You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Now that portable SIMD is available on nightly (std::simd), it might be interesting to switch to that instead of the homemade fp::simd. It might even make sense to upgrade Limb from u64 to u64x8 or something similar, to see if the autovectorization kicks in and does it better than our implementation.
The text was updated successfully, but these errors were encountered:
I've been looking into this too. It seems like for best performance we still need to manually select the number of lanes, and AVX-512 is not yet supported.
Now that portable SIMD is available on nightly (
std::simd
), it might be interesting to switch to that instead of the homemadefp::simd
. It might even make sense to upgradeLimb
fromu64
tou64x8
or something similar, to see if the autovectorization kicks in and does it better than our implementation.The text was updated successfully, but these errors were encountered: