=?gb2312?B?UmU6IFJlOiBbWHZpRC1kZXZlbF0gb3B0aW1pemUgeHZpZCBvbiBhcm0=?= =?gb2312?B?OQ==?=

[Christoph Lampert]

On Tue, 17 Aug 2004, daniel wrote:

> Oh,it is encoding.I found that encoding process cost vast data
> comparison ,not like mp3 encoding ,complex computation concentrated.So
> it same hard to optimized.

You might want to profile the encoder to find bottlenecks specifically for
your plattform then. On x86/PCs, since the SAD routines are called most
often, the difference between pure C and SIMD ASM is large there. General
memory access is a limiting factor as well, though not as much as might be
the worst case, because modern CPUs have large caches.  But if your ARM
plattform has a slower RAM interface, that might indeed play a crucial
role.

I'm curious, can you give some numbers about your hardware and 
encoding/deocding speed?

gruel