[SYCL] Remove undocumented descriptors by Robertkq · Pull Request #22542 · intel/llvm

Robertkq · 2026-07-03T17:17:58Z

Removal of non-standard descriptors from multiple classes

Robertkq · 2026-07-05T10:26:26Z

@KornevNikita can I get a review on this PR? Thanks!

I found some more non-standard descriptors for which I am going to open issues

Update: I think I misunderstood the scope of the original issue, I missed that this issue should have covered only device and for the descriptors that were present on other classes like kernel platform context I also removed them, the likes of reference_count. Not sure if I should revert the changes to the other classes or keep them, as they need to be removed either way

KornevNikita

Just a quick Sunday review, will take an additional one tomorrow.

KornevNikita · 2026-07-05T19:36:42Z

@KornevNikita can I get a review on this PR? Thanks!

I found some more non-standard descriptors for which I am going to open issues

Update: I think I misunderstood the scope of the original issue, I missed that this issue should have covered only device and for the descriptors that were present on other classes like kernel platform context I also removed them, the likes of reference_count. Not sure if I should revert the changes to the other classes or keep them, as they need to be removed either way

let's stick to device info descriptors, other classes are covered by different PRs

Robertkq · 2026-07-06T12:30:21Z

Ok.. I think this PR is getting close to final touches if not already. I am experiencing some issues with the target check-sycl-e2e, I constantly get this error now:

[0/2] Re-checking globbed directories...
ninja: error: 'tools/sycl/test-e2e/compiler-rt', needed by 'tools/sycl/test-e2e/CMakeFiles/check-sycl-e2e', missing and no known rule to make it

But I remember being able to run it, I used the builbot scripts to configure, compile and check the build, I also re-ran ninja in the build directory to try and build any remaining targets but I always get this issue. Will take a look at how the tests look to see any issues with the current stage of the PR

KornevNikita · 2026-07-06T15:17:13Z

                    "For a 3D image, the depth must be a Value >= 1 and <= "
                    "info::device::image2d_max_depth");

+#ifndef __INTEL_PREVIEW_BREAKING_CHANGES


not sure if we should remove this block, although looks like legacy. I thing we need to update checkImageValueRange() to call adapter instead (like what info::device::image_max_array_size actually calls)

I tried to do this right now but it's a bit difficult for me to properly add a workaround for it. But logically, if image_max_array_size is non-standard and will be removed, also there is no replacement for it, then we shouldn't test against it anymore, no?

KornevNikita · 2026-07-06T15:17:58Z

Sorry for new requests, didn't catch them all at first.

Robertkq · 2026-07-06T16:48:37Z

Sorry for new requests, didn't catch them all at first.

No worries! Thank you for bringing them up and looking in my PRs! Much appreciated!

Robertkq · 2026-07-06T16:51:15Z

Also I am not sure what to make of CI.. it all fails but it all looks expected

/__w/llvm/llvm/llvm/sycl/test-e2e/Basic/queue/queue_parallel_for_without_context.cpp:13:42: error: 'usm_shared_allocations' is deprecated: use sycl::aspect::usm_shared_allocations instead [-Werror,-Wdeprecated-declarations]
  # |    13 |            .get_info<sycl::info::device::usm_shared_allocations>()) {

Apparently there is -Werrorr specified, so.. its failing expectedly in this case

KornevNikita · 2026-07-06T18:13:12Z

Also I am not sure what to make of CI.. it all fails but it all looks expected
/__w/llvm/llvm/llvm/sycl/test-e2e/Basic/queue/queue_parallel_for_without_context.cpp:13:42: error: 'usm_shared_allocations' is deprecated: use sycl::aspect::usm_shared_allocations instead [-Werror,-Wdeprecated-declarations]
  # |    13 |            .get_info<sycl::info::device::usm_shared_allocations>()) {
Apparently there is -Werrorr specified, so.. its failing expectedly in this case

yep, replace get_info by has(aspect)

Robertkq · 2026-07-07T18:35:38Z

Yeah, I'll be honest, I got really confused in this PR, from all the extra modifications with different classes, also I did not expect deprecated descriptors to have extensions / aspects that are very similar or identical in naming, so it messed up with my head. Can you please re-review now? I have moved what get_info used to return for the descriptors to what has should return now, as I think that is the intended behaviour. For image_impl.cpp, there is no aspect image_max_array_size so I can't really call has for it. Sorry for the mess, this should have been easier than what it has become :)

KornevNikita · 2026-07-08T07:11:56Z

Yeah, I'll be honest, I got really confused in this PR, from all the extra modifications with different classes, also I did not expect deprecated descriptors to have extensions / aspects that are very similar or identical in naming, so it messed up with my head. Can you please re-review now? I have moved what get_info used to return for the descriptors to what has should return now, as I think that is the intended behaviour. For image_impl.cpp, there is no aspect image_max_array_size so I can't really call has for it. Sorry for the mess, this should have been easier than what it has become :)

@Robertkq thanks for working on this! In general LGTM, but I assume there will be test failures, so there is still some amount of changes to do.

YuriPlyakhin

the only file change owned by me: sycl/test-e2e/NewOffloadDriver/image_compression.cpp - LGTM

KornevNikita · 2026-07-09T10:35:56Z

@Robertkq just see the "Self build" job, all other failures are related to the tests that haven't been built and therefore found:

/__w/llvm/llvm/llvm/sycl/test-e2e/SubGroup/info.cpp:24:33: error: 'sub_group_independent_forward_progress' is deprecated: extension is deprecated [-Werror,-Wdeprecated-declarations]
  # |    24 |   Device.get_info<info::device::sub_group_independent_forward_progress>();

I suggest to just remove this - it's deprecated and I have plans to remove the support.

For the rest of the tests - use alternatives to the deprecated descriptors if possible, or guard with the preview macro, just a note, I guess you know what to do:)

remove undocumented descriptors from platform Deprecate & remove the rest of descriptors Fix uses of removed descriptors revert changes to other classes

…vice::descriptor)

Robertkq requested a review from a team as a code owner July 3, 2026 17:17

Robertkq requested a review from sergey-semenov July 3, 2026 17:17

Robertkq marked this pull request as draft July 3, 2026 17:18

Robertkq marked this pull request as ready for review July 5, 2026 10:24

KornevNikita reviewed Jul 5, 2026

View reviewed changes

Comment thread sycl/include/sycl/info/context.hpp Outdated

Comment thread sycl/include/sycl/info/device.hpp Outdated

Comment thread sycl/source/detail/device_impl.hpp