4 ms·

Does it have fallback paths for everything, though? Scalar if necessary? Projects that depend on Highway drop support for CPUs not listed in the Highway docum

by fweimer 5mo ago

Does it have fallback paths for everything, though? Scalar if necessary? Projects that depend on Highway drop support for CPUs not listed in the Highway documentation, saying that they can't support these CPUs because they are incompatible with Highway: https://google.github.io/highway/en/master/README.html#current-status https://google.github.io/highway/en/master/README.html#curre... Are these projects somehow mistaken?

janwas 5mo ago
Yes, the EMU128 target is scalar only, with for loops. This is a fun way to see how well autovectorization works, with the same source code. That works on any CPU. Curious which projects have such concerns, any link?
- fweimer 5mo ago
  People reported challenges building V8 (whether upstream or the Node.js variant) on s390x with z13 support. I don't know if it was discussed on the porters mailing list because it's not public: https://groups.google.com/g/v8-s390-ports https://groups.google.com/g/v8-s390-ports Elsewhere, some people interpreted https://github.com/google/highway/issues/1895 https://github.com/google/highway/issues/1895 as meaning that Highway code does not work on z13 at all.
  - janwas 5mo ago
    Thanks for sharing. The first link seems non public indeed. I can imagine there is some compile issue we could reasonably fix, with the help of someone who has Z13 access. Please encourage them to raise an issue. I will be back on May 26. After that, it should at least be able to use the scalar fallback. The issue with Z14 is that it lacks fp32 support. Would their usage be integer only?
    - janwas 5mo ago
      Correction (typo): Z13 lacks fp32.
    - fweimer 5mo ago
      I'll bring it up with some folks. It probably won't change much because the z13 transition has finished by now. It's still good to know because RISC-V is in the same boat regarding Highway support today: we need scalar fallback in Highway until we get RVA23 hardware deployed.

by fweimer 5mo ago

janwas 5mo ago

Yes, the EMU128 target is scalar only, with for loops. This is a fun way to see how well autovectorization works, with the same source code. That works on any CPU. Curious which projects have such concerns, any link?

fweimer 5mo ago
People reported challenges building V8 (whether upstream or the Node.js variant) on s390x with z13 support. I don't know if it was discussed on the porters mailing list because it's not public: https://groups.google.com/g/v8-s390-ports https://groups.google.com/g/v8-s390-ports Elsewhere, some people interpreted https://github.com/google/highway/issues/1895 https://github.com/google/highway/issues/1895 as meaning that Highway code does not work on z13 at all.
- janwas 5mo ago
  Thanks for sharing. The first link seems non public indeed. I can imagine there is some compile issue we could reasonably fix, with the help of someone who has Z13 access. Please encourage them to raise an issue. I will be back on May 26. After that, it should at least be able to use the scalar fallback. The issue with Z14 is that it lacks fp32 support. Would their usage be integer only?
  - janwas 5mo ago
    Correction (typo): Z13 lacks fp32.
  - fweimer 5mo ago
    I'll bring it up with some folks. It probably won't change much because the z13 transition has finished by now. It's still good to know because RISC-V is in the same boat regarding Highway support today: we need scalar fallback in Highway until we get RVA23 hardware deployed.