ITK/Release 4/GPU Acceleration/Tcon-2011-01-06: Difference between revisions

From KitwarePublic
Jump to navigationJump to search
No edit summary
 
(One intermediate revision by one other user not shown)
Line 50: Line 50:


* Using time stamp class, mutex
* Using time stamp class, mutex
* Global repository for GPU source code files and on-the-fly hard-coded headers generation


== Action Items ==
== Action Items ==

Latest revision as of 16:01, 9 December 2011

Tcon Information

Day - Time

  Thursday January 6th, 1pm EST.

Phone Line

  Please join the following teleconference line:
      Toll-Free #:      1-800-704-9804 
      International #:  1-404-920-6604
      Participant Code: 61466276

Web Conference

 http://www.readytalk.com/


 https://cc.readytalk.com/r/ygqhyoco4bml

Attendees

  • Won Ki - Harvard University
  • Jim Miller - Ge
  • Brad Lowekamp - NLM
  • Dennis Sessanna - NVIDIA
  • Kimberly Powell - NVIDIA
  • Luis Ibanez - Kitware Inc.

Topics

Overview

  • Current status of ITK-GPU
    • Design strategy
    • Implementation
    • Usage example
  • GPU code testing strategy

Questions/Comments

  • Asynchronous memory copy support
  • Multithreading support (host)
    • GPU driver will handle time-sharing
  • Using time stamp class, mutex
  • Global repository for GPU source code files and on-the-fly hard-coded headers generation

Action Items

  • itkGPU* for general GPU class, itkOpenCL* for specific GPU implementation
  • Pipelining using ITK factory mechanism
    • Try subclassing BinaryThresholdImageFilter and SigmoidImageFilter so that we can experiment with using the ITK factory mechanism to use the OpenCL versions
  • Inplace filter handling of the GPUImage