[SCCP] Simplify [us]cmp(X, Y) into X - Y #144717
Open
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
|
@llvm/pr-subscribers-llvm-transforms Author: Yingwei Zheng (dtcxzyw) ChangesIf the difference between [us]cmp's operands is not greater than 1, we can simplify it into Full diff: https://github.com/llvm/llvm-project/pull/144717.diff 2 Files Affected:
diff --git a/llvm/lib/Transforms/Utils/SCCPSolver.cpp b/llvm/lib/Transforms/Utils/SCCPSolver.cpp
index f4b378b82daec..0e3f0647d6284 100644
--- a/llvm/lib/Transforms/Utils/SCCPSolver.cpp
+++ b/llvm/lib/Transforms/Utils/SCCPSolver.cpp
@@ -245,11 +245,46 @@ static Value *simplifyInstruction(SCCPSolver &Solver,
const APInt *RHSC;
// Remove masking operations.
if (match(&Inst, m_And(m_Value(X), m_LowBitMask(RHSC)))) {
- ConstantRange LRange = GetRange(Inst.getOperand(0));
+ ConstantRange LRange = GetRange(X);
if (LRange.getUnsignedMax().ule(*RHSC))
return X;
}
+ if (auto *II = dyn_cast<IntrinsicInst>(&Inst)) {
+ Intrinsic::ID IID = II->getIntrinsicID();
+ // Check if we can simplify [us]cmp(X, Y) to X - Y.
+ if (IID == Intrinsic::scmp || IID == Intrinsic::ucmp) {
+ Value *LHS = II->getOperand(0);
+ Value *RHS = II->getOperand(1);
+ unsigned BitWidth = LHS->getType()->getScalarSizeInBits();
+ ConstantRange LRange = GetRange(LHS);
+ if (LRange.isSizeLargerThan(3))
+ return nullptr;
+ ConstantRange RRange = GetRange(RHS);
+ if (RRange.isSizeLargerThan(3))
+ return nullptr;
+ ConstantRange RHSLower = RRange.sub(APInt(BitWidth, 1));
+ ConstantRange RHSUpper = RRange.add(APInt(BitWidth, 1));
+ ICmpInst::Predicate Pred =
+ IID == Intrinsic::scmp ? CmpInst::ICMP_SLE : CmpInst::ICMP_ULE;
+ if (!RHSLower.icmp(Pred, LRange) || !LRange.icmp(Pred, RHSUpper))
+ return nullptr;
+ Instruction *Sub = BinaryOperator::CreateSub(LHS, RHS, Inst.getName(),
+ Inst.getIterator());
+ if (IID == Intrinsic::scmp)
+ Sub->setHasNoSignedWrap(true);
+ Sub->setDebugLoc(Inst.getDebugLoc());
+ InsertedValues.insert(Sub);
+ if (Sub->getType() != Inst.getType()) {
+ Sub = CastInst::CreateIntegerCast(Sub, Inst.getType(), true, "",
+ Inst.getIterator());
+ Sub->setDebugLoc(Inst.getDebugLoc());
+ InsertedValues.insert(Sub);
+ }
+ return Sub;
+ }
+ }
+
return nullptr;
}
diff --git a/llvm/test/Transforms/SCCP/uscmp.ll b/llvm/test/Transforms/SCCP/uscmp.ll
new file mode 100644
index 0000000000000..ea5a65556ac55
--- /dev/null
+++ b/llvm/test/Transforms/SCCP/uscmp.ll
@@ -0,0 +1,78 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -passes=sccp -S < %s | FileCheck %s
+
+define i32 @scmp_to_sub(i32 range(i32 -1, 2) %a) {
+; CHECK-LABEL: define i32 @scmp_to_sub(
+; CHECK-SAME: i32 range(i32 -1, 2) [[A:%.*]]) {
+; CHECK-NEXT: [[SCMP:%.*]] = sub nsw i32 [[A]], 0
+; CHECK-NEXT: ret i32 [[SCMP]]
+;
+ %scmp = call i32 @llvm.scmp(i32 %a, i32 0)
+ ret i32 %scmp
+}
+
+define i32 @scmp_zext_to_sub(i1 %a, i1 %b) {
+; CHECK-LABEL: define i32 @scmp_zext_to_sub(
+; CHECK-SAME: i1 [[A:%.*]], i1 [[B:%.*]]) {
+; CHECK-NEXT: [[ZEXT_A:%.*]] = zext i1 [[A]] to i32
+; CHECK-NEXT: [[ZEXT_B:%.*]] = zext i1 [[B]] to i32
+; CHECK-NEXT: [[SCMP:%.*]] = sub nsw i32 [[ZEXT_A]], [[ZEXT_B]]
+; CHECK-NEXT: ret i32 [[SCMP]]
+;
+ %zext_a = zext i1 %a to i32
+ %zext_b = zext i1 %b to i32
+ %scmp = call i32 @llvm.scmp(i32 %zext_a, i32 %zext_b)
+ ret i32 %scmp
+}
+
+define i8 @scmp_to_sub_trunc(i32 range(i32 -1, 2) %a) {
+; CHECK-LABEL: define i8 @scmp_to_sub_trunc(
+; CHECK-SAME: i32 range(i32 -1, 2) [[A:%.*]]) {
+; CHECK-NEXT: [[SCMP1:%.*]] = sub nsw i32 [[A]], 0
+; CHECK-NEXT: [[SCMP:%.*]] = trunc i32 [[SCMP1]] to i8
+; CHECK-NEXT: ret i8 [[SCMP]]
+;
+ %scmp = call i8 @llvm.scmp(i32 %a, i32 0)
+ ret i8 %scmp
+}
+
+define i64 @scmp_to_sub_sext(i32 range(i32 -1, 2) %a) {
+; CHECK-LABEL: define i64 @scmp_to_sub_sext(
+; CHECK-SAME: i32 range(i32 -1, 2) [[A:%.*]]) {
+; CHECK-NEXT: [[SCMP1:%.*]] = sub nsw i32 [[A]], 0
+; CHECK-NEXT: [[SCMP:%.*]] = sext i32 [[SCMP1]] to i64
+; CHECK-NEXT: ret i64 [[SCMP]]
+;
+ %scmp = call i64 @llvm.scmp(i32 %a, i32 0)
+ ret i64 %scmp
+}
+
+define i32 @scmp_to_sub_small_range(i32 range(i32 -1, 1) %a) {
+; CHECK-LABEL: define i32 @scmp_to_sub_small_range(
+; CHECK-SAME: i32 range(i32 -1, 1) [[A:%.*]]) {
+; CHECK-NEXT: [[SCMP:%.*]] = sub nsw i32 [[A]], 0
+; CHECK-NEXT: ret i32 [[SCMP]]
+;
+ %scmp = call i32 @llvm.scmp(i32 %a, i32 0)
+ ret i32 %scmp
+}
+
+define i32 @ucmp_to_sub(i32 range(i32 0, 3) %a) {
+; CHECK-LABEL: define i32 @ucmp_to_sub(
+; CHECK-SAME: i32 range(i32 0, 3) [[A:%.*]]) {
+; CHECK-NEXT: [[SCMP:%.*]] = sub nsw i32 [[A]], 1
+; CHECK-NEXT: ret i32 [[SCMP]]
+;
+ %scmp = call i32 @llvm.scmp(i32 %a, i32 1)
+ ret i32 %scmp
+}
+
+define i32 @scmp_to_sub_large_range(i32 range(i32 -1, 3) %a) {
+; CHECK-LABEL: define i32 @scmp_to_sub_large_range(
+; CHECK-SAME: i32 range(i32 -1, 3) [[A:%.*]]) {
+; CHECK-NEXT: [[SCMP:%.*]] = call i32 @llvm.scmp.i32.i32(i32 [[A]], i32 0)
+; CHECK-NEXT: ret i32 [[SCMP]]
+;
+ %scmp = call i32 @llvm.scmp(i32 %a, i32 0)
+ ret i32 %scmp
+}
|
|
@llvm/pr-subscribers-function-specialization Author: Yingwei Zheng (dtcxzyw) ChangesIf the difference between [us]cmp's operands is not greater than 1, we can simplify it into Full diff: https://github.com/llvm/llvm-project/pull/144717.diff 2 Files Affected:
diff --git a/llvm/lib/Transforms/Utils/SCCPSolver.cpp b/llvm/lib/Transforms/Utils/SCCPSolver.cpp
index f4b378b82daec..0e3f0647d6284 100644
--- a/llvm/lib/Transforms/Utils/SCCPSolver.cpp
+++ b/llvm/lib/Transforms/Utils/SCCPSolver.cpp
@@ -245,11 +245,46 @@ static Value *simplifyInstruction(SCCPSolver &Solver,
const APInt *RHSC;
// Remove masking operations.
if (match(&Inst, m_And(m_Value(X), m_LowBitMask(RHSC)))) {
- ConstantRange LRange = GetRange(Inst.getOperand(0));
+ ConstantRange LRange = GetRange(X);
if (LRange.getUnsignedMax().ule(*RHSC))
return X;
}
+ if (auto *II = dyn_cast<IntrinsicInst>(&Inst)) {
+ Intrinsic::ID IID = II->getIntrinsicID();
+ // Check if we can simplify [us]cmp(X, Y) to X - Y.
+ if (IID == Intrinsic::scmp || IID == Intrinsic::ucmp) {
+ Value *LHS = II->getOperand(0);
+ Value *RHS = II->getOperand(1);
+ unsigned BitWidth = LHS->getType()->getScalarSizeInBits();
+ ConstantRange LRange = GetRange(LHS);
+ if (LRange.isSizeLargerThan(3))
+ return nullptr;
+ ConstantRange RRange = GetRange(RHS);
+ if (RRange.isSizeLargerThan(3))
+ return nullptr;
+ ConstantRange RHSLower = RRange.sub(APInt(BitWidth, 1));
+ ConstantRange RHSUpper = RRange.add(APInt(BitWidth, 1));
+ ICmpInst::Predicate Pred =
+ IID == Intrinsic::scmp ? CmpInst::ICMP_SLE : CmpInst::ICMP_ULE;
+ if (!RHSLower.icmp(Pred, LRange) || !LRange.icmp(Pred, RHSUpper))
+ return nullptr;
+ Instruction *Sub = BinaryOperator::CreateSub(LHS, RHS, Inst.getName(),
+ Inst.getIterator());
+ if (IID == Intrinsic::scmp)
+ Sub->setHasNoSignedWrap(true);
+ Sub->setDebugLoc(Inst.getDebugLoc());
+ InsertedValues.insert(Sub);
+ if (Sub->getType() != Inst.getType()) {
+ Sub = CastInst::CreateIntegerCast(Sub, Inst.getType(), true, "",
+ Inst.getIterator());
+ Sub->setDebugLoc(Inst.getDebugLoc());
+ InsertedValues.insert(Sub);
+ }
+ return Sub;
+ }
+ }
+
return nullptr;
}
diff --git a/llvm/test/Transforms/SCCP/uscmp.ll b/llvm/test/Transforms/SCCP/uscmp.ll
new file mode 100644
index 0000000000000..ea5a65556ac55
--- /dev/null
+++ b/llvm/test/Transforms/SCCP/uscmp.ll
@@ -0,0 +1,78 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -passes=sccp -S < %s | FileCheck %s
+
+define i32 @scmp_to_sub(i32 range(i32 -1, 2) %a) {
+; CHECK-LABEL: define i32 @scmp_to_sub(
+; CHECK-SAME: i32 range(i32 -1, 2) [[A:%.*]]) {
+; CHECK-NEXT: [[SCMP:%.*]] = sub nsw i32 [[A]], 0
+; CHECK-NEXT: ret i32 [[SCMP]]
+;
+ %scmp = call i32 @llvm.scmp(i32 %a, i32 0)
+ ret i32 %scmp
+}
+
+define i32 @scmp_zext_to_sub(i1 %a, i1 %b) {
+; CHECK-LABEL: define i32 @scmp_zext_to_sub(
+; CHECK-SAME: i1 [[A:%.*]], i1 [[B:%.*]]) {
+; CHECK-NEXT: [[ZEXT_A:%.*]] = zext i1 [[A]] to i32
+; CHECK-NEXT: [[ZEXT_B:%.*]] = zext i1 [[B]] to i32
+; CHECK-NEXT: [[SCMP:%.*]] = sub nsw i32 [[ZEXT_A]], [[ZEXT_B]]
+; CHECK-NEXT: ret i32 [[SCMP]]
+;
+ %zext_a = zext i1 %a to i32
+ %zext_b = zext i1 %b to i32
+ %scmp = call i32 @llvm.scmp(i32 %zext_a, i32 %zext_b)
+ ret i32 %scmp
+}
+
+define i8 @scmp_to_sub_trunc(i32 range(i32 -1, 2) %a) {
+; CHECK-LABEL: define i8 @scmp_to_sub_trunc(
+; CHECK-SAME: i32 range(i32 -1, 2) [[A:%.*]]) {
+; CHECK-NEXT: [[SCMP1:%.*]] = sub nsw i32 [[A]], 0
+; CHECK-NEXT: [[SCMP:%.*]] = trunc i32 [[SCMP1]] to i8
+; CHECK-NEXT: ret i8 [[SCMP]]
+;
+ %scmp = call i8 @llvm.scmp(i32 %a, i32 0)
+ ret i8 %scmp
+}
+
+define i64 @scmp_to_sub_sext(i32 range(i32 -1, 2) %a) {
+; CHECK-LABEL: define i64 @scmp_to_sub_sext(
+; CHECK-SAME: i32 range(i32 -1, 2) [[A:%.*]]) {
+; CHECK-NEXT: [[SCMP1:%.*]] = sub nsw i32 [[A]], 0
+; CHECK-NEXT: [[SCMP:%.*]] = sext i32 [[SCMP1]] to i64
+; CHECK-NEXT: ret i64 [[SCMP]]
+;
+ %scmp = call i64 @llvm.scmp(i32 %a, i32 0)
+ ret i64 %scmp
+}
+
+define i32 @scmp_to_sub_small_range(i32 range(i32 -1, 1) %a) {
+; CHECK-LABEL: define i32 @scmp_to_sub_small_range(
+; CHECK-SAME: i32 range(i32 -1, 1) [[A:%.*]]) {
+; CHECK-NEXT: [[SCMP:%.*]] = sub nsw i32 [[A]], 0
+; CHECK-NEXT: ret i32 [[SCMP]]
+;
+ %scmp = call i32 @llvm.scmp(i32 %a, i32 0)
+ ret i32 %scmp
+}
+
+define i32 @ucmp_to_sub(i32 range(i32 0, 3) %a) {
+; CHECK-LABEL: define i32 @ucmp_to_sub(
+; CHECK-SAME: i32 range(i32 0, 3) [[A:%.*]]) {
+; CHECK-NEXT: [[SCMP:%.*]] = sub nsw i32 [[A]], 1
+; CHECK-NEXT: ret i32 [[SCMP]]
+;
+ %scmp = call i32 @llvm.scmp(i32 %a, i32 1)
+ ret i32 %scmp
+}
+
+define i32 @scmp_to_sub_large_range(i32 range(i32 -1, 3) %a) {
+; CHECK-LABEL: define i32 @scmp_to_sub_large_range(
+; CHECK-SAME: i32 range(i32 -1, 3) [[A:%.*]]) {
+; CHECK-NEXT: [[SCMP:%.*]] = call i32 @llvm.scmp.i32.i32(i32 [[A]], i32 0)
+; CHECK-NEXT: ret i32 [[SCMP]]
+;
+ %scmp = call i32 @llvm.scmp(i32 %a, i32 0)
+ ret i32 %scmp
+}
|
nikic
reviewed
Jun 19, 2025
See the first case in the diff: |
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.
If the difference between [us]cmp's operands is not greater than 1, we can simplify it into
X - Y.Alive2: https://alive2.llvm.org/ce/z/JS55so
llvm-opt-benchmark diff: https://github.com/dtcxzyw/llvm-opt-benchmark/pull/2464/files