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 *restrict src,
  32                                      ALuint frac, ALuint increment, ALfloat *restrict dst,
  33                                      ALuint numsamples)
  34 {
  35     const __m128i increment4 = _mm_set1_epi32(increment*4);
  36     const __m128 fracOne4 = _mm_set1_ps(1.0f/FRACTIONONE);
  37     const __m128i fracMask4 = _mm_set1_epi32(FRACTIONMASK);
  38     union { alignas(16) ALuint i[4]; float f[4]; } pos_;
  39     union { alignas(16) ALuint i[4]; float f[4]; } frac_;
  40     __m128i frac4, pos4;
  41     ALuint pos;
  42     ALuint i;
  43
  44     InitiatePositionArrays(frac, increment, frac_.i, pos_.i, 4);
  45
  46     frac4 = _mm_castps_si128(_mm_load_ps(frac_.f));
  47     pos4 = _mm_castps_si128(_mm_load_ps(pos_.f));
  48
  49     for(i = 0;numsamples-i > 3;i += 4)
  50     {
  51         const __m128 val1 = _mm_setr_ps(src[pos_.i[0]], src[pos_.i[1]], src[pos_.i[2]], src[pos_.i[3]]);
  52         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]);
  53
  54         /* val1 + (val2-val1)*mu */
  55         const __m128 r0 = _mm_sub_ps(val2, val1);
  56         const __m128 mu = _mm_mul_ps(_mm_cvtepi32_ps(frac4), fracOne4);
  57         const __m128 out = _mm_add_ps(val1, _mm_mul_ps(mu, r0));
  58
  59         _mm_store_ps(&dst[i], out);
  60
  61         frac4 = _mm_add_epi32(frac4, increment4);
  62         pos4 = _mm_add_epi32(pos4, _mm_srli_epi32(frac4, FRACTIONBITS));
  63         frac4 = _mm_and_si128(frac4, fracMask4);
  64
  65         pos_.i[0] = _mm_extract_epi32(pos4, 0);
  66         pos_.i[1] = _mm_extract_epi32(pos4, 1);
  67         pos_.i[2] = _mm_extract_epi32(pos4, 2);
  68         pos_.i[3] = _mm_extract_epi32(pos4, 3);
  69     }
  70
  71     /* NOTE: These four elements represent the position *after* the last four
  72      * samples, so the lowest element is the next position to resample.
  73      */
  74     pos = pos_.i[0];
  75     frac = _mm_cvtsi128_si32(frac4);
  76
  77     for(;i < numsamples;i++)
  78     {
  79         dst[i] = lerp(src[pos], src[pos+1], frac * (1.0f/FRACTIONONE));
  80
  81         frac += increment;
  82         pos  += frac>>FRACTIONBITS;
  83         frac &= FRACTIONMASK;
  84     }
  85     return dst;
  86 }
  87
  88 const ALfloat *Resample_fir4_32_SSE41(const BsincState* UNUSED(state), const ALfloat *restrict src,
  89                                       ALuint frac, ALuint increment, ALfloat *restrict dst,
  90                                       ALuint numsamples)
  91 {
  92     const __m128i increment4 = _mm_set1_epi32(increment*4);
  93     const __m128i fracMask4 = _mm_set1_epi32(FRACTIONMASK);
  94     union { alignas(16) ALuint i[4]; float f[4]; } pos_;
  95     union { alignas(16) ALuint i[4]; float f[4]; } frac_;
  96     __m128i frac4, pos4;
  97     ALuint pos;
  98     ALuint i;
  99
 100     InitiatePositionArrays(frac, increment, frac_.i, pos_.i, 4);
 101
 102     frac4 = _mm_castps_si128(_mm_load_ps(frac_.f));
 103     pos4 = _mm_castps_si128(_mm_load_ps(pos_.f));
 104
 105     --src;
 106     for(i = 0;numsamples-i > 3;i += 4)
 107     {
 108         const __m128 val0 = _mm_loadu_ps(&src[pos_.i[0]]);
 109         const __m128 val1 = _mm_loadu_ps(&src[pos_.i[1]]);
 110         const __m128 val2 = _mm_loadu_ps(&src[pos_.i[2]]);
 111         const __m128 val3 = _mm_loadu_ps(&src[pos_.i[3]]);
 112         __m128 k0 = _mm_load_ps(ResampleCoeffs.FIR4[frac_.i[0]]);
 113         __m128 k1 = _mm_load_ps(ResampleCoeffs.FIR4[frac_.i[1]]);
 114         __m128 k2 = _mm_load_ps(ResampleCoeffs.FIR4[frac_.i[2]]);
 115         __m128 k3 = _mm_load_ps(ResampleCoeffs.FIR4[frac_.i[3]]);
 116         __m128 out;
 117
 118         k0 = _mm_mul_ps(k0, val0);
 119         k1 = _mm_mul_ps(k1, val1);
 120         k2 = _mm_mul_ps(k2, val2);
 121         k3 = _mm_mul_ps(k3, val3);
 122         k0 = _mm_hadd_ps(k0, k1);
 123         k2 = _mm_hadd_ps(k2, k3);
 124         out = _mm_hadd_ps(k0, k2);
 125
 126         _mm_store_ps(&dst[i], out);
 127
 128         frac4 = _mm_add_epi32(frac4, increment4);
 129         pos4 = _mm_add_epi32(pos4, _mm_srli_epi32(frac4, FRACTIONBITS));
 130         frac4 = _mm_and_si128(frac4, fracMask4);
 131
 132         pos_.i[0] = _mm_extract_epi32(pos4, 0);
 133         pos_.i[1] = _mm_extract_epi32(pos4, 1);
 134         pos_.i[2] = _mm_extract_epi32(pos4, 2);
 135         pos_.i[3] = _mm_extract_epi32(pos4, 3);
 136         frac_.i[0] = _mm_extract_epi32(frac4, 0);
 137         frac_.i[1] = _mm_extract_epi32(frac4, 1);
 138         frac_.i[2] = _mm_extract_epi32(frac4, 2);
 139         frac_.i[3] = _mm_extract_epi32(frac4, 3);
 140     }
 141
 142     pos = pos_.i[0];
 143     frac = frac_.i[0];
 144
 145     for(;i < numsamples;i++)
 146     {
 147         dst[i] = resample_fir4(src[pos], src[pos+1], src[pos+2], src[pos+3], frac);
 148
 149         frac += increment;
 150         pos  += frac>>FRACTIONBITS;
 151         frac &= FRACTIONMASK;
 152     }
 153     return dst;
 154 }
 155
 156 const ALfloat *Resample_fir8_32_SSE41(const BsincState* UNUSED(state), const ALfloat *restrict src,
 157                                       ALuint frac, ALuint increment, ALfloat *restrict dst,
 158                                       ALuint numsamples)
 159 {
 160     const __m128i increment4 = _mm_set1_epi32(increment*4);
 161     const __m128i fracMask4 = _mm_set1_epi32(FRACTIONMASK);
 162     union { alignas(16) ALuint i[4]; float f[4]; } pos_;
 163     union { alignas(16) ALuint i[4]; float f[4]; } frac_;
 164     __m128i frac4, pos4;
 165     ALuint pos;
 166     ALuint i, j;
 167
 168     InitiatePositionArrays(frac, increment, frac_.i, pos_.i, 4);
 169
 170     frac4 = _mm_castps_si128(_mm_load_ps(frac_.f));
 171     pos4 = _mm_castps_si128(_mm_load_ps(pos_.f));
 172
 173     src -= 3;
 174     for(i = 0;numsamples-i > 3;i += 4)
 175     {
 176         __m128 out[2];
 177         for(j = 0;j < 8;j+=4)
 178         {
 179             const __m128 val0 = _mm_loadu_ps(&src[pos_.i[0]+j]);
 180             const __m128 val1 = _mm_loadu_ps(&src[pos_.i[1]+j]);
 181             const __m128 val2 = _mm_loadu_ps(&src[pos_.i[2]+j]);
 182             const __m128 val3 = _mm_loadu_ps(&src[pos_.i[3]+j]);
 183             __m128 k0 = _mm_load_ps(&ResampleCoeffs.FIR8[frac_.i[0]][j]);
 184             __m128 k1 = _mm_load_ps(&ResampleCoeffs.FIR8[frac_.i[1]][j]);
 185             __m128 k2 = _mm_load_ps(&ResampleCoeffs.FIR8[frac_.i[2]][j]);
 186             __m128 k3 = _mm_load_ps(&ResampleCoeffs.FIR8[frac_.i[3]][j]);
 187
 188             k0 = _mm_mul_ps(k0, val0);
 189             k1 = _mm_mul_ps(k1, val1);
 190             k2 = _mm_mul_ps(k2, val2);
 191             k3 = _mm_mul_ps(k3, val3);
 192             k0 = _mm_hadd_ps(k0, k1);
 193             k2 = _mm_hadd_ps(k2, k3);
 194             out[j>>2] = _mm_hadd_ps(k0, k2);
 195         }
 196
 197         out[0] = _mm_add_ps(out[0], out[1]);
 198         _mm_store_ps(&dst[i], out[0]);
 199
 200         frac4 = _mm_add_epi32(frac4, increment4);
 201         pos4 = _mm_add_epi32(pos4, _mm_srli_epi32(frac4, FRACTIONBITS));
 202         frac4 = _mm_and_si128(frac4, fracMask4);
 203
 204         pos_.i[0] = _mm_extract_epi32(pos4, 0);
 205         pos_.i[1] = _mm_extract_epi32(pos4, 1);
 206         pos_.i[2] = _mm_extract_epi32(pos4, 2);
 207         pos_.i[3] = _mm_extract_epi32(pos4, 3);
 208         frac_.i[0] = _mm_extract_epi32(frac4, 0);
 209         frac_.i[1] = _mm_extract_epi32(frac4, 1);
 210         frac_.i[2] = _mm_extract_epi32(frac4, 2);
 211         frac_.i[3] = _mm_extract_epi32(frac4, 3);
 212     }
 213
 214     pos = pos_.i[0];
 215     frac = frac_.i[0];
 216
 217     for(;i < numsamples;i++)
 218     {
 219         dst[i] = resample_fir8(src[pos  ], src[pos+1], src[pos+2], src[pos+3],
 220                                src[pos+4], src[pos+5], src[pos+6], src[pos+7], frac);
 221
 222         frac += increment;
 223         pos  += frac>>FRACTIONBITS;
 224         frac &= FRACTIONMASK;
 225     }
 226     return dst;
 227 }