Skip to content

Latest commit

 

History

History
150 lines (109 loc) · 3.56 KB

NEWS.md

File metadata and controls

150 lines (109 loc) · 3.56 KB

Release 0.4-0 (1/9/2022)

New:

  • Moved fml headers to separate fmlh package.
  • Updated to fmlh 0.4-1:
    • Added linalg_dot()
    • Added linalg_trinv()
    • Added linalg_rsvd()
    • Added pow() methods for cpuvec and gpuvec
    • Added dimops_rowsweep() and dimops_colsweep()

API Changes:

  • linalg_matmult() can now accept mixed matrix/vector arguments.
  • Renamed linalg_tssvd() to linalg_qrsvd().

Bug Fixes:

  • Fixed gpu/mpi bug on norm/cond dispatch.

Documentation: None

Internal Changes of Note from Latest fml Release:

  • Changed gpumat crossprod() and tcrossprod() to use Xgemm() instead of Xsyrk() for better run-time performance.
  • Switched linalg::qr_Q() and linalg::lq_Q() to use Xorgqr()/Xorglq() except in gpumat lq_Q().

Release 0.3-0 (9/6/2020)

New:

  • Updated to fml v. 0.3-0:
    • Added min/max methods for cpuvec and gpuvec.
    • Added norm functions linalg_norm_1(), linalg_norm_I(), linalg_norm_F(), linalg_norm_M(), and linalg_norm_2().
    • Added condition number functions linalg_cond_1(), linalg_cond_I(), and linalg_cond_2().

API Changes: None

Bug Fixes:

  • See fml 0.3-0 release.

Documentation: None

Release 0.2-1-1 (9/2/2020)

New:

  • Building with MPI backend support is now optional, and disabled by default. Enable by adding --enable-mpi to the package configure args (Mac, Linux), or modifying Makevars.win appropriately. This is explained in-depth in the installation guide.
  • Fixed installation issues for Windows.
  • GPU backend now available for Windows.
  • Added more benchmarks (inst/benchmarks/).
  • Added synch() method for card objects to force a synchronization.

API Changes: None

Bug Fixes:

  • Safer .NAME usage in .Call() internals.
  • Added more exception guards.

Documentation:

  • Clarified some things in the installation guide.
  • Added Windows GPU instructions.

Release 0.2-1 (5/29/2020)

New:

  • Added linalg bindings:
    • linalg_qr()
    • linalg_qr_Q()
    • linalg_qr_R()
    • linalg_lq()
    • linalg_lq_L()
    • linalg_lq_Q()
    • linalg_tssvd()
    • linalg_cpsvd()
    • linalg_chol()
  • Added dimops bindings:
    • dimops_rowsums()
    • dimops_rowmeans()
    • dimops_colsums()
    • dimops_colmeans()
    • dimops_scale()
  • Added stats bindings:
    • stats_pca()
  • Added mpihelpers bindings:
    • cpu2mpi()

API Changes: None

Bug Fixes:

  • Added Makevars.win.
  • Updated cpu inherit() methods and all class from_robj() methods (and correspondingly, as_*() methods) to accept double, int, and float data.
  • as_*() functions now choose return type based on the R object input.

Documentation:

  • Added vignettes:
    • installation
    • basic overview
    • backend management
    • data management
  • Added vignettes to pkgdown docs.

Release 0.1-0 (2/5/2020)

New:

  • Added bindings for cpumat and cpuvec classes
  • Added bindings for gpumat and gpuvec classes
  • Added bindings for mpimat class
  • Added linalg bindings:
    • linalg_add()
    • linalg_matmult()
    • linalg_crossprod()
    • linalg_tcrossprod()
    • linalg_xpose()
    • linalg_lu()
    • linalg_svd()
    • linalg_eigen_sym()
    • linalg_invert()
    • linalg_solve()
  • Added cpuhelpers bindings:
    • cpu2cpu()
  • Added gpuhelpers bindings:
    • gpu2cpu()
    • cpu2gpu()
    • gpu2gpu()
  • Added mpihelpers bindings:
    • mpi2cpu()
    • mpi2mpi()

API Changes: None

Bug Fixes: None

Documentation:

  • Added documentation for most things.