search

evaleev / libint

Libint: high-performance library for computing Gaussian integrals in quantum mechanics
Other
225 stars 95 forks source link

LIBINT2_MAX_AM_default2 definition #149

Closed loriab closed 4 years ago

loriab commented 4 years ago

In linking against libint2 built with derivatives, Psi4 folks have noticed errors like the below

/home/psilocaluser/gits/hrw-quaternary/objdir_dev10_gcc4/stage/include/libint2/engine.impl.h: In member function 'void libint2::Engine::_initialize()':
/home/psilocaluser/gits/hrw-quaternary/objdir_dev10_gcc4/stage/include/libint2/engine.impl.h:627:3: error: 'LIBINT2_MAX_AM_default1' was not declared in this scope
   BOOST_PP_LIST_FOR_EACH_PRODUCT(
   ^
/home/psilocaluser/gits/hrw-quaternary/objdir_dev10_gcc4/stage/include/libint2/engine.impl.h:627:3: note: suggested alternative: 'LIBINT2_MAX_AM_default'

We've been patching them up and proceeding (step (5) of https://github.com/psi4/psi4/pull/1721#issuecomment-537945417). This is an attempt at a more permanent solution. It does write the extra defines to libint2_params.h, but maybe it's doing untold harm somewhere else. And the fact that it's using the max ERI deriv level instead of the max of all integrals classes is admittedly bad.

#ifndef LIBINT2_MAX_AM_default                  #ifndef LIBINT2_MAX_AM_default
# define LIBINT2_MAX_AM_default 4               # define LIBINT2_MAX_AM_default 4
#endif                              #endif

                                  > #ifndef LIBINT2_MAX_AM_default1
                                  > # define LIBINT2_MAX_AM_default1 4
                                  > #endif
                                  >
                                  > #ifndef LIBINT2_MAX_AM_default2
                                  > # define LIBINT2_MAX_AM_default2 4
                                  > #endif
                                  >
#ifndef LIBINT2_MAX_AM_overlap                  #ifndef LIBINT2_MAX_AM_overlap
# define LIBINT2_MAX_AM_overlap 4               # define LIBINT2_MAX_AM_overlap 4
#endif                              #endif
andysim commented 4 years ago

Thanks for handling this, @loriab. My solution was inspired by this section of code in the Fortran interface - if you don't like the idea of dumping the max value in there, perhaps it could just be assigned to whatever the non-derivative integrals have instead?