Re: DPX Support

On Thu, 7 Apr 2005, Robin Rowe wrote:
>
>> GraphicsMagick has what I call "read/write" tests where the read/write 
>> process is:
>> 
>>  reference --write--> version 1 --read/write--> version 2 --read
>> 
>> The final image is then compared (mean distance method) with the first 
>> image written ("version 1").  This is intended to verify that writing and 
>> reading the image does not corrupt it.  It is not intended to be a 
>> performance benchmark.
>
> Does reading and writing an image to the same version of DPX change it? Is it 
> verbatim?

The first write step may (possibly) alter the image based on the 
capabilities of the format.  It is then expected that when version 1 
is read back that identical results are obtained by reading version 2. 
The existing DPX support is passing that test.  Once log support is 
added, it may be necessary to allow a small amount of error to be 
induced.  We shall see.

>> Under UltraSPARC/Solaris I am seeing performance problems when writing, so 
>> with a 2K frame this test is taking 2.620 seconds on Intel Xeon but it 
>> takes 6.732 seconds on UltraSPARC.
>
> Note that's over a minute to write one second of a 24fps movie.

Yes, but remember that this test does two writes and two reads.  More 
significantly, it does an expensive floating-point based compare 
operation between two images in memory.

Reading on SPARC:
  dlad_2048X1556.dpx=> NULL 2048x1556+0+0 DirectClass 10-bit 12.2m 0.920u 0:01 (3.3m pixels/s)

Writing on SPARC:
  dlad_2048X1556.mpc=>crap.dpx DPX 2048x1556+0+0 DirectClass 10-bit 11.4m 2.080u 0:03 (1.4m pixels/s)

Reading on Intel Xeon:
  dlad_2048X1556.dpx=> NULL 2048x1556+0+0 DirectClass 10-bit 12.2m 0.289u 0:01 (11.1m pixels/s)

Writing on Intel Xeon:
  dlad_2048X1556.mpc=>crap.dpx DPX 2048x1556+0+0 DirectClass 10-bit 11.4m 0.813u 0:01 (3.8m pixels/s)

So for SPARC a read/write operation takes about three seconds, but it 
takes about a second on Intel Xeon.

Clearly the performance should be improved (and can be improved), but 
from my point of view, at this time, functionality is more important 
than performance.

Bob
======================================
Bob Friesenhahn
[email protected], http://www.simplesystems.org/users/bfriesen/
GraphicsMagick Maintainer,    http://www.GraphicsMagick.org/


-------------------------------------------------------
SF email is sponsored by - The IT Product Guide
Read honest & candid reviews on hundreds of IT Products from real users.
Discover which products truly live up to the hype. Start reading now.
http://ads.osdn.com/?ad_id=6595&alloc_id=14396&op=click