arman68
9th September 2005, 13:35
I was benchmarking different compile of mplayer to find out which one was the fastest on a particular slow machine, when I noticed something strange with the video filters: it seemed the order of the video filter had a big impact on speed!. I decided to investigate further, and test various combinations. Here are the results:
FYI clip was encoded with xvid at 640x352 PAL 1500kbps, and I only benchmark on the first 40s (1000 frames) which are quite intensive (start of BBC Amazon Abyss documentary). Pure vide decoding, without any filter, takes 3.8s
pp=fa 1.8s
pp=fa,spp 2.1s
pp=de 2.5s
pp=ac 4.4s
pp=ac,spp 4.6s
pp=ac,fspp 4.6s
denoise3d 9.4s
hqdn3d 13.4s
hqdn3d,fspp 13.6s
pp=ac,hqdn3d 17.8s
pp=ac,spp,hqdn3d 18.0s
pp=ac,fspp,hqdn3d 18.0s
pp=ac,hqdn3d,fspp 18.0s
fspp 29.6s
spp 42.1s
fspp,hqdn3d 43.3s
fspp,hqdn3d,pp=ac 47.3s
What I am beginning to think is the spp and fspp filters must be first in the chain. Any other position and they are bypassed. Filters before and after the spp filters are correctly processed.
Could anybody with a more intimate knowledge of mplayer confirm, and explain why this is the case?
FYI clip was encoded with xvid at 640x352 PAL 1500kbps, and I only benchmark on the first 40s (1000 frames) which are quite intensive (start of BBC Amazon Abyss documentary). Pure vide decoding, without any filter, takes 3.8s
pp=fa 1.8s
pp=fa,spp 2.1s
pp=de 2.5s
pp=ac 4.4s
pp=ac,spp 4.6s
pp=ac,fspp 4.6s
denoise3d 9.4s
hqdn3d 13.4s
hqdn3d,fspp 13.6s
pp=ac,hqdn3d 17.8s
pp=ac,spp,hqdn3d 18.0s
pp=ac,fspp,hqdn3d 18.0s
pp=ac,hqdn3d,fspp 18.0s
fspp 29.6s
spp 42.1s
fspp,hqdn3d 43.3s
fspp,hqdn3d,pp=ac 47.3s
What I am beginning to think is the spp and fspp filters must be first in the chain. Any other position and they are bypassed. Filters before and after the spp filters are correctly processed.
Could anybody with a more intimate knowledge of mplayer confirm, and explain why this is the case?