[FFmpeg-devel] I dont know asm too well =( but I have a tip: mem+sync and machine learning

[Jon Maser <jonmaser8@gmail.com>]

i got this idea from postresql, a very small sql database server

i see ffmpeg is broken down into optimized asm libraries for embedding
in c etc programs, and i know a few things about asm, ie, hardware
registers on cpu+, pump in data into a register, get data out, think
it goes down the system bus

but i was thinking someone could implement an upload to memory feature
when uploading/decoding (kinda like a buffer, maybe a queue? add
matrix field of vid, range of tiles, algorithm.. pop when done,
algorithm etc ,  ) video and audio using asm, and implement a sync
mechanism (basically a queue of registers from the matrice with data
you need to process for stuff like deinterlace or fast moving video )

dont know how av is synced, maybe you can implement a buffer and a
stream pointer system?

you can do (i surmise)

scale_cuda_register_with_image img, offset

or

upload to memory
scale_cuda register_with_image image, offset
upscale_cuda_register_with_image image,  offset
sync

you can also use machine learning, which deals with matrices and i
think hardware estimation of certain numbers

it can also work well in writing/reading files, but now that i think
of it c/dma allows you to upload data to the memory and access it
until its freed, but you can do stuff like stream data into pointers,
maybe that helps, would like

it wont give a major preformance gain, but im hoping for the best


hopefully that works, ffmpeg is good software! cant wait to hit usenet B)
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".