Fix bug in the SIMD-accelerated extend (AVX2)

smarco / WFA2-lib

WFA-lib: Wavefront alignment algorithm library v2

Other

162 stars 36 forks source link

Fix bug in the SIMD-accelerated extend (AVX2) #49

Closed quim0 closed 1 year ago

quim0 commented 1 year ago

Fix bug in the AVX2-accelerated version of the extend function. In some corner cases, the function exited before all the diagonals where processed.
Update the AVX2 branch to the most recent commit (with biWFA)

pjotrp commented 1 year ago

Can you do a PR against master? It is strange to see all commits that are already in the tree.

smarco commented 1 year ago

I would not merge this code against master for the time being. This code is very experimental (and this branch has to die).

Said that @pjotrp is right. I believe that the proper thing to do here is (@pjotrp, correct me if I'm wrong): (1) New branch from the current master and push the AVX versions of the kernels. (2) When this is mature (we need to test that the build system is happy with machine-dependent code), we will merge against development. (3) Kill this branch as it has become too outdated (my bad).

@quim0 let me know if this makes sense to you (and thanks for the effort).

quim0 commented 1 year ago

As @smarco says, this code is still experimental, and I think it's not mature enough to be in master yet. I can create a new branch, but with this PR the extend_avx2 branch is already even with main (to test that it works with modern features/tests), so I don't know if it makes much sense. The extend_avx2 branch will be removed anyway when the accelerated kernels are merged into development. Let me know what you think about it.