KhronosGroup / KhronosGroup/OpenCL-Docs
MAX_DEPTH for device-side enqueue nesting
- Dominant language
- Python
- Stars
- 420
- Forks
- 131
- Avg merge
- 5d 13h
- Merged PRs (30d)
- 11
Description
Device-side-enqueue spec doesn't specify the minimum max depth that should be supported by a conformant implementation, but CTS tests for arbitrary number (currently set to 28 here https://github.com/KhronosGroup/OpenCL-CTS/blob/81cea4775e1915365d5447957699e2231d43d205/test_conformance/device_execution/enqueue_flags.cpp#L30).
This issue tracks the rational behind the choice and the resolution.
A couple of options -
1. Keep it reasonably low for CTS (8 or 16 should be more than enough) - Should be simple, but not clean enough and does not address lack of clarity in the spec.
2. Define the minimum max depth in the spec. Potentially, add a query to get the supported max depth on a given device.
Contributor guide
Research direction
Start by reviewing the device-side-enqueue specification and the OpenCL-CTS test in test_conformance/device_execution/enqueue_flags.cpp. Examine the two proposed approaches and the existing discussion; done means the rationale and resolution are recorded clearly in the specification, including any required minimum or device query.
Written by the indexing model from the issue text.
Assessment
- Domain
- documentation
- Issue type
- Documentation
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100