Skip to content

Releases: ctkelley/MultiPrecisionArrays.jl

v0.1.3-joss

27 Sep 17:56
6661fbb
Compare
Choose a tag to compare

Release for JOSS paper

v0.1.3

27 Sep 13:33
9f71191
Compare
Choose a tag to compare

MultiPrecisionArrays v0.1.3

Diff since v0.1.2

Update termination criteria. Fix performance bug. JOSS version.

Merged pull requests:

Closed issues:

  • JOSS Review Comments (#51)
  • JOSS references (#60)

v0.1.2

10 Jul 01:04
c7c2c46
Compare
Choose a tag to compare

MultiPrecisionArrays v0.1.2

Diff since v0.1.1

Version 0.1.2 adds support for residuals in high precision within the IR-Krylov functions.

Merged pull requests:

v0.1.1

19 Jun 15:50
55179ea
Compare
Choose a tag to compare

MultiPrecisionArrays v0.1.1

Diff since v0.1.0

Version .0.1.1 updates the termination criterion to normwise backward error.

Merged pull requests:

v0.1.0

22 Mar 19:03
Compare
Choose a tag to compare

MultiPrecisionArrays v0.1.0

Diff since v0.0.9

This update to version 0.1.0 has better docs with several changes.

  • I no longer export the constructors and the MPArray factorizations. You should only be using mplu, mplu!, mpglu, mpglu!, ...

  • Notation and variable name change to conform with standard practice (TH --> TW for working precision, TL --> TF for factorization precision etc). If you just use the multiprecision factorizations with no options, you will not notice this.

  • Explanation for why I am not excited about evaluating the residual in extended precision + a bit of support for that anyhow

  • Replacing Polyester in the Float16 LU with OhMyThreads v0.5 or later.

Merged pull requests:

  • CompatHelper: add new compat entry for OhMyThreads at version 0.4, (keep existing compat) (#9) (@github-actions[bot])
  • CompatHelper: add new compat entry for FLoops at version 0.2, (keep existing compat) (#10) (@github-actions[bot])

Update docs

20 Jan 06:18
ab19436
Compare
Choose a tag to compare

Fixing the Documenter problem

v0.0.9

20 Jan 04:39
Compare
Choose a tag to compare

MultiPrecisionArrays v0.0.9

Diff since v0.0.8

This is v0.0.9 with more updates, better organization, and BiCGSTAB-IR.

v0.0.8

10 Dec 19:49
Compare
Choose a tag to compare

MultiPrecisionArrays v0.0.8

Diff since v0.0.7

New support for updating multiprecision factorizations in place + better docs

First registered release

24 Nov 22:14
958c53c
Compare
Choose a tag to compare

This package provides data structures and solvers for several variants of iterative refinement (IR). It will become much more useful when half precision (aka Float16) is fully supported in LAPACK/BLAS. For now, its only general-purpose application is classical iterative refinement with double precision equations and single precision factorizations.

IR is a perfect example of a storage/time tradeoff. To solve a linear system $A x = b$ in $R^N$ with IR, one incurs the storage penalty of making a low precision copy of $A$ and reaps the benefit of only having to factor the low precision copy.

MultiPrecisionArrays v0.0.6

14 Oct 17:20
Compare
Choose a tag to compare

This version has simpler data structures and better documentation.