[Reassociate] Use uint64_t for repeat count #94232
Merged
Conversation
This file contains 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) ChangesThis patch relands #91469 and uses Full diff: https://github.com/llvm/llvm-project/pull/94232.diff 2 Files Affected:
diff --git a/llvm/lib/Transforms/Scalar/Reassociate.cpp b/llvm/lib/Transforms/Scalar/Reassociate.cpp
index c73d7c8d83bec..6cf097094ddd0 100644
--- a/llvm/lib/Transforms/Scalar/Reassociate.cpp
+++ b/llvm/lib/Transforms/Scalar/Reassociate.cpp
@@ -302,98 +302,7 @@ static BinaryOperator *LowerNegateToMultiply(Instruction *Neg) {
return Res;
}
-/// Returns k such that lambda(2^Bitwidth) = 2^k, where lambda is the Carmichael
-/// function. This means that x^(2^k) === 1 mod 2^Bitwidth for
-/// every odd x, i.e. x^(2^k) = 1 for every odd x in Bitwidth-bit arithmetic.
-/// Note that 0 <= k < Bitwidth, and if Bitwidth > 3 then x^(2^k) = 0 for every
-/// even x in Bitwidth-bit arithmetic.
-static unsigned CarmichaelShift(unsigned Bitwidth) {
- if (Bitwidth < 3)
- return Bitwidth - 1;
- return Bitwidth - 2;
-}
-
-/// Add the extra weight 'RHS' to the existing weight 'LHS',
-/// reducing the combined weight using any special properties of the operation.
-/// The existing weight LHS represents the computation X op X op ... op X where
-/// X occurs LHS times. The combined weight represents X op X op ... op X with
-/// X occurring LHS + RHS times. If op is "Xor" for example then the combined
-/// operation is equivalent to X if LHS + RHS is odd, or 0 if LHS + RHS is even;
-/// the routine returns 1 in LHS in the first case, and 0 in LHS in the second.
-static void IncorporateWeight(APInt &LHS, const APInt &RHS, unsigned Opcode) {
- // If we were working with infinite precision arithmetic then the combined
- // weight would be LHS + RHS. But we are using finite precision arithmetic,
- // and the APInt sum LHS + RHS may not be correct if it wraps (it is correct
- // for nilpotent operations and addition, but not for idempotent operations
- // and multiplication), so it is important to correctly reduce the combined
- // weight back into range if wrapping would be wrong.
-
- // If RHS is zero then the weight didn't change.
- if (RHS.isMinValue())
- return;
- // If LHS is zero then the combined weight is RHS.
- if (LHS.isMinValue()) {
- LHS = RHS;
- return;
- }
- // From this point on we know that neither LHS nor RHS is zero.
-
- if (Instruction::isIdempotent(Opcode)) {
- // Idempotent means X op X === X, so any non-zero weight is equivalent to a
- // weight of 1. Keeping weights at zero or one also means that wrapping is
- // not a problem.
- assert(LHS == 1 && RHS == 1 && "Weights not reduced!");
- return; // Return a weight of 1.
- }
- if (Instruction::isNilpotent(Opcode)) {
- // Nilpotent means X op X === 0, so reduce weights modulo 2.
- assert(LHS == 1 && RHS == 1 && "Weights not reduced!");
- LHS = 0; // 1 + 1 === 0 modulo 2.
- return;
- }
- if (Opcode == Instruction::Add || Opcode == Instruction::FAdd) {
- // TODO: Reduce the weight by exploiting nsw/nuw?
- LHS += RHS;
- return;
- }
-
- assert((Opcode == Instruction::Mul || Opcode == Instruction::FMul) &&
- "Unknown associative operation!");
- unsigned Bitwidth = LHS.getBitWidth();
- // If CM is the Carmichael number then a weight W satisfying W >= CM+Bitwidth
- // can be replaced with W-CM. That's because x^W=x^(W-CM) for every Bitwidth
- // bit number x, since either x is odd in which case x^CM = 1, or x is even in
- // which case both x^W and x^(W - CM) are zero. By subtracting off multiples
- // of CM like this weights can always be reduced to the range [0, CM+Bitwidth)
- // which by a happy accident means that they can always be represented using
- // Bitwidth bits.
- // TODO: Reduce the weight by exploiting nsw/nuw? (Could do much better than
- // the Carmichael number).
- if (Bitwidth > 3) {
- /// CM - The value of Carmichael's lambda function.
- APInt CM = APInt::getOneBitSet(Bitwidth, CarmichaelShift(Bitwidth));
- // Any weight W >= Threshold can be replaced with W - CM.
- APInt Threshold = CM + Bitwidth;
- assert(LHS.ult(Threshold) && RHS.ult(Threshold) && "Weights not reduced!");
- // For Bitwidth 4 or more the following sum does not overflow.
- LHS += RHS;
- while (LHS.uge(Threshold))
- LHS -= CM;
- } else {
- // To avoid problems with overflow do everything the same as above but using
- // a larger type.
- unsigned CM = 1U << CarmichaelShift(Bitwidth);
- unsigned Threshold = CM + Bitwidth;
- assert(LHS.getZExtValue() < Threshold && RHS.getZExtValue() < Threshold &&
- "Weights not reduced!");
- unsigned Total = LHS.getZExtValue() + RHS.getZExtValue();
- while (Total >= Threshold)
- Total -= CM;
- LHS = Total;
- }
-}
-
-using RepeatedValue = std::pair<Value*, APInt>;
+using RepeatedValue = std::pair<Value *, uint64_t>;
/// Given an associative binary expression, return the leaf
/// nodes in Ops along with their weights (how many times the leaf occurs). The
@@ -475,7 +384,6 @@ static bool LinearizeExprTree(Instruction *I,
assert((isa<UnaryOperator>(I) || isa<BinaryOperator>(I)) &&
"Expected a UnaryOperator or BinaryOperator!");
LLVM_DEBUG(dbgs() << "LINEARIZE: " << *I << '\n');
- unsigned Bitwidth = I->getType()->getScalarType()->getPrimitiveSizeInBits();
unsigned Opcode = I->getOpcode();
assert(I->isAssociative() && I->isCommutative() &&
"Expected an associative and commutative operation!");
@@ -490,8 +398,8 @@ static bool LinearizeExprTree(Instruction *I,
// with their weights, representing a certain number of paths to the operator.
// If an operator occurs in the worklist multiple times then we found multiple
// ways to get to it.
- SmallVector<std::pair<Instruction*, APInt>, 8> Worklist; // (Op, Weight)
- Worklist.push_back(std::make_pair(I, APInt(Bitwidth, 1)));
+ SmallVector<std::pair<Instruction *, uint64_t>, 8> Worklist; // (Op, Weight)
+ Worklist.push_back(std::make_pair(I, 1));
bool Changed = false;
// Leaves of the expression are values that either aren't the right kind of
@@ -509,7 +417,7 @@ static bool LinearizeExprTree(Instruction *I,
// Leaves - Keeps track of the set of putative leaves as well as the number of
// paths to each leaf seen so far.
- using LeafMap = DenseMap<Value *, APInt>;
+ using LeafMap = DenseMap<Value *, uint64_t>;
LeafMap Leaves; // Leaf -> Total weight so far.
SmallVector<Value *, 8> LeafOrder; // Ensure deterministic leaf output order.
const DataLayout DL = I->getModule()->getDataLayout();
@@ -518,8 +426,8 @@ static bool LinearizeExprTree(Instruction *I,
SmallPtrSet<Value *, 8> Visited; // For checking the iteration scheme.
#endif
while (!Worklist.empty()) {
- std::pair<Instruction*, APInt> P = Worklist.pop_back_val();
- I = P.first; // We examine the operands of this binary operator.
+ // We examine the operands of this binary operator.
+ auto [I, Weight] = Worklist.pop_back_val();
if (isa<OverflowingBinaryOperator>(I)) {
Flags.HasNUW &= I->hasNoUnsignedWrap();
@@ -528,7 +436,6 @@ static bool LinearizeExprTree(Instruction *I,
for (unsigned OpIdx = 0; OpIdx < I->getNumOperands(); ++OpIdx) { // Visit operands.
Value *Op = I->getOperand(OpIdx);
- APInt Weight = P.second; // Number of paths to this operand.
LLVM_DEBUG(dbgs() << "OPERAND: " << *Op << " (" << Weight << ")\n");
assert(!Op->use_empty() && "No uses, so how did we get to it?!");
@@ -562,7 +469,7 @@ static bool LinearizeExprTree(Instruction *I,
"In leaf map but not visited!");
// Update the number of paths to the leaf.
- IncorporateWeight(It->second, Weight, Opcode);
+ It->second += Weight;
// If we still have uses that are not accounted for by the expression
// then it is not safe to modify the value.
@@ -625,10 +532,7 @@ static bool LinearizeExprTree(Instruction *I,
// Node initially thought to be a leaf wasn't.
continue;
assert(!isReassociableOp(V, Opcode) && "Shouldn't be a leaf!");
- APInt Weight = It->second;
- if (Weight.isMinValue())
- // Leaf already output or weight reduction eliminated it.
- continue;
+ uint64_t Weight = It->second;
// Ensure the leaf is only output once.
It->second = 0;
Ops.push_back(std::make_pair(V, Weight));
@@ -642,7 +546,7 @@ static bool LinearizeExprTree(Instruction *I,
if (Ops.empty()) {
Constant *Identity = ConstantExpr::getBinOpIdentity(Opcode, I->getType());
assert(Identity && "Associative operation without identity!");
- Ops.emplace_back(Identity, APInt(Bitwidth, 1));
+ Ops.emplace_back(Identity, 1);
}
return Changed;
@@ -1188,8 +1092,7 @@ Value *ReassociatePass::RemoveFactorFromExpression(Value *V, Value *Factor) {
Factors.reserve(Tree.size());
for (unsigned i = 0, e = Tree.size(); i != e; ++i) {
RepeatedValue E = Tree[i];
- Factors.append(E.second.getZExtValue(),
- ValueEntry(getRank(E.first), E.first));
+ Factors.append(E.second, ValueEntry(getRank(E.first), E.first));
}
bool FoundFactor = false;
@@ -2368,7 +2271,7 @@ void ReassociatePass::ReassociateExpression(BinaryOperator *I) {
SmallVector<ValueEntry, 8> Ops;
Ops.reserve(Tree.size());
for (const RepeatedValue &E : Tree)
- Ops.append(E.second.getZExtValue(), ValueEntry(getRank(E.first), E.first));
+ Ops.append(E.second, ValueEntry(getRank(E.first), E.first));
LLVM_DEBUG(dbgs() << "RAIn:\t"; PrintOps(I, Ops); dbgs() << '\n');
diff --git a/llvm/test/Transforms/Reassociate/repeats.ll b/llvm/test/Transforms/Reassociate/repeats.ll
index ba25c4bfc643c..8600777877bb3 100644
--- a/llvm/test/Transforms/Reassociate/repeats.ll
+++ b/llvm/test/Transforms/Reassociate/repeats.ll
@@ -60,7 +60,8 @@ define i3 @foo3x5(i3 %x) {
; CHECK-SAME: i3 [[X:%.*]]) {
; CHECK-NEXT: [[TMP3:%.*]] = mul i3 [[X]], [[X]]
; CHECK-NEXT: [[TMP4:%.*]] = mul i3 [[TMP3]], [[X]]
-; CHECK-NEXT: ret i3 [[TMP4]]
+; CHECK-NEXT: [[TMP5:%.*]] = mul i3 [[TMP4]], [[TMP3]]
+; CHECK-NEXT: ret i3 [[TMP5]]
;
%tmp1 = mul i3 %x, %x
%tmp2 = mul i3 %tmp1, %x
@@ -74,7 +75,8 @@ define i3 @foo3x5_nsw(i3 %x) {
; CHECK-LABEL: define i3 @foo3x5_nsw(
; CHECK-SAME: i3 [[X:%.*]]) {
; CHECK-NEXT: [[TMP3:%.*]] = mul i3 [[X]], [[X]]
-; CHECK-NEXT: [[TMP4:%.*]] = mul nsw i3 [[TMP3]], [[X]]
+; CHECK-NEXT: [[TMP2:%.*]] = mul i3 [[TMP3]], [[X]]
+; CHECK-NEXT: [[TMP4:%.*]] = mul i3 [[TMP2]], [[TMP3]]
; CHECK-NEXT: ret i3 [[TMP4]]
;
%tmp1 = mul i3 %x, %x
@@ -89,7 +91,8 @@ define i3 @foo3x6(i3 %x) {
; CHECK-LABEL: define i3 @foo3x6(
; CHECK-SAME: i3 [[X:%.*]]) {
; CHECK-NEXT: [[TMP1:%.*]] = mul i3 [[X]], [[X]]
-; CHECK-NEXT: [[TMP2:%.*]] = mul i3 [[TMP1]], [[TMP1]]
+; CHECK-NEXT: [[TMP3:%.*]] = mul i3 [[TMP1]], [[X]]
+; CHECK-NEXT: [[TMP2:%.*]] = mul i3 [[TMP3]], [[TMP3]]
; CHECK-NEXT: ret i3 [[TMP2]]
;
%tmp1 = mul i3 %x, %x
@@ -106,7 +109,9 @@ define i3 @foo3x7(i3 %x) {
; CHECK-SAME: i3 [[X:%.*]]) {
; CHECK-NEXT: [[TMP5:%.*]] = mul i3 [[X]], [[X]]
; CHECK-NEXT: [[TMP6:%.*]] = mul i3 [[TMP5]], [[X]]
-; CHECK-NEXT: ret i3 [[TMP6]]
+; CHECK-NEXT: [[TMP3:%.*]] = mul i3 [[TMP6]], [[X]]
+; CHECK-NEXT: [[TMP7:%.*]] = mul i3 [[TMP3]], [[TMP6]]
+; CHECK-NEXT: ret i3 [[TMP7]]
;
%tmp1 = mul i3 %x, %x
%tmp2 = mul i3 %tmp1, %x
@@ -123,7 +128,8 @@ define i4 @foo4x8(i4 %x) {
; CHECK-SAME: i4 [[X:%.*]]) {
; CHECK-NEXT: [[TMP1:%.*]] = mul i4 [[X]], [[X]]
; CHECK-NEXT: [[TMP4:%.*]] = mul i4 [[TMP1]], [[TMP1]]
-; CHECK-NEXT: ret i4 [[TMP4]]
+; CHECK-NEXT: [[TMP3:%.*]] = mul i4 [[TMP4]], [[TMP4]]
+; CHECK-NEXT: ret i4 [[TMP3]]
;
%tmp1 = mul i4 %x, %x
%tmp2 = mul i4 %tmp1, %x
@@ -140,8 +146,9 @@ define i4 @foo4x9(i4 %x) {
; CHECK-LABEL: define i4 @foo4x9(
; CHECK-SAME: i4 [[X:%.*]]) {
; CHECK-NEXT: [[TMP1:%.*]] = mul i4 [[X]], [[X]]
-; CHECK-NEXT: [[TMP2:%.*]] = mul i4 [[TMP1]], [[X]]
-; CHECK-NEXT: [[TMP8:%.*]] = mul i4 [[TMP2]], [[TMP1]]
+; CHECK-NEXT: [[TMP2:%.*]] = mul i4 [[TMP1]], [[TMP1]]
+; CHECK-NEXT: [[TMP3:%.*]] = mul i4 [[TMP2]], [[X]]
+; CHECK-NEXT: [[TMP8:%.*]] = mul i4 [[TMP3]], [[TMP2]]
; CHECK-NEXT: ret i4 [[TMP8]]
;
%tmp1 = mul i4 %x, %x
@@ -160,7 +167,8 @@ define i4 @foo4x10(i4 %x) {
; CHECK-LABEL: define i4 @foo4x10(
; CHECK-SAME: i4 [[X:%.*]]) {
; CHECK-NEXT: [[TMP1:%.*]] = mul i4 [[X]], [[X]]
-; CHECK-NEXT: [[TMP2:%.*]] = mul i4 [[TMP1]], [[X]]
+; CHECK-NEXT: [[TMP4:%.*]] = mul i4 [[TMP1]], [[TMP1]]
+; CHECK-NEXT: [[TMP2:%.*]] = mul i4 [[TMP4]], [[X]]
; CHECK-NEXT: [[TMP3:%.*]] = mul i4 [[TMP2]], [[TMP2]]
; CHECK-NEXT: ret i4 [[TMP3]]
;
@@ -181,7 +189,8 @@ define i4 @foo4x11(i4 %x) {
; CHECK-LABEL: define i4 @foo4x11(
; CHECK-SAME: i4 [[X:%.*]]) {
; CHECK-NEXT: [[TMP1:%.*]] = mul i4 [[X]], [[X]]
-; CHECK-NEXT: [[TMP2:%.*]] = mul i4 [[TMP1]], [[X]]
+; CHECK-NEXT: [[TMP4:%.*]] = mul i4 [[TMP1]], [[TMP1]]
+; CHECK-NEXT: [[TMP2:%.*]] = mul i4 [[TMP4]], [[X]]
; CHECK-NEXT: [[TMP3:%.*]] = mul i4 [[TMP2]], [[X]]
; CHECK-NEXT: [[TMP10:%.*]] = mul i4 [[TMP3]], [[TMP2]]
; CHECK-NEXT: ret i4 [[TMP10]]
@@ -204,7 +213,9 @@ define i4 @foo4x12(i4 %x) {
; CHECK-LABEL: define i4 @foo4x12(
; CHECK-SAME: i4 [[X:%.*]]) {
; CHECK-NEXT: [[TMP1:%.*]] = mul i4 [[X]], [[X]]
-; CHECK-NEXT: [[TMP2:%.*]] = mul i4 [[TMP1]], [[TMP1]]
+; CHECK-NEXT: [[TMP4:%.*]] = mul i4 [[TMP1]], [[X]]
+; CHECK-NEXT: [[TMP3:%.*]] = mul i4 [[TMP4]], [[TMP4]]
+; CHECK-NEXT: [[TMP2:%.*]] = mul i4 [[TMP3]], [[TMP3]]
; CHECK-NEXT: ret i4 [[TMP2]]
;
%tmp1 = mul i4 %x, %x
@@ -227,7 +238,9 @@ define i4 @foo4x13(i4 %x) {
; CHECK-SAME: i4 [[X:%.*]]) {
; CHECK-NEXT: [[TMP1:%.*]] = mul i4 [[X]], [[X]]
; CHECK-NEXT: [[TMP2:%.*]] = mul i4 [[TMP1]], [[X]]
-; CHECK-NEXT: [[TMP12:%.*]] = mul i4 [[TMP2]], [[TMP1]]
+; CHECK-NEXT: [[TMP3:%.*]] = mul i4 [[TMP2]], [[TMP2]]
+; CHECK-NEXT: [[TMP4:%.*]] = mul i4 [[TMP3]], [[X]]
+; CHECK-NEXT: [[TMP12:%.*]] = mul i4 [[TMP4]], [[TMP3]]
; CHECK-NEXT: ret i4 [[TMP12]]
;
%tmp1 = mul i4 %x, %x
@@ -252,7 +265,9 @@ define i4 @foo4x14(i4 %x) {
; CHECK-NEXT: [[TMP1:%.*]] = mul i4 [[X]], [[X]]
; CHECK-NEXT: [[TMP6:%.*]] = mul i4 [[TMP1]], [[X]]
; CHECK-NEXT: [[TMP7:%.*]] = mul i4 [[TMP6]], [[TMP6]]
-; CHECK-NEXT: ret i4 [[TMP7]]
+; CHECK-NEXT: [[TMP4:%.*]] = mul i4 [[TMP7]], [[X]]
+; CHECK-NEXT: [[TMP5:%.*]] = mul i4 [[TMP4]], [[TMP4]]
+; CHECK-NEXT: ret i4 [[TMP5]]
;
%tmp1 = mul i4 %x, %x
%tmp2 = mul i4 %tmp1, %x
@@ -276,8 +291,10 @@ define i4 @foo4x15(i4 %x) {
; CHECK-SAME: i4 [[X:%.*]]) {
; CHECK-NEXT: [[TMP1:%.*]] = mul i4 [[X]], [[X]]
; CHECK-NEXT: [[TMP6:%.*]] = mul i4 [[TMP1]], [[X]]
-; CHECK-NEXT: [[TMP5:%.*]] = mul i4 [[TMP6]], [[X]]
-; CHECK-NEXT: [[TMP14:%.*]] = mul i4 [[TMP5]], [[TMP6]]
+; CHECK-NEXT: [[TMP3:%.*]] = mul i4 [[TMP6]], [[TMP6]]
+; CHECK-NEXT: [[TMP4:%.*]] = mul i4 [[TMP3]], [[X]]
+; CHECK-NEXT: [[TMP5:%.*]] = mul i4 [[TMP4]], [[X]]
+; CHECK-NEXT: [[TMP14:%.*]] = mul i4 [[TMP5]], [[TMP4]]
; CHECK-NEXT: ret i4 [[TMP14]]
;
%tmp1 = mul i4 %x, %x
|
dtcxzyw
commented
Jun 3, 2024
| if (Weight.isMinValue()) | ||
| // Leaf already output or weight reduction eliminated it. | ||
| continue; | ||
| uint64_t Weight = It->second; |
There was a problem hiding this comment.
As we don't reduce weight for nilpotent ops now (e.g., xor), the weight cannot be zero.
|
I don't see any overflow checks here... is there some reason you can't overflow a uint64_t? |
I think it overflows when there are more than 2^64 instructions in the IR. Can you provide a counterexample? |
|
I don't have a specific example, but there should probably be a comment with the invariants somewhere. (If an operand can show up multiple times in the tree, you might run into issues?) |
dtcxzyw
commented
Jun 7, 2024
| @@ -562,7 +469,8 @@ static bool LinearizeExprTree(Instruction *I, | |||
| "In leaf map but not visited!"); | |||
|
|
|||
| // Update the number of paths to the leaf. | |||
| IncorporateWeight(It->second, Weight, Opcode); | |||
| It->second += Weight; | |||
| assert(It->second >= Weight && "Weight overflows"); | |||
There was a problem hiding this comment.
See
llvm-project/llvm/lib/Support/APInt.cpp
Lines 1905 to 1909 in 374f655
nikic
approved these changes
Jun 7, 2024
There was a problem hiding this comment.
LGTM. I initially thought we could cause overflow with an exponential tree like https://gist.github.com/nikic/e7736ce11aa66b75ac8fa2b0bf112537. But apparently this is not possible, because any value with multiple uses will be considered as a leaf. That means that the weights can only increase linearly and not exponentially.
nekoshirro
pushed a commit
to nekoshirro/Alchemist-LLVM
that referenced
this pull request
Jun 9, 2024
This patch relands llvm#91469 and uses `uint64_t` for repeat count to avoid a miscompilation caused by overflow llvm#91469 (comment). Signed-off-by: Hafidz Muzakky <ais.muzakky@gmail.com>
Closed
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.
This patch relands #91469 and uses
uint64_tfor repeat count to avoid a miscompilation caused by overflow #91469 (comment).