avoid calls to blas routines when the data type is not compatible #603
Conversation
|
Yep, this is not tested in the regression tests, so it slipped through. |
There was a problem hiding this comment.
Thanks @pcarruscag for taking care of this. However, a few changes are still needed.
Also CODI_FORWARD_TYPE should be added to the ifdefs, i.e. it should become
#if (defined(HAVE_MKL) || defined(HAVE_LAPACK)) && !defined(CODI_REVERSE_TYPE) && !defined(CODI_FORWARD_TYPE)
These ifdefs should be added to blas_structure.hpp and blas_structure.cpp as well, but that is a bit trickier. In blas_structure.hpp, make the following changes
#ifdef HAVE_LIBXSMM
should become
#if defined(HAVE_LIBXSMM) && !defined(CODI_REVERSE_TYPE) && !defined(CODI_FORWARD_TYPE)
#if !defined(HAVE_LIBXSMM) && !defined(HAVE_BLAS) && !defined(HAVE_MKL)
should become
#if !defined(HAVE_LIBXSMM) && !defined(HAVE_BLAS) && !defined(HAVE_MKL) || defined(CODI_REVERSE_TYPE) || defined(CODI_FORWARD_TYPE)
In blas_structure.cpp similar changes are needed. If you want, I can give them in details.
|
@vdweide ok I think I safeguarded the required areas, if you could test this it would be great as I do not have these libraries. |
|
@pcarruscag, everything looks to be working as it should. As soon as the last regression test goes through (it failed due to a connection time out), it can be merged in. |
avoid calls to blas routines when the data type is not compatible. Merging in this one.
Proposed Changes
Avoid calls to external functions when compiling in AD and the datatype becomes incompatible.
(The HAVE_LAPACK flag is compatible with AD in the RBF interpolation but not in the HOM feature).
@vdweide please have a look at this.
PR Checklist
Put an X by all that apply. You can fill this out after submitting the PR. If you have any questions, don't hesitate to ask! We want to help. These are a guide for you to know what the reviewers will be looking for in your contribution.