Runs fine on /O0 but fails on /O2 and /O3

[ScottBoyce]

Hi All,

I have a strange bug in a large numerical simulation model that uses the Windows Intel Compiler (most recent release). I apologize that the details are so brief, but its largely do to the fact that I am completely stumped on what is wrong. The code runs all example problems perfectly in debug mode and with release only when Optimization is set to /O0 (not for /O2 or /O3).

I do have the Floating-Point Model set to Source (/fp:source) and Floating-Point Speculation to Fast. I have played with every opt turning stuff on/off, but the only thing that seems to work is /O0.

Any help would be greatly appreciated.

[ScottBoyce]

I have identified that it does work when I have Inline Function Expansion set to Disable with /O3

Any comments regarding this, would be greatly appreciated.

[ScottBoyce]

I apologize, I tried to delete this post, but can not. Shortly after writing it, I received an email say that one subroutine without implicit none was passing the a wrong variable name. The program now runs fine.

It still is very strange that with that error I would still get a correct result with debug mode, but not with release.

I apologize for anyone's time I wasted with this post.

[Steven_L_Intel1]

Thanks for letting us know. Optimization can sometimes uncover programming errors not seen without optimization.