FFmpeg: libavcodec/mpegaudiodsp_template.c Source File

00001 /*
00002  * Copyright (c) 2001, 2002 Fabrice Bellard
00003  *
00004  * This file is part of FFmpeg.
00005  *
00006  * FFmpeg is free software; you can redistribute it and/or
00007  * modify it under the terms of the GNU Lesser General Public
00008  * License as published by the Free Software Foundation; either
00009  * version 2.1 of the License, or (at your option) any later version.
00010  *
00011  * FFmpeg is distributed in the hope that it will be useful,
00012  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00013  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00014  * Lesser General Public License for more details.
00015  *
00016  * You should have received a copy of the GNU Lesser General Public
00017  * License along with FFmpeg; if not, write to the Free Software
00018  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
00019  */
00020 
00021 #include <stdint.h>
00022 
00023 #include "libavutil/mem.h"
00024 #include "dct32.h"
00025 #include "mathops.h"
00026 #include "mpegaudiodsp.h"
00027 #include "mpegaudio.h"
00028 #include "mpegaudiodata.h"
00029 
00030 #if CONFIG_FLOAT
00031 #define RENAME(n) n##_float
00032 
00033 static inline float round_sample(float *sum)
00034 {
00035     float sum1=*sum;
00036     *sum = 0;
00037     return sum1;
00038 }
00039 
00040 #define MACS(rt, ra, rb) rt+=(ra)*(rb)
00041 #define MULS(ra, rb) ((ra)*(rb))
00042 #define MULH3(x, y, s) ((s)*(y)*(x))
00043 #define MLSS(rt, ra, rb) rt-=(ra)*(rb)
00044 #define MULLx(x, y, s) ((y)*(x))
00045 #define FIXHR(x)        ((float)(x))
00046 #define FIXR(x)        ((float)(x))
00047 #define SHR(a,b)       ((a)*(1.0f/(1<<(b))))
00048 
00049 #else
00050 
00051 #define RENAME(n) n##_fixed
00052 #define OUT_SHIFT (WFRAC_BITS + FRAC_BITS - 15)
00053 
00054 static inline int round_sample(int64_t *sum)
00055 {
00056     int sum1;
00057     sum1 = (int)((*sum) >> OUT_SHIFT);
00058     *sum &= (1<<OUT_SHIFT)-1;
00059     return av_clip_int16(sum1);
00060 }
00061 
00062 #   define MULS(ra, rb) MUL64(ra, rb)
00063 #   define MACS(rt, ra, rb) MAC64(rt, ra, rb)
00064 #   define MLSS(rt, ra, rb) MLS64(rt, ra, rb)
00065 #   define MULH3(x, y, s) MULH((s)*(x), y)
00066 #   define MULLx(x, y, s) MULL(x,y,s)
00067 #   define SHR(a,b)       ((a)>>(b))
00068 #   define FIXR(a)        ((int)((a) * FRAC_ONE + 0.5))
00069 #   define FIXHR(a)       ((int)((a) * (1LL<<32) + 0.5))
00070 #endif
00071 
00072 DECLARE_ALIGNED(16, MPA_INT, RENAME(ff_mpa_synth_window))[512+256];
00073 
00074 #define SUM8(op, sum, w, p)               \
00075 {                                         \
00076     op(sum, (w)[0 * 64], (p)[0 * 64]);    \
00077     op(sum, (w)[1 * 64], (p)[1 * 64]);    \
00078     op(sum, (w)[2 * 64], (p)[2 * 64]);    \
00079     op(sum, (w)[3 * 64], (p)[3 * 64]);    \
00080     op(sum, (w)[4 * 64], (p)[4 * 64]);    \
00081     op(sum, (w)[5 * 64], (p)[5 * 64]);    \
00082     op(sum, (w)[6 * 64], (p)[6 * 64]);    \
00083     op(sum, (w)[7 * 64], (p)[7 * 64]);    \
00084 }
00085 
00086 #define SUM8P2(sum1, op1, sum2, op2, w1, w2, p) \
00087 {                                               \
00088     INTFLOAT tmp;\
00089     tmp = p[0 * 64];\
00090     op1(sum1, (w1)[0 * 64], tmp);\
00091     op2(sum2, (w2)[0 * 64], tmp);\
00092     tmp = p[1 * 64];\
00093     op1(sum1, (w1)[1 * 64], tmp);\
00094     op2(sum2, (w2)[1 * 64], tmp);\
00095     tmp = p[2 * 64];\
00096     op1(sum1, (w1)[2 * 64], tmp);\
00097     op2(sum2, (w2)[2 * 64], tmp);\
00098     tmp = p[3 * 64];\
00099     op1(sum1, (w1)[3 * 64], tmp);\
00100     op2(sum2, (w2)[3 * 64], tmp);\
00101     tmp = p[4 * 64];\
00102     op1(sum1, (w1)[4 * 64], tmp);\
00103     op2(sum2, (w2)[4 * 64], tmp);\
00104     tmp = p[5 * 64];\
00105     op1(sum1, (w1)[5 * 64], tmp);\
00106     op2(sum2, (w2)[5 * 64], tmp);\
00107     tmp = p[6 * 64];\
00108     op1(sum1, (w1)[6 * 64], tmp);\
00109     op2(sum2, (w2)[6 * 64], tmp);\
00110     tmp = p[7 * 64];\
00111     op1(sum1, (w1)[7 * 64], tmp);\
00112     op2(sum2, (w2)[7 * 64], tmp);\
00113 }
00114 
00115 void RENAME(ff_mpadsp_apply_window)(MPA_INT *synth_buf, MPA_INT *window,
00116                                   int *dither_state, OUT_INT *samples,
00117                                   int incr)
00118 {
00119     register const MPA_INT *w, *w2, *p;
00120     int j;
00121     OUT_INT *samples2;
00122 #if CONFIG_FLOAT
00123     float sum, sum2;
00124 #else
00125     int64_t sum, sum2;
00126 #endif
00127 
00128     /* copy to avoid wrap */
00129     memcpy(synth_buf + 512, synth_buf, 32 * sizeof(*synth_buf));
00130 
00131     samples2 = samples + 31 * incr;
00132     w = window;
00133     w2 = window + 31;
00134 
00135     sum = *dither_state;
00136     p = synth_buf + 16;
00137     SUM8(MACS, sum, w, p);
00138     p = synth_buf + 48;
00139     SUM8(MLSS, sum, w + 32, p);
00140     *samples = round_sample(&sum);
00141     samples += incr;
00142     w++;
00143 
00144     /* we calculate two samples at the same time to avoid one memory
00145        access per two sample */
00146     for(j=1;j<16;j++) {
00147         sum2 = 0;
00148         p = synth_buf + 16 + j;
00149         SUM8P2(sum, MACS, sum2, MLSS, w, w2, p);
00150         p = synth_buf + 48 - j;
00151         SUM8P2(sum, MLSS, sum2, MLSS, w + 32, w2 + 32, p);
00152 
00153         *samples = round_sample(&sum);
00154         samples += incr;
00155         sum += sum2;
00156         *samples2 = round_sample(&sum);
00157         samples2 -= incr;
00158         w++;
00159         w2--;
00160     }
00161 
00162     p = synth_buf + 32;
00163     SUM8(MLSS, sum, w + 32, p);
00164     *samples = round_sample(&sum);
00165     *dither_state= sum;
00166 }
00167 
00168 /* 32 sub band synthesis filter. Input: 32 sub band samples, Output:
00169    32 samples. */
00170 void RENAME(ff_mpa_synth_filter)(MPADSPContext *s, MPA_INT *synth_buf_ptr,
00171                                  int *synth_buf_offset,
00172                                  MPA_INT *window, int *dither_state,
00173                                  OUT_INT *samples, int incr,
00174                                  MPA_INT *sb_samples)
00175 {
00176     MPA_INT *synth_buf;
00177     int offset;
00178 
00179     offset = *synth_buf_offset;
00180     synth_buf = synth_buf_ptr + offset;
00181 
00182     s->RENAME(dct32)(synth_buf, sb_samples);
00183     s->RENAME(apply_window)(synth_buf, window, dither_state, samples, incr);
00184 
00185     offset = (offset - 32) & 511;
00186     *synth_buf_offset = offset;
00187 }
00188 
00189 void av_cold RENAME(ff_mpa_synth_init)(MPA_INT *window)
00190 {
00191     int i, j;
00192 
00193     /* max = 18760, max sum over all 16 coefs : 44736 */
00194     for(i=0;i<257;i++) {
00195         INTFLOAT v;
00196         v = ff_mpa_enwindow[i];
00197 #if CONFIG_FLOAT
00198         v *= 1.0 / (1LL<<(16 + FRAC_BITS));
00199 #endif
00200         window[i] = v;
00201         if ((i & 63) != 0)
00202             v = -v;
00203         if (i != 0)
00204             window[512 - i] = v;
00205     }
00206 
00207     // Needed for avoiding shuffles in ASM implementations
00208     for(i=0; i < 8; i++)
00209         for(j=0; j < 16; j++)
00210             window[512+16*i+j] = window[64*i+32-j];
00211 
00212     for(i=0; i < 8; i++)
00213         for(j=0; j < 16; j++)
00214             window[512+128+16*i+j] = window[64*i+48-j];
00215 }
00216 
00217 /* cos(pi*i/18) */
00218 #define C1 FIXHR(0.98480775301220805936/2)
00219 #define C2 FIXHR(0.93969262078590838405/2)
00220 #define C3 FIXHR(0.86602540378443864676/2)
00221 #define C4 FIXHR(0.76604444311897803520/2)
00222 #define C5 FIXHR(0.64278760968653932632/2)
00223 #define C6 FIXHR(0.5/2)
00224 #define C7 FIXHR(0.34202014332566873304/2)
00225 #define C8 FIXHR(0.17364817766693034885/2)
00226 
00227 /* 0.5 / cos(pi*(2*i+1)/36) */
00228 static const INTFLOAT icos36[9] = {
00229     FIXR(0.50190991877167369479),
00230     FIXR(0.51763809020504152469),
00231     FIXR(0.55168895948124587824),
00232     FIXR(0.61038729438072803416),
00233     FIXR(0.70710678118654752439),
00234     FIXR(0.87172339781054900991),
00235     FIXR(1.18310079157624925896),
00236     FIXR(1.93185165257813657349),
00237     FIXR(5.73685662283492756461),
00238 };
00239 
00240 /* 0.5 / cos(pi*(2*i+1)/36) */
00241 static const INTFLOAT icos36h[9] = {
00242     FIXHR(0.50190991877167369479/2),
00243     FIXHR(0.51763809020504152469/2),
00244     FIXHR(0.55168895948124587824/2),
00245     FIXHR(0.61038729438072803416/2),
00246     FIXHR(0.70710678118654752439/2),
00247     FIXHR(0.87172339781054900991/2),
00248     FIXHR(1.18310079157624925896/4),
00249     FIXHR(1.93185165257813657349/4),
00250 };
00251 
00252 
00253 /* using Lee like decomposition followed by hand coded 9 points DCT */
00254 void RENAME(ff_imdct36)(INTFLOAT *out, INTFLOAT *buf, INTFLOAT *in,
00255                         INTFLOAT *win)
00256 {
00257     int i, j;
00258     INTFLOAT t0, t1, t2, t3, s0, s1, s2, s3;
00259     INTFLOAT tmp[18], *tmp1, *in1;
00260 
00261     for(i=17;i>=1;i--)
00262         in[i] += in[i-1];
00263     for(i=17;i>=3;i-=2)
00264         in[i] += in[i-2];
00265 
00266     for(j=0;j<2;j++) {
00267         tmp1 = tmp + j;
00268         in1 = in + j;
00269 
00270         t2 = in1[2*4] + in1[2*8] - in1[2*2];
00271 
00272         t3 = in1[2*0] + SHR(in1[2*6],1);
00273         t1 = in1[2*0] - in1[2*6];
00274         tmp1[ 6] = t1 - SHR(t2,1);
00275         tmp1[16] = t1 + t2;
00276 
00277         t0 = MULH3(in1[2*2] + in1[2*4] ,    C2, 2);
00278         t1 = MULH3(in1[2*4] - in1[2*8] , -2*C8, 1);
00279         t2 = MULH3(in1[2*2] + in1[2*8] ,   -C4, 2);
00280 
00281         tmp1[10] = t3 - t0 - t2;
00282         tmp1[ 2] = t3 + t0 + t1;
00283         tmp1[14] = t3 + t2 - t1;
00284 
00285         tmp1[ 4] = MULH3(in1[2*5] + in1[2*7] - in1[2*1], -C3, 2);
00286         t2 = MULH3(in1[2*1] + in1[2*5],    C1, 2);
00287         t3 = MULH3(in1[2*5] - in1[2*7], -2*C7, 1);
00288         t0 = MULH3(in1[2*3], C3, 2);
00289 
00290         t1 = MULH3(in1[2*1] + in1[2*7],   -C5, 2);
00291 
00292         tmp1[ 0] = t2 + t3 + t0;
00293         tmp1[12] = t2 + t1 - t0;
00294         tmp1[ 8] = t3 - t1 - t0;
00295     }
00296 
00297     i = 0;
00298     for(j=0;j<4;j++) {
00299         t0 = tmp[i];
00300         t1 = tmp[i + 2];
00301         s0 = t1 + t0;
00302         s2 = t1 - t0;
00303 
00304         t2 = tmp[i + 1];
00305         t3 = tmp[i + 3];
00306         s1 = MULH3(t3 + t2, icos36h[j], 2);
00307         s3 = MULLx(t3 - t2, icos36[8 - j], FRAC_BITS);
00308 
00309         t0 = s0 + s1;
00310         t1 = s0 - s1;
00311         out[(9 + j)*SBLIMIT] =  MULH3(t1, win[9 + j], 1) + buf[9 + j];
00312         out[(8 - j)*SBLIMIT] =  MULH3(t1, win[8 - j], 1) + buf[8 - j];
00313         buf[9 + j] = MULH3(t0, win[20 + 9 + j], 1);
00314         buf[8 - j] = MULH3(t0, win[20 + 8 - j], 1);
00315 
00316         t0 = s2 + s3;
00317         t1 = s2 - s3;
00318         out[(9 + 8 - j)*SBLIMIT] =  MULH3(t1, win[9 + 8 - j], 1) + buf[9 + 8 - j];
00319         out[(        j)*SBLIMIT] =  MULH3(t1, win[        j], 1) + buf[        j];
00320         buf[9 + 8 - j] = MULH3(t0, win[20 + 9 + 8 - j], 1);
00321         buf[      + j] = MULH3(t0, win[20         + j], 1);
00322         i += 4;
00323     }
00324 
00325     s0 = tmp[16];
00326     s1 = MULH3(tmp[17], icos36h[4], 2);
00327     t0 = s0 + s1;
00328     t1 = s0 - s1;
00329     out[(9 + 4)*SBLIMIT] =  MULH3(t1, win[9 + 4], 1) + buf[9 + 4];
00330     out[(8 - 4)*SBLIMIT] =  MULH3(t1, win[8 - 4], 1) + buf[8 - 4];
00331     buf[9 + 4] = MULH3(t0, win[20 + 9 + 4], 1);
00332     buf[8 - 4] = MULH3(t0, win[20 + 8 - 4], 1);
00333 }
00334