Skip to content

Releases: mpicbg-scicomp/gearshifft

v0.4.0

02 Feb 10:51
3b68802
Compare
Choose a tag to compare
v0.4.0 Pre-release
Pre-release
  • adds rocFFT backend
  • adds CLI option 'tag'
  • cmake refactored
    • superbuild mode available for downloading and building external libraries
    • packaging support, also handles static libs and multi-architecture, ie., different compilers
    • cmake 3.7+ required
    • targets will be built in subfolders (gearshifft/gearshifft_clfft, ...)
    • improvements to finders
    • precision selector (choose to compile only for single or double precision)
    • more information in #124
  • fixes C++17 'typename' bug (uses C++14 'class' instead)
  • checks device if half precision is supported (cufft backend)

WARNING: this probably only works for Boost 1.66.0 and newer (see #145)

v0.3.0

31 Aug 19:15
Compare
Choose a tag to compare
v0.3.0 Pre-release
Pre-release

Just a pre-release to freeze current features.

  • adds MKL FFTW Wrapper back-end
  • adds travis CI scripts
  • adds float16 support for cuFFT half-precision FFTs
  • removed results and shiny app, which are now located in gearshifft_results
  • results not compatible to v0.2.0

v0.2.0

28 Mar 07:29
Compare
Choose a tag to compare
v0.2.0 Pre-release
Pre-release

With v0.2.0 gearshifft provides better latency and more accurate timings as well as more command-line arguments for fftw. Further improvements:

  • error-bound (=1e-5), number of warmups (=1) and number of runs (=10) can be set via cmake
  • the data for FFT is no more using randoms but a see-saw function
  • the csv outputs have been improved

v0.1.0

27 Nov 16:20
Compare
Choose a tag to compare
v0.1.0 Pre-release
Pre-release

This is a pre-release of gearshifft. We are still in the initial development phase, so other platforms than Linux are not supported yet and changes of the gearshifft API are likely. gearshifft v0.1.0 provides FFT client wrapper for cuFFT, clFFT and FFTW to benchmark plan creation, data transfer, FFT and inverse FFT as well as buffer allocations and context/device initialization. The goal of this release is to satisfy the upstream requirements of gearshifft_publication; DOI