Pavel Roskin
Pavel Roskin
My progress so far can be seen in the `intel-fix-wip1` branch in my fork: https://github.com/proski/mfakto/commits/intel-fix-wip1/ I limited selftests to one test (M50896831) and set `SieveOnGPU=0`. That makes the failure reproducible...
I see that calling `shl_96` from `check_barrett32_76` introduces some randomness with Intel OpenCL. The result of the shift is wrong and unpredictable. I tried inlining `shl_96`, replacing `amd_bitalign` with shifts...
The fix (WIP) is in https://github.com/primesearch/mfakto/pull/15 There is indeed some issue with vector shifts in the Intel OpenCL implementation. Closing this issue in favor of the PR.
The issue has been addressed in the https://github.com/primesearch/mfakto repository.
mfakto only links to the OpenCL library. In fact, mfakto can be compiled without ROCm.
Some ideas how to do it with git: https://stackoverflow.com/questions/10415100/exclude-file-from-git-diff
I do edit files occasionally. These are the ways I use `stg edit -d`: * Writing or improving commit description while looking at the commit * Removing chunks that don't...
You can also replace the bitmap with a one containing the disclaimer text.
Rebased on top on the current main branch. No changes were needed, as my PR did not include changes to the files removed in 6d7ff05af3896e3a661c2e3f170382ec144f4d7c
Here's the updated version. Many typos were fixed upstream already, so it's a shorter PR. However, I added one more fix - `Producta` should be `Product`.