Skip to content

[InstCombine] Handle trunc i1 pattern in eq-of-parts fold #112704

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 2 commits into from
Nov 25, 2024
+22 −16
Merged

[InstCombine] Handle trunc i1 pattern in eq-of-parts fold #112704

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
dd6eff5
[InstCombine] Handle trunc i1 pattern in eq-of-parts fold
nikic Oct 16, 2024
7117fbb
add assert
nikic Nov 25, 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
25 changes: 19 additions & 6 deletions llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1185,14 +1185,27 @@ static Value *extractIntPart(const IntPart &P, IRBuilderBase &Builder) {
/// (icmp eq X0, Y0) & (icmp eq X1, Y1) -> icmp eq X01, Y01
/// (icmp ne X0, Y0) | (icmp ne X1, Y1) -> icmp ne X01, Y01
/// where X0, X1 and Y0, Y1 are adjacent parts extracted from an integer.
Value *InstCombinerImpl::foldEqOfParts(ICmpInst *Cmp0, ICmpInst *Cmp1,
bool IsAnd) {
Value *InstCombinerImpl::foldEqOfParts(Value *Cmp0, Value *Cmp1, bool IsAnd) {
if (!Cmp0->hasOneUse() || !Cmp1->hasOneUse())
return nullptr;

CmpInst::Predicate Pred = IsAnd ? CmpInst::ICMP_EQ : CmpInst::ICMP_NE;
auto GetMatchPart = [&](ICmpInst *Cmp,
auto GetMatchPart = [&](Value *CmpV,
unsigned OpNo) -> std::optional<IntPart> {
assert(CmpV->getType()->isIntOrIntVectorTy(1) && "Must be bool");

Value *X, *Y;
// icmp ne (and x, 1), (and y, 1) <=> trunc (xor x, y) to i1
// icmp eq (and x, 1), (and y, 1) <=> not (trunc (xor x, y) to i1)
if (Pred == CmpInst::ICMP_NE
? match(CmpV, m_Trunc(m_Xor(m_Value(X), m_Value(Y))))
: match(CmpV, m_Not(m_Trunc(m_Xor(m_Value(X), m_Value(Y))))))
return {{OpNo == 0 ? X : Y, 0, 1}};

auto *Cmp = dyn_cast<ICmpInst>(CmpV);
if (!Cmp)
return std::nullopt;

if (Pred == Cmp->getPredicate())
return matchIntPart(Cmp->getOperand(OpNo));

Expand Down Expand Up @@ -3404,9 +3417,6 @@ Value *InstCombinerImpl::foldAndOrOfICmps(ICmpInst *LHS, ICmpInst *RHS,
return X;
}

if (Value *X = foldEqOfParts(LHS, RHS, IsAnd))
return X;

// (icmp ne A, 0) | (icmp ne B, 0) --> (icmp ne (A|B), 0)
// (icmp eq A, 0) & (icmp eq B, 0) --> (icmp eq (A|B), 0)
// TODO: Remove this and below when foldLogOpOfMaskedICmps can handle undefs.
Expand Down Expand Up @@ -3529,6 +3539,9 @@ Value *InstCombinerImpl::foldBooleanAndOr(Value *LHS, Value *RHS,
if (Value *Res = foldLogicOfFCmps(LHSCmp, RHSCmp, IsAnd, IsLogical))
return Res;

if (Value *Res = foldEqOfParts(LHS, RHS, IsAnd))
return Res;

return nullptr;
}

Expand Down
2 changes: 1 addition & 1 deletion llvm/lib/Transforms/InstCombine/InstCombineInternal.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -412,7 +412,7 @@ class LLVM_LIBRARY_VISIBILITY InstCombinerImpl final
bool IsAnd, bool IsLogical = false);
Value *foldXorOfICmps(ICmpInst *LHS, ICmpInst *RHS, BinaryOperator &Xor);

Value *foldEqOfParts(ICmpInst *Cmp0, ICmpInst *Cmp1, bool IsAnd);
Value *foldEqOfParts(Value *Cmp0, Value *Cmp1, bool IsAnd);

Value *foldAndOrOfICmpsUsingRanges(ICmpInst *ICmp1, ICmpInst *ICmp2,
bool IsAnd);
Expand Down
11 changes: 2 additions & 9 deletions llvm/test/Transforms/InstCombine/eq-of-parts.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1441,11 +1441,7 @@ define i1 @ne_optimized_highbits_cmp_todo_overlapping(i32 %x, i32 %y) {

define i1 @and_trunc_i1(i8 %a1, i8 %a2) {
; CHECK-LABEL: @and_trunc_i1(
; CHECK-NEXT: [[XOR:%.*]] = xor i8 [[A1:%.*]], [[A2:%.*]]
; CHECK-NEXT: [[CMP:%.*]] = icmp ult i8 [[XOR]], 2
; CHECK-NEXT: [[LOBIT:%.*]] = trunc i8 [[XOR]] to i1
; CHECK-NEXT: [[LOBIT_INV:%.*]] = xor i1 [[LOBIT]], true
; CHECK-NEXT: [[AND:%.*]] = and i1 [[CMP]], [[LOBIT_INV]]
; CHECK-NEXT: [[AND:%.*]] = icmp eq i8 [[A1:%.*]], [[A2:%.*]]
; CHECK-NEXT: ret i1 [[AND]]
;
%xor = xor i8 %a1, %a2
Expand Down Expand Up @@ -1494,10 +1490,7 @@ define i1 @and_trunc_i1_wrong_operands(i8 %a1, i8 %a2, i8 %a3) {

define i1 @or_trunc_i1(i64 %a1, i64 %a2) {
; CHECK-LABEL: @or_trunc_i1(
; CHECK-NEXT: [[XOR:%.*]] = xor i64 [[A2:%.*]], [[A1:%.*]]
; CHECK-NEXT: [[CMP:%.*]] = icmp ugt i64 [[XOR]], 1
; CHECK-NEXT: [[TRUNC:%.*]] = trunc i64 [[XOR]] to i1
; CHECK-NEXT: [[OR:%.*]] = or i1 [[CMP]], [[TRUNC]]
; CHECK-NEXT: [[OR:%.*]] = icmp ne i64 [[A2:%.*]], [[A1:%.*]]
; CHECK-NEXT: ret i1 [[OR]]
;
%xor = xor i64 %a2, %a1
Expand Down
Loading