Bump submodules 20250410

[marbre]

• Bumps all submodules, expect hipBLASLt, to remote heads:
  hipBLASLt points to a commit at develop whereas the module defaults to mainline.
• Manually keps the Tensile pin.

[marbre]

The following tests fail

• rocTHRUST
  • 17 - copy.hip (Failed)
• rocPRIM
  • 51 - rocprim.lookback_reproducibility (Failed)
  • 75 - rocprim.linking (Failed)

[marbre]

Thanks for adding the filtering @geomin12. In difference to an expected failure, tests wont run at all whit this. An alternative might be to set WILL_FAIL to the test targets at CMake level. Looking into it this unfortunately a little painful due to the way test targets are added. Another option could be to use one of the GoogleTest macros (ASSERT_NO_FATAL_FAILURE, EXPECT_NO_FATAL_FAILURE). Both options would require to patch the projects themselves, which might not be too bad as there is at least hope that those patches can be dropped sooner or later. @stellaraccident any preference?

[geomin12]

Thanks for adding the filtering @geomin12. In difference to an expected failure, tests wont run at all whit this. An alternative might be to set WILL_FAIL to the test targets at CMake level. Looking into it this unfortunately a little painful due to the way test targets are added. Another option could be to use one of the GoogleTest macros (ASSERT_NO_FATAL_FAILURE, EXPECT_NO_FATAL_FAILURE). Both options would require to patch the projects themselves, which might not be too bad as there is at least hope that those patches can be dropped sooner or later. @stellaraccident any preference?

ah i see! would we prefer the "xfail" or explicitly removing them from the test runs? i can adjust to whatever is preferred

[marbre]

Thanks for adding the filtering @geomin12. In difference to an expected failure, tests wont run at all whit this. An alternative might be to set WILL_FAIL to the test targets at CMake level. Looking into it this unfortunately a little painful due to the way test targets are added. Another option could be to use one of the GoogleTest macros (ASSERT_NO_FATAL_FAILURE, EXPECT_NO_FATAL_FAILURE). Both options would require to patch the projects themselves, which might not be too bad as there is at least hope that those patches can be dropped sooner or later. @stellaraccident any preference?

ah i see! would we prefer the "xfail" or explicitly removing them from the test runs? i can adjust to whatever is preferred

As discussed in today's meeting we probably merge as is and fill an issue for now but definitely need to follow up if more tests start to fail.

[marbre]

Filled #400 and #401.