camb compiler error with gcc5

Use of Healpix, camb, CLASS, cosmomc, compilers, etc.
Post Reply
David J. E. Marsh
Posts: 2
Joined: September 09 2015
Affiliation: King's College London

camb compiler error with gcc5

Post by David J. E. Marsh » September 09 2015

I recently upgraded gfortran compilers via macports from gfortran-mp-4.5 to gfortran-mp-5 (which is part of gcc5, the most recent gcc port as of Sep 9, 2015). A new feature of this compiler is that it displays floating point errors.

My compiler settings are

Code: Select all

F90C     = gfortran-mp-5
FFLAGS= -O3
When running CAMB (April 2014) I receive the following error message:

Code: Select all

Note: The following floating-point exceptions are signalling: IEEE_UNDERFLOW_FLAG IEEE_DENORMAL
(this is for an entirely ordinary cosmology, as far as I can tell) This error message does not appear with gfortran-mp-4.5.

I found similar floating point errors once when running CAMB in gdb (sorry, but I forget what).

Googling the error message gives https://www.cisl.ucar.edu/news/01/tips/0209.fpe.html
which suggests such errors should be found, as hey can lead to slow down, if not catasatrophic failure.

I'm not worried so much about this affecting the physics (the Cl's from this look fine), but that errors like this could be slowing CAMB down, or leading to unexpected effects elsewhere. So far, it isn't affecting me, but it is worrying and I wonder if others have become aware of this, or should be made aware of this.

David J. E. Marsh
Posts: 2
Joined: September 09 2015
Affiliation: King's College London

camb compiler error with gcc5

Post by David J. E. Marsh » September 09 2015

I see that some release/debug comments have been added related to the most recent (Feb 2015) version of CAMB. I have yet to check how these affect my post above. The debugging environment and auto-detect of compiler may allow for easier resolution of the above errors.

Post Reply