[FFmpeg-devel] [PATCH v3 1/9] lavc/vp8dsp: R-V put_vp8_pixels

[Rémi Denis-Courmont]

Le maanantaina 6. toukokuuta 2024, 6.38.01 EEST uk7b at foxmail.com a écrit :
> From: sunyuechi <sunyuechi at iscas.ac.cn>
> 
> C908:
> vp8_put_pixels4_c: 78.0
> vp8_put_pixels4_rvi: 33.7
> vp8_put_pixels8_c: 278.0
> vp8_put_pixels8_rvi: 55.0
> vp8_put_pixels16_c: 999.0
> vp8_put_pixels16_rvi: 86.7

Actually, I do think that the C compiler is doing a better job.
On SiFive U74, the figures are horrible:

By time:
vp8_put_pixels4_c:     0.2
vp8_put_pixels4_rvi:  10.5
vp8_put_pixels8_c:     1.0
vp8_put_pixels8_rvi:  30.0
vp8_put_pixels16_c:    2.2
vp8_put_pixels16_rvi: 60.0

By cycles:
vp8_put_pixels4_c:       84.0
vp8_put_pixels4_rvi:   3871.0
vp8_put_pixels8_c:      331.0
vp8_put_pixels8_rvi:  11281.2
vp8_put_pixels16_c:     826.0
vp8_put_pixels16_rvi: 22502.0

Best guess is that the data is not 64-bit aligned, otherwise I can't really 
make sense of those figures.

-- 
Rémi Denis-Courmont
http://www.remlab.net/