Coding

Most often codes are written in Python or Fortran. Here some information is given on Fortran and a little bit of C as well.

Have a look at Debugging

Fortran

Fortran is quite popular in Meteorology and Geophysics. Please find some help on solving common problems.

Get some information on Fortran: - Fortran Language, Learning, Compilers

Compilers

There are a few compilers, but most commonly GNU (Gfortran) and INTEL (ifort) are used on our servers.

	GNU Fortran	INTEL Fortran
double Fprecision real	`-fdefault-real-8`	`-r8`
check array bounds	`-fbounds-check`	`-check`
call chain traceback	`-fbacktrace`	`-traceback`
convert little/big endian	`-fconvert=big-endian/little-endian`	`-convert big_endian/little_endian`
default optimisation	`-O0`	`-O2`
highest recommended optimisation	`-O3`	`-O2` maybe `-O3` or `-fast`
position independent code (shared)	`-fPIC`	`-fPIC`

Please take a look into the compiler options guidelines from AMD or Intel, which might be very helpful for finding the right compiler flags for certain CPUs

Servers such as Aurora AMD EPYC ZEN3 compiler quick reference
Servers such as VSC5, LUMI, ... AMD EPYC ZEN4 compiler quick reference
Servers such as SRVX1, JET, VSC4 Intel compiler quick reference

more information on how to find good options can be found on the PRACE (Partnership for advanced computing in Europe) website

Generating optimized code for a certain processor

All compiler providers have special options that have optimizations for certain processors, especially intel.

Processor Type	System	Flag (intel)	Flag (gcc)
Cascade Lake	VSC5	`-xCORE-AVX512`	`-march=cascadelake`
Skylake	JET, SRVX1, VSC4	`-xCORE-AVX512`	`-march=skylake`
Broadwell	SRVX8	`-xCORE-AVX2`	`-march=broadwell`
Sandy Bridge	DEVx1	`-xAVX`	`-march=sandybridge`
Milan	VSC5, Aurora	`-xCORE-AVX2`	`-march=znver3`
Generic x64	containers	`-xCORE-AVX2`	`-march=x86-64-v4`

GCC/GFortran processor compiler flags

Intel Compiler

from P. Seibert using ifort for the fastest code (srvx1):

Makefile
# get GRIP_PATH from environment modules
INCPATH = GRIP_API/include
LIBPATH = GRIP_API/lib
# using -xAVX is for old CPUs !!!
FFLAGS = -cpp -xAVX -ipo -O3 -no-prec-div -opt-prefetch  -m64 -mcmodel=medium  -I$(INCPATH)
LDFLAGS = $(FFLAGS) -L$(LIBPATH) -Bstatic -lgrib_api_f90 -lgrib_api -lm -ljasper 

Remark: for FLEXPART, otherwise you won't need grib_api and jasper, or mcmodel=medium

Tricky Issues

record markers

On 64-bit machines, some Fortran compilers will insert record markers that are 64-bit integers instead of the standard 32-bit integers.

gfortran man page says:

Text Only
-frecord-marker=length
Specify the length of record markers for unformatted files. Valid values for length are 4 and 8. Default is 4. This is different from previous versions of gfortran, which specified a default record marker length of 8 on most systems. If you want to read or write files compatible with earlier versions of gfortran, use -frecord-marker=8.

Code Testing

pFUnit - python Parallel Fortran Unit Testing Framework GitHub

Documentation and other resources

Fortran wikibook DE EN

Automatic Documentation Generation

FORD: Overview in Fortran Wiki | Code on Github

Last update: February 1, 2024
Created: October 14, 2021