v0.9.9

Version 0.9.9

• Fix #34: Empty vector/matrix multiplication returns correct shape for result

v0.9.8

Version 0.9.8

• Fix #31: Remove ndim check from sparse vector ctypes wrapper to accommodate np.matrix arrays that can't be flattened

v0.9.7

Version 0.9.7

• Fix #28: incorrect output when out is provided but the matrix multiplication is trivially all zeros
• Added scipy sparse classes where MKL is used for matrix multiplication

v0.9.6

Version 0.9.6

• Add wrapper for MKL iterative CG and FGMRES solvers in sparse_dot.solvers

v0.9.5

Version 0.9.5

• Add wrapper for MKL pardiso solver in sparse_dot.solvers

v0.9.4

Version 0.9.4

• Replace deprecated scipy spmatrix.A calls with spmatrix.todense()

v0.9.3

Version 0.9.3

• Directly expose service functions mkl_get_max_threads, mkl_set_num_threads, mkl_set_num_threads_local, mkl_get_version, and mkl_get_version_string.

v0.9.2

Version 0.9.2

• Explicit check for interface env MKL_INTERFACE_LAYER and interface selection in python.
  Will raise a RuntimeWarning if ILP64 is requested but the 64-bit interface layer is unavailable.
  Note that the MKL_INTERFACE_LAYER env must be set before this package is imported.

v0.9.1

Version 0.9.1

• Support for out parameter with sparse-sparse multiplication when dense=True

v0.9.0

• Support for scipy sparse arrays (introduced in scipy 1.11)
• Additional fallback support for finding mkl DLLs with OneAPI