Call the patchMad.sh script from within the plugin's finalize

[roiser]

first version of calling the patch script
patch script and patch files moved into the plugin directory

[valassi]

I am doing further integration tests and patches, to make sure that calling this from generateAndCompare.sh also still works.

@roiser I am confused however because, unless I did something different from you, the patchMad.sh scripts is not called from finalize (or better, it is called but it does nothing because it fails and prints an error)

[avalassi@itscrd80 gcc11.2/cvmfs] /data/avalassi/GPU2023/MG5aMC/ghav-mg5amcnlo> more pippo 
set stdout_level DEBUG
set zerowidth_tchannel F
generate g g > t t~
output madevent CODEGEN_mad_gg_tt --hel_recycling=False --vector_size=32 --me_exporter=standalone_cudacpp

[avalassi@itscrd80 gcc11.2/cvmfs] /data/avalassi/GPU2023/MG5aMC/ghav-mg5amcnlo> ./bin/mg5_aMC pippo
...
INFO: Use Fortran compiler gfortran 
INFO: Use c++ compiler g++ 
INFO: Generate web pages 
DEBUG:  'Entering PLUGIN_ProcessExporter.finalize', self.in_madevent_mode =  Entering PLUGIN_ProcessExporter.finalize True [output.py at line 204] 
Usage: /data/avalassi/GPU2023/MG5aMC/ghav-mg5amcnlo/PLUGIN/CUDACPP_SA_OUTPUT/patchMad.sh <process.[madonly|mad]> <vecsize> <patch_dir> [--nopatch|--upstream]
Output to directory /data/avalassi/GPU2023/MG5aMC/ghav-mg5amcnlo/CODEGEN_mad_gg_tt done.
Type "launch" to generate events from this process, or see
/data/avalassi/GPU2023/MG5aMC/ghav-mg5amcnlo/CODEGEN_mad_gg_tt/README
Run "open index.html" to see more information about this process.
quit

In particular, Usage: /data/avalassi/GPU2023/MG5aMC/ghav-mg5amcnlo/PLUGIN/CUDACPP_SA_OUTPUT/patchMad.sh <process.[madonly|mad]> <vecsize> <patch_dir> [--nopatch|--upstream] is an error printout. It means that the script failed.

I will add an error check if possible. And also try to fix this

[valassi]

I have also fixed all other pending issues
To do still

• allow the patchMad to be used with any vecsize as input (now it assumes it was generated with 16384, so the hardcoded 32 in output.py will fail there)
• resurrect a "no-patch" mode in generateAndCompare: this is needed to create new patch.P1 and patch.common if needed
• final checks that the same code is generated in mg5amc and in my generateAndCompare script (modulo irrelevant details)
• regenerate all processes
• added comments about self.in_madevent_mode (but did not use the flag everywhere, it did not seem appropriate/necessary)
• I also moved to the very latest gpucpp from Olivier

Now rerunning tests, should be able to merge soon

[valassi]

I will wait for the tests running tonight to succeed, and will merge tomorrow.

This will close issue #656. Note that the strategy for integrating patchMad.sh patches upstream has changed:

• this is no longer about getting rid of patchMad.sh and moving the patches to the common LO templates (as in my WIP MR WIP: remove patchMad.sh #675, which has now been closed)
• instead, we keep patchMad.sh but we move it to the plugin itself so that it can be called from standard MG5AMC commands (this MR Call the patchMad.sh script from within the plugin's finalize #747)

[valassi]

The CI tests passsed - I am self merging