[FFmpeg-devel] [PATCH] lavc/h264chroma: RISC-V V add motion compensation for 8x8 chroma blocks

[Rémi Denis-Courmont]

Le perjantaina 19. toukokuuta 2023, 21.52.57 EEST Lynne a écrit :
> May 19, 2023, 19:16 by remi at remlab.net:
> > Le keskiviikkona 17. toukokuuta 2023, 17.54.22 EEST Lynne a écrit :
> >> Finally, run:
> >> make checkasm && ./tests/checkasm/checkasm --bench
> >> and report on the timings for both the C and assembly versions.
> >> If you've made a mistake somewhere, (forgot to restore stack, or a
> >> callee-saved register, or your function produces an incorrect result),
> >> checkasm will fail.
> > 
> > To be fair, in this particular case, the stack pointer and saved registers
> > are never used, so the risk of messing those are zero.
> > 
> > checkasm would of course verify that the function does that it is supposed
> > to do, and personally, I have kept off untested functions. But I am not
> > sure if it is fair to require adding test cases whilst other
> > architectures weren't required to have them.
> 
> Other hardware exists, and even without checkasm, bugs are found
> through fate.

There are exactly two FATE RISC-V instances of which exactly zero support 
vectors to date, so the chance for FATE to find bugs there is nul. I do 
appreciate that checkasm would make revectoring of the patchset easier and 
safer though.

That said, the argument is moot seen as you seem to have posted a fix.

-- 
Реми Дёни-Курмон
http://www.remlab.net/