Closed jhorstmann closed 1 year ago
The last run of the actions can also be seen in my fork: https://github.com/jhorstmann/packed_simd/actions Would be nice if these could be somehow compared with the last successful run in travis. Especially the wrong results in some of the mips or powerpc tests are a bit worrying.
Oof, this has some errors on master, judging by https://github.com/rust-lang/packed_simd/actions/runs/5673019417
https://github.com/rust-lang/packed_simd/pull/355 gets us half of the way there.
This sets up github actions workflows that should run the same tests as were previously setup for travis ci. Luckily none of the shell scripts needed changing, only a new set of yaml files.
There are a few FIXME issues, where I could use some help or which could maybe be solved in followup:
dox.sh
that I haven't looked into yetispc
feature for the benchmarks is currently disabled, because I haven't found an easy way to add the binary to thePATH
There is a mostly unrelated change in the instruction verification tests for vector rotates, that turned out to be a red herring. The verify tests for i586 and i686 were failing because the verifier did not expect a call instruction in the code. That call instruction however was only needed to support relocations, the code was getting inlined correctly. However, I think the target features are more correct after the change, the 128bit type m64x2 should require avx512vl, the other two tests use 512bit vectors where avx512f should suffice.