README

computeFTLE_CUDA is a small little package that implements
basic FTLE and fluid dynamics simulations on nVidia GPUs
via CUDA.

Please read INSTALL and BUGS!

The following interesting tidbits are included:
-------------------------------------------------------

- src/darwin-gpu-global.cu is a simple rigid body fluid demonstration
  that shows a base implementation for CUDA. Accuracy not guaranteed.

- computeFTLE_CUDA computes an FTLE field for an explicit
  vector flow function given by the function vec in vec.c.
  This program has the ability to output PNG plots of the
  FTLE field in conjunction with PLplot. Please modify the
  Makefile as well as settings.h to enable/disable this option.

  The current vectorfield is given in vectorfield.cu; out-of-the
  box the vector field is a time-dependent double-gyre:

  http://www.cds.caltech.edu/~shawn/LCS-tutorial/examples.html#Sec7.1


- graph_data is a small utility for plotting FTLE and tracer data.
  Small and lightweight, it outperforms PLPlot by an order of
  magnitude for these specific problems. We recommend using
  graph_data over PLPlot. To use, see TIPS and graph_data --help.


- computeFTLE computes an FTLE field for an explicit
  vector flow function, and spits the output (at the moment)
  to a file ftle_1.data, in the format:

  x	y	FTLE

  See the source to replace the vector function or to change
  t/T/solver parameters. This code is currently obsolete;
  please compile computeFTLE_CUDA in CUDA emulation mode for
  the whole suite of features if you don't have a GPU.

- examples/ has a bunch of pictures created using 
  computeCUDA_FTLE. If interested in specifics, please
  email Raymond Jimenez <raymondj@caltech.edu>.

Guide for use:
-------------------------------------------------------

  All of the above have user-servicable parts inside, please open!

  The easiest way to get computeFTLE_CUDA working for your specific
demonstration is to:

1) Open up src/vectorfield.cu and replace vec() and vec_mask() with
   your flow equations (look at the examples to see how this should work;
   it's easiest with a velocity field, but a potential function or
   stream function can be used with slight difficulty as well)
 
2) Modify settings.h to the necessary window and grid size. Note
   that the point-offset used in calculating the FTLE field is
   .001, so a grid that results in more than 500 points per unit
   may be useless.

3) Run computeFTLE_CUDA and generate output tracer/FTLE files.
   
4) Open up src/graph_data.c and modify the vector/tracer masks
   as you see fit.

5) Run graph_data to get the final output PNGs. You may want
   to automate this using a simple batch shell script.

--
Raymond Jimenez <raymondj@caltech.edu>
Joris Vankerschaver <jvk@cds.caltech.edu>