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 
00029 #if CONFIG_FLOAT
00030 #define RENAME(n) n##_float
00031 
00032 static inline float round_sample(float *sum)
00033 {
00034     float sum1=*sum;
00035     *sum = 0;
00036     return sum1;
00037 }
00038 
00039 #define MACS(rt, ra, rb) rt+=(ra)*(rb)
00040 #define MULS(ra, rb) ((ra)*(rb))
00041 #define MULH3(x, y, s) ((s)*(y)*(x))
00042 #define MLSS(rt, ra, rb) rt-=(ra)*(rb)
00043 #define MULLx(x, y, s) ((y)*(x))
00044 #define FIXHR(x)        ((float)(x))
00045 #define FIXR(x)        ((float)(x))
00046 #define SHR(a,b)       ((a)*(1.0f/(1<<(b))))
00047 
00048 #else
00049 
00050 #define RENAME(n) n##_fixed
00051 #define OUT_SHIFT (WFRAC_BITS + FRAC_BITS - 15)
00052 
00053 static inline int round_sample(int64_t *sum)
00054 {
00055     int sum1;
00056     sum1 = (int)((*sum) >> OUT_SHIFT);
00057     *sum &= (1<<OUT_SHIFT)-1;
00058     return av_clip_int16(sum1);
00059 }
00060 
00061 #   define MULS(ra, rb) MUL64(ra, rb)
00062 #   define MACS(rt, ra, rb) MAC64(rt, ra, rb)
00063 #   define MLSS(rt, ra, rb) MLS64(rt, ra, rb)
00064 #   define MULH3(x, y, s) MULH((s)*(x), y)
00065 #   define MULLx(x, y, s) MULL(x,y,s)
00066 #   define SHR(a,b)       ((a)>>(b))
00067 #   define FIXR(a)        ((int)((a) * FRAC_ONE + 0.5))
00068 #   define FIXHR(a)       ((int)((a) * (1LL<<32) + 0.5))
00069 #endif
00070 
00075 DECLARE_ALIGNED(16, INTFLOAT, RENAME(ff_mdct_win))[8][MDCT_BUF_SIZE];
00076 
00077 DECLARE_ALIGNED(16, MPA_INT, RENAME(ff_mpa_synth_window))[512+256];
00078 
00079 #define SUM8(op, sum, w, p)               \
00080 {                                         \
00081     op(sum, (w)[0 * 64], (p)[0 * 64]);    \
00082     op(sum, (w)[1 * 64], (p)[1 * 64]);    \
00083     op(sum, (w)[2 * 64], (p)[2 * 64]);    \
00084     op(sum, (w)[3 * 64], (p)[3 * 64]);    \
00085     op(sum, (w)[4 * 64], (p)[4 * 64]);    \
00086     op(sum, (w)[5 * 64], (p)[5 * 64]);    \
00087     op(sum, (w)[6 * 64], (p)[6 * 64]);    \
00088     op(sum, (w)[7 * 64], (p)[7 * 64]);    \
00089 }
00090 
00091 #define SUM8P2(sum1, op1, sum2, op2, w1, w2, p) \
00092 {                                               \
00093     INTFLOAT tmp;\
00094     tmp = p[0 * 64];\
00095     op1(sum1, (w1)[0 * 64], tmp);\
00096     op2(sum2, (w2)[0 * 64], tmp);\
00097     tmp = p[1 * 64];\
00098     op1(sum1, (w1)[1 * 64], tmp);\
00099     op2(sum2, (w2)[1 * 64], tmp);\
00100     tmp = p[2 * 64];\
00101     op1(sum1, (w1)[2 * 64], tmp);\
00102     op2(sum2, (w2)[2 * 64], tmp);\
00103     tmp = p[3 * 64];\
00104     op1(sum1, (w1)[3 * 64], tmp);\
00105     op2(sum2, (w2)[3 * 64], tmp);\
00106     tmp = p[4 * 64];\
00107     op1(sum1, (w1)[4 * 64], tmp);\
00108     op2(sum2, (w2)[4 * 64], tmp);\
00109     tmp = p[5 * 64];\
00110     op1(sum1, (w1)[5 * 64], tmp);\
00111     op2(sum2, (w2)[5 * 64], tmp);\
00112     tmp = p[6 * 64];\
00113     op1(sum1, (w1)[6 * 64], tmp);\
00114     op2(sum2, (w2)[6 * 64], tmp);\
00115     tmp = p[7 * 64];\
00116     op1(sum1, (w1)[7 * 64], tmp);\
00117     op2(sum2, (w2)[7 * 64], tmp);\
00118 }
00119 
00120 void RENAME(ff_mpadsp_apply_window)(MPA_INT *synth_buf, MPA_INT *window,
00121                                   int *dither_state, OUT_INT *samples,
00122                                   int incr)
00123 {
00124     register const MPA_INT *w, *w2, *p;
00125     int j;
00126     OUT_INT *samples2;
00127 #if CONFIG_FLOAT
00128     float sum, sum2;
00129 #else
00130     int64_t sum, sum2;
00131 #endif
00132 
00133     /* copy to avoid wrap */
00134     memcpy(synth_buf + 512, synth_buf, 32 * sizeof(*synth_buf));
00135 
00136     samples2 = samples + 31 * incr;
00137     w = window;
00138     w2 = window + 31;
00139 
00140     sum = *dither_state;
00141     p = synth_buf + 16;
00142     SUM8(MACS, sum, w, p);
00143     p = synth_buf + 48;
00144     SUM8(MLSS, sum, w + 32, p);
00145     *samples = round_sample(&sum);
00146     samples += incr;
00147     w++;
00148 
00149     /* we calculate two samples at the same time to avoid one memory
00150        access per two sample */
00151     for(j=1;j<16;j++) {
00152         sum2 = 0;
00153         p = synth_buf + 16 + j;
00154         SUM8P2(sum, MACS, sum2, MLSS, w, w2, p);
00155         p = synth_buf + 48 - j;
00156         SUM8P2(sum, MLSS, sum2, MLSS, w + 32, w2 + 32, p);
00157 
00158         *samples = round_sample(&sum);
00159         samples += incr;
00160         sum += sum2;
00161         *samples2 = round_sample(&sum);
00162         samples2 -= incr;
00163         w++;
00164         w2--;
00165     }
00166 
00167     p = synth_buf + 32;
00168     SUM8(MLSS, sum, w + 32, p);
00169     *samples = round_sample(&sum);
00170     *dither_state= sum;
00171 }
00172 
00173 /* 32 sub band synthesis filter. Input: 32 sub band samples, Output:
00174    32 samples. */
00175 void RENAME(ff_mpa_synth_filter)(MPADSPContext *s, MPA_INT *synth_buf_ptr,
00176                                  int *synth_buf_offset,
00177                                  MPA_INT *window, int *dither_state,
00178                                  OUT_INT *samples, int incr,
00179                                  MPA_INT *sb_samples)
00180 {
00181     MPA_INT *synth_buf;
00182     int offset;
00183 
00184     offset = *synth_buf_offset;
00185     synth_buf = synth_buf_ptr + offset;
00186 
00187     s->RENAME(dct32)(synth_buf, sb_samples);
00188     s->RENAME(apply_window)(synth_buf, window, dither_state, samples, incr);
00189 
00190     offset = (offset - 32) & 511;
00191     *synth_buf_offset = offset;
00192 }
00193 
00194 av_cold void RENAME(ff_mpa_synth_init)(MPA_INT *window)
00195 {
00196     int i, j;
00197 
00198     /* max = 18760, max sum over all 16 coefs : 44736 */
00199     for(i=0;i<257;i++) {
00200         INTFLOAT v;
00201         v = ff_mpa_enwindow[i];
00202 #if CONFIG_FLOAT
00203         v *= 1.0 / (1LL<<(16 + FRAC_BITS));
00204 #endif
00205         window[i] = v;
00206         if ((i & 63) != 0)
00207             v = -v;
00208         if (i != 0)
00209             window[512 - i] = v;
00210     }
00211 
00212 
00213     // Needed for avoiding shuffles in ASM implementations
00214     for(i=0; i < 8; i++)
00215         for(j=0; j < 16; j++)
00216             window[512+16*i+j] = window[64*i+32-j];
00217 
00218     for(i=0; i < 8; i++)
00219         for(j=0; j < 16; j++)
00220             window[512+128+16*i+j] = window[64*i+48-j];
00221 }
00222 
00223 void RENAME(ff_init_mpadsp_tabs)(void)
00224 {
00225     int i, j;
00226     /* compute mdct windows */
00227     for (i = 0; i < 36; i++) {
00228         for (j = 0; j < 4; j++) {
00229             double d;
00230 
00231             if (j == 2 && i % 3 != 1)
00232                 continue;
00233 
00234             d = sin(M_PI * (i + 0.5) / 36.0);
00235             if (j == 1) {
00236                 if      (i >= 30) d = 0;
00237                 else if (i >= 24) d = sin(M_PI * (i - 18 + 0.5) / 12.0);
00238                 else if (i >= 18) d = 1;
00239             } else if (j == 3) {
00240                 if      (i <   6) d = 0;
00241                 else if (i <  12) d = sin(M_PI * (i -  6 + 0.5) / 12.0);
00242                 else if (i <  18) d = 1;
00243             }
00244             //merge last stage of imdct into the window coefficients
00245             d *= 0.5 * IMDCT_SCALAR / cos(M_PI * (2 * i + 19) / 72);
00246 
00247             if (j == 2)
00248                 RENAME(ff_mdct_win)[j][i/3] = FIXHR((d / (1<<5)));
00249             else {
00250                 int idx = i < 18 ? i : i + (MDCT_BUF_SIZE/2 - 18);
00251                 RENAME(ff_mdct_win)[j][idx] = FIXHR((d / (1<<5)));
00252             }
00253         }
00254     }
00255 
00256     /* NOTE: we do frequency inversion adter the MDCT by changing
00257         the sign of the right window coefs */
00258     for (j = 0; j < 4; j++) {
00259         for (i = 0; i < MDCT_BUF_SIZE; i += 2) {
00260             RENAME(ff_mdct_win)[j + 4][i    ] =  RENAME(ff_mdct_win)[j][i    ];
00261             RENAME(ff_mdct_win)[j + 4][i + 1] = -RENAME(ff_mdct_win)[j][i + 1];
00262         }
00263     }
00264 }
00265 /* cos(pi*i/18) */
00266 #define C1 FIXHR(0.98480775301220805936/2)
00267 #define C2 FIXHR(0.93969262078590838405/2)
00268 #define C3 FIXHR(0.86602540378443864676/2)
00269 #define C4 FIXHR(0.76604444311897803520/2)
00270 #define C5 FIXHR(0.64278760968653932632/2)
00271 #define C6 FIXHR(0.5/2)
00272 #define C7 FIXHR(0.34202014332566873304/2)
00273 #define C8 FIXHR(0.17364817766693034885/2)
00274 
00275 /* 0.5 / cos(pi*(2*i+1)/36) */
00276 static const INTFLOAT icos36[9] = {
00277     FIXR(0.50190991877167369479),
00278     FIXR(0.51763809020504152469), //0
00279     FIXR(0.55168895948124587824),
00280     FIXR(0.61038729438072803416),
00281     FIXR(0.70710678118654752439), //1
00282     FIXR(0.87172339781054900991),
00283     FIXR(1.18310079157624925896),
00284     FIXR(1.93185165257813657349), //2
00285     FIXR(5.73685662283492756461),
00286 };
00287 
00288 /* 0.5 / cos(pi*(2*i+1)/36) */
00289 static const INTFLOAT icos36h[9] = {
00290     FIXHR(0.50190991877167369479/2),
00291     FIXHR(0.51763809020504152469/2), //0
00292     FIXHR(0.55168895948124587824/2),
00293     FIXHR(0.61038729438072803416/2),
00294     FIXHR(0.70710678118654752439/2), //1
00295     FIXHR(0.87172339781054900991/2),
00296     FIXHR(1.18310079157624925896/4),
00297     FIXHR(1.93185165257813657349/4), //2
00298 //    FIXHR(5.73685662283492756461),
00299 };
00300 
00301 /* using Lee like decomposition followed by hand coded 9 points DCT */
00302 static void imdct36(INTFLOAT *out, INTFLOAT *buf, INTFLOAT *in, INTFLOAT *win)
00303 {
00304     int i, j;
00305     INTFLOAT t0, t1, t2, t3, s0, s1, s2, s3;
00306     INTFLOAT tmp[18], *tmp1, *in1;
00307 
00308     for (i = 17; i >= 1; i--)
00309         in[i] += in[i-1];
00310     for (i = 17; i >= 3; i -= 2)
00311         in[i] += in[i-2];
00312 
00313     for (j = 0; j < 2; j++) {
00314         tmp1 = tmp + j;
00315         in1 = in + j;
00316 
00317         t2 = in1[2*4] + in1[2*8] - in1[2*2];
00318 
00319         t3 = in1[2*0] + SHR(in1[2*6],1);
00320         t1 = in1[2*0] - in1[2*6];
00321         tmp1[ 6] = t1 - SHR(t2,1);
00322         tmp1[16] = t1 + t2;
00323 
00324         t0 = MULH3(in1[2*2] + in1[2*4] ,    C2, 2);
00325         t1 = MULH3(in1[2*4] - in1[2*8] , -2*C8, 1);
00326         t2 = MULH3(in1[2*2] + in1[2*8] ,   -C4, 2);
00327 
00328         tmp1[10] = t3 - t0 - t2;
00329         tmp1[ 2] = t3 + t0 + t1;
00330         tmp1[14] = t3 + t2 - t1;
00331 
00332         tmp1[ 4] = MULH3(in1[2*5] + in1[2*7] - in1[2*1], -C3, 2);
00333         t2 = MULH3(in1[2*1] + in1[2*5],    C1, 2);
00334         t3 = MULH3(in1[2*5] - in1[2*7], -2*C7, 1);
00335         t0 = MULH3(in1[2*3], C3, 2);
00336 
00337         t1 = MULH3(in1[2*1] + in1[2*7],   -C5, 2);
00338 
00339         tmp1[ 0] = t2 + t3 + t0;
00340         tmp1[12] = t2 + t1 - t0;
00341         tmp1[ 8] = t3 - t1 - t0;
00342     }
00343 
00344     i = 0;
00345     for (j = 0; j < 4; j++) {
00346         t0 = tmp[i];
00347         t1 = tmp[i + 2];
00348         s0 = t1 + t0;
00349         s2 = t1 - t0;
00350 
00351         t2 = tmp[i + 1];
00352         t3 = tmp[i + 3];
00353         s1 = MULH3(t3 + t2, icos36h[    j], 2);
00354         s3 = MULLx(t3 - t2, icos36 [8 - j], FRAC_BITS);
00355 
00356         t0 = s0 + s1;
00357         t1 = s0 - s1;
00358         out[(9 + j) * SBLIMIT] = MULH3(t1, win[     9 + j], 1) + buf[4*(9 + j)];
00359         out[(8 - j) * SBLIMIT] = MULH3(t1, win[     8 - j], 1) + buf[4*(8 - j)];
00360         buf[4 * ( 9 + j     )] = MULH3(t0, win[MDCT_BUF_SIZE/2 + 9 + j], 1);
00361         buf[4 * ( 8 - j     )] = MULH3(t0, win[MDCT_BUF_SIZE/2 + 8 - j], 1);
00362 
00363         t0 = s2 + s3;
00364         t1 = s2 - s3;
00365         out[(9 + 8 - j) * SBLIMIT] = MULH3(t1, win[     9 + 8 - j], 1) + buf[4*(9 + 8 - j)];
00366         out[         j  * SBLIMIT] = MULH3(t1, win[             j], 1) + buf[4*(        j)];
00367         buf[4 * ( 9 + 8 - j     )] = MULH3(t0, win[MDCT_BUF_SIZE/2 + 9 + 8 - j], 1);
00368         buf[4 * (         j     )] = MULH3(t0, win[MDCT_BUF_SIZE/2         + j], 1);
00369         i += 4;
00370     }
00371 
00372     s0 = tmp[16];
00373     s1 = MULH3(tmp[17], icos36h[4], 2);
00374     t0 = s0 + s1;
00375     t1 = s0 - s1;
00376     out[(9 + 4) * SBLIMIT] = MULH3(t1, win[     9 + 4], 1) + buf[4*(9 + 4)];
00377     out[(8 - 4) * SBLIMIT] = MULH3(t1, win[     8 - 4], 1) + buf[4*(8 - 4)];
00378     buf[4 * ( 9 + 4     )] = MULH3(t0, win[MDCT_BUF_SIZE/2 + 9 + 4], 1);
00379     buf[4 * ( 8 - 4     )] = MULH3(t0, win[MDCT_BUF_SIZE/2 + 8 - 4], 1);
00380 }
00381 
00382 void RENAME(ff_imdct36_blocks)(INTFLOAT *out, INTFLOAT *buf, INTFLOAT *in,
00383                                int count, int switch_point, int block_type)
00384 {
00385     int j;
00386     for (j=0 ; j < count; j++) {
00387         /* apply window & overlap with previous buffer */
00388 
00389         /* select window */
00390         int win_idx = (switch_point && j < 2) ? 0 : block_type;
00391         INTFLOAT *win = RENAME(ff_mdct_win)[win_idx + (4 & -(j & 1))];
00392 
00393         imdct36(out, buf, in, win);
00394 
00395         in  += 18;
00396         buf += ((j&3) != 3 ? 1 : (72-3));
00397         out++;
00398     }
00399 }
00400