Update on VASST - PowerPoint PPT Presentation

1 / 11
About This Presentation
Title:

Update on VASST

Description:

VTK/ITK have been built on it using gcc/icc. Gcc build is straight forward as ... Tested the Volumizer sample programs and found a bug for specular shading. ... – PowerPoint PPT presentation

Number of Views:43
Avg rating:3.0/5.0
Slides: 12
Provided by: imaging1
Category:
Tags: vasst | specular | update

less

Transcript and Presenter's Notes

Title: Update on VASST


1
Update on VASST
  • Meant for computation intensive tasks also good
    for applications deal with huge date set.
  • GCC/ICC installed on it.
  • VTK/ITK have been built on it using gcc/icc.
  • Gcc build is straight forward as other linux
    build when using cmake. When build using icc, use
    O2 otherwise causes compilation errors.
  • So far we do not have a job scheduling system so
    be careful on how many processors you use.

2
VTK/ITK Quality dashboard
  • It is a testing system used to run tests and post
    results on the VTK/ITK quality dashboard. We
    added our Altix/icc configuration into the
    dashboard to let developers be aware of any
    problem specific for this configuration.
  • VTK dashboard for Altix/icc has been up for
    several months.
  • ITK dashboard for Altix/icc was just started.

3
Multi-threading in VTK/ITK
  • Fact most imaging filters in VTK/ITK are
    multi-threaded. Look for vtkImageThreadedAlgorithm
    (VTK) and itkImageSource(ITK)
  • The number of threads is set the number of
    processors by default
  • We are looking at image registration algorithm
    and trying to multi-thread some of them.
  • Example itkMutualInformationImageToImageMetric
  • Thread safety VXL uses netlib and it is not
    clear if it is thread safe so far.

4
VTK/ITK threading model
algorithm
cpu1
output
Input volume
cpu2
5
Other multi-threading techniques
  • VTK/ITK mutli-threading techniques are based on
    Pthread model.
  • OpenMP is another way for multi-threading. It is
    only available in icc, not gcc.
  • Message Passing Interface(MPI) is meant for
    distributed computing. Altix is a shared memory
    architecture. But there is a SGI MPI API on it
    called MPT.

6
Computation library
  • Freely available lib netlib
  • www.netlib.orgwww.netlib.org
  • SGI computation library SCSL
  • Signal processing routines FFT Convolution
    Correlation.
  • Direct linear equation solvers
  • Lapack and blas

7
Prism graphics system
  • Hardware 8 Intel Itanium CPUs 4 ATI Fire GL
    graphics boards image compositor.
  • Software SGI advanced Linux SGI OpenGL
    Multipipe environment ( with SGI image compositor
    ) SGI OpenGL volumizer2 API.
  • Difference between vasst and prism Meant to be a
    graphics system, similar hardware and software
    environment. Basically programs developed for
    vasst can be run directly on prism.

8
OpenGL Multipipe Environment
9
SGI OpenGL Volumizer API
  • Uses 3D texture mapping technique for volume
    rendering
  • Currently I am working on integration of it with
    VTK.
  • Tested the Volumizer sample programs and found a
    bug for specular shading. SGI just confirmed this
    bug and said it will be fixed in their next
    release.
  • Other volume rendering methods ray casting(
    multi-threaded) shear-warp( not working) 2D
    texture mapping.

10
Documentations
  • Intel compilers
  • /opt/intel_cc_80/doc/
  • SGI techpub library(http//techpubs.sgi.com)
  • SCSL
  • OpenGL Multipipe
  • OpenGL Volumizer
  • www.imaging.robarts.ca/kwang

11
Questions?
  • You are welcome to use vasst for computation
    intensive task and prism for visualization task.
  • Think your algorithm might be able to take
    advantage of multi-theading?
Write a Comment
User Comments (0)
About PowerShow.com