Libav 0.7.1
libavcodec/ac3dsp.c
Go to the documentation of this file.
00001 /*
00002  * AC-3 DSP utils
00003  * Copyright (c) 2011 Justin Ruggles
00004  *
00005  * This file is part of Libav.
00006  *
00007  * Libav is free software; you can redistribute it and/or
00008  * modify it under the terms of the GNU Lesser General Public
00009  * License as published by the Free Software Foundation; either
00010  * version 2.1 of the License, or (at your option) any later version.
00011  *
00012  * Libav is distributed in the hope that it will be useful,
00013  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00014  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00015  * Lesser General Public License for more details.
00016  *
00017  * You should have received a copy of the GNU Lesser General Public
00018  * License along with Libav; if not, write to the Free Software
00019  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
00020  */
00021 
00022 #include "libavutil/avassert.h"
00023 #include "avcodec.h"
00024 #include "ac3.h"
00025 #include "ac3dsp.h"
00026 
00027 static void ac3_exponent_min_c(uint8_t *exp, int num_reuse_blocks, int nb_coefs)
00028 {
00029     int blk, i;
00030 
00031     if (!num_reuse_blocks)
00032         return;
00033 
00034     for (i = 0; i < nb_coefs; i++) {
00035         uint8_t min_exp = *exp;
00036         uint8_t *exp1 = exp + 256;
00037         for (blk = 0; blk < num_reuse_blocks; blk++) {
00038             uint8_t next_exp = *exp1;
00039             if (next_exp < min_exp)
00040                 min_exp = next_exp;
00041             exp1 += 256;
00042         }
00043         *exp++ = min_exp;
00044     }
00045 }
00046 
00047 static int ac3_max_msb_abs_int16_c(const int16_t *src, int len)
00048 {
00049     int i, v = 0;
00050     for (i = 0; i < len; i++)
00051         v |= abs(src[i]);
00052     return v;
00053 }
00054 
00055 static void ac3_lshift_int16_c(int16_t *src, unsigned int len,
00056                                unsigned int shift)
00057 {
00058     uint32_t *src32 = (uint32_t *)src;
00059     const uint32_t mask = ~(((1 << shift) - 1) << 16);
00060     int i;
00061     len >>= 1;
00062     for (i = 0; i < len; i += 8) {
00063         src32[i  ] = (src32[i  ] << shift) & mask;
00064         src32[i+1] = (src32[i+1] << shift) & mask;
00065         src32[i+2] = (src32[i+2] << shift) & mask;
00066         src32[i+3] = (src32[i+3] << shift) & mask;
00067         src32[i+4] = (src32[i+4] << shift) & mask;
00068         src32[i+5] = (src32[i+5] << shift) & mask;
00069         src32[i+6] = (src32[i+6] << shift) & mask;
00070         src32[i+7] = (src32[i+7] << shift) & mask;
00071     }
00072 }
00073 
00074 static void ac3_rshift_int32_c(int32_t *src, unsigned int len,
00075                                unsigned int shift)
00076 {
00077     do {
00078         *src++ >>= shift;
00079         *src++ >>= shift;
00080         *src++ >>= shift;
00081         *src++ >>= shift;
00082         *src++ >>= shift;
00083         *src++ >>= shift;
00084         *src++ >>= shift;
00085         *src++ >>= shift;
00086         len -= 8;
00087     } while (len > 0);
00088 }
00089 
00090 static void float_to_fixed24_c(int32_t *dst, const float *src, unsigned int len)
00091 {
00092     const float scale = 1 << 24;
00093     do {
00094         *dst++ = lrintf(*src++ * scale);
00095         *dst++ = lrintf(*src++ * scale);
00096         *dst++ = lrintf(*src++ * scale);
00097         *dst++ = lrintf(*src++ * scale);
00098         *dst++ = lrintf(*src++ * scale);
00099         *dst++ = lrintf(*src++ * scale);
00100         *dst++ = lrintf(*src++ * scale);
00101         *dst++ = lrintf(*src++ * scale);
00102         len -= 8;
00103     } while (len > 0);
00104 }
00105 
00106 static void ac3_bit_alloc_calc_bap_c(int16_t *mask, int16_t *psd,
00107                                      int start, int end,
00108                                      int snr_offset, int floor,
00109                                      const uint8_t *bap_tab, uint8_t *bap)
00110 {
00111     int bin, band, band_end;
00112 
00113     /* special case, if snr offset is -960, set all bap's to zero */
00114     if (snr_offset == -960) {
00115         memset(bap, 0, AC3_MAX_COEFS);
00116         return;
00117     }
00118 
00119     bin  = start;
00120     band = ff_ac3_bin_to_band_tab[start];
00121     do {
00122         int m = (FFMAX(mask[band] - snr_offset - floor, 0) & 0x1FE0) + floor;
00123         band_end = ff_ac3_band_start_tab[++band];
00124         band_end = FFMIN(band_end, end);
00125 
00126         for (; bin < band_end; bin++) {
00127             int address = av_clip((psd[bin] - m) >> 5, 0, 63);
00128             bap[bin] = bap_tab[address];
00129         }
00130     } while (end > band_end);
00131 }
00132 
00133 static void ac3_update_bap_counts_c(uint16_t mant_cnt[16], uint8_t *bap,
00134                                     int len)
00135 {
00136     while (len-- > 0)
00137         mant_cnt[bap[len]]++;
00138 }
00139 
00140 DECLARE_ALIGNED(16, const uint16_t, ff_ac3_bap_bits)[16] = {
00141     0,  0,  0,  3,  0,  4,  5,  6,  7,  8,  9, 10, 11, 12, 14, 16
00142 };
00143 
00144 static int ac3_compute_mantissa_size_c(uint16_t mant_cnt[6][16])
00145 {
00146     int blk, bap;
00147     int bits = 0;
00148 
00149     for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
00150         // bap=1 : 3 mantissas in 5 bits
00151         bits += (mant_cnt[blk][1] / 3) * 5;
00152         // bap=2 : 3 mantissas in 7 bits
00153         // bap=4 : 2 mantissas in 7 bits
00154         bits += ((mant_cnt[blk][2] / 3) + (mant_cnt[blk][4] >> 1)) * 7;
00155         // bap=3 : 1 mantissa in 3 bits
00156         bits += mant_cnt[blk][3] * 3;
00157         // bap=5 to 15 : get bits per mantissa from table
00158         for (bap = 5; bap < 16; bap++)
00159             bits += mant_cnt[blk][bap] * ff_ac3_bap_bits[bap];
00160     }
00161     return bits;
00162 }
00163 
00164 static void ac3_extract_exponents_c(uint8_t *exp, int32_t *coef, int nb_coefs)
00165 {
00166     int i;
00167 
00168     for (i = 0; i < nb_coefs; i++) {
00169         int e;
00170         int v = abs(coef[i]);
00171         if (v == 0)
00172             e = 24;
00173         else {
00174             e = 23 - av_log2(v);
00175             if (e >= 24) {
00176                 e = 24;
00177                 coef[i] = 0;
00178             } else if (e < 0) {
00179                 e = 0;
00180                 coef[i] = av_clip(coef[i], -16777215, 16777215);
00181             }
00182         }
00183         exp[i] = e;
00184     }
00185 }
00186 
00187 av_cold void ff_ac3dsp_init(AC3DSPContext *c, int bit_exact)
00188 {
00189     c->ac3_exponent_min = ac3_exponent_min_c;
00190     c->ac3_max_msb_abs_int16 = ac3_max_msb_abs_int16_c;
00191     c->ac3_lshift_int16 = ac3_lshift_int16_c;
00192     c->ac3_rshift_int32 = ac3_rshift_int32_c;
00193     c->float_to_fixed24 = float_to_fixed24_c;
00194     c->bit_alloc_calc_bap = ac3_bit_alloc_calc_bap_c;
00195     c->update_bap_counts = ac3_update_bap_counts_c;
00196     c->compute_mantissa_size = ac3_compute_mantissa_size_c;
00197     c->extract_exponents = ac3_extract_exponents_c;
00198 
00199     if (ARCH_ARM)
00200         ff_ac3dsp_init_arm(c, bit_exact);
00201     if (HAVE_MMX)
00202         ff_ac3dsp_init_x86(c, bit_exact);
00203 }