HowTo: Profiling LMDZ - Historique des versions

Abarral : add minor details

2024-07-16T19:18:26Z

add minor details

Abarral : Add scalsca use

2024-07-16T18:42:27Z

Add scalsca use

Abarral : (WIP) Add scalasca install guide

2024-07-16T18:29:34Z

(WIP) Add scalasca install guide

Abarral : (WIP) Add gprof guide

2024-07-16T17:57:04Z

(WIP) Add gprof guide

Nouvelle page

== Profiling with gprof ===

The legacy way to profile the model is using <code>gprof</code>.
* ''Pros:'' available everywhere, easy to use
* ''Cons:'' old, bad handling of multithreaded applications, requires instrumentation

<code>gprof</code> is a rudimentary but simple to use tool to profile a sequential executable.

==== Instrumenting the code ====

In the <code>.fcm</code> arch file used, add <code>-pg</code> to <code>BASE_FFLAGS</code> and <code>BASE_LD</code>. Recompile the model.

==== Collecting statistics ====

Run the executable <code>gcm.e</code>. This will generate a <code>gmon.out</code> file locally.

==== Reading results ====

Run <code>gprof gcm.e gmon.out > profiling.txt</code> to get a textual view of the profiling.

For a graphical representation, we recommend using [https://github.com/jrfonseca/gprof2dot gprof2dot], via <code>gprof gcm.e | gprof2dot | dot -Tpng -o output.png</code>. A typical example is shown below.

[[Fichier:gprof.png|400px]]

== Profiling with gprofng ==

== Profiling with scorep & scalasca ==

[[Category:HowTo]]

← Version précédente		Version du 16 juillet 2024 à 19:18
Ligne 29 :		Ligne 29 :

	Scalasca ([https://www.scalasca.org/ home], [https://apps.fz-juelich.de/scalasca/releases/scalasca/2.6/docs/manual/ manual]) is a profiling suite developed by Technische Universität Darmstadt Laboratory for Parallel Programming. It is much more capable than <code>gprof</code>, but also requires a heavier setup.		Scalasca ([https://www.scalasca.org/ home], [https://apps.fz-juelich.de/scalasca/releases/scalasca/2.6/docs/manual/ manual]) is a profiling suite developed by Technische Universität Darmstadt Laboratory for Parallel Programming. It is much more capable than <code>gprof</code>, but also requires a heavier setup.
		+	* ''Pros:'' much better overall: excellent multithreaded support with separation between USR/OMP/MPI, filtering, trace profiling, great visualisation, highly customizable
		+	* ''Cons:'' requires a trickier installation, more complex

	=== Installing scalasca ===		=== Installing scalasca ===

@@ Ligne 1 : / Ligne 1 : @@
-== Profiling with gprof ===
+== Profiling with gprof ==
 The legacy way to profile the model is using <code>gprof</code>.
@@ Ligne 23 : / Ligne 23 : @@
 For a graphical representation, we recommend using [https://github.com/jrfonseca/gprof2dot gprof2dot], via <code>gprof gcm.e | gprof2dot | dot -Tpng -o output.png</code>. A typical example is shown below.
-[[Fichier:gprof.png|400px]]
+[[Fichier:gprof.png|800px]]
 == Profiling with scorep & scalasca ==
-Scalasca ([https://www.scalasca.org/ home], [https://apps.fz-juelich.de/scalasca/releases/scalasca/2.6/docs/manual/ manual]) is a profiling suite developed by Technische Universität Darmstadt Laboratory for Parallel Programming. It is much more capable than gprof, but also requires a heavier setup.
+Scalasca ([https://www.scalasca.org/ home], [https://apps.fz-juelich.de/scalasca/releases/scalasca/2.6/docs/manual/ manual]) is a profiling suite developed by Technische Universität Darmstadt Laboratory for Parallel Programming. It is much more capable than <code>gprof</code>, but also requires a heavier setup.
-==== Installing scalasca ====
+=== Installing scalasca ===
 Some supercomputers have scalasca installed already, but for completeness here's a short guide to installing it locally (written 06/24, for v2.6.1):
@@ Ligne 36 : / Ligne 36 : @@
 All dependencies can be downloaded from [https://www.scalasca.org/scalasca/software/scalasca-2.x/requirements.html here].
-===== Compiling CubeBundle / CubeLib =====
+==== Compiling CubeBundle / CubeLib ====
 ''Note: If you're on your own computer, we recommend installing CubeBundle, which includes CubeGUI. This requires Qt5 and OpenGL, which are a pain to install on supercomputers without sudo. For supercomputer local install, we recommand installing CubeW+CubeLib instead, and visualizing the results on your own computer.''
@@ Ligne 60 : / Ligne 60 : @@
 ''Note: To perform traces, Scalasca requires executables in <code>$install_dir/bin/backend/</code> to be in the path. You can safely simply copy them to <code>$install_dir/bin/</code> and add that to the path instead.''
 [[Category:HowTo]]

@@ Ligne 6 : / Ligne 6 : @@
 <code>gprof</code> is a rudimentary but simple to use tool to profile a sequential executable.
 ==== Instrumenting the code ====
@@ Ligne 23 : / Ligne 25 : @@
 [[Fichier:gprof.png|400px]]
 == Profiling with scorep & scalasca ==
 [[Category:HowTo]]