[UR][L0] Implement support for device to device copy

[winstonzhang-intel]

Bindless Image: pitched usm image to pitched usm image copy.

[DBDuncan]

Would be good if you could look at enabling the device to device copy tests in sycl/test-e2e/bindless_images/copies. Have a look at removing // REQUIRES: cuda. You might need to add // UNSUPPORTED: target-amd to prevent failures on AMD.

Also, I am currently trying to get #16862 merged which will enable sycl/test-e2e/bindless_images/copies/device_to_device_copy_3D_subregion.cpp.

[winstonzhang-intel]

Would be good if you could look at enabling the device to device copy tests in sycl/test-e2e/bindless_images/copies. Have a look at removing // REQUIRES: cuda. You might need to add // UNSUPPORTED: target-amd to prevent failures on AMD.

Also, I am currently trying to get #16862 merged which will enable sycl/test-e2e/bindless_images/copies/device_to_device_copy_3D_subregion.cpp.

I'll give that a try, thanks for the comment.

[ProGTX]

Generally looks good, what's the interaction with #17292, though? It seems like this PR should go in first.

[winstonzhang-intel]

As I have just tested, the changes that I've made are not the reason why these tests are failing. So now my question is, before we had copy from non-usm to non-usm memory, what were the tests that were conducted before?

Seems like these failures does not stem from my changes of pitched usm to pitched usm memory.

[JackAKirk]

As I have just tested, the changes that I've made are not the reason why these tests are failing. So now my question is, before we had copy from non-usm to non-usm memory, what were the tests that were conducted before?

Seems like these failures does not stem from my changes of pitched usm to pitched usm memory.

Yeah my understanding is that these tests were always failing, you haven't broken them here, but you haven't fixed them either. You wrote above that you've enabled all tests for pitched usm. I pointed out that you've only enabled tests for 1d pitched usm, and that higher dimensional pitched USM is broken.
The PR title is "Implement support for device to device copies". I think it would make sense to fix them fully, since as I wrote above a complete fix is likely to interact with your partial fix here, so it is likely more efficient to fix everything at once.

Otherwise maybe you could e.g. update the desciption to "Bindless Image: 1D pitched usm image to pitched usm image copy."?

[winstonzhang-intel]

Sounds good, this is more effort than originally projected. I will begin a triage effort on this to get multidimensional bindless copy enabled.

[JackAKirk]

Sounds good, this is more effort than originally projected. I will begin a triage effort on this to get multidimensional bindless copy enabled.

OK sounds good.

[ProGTX]

@winstonzhang-intel what's the status of this PR?