GitLab

experimanetal: improved Taylor series approx by copying LUT to shared mem.
               TODO: cleanup, expose sin/cos method on plan creation interface,
                     do proper calculation of available shared mem for sin cos LUT

experimental: -added alternative method for twiddle factor calc, using a smaller LUT (256 * float2 )
               via Taylor series to 3rd order, seems to be almost as accurate as method with 2 bigger LUTs, but faster.
              -improved method w/ 2 bigger LUTs to use LUTs of float2
              -improved method using slow sin/cos functions (now uses sincos combined function), still slow
              - preparaed plan struct to have method switchable at plan creation time.

              TODO: load smaller LUT for Taylor series approx into shared mem.

Still experimental: replace calls to native_sin in clFFT
This change explores the performance impacts of using a set of LUTs, precomputed on the CPU
to perform sin(x_i) and cos(x_i) in a grid x_i= +/- 2*pi *i/N , N fixed.

On a 6770M, this code is still ca 3% slower than the original native_sin/native_cos varaint
for a BRP4-like transform

This variant should have a very high accuracy, versions with lesser accuracy but
higher performance should be explored next. Eventually the method should be selectable
by a parameter to the plan creator as suggested by Bernd.

TODO: - remove some diagnostic code,
      - optimze total size of LUTs perhaps by using
        cos(x) = sin(x+pi/2), so no need to keep separate LUTs for sin and cos, just one slighly longer with
        an additional alias pointer
      - try caching the LUTs in shared memory (using constant memory didn't help)

* Added dedicated Makefiles
* Added support for Mac OS 10.6
* TODO: proper Linux and Windows support
* Note: we need to add licensing headers!

* Based on Apple's OpenCL FFT implementation
* Derivative work done by Gaurav Khanna