Yasushi Negishi

Let me close this PR, because this PR has been replaced by PR https://github.com/onnx/onnx-mlir/pull/1636 . Thank you very much for your comments and I am sorry for troubling you.

@jenkins-droid test this please

It seems that the imported onnx.Scan op is not correct. The newron.app outputs ( https://netron.app/ ) of an onnx.Scan op ("Scan_20") are as follows.  I generated a mlir file...

Results of other test results those may be related to the bidaf-9 compilation issue. - The backend test for "test_scan_sum_cpu" fails, although test for "test_scan9_sum_cpu" passed. - Shape-inference of onnx.ClipV6...

Found the following facts with investigation of the Bidaf-9 model, opsets of onnx.Scan, and onnx-mlir. - Bidaf-9 uses opset-8 of onnx.Scan op with dynamic shape (dynamic batch size). - Onnx-mlir...

Please look at the following outputs of the both cases. ** In the dynamic value== 1 case,** %1110 = "onnx.Mul"(%1096, %1104) : (tensor, tensor) -> tensor %1111 = "onnx.Mul"(%1110, %1105)...

I tried epeatOnnxTransform=10 . but the results were the same. Is there any reason why it works for the the dynamic value=4 case, but does not work for he dynamic...

@AlexandreEichenberger Thank you for your comments. I found that "--onnx-op-transform-threshold=5" (not "--repeatOnnxTransform=5") option can solve this issue. This option can change the maximum number of transform pass repeat times. The...

I am sorry, I used the "--onnx-op-transform-threshold=5" (not "--repeatOnnxTransform=5") option. (I updated my previous comments). It seems that the "--repeatOnnxTransform" option is used only when the "onnxOpTransformThreshold" value

> if no changes occurs at iteration n, do we still execute all of the iteration or we have a simple mechanism to stop? I would not invest a lot...