Commits

Breadcrumbs

History for

onmaster

User selector

All users

Datepicker

All time

Commit History

Commits on Nov 15, 2016

• Add support for non-power of 2 FFT sizes using Chirp z-transform a.k.a Bluestein algorithm

  linkotec

  committed
  ebac20d

  Copy full SHA for ebac20d

Commits on Nov 14, 2016

• Rename double_t to ffts_double_t, and use for all trig tables

  linkotec

  committed
  92c8a59

  Copy full SHA for 92c8a59

• New functions ffts_cexp_32f to calculate cos(pi * n / d) and sin(pi * n / d), ffts_generate_chirp_32f to generate chirp z-transform sequence, and replace old ffts_generate_cosine_sine_32f with more…

  Show description for 2353060
  linkotec

  committed
  2353060

  Copy full SHA for 2353060

Commits on Oct 19, 2016

• Improve const correctness

  linkotec

  committed
  7de1b87

  Copy full SHA for 7de1b87

• Fix SSE instruction extending to align loops

  linkotec

  committed
  f11355b

  Copy full SHA for f11355b

Commits on Oct 18, 2016

• Allocate aligned memory using non-obsolete methods, and stop using FFTS_MALLOC and FFTS_FREE macros for this

  linkotec

  committed
  132fa85

  Copy full SHA for 132fa85

Commits on Oct 4, 2016

• Use _mm_movehl_ps instead of _mm_shuffle_ps, it is significantly faster on older machines.

  linkotec

  committed
  ee39a68

  Copy full SHA for ee39a68

Commits on Oct 2, 2016

• Temporarily disable AltiVec support pending update to use new data types

  madscientist159

  committed
  417e6d2

  Copy full SHA for 417e6d2

• Fix Autotools build

  madscientist159

  committed
  c3f7d77

  Copy full SHA for c3f7d77

Commits on May 5, 2016

• Introduce ENABLE_LOG macro

  h6ah4i

  committed
  1f4b9a2

  Copy full SHA for 1f4b9a2

Commits on Apr 6, 2016

• Try to remove some of the hard coded offsets to _ffts_plan_t

  linkotec

  committed
  c85bf4b

  Copy full SHA for c85bf4b

• Silence the compiler warnings

  linkotec

  committed
  78d328e

  Copy full SHA for 78d328e

Commits on Apr 5, 2016

• Combine ffts_tranpose_scalar and ffts_transpose, and use ffts_transpose_scalar as native C fallback

  linkotec

  committed
  8b2d55e

  Copy full SHA for 8b2d55e

• Fix ffts_init_nd() for 3 or higher rank complex FFTs

  linkotec

  committed
  c71724a

  Copy full SHA for c71724a

Commits on Mar 29, 2016

• Fix neon_transpose8 for non-square matrices, move loops to assembly side, about 5% faster

  linkotec

  committed
  e464bcb

  Copy full SHA for e464bcb

Commits on Mar 28, 2016

• Rename neon_transpose to neon_transpose4, 4x4 tiled matrix transpose.

  Show description for ab556f8
  linkotec

  committed
  ab556f8

  Copy full SHA for ab556f8

• "transpose_buf" is not used

  linkotec

  committed
  7f74c87

  Copy full SHA for 7f74c87

• Improve performance of small complex 2D Neon transform by 15%

  linkotec

  committed
  2cf6816

  Copy full SHA for 2cf6816

Commits on Mar 17, 2016

• Simplify maintenance by using macros

  linkotec

  committed
  a62bc08

  Copy full SHA for a62bc08

• Use local labels to fix 'symbol already defined' errors

  linkotec

  committed
  7dea850

  Copy full SHA for 7dea850

• Combine neon_static_f.s and neon_static_i.s to neon_static.s

  linkotec

  committed
  b982322

  Copy full SHA for b982322

Commits on Mar 16, 2016

• Clean and optimize ARM Neon static transform, 4-5% faster

  linkotec

  committed
  52beaeb

  Copy full SHA for 52beaeb

Commits on Mar 15, 2016

• neon_static_x4_f and neon_static_x4_i don't use the second passed argument, and reschedule instructions for possible dual issue

  linkotec

  committed
  f64d89a

  Copy full SHA for f64d89a

Commits on Mar 14, 2016

• Coverage analysis shows unused if-else branches

  linkotec

  committed
  85a7167

  Copy full SHA for 85a7167

• Unroll to minimize recursive function call depth (overhead)

  linkotec

  committed
  10d4d45

  Copy full SHA for 10d4d45

• Peel off top-level only if-case from ARM NEON recursive implementation

  linkotec

  committed
  6116601

  Copy full SHA for 6116601

Commits on Mar 11, 2016

• Restore ARM NEON optimized recursive version

  linkotec

  committed
  e667ca5

  Copy full SHA for e667ca5

Commits on Nov 30, 2015

• Enable building shared library and start version numbering from 0.9.0. On Windows when using FFTS as a DLL, define FFTS_SHARED. This is not mandatory, but it offers a little performance increase. H…

  Show description for ae1b59d
  linkotec

  committed
  ae1b59d

  Copy full SHA for ae1b59d

Commits on Oct 12, 2015

• Fix MSVC error C2719

  linkotec

  committed
  2132b65

  Copy full SHA for 2132b65

Commits on Sep 17, 2015

• Add SSE2 optimized ffts_generate_cosine_sine_pow2_32f

  linkotec

  committed
  e013b85

  Copy full SHA for e013b85

Commits on Sep 16, 2015

• Add double-double arithmetic to generate "exact" double precision cosine and sine tables. Correct rounding verified using MPFR upto 2^28. SSE2 optimized ffts_generate_cosine_sine_pow2_64f takes twi…

  Show description for f4e533c
  linkotec

  committed
  f4e533c

  Copy full SHA for f4e533c

• Change the order of constants; cos_hi, cos_lo, sin_hi, sin_lo -> cos_hi, sin_hi, cos_lo, sin_lo to support 128 bit vectorization

  linkotec

  committed
  7ae74d3

  Copy full SHA for 7ae74d3

Commits on Sep 15, 2015

• Extended constant tables to double-double arithmetic

  linkotec

  committed
  896904f

  Copy full SHA for 896904f

Commits on Aug 28, 2015

• No need to display the size of transform

  linkotec

  committed
  5f0db6e

  Copy full SHA for 5f0db6e

Commits on Jul 30, 2015

• Control reaches end of non-void function

  linkotec

  committed
  cdf9015

  Copy full SHA for cdf9015

Pagination

Previous

Next