[SCCP] Relax two-instruction range checks #158495
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
|
✅ With the latest revision this PR passed the undef deprecator. |
|
@llvm/pr-subscribers-llvm-transforms Author: Yingwei Zheng (dtcxzyw) ChangesIf we know x in R1, the range check Compile-time impact: https://llvm-compile-time-tracker.com/compare.php?from=ead4f3e271fdf6918aef2ede3a7134811147d276&to=fcd06a82a8e04120aa24cd521953740c9ca5d3d4&stat=instructions%3Au Full diff: https://github.com/llvm/llvm-project/pull/158495.diff 2 Files Affected:
diff --git a/llvm/lib/Transforms/Utils/SCCPSolver.cpp b/llvm/lib/Transforms/Utils/SCCPSolver.cpp
index 84485176ad4ff..e239cad257141 100644
--- a/llvm/lib/Transforms/Utils/SCCPSolver.cpp
+++ b/llvm/lib/Transforms/Utils/SCCPSolver.cpp
@@ -19,8 +19,10 @@
#include "llvm/Analysis/ValueLattice.h"
#include "llvm/Analysis/ValueLatticeUtils.h"
#include "llvm/Analysis/ValueTracking.h"
+#include "llvm/IR/ConstantRange.h"
#include "llvm/IR/IRBuilder.h"
#include "llvm/IR/InstVisitor.h"
+#include "llvm/IR/Instructions.h"
#include "llvm/IR/NoFolder.h"
#include "llvm/IR/PatternMatch.h"
#include "llvm/Support/Casting.h"
@@ -284,6 +286,54 @@ static Value *simplifyInstruction(SCCPSolver &Solver,
return Sub;
}
+ // Relax range checks.
+ if (auto *ICmp = dyn_cast<ICmpInst>(&Inst)) {
+ Value *X;
+ auto MatchTwoInstructionExactRangeCheck =
+ [&]() -> std::optional<ConstantRange> {
+ const APInt *RHSC;
+ if (!match(ICmp->getOperand(1), m_APInt(RHSC)))
+ return std::nullopt;
+
+ Value *LHS = ICmp->getOperand(0);
+ ICmpInst::Predicate Pred = ICmp->getPredicate();
+ const APInt *Offset;
+ if (match(LHS, m_OneUse(m_AddLike(m_Value(X), m_APInt(Offset)))))
+ return ConstantRange::makeExactICmpRegion(Pred, *RHSC).sub(*Offset);
+ // Match icmp eq/ne X & NegPow2, C
+ if (ICmp->isEquality()) {
+ const APInt *Mask;
+ if (match(LHS, m_OneUse(m_And(m_Value(X), m_NegatedPower2(Mask)))) &&
+ RHSC->countr_zero() >= Mask->countr_zero()) {
+ ConstantRange CR(*RHSC, *RHSC - *Mask);
+ return Pred == ICmpInst::ICMP_EQ ? CR : CR.inverse();
+ }
+ }
+ return std::nullopt;
+ };
+
+ if (auto CR = MatchTwoInstructionExactRangeCheck()) {
+ ConstantRange LRange = GetRange(X);
+ if (LRange.isFullSet())
+ return nullptr;
+ auto NewCR = CR->exactUnionWith(LRange.inverse());
+ if (!NewCR)
+ return nullptr;
+ // Avoid transforming cases which do not relax the range.
+ if (NewCR.value() == *CR)
+ return nullptr;
+ ICmpInst::Predicate Pred;
+ APInt RHS;
+ if (NewCR->getEquivalentICmp(Pred, RHS)) {
+ IRBuilder<NoFolder> Builder(&Inst);
+ Value *NewICmp =
+ Builder.CreateICmp(Pred, X, ConstantInt::get(X->getType(), RHS));
+ InsertedValues.insert(NewICmp);
+ return NewICmp;
+ }
+ }
+ }
+
return nullptr;
}
diff --git a/llvm/test/Transforms/SCCP/relax-range-checks.ll b/llvm/test/Transforms/SCCP/relax-range-checks.ll
new file mode 100644
index 0000000000000..90722f350aa9e
--- /dev/null
+++ b/llvm/test/Transforms/SCCP/relax-range-checks.ll
@@ -0,0 +1,92 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
+; RUN: opt < %s -passes=sccp -S | FileCheck %s
+
+define i1 @relax_range_check(i8 range(i8 0, 5) %x) {
+; CHECK-LABEL: define i1 @relax_range_check(
+; CHECK-SAME: i8 range(i8 0, 5) [[X:%.*]]) {
+; CHECK-NEXT: [[ADD:%.*]] = add nsw i8 [[X]], -3
+; CHECK-NEXT: [[RET:%.*]] = icmp uge i8 [[X]], 3
+; CHECK-NEXT: ret i1 [[RET]]
+;
+ %add = add i8 %x, -3
+ %ret = icmp ult i8 %add, 2
+ ret i1 %ret
+}
+
+define i1 @relax_range_check_highbits_check(i8 range(i8 2, 0) %x) {
+; CHECK-LABEL: define i1 @relax_range_check_highbits_check(
+; CHECK-SAME: i8 range(i8 2, 0) [[X:%.*]]) {
+; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], -2
+; CHECK-NEXT: [[RET:%.*]] = icmp ult i8 [[X]], 4
+; CHECK-NEXT: ret i1 [[RET]]
+;
+ %and = and i8 %x, -2
+ %ret = icmp eq i8 %and, 2
+ ret i1 %ret
+}
+
+; Negative tests.
+
+define i1 @relax_range_check_one_instruction(i8 range(i8 0, 5) %x) {
+; CHECK-LABEL: define i1 @relax_range_check_one_instruction(
+; CHECK-SAME: i8 range(i8 0, 5) [[X:%.*]]) {
+; CHECK-NEXT: [[RET:%.*]] = icmp ult i8 [[X]], 2
+; CHECK-NEXT: ret i1 [[RET]]
+;
+ %ret = icmp ult i8 %x, 2
+ ret i1 %ret
+}
+
+define i1 @relax_range_check_not_profitable(i8 range(i8 0, 6) %x) {
+; CHECK-LABEL: define i1 @relax_range_check_not_profitable(
+; CHECK-SAME: i8 range(i8 0, 6) [[X:%.*]]) {
+; CHECK-NEXT: [[ADD:%.*]] = add nsw i8 [[X]], -3
+; CHECK-NEXT: [[RET:%.*]] = icmp ult i8 [[ADD]], 2
+; CHECK-NEXT: ret i1 [[RET]]
+;
+ %add = add i8 %x, -3
+ %ret = icmp ult i8 %add, 2
+ ret i1 %ret
+}
+
+define i1 @relax_range_check_unknown_range(i64 %x) {
+; CHECK-LABEL: define i1 @relax_range_check_unknown_range(
+; CHECK-SAME: i64 [[X:%.*]]) {
+; CHECK-NEXT: [[AND:%.*]] = and i64 [[X]], -67108864
+; CHECK-NEXT: [[TMP1:%.*]] = icmp eq i64 [[AND]], 0
+; CHECK-NEXT: ret i1 [[TMP1]]
+;
+ %and = and i64 %x, -67108864
+ %test = icmp eq i64 %and, 0
+ ret i1 %test
+}
+
+define i1 @relax_range_check_highbits_check_multiuse(i8 range(i8 2, 0) %x) {
+; CHECK-LABEL: define i1 @relax_range_check_highbits_check_multiuse(
+; CHECK-SAME: i8 range(i8 2, 0) [[X:%.*]]) {
+; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], -2
+; CHECK-NEXT: call void @use(i8 [[AND]])
+; CHECK-NEXT: [[RET:%.*]] = icmp eq i8 [[AND]], 2
+; CHECK-NEXT: ret i1 [[RET]]
+;
+ %and = and i8 %x, -2
+ call void @use(i8 %and)
+ %ret = icmp eq i8 %and, 2
+ ret i1 %ret
+}
+
+define i1 @relax_range_check_multiuse(i8 range(i8 0, 5) %x) {
+; CHECK-LABEL: define i1 @relax_range_check_multiuse(
+; CHECK-SAME: i8 range(i8 0, 5) [[X:%.*]]) {
+; CHECK-NEXT: [[ADD:%.*]] = add nsw i8 [[X]], -3
+; CHECK-NEXT: call void @use(i8 [[ADD]])
+; CHECK-NEXT: [[RET:%.*]] = icmp ult i8 [[ADD]], 2
+; CHECK-NEXT: ret i1 [[RET]]
+;
+ %add = add i8 %x, -3
+ call void @use(i8 %add)
+ %ret = icmp ult i8 %add, 2
+ ret i1 %ret
+}
+
+declare void @use(i8)
|
|
@llvm/pr-subscribers-function-specialization Author: Yingwei Zheng (dtcxzyw) ChangesIf we know x in R1, the range check Compile-time impact: https://llvm-compile-time-tracker.com/compare.php?from=ead4f3e271fdf6918aef2ede3a7134811147d276&to=fcd06a82a8e04120aa24cd521953740c9ca5d3d4&stat=instructions%3Au Full diff: https://github.com/llvm/llvm-project/pull/158495.diff 2 Files Affected:
diff --git a/llvm/lib/Transforms/Utils/SCCPSolver.cpp b/llvm/lib/Transforms/Utils/SCCPSolver.cpp
index 84485176ad4ff..e239cad257141 100644
--- a/llvm/lib/Transforms/Utils/SCCPSolver.cpp
+++ b/llvm/lib/Transforms/Utils/SCCPSolver.cpp
@@ -19,8 +19,10 @@
#include "llvm/Analysis/ValueLattice.h"
#include "llvm/Analysis/ValueLatticeUtils.h"
#include "llvm/Analysis/ValueTracking.h"
+#include "llvm/IR/ConstantRange.h"
#include "llvm/IR/IRBuilder.h"
#include "llvm/IR/InstVisitor.h"
+#include "llvm/IR/Instructions.h"
#include "llvm/IR/NoFolder.h"
#include "llvm/IR/PatternMatch.h"
#include "llvm/Support/Casting.h"
@@ -284,6 +286,54 @@ static Value *simplifyInstruction(SCCPSolver &Solver,
return Sub;
}
+ // Relax range checks.
+ if (auto *ICmp = dyn_cast<ICmpInst>(&Inst)) {
+ Value *X;
+ auto MatchTwoInstructionExactRangeCheck =
+ [&]() -> std::optional<ConstantRange> {
+ const APInt *RHSC;
+ if (!match(ICmp->getOperand(1), m_APInt(RHSC)))
+ return std::nullopt;
+
+ Value *LHS = ICmp->getOperand(0);
+ ICmpInst::Predicate Pred = ICmp->getPredicate();
+ const APInt *Offset;
+ if (match(LHS, m_OneUse(m_AddLike(m_Value(X), m_APInt(Offset)))))
+ return ConstantRange::makeExactICmpRegion(Pred, *RHSC).sub(*Offset);
+ // Match icmp eq/ne X & NegPow2, C
+ if (ICmp->isEquality()) {
+ const APInt *Mask;
+ if (match(LHS, m_OneUse(m_And(m_Value(X), m_NegatedPower2(Mask)))) &&
+ RHSC->countr_zero() >= Mask->countr_zero()) {
+ ConstantRange CR(*RHSC, *RHSC - *Mask);
+ return Pred == ICmpInst::ICMP_EQ ? CR : CR.inverse();
+ }
+ }
+ return std::nullopt;
+ };
+
+ if (auto CR = MatchTwoInstructionExactRangeCheck()) {
+ ConstantRange LRange = GetRange(X);
+ if (LRange.isFullSet())
+ return nullptr;
+ auto NewCR = CR->exactUnionWith(LRange.inverse());
+ if (!NewCR)
+ return nullptr;
+ // Avoid transforming cases which do not relax the range.
+ if (NewCR.value() == *CR)
+ return nullptr;
+ ICmpInst::Predicate Pred;
+ APInt RHS;
+ if (NewCR->getEquivalentICmp(Pred, RHS)) {
+ IRBuilder<NoFolder> Builder(&Inst);
+ Value *NewICmp =
+ Builder.CreateICmp(Pred, X, ConstantInt::get(X->getType(), RHS));
+ InsertedValues.insert(NewICmp);
+ return NewICmp;
+ }
+ }
+ }
+
return nullptr;
}
diff --git a/llvm/test/Transforms/SCCP/relax-range-checks.ll b/llvm/test/Transforms/SCCP/relax-range-checks.ll
new file mode 100644
index 0000000000000..90722f350aa9e
--- /dev/null
+++ b/llvm/test/Transforms/SCCP/relax-range-checks.ll
@@ -0,0 +1,92 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
+; RUN: opt < %s -passes=sccp -S | FileCheck %s
+
+define i1 @relax_range_check(i8 range(i8 0, 5) %x) {
+; CHECK-LABEL: define i1 @relax_range_check(
+; CHECK-SAME: i8 range(i8 0, 5) [[X:%.*]]) {
+; CHECK-NEXT: [[ADD:%.*]] = add nsw i8 [[X]], -3
+; CHECK-NEXT: [[RET:%.*]] = icmp uge i8 [[X]], 3
+; CHECK-NEXT: ret i1 [[RET]]
+;
+ %add = add i8 %x, -3
+ %ret = icmp ult i8 %add, 2
+ ret i1 %ret
+}
+
+define i1 @relax_range_check_highbits_check(i8 range(i8 2, 0) %x) {
+; CHECK-LABEL: define i1 @relax_range_check_highbits_check(
+; CHECK-SAME: i8 range(i8 2, 0) [[X:%.*]]) {
+; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], -2
+; CHECK-NEXT: [[RET:%.*]] = icmp ult i8 [[X]], 4
+; CHECK-NEXT: ret i1 [[RET]]
+;
+ %and = and i8 %x, -2
+ %ret = icmp eq i8 %and, 2
+ ret i1 %ret
+}
+
+; Negative tests.
+
+define i1 @relax_range_check_one_instruction(i8 range(i8 0, 5) %x) {
+; CHECK-LABEL: define i1 @relax_range_check_one_instruction(
+; CHECK-SAME: i8 range(i8 0, 5) [[X:%.*]]) {
+; CHECK-NEXT: [[RET:%.*]] = icmp ult i8 [[X]], 2
+; CHECK-NEXT: ret i1 [[RET]]
+;
+ %ret = icmp ult i8 %x, 2
+ ret i1 %ret
+}
+
+define i1 @relax_range_check_not_profitable(i8 range(i8 0, 6) %x) {
+; CHECK-LABEL: define i1 @relax_range_check_not_profitable(
+; CHECK-SAME: i8 range(i8 0, 6) [[X:%.*]]) {
+; CHECK-NEXT: [[ADD:%.*]] = add nsw i8 [[X]], -3
+; CHECK-NEXT: [[RET:%.*]] = icmp ult i8 [[ADD]], 2
+; CHECK-NEXT: ret i1 [[RET]]
+;
+ %add = add i8 %x, -3
+ %ret = icmp ult i8 %add, 2
+ ret i1 %ret
+}
+
+define i1 @relax_range_check_unknown_range(i64 %x) {
+; CHECK-LABEL: define i1 @relax_range_check_unknown_range(
+; CHECK-SAME: i64 [[X:%.*]]) {
+; CHECK-NEXT: [[AND:%.*]] = and i64 [[X]], -67108864
+; CHECK-NEXT: [[TMP1:%.*]] = icmp eq i64 [[AND]], 0
+; CHECK-NEXT: ret i1 [[TMP1]]
+;
+ %and = and i64 %x, -67108864
+ %test = icmp eq i64 %and, 0
+ ret i1 %test
+}
+
+define i1 @relax_range_check_highbits_check_multiuse(i8 range(i8 2, 0) %x) {
+; CHECK-LABEL: define i1 @relax_range_check_highbits_check_multiuse(
+; CHECK-SAME: i8 range(i8 2, 0) [[X:%.*]]) {
+; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], -2
+; CHECK-NEXT: call void @use(i8 [[AND]])
+; CHECK-NEXT: [[RET:%.*]] = icmp eq i8 [[AND]], 2
+; CHECK-NEXT: ret i1 [[RET]]
+;
+ %and = and i8 %x, -2
+ call void @use(i8 %and)
+ %ret = icmp eq i8 %and, 2
+ ret i1 %ret
+}
+
+define i1 @relax_range_check_multiuse(i8 range(i8 0, 5) %x) {
+; CHECK-LABEL: define i1 @relax_range_check_multiuse(
+; CHECK-SAME: i8 range(i8 0, 5) [[X:%.*]]) {
+; CHECK-NEXT: [[ADD:%.*]] = add nsw i8 [[X]], -3
+; CHECK-NEXT: call void @use(i8 [[ADD]])
+; CHECK-NEXT: [[RET:%.*]] = icmp ult i8 [[ADD]], 2
+; CHECK-NEXT: ret i1 [[RET]]
+;
+ %add = add i8 %x, -3
+ call void @use(i8 %add)
+ %ret = icmp ult i8 %add, 2
+ ret i1 %ret
+}
+
+declare void @use(i8)
|
andjo403
reviewed
Sep 15, 2025
| // Relax range checks. | ||
| if (auto *ICmp = dyn_cast<ICmpInst>(&Inst)) { | ||
| Value *X; | ||
| auto MatchTwoInstructionExactRangeCheck = |
There was a problem hiding this comment.
almost the same as MatchRangeCheck in #158498 is it possible to have this common somewhere?
There was a problem hiding this comment.
I used to add a common helper in ValueTracking. But I finally decided to duplicate the code because they have different one-use constraints from each other.
| Value *LHS = ICmp->getOperand(0); | ||
| ICmpInst::Predicate Pred = ICmp->getPredicate(); | ||
| const APInt *Offset; | ||
| if (match(LHS, m_OneUse(m_AddLike(m_Value(X), m_APInt(Offset))))) |
There was a problem hiding this comment.
is it not good to remove the dependency to the "add" even if there is muli use?
There was a problem hiding this comment.
It is tuned based on the feedback from opt-bench. See the ir diff in previous runs.
| ConstantRange LRange = GetRange(X); | ||
| if (LRange.isFullSet()) | ||
| return nullptr; | ||
| auto NewCR = CR->exactUnionWith(LRange.inverse()); |
There was a problem hiding this comment.
maybe good to have a comment here that describe why union of the inverse relax the range
| if (LRange.isFullSet()) | ||
| return nullptr; | ||
| auto NewCR = CR->exactUnionWith(LRange.inverse()); | ||
| if (!NewCR) |
There was a problem hiding this comment.
do we also want to check the intersection with LRange as if there only is two values it is possible to have EQ/NE predicate instead? eg if relax_range_check had the argument range [0,4) instead of [0,5)
andjo403
approved these changes
Sep 16, 2025
|
@zyw-bot csmith-quick-fuzz |
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 we know x in R1, the range check
x in R2can be relaxed intox in Union(R2, Inverse(R1)). The latter one may be more efficient if we can represent it with one icmp.Fixes regressions introduced by #156497.
Proof for
(X & -Pow2) == C -> (X - C) < Pow2: https://alive2.llvm.org/ce/z/HMgkuuCompile-time impact: https://llvm-compile-time-tracker.com/compare.php?from=ead4f3e271fdf6918aef2ede3a7134811147d276&to=bee3d902dd505cf9b11499ba4f230e4e8ae96b92&stat=instructions%3Au