v2.0.0-rc5

Version 2.0.0-rc5

v2.0.0-rc4

Version 2.0.0-rc4

v2.0.0-rc3

Version v2.0.0-rc3

v2.0.0-rc2

Version v2.0.0-rc2

v2.0.0-rc1

Version v2.0.0-rc1

v2.0.0-rc.0

This release has an improved API, therefore it requires some changes with respect to the DBCSR v1.
Summary of changes in API:

• Remove arrays for clusters in dbcsr_distribution_new and dbcsr_distribution_get #34
• Rename dbcsr_trace_ab in dbcsr_dot #68
• dbcsr_init_lib takes an MPI communicator #102
• dbcsr_finalize_lib remove input MPI communicator #102
• Print DBCSR statistics with dbcsr_print_statistics #102
• All functions in API have a dbcsr_ namespace #109
• acc_get_ndevices and acc_set_active_device have now dbcsr_ prefix, available in the API #111
• dbcsr_get_block_* has a traspose parameter (old API is will available) #109
• dbcsr_iterator_next_* has a traspose parameter (old API is will available) #109
• Add GPU V100 parameters #172
• Add support for F2008 #136
• Better CMAKE support
• Fix minor bugs

v2.0.0-alpha2

Version 2.0.0-alpha2

v2.0.0-alpha1

Version 2.0.0-alpha1

v2.0.0-alpha

Version 2.0.0-alpha

Version 1.1.0

This DBCSR version introduces predictive modeling for the CUDA generation of the kernels. A decision tree model is built from the data obtained by autotuning certain (m, n, k)-triplets and is used to predict optimal parameters for unseen (m, n, k)-triplets.

Developed as a part of the PASC proposal "Sparse Tensor Linear Algebra Library"

Thanks to @shoshijak for the hard work to implement it!