[SYCL] Temporarily disable hier_par test

[andykaylor]

The CodeGenSYCL/hier_par.cpp test is exhibiting inconsistent behavior.
This patch marks that test as UNSUPPORTED. When hierarchical parallelism
is stabilized the test should be re-enabled.

Signed-off-by: Andy Kaylor andrew.kaylor@intel.com

[bader]

FYI: @againull is also making changes to this file here #1205.
I think we should apply both PRs. @againull, do you agree?

[andykaylor]

FYI: @againull is also making changes to this file here #1205.
I think we should apply both PRs. @againull, do you agree?

Can my change just be included in #1205?

[againull]

In #1205 I was trying to fix the test and your PR is disabling it, so looks like merging them doesn't make sense. I want to update #1205 by changing the test to feed IR to LowerWGScope pass and check output IR (it looks like opt tool is available during check-clang). This should solve the problem with non-stable IR coming from frontend. Let me upload this change today so there will be non need to disable this test.

P.S. Also I am considering to move this pass to llvm because clang probably is the wrong place for it.

[andykaylor]

If #1205 fixes the test then my change can be discarded. I just wanted to avoid the unexpected pass that sometimes happens.

[bader]

In #1205 I was trying to fix the test and your PR is disabling it, so looks like merging them doesn't make sense. I want to update #1205 by changing the test to feed IR to LowerWGScope pass and check output IR (it looks like opt tool is available during check-clang).

+1 for running isolated pass instead of clang.

This should solve the problem with non-stable IR coming from frontend. Let me upload this change today so there will be non need to disable this test.

P.S. Also I am considering to move this pass to llvm because clang probably is the wrong place for it.

+1. LLVM project is the right place for LLVM passes.

Agree with Andy, if #1205 fixes the test, there is not need to disable, but I still see unfixed bugs in the pass, so I'm not sure if the test is going to be fixed by #1205.

[andykaylor]

In #1205 I was trying to fix the test and your PR is disabling it, so looks like merging them doesn't make sense. I want to update #1205 by changing the test to feed IR to LowerWGScope pass and check output IR (it looks like opt tool is available during check-clang). This should solve the problem with non-stable IR coming from frontend. Let me upload this change today so there will be non need to disable this test.

P.S. Also I am considering to move this pass to llvm because clang probably is the wrong place for it.

This should absolutely be moved. There shouldn't be any passes in the front end.

[bader]

@againull, feel free to close this PR, if you think #1205 will address hier_par test failures.

[againull]

I prepared PR where I moved this pass to llvm project and replaced hier_par test:
#1279

Closing this PR.