Michael, I tried to implement page flipping in s3c2410fb. No luck. Performance is worse than in case without flipping. I suppose SurfaceFlinger uses memcpy (or similar) for double buffering from/to "offscreen" surface(s). If offscreen surface resides in Fb memory it's non-cached, therefore performance suffer. (I may be wrong. But it's most credible explanation I found)
On 4 дек, 15:30, Michael Trimarchi <trimar...@gandalf.sssup.it> wrote: > Hi, > > Alexey Roslyakov wrote: > > Thanks, Michael, > > great work. > > > Now bionic code (libc/arch-arm/*.s) is armv4-friendly. You can safely > > use original bionic/libc/Android.mk - optimized versions of memcpy, > > strlen and memcmp. > > Android will run faster a little bit;) > > I have redone a rebase and use now the memcpu of bionic but with same > result in > performance. I will continue to investigate but I think that it can be > related > to the new graphics abstraction. I have compiled alsa, but it is far away > to work. Have you done some work on it? > > Michael > > > On 27 , 03:25, Michael Trimarchi <trimar...@gandalf.sssup.it> > > wrote: > > >> Hi, > > >> this is my patch to boot android. The PAN display ioctl is only for > >> openmoko, hope that can help > >> some user. > > >> The overall performance are not good but the keyboard is three time faster > >> from cupcake version :) > > >> Michael > > >> eclair.patch > >> 50K -- unsubscribe: android-porting+unsubscr...@googlegroups.com website: http://groups.google.com/group/android-porting
