GFFT Files

Release notes

-------------------------------------------------------------------------------------------
Version 0.32 (2015-04-16)

    - Improved single- and multi-theaded performance (OpenMP)
    - CMake files contain support for GCC, Clang, Intel C++ and MSVC
    - Examples of compile-time computations of pi (metapi.cpp) and square root (metasqrt.cpp)
    - Fixed accuracy bug in single precision FFTs

-------------------------------------------------------------------------------------------
Version 0.31 (2014-02-22)

Accuracy and performance benchmarks added in the subfolder 'test'

-------------------------------------------------------------------------------------------
Version 0.3 (2014-02-07)

The project is back after 5 years break with a couple of new powerful features:

    - Out-of-place transforms
    - Complex transforms of any length N
    - Compile-time factorization of the input length N
    - Better accuracy due to compile-time computation of the roots-of-unity as rationals

-------------------------------------------------------------------------------------------
Version 0.2 (2009-12-28)

Completely redesigned release to make the usage even more simple and 
to implement many new features in the future.

    - New template class GenerateTransform to define a set of transforms.
    - New notation for different transform types taken from SPIRAL project. 
    - The following transforms in-place are now implemented: DFT, IDFT, RDFT, IRDFT
    - Reordered template parameters for the transforms definitions, 
      where most important parameters go first and less important parameters 
      may be omitted using defaults.
    - Multithreaded parallelization using OpenMP standard
    - Bug fixes

-------------------------------------------------------------------------------------------
Version 0.11 (2008-07-04)

Bug-fixig release

-------------------------------------------------------------------------------------------
Version 0.1 (2008-02-22)

Initial release of GFFT with the following features:

    - Transforms in-place
    - Complex FFT of power-of-two length
    - Real-valued FFT of power-of-two length
    - High and cache-independent performance
    - No additional data is stored.
    - You can use all available RAM for your transform data
    - One-step transform.
      Many known FFT implementation perform to steps: initialization and transform. 
      Initialization for a given length is usually computationally expensive, 
      while transform is fast. GFFT needs only to create an object instance 
      that includes FFT-algorithm, but no additional data or computation.