On Sat, Oct 20 2018, Rafael Sadowski <rafael@sizeofvoid.org> wrote:
> On Sat Oct 20, 2018 at 01:00:12PM +0200, Jeremie Courreges-Anglas wrote:
>> On Thu, Oct 18 2018, Marc Espie <espie@nerim.net> wrote:
>>
>> [...]
>>
>> > armadillo is a C++ wrapper around lapack/blas linear algebra libraries,
>> > a required dependency of mlpack.
>> >
>> > Armadillo is a linear algebra library (matrix maths) for the C++ language,
>> > aiming towards a good balance between speed and ease of use.
>> >
>> > Provides high-level syntax and functionality deliberately similar to Matlab
>> >
>> >
>> > Attached is a tarball for math/mlpack and math/armadillo (extract under ports)
>> >
>> > I haven't checked that it builds under anything but amd64, though it should
>> > with a proper C++ compiler (hence the compiler annotations).
>>
>> Regarding armadillo,
>>
>> - could you please follow the usual layout of Makefile.template, and use
>> tabs for alignement? I find this Makefile hard to review, especially
>> with the location of DISTNAME...
>>
>> - CONFIGURE_STYLE is not needed
>>
>> - cmake says
>>
>> -- Checking for module 'hdf5'
>> --
>> -- HDF5_FOUND =
>>
>> so might pick up math/hdf5 if present (you already noticed, it seems)
>> It also looks for something called "superlu" but we have no port for
>> it (yet).
>>
>> - does not build with base-gcc but I see no compiler annotations.
>> This needs COMPILER = base-clang ports-gcc.
>>
> I fixed all your notes in the attached tarball. I'm ok with the
> attached armadillo version. jca@ could confirm, please!?
I don't think the fortran module is needed, is it? IIUC libgfortran in
WANTLIB is only here because of blas and lapack. On a related note
I don't understand the point of patch-CMakeLists_txt.
I would suggest using "base-clang ports-gcc" for COMPILER, so that only
ports-gcc is needed on non-clang archs, instead of ports-clang +
ports-gcc (the latter needed for blas and lapack).
ENOTIME to look at mlpack, sorry.
> I also review and fix a lot of stuff in mlpack. Our cmake is effected by
> https://gitlab.kitware.com/cmake/cmake/issues/15612. So we can only
> build mlpack with "USE_NINJA = No".
>
> The attached mlpack version looks good for me any objections? Maybe
> about USE_NINJA=No?
>
> From my point of view, both are ready to import, but would like to wait
> for some feedback.
>
> Rafael Sadowski
--
jca | PGP : 0x1524E7EE / 5135 92C1 AD36 5293 2BDF DDCC 0DFA 74AE 1524 E7EE
No comments:
Post a Comment