Add platform info query to identify its backend adapter

[smaslov-intel]

This allows users to select wanted adapter/platform.

[smaslov-intel]

@kbenzie : please review/merge

[smaslov-intel]

I don't see why there are errors reported in https://github.com/oneapi-src/unified-runtime/actions/runs/4601026787/jobs/8128372658?pr=407

gmake[3]: *** [CMakeFiles/check-generated.dir/build.make:70: CMakeFiles/check-generated] Error 1
gmake[2]: *** [CMakeFiles/Makefile2:1594: CMakeFiles/check-generated.dir/all] Error 2
gmake[1]: *** [CMakeFiles/Makefile2:1601: CMakeFiles/check-generated.dir/rule] Error 2
gmake: *** [Makefile:595: check-generated] Error 2
Error: Process completed with exit code 2.

[igchor]

I don't see why there are errors reported in https://github.com/oneapi-src/unified-runtime/actions/runs/4601026787/jobs/8128372658?pr=407

gmake[3]: *** [CMakeFiles/check-generated.dir/build.make:70: CMakeFiles/check-generated] Error 1
gmake[2]: *** [CMakeFiles/Makefile2:1594: CMakeFiles/check-generated.dir/all] Error 2
gmake[1]: *** [CMakeFiles/Makefile2:1601: CMakeFiles/check-generated.dir/rule] Error 2
gmake: *** [Makefile:595: check-generated] Error 2
Error: Process completed with exit code 2.

@smaslov-intel you need to include generated sources in the PR: you should run make generate locally and commit all changes.

[smaslov-intel]

I don't see why there are errors reported in https://github.com/oneapi-src/unified-runtime/actions/runs/4601026787/jobs/8128372658?pr=407

gmake[3]: *** [CMakeFiles/check-generated.dir/build.make:70: CMakeFiles/check-generated] Error 1
gmake[2]: *** [CMakeFiles/Makefile2:1594: CMakeFiles/check-generated.dir/all] Error 2
gmake[1]: *** [CMakeFiles/Makefile2:1601: CMakeFiles/check-generated.dir/rule] Error 2
gmake: *** [Makefile:595: check-generated] Error 2
Error: Process completed with exit code 2.

@smaslov-intel you need to include generated sources in the PR: you should run make generate locally and commit all changes.

Thanks @igchor !
But make generate produced tons of changes:

$ git commit -asm "make generate"
[main 8b23094] make generate
12 files changed, 31460 insertions(+), 29538 deletions(-)

Is the correct flow described anywhere?

[smaslov-intel]

I don't see why there are errors reported in https://github.com/oneapi-src/unified-runtime/actions/runs/4601026787/jobs/8128372658?pr=407

gmake[3]: *** [CMakeFiles/check-generated.dir/build.make:70: CMakeFiles/check-generated] Error 1
gmake[2]: *** [CMakeFiles/Makefile2:1594: CMakeFiles/check-generated.dir/all] Error 2
gmake[1]: *** [CMakeFiles/Makefile2:1601: CMakeFiles/check-generated.dir/rule] Error 2
gmake: *** [Makefile:595: check-generated] Error 2
Error: Process completed with exit code 2.

@smaslov-intel you need to include generated sources in the PR: you should run make generate locally and commit all changes.

Thanks @igchor ! But make generate produced tons of changes:

$ git commit -asm "make generate"
[main 8b23094] make generate
12 files changed, 31460 insertions(+), 29538 deletions(-)

Is the correct flow described anywhere?

It looks a different clang-format

[igchor]

I don't see why there are errors reported in https://github.com/oneapi-src/unified-runtime/actions/runs/4601026787/jobs/8128372658?pr=407

gmake[3]: *** [CMakeFiles/check-generated.dir/build.make:70: CMakeFiles/check-generated] Error 1
gmake[2]: *** [CMakeFiles/Makefile2:1594: CMakeFiles/check-generated.dir/all] Error 2
gmake[1]: *** [CMakeFiles/Makefile2:1601: CMakeFiles/check-generated.dir/rule] Error 2
gmake: *** [Makefile:595: check-generated] Error 2
Error: Process completed with exit code 2.

@smaslov-intel you need to include generated sources in the PR: you should run make generate locally and commit all changes.

Thanks @igchor ! But make generate produced tons of changes:

$ git commit -asm "make generate"
[main 8b23094] make generate
12 files changed, 31460 insertions(+), 29538 deletions(-)

Is the correct flow described anywhere?

I think there is only this line in the README: If you've made modifications to the specification, you can also run a custom generate target prior to building.

[igchor]

I don't see why there are errors reported in https://github.com/oneapi-src/unified-runtime/actions/runs/4601026787/jobs/8128372658?pr=407

gmake[3]: *** [CMakeFiles/check-generated.dir/build.make:70: CMakeFiles/check-generated] Error 1
gmake[2]: *** [CMakeFiles/Makefile2:1594: CMakeFiles/check-generated.dir/all] Error 2
gmake[1]: *** [CMakeFiles/Makefile2:1601: CMakeFiles/check-generated.dir/rule] Error 2
gmake: *** [Makefile:595: check-generated] Error 2
Error: Process completed with exit code 2.

@smaslov-intel you need to include generated sources in the PR: you should run make generate locally and commit all changes.

Thanks @igchor ! But make generate produced tons of changes:

$ git commit -asm "make generate"
[main 8b23094] make generate
12 files changed, 31460 insertions(+), 29538 deletions(-)

Is the correct flow described anywhere?

It looks a different clang-format

Yes, it looks like a wrong clang-format version, but it's strange - we check for specific version in CMake and provide our own rules. What version do you have installed (if any)?

I've pushed regenerated sources formatted with clang-15, the diff is much smaller.

[smaslov-intel]

I don't see why there are errors reported in https://github.com/oneapi-src/unified-runtime/actions/runs/4601026787/jobs/8128372658?pr=407

gmake[3]: *** [CMakeFiles/check-generated.dir/build.make:70: CMakeFiles/check-generated] Error 1
gmake[2]: *** [CMakeFiles/Makefile2:1594: CMakeFiles/check-generated.dir/all] Error 2
gmake[1]: *** [CMakeFiles/Makefile2:1601: CMakeFiles/check-generated.dir/rule] Error 2
gmake: *** [Makefile:595: check-generated] Error 2
Error: Process completed with exit code 2.

@smaslov-intel you need to include generated sources in the PR: you should run make generate locally and commit all changes.

Thanks @igchor ! But make generate produced tons of changes:

$ git commit -asm "make generate"
[main 8b23094] make generate
12 files changed, 31460 insertions(+), 29538 deletions(-)

Is the correct flow described anywhere?

It looks a different clang-format

Yes, it looks like a wrong clang-format version, but it's strange - we check for specific version in CMake and provide our own rules. What version do you have installed (if any)?

I've pushed regenerated sources formatted with clang-15, the diff is much smaller.

$ clang-format --version
clang-format version 17.0.0 ([email protected]:smaslov-intel/llvm.git 7719738e32500bff5a61193bfcff9c978ca52e88)

[igchor]

Ok, I think we didn't actually check for clang-version when doing 'generate': #408

[kbenzie]

Is this information not already available via the UR_PLATFORM_INFO_NAME?

I'm wary about the extensibility of an enum.

[smaslov-intel]

Is this information not already available via the UR_PLATFORM_INFO_NAME?

I'm wary about the extensibility of an enum.

The UR_PLATFORM_INFO_NAME returns a string, so we'd need a spec how platforms are named so that runtimes can map it back to what they can work with (enums). Rather than having that extra spec and map in each user, I think it is preferable to just add enum here.

[kbenzie]

The UR_PLATFORM_INFO_NAME returns a string, so we'd need a spec how platforms are named so that runtimes can map it back to what they can work with (enums). Rather than having that extra spec and map in each user, I think it is preferable to just add enum here.

Okay, while duplicating information isn't ideal I can accept it. Where do you expect this to be used, is this passed through the SYCL RT up to the application?

[smaslov-intel]

The UR_PLATFORM_INFO_NAME returns a string, so we'd need a spec how platforms are named so that runtimes can map it back to what they can work with (enums). Rather than having that extra spec and map in each user, I think it is preferable to just add enum here.

Okay, while duplicating information isn't ideal I can accept it. Where do you expect this to be used, is this passed through the SYCL RT up to the application?

The immediate use is for SYCL users to select the underlying backend (via ONEAPI_DEVICE_SELECTOR). Today this selection is via choosing a PI plugin hardcoded in SYCL RT, and with our move to UR we need this new functionality since no longer have visibility into UR adapters.

[kbenzie]

The immediate use is for SYCL users to select the underlying backend (via ONEAPI_DEVICE_SELECTOR). Today this selection is via choosing a PI plugin hardcoded in SYCL RT, and with our move to UR we need this new functionality since no longer have visibility into UR adapters.

Ah I see, so its related to #220. When we discussed this in the WG we came to the conclusion this logic would get pulled into UR once PI has been removed. Perhaps when that happens this query would no longer be necessary.

[smaslov-intel]

The immediate use is for SYCL users to select the underlying backend (via ONEAPI_DEVICE_SELECTOR). Today this selection is via choosing a PI plugin hardcoded in SYCL RT, and with our move to UR we need this new functionality since no longer have visibility into UR adapters.

Ah I see, so its related to #220. When we discussed this in the WG we came to the conclusion this logic would get pulled into UR once PI has been removed. Perhaps when that happens this query would no longer be necessary.

Unless all the device filtering is moved into UR we will need this interface to stay. Note, SYCL language has intrinsic notion of "backend", so I think this will have to stay forever. E.g. see https://github.com/intel/llvm/blob/sycl/sycl/doc/extensions/supported/sycl_ext_oneapi_filter_selector.asciidoc

[smaslov-intel]

@kbenzie : can this be merged, please?

[veselypeta]

LGTM

[kbenzie]

@kbenzie : can this be merged, please?

Sure. We aim for two approvals so I was waiting for another reviewer.