[AggressiveInstcombine] Fold away shift in or reduction chain. #137875
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: David Green (davemgreen) ChangesIf we have This is also true of ne, and true for longer or chains. Thinking out loud, this is kind of like a "any-bit is demanded" combine. I'm not sure if that exists already. Full diff: https://github.com/llvm/llvm-project/pull/137875.diff 2 Files Affected:
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index b7b0bb7361359..d2aa957b10b78 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -5046,6 +5046,29 @@ static Instruction *foldICmpOrXX(ICmpInst &I, const SimplifyQuery &Q,
return nullptr;
}
+static Value *foldShiftAwayFromOrChain(Instruction &I,
+ InstCombiner::BuilderTy &Builder) {
+ if (I.getOpcode() != Instruction::Or)
+ return nullptr;
+ Value *A, *B;
+ if (match(&I, m_c_Or(m_CombineOr(m_NSWShl(m_Value(A), m_Value()),
+ m_NUWShl(m_Value(A), m_Value())),
+ m_Value(B))))
+ return Builder.CreateOr(A, B);
+
+ Value *Op0 = I.getOperand(0);
+ if (isa<Instruction>(Op0))
+ if (auto *X = foldShiftAwayFromOrChain(*cast<Instruction>(Op0), Builder))
+ Op0 = X;
+ Value *Op1 = I.getOperand(1);
+ if (isa<Instruction>(Op1))
+ if (auto *X = foldShiftAwayFromOrChain(*cast<Instruction>(Op1), Builder))
+ Op1 = X;
+ if (Op0 != I.getOperand(0) || Op1 != I.getOperand(1))
+ return Builder.CreateOr(Op0, Op1);
+ return nullptr;
+}
+
static Instruction *foldICmpXorXX(ICmpInst &I, const SimplifyQuery &Q,
InstCombinerImpl &IC) {
Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1), *A;
@@ -7742,6 +7765,11 @@ Instruction *InstCombinerImpl::visitICmpInst(ICmpInst &I) {
}
}
+ // icmp eq/ne or(shl(a), b), 0 -> icmp eq/ne or(a, b)
+ if (I.isEquality() && match(Op1, m_Zero()) && isa<Instruction>(Op0))
+ if (auto *Res = foldShiftAwayFromOrChain(*cast<Instruction>(Op0), Builder))
+ return new ICmpInst(I.getPredicate(), Res, Op1);
+
return Changed ? &I : nullptr;
}
diff --git a/llvm/test/Transforms/InstCombine/icmp-of-or-x.ll b/llvm/test/Transforms/InstCombine/icmp-of-or-x.ll
index 993325f6ff0b0..baaa754e52894 100644
--- a/llvm/test/Transforms/InstCombine/icmp-of-or-x.ll
+++ b/llvm/test/Transforms/InstCombine/icmp-of-or-x.ll
@@ -409,3 +409,112 @@ define i1 @PR38139(i8 %arg) {
%r = icmp ne i8 %masked, %arg
ret i1 %r
}
+
+define i1 @remove_shift_nuw_ab(i8 %a, i8 %b, i8 %s) {
+; CHECK-LABEL: @remove_shift_nuw_ab(
+; CHECK-NEXT: [[OR:%.*]] = or i8 [[T:%.*]], [[B:%.*]]
+; CHECK-NEXT: [[IC:%.*]] = icmp eq i8 [[OR]], 0
+; CHECK-NEXT: ret i1 [[IC]]
+;
+ %t = shl nuw i8 %a, %s
+ %or = or i8 %t, %b
+ %ic = icmp eq i8 %or, 0
+ ret i1 %ic
+}
+
+define i1 @remove_shift_nuw_ba(i8 %a, i8 %b, i8 %s) {
+; CHECK-LABEL: @remove_shift_nuw_ba(
+; CHECK-NEXT: [[OR:%.*]] = or i8 [[B:%.*]], [[T:%.*]]
+; CHECK-NEXT: [[IC:%.*]] = icmp eq i8 [[OR]], 0
+; CHECK-NEXT: ret i1 [[IC]]
+;
+ %t = shl nuw i8 %a, %s
+ %or = or i8 %b, %t
+ %ic = icmp eq i8 %or, 0
+ ret i1 %ic
+}
+
+define i1 @remove_shift_nsw(i8 %a, i8 %b, i8 %s) {
+; CHECK-LABEL: @remove_shift_nsw(
+; CHECK-NEXT: [[OR:%.*]] = or i8 [[T:%.*]], [[B:%.*]]
+; CHECK-NEXT: [[IC:%.*]] = icmp eq i8 [[OR]], 0
+; CHECK-NEXT: ret i1 [[IC]]
+;
+ %t = shl nsw i8 %a, %s
+ %or = or i8 %t, %b
+ %ic = icmp eq i8 %or, 0
+ ret i1 %ic
+}
+
+define i1 @remove_shift_nuw_ne(i8 %a, i8 %b, i8 %s) {
+; CHECK-LABEL: @remove_shift_nuw_ne(
+; CHECK-NEXT: [[OR:%.*]] = or i8 [[T:%.*]], [[B:%.*]]
+; CHECK-NEXT: [[IC:%.*]] = icmp eq i8 [[OR]], 0
+; CHECK-NEXT: ret i1 [[IC]]
+;
+ %t = shl nuw i8 %a, %s
+ %or = or i8 %t, %b
+ %ic = icmp eq i8 %or, 0
+ ret i1 %ic
+}
+
+define i1 @remove_shift_nsw_ne(i8 %a, i8 %b, i8 %s) {
+; CHECK-LABEL: @remove_shift_nsw_ne(
+; CHECK-NEXT: [[OR:%.*]] = or i8 [[T:%.*]], [[B:%.*]]
+; CHECK-NEXT: [[IC:%.*]] = icmp eq i8 [[OR]], 0
+; CHECK-NEXT: ret i1 [[IC]]
+;
+ %t = shl nsw i8 %a, %s
+ %or = or i8 %t, %b
+ %ic = icmp eq i8 %or, 0
+ ret i1 %ic
+}
+
+define i1 @remove_shift_wraps(i8 %a, i8 %b, i8 %s) {
+; CHECK-LABEL: @remove_shift_wraps(
+; CHECK-NEXT: [[T:%.*]] = shl i8 [[A:%.*]], [[S:%.*]]
+; CHECK-NEXT: [[OR:%.*]] = or i8 [[T]], [[B:%.*]]
+; CHECK-NEXT: [[IC:%.*]] = icmp eq i8 [[OR]], 0
+; CHECK-NEXT: ret i1 [[IC]]
+;
+ %t = shl i8 %a, %s
+ %or = or i8 %t, %b
+ %ic = icmp eq i8 %or, 0
+ ret i1 %ic
+}
+
+define i1 @remove_shift_chain_d(i8 %a, i8 %b, i8 %c, i8 %d, i8 %s) {
+; CHECK-LABEL: @remove_shift_chain_d(
+; CHECK-NEXT: [[OR1:%.*]] = or i8 [[A:%.*]], [[B:%.*]]
+; CHECK-NEXT: [[OR2:%.*]] = or i8 [[C:%.*]], [[DT:%.*]]
+; CHECK-NEXT: [[OR:%.*]] = or i8 [[OR1]], [[OR2]]
+; CHECK-NEXT: [[IC:%.*]] = icmp eq i8 [[OR]], 0
+; CHECK-NEXT: ret i1 [[IC]]
+;
+ %dt = shl nuw i8 %d, %s
+ %or1 = or i8 %a, %b
+ %or2 = or i8 %c, %dt
+ %or = or i8 %or1, %or2
+ %ic = icmp eq i8 %or, 0
+ ret i1 %ic
+}
+
+define i1 @remove_shift_chain_abcd(i8 %a, i8 %b, i8 %c, i8 %d, i8 %s) {
+; CHECK-LABEL: @remove_shift_chain_abcd(
+; CHECK-NEXT: [[OR1:%.*]] = or i8 [[AT:%.*]], [[BT:%.*]]
+; CHECK-NEXT: [[OR2:%.*]] = or i8 [[CT:%.*]], [[DT:%.*]]
+; CHECK-NEXT: [[OR:%.*]] = or i8 [[OR1]], [[OR2]]
+; CHECK-NEXT: [[IC:%.*]] = icmp eq i8 [[OR]], 0
+; CHECK-NEXT: ret i1 [[IC]]
+;
+ %at = shl nuw i8 %a, %s
+ %bt = shl nuw i8 %b, 2
+ %ct = shl nuw i8 %c, 1
+ %dt = shl nuw i8 %d, %s
+ %or1 = or i8 %at, %bt
+ %or2 = or i8 %ct, %dt
+ %or = or i8 %or1, %or2
+ %ic = icmp eq i8 %or, 0
+ ret i1 %ic
+}
+
|
dtcxzyw
reviewed
Apr 30, 2025
dtcxzyw
reviewed
Apr 30, 2025
b2d83d9 to
4bc6794
Compare
This was referenced May 1, 2025
dtcxzyw
reviewed
May 1, 2025
There was a problem hiding this comment.
I am fine with this fold itself. But it prevents AggressiveInstCombine from merging consecutive loads: https://godbolt.org/z/aq94MeM3n
We should not perform this fold when the shift's LHS is zext(load).
If you think this workaround is too fragile, an alternative solution is to convert (load iN p) | (load iN p+N/8) == 0 into (load i2N p) == 0 in AggressiveInstCombine.
4bc6794 to
fc1184a
Compare
|
I've tried to move this into aggressive instcombine to allow it to check for trees of loads first. The transform is looking up through trees already, so maybe that is a better place for it. Let me know what you think. |
This was referenced May 8, 2025
…37875 If we have icmp eq or(a, shl(b)), 0 then the shift can be removed so long as it is nuw or nsw. It is still comparing that some bits are non-zero. https://alive2.llvm.org/ce/z/nhrBVX. This is also true of ne, and true for longer or chains.
fc1184a to
57ff4e8
Compare
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.
If we have
icmp eq or(a, shl(b)), 0then the shift can be removed so long as it is nuw or nsw. It is still comparing that some bits are non-zero.https://alive2.llvm.org/ce/z/nhrBVX.
This is also true of ne, and true for longer or chains.