jojers/Zorya
1
0
Fork 0
Code Issues Pull requests Projects Releases 1 Packages Wiki Activity Actions
1 release 1 tag
  • v1.0.0 8bf46b02f9

    v1.0.0 Stable

    jojers released this 2026-05-03 18:05:34 +00:00 | 9 commits to main since this release

    This release marks the first proper Zorya release where I'm confident and satisfied with the state of the project!

    Currently supported features:
    - Backend hotswap
    - Scalar backend
    - AVX2 backend
    - AVX-512 backend (requires avx512f, avx512bw, avx512dq, avx512cd, avx512vl)
    - Runtime detection of CPU features for safe backend hotswap

    Known issues:
    - Performance is inconsistent
    - Code path for tail handling with masked instructions for AVX-512 backend is slower than non masked
    - Backend unit tests are failing for AVX2 and AVX-512 due to floating point approximation and hardware rsqrt instructions being approximations with only 14 bits of precision and of course the 28 bits versions are locked to data center HPC CPUs thank you Intel and AMD

    Planned features:
    - Precision hotswap
    - Particle merging and tombstoning
    - HIP GPGPU backend!
    - Maybe some colors so it's not as boring