Skip to content

Latest commit

 

History

History
227 lines (165 loc) · 5.51 KB

NEWS.md

File metadata and controls

227 lines (165 loc) · 5.51 KB
Raw

NEWS

Release 0.4-2 (1/9/2022):

New:

  • Added stats::cov() and stats::cor() methods.
  • Added set_row() and set_col() methods for matrix classes.

API Changes:

  • Added fill_linspace() method with no args passed.

Bug Fixes:

  • None

Internal Changes of Note:

  • Greatly modified linalg internals for each backend (does not affect usage of backend/linalg.hh headers)
    • dropped linalg_ preface in file names
    • moved some internal headers to internals/
    • exported blas header to individual files (e.g. xpose.hh)
  • More granular header usage among internals

Documentation:

  • None

Release 0.4-1 (11/15/2021):

New:

  • Added set_math_mode() method for card objects
  • Added pow() method for cpuvec and gpuvec objects
  • Added dimops::rowsweep() and dimops::colsweep()

API Changes:

  • None

Bug Fixes:

  • Fixed linalg::rsvd() issues with mpimat.

Internal Changes of Note:

  • None

Documentation:

  • None

Release 0.4-0 (8/23/2020):

New:

  • Added linalg::dot()
  • Added linalg::trinv()
  • Added linalg::rsvd()
  • Added par utils get_device_count() and get_device_num()
  • Many parmat changes (still experimental).

API Changes:

  • linalg::matmult() can now accept mixed matrix/vector arguments.

Bug Fixes:

  • Changed gpumat crossprod() and tcrossprod() to use Xgemm() instead of Xsyrk() for better run-time performance.

Internal Changes of Note:

  • Internal linalg headers have been re-organized. This has no effect if you use the linalg.hh headers or the main type headers (cpu.hh, gpu.hh, mpi.hh).
  • Switched linalg::qr_Q() and linalg::lq_Q() to use Xorgqr()/Xorglq() except in gpumat lq_Q().
  • Changed uplo type in gpu_utils::lacpy() to char.

Documentation:

  • Fixed card synch() method documentation.

Release 0.3-0 (6/20/2020):

New:

  • Added singular backend headers.
    • #include <fml/cpu.hh> includes all standard headers in src/fml/cpu.
    • #include <fml/gpu.hh> includes all standard headers in src/fml/gpu.
    • etc.
  • Added fml::card_sp_t typedef for std::shared_ptr<fml::card> in card.hh.
  • Added min/max methods for cpuvec and gpuvec.
  • Added to linalg namespace for cpumat, gpumat, and mpimat:
    • norm_1()
    • norm_I()
    • norm_F()
    • norm_M()
  • More OpenMP usage in CPU and MPI backends.

API Changes:

  • All headers are now contained in the fml/ tree.
    • src/cpu/cpumat.hh becomes src/fml/cpu/cpumat.hh
    • src/gpu/gpumat.hh becomes src/fml/gpu/gpumat.hh
    • etc.
  • All classes and namespaces are now in the fml namespace.
    • cpumat<float> becomes fml::cpumat<float>
    • linalg::svd() becomes fml::linalg::svd()
    • etc.
  • All Xhelpers namespaces are renamed to the singular copy namespace.
    • cpuhelpers::cpu2cpu() becomes fml::copy::cpu2cpu()
    • gpuhelpers::cpu2gpu() becomes fml::copy::cpu2gpu()
    • etc.
  • Rename linalg::tssvd() to linalg::qrsvd(), and added LQ case when m<n.
  • Added support for m<n case (via a transpose) of gpumat linalg::svd().

Bug Fixes:

  • Fixed an indexing bug in mpimat linalg::cpsvd() when m>n.

Documentation:

  • Fixed some formatting issues in generated documentation.

Release 0.2-1 (5/28/20):

New: None

API Changes: None

Bug Fixes:

  • Fixed det() bug for cpumat.
  • Expanded cpu2mpi() to accept objects of different fundamental types.
  • Fixed indexing bug in mpi2cpu().
  • Fixed mixed type bug in mpi2cpu().
  • Fixed 0 value ownership issue with mpimat.
  • Fixed 0 value ownership issue with invert() for mpimat.

Documentation: None

Release 0.2-0 (5/20/20):

New:

  • Added to linalg namespace for cpumat, gpumat, and mpimat:
    • qr()
    • qr_Q()
    • qr_R()
    • lq()
    • lq_L()
    • lq_Q()
    • tssvd()
    • cpsvd()
    • det()
    • chol()
  • Created dimops namespace
  • Added to dimops namespace for cpumat, gpumat, and mpimat:
    • rowsums()
    • rowmeans()
    • colsums()
    • colmeans()
    • scale()
  • Created stats namespace
  • Added to stats namespace for cpumat, gpumat, and mpimat:
    • pca()

API Changes:

  • Added default argument to gpuhelpers::new_card()

Bug Fixes:

  • Fixed major resize bug in mpimat not updating descriptor arrays.
  • Changed matrix object == tests to use arraytools::fltcmp::eq() instead of == on fundamental types.
  • Changed seed setter behavior in mpimat rng functions: local seed is now computed as seed + myrow + nprow*mycol.
  • Fixed several memory issues in gpumat linalg::svd().
  • Fixed an internal logic error in gpumat linalg::svd().

Documentation:

  • Minor corrections; no major changes.

Release 0.1-0 (2/5/2020):

New:

  • Created cpumat and cpuvec classes
  • Created gpumat and gpuvec classes
  • Created mpimat class
  • Created parmat_cpu and parmat_gpu classes
  • Created linalg namespace
  • Added to linalg namespace for cpumat, gpumat, and mpimat:
    • add()
    • matmult()
    • crossprod()
    • tcrossprod()
    • xpose()
    • lu()
    • svd()
    • eigen_sym()
    • invert()
    • solve()
  • Added to linalg namespace for parmat:
    • crossprod()
  • Created cpuhelpers namespace
  • Added to cpuhelpers namespace
    • cpu2cpu()
  • Created gpuhelpers namespace
  • Added to gpuhelpers namespace
    • gpu2cpu()
    • cpu2gpu()
    • gpu2gpu()
  • Created mpihelpers namespace
  • Added to mpihelpers namespace
    • mpi2cpu()
    • cpu2mpi()
    • mpi2mpi()

API Changes: None

Bug Fixes: None

Documentation:

  • Added documentation for most things.