error code consistency for clEnqueueNDRangeKernel #1501
Conversation
Also, for clGetKernelSuggestedLocalWorkSizeKHR, while we're at it.
| * {CL_INVALID_OPERATION} | ||
| ** if SVM pointers are set as arguments for _kernel_ and the device associated with _command_queue_ does not support SVM | ||
| ** if system pointers are set as arguments for _kernel_ and the device associated with _command_queue_ does not support fine-grain system SVM | ||
| // TODO: Do we still need these explicit examples? |
There was a problem hiding this comment.
In my opinion, this can be removed as it is already specified in the footnote 7: https://registry.khronos.org/OpenCL/specs/3.0-unified/html/OpenCL_API.html#_footnotedef_7
There was a problem hiding this comment.
Thanks for the reminder!
I agree footnote 7 is a good example where these out-of-resources cases are described elsewhere in the spec. So, I'm inclined to keep the same boilerplate text here for CL_OUT_OF_RESOURCES for clEnqueueNDRangeKernel vs. including the explicit examples.
If we really want to keep text like the following, which I don't believe is documented anywhere else in the spec...
For example, the explicitly specified local_work_size causes a failure to execute the kernel because of insufficient resources such as registers or local memory.
... perhaps it should move to a footnote, also? Or, we can just remove it completely, as is done in this PR.
|
Merging as discussed in the January 6th teleconference. |
2 participants
This is another PR to partially address the way we document error conditions, see #1320 and in particular #1320 (comment).
It contains a subset of the changes from #1399, specifically for clEnqueueNDRangeKernel.
Also, for clGetKernelSuggestedLocalWorkSizeKHR, while we're at it, since they share a lot of the same error conditions.