Al Crate wrote:
Hiya Piotr ... small world eh?!
indeed! :)
Nope pulling from iif-optimisations branch, I'll check and see if
we've missed out the sse2 compiler flags or something.
We folded those changes into develop once we thought the feature was
ready for it. If you have a chance, try timing just around the actual
read to isolate from the rest of work needed.
On 20/03/13 15:51, Piotr Stanczyk wrote:
Hey Al,
Are you grabbing code from the develop branch?
There should be a input file method :
//---------------------------------------------------------------
// Check if SSE optimization is enabled
//
// Call after setFrameBuffer() to query whether optimized file
decoding
// is available - decode times will be faster if returns true
//
// Optimization depends on:
// the file type (only scanline data is supported),
// the framebuffer channels (RGB/RGBA mono or stereo)
// the framebuffer channel types (all channels half-float
format only)
// the file channels (RGB/RGBA mono or stereo)
// the file channel types (all channel half-float format only)
// whether SSE2 instruction support was detected at compile time
//
// Calling isOptimizationEnabled before setFrameBuffer will
throw an exception
//
//---------------------------------------------------------------
isOptimizationEnabled
Can you run that and let me know what you are seeing?
The read times should indeed be mostly dominated by the
(de)compressor workload. However, on my machine here I saw something
in the region of 10-15% improvement.
Are you able to try this on an uncompressed file? You should see
quite a significant speed up there.
Let me know if that is not the case and we can look into this further.
________________________________________
From: openexr-devel-bounces+pstanczyk=ilm....@nongnu.org
[openexr-devel-bounces+pstanczyk=ilm....@nongnu.org] on behalf of Al
Crate [a...@dneg.com]
Sent: 20 March 2013 03:32
To: openexr-devel@nongnu.org
Subject: [Openexr-devel] iif-optimisations branch tests
Hi all,
Just built the iff-optimised version, I tweaked the code in IlmImfTest
to spit out some timings between the aligned and unaligned tests.
The timings are almost identical, which isn't what I'd expect, is there
anything I should be doing here to make this work, I also tried using
the optimised libraries in our playback tool and that showed little to
no speed improvements.
I'm guessing I'm not using these correctly, is there some compiler
option I require or memory alignment needed?
cheers
al
_______________________________________________
Openexr-devel mailing list
Openexr-devel@nongnu.org
https://lists.nongnu.org/mailman/listinfo/openexr-devel
_______________________________________________
Openexr-devel mailing list
Openexr-devel@nongnu.org
https://lists.nongnu.org/mailman/listinfo/openexr-devel
