Skip to content

[SCEV] Fold ((-1 * C1) * D / C1) -> -1 * D. #157555

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.

Sign up for GitHub

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 4 commits into from
Sep 10, 2025
Merged

[SCEV] Fold ((-1 * C1) * D / C1) -> -1 * D. #157555

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
b635047
[SCEV] Fold ((-1 * C1) * D / C1) -> -1 * D.
fhahn Sep 8, 2025
0c1705c
!fixup adjust comment, thanks
fhahn Sep 9, 2025
44844c0
Merge branch 'main' into scev-mul-div-neg
fhahn Sep 10, 2025
ecc0ad5
Merge branch 'main' into scev-mul-div-neg
fhahn Sep 10, 2025
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
16 changes: 10 additions & 6 deletions llvm/lib/Analysis/ScalarEvolution.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3217,15 +3217,19 @@ const SCEV *ScalarEvolution::getMulExpr(SmallVectorImpl<const SCEV *> &Ops,
}

// Try to fold (C1 * D /u C2) -> C1/C2 * D, if C1 and C2 are powers-of-2,
// D is a multiple of C2, and C1 is a multiple of C1.
// D is a multiple of C2, and C1 is a multiple of C2.
const SCEV *D;
APInt C1V = LHSC->getAPInt();
// If C1 is negative, try (-1 * abs(C1)) instead.
Copy link
Collaborator

@preames preames Sep 9, 2025

Choose a reason for hiding this comment

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

Adjust the comment to something like:
(C1 * D /u C2) == -1 * -C1 * D /u C2 when C1 != INT_MIN

Though maybe that's not quite right because that equality still holds (just not in a useful way) when C1 is INT_MIN.

Basically, we need something here with more of indication of why it's correct.

Copy link
Contributor Author

@fhahn fhahn Sep 9, 2025

Choose a reason for hiding this comment

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

Updated, thanks!

if (C1V.isNegative() && !C1V.isMinSignedValue())
Copy link
Contributor

@nikic nikic Sep 10, 2025

Choose a reason for hiding this comment

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

Why is the INT_MIN check here needed? Isn't abs() just a no-op for that?

Copy link
Contributor

@nikic nikic Sep 10, 2025 •
edited
Loading

Choose a reason for hiding this comment

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

Ah, I guess this avoid doing an unnecessary getNegativeSCEV() lateron. No, that's based on the equality comparison, so it shouldn't matter if we do an extra abs() here.

Copy link
Contributor Author

@fhahn fhahn Sep 10, 2025

Choose a reason for hiding this comment

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

Yeah it's a no-op, we would end up with (-INT_MIN * D /u C2) == (-1 * -INT_MIN * D /u C2). Can change to use abs() if you prefer and remove the check?

Copy link
Contributor Author

@fhahn fhahn Sep 10, 2025

Choose a reason for hiding this comment

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

I'll leave it as-is for now, when this gets generalized to non-constants we can adjust it

C1V = C1V.abs();
const SCEVConstant *C2;
const APInt &LHSV = LHSC->getAPInt();
if (LHSV.isPowerOf2() &&
if (C1V.isPowerOf2() &&
match(Ops[1], m_scev_UDiv(m_SCEV(D), m_SCEVConstant(C2))) &&
C2->getAPInt().isPowerOf2() && LHSV.uge(C2->getAPInt()) &&
LHSV.logBase2() <= getMinTrailingZeros(D)) {
return getMulExpr(getUDivExpr(LHSC, C2), D);
C2->getAPInt().isPowerOf2() && C1V.uge(C2->getAPInt()) &&
C1V.logBase2() <= getMinTrailingZeros(D)) {
const SCEV *NewMul = getMulExpr(getUDivExpr(getConstant(C1V), C2), D);
return C1V == LHSC->getAPInt() ? NewMul : getNegativeSCEV(NewMul);
}
}
}
Expand Down
2 changes: 1 addition & 1 deletion llvm/test/Analysis/ScalarEvolution/mul-udiv-folds.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,7 +23,7 @@ define void @udiv4_and_udiv2(i1 %c, ptr %A) {
; CHECK-NEXT: %gep.16 = getelementptr i16, ptr %A, i64 %iv
; CHECK-NEXT: --> {((2 * ((zext i32 %start to i64) /u 4))<nuw><nsw> + %A),+,2}<%loop> U: full-set S: full-set Exits: ((zext i32 %start to i64) + %A) LoopDispositions: { %loop: Computable }
; CHECK-NEXT: %gep.32 = getelementptr i32, ptr %A, i64 %iv
; CHECK-NEXT: --> {((zext i32 %start to i64) + %A),+,4}<%loop> U: full-set S: full-set Exits: ((3 * (zext i32 %start to i64))<nuw><nsw> + (-4 * ((zext i32 %start to i64) /u 4))<nsw> + %A) LoopDispositions: { %loop: Computable }
; CHECK-NEXT: --> {((zext i32 %start to i64) + %A),+,4}<%loop> U: full-set S: full-set Exits: ((2 * (zext i32 %start to i64))<nuw><nsw> + %A) LoopDispositions: { %loop: Computable }
; CHECK-NEXT: %gep.40 = getelementptr <{ i32, i8 }>, ptr %A, i64 %iv
; CHECK-NEXT: --> {((5 * ((zext i32 %start to i64) /u 4))<nuw><nsw> + %A),+,5}<%loop> U: full-set S: full-set Exits: ((5 * ((zext i32 %start to i64) /u 2))<nuw><nsw> + %A) LoopDispositions: { %loop: Computable }
; CHECK-NEXT: %gep.48 = getelementptr <{ i32, i16 }>, ptr %A, i64 %iv
Expand Down