Remove smatrix counters and regenerate all processes #762
Conversation
|
Thanks @hageboeck very good. Just a few comments inline. Only thing, especially for the last point of your initial message I would ask @oliviermattelaer to check that this is all good? I also see some re-ordering happening, e.g. in Cards files or a variable ALOHARoutine, do we understand that this is all good? |
|
Hi, Yes the ordering for the variable in the makefile is irrelevant and the ordering in that file too. Thanks a lot @hageboeck, It is ready to go! |
At least for the makefile, I can easily fix this in upstream mg5. For the cards, I haven't found the code yet. @roiser Is this the coupling bug you were mentioning? |
|
No need to fix it, those ordering have no meaning (and no impact) |
This is a fix done for respecting the order of the couplings as they appear in the code generation, see also #757 There was another fix done last Friday which fixes the problem that the non-running couplings need to be filtered out before this re-ordering is applied, see #761 |
In order to facilitate the code generation, add a script to generate the processes in this repo. Pass all processes to be generated as argument, e.g: ./runCodegen.sh gg_tt.mad ee_mumu.mad It is assumed that each process directory contains mg5.in with all madgraph commands that are needed to generate the corresponding process. The script will try to stage the changes, so they can be committed.
Instead of one copy per subprocess, link these files into the subprocess directories.
Since matrix1 is called in a loop over all events, instrumenting it with a timer and counter creates significant overhead. Curiously, this counter is never read, so it can be removed. For pp -> tt~ 012j, the difference for 1M events on 16 cores is: Walltime: 35:48 m --> 7:45 m CPU time: 13:14 h --> 2:56 h
To be able to run the code generator from every directory, the path to patchMad needs to be relative to the plugin directory.
The following changes are included: - Updated mg5. mg5 sets zero-width t-channel to True. - Removed all timers from smatrix function. This signficantly slowed down the Fortran MEs. - Link counters.cc and ompnumthreads into subprocesses instead of copying them.
The files seem to be regenerated when ufomodel runs, and they only create noise in git.
hageboeck
force-pushed
the
fixTimingCounters
branch
from
612cc0c to
0e70701
Compare
This adds the following commits by Stephan (first part of PR madgraph5#762) c586208 Stephan Hageboeck Wed Sep 6 19:15:44 2023 +0200 [CODEGEN] Fix the path to patchMad 57313ae Stephan Hageboeck Fri Sep 1 17:43:19 2023 +0200 [CODEGEN] Remove matrix1 counter from Fortran MEs. 9635bbb Stephan Hageboeck Fri Sep 1 17:31:56 2023 +0200 [CODEGEN] Move counters.cc and ompnumthreads to SubProcesses. 30ca6cf Stephan Hageboeck Fri Sep 8 18:04:25 2023 +0200 Add codegen inputs for standalone directories. 3823214 Stephan Hageboeck Wed Sep 6 19:19:03 2023 +0200 Add a script for running code generator from mg5 interface. 5ca48d7 Stephan Hageboeck Fri Sep 8 18:00:34 2023 +0200 Add gitignore for cudacpp. The main two effects of these changes are - functions counters_matrix1_(start|stop)_ and calls to them are removed - a runCodegen.sh script has been added, which also leaves a file mg5.in in each generated directory Fix conflicts: epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/output.py epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/patchMad.sh
…an's excellent patch in PR madgraph5#762)
…ts with Stephan's PR madgraph5#762 git checkout c586208 $(git ls-tree --name-only HEAD *.mad) git checkout c586208 $(git ls-tree --name-only HEAD *.sa)
This adds the following commits by Stephan (second part of PR madgraph5#762) f1244bf Stephan Hageboeck Fri Sep 8 18:14:31 2023 +0200 Regenerate code for standalone directories. 4c02578 Stephan Hageboeck Fri Sep 8 17:39:53 2023 +0200 Regenerate code for all .mad directories. I will in any case regenerate all processes with my current oct23av (and compare)
…ndo temporary changes before merging PR madgraph5#762) git checkout 8a0c4d7 $(git ls-tree --name-only HEAD *.mad) git checkout 8a0c4d7 $(git ls-tree --name-only HEAD *.sa)
This adds the following commit by Stephan (third and last part of PR madgraph5#762) f75e994 Stephan Hageboeck Fri Sep 8 18:41:44 2023 +0200 Remove pickle files from version control. Fix conflicts. Remove manually all 8 *.mad/bin/internal/ufomodel/py3_model.pkl as intended by Stephan git rm $(git ls-tree --name-only HEAD *.mad/bin/internal/ufomodel/py3_model.pkl)
FK_MDL_WTfor FFV functions changes toZERO