[SYCLomatic] USM_NONE testing fixes - #243
Conversation
Sorry, something went wrong.
|
It does not seem like using TEMPLATE_help_function_usm.xml actually prevents it from running with USM_NONE defined. Am I misunderstanding how this should be used? Perhaps this xml need to be updated with an appropriate rule to exclude USM_NONE? I'm also surprised that onedpl_test_device_malloc_free and onedpl_test_uninitialized_fill are passing with USM_NONE defined, as they are using USM malloc together with dpct::device_pointer. |
Sorry, something went wrong.
tomflinda
left a comment
There was a problem hiding this comment.
The changes are OK to me, pls make sure CI test pass and all conflicts resolved before merging.
Sorry, something went wrong.
Thanks. I've rebased and resolved the conflicts. Some tests are failing, but it seems to be a configuration issue of the CUDA environment variables on Windows. Is this a known issue? |
Sorry, something went wrong.
tomflinda
left a comment
There was a problem hiding this comment.
Pls rebase your PR to latest code repo to trigger the CI test again.
Sorry, something went wrong.
Signed-off-by: Dan Hoeflinger <dan.hoeflinger@intel.com>
Signed-off-by: Dan Hoeflinger <dan.hoeflinger@intel.com>
Signed-off-by: Dan Hoeflinger <dan.hoeflinger@intel.com>
Signed-off-by: Dan Hoeflinger <dan.hoeflinger@intel.com>
Signed-off-by: Dan Hoeflinger <dan.hoeflinger@intel.com>
Signed-off-by: Dan Hoeflinger <dan.hoeflinger@intel.com>
A number of onedpl_test_* tests assume the presence of USM.
This PR attempts to address the usage of dpct::device_vector and dpct::device_pointer and assumptions that USM is available in a few ways:
In onedpl_test_copy_if, onedpl_test_fill.cpp, onedpl_test_sort_by_key.cpp, onedpl_test_for_each.cpp and onedpl_test_transform_reduce.cpp : we change the usage pattern of some tests to use dpct::device_vector in such a way to fit both USM and USM_NONE options
In onedpl_test_exclusive_scan.cpp, onedpl_test_fill.cpp, and onedpl_test_sort_by_key.cpp: we add checks to avoid some tests when DPCT_USM_LEVEL_NONE is defined.
This PR also adds 2 tests which were missing from help_function/help_function.xml: onedpl_test_device_malloc_free, and onedpl_test_transform, but only to run for the option where USM is present.
Edit: removing onedpl_test_transform as it is failing in all cases it seems. It seems this should be dealt with separately.
This PR also changes onedpl_test_uninitialized_fill to only run with USM available, as all of its tests assume the USM is available.
Closing #240 in favor of this PR which has a larger scope.