Release Notes: This release fixes division by zero bugs in xLAED6 and xLANHF, unscaling bugs in xGGEV and xGGEVX, uninitialized variables in ZHERFSX, other assorted bugs, and issues with the cmake build system.
Release Notes: This release adds xGEQRT, a QR factorization facility that enables better performance when the blocked reflectors need to be reused. It adds xGEQRT3, a recursive QR factorization facility that has high performance on tall and skinny matrices. It adds xTPQRT, a collection of Communication-Avoiding QR sequential kernels. It replaces the build system with CMAKE for better portability. It adds Doxygen documentation. It integrates LAPACKE's C language APIs into LAPACK.
Release Notes: Numerous bugfixes.
Release Notes: The APIs for C and Fortran have been cleaned up to make it easy to use both. Functions have been added for computing the complete CS decomposition of a partitioned unitary matrix. xSYTRF and xSYTRI have been accelerated. SLAMCH and DLAMCH have been made thread-safe.
Release Notes: Minor bugfixes and documentation cleanups.
Release Notes: Extra-precise iterative refinement. XBLAS. Non-negative diagonals from Householder QR. High-performance QR and Householder Reflections on low-profile matrices. A new fast and accurate Jacobi SVD. Routines for Rectangular Full Packed format. Pivoted Cholesky. Mixed precision iterative refinement routines. Some new variants for one-sided factorization. A more robust DQDS algorithm. Improvements to the multi-shift Hessenberg QR algorithm.
Release Notes: A complex version of the SVD driver was added, and I/O has been simplified. Dependencies on LAPACK have been externalized, and assorted bugfixes have been made.
Release Notes: The BLAS implementation was completed. The number of timers accessible was increased. Code cleanups and bugfixes were made.
Release Notes: Bugfixes for errors that could produce inaccurate results or incorrect table sizes in certain cases. A Complex SVD driver has been added. Some documentation bugs have been fixed.
Release Notes: Many new algorithms. Several implementations have been replaced with much faster and/or much more accurate versions. Support has been added for exploiting fast single- precision hardware where possible. Thread safety has been added. Reliability on SMP systems has been improved. There is additional support for matrices with NaN/subnormal elements, and optimization of the balancing routines. Several bugfixes, including a rare bug that could produce wildly inaccurate results with QR factorization.