[x264-devel] PPC profile/benchmarks/computational complexity

[R. Brouard]

Thx.  I'm not asking for complete profiles, I'm just asking for the
total processing time, the computer system or basically the PPC
involved, memory used,  the x264 flags used and the size of the input
file used.    If it isn't a 1 minute easy thing to do I don't want to
burden anyone.  Just looking for some fast numbers.  

Benchmarking is a job in and of itself and why I'm doing a quick and
dirty estimate,  and a very rough theoretical mapping ratio to the
system.  

thx for the link I'll check that out.   

On Tue, 2008-12-09 at 23:05 +0100, Guillaume POIRIER wrote:
> Hello,
> 
> On Tue, Dec 9, 2008 at 10:29 PM, R. Brouard <opensource at musemagic.com> wrote:
> 
> > ;)  I'm looking at some embedded ports where they are using a PPC based
> > core.
> >
> > Thx for the below, I'm also looking for, with all vectorized instr. on,
> > "extreme" profile, target bitrate, input mode.  As you would expect to
> > see in a computational complexity or a "MIPS" document from say ARM or
> > TI. So total CPU time, total size of input test stream, fps, memory
> > usage requirements and then a breakdown of each module
> 
> Benchmarking/analysing such things is a job on its own. This takes
> time, which is resource I lack.
> 
> Since I'm not quite sure I understand your final goal is, you should
> consider that x86 port is already extremely optimized and would
> probably run great on an Atom CPU, and that PPC port is fairly well
> optimized and would run somewhat well on embedded PPC7450.
> 
> Any other general purpose CPU is, IMHO not worth being considered
> because there aren't any SIMD routines written for them and I will
> take time and money to write them for new CPUs.
> 
> 
> > My apologies for the newbie nomenclature vs. what you are using.  I am
> > looking to expand, leverage on your existing optimizations and port and
> > just joined this list.  So, also ramping up.
> 
> Good.
> 
> 
> > But first I have to give benchmark feasibility that x264 will "fit" to
> > move forward, thus your G4 benchmarking is most applicable to map an
> > est. over.
> 
> If you need to access a PPC machine, you could request an account
> here: http://powerdev.osuosl.org/. Note however that the machine they
> offer access to is a PPC970, which is quite the opposite of an
> embedded CPU.
> 
> Guillaume