Hi, List,

We are using libpixelflinger on gingerbread. we don't have GPU right now and
do not expect  to have one in the future.  we don't have a very mature
OpenGL implementation yet so we end up using libagl. hopefully,  we will
refine libpixelflinger as much as possible to deliver *modest* graphics
performance. I understand that many SoCs have GPU and they don't have such
issue. feel free to give us advices if you are on happy land.  for google's
guys, could you tell me whether pixelflinger exists in the upcoming release?
what's the alternative if not.

first of all, I am not graphics guy but have compiler knowledge. Quote
Mentorgraphics slice here, "Pixelflinger is a graphics output sink, renders
a final frame buffer image". "Codeflinger JIT generates simple assembly
routines for frequently used graphic operations"

IPP(instruction per pixel) is very important indicator of code quality from
CodeFlinger. IPP represents the number of instructions of the only loop in
an assembly snippet. I am trying to reduce IPP by peekhole and loop
optimizations. but I am frustrated with the IR. it's too low-level to
perform any meaningful analysis.  Should we introduce a simple IR to conduct
"loop invariance motion" and "strength reduction"?

The performance of Pixelflinger is critical. if you take default
libpixelflinger, the responsiveness of android UI is crazy.  View redrawing
usually consume several seconds.  With trivial codeflinger JIT, the
responsiveness has been significantly improved. we measured about 10 times
speedup in simple opengl testcases.  Therefore, further tuning
libpixelflinger is promising.

i don't know what guys from other architectures(i mean non-ARM)  are doing.
Mips has a simple patch to convert ARM to MIPS instruction. but they don't
have any optimization.  does it matter on your platform?

thanks,
--lx

-- 
unsubscribe: [email protected]
website: http://groups.google.com/group/android-porting

Reply via email to