Tags give the ability to mark specific points in history as being important
  • v0.1   Version 0.1

    Initial release of Catamari. There is support for multithreaded (via OpenMP tasking) supernodal sparse-direct factorizations (including with iterative refinement) and dense and sparse-direct multithreaded Determinantal Point Process sampling.

  • v0.2   Version 0.2

    This release includes two major performance improvements to the sparse-direct LDL factorizations and DPP sampling: (1) Simplicial, up-looking factorizations are now automatically used for small and/or low-intensity factorizations, leading to orders of magnitude speedups on some matrices, and (2) Nested OpenMP tasking now terminates based upon the workload of the subtree rather than its depth. Beyond the performance improvements, the factorizations and DPP sampling can now be instrumented -- via the -Denable_timers=true configuration option -- to output a Graphviz visualization of the inclusive and exclusive processing times at the top of the supernodal assembly tree. And an example driver was added to visualize -- via ASCII and/or TIFF -- uniform and/or maximum likelihood spanning trees from 2D and 3D grid graphs.

  • v0.2.1   Version 0.2.1

    This release includes two major performance improvements to the sparse-direct LDL factorizations and DPP sampling: (1) Simplicial, up-looking factorizations are now automatically used for small and/or low-intensity factorizations, leading to orders of magnitude speedups on some matrices, and (2) Nested OpenMP tasking now terminates based upon the workload of the subtree rather than its depth. Beyond the performance improvements, the factorizations and DPP sampling can now be instrumented -- via the -Denable_timers=true configuration option -- to output a Graphviz visualization of the inclusive and exclusive processing times at the top of the supernodal assembly tree. And an example driver was added to visualize -- via ASCII and/or TIFF -- uniform and/or maximum likelihood spanning trees from Z^3 or a hexagonal tiling of the plane.

  • v0.2.2   Version 0.2.2

    This release includes two major performance improvements to the sparse-direct LDL factorizations and DPP sampling: (1) Simplicial, up-looking factorizations are now automatically used for small and/or low-intensity factorizations, leading to orders of magnitude speedups on some matrices, and (2) Nested OpenMP tasking now terminates based upon the workload of the subtree rather than its depth. It also contains sequential and DAG-scheduled dense, nonsymmetric DPP samplers and a routine for a posteriori computation of the likelihood of a DPP sample. The factorizations and DPP sampling can now also be instrumented -- via the -Denable_timers=true configuration option -- to output a Graphviz visualization of the inclusive and exclusive processing times at the top of the supernodal assembly tree. And example drivers were added for sampling and visualizing -- via ASCII and/or TIFF -- uniform domino tilings of the Aztec diamond and uniform spanning trees from Z^3 or a hexagonal tiling of the plane.

  • v0.2.3   Version 0.2.3

    This release includes two major performance improvements to the sparse-direct LDL factorizations and DPP sampling: (1) Simplicial, up-looking factorizations are now automatically used for small and/or low-intensity factorizations, leading to orders of magnitude speedups on some matrices, and (2) Nested OpenMP tasking now terminates based upon the workload of the subtree rather than its depth. It also contains sequential and DAG-scheduled dense, non-Hermitian DPP samplers and a routine for a posteriori computation of the likelihood of dense and sparse DPP samples. The factorizations and DPP sampling can now also be instrumented -- via the -Denable_timers=true configuration option -- to output a Graphviz visualization of the inclusive and exclusive processing times at the top of the supernodal assembly tree. And example drivers were added for sampling and visualizing -- via ASCII and/or TIFF -- uniform domino tilings of the Aztec diamond and uniform spanning trees from Z^3 or a hexagonal tiling of the plane.

  • v0.2.4   Version 0.2.4

    This release includes two major performance improvements to the sparse-direct LDL factorizations and DPP sampling: (1) Simplicial, up-looking factorizations are now automatically used for small and/or low-intensity factorizations, leading to orders of magnitude speedups on some matrices, and (2) Nested OpenMP tasking now terminates based upon the workload of the subtree rather than its depth. It also contains sequential and DAG-scheduled dense, non-Hermitian DPP samplers and a routine for a posteriori computation of the likelihood of dense and sparse DPP samples. The factorizations and DPP sampling can now also be instrumented -- via the -Denable_timers=true configuration option -- to output a Graphviz visualization of the inclusive and exclusive processing times at the top of the supernodal assembly tree. And example drivers were added for sampling and visualizing -- via ASCII and/or TIFF -- uniform domino tilings of the Aztec diamond and uniform spanning trees from Z^3 or a hexagonal tiling of the plane.

  • v0.2.5   Version 0.2.5

    This release includes two major performance improvements to the sparse-direct LDL factorizations and DPP sampling: (1) Simplicial, up-looking factorizations are now automatically used for small and/or low-intensity factorizations, leading to orders of magnitude speedups on some matrices, and (2) Nested OpenMP tasking now terminates based upon the workload of the subtree rather than its depth. It also contains sequential and DAG-scheduled dense, non-Hermitian DPP samplers and a routine for a posteriori computation of the likelihood of dense and sparse DPP samples. The factorizations and DPP sampling can now also be instrumented -- via the -Denable_timers=true configuration option -- to output a Graphviz visualization of the inclusive and exclusive processing times at the top of the supernodal assembly tree. And example drivers were added for sampling and visualizing -- via ASCII and/or TIFF -- uniform domino tilings of the Aztec diamond and uniform spanning trees from Z^3 or a hexagonal tiling of the plane. There is also support for elementary DPP sampling.