git.sesse.net Git - ffmpeg/blob - libavcodec/ppc/fft_vsx.c

   1 /*
   2  * FFT  transform, optimized with VSX built-in functions
   3  * Copyright (c) 2014 Rong Yan
   4  *
   5  * This algorithm (though not any of the implementation details) is
   6  * based on libdjbfft by D. J. Bernstein.
   7  *
   8  * This file is part of FFmpeg.
   9  *
  10  * FFmpeg is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU Lesser General Public
  12  * License as published by the Free Software Foundation; either
  13  * version 2.1 of the License, or (at your option) any later version.
  14  *
  15  * FFmpeg is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18  * Lesser General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU Lesser General Public
  21  * License along with FFmpeg; if not, write to the Free Software
  22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  23  */
  24
  25
  26 #include "config.h"
  27 #include "libavutil/cpu.h"
  28 #include "libavutil/ppc/util_altivec.h"
  29 #include "libavcodec/fft.h"
  30 #include "libavcodec/fft-internal.h"
  31 #include "fft_vsx.h"
  32
  33 #if HAVE_VSX
  34
  35 static void fft32_vsx_interleave(FFTComplex *z)
  36 {
  37     fft16_vsx_interleave(z);
  38     fft8_vsx_interleave(z+16);
  39     fft8_vsx_interleave(z+24);
  40     pass_vsx_interleave(z,ff_cos_32,4);
  41 }
  42
  43 static void fft64_vsx_interleave(FFTComplex *z)
  44 {
  45     fft32_vsx_interleave(z);
  46     fft16_vsx_interleave(z+32);
  47     fft16_vsx_interleave(z+48);
  48     pass_vsx_interleave(z,ff_cos_64, 8);
  49 }
  50 static void fft128_vsx_interleave(FFTComplex *z)
  51 {
  52     fft64_vsx_interleave(z);
  53     fft32_vsx_interleave(z+64);
  54     fft32_vsx_interleave(z+96);
  55     pass_vsx_interleave(z,ff_cos_128,16);
  56 }
  57 static void fft256_vsx_interleave(FFTComplex *z)
  58 {
  59     fft128_vsx_interleave(z);
  60     fft64_vsx_interleave(z+128);
  61     fft64_vsx_interleave(z+192);
  62     pass_vsx_interleave(z,ff_cos_256,32);
  63 }
  64 static void fft512_vsx_interleave(FFTComplex *z)
  65 {
  66     fft256_vsx_interleave(z);
  67     fft128_vsx_interleave(z+256);
  68     fft128_vsx_interleave(z+384);
  69     pass_vsx_interleave(z,ff_cos_512,64);
  70 }
  71 static void fft1024_vsx_interleave(FFTComplex *z)
  72 {
  73     fft512_vsx_interleave(z);
  74     fft256_vsx_interleave(z+512);
  75     fft256_vsx_interleave(z+768);
  76     pass_vsx_interleave(z,ff_cos_1024,128);
  77
  78 }
  79 static void fft2048_vsx_interleave(FFTComplex *z)
  80 {
  81     fft1024_vsx_interleave(z);
  82     fft512_vsx_interleave(z+1024);
  83     fft512_vsx_interleave(z+1536);
  84     pass_vsx_interleave(z,ff_cos_2048,256);
  85 }
  86 static void fft4096_vsx_interleave(FFTComplex *z)
  87 {
  88     fft2048_vsx_interleave(z);
  89     fft1024_vsx_interleave(z+2048);
  90     fft1024_vsx_interleave(z+3072);
  91     pass_vsx_interleave(z,ff_cos_4096, 512);
  92 }
  93 static void fft8192_vsx_interleave(FFTComplex *z)
  94 {
  95     fft4096_vsx_interleave(z);
  96     fft2048_vsx_interleave(z+4096);
  97     fft2048_vsx_interleave(z+6144);
  98     pass_vsx_interleave(z,ff_cos_8192,1024);
  99 }
 100 static void fft16384_vsx_interleave(FFTComplex *z)
 101 {
 102     fft8192_vsx_interleave(z);
 103     fft4096_vsx_interleave(z+8192);
 104     fft4096_vsx_interleave(z+12288);
 105     pass_vsx_interleave(z,ff_cos_16384,2048);
 106 }
 107 static void fft32768_vsx_interleave(FFTComplex *z)
 108 {
 109     fft16384_vsx_interleave(z);
 110     fft8192_vsx_interleave(z+16384);
 111     fft8192_vsx_interleave(z+24576);
 112     pass_vsx_interleave(z,ff_cos_32768,4096);
 113 }
 114 static void fft65536_vsx_interleave(FFTComplex *z)
 115 {
 116     fft32768_vsx_interleave(z);
 117     fft16384_vsx_interleave(z+32768);
 118     fft16384_vsx_interleave(z+49152);
 119     pass_vsx_interleave(z,ff_cos_65536,8192);
 120 }
 121
 122 static void fft32_vsx(FFTComplex *z)
 123 {
 124     fft16_vsx(z);
 125     fft8_vsx(z+16);
 126     fft8_vsx(z+24);
 127     pass_vsx(z,ff_cos_32,4);
 128 }
 129
 130 static void fft64_vsx(FFTComplex *z)
 131 {
 132     fft32_vsx(z);
 133     fft16_vsx(z+32);
 134     fft16_vsx(z+48);
 135     pass_vsx(z,ff_cos_64, 8);
 136 }
 137 static void fft128_vsx(FFTComplex *z)
 138 {
 139     fft64_vsx(z);
 140     fft32_vsx(z+64);
 141     fft32_vsx(z+96);
 142     pass_vsx(z,ff_cos_128,16);
 143 }
 144 static void fft256_vsx(FFTComplex *z)
 145 {
 146     fft128_vsx(z);
 147     fft64_vsx(z+128);
 148     fft64_vsx(z+192);
 149     pass_vsx(z,ff_cos_256,32);
 150 }
 151 static void fft512_vsx(FFTComplex *z)
 152 {
 153     fft256_vsx(z);
 154     fft128_vsx(z+256);
 155     fft128_vsx(z+384);
 156     pass_vsx(z,ff_cos_512,64);
 157 }
 158 static void fft1024_vsx(FFTComplex *z)
 159 {
 160     fft512_vsx(z);
 161     fft256_vsx(z+512);
 162     fft256_vsx(z+768);
 163     pass_vsx(z,ff_cos_1024,128);
 164
 165 }
 166 static void fft2048_vsx(FFTComplex *z)
 167 {
 168     fft1024_vsx(z);
 169     fft512_vsx(z+1024);
 170     fft512_vsx(z+1536);
 171     pass_vsx(z,ff_cos_2048,256);
 172 }
 173 static void fft4096_vsx(FFTComplex *z)
 174 {
 175     fft2048_vsx(z);
 176     fft1024_vsx(z+2048);
 177     fft1024_vsx(z+3072);
 178     pass_vsx(z,ff_cos_4096, 512);
 179 }
 180 static void fft8192_vsx(FFTComplex *z)
 181 {
 182     fft4096_vsx(z);
 183     fft2048_vsx(z+4096);
 184     fft2048_vsx(z+6144);
 185     pass_vsx(z,ff_cos_8192,1024);
 186 }
 187 static void fft16384_vsx(FFTComplex *z)
 188 {
 189     fft8192_vsx(z);
 190     fft4096_vsx(z+8192);
 191     fft4096_vsx(z+12288);
 192     pass_vsx(z,ff_cos_16384,2048);
 193 }
 194 static void fft32768_vsx(FFTComplex *z)
 195 {
 196     fft16384_vsx(z);
 197     fft8192_vsx(z+16384);
 198     fft8192_vsx(z+24576);
 199     pass_vsx(z,ff_cos_32768,4096);
 200 }
 201 static void fft65536_vsx(FFTComplex *z)
 202 {
 203     fft32768_vsx(z);
 204     fft16384_vsx(z+32768);
 205     fft16384_vsx(z+49152);
 206     pass_vsx(z,ff_cos_65536,8192);
 207 }
 208
 209 static void (* const fft_dispatch_vsx[])(FFTComplex*) = {
 210     fft4_vsx, fft8_vsx, fft16_vsx, fft32_vsx, fft64_vsx, fft128_vsx, fft256_vsx, fft512_vsx, fft1024_vsx,
 211     fft2048_vsx, fft4096_vsx, fft8192_vsx, fft16384_vsx, fft32768_vsx, fft65536_vsx,
 212 };
 213 static void (* const fft_dispatch_vsx_interleave[])(FFTComplex*) = {
 214     fft4_vsx_interleave, fft8_vsx_interleave, fft16_vsx_interleave, fft32_vsx_interleave, fft64_vsx_interleave,
 215     fft128_vsx_interleave, fft256_vsx_interleave, fft512_vsx_interleave, fft1024_vsx_interleave,
 216     fft2048_vsx_interleave, fft4096_vsx_interleave, fft8192_vsx_interleave, fft16384_vsx_interleave, fft32768_vsx_interleave, fft65536_vsx_interleave,
 217 };
 218 void ff_fft_calc_interleave_vsx(FFTContext *s, FFTComplex *z)
 219 {
 220      fft_dispatch_vsx_interleave[s->nbits-2](z);
 221 }
 222 void ff_fft_calc_vsx(FFTContext *s, FFTComplex *z)
 223 {
 224      fft_dispatch_vsx[s->nbits-2](z);
 225 }
 226 #endif /* HAVE_VSX */