-
Notifications
You must be signed in to change notification settings - Fork 14.3k
[mlir] Guard sccp pass from crashing with different source type #120656
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Conversation
@llvm/pr-subscribers-mlir-vector @llvm/pr-subscribers-mlir Author: Kai Sasaki (Lewuathe) ChangesVector::BroadCastOp expects the identical element type in folding. It causes the crash if the different source type is given to the SCCP pass. We need to guard the pass from crashing if the nonidentical element type is given, but still compatible. (e.g. index vs integer type) Full diff: https://github.com/llvm/llvm-project/pull/120656.diff 2 Files Affected:
diff --git a/mlir/lib/Dialect/Vector/IR/VectorOps.cpp b/mlir/lib/Dialect/Vector/IR/VectorOps.cpp
index ad709813c6216a..4db3ba239eeaf3 100644
--- a/mlir/lib/Dialect/Vector/IR/VectorOps.cpp
+++ b/mlir/lib/Dialect/Vector/IR/VectorOps.cpp
@@ -2523,8 +2523,16 @@ OpFoldResult BroadcastOp::fold(FoldAdaptor adaptor) {
if (!adaptor.getSource())
return {};
auto vectorType = getResultVectorType();
- if (llvm::isa<IntegerAttr, FloatAttr>(adaptor.getSource()))
- return DenseElementsAttr::get(vectorType, adaptor.getSource());
+ if (auto attr = llvm::dyn_cast<IntegerAttr>(adaptor.getSource())) {
+ if (vectorType.getElementType() != attr.getType())
+ return {};
+ return DenseElementsAttr::get(vectorType, attr);
+ }
+ if (auto attr = llvm::dyn_cast<FloatAttr>(adaptor.getSource())) {
+ if (vectorType.getElementType() != attr.getType())
+ return {};
+ return DenseElementsAttr::get(vectorType, attr);
+ }
if (auto attr = llvm::dyn_cast<SplatElementsAttr>(adaptor.getSource()))
return DenseElementsAttr::get(vectorType, attr.getSplatValue<Attribute>());
return {};
diff --git a/mlir/test/Transforms/sccp.mlir b/mlir/test/Transforms/sccp.mlir
index dcae052c29c248..035fc5c21a38ef 100644
--- a/mlir/test/Transforms/sccp.mlir
+++ b/mlir/test/Transforms/sccp.mlir
@@ -246,3 +246,16 @@ func.func @op_with_region() -> (i32) {
^b:
return %1 : i32
}
+
+// CHECK-LABEL: no_crash_with_different_source_type
+func.func @no_crash_with_different_source_type() {
+ // CHECK: llvm.mlir.constant(0 : index) : i64
+ %0 = llvm.mlir.constant(0 : index) : i64
+ llvm.br ^b1(%0 : i64)
+^b1(%1: i64):
+ llvm.br ^b2
+^b2:
+ // CHECK: vector.broadcast %[[CST:.*]] : i64 to vector<128xi64>
+ %2 = vector.broadcast %1 : i64 to vector<128xi64>
+ llvm.return
+}
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Others, LGTM.
if (vectorType.getElementType() != attr.getType()) | ||
return {}; | ||
return DenseElementsAttr::get(vectorType, attr); | ||
} |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
What is the reason for separating the handling of int and float types?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
dyn_cast
cannot take multiple arguments to be casted.
if (auto attr = (llvm::dyn_cast<IntegerAttr>(adaptor.getSource()) || llvm::dyn_cast<FloatAttr>(adaptor.getSource()))) {
Even if I unioned the result of the cast like this, it causes the following error.
llvm-project/mlir/lib/Dialect/Vector/IR/VectorOps.cpp:2527:44: error: member reference base type 'bool' is not a structure or union
if (vectorType.getElementType() != attr.getType())
So we can only separate the logic to handle int and float separetely.
Vector::BroadCastOp expects the idential element type in folding. It causes the crash if the different source type is given to the SCCP pass. We need to guard the pass from crashing if the non-idential element type is given, but still compatible. (e.g. index vs integer type)
c062637
to
ade75d1
Compare
Vector::BroadCastOp expects the identical element type in folding. It causes the crash if the different source type is given to the SCCP pass. We need to guard the pass from crashing if the nonidentical element type is given, but still compatible. (e.g. index vs integer type)
#120193