CphCT Toolbox NEWS
Last updated/generated 31-08-2016 15:23


  1. CphCT Toolbox News
    1.1. CphCT 1.0.0
    1.2. CphCT 1.0.1
    1.3. CphCT 1.0.2
    1.4. CphCT 1.0.3
    1.5. CphCT 1.0.4
    1.6. CphCT 1.1.0
    1.7. CphCT 1.1.1
    1.8. CphCT 1.1.2
    1.9. CphCT 1.1.3


  1. CphCT Toolbox News
  =====================

This document contains a summary of project news and updates.
The sub sections describe the most important changes and addition for
each public release.


	1.1. CphCT 1.0.0
	================

Our first public release of the toolbox with fan and cone beam
implementations. We provide Numpy and CUDA engines so far, but we are
working on OpenCL engines for the next release.

 - Center Feldkamp-Davis-Kress (CenterFDK) Circular Fan Beam Reconstruction
 - Feldkamp-Davis-Kress (FDK) Circular Cone Beam Reconstruction
 - Katsevich Spiral Cone Beam Reconstruction
 - Basic instructions for using each of the applications


	1.2. CphCT 1.0.1
	================

Improved scalability of GPU apps to allow bigger problem sizes.
Minor bugfixes in e.g. the automatic GPU block layout code.


	1.3. CphCT 1.0.2
	================

Major bugfix in the numpy application initialization and in the
Katsevich GPU block layout code for irregular dimensions.


	1.4. CphCT 1.0.3
	================

Feature update to improve scalability of the CUDA FDK and Katsevich
implementations. 

Previous CUDA Katsevich benchmarks were limited to 
2048x2048x2048 voxels with 360 512x2048 pixel projections per rotation
and a chunk size of at most 64 even on high end GPUs. Now it can scale further
with a smaller GPU memory footprint.

The CUDA FDK implementation previously had increased execution time per
voxel when the chunk size was decreased. Now the execution time per
voxel is nearly constant with only a small overhead introduced as 
the chunk size decreases.


	1.5. CphCT 1.0.4
	================

Bugfix release to address problems with the cuda kernel source files not
getting properly installed when directly using pip or setup.py install
rather than going through distro package building and install.
Additionally adds a workaround to address missing float128 support in
some (32-bit) NumPY installations, thus allowing the toolbox to run
there as well.


	1.6. CphCT 1.1.0
	================

Introduction of OpenCL versions to increase platform support quite a
lot. Reworked some gpu layout choices to remove a number of unnecessary
scalability limitations.


	1.7. CphCT 1.1.1
	================

Bugfix release to properly filter Katsevich projections independently of
the value of gpu-target-threads. 


	1.8. CphCT 1.1.2
	================

Minor bugfix release to address a bug in some of the GPU plugins that
made them crash.


	1.9. CphCT 1.1.3
	================

Minor release to fix an API breakage issue in save plugins with recent PIL(low)
versions. Update references to project home at SourceForge rather than Google
Code.

