Alc/mixer_sse3.c

   1 /**
   2  * OpenAL cross platform audio library, SSE3 mixer functions
   3  *
   4  * Copyright (C) 2014 by Timothy Arceri <t_arceri@yahoo.com.au>.
   5  * Copyright (C) 2015 by Chris Robinson <chris.kcat@gmail.com>.
   6  *
   7  * This library is free software; you can redistribute it and/or
   8  *  modify it under the terms of the GNU Library General Public
   9  *  License as published by the Free Software Foundation; either
  10  *  version 2 of the License, or (at your option) any later version.
  11  *
  12  * This library is distributed in the hope that it will be useful,
  13  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  *  Library General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU Library General Public
  18  *  License along with this library; if not, write to the
  19  *  Free Software Foundation, Inc.,
  20  *  51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  21  * Or go to http://www.gnu.org/copyleft/lgpl.html
  22  */
  23
  24 #include "config.h"
  25
  26 #include <xmmintrin.h>
  27 #include <emmintrin.h>
  28 #include <pmmintrin.h>
  29
  30 #include "alu.h"
  31 #include "mixer_defs.h"
  32
  33
  34 const ALfloat *Resample_fir4_32_SSE3(const BsincState* UNUSED(state), const ALfloat *restrict src,
  35                                      ALuint frac, ALuint increment, ALfloat *restrict dst,
  36                                      ALuint numsamples)
  37 {
  38     const __m128i increment4 = _mm_set1_epi32(increment*4);
  39     const __m128i fracMask4 = _mm_set1_epi32(FRACTIONMASK);
  40     union { alignas(16) ALuint i[4]; float f[4]; } pos_;
  41     union { alignas(16) ALuint i[4]; float f[4]; } frac_;
  42     __m128i frac4, pos4;
  43     ALuint pos;
  44     ALuint i;
  45
  46     InitiatePositionArrays(frac, increment, frac_.i, pos_.i, 4);
  47
  48     frac4 = _mm_castps_si128(_mm_load_ps(frac_.f));
  49     pos4 = _mm_castps_si128(_mm_load_ps(pos_.f));
  50
  51     --src;
  52     for(i = 0;numsamples-i > 3;i += 4)
  53     {
  54         const __m128 val0 = _mm_loadu_ps(&src[pos_.i[0]]);
  55         const __m128 val1 = _mm_loadu_ps(&src[pos_.i[1]]);
  56         const __m128 val2 = _mm_loadu_ps(&src[pos_.i[2]]);
  57         const __m128 val3 = _mm_loadu_ps(&src[pos_.i[3]]);
  58         __m128 k0 = _mm_load_ps(ResampleCoeffs.FIR4[frac_.i[0]]);
  59         __m128 k1 = _mm_load_ps(ResampleCoeffs.FIR4[frac_.i[1]]);
  60         __m128 k2 = _mm_load_ps(ResampleCoeffs.FIR4[frac_.i[2]]);
  61         __m128 k3 = _mm_load_ps(ResampleCoeffs.FIR4[frac_.i[3]]);
  62         __m128 out;
  63
  64         k0 = _mm_mul_ps(k0, val0);
  65         k1 = _mm_mul_ps(k1, val1);
  66         k2 = _mm_mul_ps(k2, val2);
  67         k3 = _mm_mul_ps(k3, val3);
  68         k0 = _mm_hadd_ps(k0, k1);
  69         k2 = _mm_hadd_ps(k2, k3);
  70         out = _mm_hadd_ps(k0, k2);
  71
  72         _mm_store_ps(&dst[i], out);
  73
  74         frac4 = _mm_add_epi32(frac4, increment4);
  75         pos4 = _mm_add_epi32(pos4, _mm_srli_epi32(frac4, FRACTIONBITS));
  76         frac4 = _mm_and_si128(frac4, fracMask4);
  77
  78         _mm_store_ps(pos_.f, _mm_castsi128_ps(pos4));
  79         _mm_store_ps(frac_.f, _mm_castsi128_ps(frac4));
  80     }
  81
  82     /* NOTE: These four elements represent the position *after* the last four
  83      * samples, so the lowest element is the next position to resample.
  84      */
  85     pos = pos_.i[0];
  86     frac = frac_.i[0];
  87
  88     for(;i < numsamples;i++)
  89     {
  90         dst[i] = resample_fir4(src[pos], src[pos+1], src[pos+2], src[pos+3], frac);
  91
  92         frac += increment;
  93         pos  += frac>>FRACTIONBITS;
  94         frac &= FRACTIONMASK;
  95     }
  96     return dst;
  97 }
  98
  99 const ALfloat *Resample_fir8_32_SSE3(const BsincState* UNUSED(state), const ALfloat *restrict src,
 100                                      ALuint frac, ALuint increment, ALfloat *restrict dst,
 101                                      ALuint numsamples)
 102 {
 103     const __m128i increment4 = _mm_set1_epi32(increment*4);
 104     const __m128i fracMask4 = _mm_set1_epi32(FRACTIONMASK);
 105     union { alignas(16) ALuint i[4]; float f[4]; } pos_;
 106     union { alignas(16) ALuint i[4]; float f[4]; } frac_;
 107     __m128i frac4, pos4;
 108     ALuint pos;
 109     ALuint i, j;
 110
 111     InitiatePositionArrays(frac, increment, frac_.i, pos_.i, 4);
 112
 113     frac4 = _mm_castps_si128(_mm_load_ps(frac_.f));
 114     pos4 = _mm_castps_si128(_mm_load_ps(pos_.f));
 115
 116     src -= 3;
 117     for(i = 0;numsamples-i > 3;i += 4)
 118     {
 119         __m128 out[2];
 120         for(j = 0;j < 8;j+=4)
 121         {
 122             const __m128 val0 = _mm_loadu_ps(&src[pos_.i[0]+j]);
 123             const __m128 val1 = _mm_loadu_ps(&src[pos_.i[1]+j]);
 124             const __m128 val2 = _mm_loadu_ps(&src[pos_.i[2]+j]);
 125             const __m128 val3 = _mm_loadu_ps(&src[pos_.i[3]+j]);
 126             __m128 k0 = _mm_load_ps(&ResampleCoeffs.FIR8[frac_.i[0]][j]);
 127             __m128 k1 = _mm_load_ps(&ResampleCoeffs.FIR8[frac_.i[1]][j]);
 128             __m128 k2 = _mm_load_ps(&ResampleCoeffs.FIR8[frac_.i[2]][j]);
 129             __m128 k3 = _mm_load_ps(&ResampleCoeffs.FIR8[frac_.i[3]][j]);
 130
 131             k0 = _mm_mul_ps(k0, val0);
 132             k1 = _mm_mul_ps(k1, val1);
 133             k2 = _mm_mul_ps(k2, val2);
 134             k3 = _mm_mul_ps(k3, val3);
 135             k0 = _mm_hadd_ps(k0, k1);
 136             k2 = _mm_hadd_ps(k2, k3);
 137             out[j>>2] = _mm_hadd_ps(k0, k2);
 138         }
 139
 140         out[0] = _mm_add_ps(out[0], out[1]);
 141         _mm_store_ps(&dst[i], out[0]);
 142
 143         frac4 = _mm_add_epi32(frac4, increment4);
 144         pos4 = _mm_add_epi32(pos4, _mm_srli_epi32(frac4, FRACTIONBITS));
 145         frac4 = _mm_and_si128(frac4, fracMask4);
 146
 147         _mm_store_ps(pos_.f, _mm_castsi128_ps(pos4));
 148         _mm_store_ps(frac_.f, _mm_castsi128_ps(frac4));
 149     }
 150
 151     pos = pos_.i[0];
 152     frac = frac_.i[0];
 153
 154     for(;i < numsamples;i++)
 155     {
 156         dst[i] = resample_fir8(src[pos  ], src[pos+1], src[pos+2], src[pos+3],
 157                                src[pos+4], src[pos+5], src[pos+6], src[pos+7], frac);
 158
 159         frac += increment;
 160         pos  += frac>>FRACTIONBITS;
 161         frac &= FRACTIONMASK;
 162     }
 163     return dst;
 164 }