Skip to content

gpuR Roadmap

Jump to bottom Edit New page
Charles Determan edited this page Sep 15, 2017 · 10 revisions

General Roadmap for gpuR development. Subject to changing as needed either to meet pre-determined deadlines or changes in development.

gpuR 1.2.x

Planned Features

  • Add Factorizations
    • QR
    • SVD
    • Cholesky
  • Control over OpenCL context via setContext (i.e. allow use of desired platform/device for each object)
    • This will allow use of parallel functions such as foreach and mclapply
      • Stalled until can find a way to handle XPtr objects in foreach/mclapply calls
  • Try to get rid of BH - only minimally used and is a big package

gpuR 2.0.x

Planned Features

  • Accept custom OpenCL kernels
  • In-Place operations

gpuR 2.1.x

  • Ability to createContext to choose multiple devices for context
  • Use of multiple GPU's on individual objects (within same context) 'automatically'
    • Likely just put in backend (if devices > 1 for object then 'chunk')
    • This will be problematic for factorization/solvers as not implemented by ViennaCL and not a simple process (such as most BLAS operations)

Please submit additional issues on the issues tab.

Add a custom sidebar
Clone this wiki locally