openexr-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Openexr-devel] Multipart Performance Question


From: peterh
Subject: Re: [Openexr-devel] Multipart Performance Question
Date: Sat, 13 Apr 2013 00:07:44 +1200
User-agent: Roundcube Webmail/0.8.4

We haven't done a great deal of testing comparing reading all channels - we have seen very satisfactory speedups reading only one part from a file, particularly when the SSE optimised path gets used. Theoretically, multipart EXRs will be a little slower when reading all parts, though larger images won't make any difference.  If your hobby is writing hundreds of channels into very tall, very thin images you might notice the slowdown!

You may also notice an increase in filesize if the channels are similar, since the compression can take advantage of the redundancy if all channels are compressed together, but not if they are compressed separately into different parts.

You do have some freedom in how you map channels into different parts: it might pay to group channels that are usually read together into the same part, storing less commonly read channels separately - textures for rendering might well benefit from some optimisation like this

Basically, if there's any chance you might only want a subset of the channels in an EXR, the speed gain you'll get from being able to ignore a part of a multipart file will outweigh any slowdown you suffer when you need to read all parts.

Since parts can have different datawindows, multipart images might be faster even when always reading all channels, though compression tends to mitigate that overhead anyway.

Peter

On 2013-04-12 22:15, Thorsten Kaufmann wrote:

Hey everyone,

 

Due to using a lot of channels we are quite often seeing slowdowns due to scanline interleaving. Now that we have parts i

Was wondering if parts come at a cost and how much of a cost.

 

Say I would put each layer (RGB) in its own part. How much would that impact performance in the case of using a single layer

At the end, the beginning or using half of the layers or all?

 

Basically the question is if a multipart image with each RGB as a part would perform as fast as a single part scanline image

When actually using all channels vs. using just a few.

 

Regards and a late congrats on 2.0 J,

Thorsten

 

Thorsten Kaufmann
Head of Production


 

 

 

Mackevision Medien Design GmbH
Forststraße 7
D-70174 Stuttgart

T  +49 711 93 30 48 59
F  +49 711 93 30 48 90
M +49 151 19 55 55 02

address@hidden
http://www.mackevision.de

Geschäftsführer: Armin Pohl, Joachim Lincke, Karin Suttheimer
HRB 243735 Amtsgericht Stuttgart

 

 

 

 


 

F_BOX picture shooter 2.0: High-End-3D as a web service

Erfahren Sie mehr:www.mackevision.de/de/technology/f_box_produkt_familie/picture_shooter.html

Read more about it:www.mackevision.de/en/technology/f_box_product_family/picture_shooter.html

 


 

 

 

 



_______________________________________________
Openexr-devel mailing list
address@hidden
https://lists.nongnu.org/mailman/listinfo/openexr-devel
 

reply via email to

[Prev in Thread] Current Thread [Next in Thread]