[onnxruntime/build] Add CI testing for ORT build with generic interface

[karim-vad]

[onnxruntime/build] Add CI testing for ORT build with generic interface

Summary:
- Remove unused cmake variables
- Add target specific logic when generic interface is used.
- Add QNN EP test case that use ORT generic interface build

[github-actions]

You can commit the suggested changes from lintrunner.

[github-actions]

You can commit the suggested changes from lintrunner.

[adrianlizarraga]

Hi @karim-vad, I'm not fully understanding what this PR is testing. Is it essentially running CI/tests with the combinations:

• --enable_generic_interface --use_qnn
• --enable_generic_interface --use_tensorrt
• etc

[karim-vad]

Hi @karim-vad, I'm not fully understanding what this PR is testing. Is it essentially running CI/tests with the combinations:

• --enable_generic_interface --use_qnn
• --enable_generic_interface --use_tensorrt
• etc

@adrianlizarraga this PR is forcing all the onnxruntime.dll builds with enable_generic_interface flags when the existing CI tests uses one of the primary EPs ( OpenVino, TensorRT, QNN) . The successful execution of the existing CI test validates the Generic interface for ORT is build correctly.

In addition to that, cmake system is incorrectly settings certain pre-processor macros when generic interface is enabled. so this PR fixes those as well.

The next steps is to figure out how we want to enable CI testing with generic interface flag of ORT ?

1. Create new set of CI tests that validates ORT + EP combos ?
2. Leverage existing CI tests to set the generic interface flag when OpenVino or TensorRT or QNN EP in use ?

[adrianlizarraga]

In addition to that, cmake system is incorrectly settings certain pre-processor macros when generic interface is enabled. so this PR fixes those as well.

I see. If these cmake changes should be merged into main, they should probably be extracted and put into a separate PR. (since we are not on merging this experimental PR). This would allow us to get reviews from others who maintain build configs.

The next steps is to figure out how we want to enable CI testing with generic interface flag of ORT ?

1. Create new set of CI tests that validates ORT + EP combos ?
2. Leverage existing CI tests to set the generic interface flag when OpenVino or TensorRT or QNN EP in use ?

From this comment, @snnn indicated that testing all EP combos would not be acceptable due to doubling of resources. Perhaps we can limit this to QNN, TensorRT, OpenVINO. We can start with a PR that only enables this with QNN (as an example). We can use the yaml "strategy matrix" to run tests with --enable_generic_interface --use_qnn:

onnxruntime/tools/ci_build/github/azure-pipelines/win-qnn-arm64-ci-pipeline.yml

Lines 49 to 54 in 989c96c

	strategy:
	matrix:
	SHARED_LIB:
	QnnLibKind: 'shared_lib'
	STATIC_LIB:
	QnnLibKind: 'static_lib'

  strategy:
    matrix:
      SHARED_LIB:
        QnnLibKind: 'shared_lib'
        qnn_build_args: '--use_qnn'
      STATIC_LIB:
        QnnLibKind: 'static_lib'
        qnn_build_args: '--use_qnn'
      SHARED_LIB_GENERIC_INTERFACE:
        QnnLibKind: 'shared_lib'
        qnn_build_args: '--use_qnn --use_generic_interface'

Maybe something like the above to get us going.

[snnn]

/azp run Linux GPU CI Pipeline

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).