KhronosGroup / KhronosGroup/OpenCL-Docs
Can we relax requirements on size of the masks used with the `shuffle` and `shuffle2` built-in functions?
- Dominant language
- Python
- Stars
- 420
- Forks
- 131
- Avg merge
- 5d 13h
- Merged PRs (30d)
- 11
Description
The spec currently states:
> The size of each element in the mask must match the size of each element in the result.
This seems rather overkill and doesn't seem helpful for either implementations or applications. For example it means than shuffling a vector of floats requires 32-bit masks when 8-bit masks would be more than sufficient for all the vector sizes currently defined by OpenCL C.
Could we relax this requirement in a future version of OpenCL C?
Contributor guide
Research direction
Review the OpenCL C specification text for the shuffle and shuffle2 built-in functions, focusing on the requirement that mask element size match result element size. Compare that requirement with the currently defined vector sizes and evaluate the proposed relaxation. Done means the future-version specification decision and resulting wording are resolved.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- c
- Domain
- documentation
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 25/100