[SYCL][CI] Update workflow files to match sycl branch

[AlexeySachkov]

This is an attempt to fix pre-commit on sycl-rel-6_1_0 branch by copying contents of CI scripts from sycl branch.

This PR specifically ignores changes to upstream workflows, because we do not have them enabled at intel/llvm.

[AlexeySachkov]

I took a look at Windows issues and it seems like all of them require some test tweaks for us to have a successful pre-commit. I don't want to include all those patches here to avoid overcomplicating the PR, but we should not consider failures on Windows to be blocking the merge.

Gen 12 Windows:

Failed Tests (2): 
 SYCL :: WorkGroupMemory/Dynamic/copy_dynamic_size.cpp 
 SYCL :: WorkGroupMemory/Dynamic/dynamic_unused.cpp 

Could be fixed by https://github.com/intel/llvm/pull/17046 and https://github.com/intel/llvm/pull/17100/files


BMG Windows

hanging test is a known issue: https://github.com/intel/llvm/issues/17255
Adapters/level_zero_eager_init.cpp: https://github.com/intel/llvm/issues/17165
DeviceLib/std_complex_math_fp64_test.cpp: https://github.com/intel/llvm/issues/17165
Graph: https://github.com/intel/llvm/commit/acfbc00124e3fb9119c06c773ba3320bce2a0aab - disabled on Windows BMG

[AlexeySachkov]

Linux failures analysis:

• It seems like something goes wrong during E2E tests build, because they are not being built for CUDA & HIP: binaries are present, but there are no kernels in them
• Bunch of AOT failures due to missing [SYCL][E2E] Clean-up test-modes by removing workarounds #17023

I haven't investigated other failures yet, but I also expect them to be mostly caused by missing patches.

Considering that the only thing this patch changes is the infrastructure, I think it is safe to merge it.
Note that even despite all those failures, it is still an improvement, because previously we couldn't even do builds in pre-commit.

As I said above, test tweaks to fix failures will be cherry-picked separately to reduce scope of the patches.

[AlexeySachkov]

@intel/sycl-matrix-reviewers, @intel/dpcpp-esimd-reviewers, FYI: you were summoned because I changed devops/dependencies-igc-dev.json. Note that I've dropped testing with dev IGC from the release branch, so your approval should be mostly no-op

[MrSidims]

@AlexeySachkov I'm fine with dropping igc-dev from testing. The question though is - what drivers we recommend users to install for open source releases?

[AlexeySachkov]

@AlexeySachkov I'm fine with dropping igc-dev from testing. The question though is - what drivers we recommend users to install for open source releases?

I think that our recommendation is to always go for the latest one, because the interfaces should be stable and we are not allowed to break ABI.

What we have been doing so far (and will continue doing so) is documenting which version we have tested against so that our customers know what we used as a "reference". See example in v6.0.0

[sarnex]

lgtm assuming everything is a legit copy-paste from sycl