md5crypt-opencl often autotuning poorly (on nvidia 2080ti)
Open
Nobody has claimed this yet.
bug
enhancement
- Dominant language
- C
- Stars
- 13.6k
- Forks
- 2.6k
- PR merge metrics
- No merged PRs in 30d
Description
As mentioned in #5006, I noticed md5crypt-opencl would often autotune to too low values resulting in significantly lower speed.
In 10 runs on a 2080ti:
| ws | speed (many salts) | num times |
|---|---|---|
| 256, 34816 | 12916K c/s | 6 |
| 256, 4456448 | 23014K c/s | 2 |
| 256, 2228224 | 22825K c/s | 1 |
| 256, 139264 | 21547K c/s | 1 |
I haven't seen this behavior with any other format but it would sure be nice to avoid that lowest setting.
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start by locating the md5crypt-opencl format's autotuning and work-size selection code, then reproduce the reported runs on an NVIDIA 2080 Ti. Compare the selected settings and speeds across repeated runs; done means the tuner no longer frequently chooses the lowest-performing setting without reducing performance elsewhere.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- c
- Domain
- performance
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 45/100