[SYCL][CUDA] Add mem_advise reset and managed mem check

[AidanBeltonS]

This PR adds the mem_advise reset case and a managed memory check in cuda_piextUSMEnqueueMemAdvise.

cuMemAdvise only works on managed memory and returns an error if host or device memory is passed.
The SYCL-CTS tests mem_advise with host and device memory, which are not managed. This PR prevents an error from being thrown in the cts usm test.

In addition, the sycl spec specifies for mem_advise A value of 0 reverts the advice for ptr to the default behavior, currently a value of 0 is treated as unknown and throws an error in the cts as well. This reset case is added to mem_advise.

This patch along with #5446 resolves issue #5209

This change breaks the ABI

[s-kanaev]

Please, don't change user-visible enum values. Only add new ones.

[AidanBeltonS]

Non ABI breaking changes also being proposed in: #6281

[steffenlarsen]

Non ABI breaking changes also being proposed in: #6281

Sorry @AidanBeltonS . I completely missed this PR. Does #6281 cover everything that this changes or should we coordinate the patches?

[AidanBeltonS]

Sorry @AidanBeltonS . I completely missed this PR. Does #6281 cover everything that this changes or should we coordinate the patches?

No worries. I think #6281 covers all everything that does not require an ABI breaking window so I don't think anything needs to be coordinated. I can update this PR later so that it just makes the enum changes.

[AidanBeltonS]

The alternative is to translate a 0 in mem_advise to whatever value we give PI_MEM_ADVICE_RESET here. Since ABI/API breaks are allowed currently I think it's fine to do like this, though I wonder if we should take the opportunity and move PI_MEM_ADVICE_UNKNOWN to a value that is unlikely to conflict with future advice.

I have set PI_MEM_ADVICE_UNKNOWN to 999, so this should hopefully be a value such that it does not cause any conflicts.

Either way, if this changes ABI/API we need to increment the version number.

Done

[steffenlarsen]

Thanks, @AidanBeltonS! Changes look good to me, but since it changes the PI API and break backwards compatability, the PI version should be incremented as detailed at the top of pi.h.

[steffenlarsen]

LGTM!

[steffenlarsen]

Test failure in Basic/image/image_write_fp16.cpp. @AidanBeltonS - Is this related to this patch?

[AidanBeltonS]

It appears to be unrelated to this patch. There is no use of mem_advice or USM within the test.

[steffenlarsen]

Do you know if it's a known failure? If not, could you please open an issue for investigating it?

[AidanBeltonS]

I do not think it is a known problem. I have created an issue to track it #6491.

[steffenlarsen]

Perfect, thank you! In that case, let's get this merged.