Skip to content

[InstCombine] Fold (X / C) < X and (X >> C) < X into X > 0 #85555

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
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
34 changes: 34 additions & 0 deletions llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -7103,6 +7103,40 @@ Instruction *InstCombinerImpl::foldICmpCommutative(ICmpInst::Predicate Pred,
if (Value *V = foldICmpWithLowBitMaskedVal(Pred, Op0, Op1, Q, *this))
return replaceInstUsesWith(CxtI, V);

// Folding (X / Y) pred X => X swap(pred) 0 for constant Y other than 0 or 1
{
const APInt *Divisor;
if (match(Op0, m_UDiv(m_Specific(Op1), m_APInt(Divisor))) &&
Divisor->ugt(1)) {
return new ICmpInst(ICmpInst::getSwappedPredicate(Pred), Op1,
Constant::getNullValue(Op1->getType()));
}

if (!ICmpInst::isUnsigned(Pred) &&
match(Op0, m_SDiv(m_Specific(Op1), m_APInt(Divisor))) &&
Divisor->ugt(1)) {
return new ICmpInst(ICmpInst::getSwappedPredicate(Pred), Op1,
Constant::getNullValue(Op1->getType()));
}
}

// Another case of this fold is (X >> Y) pred X => X swap(pred) 0 if Y != 0
{
const APInt *Shift;
if (match(Op0, m_LShr(m_Specific(Op1), m_APInt(Shift))) &&
!Shift->isZero()) {
return new ICmpInst(ICmpInst::getSwappedPredicate(Pred), Op1,
Constant::getNullValue(Op1->getType()));
}

if ((Pred == CmpInst::ICMP_SLT || Pred == CmpInst::ICMP_SGE) &&
match(Op0, m_AShr(m_Specific(Op1), m_APInt(Shift))) &&
!Shift->isZero()) {
return new ICmpInst(ICmpInst::getSwappedPredicate(Pred), Op1,
Constant::getNullValue(Op1->getType()));
}
}
Copy link
Contributor

Choose a reason for hiding this comment

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

nit: There is a bit of redundancy in that each of the return cases is identical, but at least imo its fine.


return nullptr;
}

Expand Down
141 changes: 141 additions & 0 deletions llvm/test/Transforms/InstCombine/icmp-div-constant.ll
Original file line number Diff line number Diff line change
Expand Up @@ -375,3 +375,144 @@ define i1 @sdiv_eq_smin_use(i32 %x, i32 %y) {
%r = icmp eq i32 %d, -2147483648
ret i1 %r
}

; Fold (X / C) cmp X into X ~cmp 0 (~cmp is the inverse predicate of cmp), for some C != 1
; Alternative form of this fold is when division is replaced with logic right shift

define i1 @sdiv_x_by_const_cmp_x(i32 %x) {
; CHECK-LABEL: @sdiv_x_by_const_cmp_x(
; CHECK-NEXT: [[TMP1:%.*]] = icmp eq i32 [[X:%.*]], 0
; CHECK-NEXT: ret i1 [[TMP1]]
;
%v = sdiv i32 %x, 13
%r = icmp eq i32 %v, %x
ret i1 %r
}

define i1 @udiv_x_by_const_cmp_x(i32 %x) {
; CHECK-LABEL: @udiv_x_by_const_cmp_x(
; CHECK-NEXT: [[TMP1:%.*]] = icmp sgt i32 [[X:%.*]], 0
; CHECK-NEXT: ret i1 [[TMP1]]
;
%1 = udiv i32 %x, 123
%2 = icmp slt i32 %1, %x
ret i1 %2
}

; Same as above but with right shift instead of division (C != 0)

define i1 @lshr_x_by_const_cmp_x(i32 %x) {
; CHECK-LABEL: @lshr_x_by_const_cmp_x(
; CHECK-NEXT: [[TMP1:%.*]] = icmp eq i32 [[X:%.*]], 0
; CHECK-NEXT: ret i1 [[TMP1]]
;
%v = lshr i32 %x, 1
%r = icmp eq i32 %v, %x
ret i1 %r
}

define <4 x i1> @lshr_by_const_cmp_sle_value(<4 x i32> %x) {
; CHECK-LABEL: @lshr_by_const_cmp_sle_value(
; CHECK-NEXT: [[R:%.*]] = icmp sgt <4 x i32> [[X:%.*]], <i32 -1, i32 -1, i32 -1, i32 -1>
; CHECK-NEXT: ret <4 x i1> [[R]]
;
%v = lshr <4 x i32> %x, <i32 3, i32 3, i32 3, i32 3>
%r = icmp sle <4 x i32> %v, %x
ret <4 x i1> %r
}

define i1 @lshr_by_const_cmp_sge_value(i32 %x) {
; CHECK-LABEL: @lshr_by_const_cmp_sge_value(
; CHECK-NEXT: [[R:%.*]] = icmp slt i32 [[X:%.*]], 1
; CHECK-NEXT: ret i1 [[R]]
;
%v = lshr i32 %x, 3
%r = icmp sge i32 %v, %x
ret i1 %r
}

define i1 @ashr_x_by_const_cmp_sge_x(i32 %x) {
; CHECK-LABEL: @ashr_x_by_const_cmp_sge_x(
; CHECK-NEXT: [[R:%.*]] = icmp slt i32 [[X:%.*]], 1
; CHECK-NEXT: ret i1 [[R]]
;
%v = ashr i32 %x, 5
%r = icmp sge i32 %v, %x
ret i1 %r
}

; Negative test - constant is 1

define <2 x i1> @udiv_x_by_const_cmp_eq_value_neg(<2 x i32> %x) {
; CHECK-LABEL: @udiv_x_by_const_cmp_eq_value_neg(
; CHECK-NEXT: [[V:%.*]] = udiv <2 x i32> [[X:%.*]], <i32 1, i32 3>
; CHECK-NEXT: [[R:%.*]] = icmp eq <2 x i32> [[V]], [[X]]
; CHECK-NEXT: ret <2 x i1> [[R]]
;
%v = udiv <2 x i32> %x, <i32 1, i32 3>
%r = icmp eq <2 x i32> %v, %x
ret <2 x i1> %r
}

define <2 x i1> @sdiv_x_by_const_cmp_eq_value_neg(<2 x i32> %x) {
; CHECK-LABEL: @sdiv_x_by_const_cmp_eq_value_neg(
; CHECK-NEXT: [[V:%.*]] = sdiv <2 x i32> [[X:%.*]], <i32 1, i32 3>
; CHECK-NEXT: [[R:%.*]] = icmp eq <2 x i32> [[V]], [[X]]
; CHECK-NEXT: ret <2 x i1> [[R]]
;
%v = sdiv <2 x i32> %x, <i32 1, i32 3>
%r = icmp eq <2 x i32> %v, %x
ret <2 x i1> %r
}

; Negative test - constant is 0

define <2 x i1> @lshr_x_by_const_cmp_slt_value_neg(<2 x i32> %x) {
; CHECK-LABEL: @lshr_x_by_const_cmp_slt_value_neg(
; CHECK-NEXT: [[V:%.*]] = lshr <2 x i32> [[X:%.*]], <i32 0, i32 2>
; CHECK-NEXT: [[R:%.*]] = icmp slt <2 x i32> [[V]], [[X]]
; CHECK-NEXT: ret <2 x i1> [[R]]
;
%v = lshr <2 x i32> %x, <i32 0, i32 2>
%r = icmp slt <2 x i32> %v, %x
ret <2 x i1> %r
}

; Negative test - unsigned predicate with sdiv

define i1 @sdiv_x_by_const_cmp_ult_value_neg(i32 %x) {
; CHECK-LABEL: @sdiv_x_by_const_cmp_ult_value_neg(
; CHECK-NEXT: [[V:%.*]] = sdiv i32 [[X:%.*]], 3
; CHECK-NEXT: [[R:%.*]] = icmp ult i32 [[V]], [[X]]
; CHECK-NEXT: ret i1 [[R]]
;
%v = sdiv i32 %x, 3
%r = icmp ult i32 %v, %x
ret i1 %r
}

; Negative case - one of the components of a vector is 1

define <4 x i1> @sdiv_x_by_const_cmp_sgt_value_neg(<4 x i32> %x) {
; CHECK-LABEL: @sdiv_x_by_const_cmp_sgt_value_neg(
; CHECK-NEXT: [[V:%.*]] = sdiv <4 x i32> [[X:%.*]], <i32 1, i32 2, i32 3, i32 4>
; CHECK-NEXT: [[R:%.*]] = icmp sgt <4 x i32> [[V]], [[X]]
; CHECK-NEXT: ret <4 x i1> [[R]]
;
%v = sdiv <4 x i32> %x, <i32 1, i32 2, i32 3, i32 4>
%r = icmp sgt <4 x i32> %v, %x
ret <4 x i1> %r
}

; Negative case - ashr only allows sge/slt predicates

define i1 @ashr_x_by_const_cmp_sle_value_neg(i32 %x) {
; CHECK-LABEL: @ashr_x_by_const_cmp_sle_value_neg(
; CHECK-NEXT: [[V:%.*]] = ashr i32 [[X:%.*]], 3
; CHECK-NEXT: [[R:%.*]] = icmp sle i32 [[V]], [[X]]
; CHECK-NEXT: ret i1 [[R]]
;
%v = ashr i32 %x, 3
%r = icmp sle i32 %v, %x
ret i1 %r
}
Loading