@im_yeyito "mkver" is removing our old SIMD paths and upgrading them to newer ones

@digital_n01se Not an important enough codec to do it

@FFmpeg That hand tuned patch from MMX to SSSE3 by FFmpeg demonstrates how targeted assembly keeps performance ahead in real world video workloads

@FFmpeg BuT wHy CaNt yoU WriTe iT iN RuST?

@FFmpeg Me anytime I read "Hand written assembly". You people are wizards

@FFmpeg 14 times the speed of light? 😲
It must've come from the future.

@FFmpeg At this rate FFMPEG will be rewritten in assembly

@FFmpeg Isn't SSSE3 20 years old?

@FFmpeg nice, this is what I want to see...

btw... did you guys stopped accepting patch from mythos?

@FFmpeg still wild that handwritten asm can blow c out of the water like that.

@FFmpeg MMX chips were peak society. Windows 95 version B, mIRC, and ftp client and you were in business

@FFmpeg Really wonder if giving AI the cpu physical layout it might come up with some insane assembly taking advantage of physics and some timing wizardry that’d be incomprehensible yet usable

@FFmpeg had any luck with the recently hyped-up tools on generating good assembly?

@FFmpeg MMX...

@FFmpeg The Rustaceans will be seething about this because it's not memory safe.

@FFmpeg Hand-written SIMD still has its place. We hit similar trade-offs in GCP video encoding pipelines — sometimes the compiler just can't see the optimization path. The maintainability vs performance balance is real, but for hot paths the payoff is worth it.

@FFmpeg The legends at work

@FFmpeg Thanks

@FFmpeg legend

@FFmpeg What's the threshold for keeping multiple implementations around? Surely a function that is only 2 % faster than the C version would be considered for pruning.

@FFmpeg Getting ready to apt update!

@FFmpeg legendary status for assembly coders.

@FFmpeg wow thats so cool, the C compiler produced slower code even with flags to use those wide instructions?

@FFmpeg how fast is it if compare with mmx assembly?

why compare it with c if it was already written in assembly?

@FFmpeg Mmmmm supplementary streaming single instruction multiple data extensions....

@FFmpeg You should have used certain other language, It would bem safer and 100x master🤥

@FFmpegitten assembly can give massive gains. Many years ago, I worked at Matrox and did the conversion from 32-bit to 64-bit of their mpeg-2 video codec. 100% pure assembly, unrolled loop, no function, it was 10k+ thousands line of code, the 64-bit gave us 15-20% performance boost, just for the extra 8 cpu register we could use instead of the stack! This codec was blasting fast, it was 4 to 8X faster then competition, we could encode 4 1080p video in real time on a dual opteron system. (at that time, Adobe Premiere could not even encode a single 480p video in real time ....)

@FFmpeg 😳

@FFmpeg @grok que es esto explicado en wspañol

@FFmpeg Where's the rust dev? Don't you have passions to rewrite this in rust?

@FFmpeg Is anyone at Ffmpeg doing Autoresearch style experiments to wring out extra speed from the codebase?

@FFmpeg This is going to be fun.

@FFmpeg Cool