Skip to content

[InstCombine] Extend #125676 to handle variable power of 2 #125855

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

Merged
merged 3 commits into from
Feb 6, 2025
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
10 changes: 5 additions & 5 deletions llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -4200,14 +4200,14 @@ Value *InstCombinerImpl::foldXorOfICmps(ICmpInst *LHS, ICmpInst *RHS,

// Fold (icmp eq/ne (X & Pow2), 0) ^ (icmp eq/ne (Y & Pow2), 0) into
// (icmp eq/ne ((X ^ Y) & Pow2), 0)
Value *X, *Y;
const APInt *Mask;
Value *X, *Y, *Pow2;
if (ICmpInst::isEquality(PredL) && ICmpInst::isEquality(PredR) &&
LC->isZero() && RC->isZero() && LHS->hasOneUse() && RHS->hasOneUse() &&
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

can it not be beneficial to make the fold if either LHS or RHS have one use? it will result in the same amount of instructions but the other argument will have one less use.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The cmp that has one use would also need its op to have one use i.e (LHS->hasOneUse() && LHS0->hasOneUse()) || (RHS->hasOneUse() && RHS0->hasOneUse()). And that would only make it break even.

match(LHS0, m_And(m_Value(X), m_Power2(Mask))) &&
match(RHS0, m_And(m_Value(Y), m_SpecificInt(*Mask)))) {
match(LHS0, m_And(m_Value(X), m_Value(Pow2))) &&
match(RHS0, m_And(m_Value(Y), m_Specific(Pow2))) &&
isKnownToBeAPowerOfTwo(Pow2, /*OrZero=*/true, /*Depth=*/0, &I)) {
Value *Xor = Builder.CreateXor(X, Y);
Value *And = Builder.CreateAnd(Xor, *Mask);
Value *And = Builder.CreateAnd(Xor, Pow2);
return Builder.CreateICmp(PredL == PredR ? ICmpInst::ICMP_NE
: ICmpInst::ICMP_EQ,
And, ConstantInt::getNullValue(Xor->getType()));
Expand Down
36 changes: 36 additions & 0 deletions llvm/test/Transforms/InstCombine/xor-icmps.ll
Original file line number Diff line number Diff line change
Expand Up @@ -335,6 +335,42 @@ define i1 @test_xor_of_bittest_ne_ne(i8 %x, i8 %y) {
ret i1 %xor
}

define i1 @test_xor_of_bittest_ne_ne_var_pow2(i8 %x, i8 %y, i8 %shamt) {
; CHECK-LABEL: @test_xor_of_bittest_ne_ne_var_pow2(
; CHECK-NEXT: [[POW2:%.*]] = shl nuw i8 1, [[SHAMT:%.*]]
; CHECK-NEXT: [[Y:%.*]] = xor i8 [[X:%.*]], [[Y1:%.*]]
; CHECK-NEXT: [[MASK2:%.*]] = and i8 [[Y]], [[POW2]]
; CHECK-NEXT: [[CMP2:%.*]] = icmp ne i8 [[MASK2]], 0
; CHECK-NEXT: ret i1 [[CMP2]]
;
%pow2 = shl nuw i8 1, %shamt
%mask1 = and i8 %x, %pow2
%cmp1 = icmp ne i8 %mask1, 0
%mask2 = and i8 %y, %pow2
%cmp2 = icmp ne i8 %mask2, 0
%xor = xor i1 %cmp1, %cmp2
ret i1 %xor
}

define i1 @test_xor_of_bittest_ne_ne_var_pow2_or_zero(i8 %x, i8 %y, i8 %z) {
; CHECK-LABEL: @test_xor_of_bittest_ne_ne_var_pow2_or_zero(
; CHECK-NEXT: [[NZ:%.*]] = sub i8 0, [[Z:%.*]]
; CHECK-NEXT: [[TMP1:%.*]] = xor i8 [[X:%.*]], [[Y:%.*]]
; CHECK-NEXT: [[TMP2:%.*]] = and i8 [[TMP1]], [[NZ]]
; CHECK-NEXT: [[MASK2:%.*]] = and i8 [[TMP2]], [[Z]]
; CHECK-NEXT: [[CMP2:%.*]] = icmp ne i8 [[MASK2]], 0
; CHECK-NEXT: ret i1 [[CMP2]]
;
%nz = sub i8 0, %z
%pow2 = and i8 %z, %nz
%mask1 = and i8 %x, %pow2
%cmp1 = icmp ne i8 %mask1, 0
%mask2 = and i8 %y, %pow2
%cmp2 = icmp ne i8 %mask2, 0
%xor = xor i1 %cmp1, %cmp2
ret i1 %xor
}

define i1 @test_xor_of_bittest_eq_eq(i8 %x, i8 %y) {
; CHECK-LABEL: @test_xor_of_bittest_eq_eq(
; CHECK-NEXT: [[Y:%.*]] = xor i8 [[X:%.*]], [[Y1:%.*]]
Expand Down