[InstSimplify] Generalize simplification of icmps with monotonic operands #69471
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
goldsteinn
reviewed
Oct 18, 2023
goldsteinn
reviewed
Oct 18, 2023
| @@ -3103,6 +3103,54 @@ static Value *simplifyICmpWithConstant(CmpInst::Predicate Pred, Value *LHS, | |||
| return nullptr; | |||
| } | |||
|
|
|||
| /// Get values V_i such that V uge V_i (Greater) or V ule V_i (!Greater). | |||
| static void getUnsignedMonotonicValues(SmallPtrSetImpl<Value *> &Res, Value *V, | |||
| bool Greater, unsigned Depth = 0) { | |||
There was a problem hiding this comment.
Would replace greater with an enum that also support GreaterOrEq and LessOrEq. Also think a bool for Signed makes sense.
goldsteinn
reviewed
Oct 18, 2023
goldsteinn
reviewed
Oct 18, 2023
| match(V, m_UDiv(m_Value(X), m_Value())) || | ||
| match(V, m_LShr(m_Value(X), m_Value())) || | ||
| match(V, m_Intrinsic<Intrinsic::usub_sat>(m_Value(X)))) { | ||
| getUnsignedMonotonicValues(Res, X, Greater, Depth); |
There was a problem hiding this comment.
Don't need complete coverage here, but can you add TODO for some additional freebies (min/max/add+nuw/sub+nuw/mul+nuw/shl+nuw/ashr+signbit/etc...)
nikic
force-pushed
the
perf/simplify-cmp-monotonic
branch
from
f8b434f to
6d17739
Compare
llvmbot
added
llvm:instcombine
|
@llvm/pr-subscribers-llvm-analysis @llvm/pr-subscribers-llvm-transforms Author: Nikita Popov (nikic) ChangesInstSimplify currently folds patterns like To support this, recursively collect operands of monotonic instructions (that preserve either a greater-or-equal or less-or-equal relationship) and then check whether any of them match. Fixes #69333. Full diff: https://github.com/llvm/llvm-project/pull/69471.diff 2 Files Affected:
diff --git a/llvm/lib/Analysis/InstructionSimplify.cpp b/llvm/lib/Analysis/InstructionSimplify.cpp
index 01b0a089aab718..be8c2e3c520d3b 100644
--- a/llvm/lib/Analysis/InstructionSimplify.cpp
+++ b/llvm/lib/Analysis/InstructionSimplify.cpp
@@ -3070,6 +3070,69 @@ static Value *simplifyICmpWithConstant(CmpInst::Predicate Pred, Value *LHS,
return nullptr;
}
+enum class MonotonicType { GreaterEq, LowerEq };
+
+/// Get values V_i such that V uge V_i (GreaterEq) or V ule V_i (LowerEq).
+static void getUnsignedMonotonicValues(SmallPtrSetImpl<Value *> &Res, Value *V,
+ MonotonicType Type, unsigned Depth = 0) {
+ if (!Res.insert(V).second)
+ return;
+
+ // Can be increased if useful.
+ if (++Depth > 1)
+ return;
+
+ auto *I = dyn_cast<Instruction>(V);
+ if (!I)
+ return;
+
+ Value *X, *Y;
+ if (Type == MonotonicType::GreaterEq) {
+ if (match(I, m_Or(m_Value(X), m_Value(Y))) ||
+ match(I, m_Intrinsic<Intrinsic::uadd_sat>(m_Value(X), m_Value(Y)))) {
+ getUnsignedMonotonicValues(Res, X, Type, Depth);
+ getUnsignedMonotonicValues(Res, Y, Type, Depth);
+ }
+ } else {
+ assert(Type == MonotonicType::LowerEq);
+ switch (I->getOpcode()) {
+ case Instruction::And:
+ getUnsignedMonotonicValues(Res, I->getOperand(0), Type, Depth);
+ getUnsignedMonotonicValues(Res, I->getOperand(1), Type, Depth);
+ break;
+ case Instruction::URem:
+ case Instruction::UDiv:
+ case Instruction::LShr:
+ getUnsignedMonotonicValues(Res, I->getOperand(0), Type, Depth);
+ break;
+ case Instruction::Call:
+ if (match(I, m_Intrinsic<Intrinsic::usub_sat>(m_Value(X))))
+ getUnsignedMonotonicValues(Res, X, Type, Depth);
+ break;
+ default:
+ break;
+ }
+ }
+}
+
+static Value *simplifyICmpUsingMonotonicValues(ICmpInst::Predicate Pred,
+ Value *LHS, Value *RHS) {
+ if (Pred != ICmpInst::ICMP_UGE && Pred != ICmpInst::ICMP_ULT)
+ return nullptr;
+
+ // We have LHS uge GreaterValues and LowerValues uge RHS. If any of the
+ // GreaterValues and LowerValues are the same, it follows that LHS uge RHS.
+ SmallPtrSet<Value *, 4> GreaterValues;
+ SmallPtrSet<Value *, 4> LowerValues;
+ getUnsignedMonotonicValues(GreaterValues, LHS, MonotonicType::GreaterEq);
+ getUnsignedMonotonicValues(LowerValues, RHS, MonotonicType::LowerEq);
+ for (Value *GV : GreaterValues)
+ if (LowerValues.contains(GV))
+ return ConstantInt::getBool(getCompareTy(LHS),
+ Pred == ICmpInst::ICMP_UGE);
+ return nullptr;
+}
+
static Value *simplifyICmpWithBinOpOnLHS(CmpInst::Predicate Pred,
BinaryOperator *LBO, Value *RHS,
const SimplifyQuery &Q,
@@ -3079,11 +3142,6 @@ static Value *simplifyICmpWithBinOpOnLHS(CmpInst::Predicate Pred,
Value *Y = nullptr;
// icmp pred (or X, Y), X
if (match(LBO, m_c_Or(m_Value(Y), m_Specific(RHS)))) {
- if (Pred == ICmpInst::ICMP_ULT)
- return getFalse(ITy);
- if (Pred == ICmpInst::ICMP_UGE)
- return getTrue(ITy);
-
if (Pred == ICmpInst::ICMP_SLT || Pred == ICmpInst::ICMP_SGE) {
KnownBits RHSKnown = computeKnownBits(RHS, /* Depth */ 0, Q);
KnownBits YKnown = computeKnownBits(Y, /* Depth */ 0, Q);
@@ -3094,14 +3152,6 @@ static Value *simplifyICmpWithBinOpOnLHS(CmpInst::Predicate Pred,
}
}
- // icmp pred (and X, Y), X
- if (match(LBO, m_c_And(m_Value(), m_Specific(RHS)))) {
- if (Pred == ICmpInst::ICMP_UGT)
- return getFalse(ITy);
- if (Pred == ICmpInst::ICMP_ULE)
- return getTrue(ITy);
- }
-
// icmp pred (urem X, Y), Y
if (match(LBO, m_URem(m_Value(), m_Specific(RHS)))) {
switch (Pred) {
@@ -3132,27 +3182,6 @@ static Value *simplifyICmpWithBinOpOnLHS(CmpInst::Predicate Pred,
}
}
- // icmp pred (urem X, Y), X
- if (match(LBO, m_URem(m_Specific(RHS), m_Value()))) {
- if (Pred == ICmpInst::ICMP_ULE)
- return getTrue(ITy);
- if (Pred == ICmpInst::ICMP_UGT)
- return getFalse(ITy);
- }
-
- // x >>u y <=u x --> true.
- // x >>u y >u x --> false.
- // x udiv y <=u x --> true.
- // x udiv y >u x --> false.
- if (match(LBO, m_LShr(m_Specific(RHS), m_Value())) ||
- match(LBO, m_UDiv(m_Specific(RHS), m_Value()))) {
- // icmp pred (X op Y), X
- if (Pred == ICmpInst::ICMP_UGT)
- return getFalse(ITy);
- if (Pred == ICmpInst::ICMP_ULE)
- return getTrue(ITy);
- }
-
// If x is nonzero:
// x >>u C <u x --> true for C != 0.
// x >>u C != x --> true for C != 0.
@@ -3702,13 +3731,6 @@ static Value *simplifyICmpWithIntrinsicOnLHS(CmpInst::Predicate Pred,
switch (II->getIntrinsicID()) {
case Intrinsic::uadd_sat:
- // uadd.sat(X, Y) uge X, uadd.sat(X, Y) uge Y
- if (II->getArgOperand(0) == RHS || II->getArgOperand(1) == RHS) {
- if (Pred == ICmpInst::ICMP_UGE)
- return ConstantInt::getTrue(getCompareTy(II));
- if (Pred == ICmpInst::ICMP_ULT)
- return ConstantInt::getFalse(getCompareTy(II));
- }
// uadd.sat(X, Y) uge X + Y
if (match(RHS, m_c_Add(m_Specific(II->getArgOperand(0)),
m_Specific(II->getArgOperand(1))))) {
@@ -3719,13 +3741,6 @@ static Value *simplifyICmpWithIntrinsicOnLHS(CmpInst::Predicate Pred,
}
return nullptr;
case Intrinsic::usub_sat:
- // usub.sat(X, Y) ule X
- if (II->getArgOperand(0) == RHS) {
- if (Pred == ICmpInst::ICMP_ULE)
- return ConstantInt::getTrue(getCompareTy(II));
- if (Pred == ICmpInst::ICMP_UGT)
- return ConstantInt::getFalse(getCompareTy(II));
- }
// usub.sat(X, Y) ule X - Y
if (match(RHS, m_Sub(m_Specific(II->getArgOperand(0)),
m_Specific(II->getArgOperand(1))))) {
@@ -4030,6 +4045,12 @@ static Value *simplifyICmpInst(unsigned Predicate, Value *LHS, Value *RHS,
ICmpInst::getSwappedPredicate(Pred), RHS, LHS))
return V;
+ if (Value *V = simplifyICmpUsingMonotonicValues(Pred, LHS, RHS))
+ return V;
+ if (Value *V = simplifyICmpUsingMonotonicValues(
+ ICmpInst::getSwappedPredicate(Pred), RHS, LHS))
+ return V;
+
if (Value *V = simplifyICmpWithDominatingAssume(Pred, LHS, RHS, Q))
return V;
diff --git a/llvm/test/Transforms/InstSimplify/icmp-monotonic.ll b/llvm/test/Transforms/InstSimplify/icmp-monotonic.ll
index a1daa6bd7b4021..e1a4ee91bd15c5 100644
--- a/llvm/test/Transforms/InstSimplify/icmp-monotonic.ll
+++ b/llvm/test/Transforms/InstSimplify/icmp-monotonic.ll
@@ -4,10 +4,7 @@
define i1 @lshr_or_ule(i32 %x, i32 %y, i32 %z) {
; CHECK-LABEL: define i1 @lshr_or_ule(
; CHECK-SAME: i32 [[X:%.*]], i32 [[Y:%.*]], i32 [[Z:%.*]]) {
-; CHECK-NEXT: [[OP1:%.*]] = lshr i32 [[X]], [[Y]]
-; CHECK-NEXT: [[OP2:%.*]] = or i32 [[X]], [[Z]]
-; CHECK-NEXT: [[CMP:%.*]] = icmp ule i32 [[OP1]], [[OP2]]
-; CHECK-NEXT: ret i1 [[CMP]]
+; CHECK-NEXT: ret i1 true
;
%op1 = lshr i32 %x, %y
%op2 = or i32 %x, %z
@@ -18,10 +15,7 @@ define i1 @lshr_or_ule(i32 %x, i32 %y, i32 %z) {
define i1 @lshr_or_uge_swapped(i32 %x, i32 %y, i32 %z) {
; CHECK-LABEL: define i1 @lshr_or_uge_swapped(
; CHECK-SAME: i32 [[X:%.*]], i32 [[Y:%.*]], i32 [[Z:%.*]]) {
-; CHECK-NEXT: [[OP1:%.*]] = lshr i32 [[X]], [[Y]]
-; CHECK-NEXT: [[OP2:%.*]] = or i32 [[X]], [[Z]]
-; CHECK-NEXT: [[CMP:%.*]] = icmp uge i32 [[OP2]], [[OP1]]
-; CHECK-NEXT: ret i1 [[CMP]]
+; CHECK-NEXT: ret i1 true
;
%op1 = lshr i32 %x, %y
%op2 = or i32 %x, %z
@@ -32,10 +26,7 @@ define i1 @lshr_or_uge_swapped(i32 %x, i32 %y, i32 %z) {
define i1 @lshr_or_ugt(i32 %x, i32 %y, i32 %z) {
; CHECK-LABEL: define i1 @lshr_or_ugt(
; CHECK-SAME: i32 [[X:%.*]], i32 [[Y:%.*]], i32 [[Z:%.*]]) {
-; CHECK-NEXT: [[OP1:%.*]] = lshr i32 [[X]], [[Y]]
-; CHECK-NEXT: [[OP2:%.*]] = or i32 [[X]], [[Z]]
-; CHECK-NEXT: [[CMP:%.*]] = icmp ugt i32 [[OP1]], [[OP2]]
-; CHECK-NEXT: ret i1 [[CMP]]
+; CHECK-NEXT: ret i1 false
;
%op1 = lshr i32 %x, %y
%op2 = or i32 %x, %z
@@ -74,10 +65,7 @@ define i1 @lshr_or_sle_wrong_pred(i32 %x, i32 %y, i32 %z) {
define i1 @lshr_or_swapped_ule(i32 %x, i32 %y, i32 %z) {
; CHECK-LABEL: define i1 @lshr_or_swapped_ule(
; CHECK-SAME: i32 [[X:%.*]], i32 [[Y:%.*]], i32 [[Z:%.*]]) {
-; CHECK-NEXT: [[OP1:%.*]] = lshr i32 [[X]], [[Y]]
-; CHECK-NEXT: [[OP2:%.*]] = or i32 [[Z]], [[X]]
-; CHECK-NEXT: [[CMP:%.*]] = icmp ule i32 [[OP1]], [[OP2]]
-; CHECK-NEXT: ret i1 [[CMP]]
+; CHECK-NEXT: ret i1 true
;
%op1 = lshr i32 %x, %y
%op2 = or i32 %z, %x
@@ -102,10 +90,7 @@ define i1 @lshr_or_ule_invalid_swapped(i32 %x, i32 %y, i32 %z) {
define i1 @and_uadd_sat_ule(i32 %x, i32 %y, i32 %z) {
; CHECK-LABEL: define i1 @and_uadd_sat_ule(
; CHECK-SAME: i32 [[X:%.*]], i32 [[Y:%.*]], i32 [[Z:%.*]]) {
-; CHECK-NEXT: [[OP1:%.*]] = and i32 [[X]], [[Y]]
-; CHECK-NEXT: [[OP2:%.*]] = call i32 @llvm.uadd.sat.i32(i32 [[X]], i32 [[Z]])
-; CHECK-NEXT: [[CMP:%.*]] = icmp ule i32 [[OP1]], [[OP2]]
-; CHECK-NEXT: ret i1 [[CMP]]
+; CHECK-NEXT: ret i1 true
;
%op1 = and i32 %x, %y
%op2 = call i32 @llvm.uadd.sat(i32 %x, i32 %z)
@@ -116,10 +101,7 @@ define i1 @and_uadd_sat_ule(i32 %x, i32 %y, i32 %z) {
define i1 @urem_or_ule(i32 %x, i32 %y, i32 %z) {
; CHECK-LABEL: define i1 @urem_or_ule(
; CHECK-SAME: i32 [[X:%.*]], i32 [[Y:%.*]], i32 [[Z:%.*]]) {
-; CHECK-NEXT: [[OP1:%.*]] = urem i32 [[X]], [[Y]]
-; CHECK-NEXT: [[OP2:%.*]] = or i32 [[X]], [[Z]]
-; CHECK-NEXT: [[CMP:%.*]] = icmp ule i32 [[OP1]], [[OP2]]
-; CHECK-NEXT: ret i1 [[CMP]]
+; CHECK-NEXT: ret i1 true
;
%op1 = urem i32 %x, %y
%op2 = or i32 %x, %z
@@ -144,10 +126,7 @@ define i1 @urem_or_ule_invalid_swapped(i32 %x, i32 %y, i32 %z) {
define i1 @udiv_or_ule(i32 %x, i32 %y, i32 %z) {
; CHECK-LABEL: define i1 @udiv_or_ule(
; CHECK-SAME: i32 [[X:%.*]], i32 [[Y:%.*]], i32 [[Z:%.*]]) {
-; CHECK-NEXT: [[OP1:%.*]] = udiv i32 [[X]], [[Y]]
-; CHECK-NEXT: [[OP2:%.*]] = or i32 [[X]], [[Z]]
-; CHECK-NEXT: [[CMP:%.*]] = icmp ule i32 [[OP1]], [[OP2]]
-; CHECK-NEXT: ret i1 [[CMP]]
+; CHECK-NEXT: ret i1 true
;
%op1 = udiv i32 %x, %y
%op2 = or i32 %x, %z
@@ -172,10 +151,7 @@ define i1 @udiv_or_ule_invalid_swapped(i32 %x, i32 %y, i32 %z) {
define i1 @usub_sat_uadd_sat_ule(i32 %x, i32 %y, i32 %z) {
; CHECK-LABEL: define i1 @usub_sat_uadd_sat_ule(
; CHECK-SAME: i32 [[X:%.*]], i32 [[Y:%.*]], i32 [[Z:%.*]]) {
-; CHECK-NEXT: [[OP1:%.*]] = call i32 @llvm.usub.sat.i32(i32 [[X]], i32 [[Y]])
-; CHECK-NEXT: [[OP2:%.*]] = call i32 @llvm.uadd.sat.i32(i32 [[X]], i32 [[Z]])
-; CHECK-NEXT: [[CMP:%.*]] = icmp ule i32 [[OP1]], [[OP2]]
-; CHECK-NEXT: ret i1 [[CMP]]
+; CHECK-NEXT: ret i1 true
;
%op1 = call i32 @llvm.usub.sat(i32 %x, i32 %y)
%op2 = call i32 @llvm.uadd.sat(i32 %x, i32 %z)
|
|
Some crashes on llvm-opt-benchmark. |
nikic
force-pushed
the
perf/simplify-cmp-monotonic
branch
from
6d17739 to
bdf2ae9
Compare
|
Fixed the crash. |
Given that it's easier to handle than not handle these cases, avoid the order-dependence of transforms.
nikic
force-pushed
the
perf/simplify-cmp-monotonic
branch
from
bdf2ae9 to
29d1226
Compare
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.
InstSimplify currently folds patterns like
(x | y) uge xand(x & y) ule xto true. However, it cannot handle combinations of such situations, such as(x | y) uge (x & z)etc.To support this, recursively collect operands of monotonic instructions (that preserve either a greater-or-equal or less-or-equal relationship) and then check whether any of them match.
Fixes #69333.