Skip to content

[InstCombine] Fold (X * (Y << K)) u>> K -> X * Y when highbits are not demanded #111151

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
dtcxzyw merged 2 commits into from
Dec 3, 2024
Merged

[InstCombine] Fold (X * (Y << K)) u>> K -> X * Y when highbits are not demanded #111151

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
5bee7a7
[InstCombine] Add pre-commit tests. NFC.
dtcxzyw Oct 4, 2024
7de2345
[InstCombine] Fold `(X * (Y << K)) u>> K -> X * Y` when highbits are …
dtcxzyw Oct 4, 2024
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
9 changes: 9 additions & 0 deletions llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -770,6 +770,15 @@ Value *InstCombinerImpl::SimplifyDemandedUseBits(Instruction *I,
return InsertNewInstWith(Shl, I->getIterator());
}
}

const APInt *Factor;
if (match(I->getOperand(0),
m_OneUse(m_Mul(m_Value(X), m_APInt(Factor)))) &&
Factor->countr_zero() >= ShiftAmt) {
BinaryOperator *Mul = BinaryOperator::CreateMul(
Copy link
Contributor

@goldsteinn goldsteinn Oct 4, 2024

Choose a reason for hiding this comment

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

You can preserve nuw and nsw iff you have nuw:
https://alive2.llvm.org/ce/z/P9NudG

Copy link
Member Author

@dtcxzyw dtcxzyw Oct 4, 2024

Choose a reason for hiding this comment

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

(X *nuw (Y << K)) u>> K has been handled in other places: https://godbolt.org/z/43W4vE8bK

Copy link
Contributor

@goldsteinn goldsteinn Oct 4, 2024

Choose a reason for hiding this comment

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

Can we put the two impls in the same place?

Copy link
Contributor

@nikic nikic Dec 2, 2024

Choose a reason for hiding this comment

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

I don't think the two patterns would combine well. In one case we can optimize because there is no overflow. In the other case we can optimize because the bits affected by overflow are not demanded.

X, ConstantInt::get(X->getType(), Factor->lshr(ShiftAmt)));
return InsertNewInstWith(Mul, I->getIterator());
}
}

// Unsigned shift right.
Expand Down
62 changes: 62 additions & 0 deletions llvm/test/Transforms/InstCombine/lshr.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1523,3 +1523,65 @@ define <2 x i8> @bool_add_lshr_vec_wrong_shift_amt(<2 x i1> %a, <2 x i1> %b) {
%lshr = lshr <2 x i8> %add, <i8 1, i8 2>
ret <2 x i8> %lshr
}

define i32 @lowbits_of_lshr_mul(i64 %x) {
; CHECK-LABEL: @lowbits_of_lshr_mul(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[TMP0:%.*]] = trunc i64 [[X:%.*]] to i32
; CHECK-NEXT: [[CONV:%.*]] = mul i32 [[TMP0]], 15
; CHECK-NEXT: ret i32 [[CONV]]
;
entry:
%mul = mul i64 %x, 64424509440
%shift = lshr i64 %mul, 32
%conv = trunc i64 %shift to i32
ret i32 %conv
}

define i32 @lowbits_of_lshr_mul_mask(i32 %x) {
; CHECK-LABEL: @lowbits_of_lshr_mul_mask(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[TMP0:%.*]] = mul i32 [[X:%.*]], 1600
; CHECK-NEXT: [[CONV:%.*]] = and i32 [[TMP0]], 32704
; CHECK-NEXT: ret i32 [[CONV]]
;
entry:
%mul = mul i32 %x, 104857600
%shift = lshr i32 %mul, 16
%conv = and i32 %shift, 32767
ret i32 %conv
}

; Negative tests

define i32 @lowbits_of_lshr_mul_mask_multiuse(i32 %x) {
; CHECK-LABEL: @lowbits_of_lshr_mul_mask_multiuse(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[MUL:%.*]] = mul i32 [[X:%.*]], 104857600
; CHECK-NEXT: call void @use(i32 [[MUL]])
; CHECK-NEXT: [[SHIFT:%.*]] = lshr exact i32 [[MUL]], 16
; CHECK-NEXT: [[CONV:%.*]] = and i32 [[SHIFT]], 32704
; CHECK-NEXT: ret i32 [[CONV]]
;
entry:
%mul = mul i32 %x, 104857600
call void @use(i32 %mul)
%shift = lshr i32 %mul, 16
%conv = and i32 %shift, 32767
ret i32 %conv
}

define i32 @lowbits_of_lshr_mul_mask_indivisible(i32 %x) {
; CHECK-LABEL: @lowbits_of_lshr_mul_mask_indivisible(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[MUL:%.*]] = mul i32 [[X:%.*]], 25600
; CHECK-NEXT: [[SHIFT:%.*]] = lshr i32 [[MUL]], 16
; CHECK-NEXT: [[CONV:%.*]] = and i32 [[SHIFT]], 32767
; CHECK-NEXT: ret i32 [[CONV]]
;
entry:
%mul = mul i32 %x, 25600
%shift = lshr i32 %mul, 16
%conv = and i32 %shift, 32767
ret i32 %conv
}
Loading