MAGMA Forum

http://icl.utk.edu/magma/
https://bitbucket.org/icl/magma

Mark

As noted above, it's an issue in CMake. They don't seem to be making any progress on it.
https://gitlab.kitware.com/cmake/cmake/ ... ote_642545

Mark

It looks like CUDA renamed a field from `memoryType` to `type`. Changing that one line should fix the issue (but I haven't tested). In interface_cuda/interface.cpp, change: return (attr.memoryType == cudaMemoryTypeDevice); to: return (attr.type == cudaMemoryTypeDevice); Note that CUDA deleted a whol...

Note that CUDA 11 deleted (not just deprecated) a number of functions, at least in cuSparse, breaking MAGMA sparse.

Mark

We have had difficulty with CMake compiling MAGMA on Windows. This appears to be a bug or deficiency in CMake. It is unclear when this occurred -- whether some new version of CUDA, CMake, or MAGMA broke it -- as we have previously compiled MAGMA with CMake on Windows. See this thread: https://icl.cs...

This forum is for MAGMA, the linear algebra library for GPUs.
http://icl.utk.edu/magma/

It seems you are refering to the Magma Computational Algebra System,
http://magma.maths.usyd.edu.au/magma/

Mark

Matlab uses both dense and sparse components from MAGMA. It also uses components from SuiteSparse.
We do some collaboration with Tim Davis, the primary author of SuiteSparse. Applications can use both, as Matlab does, but there are no current plans to integrate MAGMA and SuiteSparse together.

Mark

This is an issue with CUDA, not with MAGMA. See: https://forums.developer.nvidia.com/t/cublas-for-10-1-is-missing/71015 What specific OS are you using (e.g., Ubuntu 20.04)? What did you do to install cuda and cublas? From the link above, it sounds like these may be separate packages that need to be ...

The easiest fix is to define $MKLROOT in your environment, so MAGMA knows where to find MKL. E.g., source /opt/intel/bin/compilervars.csh intel64 # sets MKLROOT echo $MKLROOT /opt/intel/compilers_and_libraries_2020.1.216/mac/mkl # in magma directory mkdir build cd build rm * # clean out stale cache ...

As shown in example/example_v2.c, you should include magma_v2.h. Then replace LAPACK dgesdd_ with magma_dgesdd. The job parameter changes from a string to an enum. The dimensions change from int pointers to plain ints. I highly suggest doing a workspace query, as LAPACK and MAGMA will have different...