[llvm] [InstCombine] Decompose more icmps into masks (PR #110836)
via llvm-commits
llvm-commits at lists.llvm.org
Wed Oct 2 05:50:56 PDT 2024
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-llvm-transforms
Author: Nikita Popov (nikic)
<details>
<summary>Changes</summary>
Extend decomposeBitTestICmp() to handle cases where the resulting comparison is of the form `icmp (X & Mask) pred Cmp` with non-zero `Cmp`. Add a flag to allow code to opt-in to this behavior and use it in the "log op of icmp" fold infrastructure.
This addresses regressions from #<!-- -->97289.
Proofs: https://alive2.llvm.org/ce/z/hUhdbU
---
Full diff: https://github.com/llvm/llvm-project/pull/110836.diff
5 Files Affected:
- (modified) llvm/include/llvm/Analysis/CmpInstAnalysis.h (+6-3)
- (modified) llvm/lib/Analysis/CmpInstAnalysis.cpp (+48-12)
- (modified) llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp (+3-2)
- (modified) llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp (+3-18)
- (modified) llvm/test/Transforms/InstCombine/and-or-icmps.ll (+6-12)
``````````diff
diff --git a/llvm/include/llvm/Analysis/CmpInstAnalysis.h b/llvm/include/llvm/Analysis/CmpInstAnalysis.h
index 406dacd930605e..79b325c620f60f 100644
--- a/llvm/include/llvm/Analysis/CmpInstAnalysis.h
+++ b/llvm/include/llvm/Analysis/CmpInstAnalysis.h
@@ -92,18 +92,21 @@ namespace llvm {
Constant *getPredForFCmpCode(unsigned Code, Type *OpTy,
CmpInst::Predicate &Pred);
- /// Represents the operation icmp (X & Mask) pred 0, where pred can only be
+ /// Represents the operation icmp (X & Mask) pred Cmp, where pred can only be
/// eq or ne.
struct DecomposedBitTest {
Value *X;
CmpInst::Predicate Pred;
APInt Mask;
+ APInt Cmp;
};
- /// Decompose an icmp into the form ((X & Mask) pred 0) if possible.
+ /// Decompose an icmp into the form ((X & Mask) pred Cmp) if possible.
+ /// Unless \p AllowNonZeroCmp is true, Cmp will always be 0.
std::optional<DecomposedBitTest>
decomposeBitTestICmp(Value *LHS, Value *RHS, CmpInst::Predicate Pred,
- bool LookThroughTrunc = true);
+ bool LookThroughTrunc = true,
+ bool AllowNonZeroCmp = false);
} // end namespace llvm
diff --git a/llvm/lib/Analysis/CmpInstAnalysis.cpp b/llvm/lib/Analysis/CmpInstAnalysis.cpp
index ad111559b0d850..2fe3aebe8190c1 100644
--- a/llvm/lib/Analysis/CmpInstAnalysis.cpp
+++ b/llvm/lib/Analysis/CmpInstAnalysis.cpp
@@ -75,7 +75,7 @@ Constant *llvm::getPredForFCmpCode(unsigned Code, Type *OpTy,
std::optional<DecomposedBitTest>
llvm::decomposeBitTestICmp(Value *LHS, Value *RHS, CmpInst::Predicate Pred,
- bool LookThruTrunc) {
+ bool LookThruTrunc, bool AllowNonZeroCmp) {
using namespace PatternMatch;
const APInt *OrigC;
@@ -100,22 +100,57 @@ llvm::decomposeBitTestICmp(Value *LHS, Value *RHS, CmpInst::Predicate Pred,
switch (Pred) {
default:
llvm_unreachable("Unexpected predicate");
- case ICmpInst::ICMP_SLT:
+ case ICmpInst::ICMP_SLT: {
// X < 0 is equivalent to (X & SignMask) != 0.
- if (!C.isZero())
- return std::nullopt;
- Result.Mask = APInt::getSignMask(C.getBitWidth());
- Result.Pred = ICmpInst::ICMP_NE;
- break;
+ if (C.isZero()) {
+ Result.Mask = APInt::getSignMask(C.getBitWidth());
+ Result.Cmp = APInt::getZero(C.getBitWidth());
+ Result.Pred = ICmpInst::ICMP_NE;
+ break;
+ }
+
+ APInt FlippedSign = C ^ APInt::getSignMask(C.getBitWidth());
+ if (FlippedSign.isPowerOf2()) {
+ // X s< 10000100 is equivalent to (X & 11111100 == 10000000)
+ Result.Mask = -FlippedSign;
+ Result.Cmp = APInt::getSignMask(C.getBitWidth());
+ Result.Pred = ICmpInst::ICMP_EQ;
+ break;
+ }
+
+ if (FlippedSign.isNegatedPowerOf2()) {
+ // X s< 01111100 is equivalent to (X & 11111100 != 01111100)
+ Result.Mask = FlippedSign;
+ Result.Cmp = C;
+ Result.Pred = ICmpInst::ICMP_NE;
+ break;
+ }
+
+ return std::nullopt;
+ }
case ICmpInst::ICMP_ULT:
// X <u 2^n is equivalent to (X & ~(2^n-1)) == 0.
- if (!C.isPowerOf2())
- return std::nullopt;
- Result.Mask = -C;
- Result.Pred = ICmpInst::ICMP_EQ;
- break;
+ if (C.isPowerOf2()) {
+ Result.Mask = -C;
+ Result.Cmp = APInt::getZero(C.getBitWidth());
+ Result.Pred = ICmpInst::ICMP_EQ;
+ break;
+ }
+
+ // X u< 11111100 is equivalent to (X & 11111100 != 11111100)
+ if (C.isNegatedPowerOf2()) {
+ Result.Mask = C;
+ Result.Cmp = C;
+ Result.Pred = ICmpInst::ICMP_NE;
+ break;
+ }
+
+ return std::nullopt;
}
+ if (!AllowNonZeroCmp && !Result.Cmp.isZero())
+ return std::nullopt;
+
if (Inverted)
Result.Pred = ICmpInst::getInversePredicate(Result.Pred);
@@ -123,6 +158,7 @@ llvm::decomposeBitTestICmp(Value *LHS, Value *RHS, CmpInst::Predicate Pred,
if (LookThruTrunc && match(LHS, m_Trunc(m_Value(X)))) {
Result.X = X;
Result.Mask = Result.Mask.zext(X->getType()->getScalarSizeInBits());
+ Result.Cmp = Result.Cmp.zext(X->getType()->getScalarSizeInBits());
} else {
Result.X = LHS;
}
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp b/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
index e8c0b006616543..52eda8cdee46ba 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
@@ -181,14 +181,15 @@ static unsigned conjugateICmpMask(unsigned Mask) {
// Adapts the external decomposeBitTestICmp for local use.
static bool decomposeBitTestICmp(Value *LHS, Value *RHS, CmpInst::Predicate &Pred,
Value *&X, Value *&Y, Value *&Z) {
- auto Res = llvm::decomposeBitTestICmp(LHS, RHS, Pred);
+ auto Res = llvm::decomposeBitTestICmp(
+ LHS, RHS, Pred, /*LookThroughTrunc=*/true, /*AllowNonZeroCmp=*/true);
if (!Res)
return false;
Pred = Res->Pred;
X = Res->X;
Y = ConstantInt::get(X->getType(), Res->Mask);
- Z = ConstantInt::get(X->getType(), 0);
+ Z = ConstantInt::get(X->getType(), Res->Cmp);
return true;
}
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index d4d45384ec90e3..687e93f948679d 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -5919,29 +5919,14 @@ Instruction *InstCombinerImpl::foldICmpWithTrunc(ICmpInst &ICmp) {
// This matches patterns corresponding to tests of the signbit as well as:
// (trunc X) u< C --> (X & -C) == 0 (are all masked-high-bits clear?)
// (trunc X) u> C --> (X & ~C) != 0 (are any masked-high-bits set?)
- if (auto Res = decomposeBitTestICmp(Op0, Op1, Pred, /*WithTrunc=*/true)) {
+ if (auto Res = decomposeBitTestICmp(Op0, Op1, Pred, /*WithTrunc=*/true,
+ /*AllowNonZeroCmp=*/true)) {
Value *And = Builder.CreateAnd(Res->X, Res->Mask);
- Constant *Zero = ConstantInt::getNullValue(Res->X->getType());
+ Constant *Zero = ConstantInt::get(Res->X->getType(), Res->Cmp);
return new ICmpInst(Res->Pred, And, Zero);
}
unsigned SrcBits = X->getType()->getScalarSizeInBits();
- if (Pred == ICmpInst::ICMP_ULT && C->isNegatedPowerOf2()) {
- // If C is a negative power-of-2 (high-bit mask):
- // (trunc X) u< C --> (X & C) != C (are any masked-high-bits clear?)
- Constant *MaskC = ConstantInt::get(X->getType(), C->zext(SrcBits));
- Value *And = Builder.CreateAnd(X, MaskC);
- return new ICmpInst(ICmpInst::ICMP_NE, And, MaskC);
- }
-
- if (Pred == ICmpInst::ICMP_UGT && (~*C).isPowerOf2()) {
- // If C is not-of-power-of-2 (one clear bit):
- // (trunc X) u> C --> (X & (C+1)) == C+1 (are all masked-high-bits set?)
- Constant *MaskC = ConstantInt::get(X->getType(), (*C + 1).zext(SrcBits));
- Value *And = Builder.CreateAnd(X, MaskC);
- return new ICmpInst(ICmpInst::ICMP_EQ, And, MaskC);
- }
-
if (auto *II = dyn_cast<IntrinsicInst>(X)) {
if (II->getIntrinsicID() == Intrinsic::cttz ||
II->getIntrinsicID() == Intrinsic::ctlz) {
diff --git a/llvm/test/Transforms/InstCombine/and-or-icmps.ll b/llvm/test/Transforms/InstCombine/and-or-icmps.ll
index 26f708dc787c7d..ad28ad980de5b4 100644
--- a/llvm/test/Transforms/InstCombine/and-or-icmps.ll
+++ b/llvm/test/Transforms/InstCombine/and-or-icmps.ll
@@ -3335,10 +3335,8 @@ define i1 @icmp_eq_or_z_or_pow2orz_fail_bad_pred2(i8 %x, i8 %y) {
define i1 @and_slt_to_mask(i8 %x) {
; CHECK-LABEL: @and_slt_to_mask(
-; CHECK-NEXT: [[CMP:%.*]] = icmp slt i8 [[X:%.*]], -124
-; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], 2
-; CHECK-NEXT: [[CMP2:%.*]] = icmp eq i8 [[AND]], 0
-; CHECK-NEXT: [[AND2:%.*]] = and i1 [[CMP]], [[CMP2]]
+; CHECK-NEXT: [[TMP1:%.*]] = and i8 [[X:%.*]], -2
+; CHECK-NEXT: [[AND2:%.*]] = icmp eq i8 [[TMP1]], -128
; CHECK-NEXT: ret i1 [[AND2]]
;
%cmp = icmp slt i8 %x, -124
@@ -3365,10 +3363,8 @@ define i1 @and_slt_to_mask_off_by_one(i8 %x) {
define i1 @and_sgt_to_mask(i8 %x) {
; CHECK-LABEL: @and_sgt_to_mask(
-; CHECK-NEXT: [[CMP:%.*]] = icmp sgt i8 [[X:%.*]], 123
-; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], 2
-; CHECK-NEXT: [[CMP2:%.*]] = icmp eq i8 [[AND]], 0
-; CHECK-NEXT: [[AND2:%.*]] = and i1 [[CMP]], [[CMP2]]
+; CHECK-NEXT: [[TMP1:%.*]] = and i8 [[X:%.*]], -2
+; CHECK-NEXT: [[AND2:%.*]] = icmp eq i8 [[TMP1]], 124
; CHECK-NEXT: ret i1 [[AND2]]
;
%cmp = icmp sgt i8 %x, 123
@@ -3395,10 +3391,8 @@ define i1 @and_sgt_to_mask_off_by_one(i8 %x) {
define i1 @and_ugt_to_mask(i8 %x) {
; CHECK-LABEL: @and_ugt_to_mask(
-; CHECK-NEXT: [[CMP:%.*]] = icmp ugt i8 [[X:%.*]], -5
-; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], 2
-; CHECK-NEXT: [[CMP2:%.*]] = icmp eq i8 [[AND]], 0
-; CHECK-NEXT: [[AND2:%.*]] = and i1 [[CMP]], [[CMP2]]
+; CHECK-NEXT: [[TMP1:%.*]] = and i8 [[X:%.*]], -2
+; CHECK-NEXT: [[AND2:%.*]] = icmp eq i8 [[TMP1]], -4
; CHECK-NEXT: ret i1 [[AND2]]
;
%cmp = icmp ugt i8 %x, -5
``````````
</details>
https://github.com/llvm/llvm-project/pull/110836
More information about the llvm-commits
mailing list