The ABINIT Users Mailing List ( CLOSED )

[Alain Jacques <ajs.jacques@gmail.com>]

Hello Samuel,

Samuel Sarholz wrote:
> Hi Alain,
>
> thanks for your hints.
> The relative error is always below 0.001 on the tests that failed.
> So it seems that all compiler give reasonable, but different, results.
>
> (Actually I am a bit surprised by this, as I (tried to) instructed the
> compiler not to do floating point optimizations. Are the results
> compute speed dependent e.g. does the number of iterations done depend
> on it)

Glad to know that your builds are consistent.
I don't exactly catch what you mean by "not to do floating point
optimization" .... what flags are you using to obtain this, let's say on
Intel's compiler?
Computing speed will be affected by compilers performance and
optimization flags. I don't want to start a flame war about compilers
comparison but Intel ifc generally provides the fastest code and
gfortran is a bit slower - nothing catastrophic, a few % - on x86
processors assuming that optimal flags have been invoked. You can find
interesting  clues in the  distribution "KNOWN_PROBLEMS" file. Yann -
I'm not sure anymore if it is Yann, Damien, Pierre-Matthieu or someone
else, thousands of apologies to the forgotten hero - has made a nice job
collecting compilers-processors "tricks" included in the config/optflags
subdirectories. Generally -O2 optimization flags are safe and quite
efficient, -O3 may squeeze some more juice out of the silicon with some
risk of code instability.

Alain