[WebAssembly] Constant fold wasm.dot #149619
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
llvmbot
added
llvm:instcombine
|
@llvm/pr-subscribers-llvm-analysis Author: Jasmine Tang (badumbatish) ChangesConstant fold wasm.dot of constant vectors/splats. Test case added in Full diff: https://github.com/llvm/llvm-project/pull/149619.diff 2 Files Affected:
diff --git a/llvm/lib/Analysis/ConstantFolding.cpp b/llvm/lib/Analysis/ConstantFolding.cpp
index 9c1c2c6e60f02..2304c58b3f95f 100644
--- a/llvm/lib/Analysis/ConstantFolding.cpp
+++ b/llvm/lib/Analysis/ConstantFolding.cpp
@@ -1657,6 +1657,7 @@ bool llvm::canConstantFoldCallTo(const CallBase *Call, const Function *F) {
case Intrinsic::aarch64_sve_convert_from_svbool:
case Intrinsic::wasm_alltrue:
case Intrinsic::wasm_anytrue:
+ case Intrinsic::wasm_dot:
// WebAssembly float semantics are always known
case Intrinsic::wasm_trunc_signed:
case Intrinsic::wasm_trunc_unsigned:
@@ -3826,6 +3827,36 @@ static Constant *ConstantFoldFixedVectorCall(
}
return ConstantVector::get(Result);
}
+ case Intrinsic::wasm_dot: {
+ unsigned NumElements =
+ cast<FixedVectorType>(Operands[0]->getType())->getNumElements();
+
+ assert(NumElements == 8 && NumElements / 2 == Result.size() &&
+ "wasm dot takes i16x8 and produce i32x4");
+ assert(Ty->isIntegerTy());
+ SmallVector<APInt, 8> MulVector;
+
+ for (unsigned I = 0; I < NumElements; ++I) {
+ ConstantInt *Elt0 =
+ cast<ConstantInt>(Operands[0]->getAggregateElement(I));
+ ConstantInt *Elt1 =
+ cast<ConstantInt>(Operands[1]->getAggregateElement(I));
+
+ // sext 32 first, according to specs
+ APInt IMul = Elt0->getValue().sext(32) * Elt1->getValue().sext(32);
+
+ // TODO: imul in specs includes a modulo operation
+ // Is this performed automatically via trunc = true in APInt creation of *
+ MulVector.push_back(IMul);
+ }
+ for (unsigned I = 0; I < Result.size(); ++I) {
+ // Same case as with imul
+ APInt IAdd = MulVector[I] + MulVector[I + Result.size()];
+ Result[I] = ConstantInt::get(Ty, IAdd);
+ }
+
+ return ConstantVector::get(Result);
+ }
default:
break;
}
diff --git a/llvm/test/Transforms/InstSimplify/ConstProp/WebAssembly/dot.ll b/llvm/test/Transforms/InstSimplify/ConstProp/WebAssembly/dot.ll
new file mode 100644
index 0000000000000..02c6649becbce
--- /dev/null
+++ b/llvm/test/Transforms/InstSimplify/ConstProp/WebAssembly/dot.ll
@@ -0,0 +1,39 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+
+; RUN: opt -passes=instsimplify -S < %s | FileCheck %s
+
+; Test that intrinsics wasm dot call are constant folded
+
+target triple = "wasm32-unknown-unknown"
+
+
+define <4 x i32> @dot_zero() {
+; CHECK-LABEL: define <4 x i32> @dot_zero() {
+; CHECK-NEXT: ret <4 x i32> zeroinitializer
+;
+ %res = tail call <4 x i32> @llvm.wasm.dot(<8 x i16> zeroinitializer, <8 x i16> zeroinitializer)
+ ret <4 x i32> %res
+}
+
+; a = 1 2 3 4 5 6 7 8
+; b = 1 2 3 4 5 6 7 8
+; k1|k2 = a * b = 1 4 9 16 25 36 49 64
+; k1 + k2 = (1+25) | (4+36) | (9+49) | (16+64)
+; result = 26 | 40 | 58 | 80
+define <4 x i32> @dot_nonzero() {
+; CHECK-LABEL: define <4 x i32> @dot_nonzero() {
+; CHECK-NEXT: ret <4 x i32> <i32 26, i32 40, i32 58, i32 80>
+;
+ %res = tail call <4 x i32> @llvm.wasm.dot(<8 x i16> <i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8>, <8 x i16> <i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8>)
+ ret <4 x i32> %res
+}
+
+define <4 x i32> @dot_doubly_negative() {
+; CHECK-LABEL: define <4 x i32> @dot_doubly_negative() {
+; CHECK-NEXT: ret <4 x i32> splat (i32 2)
+;
+ %res = tail call <4 x i32> @llvm.wasm.dot(<8 x i16> <i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1>, <8 x i16> <i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1>)
+ ret <4 x i32> %res
+}
+
+
|
|
@llvm/pr-subscribers-llvm-transforms Author: Jasmine Tang (badumbatish) ChangesConstant fold wasm.dot of constant vectors/splats. Test case added in Full diff: https://github.com/llvm/llvm-project/pull/149619.diff 2 Files Affected:
diff --git a/llvm/lib/Analysis/ConstantFolding.cpp b/llvm/lib/Analysis/ConstantFolding.cpp
index 9c1c2c6e60f02..2304c58b3f95f 100644
--- a/llvm/lib/Analysis/ConstantFolding.cpp
+++ b/llvm/lib/Analysis/ConstantFolding.cpp
@@ -1657,6 +1657,7 @@ bool llvm::canConstantFoldCallTo(const CallBase *Call, const Function *F) {
case Intrinsic::aarch64_sve_convert_from_svbool:
case Intrinsic::wasm_alltrue:
case Intrinsic::wasm_anytrue:
+ case Intrinsic::wasm_dot:
// WebAssembly float semantics are always known
case Intrinsic::wasm_trunc_signed:
case Intrinsic::wasm_trunc_unsigned:
@@ -3826,6 +3827,36 @@ static Constant *ConstantFoldFixedVectorCall(
}
return ConstantVector::get(Result);
}
+ case Intrinsic::wasm_dot: {
+ unsigned NumElements =
+ cast<FixedVectorType>(Operands[0]->getType())->getNumElements();
+
+ assert(NumElements == 8 && NumElements / 2 == Result.size() &&
+ "wasm dot takes i16x8 and produce i32x4");
+ assert(Ty->isIntegerTy());
+ SmallVector<APInt, 8> MulVector;
+
+ for (unsigned I = 0; I < NumElements; ++I) {
+ ConstantInt *Elt0 =
+ cast<ConstantInt>(Operands[0]->getAggregateElement(I));
+ ConstantInt *Elt1 =
+ cast<ConstantInt>(Operands[1]->getAggregateElement(I));
+
+ // sext 32 first, according to specs
+ APInt IMul = Elt0->getValue().sext(32) * Elt1->getValue().sext(32);
+
+ // TODO: imul in specs includes a modulo operation
+ // Is this performed automatically via trunc = true in APInt creation of *
+ MulVector.push_back(IMul);
+ }
+ for (unsigned I = 0; I < Result.size(); ++I) {
+ // Same case as with imul
+ APInt IAdd = MulVector[I] + MulVector[I + Result.size()];
+ Result[I] = ConstantInt::get(Ty, IAdd);
+ }
+
+ return ConstantVector::get(Result);
+ }
default:
break;
}
diff --git a/llvm/test/Transforms/InstSimplify/ConstProp/WebAssembly/dot.ll b/llvm/test/Transforms/InstSimplify/ConstProp/WebAssembly/dot.ll
new file mode 100644
index 0000000000000..02c6649becbce
--- /dev/null
+++ b/llvm/test/Transforms/InstSimplify/ConstProp/WebAssembly/dot.ll
@@ -0,0 +1,39 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+
+; RUN: opt -passes=instsimplify -S < %s | FileCheck %s
+
+; Test that intrinsics wasm dot call are constant folded
+
+target triple = "wasm32-unknown-unknown"
+
+
+define <4 x i32> @dot_zero() {
+; CHECK-LABEL: define <4 x i32> @dot_zero() {
+; CHECK-NEXT: ret <4 x i32> zeroinitializer
+;
+ %res = tail call <4 x i32> @llvm.wasm.dot(<8 x i16> zeroinitializer, <8 x i16> zeroinitializer)
+ ret <4 x i32> %res
+}
+
+; a = 1 2 3 4 5 6 7 8
+; b = 1 2 3 4 5 6 7 8
+; k1|k2 = a * b = 1 4 9 16 25 36 49 64
+; k1 + k2 = (1+25) | (4+36) | (9+49) | (16+64)
+; result = 26 | 40 | 58 | 80
+define <4 x i32> @dot_nonzero() {
+; CHECK-LABEL: define <4 x i32> @dot_nonzero() {
+; CHECK-NEXT: ret <4 x i32> <i32 26, i32 40, i32 58, i32 80>
+;
+ %res = tail call <4 x i32> @llvm.wasm.dot(<8 x i16> <i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8>, <8 x i16> <i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8>)
+ ret <4 x i32> %res
+}
+
+define <4 x i32> @dot_doubly_negative() {
+; CHECK-LABEL: define <4 x i32> @dot_doubly_negative() {
+; CHECK-NEXT: ret <4 x i32> splat (i32 2)
+;
+ %res = tail call <4 x i32> @llvm.wasm.dot(<8 x i16> <i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1>, <8 x i16> <i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1>)
+ ret <4 x i32> %res
+}
+
+
|
tlively
reviewed
Jul 21, 2025
tlively
reviewed
Jul 25, 2025
| // sext 32 first, according to specs | ||
| APInt IMul = Elt0->getValue().sext(32) * Elt1->getValue().sext(32); | ||
|
|
||
| // i16 -> i32 bypasses specs modulo on imul |
There was a problem hiding this comment.
Can you elaborate on what "bypasses specs modulo" means? Would it be correct to say "Do not truncate the 32-bit result of the multiplication"?
There was a problem hiding this comment.
yes, i was thinking since we're just multiplying two i16 together, there's no way that the modulo operation with 2^32 on this multiplication will affect the result, thus the word bypass.
Should i change this to something like multilplying two originally 16bit integers doesn't need a modulo of 2^32?
There was a problem hiding this comment.
I would just leave out the comment if the spec doesn't mention anything about truncation. No need to worry about it if the spec doesn't!
lukel97
reviewed
Jul 28, 2025
Comment on lines
+3834
to
+3835
| assert(NumElements == 8 && Result.size() == 4 && | ||
| "wasm dot takes i16x8 and produces i32x4"); |
There was a problem hiding this comment.
Nit
Suggested change
| assert(NumElements == 8 && Result.size() == 4 && | |
| "wasm dot takes i16x8 and produces i32x4"); | |
| assert(NumElements == 8 && Result.size() == 4 && | |
| "wasm_dot takes i16x8 and produces i32x4"); |
| assert(NumElements == 8 && Result.size() == 4 && | ||
| "wasm dot takes i16x8 and produces i32x4"); | ||
| assert(Ty->isIntegerTy()); | ||
| SmallVector<APInt, 8> MulVector; |
There was a problem hiding this comment.
Nit, if we don't need the arbitrary precisioness of APInt or the dynamicness of a vector, we can just use a plain old int32_t array:
Suggested change
| SmallVector<APInt, 8> MulVector; | |
| int32_t MulVector[8]; |
And then you can use ConstantInt::getSExtValue to get the values out. It should be a little bit faster than using an APInt.
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.
Constant fold wasm.dot of constant vectors/splats.
Test case added in
llvm/test/Transforms/InstSimplify/ConstProp/WebAssembly/dot.llRelated to #55933