RV670 AGP running git ddx, xserver, libdrm and drt + nopat/AGP ttm caching fix patch. Problem is not new.
I think there is a caching problem with KMS that does not happen with UMS which makes video that doesn't use xv like flash or mplayer -vo x11 eat too much Cpu.
A partial work around can be achieved by disabling accel UTS, but it's still not as good as UMS.
Some top figures - this is with mplayer -vo x11 704x576@25fps Cpu is single core 2.1GHz.
Cpu(s): 18.7%us, 2.3%sy, 0.0%ni, 79.0%id, 0.0%wa, 0.0%hi, 0.0%si, 0.0%st
2504 andy 40 0 39008 13m 7748 S 13.3 1.5 0:01.10 mplayer
2472 root 40 0 591m 8508 5872 S 7.3 0.9 0:02.51 Xorg
KMS + ExaNoUploadToScreen
Cpu(s): 36.3%us, 0.3%sy, 0.0%ni, 63.3%id, 0.0%wa, 0.0%hi, 0.0%si, 0.0%st
4955 root 40 0 21140 9052 5760 S 22.9 1.0 0:04.76 Xorg
5006 andy 40 0 37980 13m 7836 S 13.6 1.5 0:00.71 mplayer
Cpu(s): 17.9%us, 60.8%sy, 0.0%ni, 21.3%id, 0.0%wa, 0.0%hi, 0.0%si, 0.0%st
4897 root 40 0 19628 8684 6040 R 65.8 1.0 0:29.11 Xorg
4925 andy 40 0 37980 13m 7836 S 13.0 1.5 0:00.91 mplayer
Will try and get a useful sysprof current effort just says in kernel for the last case.
Created attachment 33664 [details] [review]
pool allocator to improve AGP performance.
Does this kernel patch help your performance problems? It is made against vanila git sources.
To profile kernel you will need 1.1.5 version of sysprofile and kernel debug symbols.
(In reply to comment #1)
> Created an attachment (id=33664) [details]
> pool allocator to improve AGP performance.
> Does this kernel patch help your performance problems? It is made against
> vanila git sources.
Yes, that does fix it, KMS with UTS enabled is now the same as UMS.
KMS + ExaNoUploadToScreen is unchanged.
> To profile kernel you will need 1.1.5 version of sysprofile and kernel debug
Ahh, I was hoping to get away with the stable as I can't compile the newer one - it doesn't like my old gtk AFAICT.
*** This bug has been marked as a duplicate of bug 26641 ***