softclip.hpp

   1 //  softclip
   2 //  Copyright (C) 2008, 2009 Tim Blechmann
   3 //
   4 //  This program is free software; you can redistribute it and/or modify
   5 //  it under the terms of the GNU General Public License as published by
   6 //  the Free Software Foundation; either version 2 of the License, or
   7 //  (at your option) any later version.
   8 //
   9 //  This program is distributed in the hope that it will be useful,
  10 //  but WITHOUT ANY WARRANTY; without even the implied warranty of
  11 //  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12 //  GNU General Public License for more details.
  13 //
  14 //  You should have received a copy of the GNU General Public License
  15 //  along with this program; see the file COPYING.  If not, write to
  16 //  the Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  17 //  Boston, MA 02111-1307, USA.
  18
  19 #ifndef SIMD_SOFTCLIP_HPP
  20 #define SIMD_SOFTCLIP_HPP
  21
  22 #include <cassert>
  23 #include <cmath>
  24
  25 #include "vec.hpp"
  26
  27 #include "detail/define_macros.hpp"
  28
  29 #if defined(__GNUC__) && defined(NDEBUG)
  30 #define always_inline inline  __attribute__((always_inline))
  31 #else
  32 #define always_inline inline
  33 #endif
  34
  35 namespace nova   {
  36 namespace detail {
  37
  38 struct softclip
  39 {
  40     template <typename FloatType>
  41     static FloatType s_run(FloatType arg)
  42     {
  43         FloatType abs = std::fabs(arg);
  44         if (abs < FloatType(0.5))
  45             return arg;
  46         else
  47             return (abs - FloatType(0.25)) / arg;
  48     }
  49
  50     template <typename FloatType>
  51     always_inline FloatType operator()(FloatType arg) const
  52     {
  53         return s_run(arg);
  54     }
  55
  56 #if defined(__SSE__) || defined(__AVX__)
  57     /* this computes both parts of the branch
  58      *
  59      * benchmarks (core2) showed:
  60      * 6.5 seconds for simdfied code
  61      * 5.3 seconds for non-simd code for samples with abs(sample) < 0.5
  62      * 17 seconds                    when 50% of the samples have abs(sample) < 0.5
  63      * 26 seconds                    for samples with abs(sample) > 0.5
  64      *
  65      * on intel nethalem cpus, the simdfied code is faster than all non-simd code
  66      * */
  67
  68     template <typename FloatType>
  69     always_inline vec<FloatType> operator()(vec<FloatType> arg) const
  70     {
  71         typedef vec<FloatType> vec_type;
  72
  73         const vec_type const05 = vec_type::gen_05();
  74         const vec_type const025 (0.25);
  75
  76         vec_type abs_ = abs(arg);
  77         vec_type selecter = mask_lt(abs_, const05);
  78         vec_type alt_ret = (abs_ - const025) / arg;
  79
  80         return select(alt_ret, arg, selecter);
  81     }
  82 #else
  83     template <typename FloatType>
  84     always_inline vec<FloatType> operator()(vec<FloatType> arg) const
  85     {
  86         return arg.collect(softclip::s_run<FloatType>);
  87     }
  88 #endif
  89 };
  90
  91 } /* namespace detail */
  92
  93 NOVA_SIMD_DEFINE_UNARY_WRAPPER(softclip, detail::softclip)
  94
  95
  96 } /* namespace nova */
  97
  98 #undef always_inline
  99
 100 #endif /* SIMD_SOFTCLIP_HPP */