Making the 2x2 and 3x3 Eigen stuff more consistent.

  Code cleanup

Updated the cl-target kernel directory and files to map naming conventions 

for symmetric 2x2 eigensolve

Updating the OpenCL backend to include update method template with placeholders 

removing mac-specific line endings

Adding the cl_kernels to the lib directory and the update kernel. 

Bug Fix: After the previous commit, the output was not correct for the OpenCL backend because it uses the shadow strand size rather than the original strand size. Fixed 

  Factored out common part of Diderot_World_t struct (see Diderot/world.h) and
  fixed output routines for the C and Parallel targets.

  Deal with the different sizes of status flags statically, instead of dynamically.

  Fix compiler warnings

Update includes a working version of the OpenCL backend and output support. Upated the output function args to include a size_t output argument for the output data. 

  Another indexing fix

  Fix vector indexing syntax for gcc.

  Port eigen code to Diderot runtime

Fixed the bug where the strand blocks were not copied from __local to __global data. All regression tests are passing except sum-ghp and vr-taxi

-This line, and those below, will be ignored--

M    pure-cfg/src/include/cl_kernels/kernels.cl
M    pure-cfg/src/include/Diderot/cl-scheduler.h
M    pure-cfg/src/lib/cl-target/main.c
M    pure-cfg/src/compiler/cl-target/cl-target.sml
M    pure-cfg/src/compiler/cl-target/runtime-names.sml

Fixed the compaction algorithm for phase 2 and create clEnqueueTasks for reseting the scheduler meta data between iterations. 
-This line, and those below, will be ignored--

M    pure-cfg/src/include/cl_kernels/kernels.cl
M    pure-cfg/src/include/Diderot/cl-scheduler.h
M    pure-cfg/src/lib/cl-target/main.c
M    pure-cfg/src/compiler/cl-target/cl-target.sml

  Add calls to stabilize method to sequential and parallel runtimes

  style fixes

Removed unwanted debugging code

Removed a clfinish statement 

Added some comments to the files 

Updating files with the new persistent scheduler 

  minor code cleanup

  Switched parallel runtime to use Diderot_Output and Diderot_Print.  Also fixed a bug
  in nrrd output code.

  Split output code into two functions: one for nrrd files and one for text files.

  Added comments and fixed indentation

Moved output proceedure to seperate file and updated c-target to use new output.

  More runtime system name changes

  Determine workgroup and local-memory size of kernels.

  Switch to using checked allocation function; added cuWidth info to OpenCL target.

  Fixed coding conventions and added a FIXME comment about the global work size

OpenCL scheduler without many-to-many mapping. 

  code formatting fixes

Added a comment and indexing fix for the regression test 

Added Support for single dimension work queue

  Fix to how shadow types are handled

Added Support for converting host strands to its shadow types 

Fixed the remaining errors with the OpenCL scheduler code 

Updated the OpenCL runtime system to include the scheduler implementation

Reverting last change to cl-target/main.c so that it will compile again.

  minor formatting changes

Nrrd output.  Use with the -nrrd command line option.

  Working on stabilize support

  Get more information about devce

  add "-w" flag to disable warnings in OpenCL compiler

  minor cleanup

  Removed unnecessary clFinish calls and made printing more like C version.

Added support for 1D and 3D access for OpenCL

  Switch to using DIDEROT_CL_VERSION instead of __OPENCL_VERSION__.

  Minor OpenCL runtime improvements

  Get the OpenCL version number supported by the device

  minor tweaks to runtime

Removed duplicate host vector to shadow vector code

Fixed the bug with converting global fields to its shadow global fields

Removed embedded pointers on both the host and gpu side

  More OpenCL info

  Improved OpenCL info and added mechanism to print the OpenCL profile

  minor tweaks

  Get more device info.

  fix calculation of image size in bytes

  Working on OpenCL support

  Add shadow types for marshalling data to the GPU

 Fix CFLAGS in configuration

  Added functions to shadow vectors and matrices

  Added shadow types for shadowing globals that are to be copied to the client.

  misc

Removed buildinfo log from runtime and fixed strand init on GPU side 

Added barrier synch to .cl file and fixed more address space bugs and upated the print function for OpenCL

  Fixing indentation etc.

Fixed more address space bugs in OpenCL

Fixed the bug with Diderot_KernelMain paramters

  Fixed indentation and minor tweaks

Fixed problem with local work size

Fixed problems with OpenCL generation code to match with headers

OpenCL runtime now compiling correctly and contains hookups

  Minor tweaks

  Working on OpenCL support

  Working on OpenCL runtime

  Added name to world structure in OpenCL runtime.  Also added more error
  checking to Diderot_LoadSource.

  OpenCL hacking

  Use the program name as the default output file name.

  Fix bug in the stabilization of strands

  Added timing support and fixed a build typo

  Fixed unclosed comment

  Added target defines to CPPFLAGS

  Working on OpenCL support

Commiting OpenCL implementation does not include Diderot library functions support

  Working on OpenCL runtime support.  Also, added configure support for generating OpenCL
  runtime makefiles.

  expand tabs to avoid editor differences

  Start of runtime support for OpenCL

  Adding support for command-line arguments

  Working on command-line option support

adding more comments, and cosmetic tweaks

incomplete commit of the previous bugfix

fixed hest-related memory bug triggered anytime that struct Diderot_Options_s's optDef_t *odef array had to be re-allocated. Basically: you can't give hest a pointer (to store a variable in) that points into an array that might be reallocated later, since the reallocation will end up at a different address

working implementation of hest wrappers for command-line arguments

  Reclaim space for option descriptors

  Added hest wrappers

  Split input functions into their own file and added Diderot_InputVec2

  Use atomic add to update numActive

  Working on support for parallel runtime

  Working on parallel runtime

  Use double precision to compute transforms for images.  We then convert these to the
  target precision.

  Fix cleanup

  Working on parallel runtime

  Add parallel runtime support

  Port eignevalue stuff to diderot runtime

  Working on porting eigenvector code to Diderot runtime conventions

  Working on multiple-target support

  Better cleanup

  Runtime system updates to support Die and to eliminate race for stabilizing

  Fix syntax error

  Some runtime system changes

  One more -m64 fix

  More work on runtime build infrastructure

  Working on new runtime build infrastructure

  After strand initialization, we copy the initial state into the other state
  object so that invariant state variables have the correct value in both
  copies.

  Reworked representation of coordinate-space transforms in image objects

  Fixed typo in Makefile.in

  Tweak error checking on image files

  Be more tolerant when checking the kinds of the axes.

  Added some consistency checking to image loading routines.

  More adjustments for non-scalar images

  bug fix to runtime: account for voxel dimensions when loading transform matrices

  Simplify matrix loading code

  Adding support for loading 1D image data

  Update INSTALL instructions

  Adding 3x3 matrix and 2D image support

  Added -m64 flag to C compiles and fixed some warnings.

  Added support for tracking output state variables through ILs and generating
  a print function in the C target.

  print range of results for debugging

  bug fix: out and in were backwards

  Switch to compiler-generated initially code.

  Added runtime interface to support creation of initial array/collection
  of strands

  Don't map negative values to 0

  modified debug printing

  Minor code cleanup/fixes

  Fixing bugs in runtime code

  Debugging vr-MIP code

  Working on vr-MIP example

  Add support for building runtime library to top-level makefile

  Working on Diderot runtime

  Working on build scripts for Diderot test programs

  Move diderot.h header to include/Diderot

  Minor edit

  Working on header

  Creating a branch to explore a pure CFG IL

  Added comments about SSE

  Working on C version of MIP

  Add Diderot_Mat types

  added InvertTranspose operation

  Bug fix: change + to *

  Working on runtime

  More runtime work

  Working on Diderot runtime

  Working on C version of runtime