5.2 | bf23ef4 | 20190809
------------------------
 o Lots of build system updates, including:
   - Bring aspects of build system up-to-date, especially relative to
    that of BLIS.
   - Support out-of-tree builds.
   - Abide by GNU conventions for shared library names.
   - Abide by GNU conventions for prefix, exec_prefix, libdir, and
     include_dir variables and their recognition within the top-level
     Makefile.
   - Flatten various headers.
   - ARG_MAX hack fixes/cleanups.
   - Prepend preexisting contents of CFLAGS to CFLAGS.
   - Support V=1 in top-level Makefile.
   - Unconditionally define _POSIX_C_SOURCE=200809L.
 o Added mixing I/O-related libf2c files.
 o Initialize global scalar constants (e.g. FLA_ZERO, etc.) via
   static initializers rather than at runtime. 
 o Integrated netlib LAPACK test suite.
 o Diabled lapack2flamec layer's xblas.
 o Prepend license notice to most .c and .h files.
 o Various bug fixes.

5.1 | bf23ef4 | 20140318
------------------------
 o Initial import after moving from subversion to git.

5.0 | r4648 | 20101028
----------------------
 o Implemented new operations:
   - Reduction of a Hermitian-definite generalized eigenproblem to
     standardized form. Sequential and hierarchical interfaces are available
     via FLA_Eig_gest() and FLASH_Eig_gest(), respectively.
   - Reduction to upper Hessenberg form, via FLA_Hess_UT().
   - Reduction to tridiagonal form, via FLA_Tridiag_UT(). (lower triangular
     storage only)
   - Reduction to bidiagonal form, via FLA_Bidiag_UT(). (upper bidiagonal
     case only, ie: m >= n)
 o Changed UT Householder transform functions so that alpha is no longer
   constrained to be in the real domain, allowing tau to be complex. This
   allows the transform to retain the property of being a reflector in the
   complex domain.

 o Various improvements to SuperMatrix runtime system, including support for
   GPU execution if GPUs are available and the operation consists of sub-tasks
   that are supported via CUBLAS.
 o Various improvements to control tree implementation, including better
   error checking to prevent the user from trying to run with control trees
   that execute non-existent variants.
 o Fixed some bugs with beta scaling in gemm, hemm, symm, trmm, and trsm when
   beta is non-unit.
 o Various improvments to the BLIS.
   - Implemented the following new operations:
     - bli_trmvsx (trmv to different vector)
     - bli_trsvsx (trsv to different vector)
     - bli_fnorm (Frobenius norm)
     - bli_setv (set vector to scalar)
     - bli_setm (set matrix to scalar)
     - bli_setmr (set triangular matrix to scalar)
     - bli_setdiag (set diagonal to scalar)
   - Implemented missing row-major support for syr2k and her2k.
   - Added various wrapper routines to map calls to Hermitian operations to
     symmetric equivalents for real datatypes.
   - Rewrote BLIS macros to look and act more like functions.
   - Changed semantics of axpymt, axpysmt, copymt, swapmt to match BLAS/LAPACK.
   - Fixed a subtle scaling bug in herk and her2k.
   - Various bugfixes.
 o Many minor interface and implementation improvements.
 o Many other bug fixes and cleanups.

 o Added configure-time switches to allow the building of static and/or dynamic
   libraries. (GNU/Linux only)
 o Added the ability to disable Fortran-77-based underscoring and LDFLAGS
   queries at configure-time, which allows one to build libflame in an
   environment that has absolutely no Fortran compiler.
 o Added a configure-time option to enable SuperMatrix visualization, which
   outputs DAG information which may be fed into graphviz.
 o Added the ability for the user to specify a supplementary set of compiler
   FLAGS. These flags get added to those that are automatically determined by
   configure.

4.0 | r3815 | 20100213
----------------------
 o Implemented blocked algorithm and algorithm-by-blocks for new
   up-and-downdating operation, which updates the Cholesky/QR factor used in
   solving a linear least-squares system.
 o Implemented algorithm-by-blocks for LU with partial pivoting.
 o Added advanced scheduling options to SuperMatrix runtime system.
 o Implemented new BLAS-like interfaces (BLIS) layer.
 o Implemented row-major support in BLIS layer for all operations except her2k
   and syr2k.
 o Added optimized unblocked routines written in C for the following operations
   using the new BLIS layer to avoid LAPACK/Fortran dependence.
   - lu_nopiv
   - lu_piv
   - qrut
   - qr2ut
   - lqut
   - uddateut
   - chol (l;u)
   - ttmm (l;u)
   - trinv (ln;lu;un;uu)
   - sylv (nn;nh;hn;hh)
   - aphudut (lh)
   - aph2ut (lh;rn;rh)
   - accumtut (fc;fr)
 o Implemented row-major support in all LAPACK-level operations, leveraging
   the new row-major support in the BLIS.
 o Removed all dependencies on LAPACK.
 o Removed all dependnecies on Fortran.
 o Added _solve() routines for the following operations:
   - chol
   - lu_piv
   - lu_incpiv
   - qrut
   - lqut
   - qrutinc
 o Renamed the following operations:
   - QR_UT_UD      -> QR2_UT
   - Apply_Q_UT_UD -> Apply_Q2_UT
   - Apply_H_UT    -> Apply_H2_UT
 o Modified hemv, hemvc, her, herc, her2, her2c, hemm, herk, her2k wrappers so
   that their real symmetric brethren are invoked if the arguments are real.
 o Reimplemented blocked QR and LQ algorithmic variants in terms of Apply_Q_UT
   operation, resulting in greatly simplified implementations.
 o Added algorithmic variants for:
   - axpyt
   - copyt
   - gemv
   - trsv
 o Created new global scalar constants:
   - FLA_ONE_HALF
   - FLA_MINUS_ONE_HALF
   - FLA_MINUS_TWO
 o Added routines to recover tau vector from T matrix stored by QR_UT and LQ_UT
   functions. This allows interoperability with LAPACK's QR/LQ interfaces and
   is used in the new lapack2flame layer.
 o Rewrote lapack2flame files in C and adjusted build system to bundle
   the object files with libflame rather than put them in a separate library.
 o Changed many signed ints that did not need to be signed to unsigned ints,
   mostly in the form of a new type, dim_t.
 o Added safeguards and error checking into partition/repartition routines.
 o Adjusted safeguards and error checking for several other routines.
 o Defined and implemented three error checking levels (full, minimal, and none)
   where the runtime default can be set via a new configure option.
 o Allow the user to enable/disable the memory leak counter at runtime.
 o Provide options for disabling/enabling:
   - lapack2flame compatibility layer
   - use of external LAPACK implementations via external wrappers
   - use of external LAPACK implementations for subproblems within blocked
     algorithms and algorithms-by-blocks
   via new configure options, as well as checks that prevent undefined and
   unsupported combinations of these options.
 o Added revision number to libflame book cover.
 o Many miscellaneous bug fixes.

3.0 | r2991 | 20090501
----------------------
 o Added LaTeX source for libflame reference guide.
 o Added a build system for use with Microsoft Windows operating systems.
 o Improved SuperMatrix abstractions and performance.
 o Changed control trees and internal back-ends to support arbitrary
   depth hierarchical matrices for many operations.
 o Changed build system to use revision number instead of version number
   when naming build products.
 o Implemented much more robust parameter checking for most routines.
 o Added support for new BLAS-like operations:
   - axpys
   - axpyt
   - copyr
   - copyt
   - dotc
   - dots
   - dotcs
   - dot2s
   - dot2cs
   - inv_scalc
   - scalc
   - scalr
   - swapt
   - gemvc
   - gerc
   - hemvc
   - herc
   - her2c
   - trmvsx
   - trsvsx
   - trmmsx
   - trsmsx
 o Added front-end wrappers for all external wrappers to level-1 and
   level-2 BLAS routines.
 o Added complex datatype support to qrut and lqut operations.
 o Added algorithm-by-blocks implementations for LAPACK-level operations:
   - luincpiv: LU factorization with incremental pivoting.
   - qrutinc: incremental QR factorization via the UT transform.
   - apqutinc: Apply Q^T (or Q^H) obtained via qrutinc to a rhs matrix.
 o Added helper routines to aid the user in creating the block Householder
   transform matrix T for qrut, lqut, and qrutinc.
 o Changed many function names and interfaces to be more clear.
 o Added many new query/utility functions.
 o Re-implemented many utility functions.
 o Removed isolated instances where libflame relied on C99 behavior to allow
   compatibility with C89 compilers.
 o Many miscellaneous bug fixes.
 o Removed reduction to upper Hessenberg form implementation (due to IP
   concerns).
 o Removed implementations for classic QR and LQ factorizations based on
   LAPACK-style Householder transforms.
 o Removed plapack2e code.
 o Removed experimental (and non-functional) Cell support.

2.0 | r1754 | 20080401
----------------------
 o Integrated FLASH and SuperMatrix with FLAME/C variants via control trees.
 o Improved SuperMatrix abstractions and performance.
 o Added POSIX threads support in SuperMatrix.
 o New and expanded interfaces for FLASH along with improved, easier-to-read
   implementations.
 o Added new and/or extended support for the following operations:
   - lu_nopiv: LU factorization without pivoting
   - chol: Cholesky factorization
   - ttmm: triangular transpose matrix multiply
   - trinv: triangular matrix inversion
   - spdinv: symmetric/Hermitian positive definite matrix inversion
   - sylv: triangular Sylvester equation solver
 o Implemented remaining unblocked FLAME/C variants for level-3 BLAS
   operations.
 o Added blocked in-place transpose implementation.
 o New error- and parameter-checking implementations.
 o Modified and improved various utility scripts.
 o Fixed a prominent 32-bit integer bug, allowing the code to malloc()
   regions of memory greater than 2GB, or, double-precision matrices larger
   than 16384-by-16384.
 o Many other bugfixes and cleanups.
 o Added the option of disabling non-critical FLAME code.
 o Added the option of compiling and including into libflame various netlib
   implementations of the files that are need for external wrappers to
   LAPACK-level operations. This is mostly useful because many FLAME
   implementations invoke wrappers to unblocked codes for the smaller
   subproblems. This applies for most LAPACK-level operations, such as
   Cholesky, LU, LQ, QR, triangular inversion, and the triangular Sylvester
   equation solver.
 o Added the option of disabling control trees in level-3 BLAS front-end.
 o Added the option of aligning memory to arbitrary base-2 boundaries via
   posix_memalign().
 o Added the option of aligning each column of a matrix to base-2 boundaries.
 o Added the option of interfacing to CBLAS in the external wrappers.
 o Removed lots of outdated and unused build system cruft.
 o Renamed "parameter checking" option to "internal error checking".
 o Updated configure to use gfortran over g77, if present.
 o Combined libflame-base.a, libflame-blas.a, and libflame-lapack.a into a
   single library archive, libflame.a. (liblapack2flame.a is still a separate
   library due to the potential for linker symbol conflicts.)
 o Produce build products in completely separate directories, allowing
   builds for multiple architectures to be maintained simultaneously with the
   same source tree.
 o Updated config.guess and config.sub scripts (previously circa 2004).

1.0 | r1307 | 20070401
----------------------
 o Additions, improvements, and cleanups to QR, LQ, QR_UT, LQ_UT, Triangular
   Inversion, and Hessenberg Reduction implementations.
 o More complex support throughout libflame, including within utility routines.
 o Improved error checking for some operations.
 o Improved leaf-level test drivers, providing easier datatype switching.
 o New test drivers for front-end interfaces in src/blas/3 and src/lapack.
 o Support for interfacing into internal libgoto blocksize symbols.
 o Added non-double datatype lapack2flamec wrappers for Cholesky.
 o Added non-double datatype lapack2flamec wrappers for LU with pivoting.
 o Added single precision lapack2flamec wrappers for QR and LQ.
 o Added full datatype support for ?trtri() lapack2flame wrappers.
 o Added several netlib LAPACK files that depend on operations for which we have
   lapack2flamec wrappers.
 o Experimental SuperMatrix code is now included.
 o Further PLAPACK2e code development.
 o Improved misc. utility scripts, build tools, and configure script.
 o Added a basic doxygen config file capable of mapping out the source tree and 
   providing output conveniently in HTML format.
 o Many bugfixes and cleanups.

0.9 | r897 | 20060901
----------------------
 o Initial release.