aboutsummaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorGravatar Axel Huebl <axel.huebl@plasma.ninja> 2022-11-08 10:59:13 -0800
committerGravatar GitHub <noreply@github.com> 2022-11-08 10:59:13 -0800
commit6beaa9fc0f5a68694d8dfbb70d6eae01446e4490 (patch)
treead0691dc3ae217133b32644212174996c3fe4c66
parentbf535730cb6c170394da8375887b3ae57ad6ecf5 (diff)
downloadWarpX-6beaa9fc0f5a68694d8dfbb70d6eae01446e4490.tar.gz
WarpX-6beaa9fc0f5a68694d8dfbb70d6eae01446e4490.tar.zst
WarpX-6beaa9fc0f5a68694d8dfbb70d6eae01446e4490.zip
Docs: Improve MPI Threading User FAQ (#3501)
* Docs: Improve MPI Threading User FAQ Clarify that this FAQ entry explains the initial output of WarpX in stdout. * Mention: `amrex.async_out_nfiles` * Link Build Options, too
-rw-r--r--Docs/source/usage/faq.rst21
1 files changed, 16 insertions, 5 deletions
diff --git a/Docs/source/usage/faq.rst b/Docs/source/usage/faq.rst
index 59c8c3aee..67cea8d66 100644
--- a/Docs/source/usage/faq.rst
+++ b/Docs/source/usage/faq.rst
@@ -6,16 +6,27 @@ FAQ
This section lists frequently asked usage questions.
-What is MPI thread support level?
----------------------------------
+What is "MPI initialized with thread support level ..."?
+--------------------------------------------------------
-We report this in output on startup together with other information.
+When we start up WarpX, we report a couple of information on used MPI processes across parallel compute processes, CPU threads or GPUs and further capabilities.
+For instance, a parallel, multi-process, multi-threaded CPU run could output::
-That is the `MPI support for threaded execution <https://www.mpich.org/static/docs/v3.1/www3/MPI_Init_thread.html>`__, e.g., with OpenMP or system threads.
+ MPI initialized with 4 MPI processes
+ MPI initialized with thread support level 3
+ OMP initialized with 8 OMP threads
+ AMReX (22.10-20-g3082028e4287) initialized
+ ...
+The 1st line is the number of parallel MPI processes (also called *MPI ranks*).
+
+The 2nd line reports on the `support level of MPI functions to be called from threads <https://www.mpich.org/static/docs/v3.1/www3/MPI_Init_thread.html>`__.
We currently only use this for optional, :ref:`async IO with AMReX plotfiles <running-cpp-parameters-diagnostics>`.
In the past, requesting MPI threading support had performance penalties, but we have not seen such anymore on recent systems.
-Thus, we request it by default but you can overwrite it with a compile time option if it ever becomes needed.
+Thus, we request it by default but you can overwrite it with a :ref:`compile time option <building-cmake-options>` if it ever becomes needed.
+
+The 3rd line is the number of CPU OpenMP (OMP) threads per MPI process.
+After that, information on software versions follow.
How do I suppress tiny profiler output if I do not care to see it?