Alc/mixer_sse41.c

   1 /**
   2  * OpenAL cross platform audio library
   3  * Copyright (C) 2014 by Timothy Arceri <t_arceri@yahoo.com.au>.
   4  * This library is free software; you can redistribute it and/or
   5  *  modify it under the terms of the GNU Library General Public
   6  *  License as published by the Free Software Foundation; either
   7  *  version 2 of the License, or (at your option) any later version.
   8  *
   9  * This library 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 GNU
  12  *  Library General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU Library General Public
  15  *  License along with this library; if not, write to the
  16  *  Free Software Foundation, Inc.,
  17  *  51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  18  * Or go to http://www.gnu.org/copyleft/lgpl.html
  19  */
  20
  21 #include "config.h"
  22
  23 #include <xmmintrin.h>
  24 #include <emmintrin.h>
  25 #include <smmintrin.h>
  26
  27 #include "alu.h"
  28 #include "mixer_defs.h"
  29
  30
  31 const ALfloat *Resample_lerp32_SSE41(const BsincState* UNUSED(state), const ALfloat *src, ALuint frac, ALuint increment,
  32                                      ALfloat *restrict dst, ALuint numsamples)
  33 {
  34     const __m128i increment4 = _mm_set1_epi32(increment*4);
  35     const __m128 fracOne4 = _mm_set1_ps(1.0f/FRACTIONONE);
  36     const __m128i fracMask4 = _mm_set1_epi32(FRACTIONMASK);
  37     union { alignas(16) ALuint i[4]; float f[4]; } pos_;
  38     union { alignas(16) ALuint i[4]; float f[4]; } frac_;
  39     __m128i frac4, pos4;
  40     ALuint pos;
  41     ALuint i;
  42
  43     InitiatePositionArrays(frac, increment, frac_.i, pos_.i, 4);
  44
  45     frac4 = _mm_castps_si128(_mm_load_ps(frac_.f));
  46     pos4 = _mm_castps_si128(_mm_load_ps(pos_.f));
  47
  48     for(i = 0;numsamples-i > 3;i += 4)
  49     {
  50         const __m128 val1 = _mm_setr_ps(src[pos_.i[0]], src[pos_.i[1]], src[pos_.i[2]], src[pos_.i[3]]);
  51         const __m128 val2 = _mm_setr_ps(src[pos_.i[0]+1], src[pos_.i[1]+1], src[pos_.i[2]+1], src[pos_.i[3]+1]);
  52
  53         /* val1 + (val2-val1)*mu */
  54         const __m128 r0 = _mm_sub_ps(val2, val1);
  55         const __m128 mu = _mm_mul_ps(_mm_cvtepi32_ps(frac4), fracOne4);
  56         const __m128 out = _mm_add_ps(val1, _mm_mul_ps(mu, r0));
  57
  58         _mm_store_ps(&dst[i], out);
  59
  60         frac4 = _mm_add_epi32(frac4, increment4);
  61         pos4 = _mm_add_epi32(pos4, _mm_srli_epi32(frac4, FRACTIONBITS));
  62         frac4 = _mm_and_si128(frac4, fracMask4);
  63
  64         pos_.i[0] = _mm_extract_epi32(pos4, 0);
  65         pos_.i[1] = _mm_extract_epi32(pos4, 1);
  66         pos_.i[2] = _mm_extract_epi32(pos4, 2);
  67         pos_.i[3] = _mm_extract_epi32(pos4, 3);
  68     }
  69
  70     /* NOTE: These four elements represent the position *after* the last four
  71      * samples, so the lowest element is the next position to resample.
  72      */
  73     pos = pos_.i[0];
  74     frac = _mm_cvtsi128_si32(frac4);
  75
  76     for(;i < numsamples;i++)
  77     {
  78         dst[i] = lerp(src[pos], src[pos+1], frac * (1.0f/FRACTIONONE));
  79
  80         frac += increment;
  81         pos  += frac>>FRACTIONBITS;
  82         frac &= FRACTIONMASK;
  83     }
  84     return dst;
  85 }
  86
  87 const ALfloat *Resample_fir4_32_SSE41(const BsincState* UNUSED(state), const ALfloat *src, ALuint frac, ALuint increment,
  88                                       ALfloat *restrict dst, ALuint numsamples)
  89 {
  90     const __m128i increment4 = _mm_set1_epi32(increment*4);
  91     const __m128i fracMask4 = _mm_set1_epi32(FRACTIONMASK);
  92     union { alignas(16) ALuint i[4]; float f[4]; } pos_;
  93     union { alignas(16) ALuint i[4]; float f[4]; } frac_;
  94     __m128i frac4, pos4;
  95     ALuint pos;
  96     ALuint i;
  97
  98     InitiatePositionArrays(frac, increment, frac_.i, pos_.i, 4);
  99
 100     frac4 = _mm_castps_si128(_mm_load_ps(frac_.f));
 101     pos4 = _mm_castps_si128(_mm_load_ps(pos_.f));
 102
 103     --src;
 104     for(i = 0;numsamples-i > 3;i += 4)
 105     {
 106         const __m128 val0 = _mm_loadu_ps(&src[pos_.i[0]]);
 107         const __m128 val1 = _mm_loadu_ps(&src[pos_.i[1]]);
 108         const __m128 val2 = _mm_loadu_ps(&src[pos_.i[2]]);
 109         const __m128 val3 = _mm_loadu_ps(&src[pos_.i[3]]);
 110         __m128 k0 = _mm_load_ps(ResampleCoeffs.FIR4[frac_.i[0]]);
 111         __m128 k1 = _mm_load_ps(ResampleCoeffs.FIR4[frac_.i[1]]);
 112         __m128 k2 = _mm_load_ps(ResampleCoeffs.FIR4[frac_.i[2]]);
 113         __m128 k3 = _mm_load_ps(ResampleCoeffs.FIR4[frac_.i[3]]);
 114         __m128 out;
 115
 116         k0 = _mm_mul_ps(k0, val0);
 117         k1 = _mm_mul_ps(k1, val1);
 118         k2 = _mm_mul_ps(k2, val2);
 119         k3 = _mm_mul_ps(k3, val3);
 120         k0 = _mm_hadd_ps(k0, k1);
 121         k2 = _mm_hadd_ps(k2, k3);
 122         out = _mm_hadd_ps(k0, k2);
 123
 124         _mm_store_ps(&dst[i], out);
 125
 126         frac4 = _mm_add_epi32(frac4, increment4);
 127         pos4 = _mm_add_epi32(pos4, _mm_srli_epi32(frac4, FRACTIONBITS));
 128         frac4 = _mm_and_si128(frac4, fracMask4);
 129
 130         pos_.i[0] = _mm_extract_epi32(pos4, 0);
 131         pos_.i[1] = _mm_extract_epi32(pos4, 1);
 132         pos_.i[2] = _mm_extract_epi32(pos4, 2);
 133         pos_.i[3] = _mm_extract_epi32(pos4, 3);
 134         frac_.i[0] = _mm_extract_epi32(frac4, 0);
 135         frac_.i[1] = _mm_extract_epi32(frac4, 1);
 136         frac_.i[2] = _mm_extract_epi32(frac4, 2);
 137         frac_.i[3] = _mm_extract_epi32(frac4, 3);
 138     }
 139
 140     pos = pos_.i[0];
 141     frac = frac_.i[0];
 142
 143     for(;i < numsamples;i++)
 144     {
 145         dst[i] = resample_fir4(src[pos], src[pos+1], src[pos+2], src[pos+3], frac);
 146
 147         frac += increment;
 148         pos  += frac>>FRACTIONBITS;
 149         frac &= FRACTIONMASK;
 150     }
 151     return dst;
 152 }
 153
 154 const ALfloat *Resample_fir8_32_SSE41(const BsincState* UNUSED(state), const ALfloat *src, ALuint frac, ALuint increment,
 155                                       ALfloat *restrict dst, ALuint numsamples)
 156 {
 157     const __m128i increment4 = _mm_set1_epi32(increment*4);
 158     const __m128i fracMask4 = _mm_set1_epi32(FRACTIONMASK);
 159     union { alignas(16) ALuint i[4]; float f[4]; } pos_;
 160     union { alignas(16) ALuint i[4]; float f[4]; } frac_;
 161     __m128i frac4, pos4;
 162     ALuint pos;
 163     ALuint i, j;
 164
 165     InitiatePositionArrays(frac, increment, frac_.i, pos_.i, 4);
 166
 167     frac4 = _mm_castps_si128(_mm_load_ps(frac_.f));
 168     pos4 = _mm_castps_si128(_mm_load_ps(pos_.f));
 169
 170     src -= 3;
 171     for(i = 0;numsamples-i > 3;i += 4)
 172     {
 173         __m128 out[2];
 174         for(j = 0;j < 8;j+=4)
 175         {
 176             const __m128 val0 = _mm_loadu_ps(&src[pos_.i[0]+j]);
 177             const __m128 val1 = _mm_loadu_ps(&src[pos_.i[1]+j]);
 178             const __m128 val2 = _mm_loadu_ps(&src[pos_.i[2]+j]);
 179             const __m128 val3 = _mm_loadu_ps(&src[pos_.i[3]+j]);
 180             __m128 k0 = _mm_load_ps(&ResampleCoeffs.FIR8[frac_.i[0]][j]);
 181             __m128 k1 = _mm_load_ps(&ResampleCoeffs.FIR8[frac_.i[1]][j]);
 182             __m128 k2 = _mm_load_ps(&ResampleCoeffs.FIR8[frac_.i[2]][j]);
 183             __m128 k3 = _mm_load_ps(&ResampleCoeffs.FIR8[frac_.i[3]][j]);
 184
 185             k0 = _mm_mul_ps(k0, val0);
 186             k1 = _mm_mul_ps(k1, val1);
 187             k2 = _mm_mul_ps(k2, val2);
 188             k3 = _mm_mul_ps(k3, val3);
 189             k0 = _mm_hadd_ps(k0, k1);
 190             k2 = _mm_hadd_ps(k2, k3);
 191             out[j>>2] = _mm_hadd_ps(k0, k2);
 192         }
 193
 194         out[0] = _mm_add_ps(out[0], out[1]);
 195         _mm_store_ps(&dst[i], out[0]);
 196
 197         frac4 = _mm_add_epi32(frac4, increment4);
 198         pos4 = _mm_add_epi32(pos4, _mm_srli_epi32(frac4, FRACTIONBITS));
 199         frac4 = _mm_and_si128(frac4, fracMask4);
 200
 201         pos_.i[0] = _mm_extract_epi32(pos4, 0);
 202         pos_.i[1] = _mm_extract_epi32(pos4, 1);
 203         pos_.i[2] = _mm_extract_epi32(pos4, 2);
 204         pos_.i[3] = _mm_extract_epi32(pos4, 3);
 205         frac_.i[0] = _mm_extract_epi32(frac4, 0);
 206         frac_.i[1] = _mm_extract_epi32(frac4, 1);
 207         frac_.i[2] = _mm_extract_epi32(frac4, 2);
 208         frac_.i[3] = _mm_extract_epi32(frac4, 3);
 209     }
 210
 211     pos = pos_.i[0];
 212     frac = frac_.i[0];
 213
 214     for(;i < numsamples;i++)
 215     {
 216         dst[i] = resample_fir8(src[pos  ], src[pos+1], src[pos+2], src[pos+3],
 217                                src[pos+4], src[pos+5], src[pos+6], src[pos+7], frac);
 218
 219         frac += increment;
 220         pos  += frac>>FRACTIONBITS;
 221         frac &= FRACTIONMASK;
 222     }
 223     return dst;
 224 }