optimization issues in intel fortran 10.1.015 and 10.1.018

[randysuggs]

We have a simulation that we have validated the results using the following baseline:

redhat fedora core 3

gcc 3.4.2

intel fortran 9.0.024

kernel ver 2.6.12.1

We use optimization level 2 (-O2)

When we upgrad the fortran compiler to 10.1.015 or 10.1.018 and use optimization, our results are nowhere near expected. When we turn off optimization, our results are as expected. So, we built a system that is documented by intel as being tested as follows:

redhat fedora core 7

gcc 4.1.2

kernel ver 2.6.21

Intel fortran 10.1.015

Using optimization, our results are way out of the expected range, however when we turn off optimization, our results are as expected. We need to be able to optimize our code so that we can run near real time. With no optimization, we cannot even get close to real time.

Please advise.

Randy Suggs

[Steven_L_Intel1]

Please submit a test case to Intel Premier Support. See if you can isolate the problem to a particular calculation that creates the difference. You may also want to experiment with options such as -fp:source or -C or -assume dummy_aliases to see if the results change. Be aware that if your algorithm is computationally unstable, any variation in the order of operations can lead to small FP differences that can build up.