[SYCL][Graph] Modified the adapters such that it is valid to call release on... #18619
Conversation
konradkusiak97
had a problem deploying
to
WindowsCILock
GitHub Actions
Error
…le it is executing
konradkusiak97
force-pushed
the
konrad/AddSyncToCBRelease
branch
from
cce454f to
8e70e5c
Compare
konradkusiak97
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
konradkusiak97
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
konradkusiak97
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
| @@ -1175,6 +1178,7 @@ UR_APIEXPORT ur_result_t UR_APICALL urEnqueueCommandBufferExp( | |||
| if (phEvent) { | |||
| UR_CHECK_ERROR(RetImplEvent->record()); | |||
| *phEvent = RetImplEvent.release(); | |||
| hCommandBuffer->CurrentExecution = *phEvent; | |||
There was a problem hiding this comment.
Will this solution work if the UR caller doesn't pass a phEvent to urEnqueueCommandBufferExp, and then does a command-buffer release? Note, this is what your CTS test does.
| @@ -310,7 +310,7 @@ returns: | |||
| - $X_RESULT_ERROR_OUT_OF_HOST_MEMORY | |||
| --- #-------------------------------------------------------------------------- | |||
| type: function | |||
| desc: "Decrement the command-buffer object's reference count and delete the command-buffer object if the reference count becomes zero." | |||
| desc: "Decrement the command-buffer object's reference count and delete the command-buffer object if the reference count becomes zero. It will try synchronizing the command-buffer, hence it is legal to call it while command-buffer is still executing." | |||
There was a problem hiding this comment.
"It will try synchronizing the command-buffer," makes it sound like this will happen even if the reference count isn't zero. So would flip round the two phrases, something like
It is legal to call the entry-point while
hCommandBufferis still executing, which will block on completion if the reference count ofhCommandBufferbecomes zero.
There was a problem hiding this comment.
That sounds better, thanks 👍
There was a problem hiding this comment.
I think we need to call the generate build target to regenerate the doyxgen in the headers.
Command Buffer, while it is still executing.