15.0.2.179 and Multithread libraries problem

[Jauch]

Hello,

Today I started to have a problem when trying to run my code.
The code must read an HDF5 file and at some point, return the biggest value on one specific dataset.

The dataset is set to 0.0 on all its values, but the value found was much bigger.

I try to run the code in DEBUG to find the error, as there was some changes in the code since the last time I compiled, but in DEBUG, the error goes away.
I thought that could be a problem with OpenMP, so I disabled the OpenMP on Release mode. But the error was still there.
After many tests, I could determine that the error only happens when I use the Multithread option in the Compiler Runtime Libraries. Even in Release Mode, if I change it to Debug Multithread, the program seems to works as supposed, like when in Debug mode.

Could you give me some advice on how to look to this problem?
I'll try to compile the code without changes and than add the different pieces of code until I find the culprit, but I would like to know if a broken code could cause a problem when using the Runtime Multithread libraries without cause the same problems when using the Debug Multithread runtime libraries.
 

Thanks for your time :)
Eduardo

[Jauch]

Well,

I found the error.
It was a maxval evaluation that was being done in the entire matrix, but should not.

For some reason, with the Debug Multithread, the matrix indexes that the program shouldn't be looking at where all set to zero, but with the Multithread library not.

Sometimes it's just a question of dig deeper... :)

Cheers,

Eduardo