lacaml

CHANGES:

• Fixed platform configuration bug

CHANGES:

• Fixed a bug in the bytecode bindings for lamch, which also affected
  syevr. These functions should now not crash anymore when compiling
  to bytecode or using the OCaml interpreter.

  Thanks to Philippe Veber [email protected] for the bug report!

• Added support for Apple Silicon.

  Thanks to Marcello Seri [email protected] for this contribution!

• Improved Dune build rules

CHANGES:

• Improved detection of (complex) dotu and dotc calling conventions.

CHANGES:

• Removed package dependencies on stdio and base. This allows
  testing with the new OCaml multicore compiler, which as of now does not
  yet support these libraries.

  Thanks to Anthony Scemama for this contribution!

  • Added support for const char strings in stubs due to stricter handling
    in newer OCaml runtimes. This eliminates C-compiler warnings.

CHANGES:

• Switched to OPAM file generation via dune-project

  • Added missing autogenerated dependencies to dune file to avoid
    build system race conditions.

CHANGES:

• Fixed warnings in C-stubs

CHANGES:

• Fixed an API misspecification in ptsv, which leads to incorrect results.

  Thanks to Nicolas Ratier [email protected] for the bug report!

CHANGES:

• Fixed a portability bug in (complex) dotu and dotc due to BLAS API
  incompatibilities between different platforms.

CHANGES:

• Switched to dune, dune-release, and OPAM 2.0

11.0.1 (2018-05-23)

• Fixed warnings and errors in configuration code due to upstream changes.

• Fixed bug in Vec.ssqr.

11.0.0 (2018-02-28)

WARNINIG WARNING WARNING

• User code changes in existing code required! User code may still compile,
  but can behave differently!

  The following functions are affected:

  • potrf
  • potri
  • potrs

  The above functions now do not support the jitter argument anymore.
  Users should remove the flag from calls to the above functions and
  call the new Mat.add_const_diag function if they need to add jitter.
  This call should happen right before the (now required) call to potrf.

  More importantly, potri and potrs now do not support the factorize
  flag anymore, which would call potrf automatically beforehand. This was
  the (ill-conceived) default, which makes it harder to port LAPACK code
  to Lacaml. In order to upgrade your code, please do the following:

  • If potri or potrs were passed ~factorize:false, just remove
    the flag.

  • If potri or potrs were passed ~factorize:true or were called
    without the factorize flag, remove the flag if necessary and call
    potrf with the exactly corresponding arguments before.

  Luckily, these functions are typically used rarely, and the changes
  are trivial. Apologies anyway for the churn!

• New functions

  • Mat.add_const_diag for adding a constant to the diagonal of a
    (sub-)matrix.

• orgqr now detects if m < n and raises an exception instead of printing
  a Fortran error message and continuing.

• Eigenvalue offsets should now work correctly.

• Improved documentation.

10.0.2 (2017-11-08)

• Fixed bugs accessing lower pentagonal matrix patterns

• Fixed library override issue on Mac OS X

10.0.1 (2017-10-21)

• Fixed wrongly capitalized build targets missed due to Mac OS X file
  system case insensitivity.

10.0.0 (2017-10-20)

• Switched to jbuilder and topkg

• API changes

  • trmm and trsm now do not label argument a anymore

  • Many matrix functions now support an optional patt argument,
    which can be used to specify rectangular, triagonal, trapezoidal, and
    pentagonal patterns on which to perform an operation.

  • New functions

    • Mat.sum_prod computes the sum of elemenet-wise products of
      two matrices. Some use cases are already covered by Mat.gemm_trace,
      but the latter does not support patterns.

• Improved C-code to better support SIMD compiler optimizations

• Many internal improvements, including untagged and unboxed passing of
  parameters to and from external functions.

• Compilation now uses -march=native -O3 -ffast-math by default, which
  should be safe and exploit SIMD on platforms that support it to greatly
  improve performance of some operations.

• Improved documentation

• Improved configuration and build process