-
Notifications
You must be signed in to change notification settings - Fork 0
Tools
Information is only about tools irek has tested and which are available as specified below
- vtune (available on triolith, akka, abisko)
I do not have information about using vtune on akka and abisko. I was using it on triolith.
Compile program in debuging mode.
ifort -g
Analysis program amplxe-cl <action> [action-option] [global-option] [[--] exe [exe-options]]
$ amplxe-cl -collect hotspots -user-data-dir ~/ -result-dir r001hs -- ./exe exe-options
$ /MPIRUNDIR/mpirun -n 8 amplxe-cl -collect hotspots -user-data-dir ~/ -result-dir r001hs -- ./exe exe-options
where:
-collect is an action; hotspots is an argument of collecting type.
-user-data-dir is an action-option; ~/ is an argument of results location
-result-dir is an action-option; r001hs is an argument of results directory name.
For visualizing the results
$ amplexe-gui results-folder
You can analysis in GUI mode
To use the VTune in the GUI mode use:
$ amplexe-gui
once the gui starts create a new project. As the new project is created it will ask for binary name, path and binary parameters, if any. Then start the analysis
- scalasca (available on triolith)
Add module
module load scalasca
work with intel 12.1.4 and impi/4.0.3.008
Compile program adding scorep before compiler
scorep gfortran ...
scorep ifort ...
scorep mpif90 ...
Analysis program
scalasca –analyze ./myprog.x
or
scan ./myprog.x
MPI
scalasca –analyze /MPIRUNDIR/mpirun -n 8 ./myprog.x
or
scan /MPIRUNDIR/mpirun -n 8 ./myprog.x
For visualizing the results
scalasca –examine score_directory
or
square score_directory
- gprof
Compile program adding -pg flag to compiler
gfortran -pg
Analysis program
./myprog.x (running program generat gmon.out file)
gprof myprog.x gmon.out > output.file
For visualizing the results
EDITOR(e.g. vim, nano, emacs) output.file
Good debuger if you create parralel applications. To run it use:
mpirun -np x -tv ./app
For more info look here: tutorial
Here are informations about some flags and reason why used.
-fp-model precise, -fp-model except - to improve the consistency and reproducibility of floating-point results while limiting the impact on performance
-r8 - set default real to double precision
-fdefault-real-8 - set default real to double precision
###Development
###Milestones