OpenCL autotune: use a subset of test vectors
Nobody has claimed this yet.
- Dominant language
- C
- Stars
- 13.6k
- Forks
- 2.6k
- PR merge metrics
- No merged PRs in 30d
Description
Originally written as comment https://github.com/magnumripper/JohnTheRipper/pull/3627#issuecomment-459991983
It looks like autotune is doing a worse job with the revised md5crypt-opencl kernel, perhaps because our test vectors differ from actual passwords and hashes so much.
It'd be very useful for us to be able to specify a reduced set of test vectors for autotune and benchmarks. For the benchmarks, we have related issue #3625. This one is for autotune.
Also, when running with specific password length(s) - such as with mask mode and/or with min/max lengths specified with command-line options or for an incremental mode - maybe the autotune could limit itself to using test vectors that fall within this lengths range?
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 tracing how autotune selects test vectors and how password-length ranges are represented for mask and incremental modes. Compare the proposed autotune behavior with related benchmark issue #3625. Done means autotune can use an appropriate reduced or length-matched vector subset without changing unrelated benchmark behavior.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- c
- Domain
- performance
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100