[tremor] [PATCH] 12% global performance gain on a StrongARM

[Nicolas Pitre]

On Sun, 22 Sep 2002, marc dukette wrote:

> yep, I just tried it and performance is right back to where it was.

Hmmm... interesting.

> Unfortunately that means the CE based devices can't currently share in the
> performance improvements you have implimented.  It would be interesteing to
> do a comparison between the two compiled versions and see which is doing a
> better job compiling. 

If you can have your compiler generate an assembly version of mdct.c with
and without the macro trick I'd be interested to see them.

> Do you have a performance test tool and stream you
> use?  I'd like to run the same test and see what the differences are. 

Well, on Unix it's easy with the "time" command.  Simply using

        time ivorbisfile_example < test.ogg > /dev/null

gives you the actual amount of time the CPU spent on executing the given
task.  Then I compare decoding of the same stream across different versions.

I can't help you with CE though.

> Once note is that your original port still far outperforms Tremor on an SA
> compiled with eVC.

Well with the improvements I made to Tremor it's now faster than my original 
fixed point port by about 7% when both are compiled with gcc.

Inspecting the asm output of eVC should give a clue to what could be done to 
improve things there as well.

<p>Nicolas

--- >8 ----
List archives:  http://www.xiph.org/archives/
Ogg project homepage: http://www.xiph.org/ogg/
To unsubscribe from this list, send a message to 'tremor-request at xiph.org'
containing only the word 'unsubscribe' in the body.  No subject is needed.
Unsubscribe messages sent to the list will be ignored/filtered.