Intel Compiler Options for AMD Processor

[draceswbell_net]

We have a customer who is trying to use a single compiler suite across platforms, but the intel compiler is giving slower performance than the pgi compiler on the AMD processor. The current options used for the compiler are

O3 -g -xHost -fno-fnalias -fno-alias -sox -traceback -complex-limited-range

Is there a better set of generic options that should be used for the latest AMD processor?

I have checked the knowledge base and I could have missed a thread here, but none of these seem to address this issue very much. Any information is appreciated.

[JenniferJ]

Quoting - draceswbell.net

O3 -g -xHost -fno-fnalias -fno-alias -sox -traceback -complex-limited-range

Is there a better set of generic options that should be used for the latest AMD processor?

If you're building and running from the same system, -xHost is the one to use. But if you're running on a different system, you should use -msse2 or -msse3.

Also is it possible to send a testcase?

Thank you,
Jennifer

[TimP]

Quoting - draceswbell.net

O3 -g -xHost -fno-fnalias -fno-alias -sox -traceback -complex-limited-range

-ansi-alias is a more "generic" option than the alias options you used. It asserts that your code complies with the standard on typed aliasing, corresponding to the default of most linux compilers.
If your code depends for performance or accuracy on performing operations in order specified by parentheses, you may find better results by adding -fp-model source .
The optimization performed by icc is more effective on barcelona and newer AMD CPUs; there is no choice specifically for older CPUs.

[draceswbell_net]

Quoting - draceswbell.net

We have a customer who is trying to use a single compiler suite across platforms, but the intel compiler is giving slower performance than the pgi compiler on the AMD processor. The current options used for the compiler are

O3 -g -xHost -fno-fnalias -fno-alias -sox -traceback -complex-limited-range

Is there a better set of generic options that should be used for the latest AMD processor?

I have checked the knowledge base and I could have missed a thread here, but none of these seem to address this issue very much. Any information is appreciated.

Thanks. This is in line with the option tests we tried. We will try to get a test case that shows the difference in performance, but this is proprietary code that is very large so it will take a while to reduce this to a small test case.