Age | Commit message (Collapse) | Author |
|
* fix compilation when deactivating SIMD with cmake options
* cmake: replaced 'USE_AVX' option by 'USE_DOUBLE' (and 'USE_FLOAT')
* cmake: option 'USE_SCALVEC',
fixed big precision errors in test_pffastconv
* cmake: renamed options to be sorted by 'group' in ccmake
* added API function pffft_simd_arch()
* pffft_bench for float and double
double uses fftw3 for comparison
* moved all simd macro files into subfolder
* fix buffer overrun of acCsvFilename[] in bench_pffft.c
Signed-off-by: hayati ayguen <h_ayguen@web.de>
|
|
* bugfix test_pffft.cpp: missing free
* bugfix test_pffft.cpp: normalization of inverse() was on wrong index or variable
* retrospection
- wrapper class with typedef to initial template type
- static helpers: isComplexTransform(), isFloatScalar() and isDoubleScalar()
* allocation helpers:
- member helpers: getSpectrumSize(), getInternalLayoutSize() and
allocateOrigin(), allocateSpectrum() and allocateInternalLayout()
* test_pffft float/double/cpp:
float/double: use/test unordered pffft_transform() for inverse
cpp: use/test inverseInternalLayout() when not useOrdered
Signed-off-by: hayati ayguen <h_ayguen@web.de>
|
|
* replaced the c++ style '//' comments
* added explicit math defines M_PI, .. - if not already defined
* gcc function (inline) attributes still require the extensions :-(
Signed-off-by: hayati ayguen <h_ayguen@web.de>
|
|
* added pffft_zconvolve_no_accu()
* added pffft_min_fft_size()
* pffft_next_power_of_two(), pffft_is_power_of_two()
* added option PFFASTCONV_CORRELATION
* fixed calculation of procSmpPerSec[] in test_pffastconv.c
* minor test additions
Signed-off-by: hayati ayguen <h_ayguen@web.de>
|
|
* API for fast convolution: pffastconv.h
* move simd macros from pffft.c into pfsimd_macros.h for reusability
Signed-off-by: hayati ayguen <h_ayguen@web.de>
|
|
Signed-off-by: hayati ayguen <h_ayguen@web.de>
|
|
Signed-off-by: hayati ayguen <h_ayguen@web.de>
|
|
* renamed existing pffft benchmark to "-U" for unordered
* added pffft ordered benchmark
* stop benchmark after 150 ms per fft algorithm
Signed-off-by: hayati ayguen <h_ayguen@web.de>
|
|
* simplified usage of other gcc compiler versions
Signed-off-by: hayati ayguen <h_ayguen@web.de>
|
|
* test_pffft now produces .csv files directly
* automatic compilation/testing and plots with bench_all.sh
Signed-off-by: hayati ayguen <h_ayguen@web.de>
|
|
* bugfix table output header names
detailed output of pffft algorithm in header (simd/scalar)
* use table output as default; added command line option '--no-tab'
* add execution of benchmark as ctest
Signed-off-by: hayati ayguen <h_ayguen@web.de>
|
|
* added kissfft as git submodule
* added greenfft as git submodule
* enhanced test_pffft.c to output csv tables .. suitable to generate plots
Signed-off-by: hayati ayguen <h_ayguen@web.de>
|
|
benchmark on ipad air 2
merge from https://bitbucket.org/jpommier/pffft/src/default/
Signed-off-by: hayati ayguen <h_ayguen@web.de>
|
|
with clang 3.2 for arm on linux
|
|
|
|
|
|
|
|
|
|
|