I have a program that takes an image as input, convolves it repeatedly to create an image pyramid, computes functions of the pyramid, and stores the outputs in a file.
If I run this program twice on a set of images, only about 30% of the resulting pairs of files are equal, and it's a different 30% every time. I know the problem is in fwiFilterRow_32f_C1R and fwiFilterColumn_32f_C1R because I substituted my own code for these functions and got deterministic results.
I am using the MS Visual C++ compiler 2005 on Windows XP with Framewave 1.1. Nothing in the 1.2 or 1.3 release notes indicate that this issue has been addressed. I am also setting the number of threads to 1 to avoid threading issues, though it does not seem to help.
This problem does not occur in the Linux version of my code (g++ 3.4.6).
Hi Mark,
It would be great if you send us the Code file/snippet.
Though I shall try and run this test with a row Filter kernel of
[1 1 1 1 1] and see the behaviour.
Thanks and Regards,
Banger.
I tried rewriting this code (the bug is part of a large project) to create something I could submit here and have been unsuccessful in making the bug re-appear. I will need to start with a copy of the entire project and remove pieces until I make the bug disappear. This will take more time.
I tried rewriting this code (the bug is part of a large project) to create something I could submit here and have been unsuccessful in making the bug re-appear. I will need to start with a copy of the entire project and remove pieces until I make the bug disappear. This will take more time.
Sourceforge.. Dandy :)
Sourceforge.. Very nice :)
Sourceforge.. Slap-up :)