[EXP][CUDA] Enable Large cluster sizes and fix cluster dimensions being set for dimensions less than 3

[AD2605]

Fix the ordering of cluster dimension in accordance to the grid Dimensions.

Also adds a call to CU_FUNC_ATTRIBUTE_NON_PORTABLE_CLUSTER_SIZE_ALLOWED to allow
cluster sizes greater than 8

[AD2605]

Could use some help Here -

I was not able to figure out where this flipping of order happens,
I see it's being set in setKernelParams but how it flips it I was not able to understand it

[AD2605]

@oneapi-src/unified-runtime-cuda-write gentle ping r.e. this PR.
E2E Cuda fails because It was unable to find a cuda GPU, seems unrelated to my changes -

lit.py: /home/test-user/actions-runners/01/_work/unified-runtime/unified-runtime/sycl-repo/sycl/test-e2e/lit.cfg.py:718: error: Cannot detect device aspect for cuda:gpu
stdout:

Platforms: 0
default_selector()      : No device of requested type available. -1 (PI_ERRO...
accelerator_selector()  : No device of requested type available. -1 (PI_ERRO...
cpu_selector()          : No device of requested type available. -1 (PI_ERRO...
gpu_selector()          : No device of requested type available. -1 (PI_ERRO...
custom_selector(gpu)    : No device of requested type available. -1 (PI_ERRO...
custom_selector(cpu)    : No device of requested type available. -1 (PI_ERRO...
custom_selector(acc)    : No device of requested type available. -1 (PI_ERRO...

Maybe re-triggering the job might help ?

[JackAKirk]

yeh CI seems to be broken currently, same here: https://github.com/oneapi-src/unified-runtime/actions/runs/9584385509/job/26430113727?pr=1774

[MartinWehking]

Hi, thanks for your patch.
Are there any test cases that fail without your fix and if not, could you provide one?

[AD2605]

Hi, thanks for your patch. Are there any test cases that fail without your fix and if not, could you provide one?

So the test cases that are being added, are in this PR here - intel/llvm#14113, which would test this PR fully. I do not suppose I can add a test which will check the ordering of the cluster dimensions in this UR PR, however I can change increase the cluster size in the tests added in #1643 and increase the cluster size, such that it tests CU_FUNC_ATTRIBUTE_NON_PORTABLE_CLUSTER_SIZE_ALLOWED flag being added

However, note that this runs on SM_90 only, which I do not suppose is on the CI. If you prefer, I can add like a log of the test run on H100 ?

[callumfare]

I've removed the ready-to-merge label since intel/llvm#14113 isn't passing CI and also has the abi-break label - it will need to wait for the ABI breaking window to open before it can be merged.

[MartinWehking]

Hi, thanks for your patch. Are there any test cases that fail without your fix and if not, could you provide one?

So the test cases that are being added, are in this PR here - intel/llvm#14113, which would test this PR fully. I do not suppose I can add a test which will check the ordering of the cluster dimensions in this UR PR, however I can change increase the cluster size in the tests added in #1643 and increase the cluster size, such that it tests CU_FUNC_ATTRIBUTE_NON_PORTABLE_CLUSTER_SIZE_ALLOWED flag being added

However, note that this runs on SM_90 only, which I do not suppose is on the CI. If you prefer, I can add like a log of the test run on H100 ?

Thanks, that would be great!

[callumfare]

This was included in #1804 which has now merged