]] = icmp eq i32 [[B]], 1023
+; CHECK-NEXT: ret i1 [[CMP]]
+;
+entry:
+ %b = and i32 %a, 1023
+ switch i32 %b, label %sw.bb [
+ i32 37, label %sw.bb
+ i32 38, label %sw.bb
+ ]
+
+sw.bb:
+ %cmp = icmp eq i32 %b, 1023
+ ret i1 %cmp
+}
+
+define i1 @test_switch_with_same_dest_fail2(i32 %a, i1 %cond) {
+; CHECK-LABEL: @test_switch_with_same_dest_fail2(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: [[B:%.*]] = and i32 [[A:%.*]], 1023
+; CHECK-NEXT: br i1 [[COND:%.*]], label [[IF_THEN:%.*]], label [[SW_BB:%.*]]
+; CHECK: if.then:
+; CHECK-NEXT: switch i32 [[B]], label [[SW_BB]] [
+; CHECK-NEXT: i32 37, label [[SW_BB]]
+; CHECK-NEXT: i32 38, label [[SW_BB]]
+; CHECK-NEXT: ]
+; CHECK: sw.bb:
+; CHECK-NEXT: [[CMP:%.*]] = icmp eq i32 [[B]], 1023
+; CHECK-NEXT: ret i1 [[CMP]]
+;
+entry:
+ %b = and i32 %a, 1023
+ br i1 %cond, label %if.then, label %sw.bb
+
+if.then:
+ switch i32 %b, label %sw.bb [
+ i32 37, label %sw.bb
+ i32 38, label %sw.bb
+ ]
+
+sw.bb:
+ %cmp = icmp eq i32 %b, 1023
+ ret i1 %cmp
+}
+
declare void @use(i1)
|
dtcxzyw
added a commit
to dtcxzyw/llvm-opt-benchmark
that referenced
this pull request
Jan 4, 2024
nikic
reviewed
Jan 4, 2024
We have handled this in CVP: llvm-project/llvm/lib/Analysis/LazyValueInfo.cpp Lines 1428 to 1475 in ce61b0e However, it cannot fix the original issue because the condition of switch is godbolt: https://godbolt.org/z/38TMcv4xE |
|
@dtcxzyw Can we solve this by folding |
dtcxzyw
added a commit
that referenced
this pull request
Jan 5, 2024
This patch folds `switch(zext/sext(X))` into `switch(X)`.
The original motivation of this patch is to optimize a pattern found in
cvc5. For example:
```
%bf.load.i = load i16, ptr %d_kind.i, align 8
%bf.clear.i = and i16 %bf.load.i, 1023
%bf.cast.i = zext nneg i16 %bf.clear.i to i32
switch i32 %bf.cast.i, label %if.else [
i32 335, label %if.then
i32 303, label %if.then
]
if.then: ; preds = %entry, %entry
%d_children.i.i = getelementptr inbounds %"class.cvc5::internal::expr::NodeValue", ptr %0, i64 0, i32 3
%cmp.i.i.i.i.i = icmp eq i16 %bf.clear.i, 1023
%cond.i.i.i.i.i = select i1 %cmp.i.i.i.i.i, i32 -1, i32 %bf.cast.i
```
`%cmp.i.i.i.i.i` always evaluates to false because `%bf.clear.i` can
only be 335 or 303.
Folding `switch i32 %bf.cast.i` to `switch i16 %bf.clear.i` will help
`CVP` to handle this case.
See also
#76928 (comment).
Compile-time impact:
http://llvm-compile-time-tracker.com/compare.php?from=7954c57124b495fbdc73674d71f2e366e4afe522&to=502b13ed34e561d995ae1f724cf06d20008bd86f&stat=instructions:u
|stage1-O3|stage1-ReleaseThinLTO|stage1-ReleaseLTO-g|stage1-O0-g|stage2-O3|stage2-O0-g|stage2-clang|
|--|--|--|--|--|--|--|
|+0.03%|+0.06%|+0.07%|+0.00%|-0.02%|-0.03%|+0.02%|
dtcxzyw
added a commit
to dtcxzyw/llvm-opt-benchmark
that referenced
this pull request
Jan 6, 2024
|
Is this still needed now that the other changes have landed? |
Looks like there are other cases which are not covered by #76988.
|
justinfargnoli
pushed a commit
to justinfargnoli/llvm-project
that referenced
this pull request
Jan 28, 2024
This patch folds `switch(zext/sext(X))` into `switch(X)`.
The original motivation of this patch is to optimize a pattern found in
cvc5. For example:
```
%bf.load.i = load i16, ptr %d_kind.i, align 8
%bf.clear.i = and i16 %bf.load.i, 1023
%bf.cast.i = zext nneg i16 %bf.clear.i to i32
switch i32 %bf.cast.i, label %if.else [
i32 335, label %if.then
i32 303, label %if.then
]
if.then: ; preds = %entry, %entry
%d_children.i.i = getelementptr inbounds %"class.cvc5::internal::expr::NodeValue", ptr %0, i64 0, i32 3
%cmp.i.i.i.i.i = icmp eq i16 %bf.clear.i, 1023
%cond.i.i.i.i.i = select i1 %cmp.i.i.i.i.i, i32 -1, i32 %bf.cast.i
```
`%cmp.i.i.i.i.i` always evaluates to false because `%bf.clear.i` can
only be 335 or 303.
Folding `switch i32 %bf.cast.i` to `switch i16 %bf.clear.i` will help
`CVP` to handle this case.
See also
llvm#76928 (comment).
Compile-time impact:
http://llvm-compile-time-tracker.com/compare.php?from=7954c57124b495fbdc73674d71f2e366e4afe522&to=502b13ed34e561d995ae1f724cf06d20008bd86f&stat=instructions:u
|stage1-O3|stage1-ReleaseThinLTO|stage1-ReleaseLTO-g|stage1-O0-g|stage2-O3|stage2-O0-g|stage2-clang|
|--|--|--|--|--|--|--|
|+0.03%|+0.06%|+0.07%|+0.00%|-0.02%|-0.03%|+0.02%|
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
The original implementation only adds the fact
cond == Cif BB dominates DestBB and there is a unique edge from BB to DestBB.This patch adds more range information for switch cases with the same destination if BB is the unique predecessor of DestBB.
The original motivation of this patch is to optimize a pattern in cvc5. For example:
%cmp.i.i.i.i.ialways evaluates to false because%bf.clear.ican only be 335 or 303.Compile-time impact: http://llvm-compile-time-tracker.com/compare.php?from=8c72ff716b3e4b298695fa3faf6add860c6dbcb2&to=24a93e6c04b60b5d7d567379b8a0cd082c6554b0&stat=instructions%3Au