Port epochX CODEGEN from 270gpu to the 311lovec branch (and pick up the new features there!)

[valassi]

Hi @oliviermattelaer I assign this on myself - then we need to discuss when I should start.

All of my epochX CODEGEN is now based on the branch 270gpu of launchpad and specifically on revision 370

madgraph4gpu/epochX/cudacpp/CODEGEN/MG5aMC_patches/2.7.0_gpu/revision.BZR

Line 1 in aa65924

370

This is this revision
https://bazaar.launchpad.net/~maddevelopers/mg5amcnlo/2.7.0_gpu/revision/370

I understand I should be moving instead to this other branch
https://code.launchpad.net/~maddevelopers/mg5amcnlo/3.1.1_lo_vectorization

For instance this includes the patch for issue #343 about nprocesses:
https://bazaar.launchpad.net/~maddevelopers/mg5amcnlo/3.1.1_lo_vectorization/revision/990

(En passant this is also eventually needed as a prerequisite for helicity recycling I guess, #279).

Details to be discussed, thanks Andrea

[valassi]

In doing this upgrade, I should try to stick closer to the original code - adressing issues such as #341 reported by Olivier

[valassi]

As discusset today with Olivier, note that the 311_lo_vec branch includes multichannel outputs. IIUC there is one more python function with one more argument, which creates the C++ with the additional argument needed for multichannel.

Note also Olivier's suggestion to try out this process to check all looks good

add process p p  > t t~
output PROC_vec
--vector_size=16
--me_exporter=standalone_gpu

(The "vector_size" may be renamed to something else, this is the number of events in one "iteration" of our standalone app).

[valassi]

Hi @oliviermattelaer I have started to work on this in PR #391.

I changed the upstream MG5AMC that I am using with my plugins, from 2.7.0_gpu to 3.1.1_lo_vectorization. Surprisingly (or maybe not?), I see no changes at all in source code except for comments.

Maybe I should actually change the templates I am using to realign them to those from the new branch?...

[valassi]

I am splitting this into two separate parts

• In PR Use the upstream 3.1.1 branch for code generation (without picking up new features yet) #391, which I am about to merge, I am just moving my scripts to the 3.1.1 branch, but I am not modifying th eplugin
• In an upcoming second PR, I will need to port the plugin to pick up the new features of the 311 branch (I have seen that there are some differences, indeed)

The resaons is that I want to have a single environment variable MG5AMC_HOME pointing to a single (311) installation, while doing other work

[valassi]

One of the new features in v311 (not the only one) is helicity recycling.

THis is the third time I repeat this comment: Within PR #401, note that I had to introduce this fix while moving to v311
04d4b8e
This indeed picks up a new feature of v311 (#360), but it is only relevant to helicit yrecycling (#279).

In practice

• initially I started generating code (madevent only, or madevent+cudacpp) without disabling heliicty recycling
• under this configuration, code generation of madevent+cudacpp did not work, and I had to fix it as above
• this was especially puxxling because I had to change some code generation lines (for generating cudacpp) which were not used in my normal cudacpp-only version
• after fixing it, I got code with helicity recycling, which however did not build (not clear why, but not an issue today): to fix the build I disabled helicity recycling
• in retrospective, the reason why I had to fix those lines is thus for helicity recycling code, which in any case I later decided to disable: if I had disabled helicity recycling before trying to generate madevent+cudacpp code, probably that patch would not hav ebeen necessary

[valassi]

Reminder for myself:

• do a diff of madgraph/iolibs/export_cpp.py in the 270 and 311 installations, this is where to start

[valassi]

I am closing this.

The "picking up new features from the 311 branch" is mainly the multichannel stuff. This is discussed in #342. I have just renamed that to mention the 311 branch:

• Olivier has done the code in the 311 branch (in cudacpp)
• We need to integrate that in codegen ("do a diff of madgraph/iolibs/export_cpp.py in the 270 and 311 installations, this is where to start")