Initial skeleton of a more comprehensive CI test suite

[valassi]

Hi @roiser @hageboeck @oliviermattelaer @zeniheisser as discussed this is a PR with an initial skeleton of a more comprehensive CI test suite.

Presently this does

• code generation (and check of clang formatting)
• builds
• runTest.exe, check.exe (no gcheck.exe as there is no GPU) And this is for eight processes both in mad and sa versions. Only on Linux. For all SIMD built on the node. This is using ccache and github caches, so you can repeat a test and the build goes way faster.

Amongst things to add

• same runTest.exe and chec.exe, with FPEs enabled (this currently would fail a few processes)
• more complete 'tput' tests including some comparison of physics results
• 'tmad' tests for a single madevent, comparing xsec and lhe files
• 'tlau' tests for a full workflow launch or generate events
• more processes beyond SM
• Mac, not only Linux
• other compilers on Linux

Anyway: for the moment these tests wich are in the PR should all succeed. So this should already be a first good approximation of something to check that we are not introducing new issues with some CODEGEN changes.

(Ah, the codegen test presently compares the code to the git repo... but I will disable this for the moment. I can take care pf popoulating the generated code in the repo if you want. But it would be great if you can run at least these tests when making new changes to codegen).

I am back in a few days and we can discuss this later on. Thanks! Andrea

[valassi]

Here you see that the full test suite is successful https://github.com/madgraph5/madgraph4gpu/actions/runs/6757359626 The full initial build and test of pp_tt012j.mad takes 19 minutes when not cached https://github.com/madgraph5/madgraph4gpu/actions/runs/6757359626/job/18367850043

I have now done a forced push just to retrigger the workflow (after it is merged, it will be possible to MANUALLY trigger it from https://github.com/madgraph5/madgraph4gpu/actions/workflows/testsuite_allprocesses.yml). The full build and test of pp_tt012j.mad when it is cached via ccache is now just over 2 minutes! https://github.com/madgraph5/madgraph4gpu/actions/runs/6757482392/job/18368091997

In other words, when the CODEGEN changes are minor and do not result in real code changes, these tests are fast. However when the generated code does change, one must rebuild it from scratch an dit takes longer.

[oliviermattelaer]

Thanks for this Andrea having tests is always a good things. So I approve. Now we will learn in the long run which tests are "problematic" to maintain and which tests we are missing. Both will need to be fixed in due time (if needed obviously).

Thanks for this,

Olivier

[valassi]

Thanks Olivier very good I merge this then :-)

I have a few other things in the pipeline, this helps me get things out of the way.

NB THE CI TESTS WILL BE A BIT LONGER NOW (20 minutes? I can disable the longest ones if this is a nuisance, initially)