Add dynamic executor support to TF plugin. #5686
Conversation
Signed-off-by: Michal Zientkiewicz <michalz@nvidia.com>
| << "B reserved" << meta[i].max_reserved[j] << "B max reserved"; | ||
| if (j != meta[i].out_num - 1) { | ||
| std::cout << ","; | ||
| if (pipeline_handle_) { |
There was a problem hiding this comment.
This was actually a bug, that could cause a SIGSEGV. One possible trigger was to specify an incomplete output shape in Python.
|
CI MESSAGE: [19611845]: BUILD STARTED |
|
CI MESSAGE: [19611845]: BUILD FAILED |
| @@ -445,6 +451,9 @@ def __init__( | |||
| output_shapes = self._handle_deprecation(output_shapes, shapes, "shapes") | |||
| output_dtypes = self._handle_deprecation(output_dtypes, dtypes, "dtypes") | |||
|
|
|||
| if pipeline._exec_dynamic: | |||
| exec_dynamic = True | |||
There was a problem hiding this comment.
Why not move it to the line 487 and do the following there:
self.exec_dynamic = True if pipeline._exec_dynamic else exec_dynamic
There was a problem hiding this comment.
I thought it was more logical to set up arguments first and then do the "business as usual" of populating the fields...
Signed-off-by: Michal Zientkiewicz <michalz@nvidia.com>
|
CI MESSAGE: [19646278]: BUILD STARTED |
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
| """Test that exec_dynamic is propagated to DALI pipeline from dali_tf.DALIIterator""" | ||
| try: | ||
| tf.compat.v1.disable_eager_execution() | ||
| except ModuleNotFoundError: |
Check notice
Code scanning / CodeQL
Empty except Note test
|
CI MESSAGE: [19646278]: BUILD PASSED |
Category:
New feature (non-breaking change which adds functionality)
Description:
This PR adds support for the new dynamic executor to DALI Tensorflow plugin.
Additional information:
Affected modules and functionalities:
Key points relevant for the review:
Tests:
Checklist
Documentation
DALI team only
Requirements
REQ IDs: N/A
JIRA TASK: DALI-4086