[InstCombine] Fold icmp of trunc nuw/nsw #90436
Merged
Conversation
This file contains hidden or 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
Convert the existing foldICmpTruncWithTruncOrExt() fold to work with trunc nowrap flags instead of computeKnownBits(). This also allows us to generalize the fold to work with signed comparisons. Interesting, apart from the obvious combinations like signed predicates with trunc nsw, some non-obvious ones are also legal. For example for unsigned predicated we can do the transform for two trunc nsw as well (rather than only trunc nuw). Proofs: https://alive2.llvm.org/ce/z/ndewwK
|
@llvm/pr-subscribers-llvm-ir Author: Nikita Popov (nikic) ChangesConvert the existing foldICmpTruncWithTruncOrExt() fold to work with trunc nowrap flags instead of computeKnownBits(). This also allows us to generalize the fold to work with signed comparisons. Interestingly, apart from the obvious combinations like signed predicates with trunc nsw, some non-obvious ones are also legal. For example for unsigned predicates we can do the transform for two trunc nsw as well (rather than only trunc nuw). Proofs: https://alive2.llvm.org/ce/z/ndewwK Full diff: https://github.com/llvm/llvm-project/pull/90436.diff 3 Files Affected:
diff --git a/llvm/include/llvm/IR/PatternMatch.h b/llvm/include/llvm/IR/PatternMatch.h
index 0b13b4aad9c326..f063e6531dda4f 100644
--- a/llvm/include/llvm/IR/PatternMatch.h
+++ b/llvm/include/llvm/IR/PatternMatch.h
@@ -1839,6 +1839,19 @@ template <typename Op_t> struct NNegZExt_match {
}
};
+template <typename Op_t, unsigned WrapFlags = 0> struct NoWrapTrunc_match {
+ Op_t Op;
+
+ NoWrapTrunc_match(const Op_t &OpMatch) : Op(OpMatch) {}
+
+ template <typename OpTy> bool match(OpTy *V) {
+ if (auto *I = dyn_cast<TruncInst>(V))
+ return (I->getNoWrapKind() & WrapFlags) == WrapFlags &&
+ Op.match(I->getOperand(0));
+ return false;
+ }
+};
+
/// Matches BitCast.
template <typename OpTy>
inline CastOperator_match<OpTy, Instruction::BitCast>
@@ -1900,6 +1913,20 @@ inline CastOperator_match<OpTy, Instruction::Trunc> m_Trunc(const OpTy &Op) {
return CastOperator_match<OpTy, Instruction::Trunc>(Op);
}
+/// Matches trunc nuw.
+template <typename OpTy>
+inline NoWrapTrunc_match<OpTy, TruncInst::NoUnsignedWrap>
+m_NUWTrunc(const OpTy &Op) {
+ return NoWrapTrunc_match<OpTy, TruncInst::NoUnsignedWrap>(Op);
+}
+
+/// Matches trunc nsw.
+template <typename OpTy>
+inline NoWrapTrunc_match<OpTy, TruncInst::NoSignedWrap>
+m_NSWTrunc(const OpTy &Op) {
+ return NoWrapTrunc_match<OpTy, TruncInst::NoSignedWrap>(Op);
+}
+
template <typename OpTy>
inline match_combine_or<CastOperator_match<OpTy, Instruction::Trunc>, OpTy>
m_TruncOrSelf(const OpTy &Op) {
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index 1064340cb53661..317b3aef40a97c 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -1479,19 +1479,29 @@ Instruction *InstCombinerImpl::foldICmpTruncConstant(ICmpInst &Cmp,
return nullptr;
}
-/// Fold icmp (trunc X), (trunc Y).
-/// Fold icmp (trunc X), (zext Y).
+/// Fold icmp (trunc nuw/nsw X), (trunc nuw/nsw Y).
+/// Fold icmp (trunc nuw/nsw X), (zext/sext Y).
Instruction *
InstCombinerImpl::foldICmpTruncWithTruncOrExt(ICmpInst &Cmp,
const SimplifyQuery &Q) {
- if (Cmp.isSigned())
- return nullptr;
-
Value *X, *Y;
ICmpInst::Predicate Pred;
- bool YIsZext = false;
+ bool YIsSExt = false;
// Try to match icmp (trunc X), (trunc Y)
if (match(&Cmp, m_ICmp(Pred, m_Trunc(m_Value(X)), m_Trunc(m_Value(Y))))) {
+ unsigned NoWrapFlags = cast<TruncInst>(Cmp.getOperand(0))->getNoWrapKind() &
+ cast<TruncInst>(Cmp.getOperand(1))->getNoWrapKind();
+ if (Cmp.isSigned()) {
+ // For signed comparisons, both truncs must be nsw.
+ if (!(NoWrapFlags & TruncInst::NoSignedWrap))
+ return nullptr;
+ } else {
+ // For unsigned and equality comparisons, either both must be nuw or
+ // both must be nsw, we don't care which.
+ if (!NoWrapFlags)
+ return nullptr;
+ }
+
if (X->getType() != Y->getType() &&
(!Cmp.getOperand(0)->hasOneUse() || !Cmp.getOperand(1)->hasOneUse()))
return nullptr;
@@ -1501,12 +1511,19 @@ InstCombinerImpl::foldICmpTruncWithTruncOrExt(ICmpInst &Cmp,
Pred = Cmp.getSwappedPredicate(Pred);
}
}
- // Try to match icmp (trunc X), (zext Y)
- else if (match(&Cmp, m_c_ICmp(Pred, m_Trunc(m_Value(X)),
- m_OneUse(m_ZExt(m_Value(Y))))))
-
- YIsZext = true;
- else
+ // Try to match icmp (trunc nuw X), (zext Y)
+ else if (!Cmp.isSigned() &&
+ match(&Cmp, m_c_ICmp(Pred, m_NUWTrunc(m_Value(X)),
+ m_OneUse(m_ZExt(m_Value(Y)))))) {
+ // Can fold trunc nuw + zext for unsigned and equality predicates.
+ }
+ // Try to match icmp (trunc nsw X), (sext Y)
+ else if (match(&Cmp, m_c_ICmp(Pred, m_NSWTrunc(m_Value(X)),
+ m_OneUse(m_ZExtOrSExt(m_Value(Y)))))) {
+ // Can fold trunc nsw + zext/sext for all predicates.
+ YIsSExt = isa<SExtInst>(Cmp.getOperand(0)) ||
+ isa<SExtInst>(Cmp.getOperand(1));
+ } else
return nullptr;
Type *TruncTy = Cmp.getOperand(0)->getType();
@@ -1518,19 +1535,7 @@ InstCombinerImpl::foldICmpTruncWithTruncOrExt(ICmpInst &Cmp,
!isDesirableIntType(X->getType()->getScalarSizeInBits()))
return nullptr;
- // Check if the trunc is unneeded.
- KnownBits KnownX = llvm::computeKnownBits(X, /*Depth*/ 0, Q);
- if (KnownX.countMaxActiveBits() > TruncBits)
- return nullptr;
-
- if (!YIsZext) {
- // If Y is also a trunc, make sure it is unneeded.
- KnownBits KnownY = llvm::computeKnownBits(Y, /*Depth*/ 0, Q);
- if (KnownY.countMaxActiveBits() > TruncBits)
- return nullptr;
- }
-
- Value *NewY = Builder.CreateZExtOrTrunc(Y, X->getType());
+ Value *NewY = Builder.CreateIntCast(Y, X->getType(), YIsSExt);
return new ICmpInst(Pred, X, NewY);
}
diff --git a/llvm/test/Transforms/InstCombine/icmp-of-trunc-ext.ll b/llvm/test/Transforms/InstCombine/icmp-of-trunc-ext.ll
index a61694919ab040..0a8bf013dc55d2 100644
--- a/llvm/test/Transforms/InstCombine/icmp-of-trunc-ext.ll
+++ b/llvm/test/Transforms/InstCombine/icmp-of-trunc-ext.ll
@@ -271,9 +271,7 @@ define i1 @icmp_trunc_x_zext_y_fail_multiuse(i32 %x, i8 %y) {
define i1 @trunc_unsigned_nuw(i16 %x, i16 %y) {
; CHECK-LABEL: @trunc_unsigned_nuw(
-; CHECK-NEXT: [[XT:%.*]] = trunc nuw i16 [[X:%.*]] to i8
-; CHECK-NEXT: [[YT:%.*]] = trunc nuw i16 [[Y:%.*]] to i8
-; CHECK-NEXT: [[C:%.*]] = icmp ult i8 [[XT]], [[YT]]
+; CHECK-NEXT: [[C:%.*]] = icmp ult i16 [[X:%.*]], [[Y:%.*]]
; CHECK-NEXT: ret i1 [[C]]
;
%xt = trunc nuw i16 %x to i8
@@ -284,9 +282,7 @@ define i1 @trunc_unsigned_nuw(i16 %x, i16 %y) {
define i1 @trunc_unsigned_nsw(i16 %x, i16 %y) {
; CHECK-LABEL: @trunc_unsigned_nsw(
-; CHECK-NEXT: [[XT:%.*]] = trunc nsw i16 [[X:%.*]] to i8
-; CHECK-NEXT: [[YT:%.*]] = trunc nsw i16 [[Y:%.*]] to i8
-; CHECK-NEXT: [[C:%.*]] = icmp ult i8 [[XT]], [[YT]]
+; CHECK-NEXT: [[C:%.*]] = icmp ult i16 [[X:%.*]], [[Y:%.*]]
; CHECK-NEXT: ret i1 [[C]]
;
%xt = trunc nsw i16 %x to i8
@@ -297,9 +293,7 @@ define i1 @trunc_unsigned_nsw(i16 %x, i16 %y) {
define i1 @trunc_unsigned_both(i16 %x, i16 %y) {
; CHECK-LABEL: @trunc_unsigned_both(
-; CHECK-NEXT: [[XT:%.*]] = trunc nuw nsw i16 [[X:%.*]] to i8
-; CHECK-NEXT: [[YT:%.*]] = trunc nuw nsw i16 [[Y:%.*]] to i8
-; CHECK-NEXT: [[C:%.*]] = icmp ult i8 [[XT]], [[YT]]
+; CHECK-NEXT: [[C:%.*]] = icmp ult i16 [[X:%.*]], [[Y:%.*]]
; CHECK-NEXT: ret i1 [[C]]
;
%xt = trunc nuw nsw i16 %x to i8
@@ -336,9 +330,7 @@ define i1 @trunc_signed_nuw(i16 %x, i16 %y) {
define i1 @trunc_signed_nsw(i16 %x, i16 %y) {
; CHECK-LABEL: @trunc_signed_nsw(
-; CHECK-NEXT: [[XT:%.*]] = trunc nsw i16 [[X:%.*]] to i8
-; CHECK-NEXT: [[YT:%.*]] = trunc nsw i16 [[Y:%.*]] to i8
-; CHECK-NEXT: [[C:%.*]] = icmp slt i8 [[XT]], [[YT]]
+; CHECK-NEXT: [[C:%.*]] = icmp slt i16 [[X:%.*]], [[Y:%.*]]
; CHECK-NEXT: ret i1 [[C]]
;
%xt = trunc nsw i16 %x to i8
@@ -349,9 +341,7 @@ define i1 @trunc_signed_nsw(i16 %x, i16 %y) {
define i1 @trunc_signed_both(i16 %x, i16 %y) {
; CHECK-LABEL: @trunc_signed_both(
-; CHECK-NEXT: [[XT:%.*]] = trunc nuw nsw i16 [[X:%.*]] to i8
-; CHECK-NEXT: [[YT:%.*]] = trunc nuw nsw i16 [[Y:%.*]] to i8
-; CHECK-NEXT: [[C:%.*]] = icmp slt i8 [[XT]], [[YT]]
+; CHECK-NEXT: [[C:%.*]] = icmp slt i16 [[X:%.*]], [[Y:%.*]]
; CHECK-NEXT: ret i1 [[C]]
;
%xt = trunc nuw nsw i16 %x to i8
@@ -375,9 +365,7 @@ define i1 @trunc_signed_either(i16 %x, i16 %y) {
define i1 @trunc_equality_nuw(i16 %x, i16 %y) {
; CHECK-LABEL: @trunc_equality_nuw(
-; CHECK-NEXT: [[XT:%.*]] = trunc nuw i16 [[X:%.*]] to i8
-; CHECK-NEXT: [[YT:%.*]] = trunc nuw i16 [[Y:%.*]] to i8
-; CHECK-NEXT: [[C:%.*]] = icmp eq i8 [[XT]], [[YT]]
+; CHECK-NEXT: [[C:%.*]] = icmp eq i16 [[X:%.*]], [[Y:%.*]]
; CHECK-NEXT: ret i1 [[C]]
;
%xt = trunc nuw i16 %x to i8
@@ -388,9 +376,7 @@ define i1 @trunc_equality_nuw(i16 %x, i16 %y) {
define i1 @trunc_equality_nsw(i16 %x, i16 %y) {
; CHECK-LABEL: @trunc_equality_nsw(
-; CHECK-NEXT: [[XT:%.*]] = trunc nsw i16 [[X:%.*]] to i8
-; CHECK-NEXT: [[YT:%.*]] = trunc nsw i16 [[Y:%.*]] to i8
-; CHECK-NEXT: [[C:%.*]] = icmp eq i8 [[XT]], [[YT]]
+; CHECK-NEXT: [[C:%.*]] = icmp eq i16 [[X:%.*]], [[Y:%.*]]
; CHECK-NEXT: ret i1 [[C]]
;
%xt = trunc nsw i16 %x to i8
@@ -401,9 +387,7 @@ define i1 @trunc_equality_nsw(i16 %x, i16 %y) {
define i1 @trunc_equality_both(i16 %x, i16 %y) {
; CHECK-LABEL: @trunc_equality_both(
-; CHECK-NEXT: [[XT:%.*]] = trunc nuw nsw i16 [[X:%.*]] to i8
-; CHECK-NEXT: [[YT:%.*]] = trunc nuw nsw i16 [[Y:%.*]] to i8
-; CHECK-NEXT: [[C:%.*]] = icmp eq i8 [[XT]], [[YT]]
+; CHECK-NEXT: [[C:%.*]] = icmp eq i16 [[X:%.*]], [[Y:%.*]]
; CHECK-NEXT: ret i1 [[C]]
;
%xt = trunc nuw nsw i16 %x to i8
@@ -427,9 +411,8 @@ define i1 @trunc_equality_either(i16 %x, i16 %y) {
define i1 @trunc_unsigned_nuw_zext(i32 %x, i8 %y) {
; CHECK-LABEL: @trunc_unsigned_nuw_zext(
-; CHECK-NEXT: [[XT:%.*]] = trunc nuw i32 [[X:%.*]] to i16
-; CHECK-NEXT: [[YE:%.*]] = zext i8 [[Y:%.*]] to i16
-; CHECK-NEXT: [[C:%.*]] = icmp ult i16 [[XT]], [[YE]]
+; CHECK-NEXT: [[TMP1:%.*]] = zext i8 [[Y:%.*]] to i32
+; CHECK-NEXT: [[C:%.*]] = icmp ugt i32 [[TMP1]], [[X:%.*]]
; CHECK-NEXT: ret i1 [[C]]
;
%xt = trunc nuw i32 %x to i16
@@ -453,9 +436,8 @@ define i1 @trunc_unsigned_nuw_sext(i32 %x, i8 %y) {
define i1 @trunc_unsigned_nsw_zext(i32 %x, i8 %y) {
; CHECK-LABEL: @trunc_unsigned_nsw_zext(
-; CHECK-NEXT: [[XT:%.*]] = trunc nsw i32 [[X:%.*]] to i16
-; CHECK-NEXT: [[YE:%.*]] = zext i8 [[Y:%.*]] to i16
-; CHECK-NEXT: [[C:%.*]] = icmp ult i16 [[XT]], [[YE]]
+; CHECK-NEXT: [[TMP1:%.*]] = zext i8 [[Y:%.*]] to i32
+; CHECK-NEXT: [[C:%.*]] = icmp ugt i32 [[TMP1]], [[X:%.*]]
; CHECK-NEXT: ret i1 [[C]]
;
%xt = trunc nsw i32 %x to i16
@@ -466,9 +448,8 @@ define i1 @trunc_unsigned_nsw_zext(i32 %x, i8 %y) {
define i1 @trunc_unsigned_nsw_sext(i32 %x, i8 %y) {
; CHECK-LABEL: @trunc_unsigned_nsw_sext(
-; CHECK-NEXT: [[XT:%.*]] = trunc nsw i32 [[X:%.*]] to i16
-; CHECK-NEXT: [[YE:%.*]] = sext i8 [[Y:%.*]] to i16
-; CHECK-NEXT: [[C:%.*]] = icmp ult i16 [[XT]], [[YE]]
+; CHECK-NEXT: [[TMP1:%.*]] = sext i8 [[Y:%.*]] to i32
+; CHECK-NEXT: [[C:%.*]] = icmp ugt i32 [[TMP1]], [[X:%.*]]
; CHECK-NEXT: ret i1 [[C]]
;
%xt = trunc nsw i32 %x to i16
@@ -479,9 +460,8 @@ define i1 @trunc_unsigned_nsw_sext(i32 %x, i8 %y) {
define i1 @trunc_signed_nsw_sext(i32 %x, i8 %y) {
; CHECK-LABEL: @trunc_signed_nsw_sext(
-; CHECK-NEXT: [[XT:%.*]] = trunc nsw i32 [[X:%.*]] to i16
-; CHECK-NEXT: [[YE:%.*]] = sext i8 [[Y:%.*]] to i16
-; CHECK-NEXT: [[C:%.*]] = icmp slt i16 [[XT]], [[YE]]
+; CHECK-NEXT: [[TMP1:%.*]] = sext i8 [[Y:%.*]] to i32
+; CHECK-NEXT: [[C:%.*]] = icmp sgt i32 [[TMP1]], [[X:%.*]]
; CHECK-NEXT: ret i1 [[C]]
;
%xt = trunc nsw i32 %x to i16
@@ -492,9 +472,8 @@ define i1 @trunc_signed_nsw_sext(i32 %x, i8 %y) {
define i1 @trunc_signed_nsw_zext(i32 %x, i8 %y) {
; CHECK-LABEL: @trunc_signed_nsw_zext(
-; CHECK-NEXT: [[XT:%.*]] = trunc nsw i32 [[X:%.*]] to i16
-; CHECK-NEXT: [[YE:%.*]] = zext i8 [[Y:%.*]] to i16
-; CHECK-NEXT: [[C:%.*]] = icmp slt i16 [[XT]], [[YE]]
+; CHECK-NEXT: [[TMP1:%.*]] = zext i8 [[Y:%.*]] to i32
+; CHECK-NEXT: [[C:%.*]] = icmp sgt i32 [[TMP1]], [[X:%.*]]
; CHECK-NEXT: ret i1 [[C]]
;
%xt = trunc nsw i32 %x to i16
@@ -531,9 +510,8 @@ define i1 @trunc_signed_nuw_zext(i32 %x, i8 %y) {
define i1 @trunc_equality_nuw_zext(i32 %x, i8 %y) {
; CHECK-LABEL: @trunc_equality_nuw_zext(
-; CHECK-NEXT: [[XT:%.*]] = trunc nuw i32 [[X:%.*]] to i16
-; CHECK-NEXT: [[YE:%.*]] = zext i8 [[Y:%.*]] to i16
-; CHECK-NEXT: [[C:%.*]] = icmp ne i16 [[XT]], [[YE]]
+; CHECK-NEXT: [[TMP1:%.*]] = zext i8 [[Y:%.*]] to i32
+; CHECK-NEXT: [[C:%.*]] = icmp ne i32 [[TMP1]], [[X:%.*]]
; CHECK-NEXT: ret i1 [[C]]
;
%xt = trunc nuw i32 %x to i16
@@ -557,9 +535,8 @@ define i1 @trunc_equality_nuw_sext(i32 %x, i8 %y) {
define i1 @trunc_equality_nsw_zext(i32 %x, i8 %y) {
; CHECK-LABEL: @trunc_equality_nsw_zext(
-; CHECK-NEXT: [[XT:%.*]] = trunc nsw i32 [[X:%.*]] to i16
-; CHECK-NEXT: [[YE:%.*]] = zext i8 [[Y:%.*]] to i16
-; CHECK-NEXT: [[C:%.*]] = icmp ne i16 [[XT]], [[YE]]
+; CHECK-NEXT: [[TMP1:%.*]] = zext i8 [[Y:%.*]] to i32
+; CHECK-NEXT: [[C:%.*]] = icmp ne i32 [[TMP1]], [[X:%.*]]
; CHECK-NEXT: ret i1 [[C]]
;
%xt = trunc nsw i32 %x to i16
@@ -570,9 +547,8 @@ define i1 @trunc_equality_nsw_zext(i32 %x, i8 %y) {
define i1 @trunc_equality_nsw_sext(i32 %x, i8 %y) {
; CHECK-LABEL: @trunc_equality_nsw_sext(
-; CHECK-NEXT: [[XT:%.*]] = trunc nsw i32 [[X:%.*]] to i16
-; CHECK-NEXT: [[YE:%.*]] = sext i8 [[Y:%.*]] to i16
-; CHECK-NEXT: [[C:%.*]] = icmp ne i16 [[XT]], [[YE]]
+; CHECK-NEXT: [[TMP1:%.*]] = sext i8 [[Y:%.*]] to i32
+; CHECK-NEXT: [[C:%.*]] = icmp ne i32 [[TMP1]], [[X:%.*]]
; CHECK-NEXT: ret i1 [[C]]
;
%xt = trunc nsw i32 %x to i16
@@ -583,9 +559,8 @@ define i1 @trunc_equality_nsw_sext(i32 %x, i8 %y) {
define i1 @trunc_equality_both_sext(i32 %x, i8 %y) {
; CHECK-LABEL: @trunc_equality_both_sext(
-; CHECK-NEXT: [[XT:%.*]] = trunc nuw nsw i32 [[X:%.*]] to i16
-; CHECK-NEXT: [[YE:%.*]] = sext i8 [[Y:%.*]] to i16
-; CHECK-NEXT: [[C:%.*]] = icmp ne i16 [[XT]], [[YE]]
+; CHECK-NEXT: [[TMP1:%.*]] = sext i8 [[Y:%.*]] to i32
+; CHECK-NEXT: [[C:%.*]] = icmp ne i32 [[TMP1]], [[X:%.*]]
; CHECK-NEXT: ret i1 [[C]]
;
%xt = trunc nuw nsw i32 %x to i16
|
|
@llvm/pr-subscribers-llvm-transforms Author: Nikita Popov (nikic) ChangesConvert the existing foldICmpTruncWithTruncOrExt() fold to work with trunc nowrap flags instead of computeKnownBits(). This also allows us to generalize the fold to work with signed comparisons. Interestingly, apart from the obvious combinations like signed predicates with trunc nsw, some non-obvious ones are also legal. For example for unsigned predicates we can do the transform for two trunc nsw as well (rather than only trunc nuw). Proofs: https://alive2.llvm.org/ce/z/ndewwK Full diff: https://github.com/llvm/llvm-project/pull/90436.diff 3 Files Affected:
diff --git a/llvm/include/llvm/IR/PatternMatch.h b/llvm/include/llvm/IR/PatternMatch.h
index 0b13b4aad9c326..f063e6531dda4f 100644
--- a/llvm/include/llvm/IR/PatternMatch.h
+++ b/llvm/include/llvm/IR/PatternMatch.h
@@ -1839,6 +1839,19 @@ template <typename Op_t> struct NNegZExt_match {
}
};
+template <typename Op_t, unsigned WrapFlags = 0> struct NoWrapTrunc_match {
+ Op_t Op;
+
+ NoWrapTrunc_match(const Op_t &OpMatch) : Op(OpMatch) {}
+
+ template <typename OpTy> bool match(OpTy *V) {
+ if (auto *I = dyn_cast<TruncInst>(V))
+ return (I->getNoWrapKind() & WrapFlags) == WrapFlags &&
+ Op.match(I->getOperand(0));
+ return false;
+ }
+};
+
/// Matches BitCast.
template <typename OpTy>
inline CastOperator_match<OpTy, Instruction::BitCast>
@@ -1900,6 +1913,20 @@ inline CastOperator_match<OpTy, Instruction::Trunc> m_Trunc(const OpTy &Op) {
return CastOperator_match<OpTy, Instruction::Trunc>(Op);
}
+/// Matches trunc nuw.
+template <typename OpTy>
+inline NoWrapTrunc_match<OpTy, TruncInst::NoUnsignedWrap>
+m_NUWTrunc(const OpTy &Op) {
+ return NoWrapTrunc_match<OpTy, TruncInst::NoUnsignedWrap>(Op);
+}
+
+/// Matches trunc nsw.
+template <typename OpTy>
+inline NoWrapTrunc_match<OpTy, TruncInst::NoSignedWrap>
+m_NSWTrunc(const OpTy &Op) {
+ return NoWrapTrunc_match<OpTy, TruncInst::NoSignedWrap>(Op);
+}
+
template <typename OpTy>
inline match_combine_or<CastOperator_match<OpTy, Instruction::Trunc>, OpTy>
m_TruncOrSelf(const OpTy &Op) {
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index 1064340cb53661..317b3aef40a97c 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -1479,19 +1479,29 @@ Instruction *InstCombinerImpl::foldICmpTruncConstant(ICmpInst &Cmp,
return nullptr;
}
-/// Fold icmp (trunc X), (trunc Y).
-/// Fold icmp (trunc X), (zext Y).
+/// Fold icmp (trunc nuw/nsw X), (trunc nuw/nsw Y).
+/// Fold icmp (trunc nuw/nsw X), (zext/sext Y).
Instruction *
InstCombinerImpl::foldICmpTruncWithTruncOrExt(ICmpInst &Cmp,
const SimplifyQuery &Q) {
- if (Cmp.isSigned())
- return nullptr;
-
Value *X, *Y;
ICmpInst::Predicate Pred;
- bool YIsZext = false;
+ bool YIsSExt = false;
// Try to match icmp (trunc X), (trunc Y)
if (match(&Cmp, m_ICmp(Pred, m_Trunc(m_Value(X)), m_Trunc(m_Value(Y))))) {
+ unsigned NoWrapFlags = cast<TruncInst>(Cmp.getOperand(0))->getNoWrapKind() &
+ cast<TruncInst>(Cmp.getOperand(1))->getNoWrapKind();
+ if (Cmp.isSigned()) {
+ // For signed comparisons, both truncs must be nsw.
+ if (!(NoWrapFlags & TruncInst::NoSignedWrap))
+ return nullptr;
+ } else {
+ // For unsigned and equality comparisons, either both must be nuw or
+ // both must be nsw, we don't care which.
+ if (!NoWrapFlags)
+ return nullptr;
+ }
+
if (X->getType() != Y->getType() &&
(!Cmp.getOperand(0)->hasOneUse() || !Cmp.getOperand(1)->hasOneUse()))
return nullptr;
@@ -1501,12 +1511,19 @@ InstCombinerImpl::foldICmpTruncWithTruncOrExt(ICmpInst &Cmp,
Pred = Cmp.getSwappedPredicate(Pred);
}
}
- // Try to match icmp (trunc X), (zext Y)
- else if (match(&Cmp, m_c_ICmp(Pred, m_Trunc(m_Value(X)),
- m_OneUse(m_ZExt(m_Value(Y))))))
-
- YIsZext = true;
- else
+ // Try to match icmp (trunc nuw X), (zext Y)
+ else if (!Cmp.isSigned() &&
+ match(&Cmp, m_c_ICmp(Pred, m_NUWTrunc(m_Value(X)),
+ m_OneUse(m_ZExt(m_Value(Y)))))) {
+ // Can fold trunc nuw + zext for unsigned and equality predicates.
+ }
+ // Try to match icmp (trunc nsw X), (sext Y)
+ else if (match(&Cmp, m_c_ICmp(Pred, m_NSWTrunc(m_Value(X)),
+ m_OneUse(m_ZExtOrSExt(m_Value(Y)))))) {
+ // Can fold trunc nsw + zext/sext for all predicates.
+ YIsSExt = isa<SExtInst>(Cmp.getOperand(0)) ||
+ isa<SExtInst>(Cmp.getOperand(1));
+ } else
return nullptr;
Type *TruncTy = Cmp.getOperand(0)->getType();
@@ -1518,19 +1535,7 @@ InstCombinerImpl::foldICmpTruncWithTruncOrExt(ICmpInst &Cmp,
!isDesirableIntType(X->getType()->getScalarSizeInBits()))
return nullptr;
- // Check if the trunc is unneeded.
- KnownBits KnownX = llvm::computeKnownBits(X, /*Depth*/ 0, Q);
- if (KnownX.countMaxActiveBits() > TruncBits)
- return nullptr;
-
- if (!YIsZext) {
- // If Y is also a trunc, make sure it is unneeded.
- KnownBits KnownY = llvm::computeKnownBits(Y, /*Depth*/ 0, Q);
- if (KnownY.countMaxActiveBits() > TruncBits)
- return nullptr;
- }
-
- Value *NewY = Builder.CreateZExtOrTrunc(Y, X->getType());
+ Value *NewY = Builder.CreateIntCast(Y, X->getType(), YIsSExt);
return new ICmpInst(Pred, X, NewY);
}
diff --git a/llvm/test/Transforms/InstCombine/icmp-of-trunc-ext.ll b/llvm/test/Transforms/InstCombine/icmp-of-trunc-ext.ll
index a61694919ab040..0a8bf013dc55d2 100644
--- a/llvm/test/Transforms/InstCombine/icmp-of-trunc-ext.ll
+++ b/llvm/test/Transforms/InstCombine/icmp-of-trunc-ext.ll
@@ -271,9 +271,7 @@ define i1 @icmp_trunc_x_zext_y_fail_multiuse(i32 %x, i8 %y) {
define i1 @trunc_unsigned_nuw(i16 %x, i16 %y) {
; CHECK-LABEL: @trunc_unsigned_nuw(
-; CHECK-NEXT: [[XT:%.*]] = trunc nuw i16 [[X:%.*]] to i8
-; CHECK-NEXT: [[YT:%.*]] = trunc nuw i16 [[Y:%.*]] to i8
-; CHECK-NEXT: [[C:%.*]] = icmp ult i8 [[XT]], [[YT]]
+; CHECK-NEXT: [[C:%.*]] = icmp ult i16 [[X:%.*]], [[Y:%.*]]
; CHECK-NEXT: ret i1 [[C]]
;
%xt = trunc nuw i16 %x to i8
@@ -284,9 +282,7 @@ define i1 @trunc_unsigned_nuw(i16 %x, i16 %y) {
define i1 @trunc_unsigned_nsw(i16 %x, i16 %y) {
; CHECK-LABEL: @trunc_unsigned_nsw(
-; CHECK-NEXT: [[XT:%.*]] = trunc nsw i16 [[X:%.*]] to i8
-; CHECK-NEXT: [[YT:%.*]] = trunc nsw i16 [[Y:%.*]] to i8
-; CHECK-NEXT: [[C:%.*]] = icmp ult i8 [[XT]], [[YT]]
+; CHECK-NEXT: [[C:%.*]] = icmp ult i16 [[X:%.*]], [[Y:%.*]]
; CHECK-NEXT: ret i1 [[C]]
;
%xt = trunc nsw i16 %x to i8
@@ -297,9 +293,7 @@ define i1 @trunc_unsigned_nsw(i16 %x, i16 %y) {
define i1 @trunc_unsigned_both(i16 %x, i16 %y) {
; CHECK-LABEL: @trunc_unsigned_both(
-; CHECK-NEXT: [[XT:%.*]] = trunc nuw nsw i16 [[X:%.*]] to i8
-; CHECK-NEXT: [[YT:%.*]] = trunc nuw nsw i16 [[Y:%.*]] to i8
-; CHECK-NEXT: [[C:%.*]] = icmp ult i8 [[XT]], [[YT]]
+; CHECK-NEXT: [[C:%.*]] = icmp ult i16 [[X:%.*]], [[Y:%.*]]
; CHECK-NEXT: ret i1 [[C]]
;
%xt = trunc nuw nsw i16 %x to i8
@@ -336,9 +330,7 @@ define i1 @trunc_signed_nuw(i16 %x, i16 %y) {
define i1 @trunc_signed_nsw(i16 %x, i16 %y) {
; CHECK-LABEL: @trunc_signed_nsw(
-; CHECK-NEXT: [[XT:%.*]] = trunc nsw i16 [[X:%.*]] to i8
-; CHECK-NEXT: [[YT:%.*]] = trunc nsw i16 [[Y:%.*]] to i8
-; CHECK-NEXT: [[C:%.*]] = icmp slt i8 [[XT]], [[YT]]
+; CHECK-NEXT: [[C:%.*]] = icmp slt i16 [[X:%.*]], [[Y:%.*]]
; CHECK-NEXT: ret i1 [[C]]
;
%xt = trunc nsw i16 %x to i8
@@ -349,9 +341,7 @@ define i1 @trunc_signed_nsw(i16 %x, i16 %y) {
define i1 @trunc_signed_both(i16 %x, i16 %y) {
; CHECK-LABEL: @trunc_signed_both(
-; CHECK-NEXT: [[XT:%.*]] = trunc nuw nsw i16 [[X:%.*]] to i8
-; CHECK-NEXT: [[YT:%.*]] = trunc nuw nsw i16 [[Y:%.*]] to i8
-; CHECK-NEXT: [[C:%.*]] = icmp slt i8 [[XT]], [[YT]]
+; CHECK-NEXT: [[C:%.*]] = icmp slt i16 [[X:%.*]], [[Y:%.*]]
; CHECK-NEXT: ret i1 [[C]]
;
%xt = trunc nuw nsw i16 %x to i8
@@ -375,9 +365,7 @@ define i1 @trunc_signed_either(i16 %x, i16 %y) {
define i1 @trunc_equality_nuw(i16 %x, i16 %y) {
; CHECK-LABEL: @trunc_equality_nuw(
-; CHECK-NEXT: [[XT:%.*]] = trunc nuw i16 [[X:%.*]] to i8
-; CHECK-NEXT: [[YT:%.*]] = trunc nuw i16 [[Y:%.*]] to i8
-; CHECK-NEXT: [[C:%.*]] = icmp eq i8 [[XT]], [[YT]]
+; CHECK-NEXT: [[C:%.*]] = icmp eq i16 [[X:%.*]], [[Y:%.*]]
; CHECK-NEXT: ret i1 [[C]]
;
%xt = trunc nuw i16 %x to i8
@@ -388,9 +376,7 @@ define i1 @trunc_equality_nuw(i16 %x, i16 %y) {
define i1 @trunc_equality_nsw(i16 %x, i16 %y) {
; CHECK-LABEL: @trunc_equality_nsw(
-; CHECK-NEXT: [[XT:%.*]] = trunc nsw i16 [[X:%.*]] to i8
-; CHECK-NEXT: [[YT:%.*]] = trunc nsw i16 [[Y:%.*]] to i8
-; CHECK-NEXT: [[C:%.*]] = icmp eq i8 [[XT]], [[YT]]
+; CHECK-NEXT: [[C:%.*]] = icmp eq i16 [[X:%.*]], [[Y:%.*]]
; CHECK-NEXT: ret i1 [[C]]
;
%xt = trunc nsw i16 %x to i8
@@ -401,9 +387,7 @@ define i1 @trunc_equality_nsw(i16 %x, i16 %y) {
define i1 @trunc_equality_both(i16 %x, i16 %y) {
; CHECK-LABEL: @trunc_equality_both(
-; CHECK-NEXT: [[XT:%.*]] = trunc nuw nsw i16 [[X:%.*]] to i8
-; CHECK-NEXT: [[YT:%.*]] = trunc nuw nsw i16 [[Y:%.*]] to i8
-; CHECK-NEXT: [[C:%.*]] = icmp eq i8 [[XT]], [[YT]]
+; CHECK-NEXT: [[C:%.*]] = icmp eq i16 [[X:%.*]], [[Y:%.*]]
; CHECK-NEXT: ret i1 [[C]]
;
%xt = trunc nuw nsw i16 %x to i8
@@ -427,9 +411,8 @@ define i1 @trunc_equality_either(i16 %x, i16 %y) {
define i1 @trunc_unsigned_nuw_zext(i32 %x, i8 %y) {
; CHECK-LABEL: @trunc_unsigned_nuw_zext(
-; CHECK-NEXT: [[XT:%.*]] = trunc nuw i32 [[X:%.*]] to i16
-; CHECK-NEXT: [[YE:%.*]] = zext i8 [[Y:%.*]] to i16
-; CHECK-NEXT: [[C:%.*]] = icmp ult i16 [[XT]], [[YE]]
+; CHECK-NEXT: [[TMP1:%.*]] = zext i8 [[Y:%.*]] to i32
+; CHECK-NEXT: [[C:%.*]] = icmp ugt i32 [[TMP1]], [[X:%.*]]
; CHECK-NEXT: ret i1 [[C]]
;
%xt = trunc nuw i32 %x to i16
@@ -453,9 +436,8 @@ define i1 @trunc_unsigned_nuw_sext(i32 %x, i8 %y) {
define i1 @trunc_unsigned_nsw_zext(i32 %x, i8 %y) {
; CHECK-LABEL: @trunc_unsigned_nsw_zext(
-; CHECK-NEXT: [[XT:%.*]] = trunc nsw i32 [[X:%.*]] to i16
-; CHECK-NEXT: [[YE:%.*]] = zext i8 [[Y:%.*]] to i16
-; CHECK-NEXT: [[C:%.*]] = icmp ult i16 [[XT]], [[YE]]
+; CHECK-NEXT: [[TMP1:%.*]] = zext i8 [[Y:%.*]] to i32
+; CHECK-NEXT: [[C:%.*]] = icmp ugt i32 [[TMP1]], [[X:%.*]]
; CHECK-NEXT: ret i1 [[C]]
;
%xt = trunc nsw i32 %x to i16
@@ -466,9 +448,8 @@ define i1 @trunc_unsigned_nsw_zext(i32 %x, i8 %y) {
define i1 @trunc_unsigned_nsw_sext(i32 %x, i8 %y) {
; CHECK-LABEL: @trunc_unsigned_nsw_sext(
-; CHECK-NEXT: [[XT:%.*]] = trunc nsw i32 [[X:%.*]] to i16
-; CHECK-NEXT: [[YE:%.*]] = sext i8 [[Y:%.*]] to i16
-; CHECK-NEXT: [[C:%.*]] = icmp ult i16 [[XT]], [[YE]]
+; CHECK-NEXT: [[TMP1:%.*]] = sext i8 [[Y:%.*]] to i32
+; CHECK-NEXT: [[C:%.*]] = icmp ugt i32 [[TMP1]], [[X:%.*]]
; CHECK-NEXT: ret i1 [[C]]
;
%xt = trunc nsw i32 %x to i16
@@ -479,9 +460,8 @@ define i1 @trunc_unsigned_nsw_sext(i32 %x, i8 %y) {
define i1 @trunc_signed_nsw_sext(i32 %x, i8 %y) {
; CHECK-LABEL: @trunc_signed_nsw_sext(
-; CHECK-NEXT: [[XT:%.*]] = trunc nsw i32 [[X:%.*]] to i16
-; CHECK-NEXT: [[YE:%.*]] = sext i8 [[Y:%.*]] to i16
-; CHECK-NEXT: [[C:%.*]] = icmp slt i16 [[XT]], [[YE]]
+; CHECK-NEXT: [[TMP1:%.*]] = sext i8 [[Y:%.*]] to i32
+; CHECK-NEXT: [[C:%.*]] = icmp sgt i32 [[TMP1]], [[X:%.*]]
; CHECK-NEXT: ret i1 [[C]]
;
%xt = trunc nsw i32 %x to i16
@@ -492,9 +472,8 @@ define i1 @trunc_signed_nsw_sext(i32 %x, i8 %y) {
define i1 @trunc_signed_nsw_zext(i32 %x, i8 %y) {
; CHECK-LABEL: @trunc_signed_nsw_zext(
-; CHECK-NEXT: [[XT:%.*]] = trunc nsw i32 [[X:%.*]] to i16
-; CHECK-NEXT: [[YE:%.*]] = zext i8 [[Y:%.*]] to i16
-; CHECK-NEXT: [[C:%.*]] = icmp slt i16 [[XT]], [[YE]]
+; CHECK-NEXT: [[TMP1:%.*]] = zext i8 [[Y:%.*]] to i32
+; CHECK-NEXT: [[C:%.*]] = icmp sgt i32 [[TMP1]], [[X:%.*]]
; CHECK-NEXT: ret i1 [[C]]
;
%xt = trunc nsw i32 %x to i16
@@ -531,9 +510,8 @@ define i1 @trunc_signed_nuw_zext(i32 %x, i8 %y) {
define i1 @trunc_equality_nuw_zext(i32 %x, i8 %y) {
; CHECK-LABEL: @trunc_equality_nuw_zext(
-; CHECK-NEXT: [[XT:%.*]] = trunc nuw i32 [[X:%.*]] to i16
-; CHECK-NEXT: [[YE:%.*]] = zext i8 [[Y:%.*]] to i16
-; CHECK-NEXT: [[C:%.*]] = icmp ne i16 [[XT]], [[YE]]
+; CHECK-NEXT: [[TMP1:%.*]] = zext i8 [[Y:%.*]] to i32
+; CHECK-NEXT: [[C:%.*]] = icmp ne i32 [[TMP1]], [[X:%.*]]
; CHECK-NEXT: ret i1 [[C]]
;
%xt = trunc nuw i32 %x to i16
@@ -557,9 +535,8 @@ define i1 @trunc_equality_nuw_sext(i32 %x, i8 %y) {
define i1 @trunc_equality_nsw_zext(i32 %x, i8 %y) {
; CHECK-LABEL: @trunc_equality_nsw_zext(
-; CHECK-NEXT: [[XT:%.*]] = trunc nsw i32 [[X:%.*]] to i16
-; CHECK-NEXT: [[YE:%.*]] = zext i8 [[Y:%.*]] to i16
-; CHECK-NEXT: [[C:%.*]] = icmp ne i16 [[XT]], [[YE]]
+; CHECK-NEXT: [[TMP1:%.*]] = zext i8 [[Y:%.*]] to i32
+; CHECK-NEXT: [[C:%.*]] = icmp ne i32 [[TMP1]], [[X:%.*]]
; CHECK-NEXT: ret i1 [[C]]
;
%xt = trunc nsw i32 %x to i16
@@ -570,9 +547,8 @@ define i1 @trunc_equality_nsw_zext(i32 %x, i8 %y) {
define i1 @trunc_equality_nsw_sext(i32 %x, i8 %y) {
; CHECK-LABEL: @trunc_equality_nsw_sext(
-; CHECK-NEXT: [[XT:%.*]] = trunc nsw i32 [[X:%.*]] to i16
-; CHECK-NEXT: [[YE:%.*]] = sext i8 [[Y:%.*]] to i16
-; CHECK-NEXT: [[C:%.*]] = icmp ne i16 [[XT]], [[YE]]
+; CHECK-NEXT: [[TMP1:%.*]] = sext i8 [[Y:%.*]] to i32
+; CHECK-NEXT: [[C:%.*]] = icmp ne i32 [[TMP1]], [[X:%.*]]
; CHECK-NEXT: ret i1 [[C]]
;
%xt = trunc nsw i32 %x to i16
@@ -583,9 +559,8 @@ define i1 @trunc_equality_nsw_sext(i32 %x, i8 %y) {
define i1 @trunc_equality_both_sext(i32 %x, i8 %y) {
; CHECK-LABEL: @trunc_equality_both_sext(
-; CHECK-NEXT: [[XT:%.*]] = trunc nuw nsw i32 [[X:%.*]] to i16
-; CHECK-NEXT: [[YE:%.*]] = sext i8 [[Y:%.*]] to i16
-; CHECK-NEXT: [[C:%.*]] = icmp ne i16 [[XT]], [[YE]]
+; CHECK-NEXT: [[TMP1:%.*]] = sext i8 [[Y:%.*]] to i32
+; CHECK-NEXT: [[C:%.*]] = icmp ne i32 [[TMP1]], [[X:%.*]]
; CHECK-NEXT: ret i1 [[C]]
;
%xt = trunc nuw nsw i32 %x to i16
|
|
✅ With the latest revision this PR passed the C/C++ code formatter. |
Does it improve the compile-time? |
dtcxzyw
added a commit
to dtcxzyw/llvm-opt-benchmark
that referenced
this pull request
Apr 29, 2024
I don't see any significant compile-time change: https://llvm-compile-time-tracker.com/compare.php?from=e1622e189e8c0ef457bfac528f90a7a930d9aad2&to=b48d51be8e5956645d323123550638ba911eef69&stat=instructions:u |
dtcxzyw
added a commit
to dtcxzyw/llvm-opt-benchmark
that referenced
this pull request
Apr 29, 2024
|
We should handle |
There is a PR for that one already here: #87935 |
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.
Convert the existing foldICmpTruncWithTruncOrExt() fold to work with trunc nowrap flags instead of computeKnownBits(). This also allows us to generalize the fold to work with signed comparisons.
Interestingly, apart from the obvious combinations like signed predicates with trunc nsw, some non-obvious ones are also legal. For example for unsigned predicates we can do the transform for two trunc nsw as well (rather than only trunc nuw).
Proofs: https://alive2.llvm.org/ce/z/ndewwK