[InstCombine] Use existing (icmp pred (add x, y), C) folds with or disjoint
#87937
+72
−49
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…or x, y), 0)` `(icmp eq/ne (or x, y), 0)` is probably easier to analyze than `(icmp eq/ne x, -y)` Proof: https://alive2.llvm.org/ce/z/2-VTb6
… disjoint` No significant changes, just some missed plumbing.
goldsteinn
changed the title
(icmp pred (add x, y), C) folds with or disjoint
|
@llvm/pr-subscribers-llvm-transforms Author: None (goldsteinn) Changes
Full diff: https://github.com/llvm/llvm-project/pull/87937.diff 3 Files Affected:
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index db302d7e526844..f805fc21f97de6 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -2986,10 +2986,10 @@ Instruction *InstCombinerImpl::foldICmpAddConstant(ICmpInst &Cmp,
Value *Op0, *Op1;
Instruction *Ext0, *Ext1;
const CmpInst::Predicate Pred = Cmp.getPredicate();
- if (match(Add,
- m_Add(m_CombineAnd(m_Instruction(Ext0), m_ZExtOrSExt(m_Value(Op0))),
- m_CombineAnd(m_Instruction(Ext1),
- m_ZExtOrSExt(m_Value(Op1))))) &&
+ if (match(Add, m_AddLike(m_CombineAnd(m_Instruction(Ext0),
+ m_ZExtOrSExt(m_Value(Op0))),
+ m_CombineAnd(m_Instruction(Ext1),
+ m_ZExtOrSExt(m_Value(Op1))))) &&
Op0->getType()->isIntOrIntVectorTy(1) &&
Op1->getType()->isIntOrIntVectorTy(1)) {
unsigned BW = C.getBitWidth();
@@ -3021,9 +3021,9 @@ Instruction *InstCombinerImpl::foldICmpAddConstant(ICmpInst &Cmp,
// If the add does not wrap, we can always adjust the compare by subtracting
// the constants. Equality comparisons are handled elsewhere. SGE/SLE/UGE/ULE
// are canonicalized to SGT/SLT/UGT/ULT.
- if ((Add->hasNoSignedWrap() &&
+ if ((match(Add, m_NSWAddLike(m_Value(), m_Value())) &&
(Pred == ICmpInst::ICMP_SGT || Pred == ICmpInst::ICMP_SLT)) ||
- (Add->hasNoUnsignedWrap() &&
+ (match(Add, m_NUWAddLike(m_Value(), m_Value())) &&
(Pred == ICmpInst::ICMP_UGT || Pred == ICmpInst::ICMP_ULT))) {
bool Overflow;
APInt NewC =
@@ -3438,6 +3438,20 @@ Instruction *InstCombinerImpl::foldICmpBinOpEqualityWithConstant(
}
}
break;
+ case Instruction::Or: {
+ const APInt *BOC;
+ if (match(BOp1, m_APInt(BOC)) && BO->hasOneUse() && RHS->isAllOnesValue()) {
+ // Comparing if all bits outside of a constant mask are set?
+ // Replace (X | C) == -1 with (X & ~C) == ~C.
+ // This removes the -1 constant.
+ Constant *NotBOC = ConstantExpr::getNot(cast<Constant>(BOp1));
+ Value *And = Builder.CreateAnd(BOp0, NotBOC);
+ return new ICmpInst(Pred, And, NotBOC);
+ }
+ if (!cast<PossiblyDisjointInst>(BO)->isDisjoint())
+ break;
+ [[fallthrough]];
+ }
case Instruction::Add: {
// (A + C2) == C --> A == (C - C2)
// (A + C2) != C --> A != (C - C2)
@@ -3452,7 +3466,12 @@ Instruction *InstCombinerImpl::foldICmpBinOpEqualityWithConstant(
return new ICmpInst(Pred, BOp0, NegVal);
if (Value *NegVal = dyn_castNegVal(BOp0))
return new ICmpInst(Pred, NegVal, BOp1);
- if (BO->hasOneUse()) {
+ if (BO->hasOneUse() && BO->getOpcode() != Instruction::Or) {
+ // (add nuw A, B) != 0 -> (or A, B) != 0
+ if (match(BO, m_NUWAdd(m_Value(), m_Value()))) {
+ Value *Or = Builder.CreateOr(BOp0, BOp1);
+ return new ICmpInst(Pred, Or, Constant::getNullValue(BO->getType()));
+ }
Value *Neg = Builder.CreateNeg(BOp1);
Neg->takeName(BO);
return new ICmpInst(Pred, BOp0, Neg);
@@ -3472,18 +3491,6 @@ Instruction *InstCombinerImpl::foldICmpBinOpEqualityWithConstant(
}
}
break;
- case Instruction::Or: {
- const APInt *BOC;
- if (match(BOp1, m_APInt(BOC)) && BO->hasOneUse() && RHS->isAllOnesValue()) {
- // Comparing if all bits outside of a constant mask are set?
- // Replace (X | C) == -1 with (X & ~C) == ~C.
- // This removes the -1 constant.
- Constant *NotBOC = ConstantExpr::getNot(cast<Constant>(BOp1));
- Value *And = Builder.CreateAnd(BOp0, NotBOC);
- return new ICmpInst(Pred, And, NotBOC);
- }
- break;
- }
case Instruction::UDiv:
case Instruction::SDiv:
if (BO->isExact()) {
@@ -3756,10 +3763,6 @@ Instruction *InstCombinerImpl::foldICmpBinOpWithConstant(ICmpInst &Cmp,
if (Instruction *I = foldICmpAndConstant(Cmp, BO, C))
return I;
break;
- case Instruction::Or:
- if (Instruction *I = foldICmpOrConstant(Cmp, BO, C))
- return I;
- break;
case Instruction::Mul:
if (Instruction *I = foldICmpMulConstant(Cmp, BO, C))
return I;
@@ -3789,6 +3792,12 @@ Instruction *InstCombinerImpl::foldICmpBinOpWithConstant(ICmpInst &Cmp,
if (Instruction *I = foldICmpSubConstant(Cmp, BO, C))
return I;
break;
+ case Instruction::Or:
+ if (Instruction *I = foldICmpOrConstant(Cmp, BO, C))
+ return I;
+ if (!cast<PossiblyDisjointInst>(BO)->isDisjoint())
+ break;
+ [[fallthrough]];
case Instruction::Add:
if (Instruction *I = foldICmpAddConstant(Cmp, BO, C))
return I;
diff --git a/llvm/test/Transforms/InstCombine/icmp-add.ll b/llvm/test/Transforms/InstCombine/icmp-add.ll
index b99ed20d7d431c..e71a3af4b5be4f 100644
--- a/llvm/test/Transforms/InstCombine/icmp-add.ll
+++ b/llvm/test/Transforms/InstCombine/icmp-add.ll
@@ -9,10 +9,8 @@ declare void @use(i32)
define i1 @cvt_icmp_0_zext_plus_zext_eq_i16(i16 %arg, i16 %arg1) {
; CHECK-LABEL: @cvt_icmp_0_zext_plus_zext_eq_i16(
; CHECK-NEXT: bb:
-; CHECK-NEXT: [[I:%.*]] = zext i16 [[ARG:%.*]] to i32
-; CHECK-NEXT: [[I2:%.*]] = zext i16 [[ARG1:%.*]] to i32
-; CHECK-NEXT: [[I3:%.*]] = sub nsw i32 0, [[I]]
-; CHECK-NEXT: [[I4:%.*]] = icmp eq i32 [[I2]], [[I3]]
+; CHECK-NEXT: [[TMP0:%.*]] = or i16 [[ARG1:%.*]], [[ARG:%.*]]
+; CHECK-NEXT: [[I4:%.*]] = icmp eq i16 [[TMP0]], 0
; CHECK-NEXT: ret i1 [[I4]]
;
bb:
@@ -27,10 +25,8 @@ bb:
define i1 @cvt_icmp_0_zext_plus_zext_eq_i8(i8 %arg, i8 %arg1) {
; CHECK-LABEL: @cvt_icmp_0_zext_plus_zext_eq_i8(
; CHECK-NEXT: bb:
-; CHECK-NEXT: [[I:%.*]] = zext i8 [[ARG:%.*]] to i32
-; CHECK-NEXT: [[I2:%.*]] = zext i8 [[ARG1:%.*]] to i32
-; CHECK-NEXT: [[I3:%.*]] = sub nsw i32 0, [[I]]
-; CHECK-NEXT: [[I4:%.*]] = icmp eq i32 [[I2]], [[I3]]
+; CHECK-NEXT: [[TMP0:%.*]] = or i8 [[ARG1:%.*]], [[ARG:%.*]]
+; CHECK-NEXT: [[I4:%.*]] = icmp eq i8 [[TMP0]], 0
; CHECK-NEXT: ret i1 [[I4]]
;
bb:
@@ -1802,22 +1798,17 @@ define i1 @test4(i32 %a) {
ret i1 %c
}
-define { i32, i1 } @test4multiuse(i32 %a) {
-; CHECK-LABEL: @test4multiuse(
-; CHECK-NEXT: [[B:%.*]] = add nsw i32 [[A:%.*]], -2147483644
-; CHECK-NEXT: [[C:%.*]] = icmp slt i32 [[A]], 2147483640
-; CHECK-NEXT: [[TMP:%.*]] = insertvalue { i32, i1 } undef, i32 [[B]], 0
-; CHECK-NEXT: [[RES:%.*]] = insertvalue { i32, i1 } [[TMP]], i1 [[C]], 1
-; CHECK-NEXT: ret { i32, i1 } [[RES]]
+define {
+i32, i1 } @test4multiuse(i32 %a) {
;
- %b = add nsw i32 %a, -2147483644
- %c = icmp slt i32 %b, -4
+%b = add nsw i32 %a, -2147483644
+%c = icmp slt i32 %b, -4
- %tmp = insertvalue { i32, i1 } undef, i32 %b, 0
- %res = insertvalue { i32, i1 } %tmp, i1 %c, 1
+%tmp = insertvalue { i32, i1 } undef, i32 %b, 0
+%res = insertvalue { i32, i1 } %tmp, i1 %c, 1
- ret { i32, i1 } %res
+ret { i32, i1 } %res
}
define <2 x i1> @test4vec(<2 x i32> %a) {
@@ -2857,7 +2848,7 @@ define i1 @icmp_add_add_C_comm2(i32 %X, i32 %b) {
; CHECK-NEXT: [[CMP:%.*]] = icmp ugt i32 [[A]], [[TMP1]]
; CHECK-NEXT: ret i1 [[CMP]]
;
- %a = udiv i32 42, %X ; thwart complexity-based canonicalization
+ %a = udiv i32 42, %X ; thwart complexity-based canonicalization
%add1 = add i32 %a, %b
%add2 = add i32 %add1, -1
%cmp = icmp ugt i32 %a, %add2
@@ -2871,7 +2862,7 @@ define i1 @icmp_add_add_C_comm2_pred(i32 %X, i32 %b) {
; CHECK-NEXT: [[CMP:%.*]] = icmp ule i32 [[A]], [[TMP1]]
; CHECK-NEXT: ret i1 [[CMP]]
;
- %a = udiv i32 42, %X ; thwart complexity-based canonicalization
+ %a = udiv i32 42, %X ; thwart complexity-based canonicalization
%add1 = add i32 %a, %b
%add2 = add i32 %add1, -1
%cmp = icmp ule i32 %a, %add2
@@ -2886,7 +2877,7 @@ define i1 @icmp_add_add_C_comm2_wrong_pred(i32 %X, i32 %b) {
; CHECK-NEXT: [[CMP:%.*]] = icmp ult i32 [[A]], [[ADD2]]
; CHECK-NEXT: ret i1 [[CMP]]
;
- %a = udiv i32 42, %X ; thwart complexity-based canonicalization
+ %a = udiv i32 42, %X ; thwart complexity-based canonicalization
%add1 = add i32 %a, %b
%add2 = add i32 %add1, -1
%cmp = icmp ult i32 %a, %add2
@@ -2900,7 +2891,7 @@ define i1 @icmp_add_add_C_comm3(i32 %X, i32 %b) {
; CHECK-NEXT: [[CMP:%.*]] = icmp ugt i32 [[A]], [[TMP1]]
; CHECK-NEXT: ret i1 [[CMP]]
;
- %a = udiv i32 42, %X ; thwart complexity-based canonicalization
+ %a = udiv i32 42, %X ; thwart complexity-based canonicalization
%add1 = add i32 %b, %a
%add2 = add i32 %add1, -1
%cmp = icmp ugt i32 %a, %add2
@@ -3003,4 +2994,28 @@ define i1 @icmp_dec_notnonzero(i8 %x) {
ret i1 %c
}
+define i1 @icmp_addnuw_nonzero(i8 %x, i8 %y) {
+; CHECK-LABEL: @icmp_addnuw_nonzero(
+; CHECK-NEXT: [[TMP1:%.*]] = or i8 [[X:%.*]], [[Y:%.*]]
+; CHECK-NEXT: [[C:%.*]] = icmp eq i8 [[TMP1]], 0
+; CHECK-NEXT: ret i1 [[C]]
+;
+ %i = add nuw i8 %x, %y
+ %c = icmp eq i8 %i, 0
+ ret i1 %c
+}
+
+define i1 @icmp_addnuw_nonzero_fail_multiuse(i32 %x, i32 %y) {
+; CHECK-LABEL: @icmp_addnuw_nonzero_fail_multiuse(
+; CHECK-NEXT: [[I:%.*]] = add nuw i32 [[X:%.*]], [[Y:%.*]]
+; CHECK-NEXT: [[C:%.*]] = icmp eq i32 [[I]], 0
+; CHECK-NEXT: call void @use(i32 [[I]])
+; CHECK-NEXT: ret i1 [[C]]
+;
+ %i = add nuw i32 %x, %y
+ %c = icmp eq i32 %i, 0
+ call void @use(i32 %i)
+ ret i1 %c
+}
+
declare void @llvm.assume(i1)
diff --git a/llvm/test/Transforms/InstCombine/known-bits.ll b/llvm/test/Transforms/InstCombine/known-bits.ll
index 769f7661fc8dc0..ba5619a25a49df 100644
--- a/llvm/test/Transforms/InstCombine/known-bits.ll
+++ b/llvm/test/Transforms/InstCombine/known-bits.ll
@@ -483,8 +483,7 @@ if.else:
define i1 @test_icmp_or_distjoint(i8 %n, i1 %other) {
; CHECK-LABEL: @test_icmp_or_distjoint(
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[N_OR:%.*]] = or disjoint i8 [[N:%.*]], 16
-; CHECK-NEXT: [[CMP:%.*]] = icmp ugt i8 [[N_OR]], -111
+; CHECK-NEXT: [[CMP:%.*]] = icmp ugt i8 [[N:%.*]], -127
; CHECK-NEXT: br i1 [[CMP]], label [[IF_THEN:%.*]], label [[IF_ELSE:%.*]]
; CHECK: if.then:
; CHECK-NEXT: ret i1 true
|
dtcxzyw
reviewed
Apr 8, 2024
| m_Add(m_CombineAnd(m_Instruction(Ext0), m_ZExtOrSExt(m_Value(Op0))), | ||
| m_CombineAnd(m_Instruction(Ext1), | ||
| m_ZExtOrSExt(m_Value(Op1))))) && | ||
| if (match(Add, m_AddLike(m_CombineAnd(m_Instruction(Ext0), |
There was a problem hiding this comment.
Alright, ill sit on this for a bit, if he gets back in the next month or so ill close. Ill pull out the add nuw patch.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
(icmp eq/ne (add nuw x, y), 0); NFC(icmp eq/ne (add nuw x, y), 0)->(icmp eq/ne (or x, y), 0)(icmp pred (add x, y), C)folds withor disjoint