Remove MakeContiguous before CPU inputs of GPU ops.

[mzient]

Category:

Refactoring (Redesign of existing code that doesn't affect functionality)

Description:

• Remove MakeContiguous before CPU inputs of GPU ops.
• Remove redundant (and dangerous) checks in workspace_policy.

Before the unification of TensorList it was necessary to insert a MakeContiguous operator between CPU and other backends because TensorVector needed to be converted to TensorList. It's no longer the case and the operation is superfluous.
The checks used a no-op function in workspace_policy now kicked in and created an invalid workspace - they are no longer necessary and were removed.

Additional information:

Affected modules and functionalities:

Key points relevant for the review:

test_pipeline.py - all of it

• Existing tests apply
• New tests added
  • Python tests
  • GTests
  • Benchmark
  • Other
• N/A

Checklist

Documentation

• Existing documentation applies
• Documentation updated
  • Docstring
  • Doxygen
  • RST
  • Jupyter
  • Other
• N/A

DALI team only

Requirements

• Implements new requirements
• Affects existing requirements
• N/A

REQ IDs: N/A

JIRA TASK: DALI-4030

[mzient]

This test checked for the presence of the feature we've just removed.

[dali-automaton]

CI MESSAGE: [17143006]: BUILD STARTED

[dali-automaton]

CI MESSAGE: [17143849]: BUILD STARTED

[mzient]

I introduced a delay in Executor RunGPU (in different places) and tested it with the following code:

@params((2, 5), (5, 2))
def test_separated_queues(cpu_queue_depth, gpu_queue_depth):
    data = [np.array((10 * (i + 1)), dtype=np.float32) for i in range(10)]
    @pipeline_def(batch_size=1, num_threads=1, device_id=0)
    def pipe():
        inp = fn.external_source(data, batch=False, cycle=True) + 0
        img = types.Constant(np.zeros(shape=(1,1,3), dtype=np.uint8)).gpu()
        return fn.resize(img, resize_x=inp, resize_y=1)  # pass argument input to resize_x

    p = pipe(prefetch_queue_depth={"cpu_size":cpu_queue_depth, "gpu_size":gpu_queue_depth})
    p.build()
    for i in range(10):
        o, = p.run()
        assert o[0].shape()[1] == data[i]

The results were correct, so this seems not necessary now (I wonder if it ever was).

I'm not adding this test to our regular tests, because it's of little value without the delays (which we certainly don't want to add).

[dali-automaton]

CI MESSAGE: [17143849]: BUILD PASSED