[SYCL][CUDA] Change hint functions to return UR_SUCCESS #11296
Conversation
fabiomestre
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
fabiomestre
force-pushed
the
fabio/cuda_update_hint_ep
branch
from
c6f3316 to
da555b6
Compare
- The UR spec was recently changed to make hint entryponts always return UR_RESULT_SUCCESS. This commit changes the CUDA adapter to be conformant with this change. - This commit also changes the type of PointerRangeSize which was causing a stack corruption.
fabiomestre
force-pushed
the
fabio/cuda_update_hint_ep
branch
from
da555b6 to
02dd7ba
Compare
fabiomestre
changed the title
fabiomestre
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
fabiomestre
had a problem deploying
to
WindowsCILock
GitHub Actions
Failure
|
I have a question. The deleted message in "setErrorMessage()" may be informative for users. If they are deleted, will a user still know about the cause of lack of support ? |
That's a good point. The aim of this PR is to make all the adapters have consistent behaviours. Since performance hints don't have an impact in the behaviour of a program, it's usually an accepted pattern that such functions should return success even when the performance hint is ignored. The current behaviour of the CUDA adapter returning an error in this situations is an exception to this. Maybe, in the future, this could be turned into warnings if support for them is added to the UR specification (oneapi-src/unified-runtime#762). |
always return UR_RESULT_SUCCESS. This commit changes the CUDA
adapter to be conformant with this change.
was causing a stack corruption.