[FFmpeg-devel] [PATCH v7] libavcodec/riscv: add RVV optimized idct_32x32_8 for HEVC

Rémi Denis-Courmont remi at remlab.net
Sat Jul 5 14:30:00 EEST 2025


Hi,

I'm aware that this patch is being held an inappropriate amount of time 
waiting for review, but as I said before, I have NO experience dealing with 
such a large transform, and I am kinda busy.

Note that I don't really see the point in specialising the code by VLEN. 
Properly designed hardware should handle VLEN-agnostic code just fine. The 
spirit of the RVV spec is to scale execution time by VL, not by VLEN. This is 
also how SiFive chips work, and I don't think we should be filling FFmpeg code 
with specialised code for subpar hardware designs. (RISC-V being open, there 
will always be crappy, insane and/or weird implementations; we can't handle 
them all.)

-- 
ヅニ-クーモン・レミ
Villeneuve de Tapiola, ex-République finlandaise d´Uusimaa





More information about the ffmpeg-devel mailing list