Skip to content

[InstCombine] Fold the bound check idiom into sign bit test #76439

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

Open
wants to merge 3 commits into
base: main
Choose a base branch
from
Open

[InstCombine] Fold the bound check idiom into sign bit test #76439

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
578c296
[InstCombine] Add pre-commit tests. NFC.
dtcxzyw Dec 27, 2023
3a386e3
[InstCombine] Fold the bound check idiom into sign bit test
dtcxzyw Dec 27, 2023
e55b9de
[InstCombine] Add one-use constraint for the simplification
dtcxzyw Dec 27, 2023
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
20 changes: 20 additions & 0 deletions llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2641,6 +2641,26 @@ Instruction *InstCombinerImpl::foldICmpDivConstant(ICmpInst &Cmp,
if (!match(Y, m_APInt(C2)))
return nullptr;

// Fold the bound check idiom:
// T *end, *start;
// (size_t)(end - start) > (size_t)(PTRDIFF_MAX / sizeof(T))
// into:
// (ptrdiff_t)(end - start) < 0
// i.e.:
// icmp ugt (sdiv exact X, C2), (sdiv signed_max, C2) --> icmp slt X, 0
// icmp ult (sdiv exact X, C2), (sdiv signed_max, C2) + 1 --> icmp sgt X, -1
// where C2 is positive.
if (DivIsSigned && Div->hasOneUse() && Div->isExact() &&
(Pred == ICmpInst::ICMP_UGT || Pred == ICmpInst::ICMP_ULT) &&
C2->isStrictlyPositive() &&
APInt::getSignedMaxValue(C2->getBitWidth()).sdiv(*C2) +
APInt(C2->getBitWidth(), Pred == ICmpInst::ICMP_UGT ? 0 : 1) ==
C)
Copy link
Contributor

Choose a reason for hiding this comment

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

Why the special case to signed max? Doesn't this hold pretty generally? https://alive2.llvm.org/ce/z/zv2kqx (The ugt would become slt 0 for the right boundary.)

Copy link
Member Author

Choose a reason for hiding this comment

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

The ugt would become slt 0 for the right boundary.

We will fold icmp ugt i64 (sdiv exact i64 %x, 24), 384307168202282325 into icmp ugt i64 %x, 9223372036854775800. The latter form cannot be simplified into icmp slt i64 %x, 0.
Alive2: https://alive2.llvm.org/ce/z/WEPmRN

Copy link
Contributor

Choose a reason for hiding this comment

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

As the division is exact, there are multiple legal constants that can be used (we can add up to C2-1 assuming it does not overflow). We could special case this to produce a better constant for the sign boundary. Or just leave the ugt, as this is still much better than the division, while covering more cases.

return new ICmpInst(
Pred == ICmpInst::ICMP_UGT ? ICmpInst::ICMP_SLT : ICmpInst::ICMP_SGT, X,
Pred == ICmpInst::ICMP_UGT ? Constant::getNullValue(X->getType())
: Constant::getAllOnesValue(X->getType()));

// FIXME: If the operand types don't match the type of the divide
// then don't attempt this transform. The code below doesn't have the
// logic to deal with a signed divide and an unsigned compare (and
Expand Down
109 changes: 108 additions & 1 deletion llvm/test/Transforms/InstCombine/icmp.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5006,7 +5006,6 @@ define i1 @or_positive_sgt_zero_multi_use(i8 %a) {
ret i1 %cmp
}


define i1 @disjoint_or_sgt_1(i8 %a, i8 %b) {
; CHECK-LABEL: @disjoint_or_sgt_1(
; CHECK-NEXT: [[B1:%.*]] = add nsw i8 [[B:%.*]], 2
Expand Down Expand Up @@ -5138,3 +5137,111 @@ entry:
%cmp = icmp eq i8 %add2, %add1
ret i1 %cmp
}

define i1 @icmp_ugt_sdiv_by_constant(i64 %x) {
; CHECK-LABEL: @icmp_ugt_sdiv_by_constant(
; CHECK-NEXT: [[CMP:%.*]] = icmp slt i64 [[X:%.*]], 0
; CHECK-NEXT: ret i1 [[CMP]]
;
%sdiv = sdiv exact i64 %x, 24
%cmp = icmp ugt i64 %sdiv, 384307168202282325
ret i1 %cmp
}

define i1 @icmp_ult_sdiv_by_constant(i64 %x) {
; CHECK-LABEL: @icmp_ult_sdiv_by_constant(
; CHECK-NEXT: [[CMP:%.*]] = icmp sgt i64 [[X:%.*]], -1
; CHECK-NEXT: ret i1 [[CMP]]
;
%sdiv = sdiv exact i64 %x, 24
%cmp = icmp ult i64 %sdiv, 384307168202282326
ret i1 %cmp
}

; TODO: This should be simplified to icmp slt i64 %x, 0
define i1 @icmp_ugt_ashr_by_constant(i64 %x) {
; CHECK-LABEL: @icmp_ugt_ashr_by_constant(
; CHECK-NEXT: [[CMP:%.*]] = icmp ugt i64 [[X:%.*]], 9223372036854775804
; CHECK-NEXT: ret i1 [[CMP]]
;
%sdiv = ashr exact i64 %x, 2
%cmp = icmp ugt i64 %sdiv, 2305843009213693951
ret i1 %cmp
}

define i1 @icmp_ult_ashr_by_constant(i64 %x) {
; CHECK-LABEL: @icmp_ult_ashr_by_constant(
; CHECK-NEXT: [[CMP:%.*]] = icmp sgt i64 [[X:%.*]], -1
; CHECK-NEXT: ret i1 [[CMP]]
;
%sdiv = ashr exact i64 %x, 2
%cmp = icmp ult i64 %sdiv, 2305843009213693952
ret i1 %cmp
}

; Negative tests
define i1 @icmp_ugt_sdiv_by_constant_without_exact(i64 %x) {
; CHECK-LABEL: @icmp_ugt_sdiv_by_constant_without_exact(
; CHECK-NEXT: [[SDIV:%.*]] = sdiv i64 [[X:%.*]], 24
; CHECK-NEXT: [[CMP:%.*]] = icmp ugt i64 [[SDIV]], 384307168202282325
; CHECK-NEXT: ret i1 [[CMP]]
;
%sdiv = sdiv i64 %x, 24
%cmp = icmp ugt i64 %sdiv, 384307168202282325
ret i1 %cmp
}

define i1 @icmp_ugt_udiv_by_constant(i64 %x) {
; CHECK-LABEL: @icmp_ugt_udiv_by_constant(
; CHECK-NEXT: [[CMP:%.*]] = icmp ugt i64 [[X:%.*]], 9223372036854775800
; CHECK-NEXT: ret i1 [[CMP]]
;
%sdiv = udiv exact i64 %x, 24
%cmp = icmp ugt i64 %sdiv, 384307168202282325
ret i1 %cmp
}

define i1 @icmp_ne_sdiv_by_constant(i64 %x) {
; CHECK-LABEL: @icmp_ne_sdiv_by_constant(
; CHECK-NEXT: [[CMP:%.*]] = icmp ne i64 [[X:%.*]], 9223372036854775800
; CHECK-NEXT: ret i1 [[CMP]]
;
%sdiv = sdiv exact i64 %x, 24
%cmp = icmp ne i64 %sdiv, 384307168202282325
ret i1 %cmp
}

define i1 @icmp_ugt_sdiv_by_constant_wrong_rhs(i64 %x) {
; CHECK-LABEL: @icmp_ugt_sdiv_by_constant_wrong_rhs(
; CHECK-NEXT: [[SDIV:%.*]] = sdiv exact i64 [[X:%.*]], 24
; CHECK-NEXT: [[CMP:%.*]] = icmp ugt i64 [[SDIV]], 384307168202282324
; CHECK-NEXT: ret i1 [[CMP]]
;
%sdiv = sdiv exact i64 %x, 24
%cmp = icmp ugt i64 %sdiv, 384307168202282324
ret i1 %cmp
}

define i1 @icmp_ugt_sdiv_by_negative_constant(i64 %x) {
; CHECK-LABEL: @icmp_ugt_sdiv_by_negative_constant(
; CHECK-NEXT: [[SDIV:%.*]] = sdiv i64 [[X:%.*]], -24
; CHECK-NEXT: [[CMP:%.*]] = icmp ugt i64 [[SDIV]], -384307168202282326
; CHECK-NEXT: ret i1 [[CMP]]
;
%sdiv = sdiv i64 %x, -24
%cmp = icmp ugt i64 %sdiv, -384307168202282326
ret i1 %cmp
}

define i1 @icmp_ugt_sdiv_by_constant_multiuse(i64 %x) {
; CHECK-LABEL: @icmp_ugt_sdiv_by_constant_multiuse(
; CHECK-NEXT: [[SDIV:%.*]] = sdiv exact i64 [[X:%.*]], 24
; CHECK-NEXT: call void @use_i64(i64 [[SDIV]])
; CHECK-NEXT: [[CMP:%.*]] = icmp ugt i64 [[SDIV]], 384307168202282325
; CHECK-NEXT: ret i1 [[CMP]]
;
%sdiv = sdiv exact i64 %x, 24
call void @use_i64(i64 %sdiv)
%cmp = icmp ugt i64 %sdiv, 384307168202282325
ret i1 %cmp
}