Sorry, the previous logic was indeed a bit confusing. I rewrote it, renaming the third parameter to max_lmul.
If e16 is involved in the function, then vsetvlstatic8 uses max_lmul == m4. If e32 is involved in the function, then vsetvlstatic8 uses max_lmul == m2. If e16 is involved in the function, then vsetvlstatic16 uses max_lmul == m8. If e32 is involved in the function, then vsetvlstatic16 uses max_lmul == m4. I think it is clearer now. Rémi Denis-Courmont <r...@remlab.net> 于2024年7月8日周一 23:41写道： > Le maanantaina 1. heinäkuuta 2024, 19.09.01 EEST flow gg a écrit : > > I reviewed it again, the purpose of is_w is to limit lmul to a maximum of > > 1/4 of vlen, > > 1/4 of vlen? Do you mean limit to EMUL=1 for EEW=32 and EMUL=1/4 for EEW=8? > > Limiting LMUL to less than 1 at maximum EEW is useless from a functional > standpoint, since fractional registers cannot be addressed individually. > (Of > course it might still be useful for performance reasons.) > > > to prevent vector register shortage, which can also be > > considered as vset limiting lmul. I renamed it to quarter_len_limit. > > TBH, I don't really understand. > > If a lower LMUL limit is reached, then specialisations for the > corresponding > VLEN are simply unncessary/infeasible and the code for lower VLEN should > be > used. > > If a higher LMUL limit is reached due to register pressure (or the 8 hard > limit), then the given VLEN cannot be supported at all, or requires some > completely different code. > > Either way, I don't really follow why vsetvlfixed macros need to be > involved. > > -- > 雷米‧德尼-库尔蒙 > http://www.remlab.net/ > > > > _______________________________________________ > ffmpeg-devel mailing list > ffmpeg-devel@ffmpeg.org > https://ffmpeg.org/mailman/listinfo/ffmpeg-devel > > To unsubscribe, visit link above, or email > ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe". > _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".
