onnx-mlir
onnx-mlir copied to clipboard
onnx
Split inferShapes and add verify to PRelu Op.
First pull request on onnx-mlir, mostly follow the code like Pow Op. Maybe need to add some error test, but not quite sure.
Thanks @AlexandreEichenberger and @tungld .
I also notice the
func @test_pow_verifier_1() in invalud.mlir with no expected-error,
And need another patch to fix it.
@redbopo could you resolve the conflict and fix the DCO issue? You can click on Details of DCO and follow the instruction.
I also notice the func @test_pow_verifier_1() in invalud.mlir with no expected-error,
Thanks for finding this! It's ok to fix it here if you are interested.
Thanks @tungld , fix DCO and currently I not deal with test_pow_verifier_1() in this patch , will do it in another .
Hello @tungld , aggreed to CLA now, and merge the main branch. Is that only first time to contribute , will need this CLA and jenkins-droid cannot auto-run ?
I believe the first few times you need approval of folks with more privileges.
@redbopo could you please resolve the conflicts and address the above comment? Thanks!
@redbopo I saw clang-format failed with this info:
--- ./src/Dialect/ONNX/ONNXOps.cpp (original)
+++ ./src/Dialect/ONNX/ONNXOps.cpp (reformatted)
@@ -1133,7 +1133,8 @@
LogicalResult ONNXPReluOp::verify() {
ArrayRef<int64_t> xShape = X().getType().cast<ShapedType>().getShape();
- ArrayRef<int64_t> slopeShape = slope().getType().cast<ShapedType>().getShape();
+ ArrayRef<int64_t> slopeShape =
+ slope().getType().cast<ShapedType>().getShape();
// PRelu supports unidirectional broadcasting, that is slope should be
// unidirectional broadcastable to input X.
Please update your code accordingly. Thanks!
@redbopo you forgot signing DCO for some last commits, please sign and it's ready to merge. Thanks!