Option to produce multi-build (fortran, cuda, cpp) gridpack

[valassi]

I am looking at #943: why does CMS see speedups in DY+4jets but not DY+3 jets

The way I would like to address this is to produce a multi-build (fortran, cuda, cpp) gridpack

That is to say:

• do a 'make -j bldall' once, which includes fortran, cuda, cpp builds
• do a vegas optimisation once (possibly in cuda which is the fastest)
• then compare only event generation in the various ME implementations

This is related to #674 about madevent_fortran, madevent_cuda, madevent_cpp.

• Our current strategy is about having a symlink to madevent which is then used everywhere. Another idea I had in mind for some time (not sure it is even written in 674 or elsewhere yet) is to have madevent as a script which then calls the appropriate madevent_cuda etc executable in its build directory. This may be more flexible.
• A second question, described in madeventfortran vs madeventcpp vs madeventcuda #674, is the usual thing of vecsize_memmax etc. Strictly speaking, an optimal fortran or cpp should have vecsize 32, not larger. Instead cuda has typically 8192. So one would need separate build directories (or dynamic memory allocations) etc. Anyway for the moment I'd be happy even to just compare cuda, fortran, cpp all with 8192 vecsizes.