yading@10: /*
yading@10:  * AC-3 DSP utils
yading@10:  * Copyright (c) 2011 Justin Ruggles
yading@10:  *
yading@10:  * This file is part of FFmpeg.
yading@10:  *
yading@10:  * FFmpeg is free software; you can redistribute it and/or
yading@10:  * modify it under the terms of the GNU Lesser General Public
yading@10:  * License as published by the Free Software Foundation; either
yading@10:  * version 2.1 of the License, or (at your option) any later version.
yading@10:  *
yading@10:  * FFmpeg is distributed in the hope that it will be useful,
yading@10:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
yading@10:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
yading@10:  * Lesser General Public License for more details.
yading@10:  *
yading@10:  * You should have received a copy of the GNU Lesser General Public
yading@10:  * License along with FFmpeg; if not, write to the Free Software
yading@10:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
yading@10:  */
yading@10: 
yading@10: #include "libavutil/avassert.h"
yading@10: #include "avcodec.h"
yading@10: #include "ac3.h"
yading@10: #include "ac3dsp.h"
yading@10: #include "mathops.h"
yading@10: 
yading@10: static void ac3_exponent_min_c(uint8_t *exp, int num_reuse_blocks, int nb_coefs)
yading@10: {
yading@10:     int blk, i;
yading@10: 
yading@10:     if (!num_reuse_blocks)
yading@10:         return;
yading@10: 
yading@10:     for (i = 0; i < nb_coefs; i++) {
yading@10:         uint8_t min_exp = *exp;
yading@10:         uint8_t *exp1 = exp + 256;
yading@10:         for (blk = 0; blk < num_reuse_blocks; blk++) {
yading@10:             uint8_t next_exp = *exp1;
yading@10:             if (next_exp < min_exp)
yading@10:                 min_exp = next_exp;
yading@10:             exp1 += 256;
yading@10:         }
yading@10:         *exp++ = min_exp;
yading@10:     }
yading@10: }
yading@10: 
yading@10: static int ac3_max_msb_abs_int16_c(const int16_t *src, int len)
yading@10: {
yading@10:     int i, v = 0;
yading@10:     for (i = 0; i < len; i++)
yading@10:         v |= abs(src[i]);
yading@10:     return v;
yading@10: }
yading@10: 
yading@10: static void ac3_lshift_int16_c(int16_t *src, unsigned int len,
yading@10:                                unsigned int shift)
yading@10: {
yading@10:     uint32_t *src32 = (uint32_t *)src;
yading@10:     const uint32_t mask = ~(((1 << shift) - 1) << 16);
yading@10:     int i;
yading@10:     len >>= 1;
yading@10:     for (i = 0; i < len; i += 8) {
yading@10:         src32[i  ] = (src32[i  ] << shift) & mask;
yading@10:         src32[i+1] = (src32[i+1] << shift) & mask;
yading@10:         src32[i+2] = (src32[i+2] << shift) & mask;
yading@10:         src32[i+3] = (src32[i+3] << shift) & mask;
yading@10:         src32[i+4] = (src32[i+4] << shift) & mask;
yading@10:         src32[i+5] = (src32[i+5] << shift) & mask;
yading@10:         src32[i+6] = (src32[i+6] << shift) & mask;
yading@10:         src32[i+7] = (src32[i+7] << shift) & mask;
yading@10:     }
yading@10: }
yading@10: 
yading@10: static void ac3_rshift_int32_c(int32_t *src, unsigned int len,
yading@10:                                unsigned int shift)
yading@10: {
yading@10:     do {
yading@10:         *src++ >>= shift;
yading@10:         *src++ >>= shift;
yading@10:         *src++ >>= shift;
yading@10:         *src++ >>= shift;
yading@10:         *src++ >>= shift;
yading@10:         *src++ >>= shift;
yading@10:         *src++ >>= shift;
yading@10:         *src++ >>= shift;
yading@10:         len -= 8;
yading@10:     } while (len > 0);
yading@10: }
yading@10: 
yading@10: static void float_to_fixed24_c(int32_t *dst, const float *src, unsigned int len)
yading@10: {
yading@10:     const float scale = 1 << 24;
yading@10:     do {
yading@10:         *dst++ = lrintf(*src++ * scale);
yading@10:         *dst++ = lrintf(*src++ * scale);
yading@10:         *dst++ = lrintf(*src++ * scale);
yading@10:         *dst++ = lrintf(*src++ * scale);
yading@10:         *dst++ = lrintf(*src++ * scale);
yading@10:         *dst++ = lrintf(*src++ * scale);
yading@10:         *dst++ = lrintf(*src++ * scale);
yading@10:         *dst++ = lrintf(*src++ * scale);
yading@10:         len -= 8;
yading@10:     } while (len > 0);
yading@10: }
yading@10: 
yading@10: static void ac3_bit_alloc_calc_bap_c(int16_t *mask, int16_t *psd,
yading@10:                                      int start, int end,
yading@10:                                      int snr_offset, int floor,
yading@10:                                      const uint8_t *bap_tab, uint8_t *bap)
yading@10: {
yading@10:     int bin, band, band_end;
yading@10: 
yading@10:     /* special case, if snr offset is -960, set all bap's to zero */
yading@10:     if (snr_offset == -960) {
yading@10:         memset(bap, 0, AC3_MAX_COEFS);
yading@10:         return;
yading@10:     }
yading@10: 
yading@10:     bin  = start;
yading@10:     band = ff_ac3_bin_to_band_tab[start];
yading@10:     do {
yading@10:         int m = (FFMAX(mask[band] - snr_offset - floor, 0) & 0x1FE0) + floor;
yading@10:         band_end = ff_ac3_band_start_tab[++band];
yading@10:         band_end = FFMIN(band_end, end);
yading@10: 
yading@10:         for (; bin < band_end; bin++) {
yading@10:             int address = av_clip((psd[bin] - m) >> 5, 0, 63);
yading@10:             bap[bin] = bap_tab[address];
yading@10:         }
yading@10:     } while (end > band_end);
yading@10: }
yading@10: 
yading@10: static void ac3_update_bap_counts_c(uint16_t mant_cnt[16], uint8_t *bap,
yading@10:                                     int len)
yading@10: {
yading@10:     while (len-- > 0)
yading@10:         mant_cnt[bap[len]]++;
yading@10: }
yading@10: 
yading@10: DECLARE_ALIGNED(16, const uint16_t, ff_ac3_bap_bits)[16] = {
yading@10:     0,  0,  0,  3,  0,  4,  5,  6,  7,  8,  9, 10, 11, 12, 14, 16
yading@10: };
yading@10: 
yading@10: static int ac3_compute_mantissa_size_c(uint16_t mant_cnt[6][16])
yading@10: {
yading@10:     int blk, bap;
yading@10:     int bits = 0;
yading@10: 
yading@10:     for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
yading@10:         // bap=1 : 3 mantissas in 5 bits
yading@10:         bits += (mant_cnt[blk][1] / 3) * 5;
yading@10:         // bap=2 : 3 mantissas in 7 bits
yading@10:         // bap=4 : 2 mantissas in 7 bits
yading@10:         bits += ((mant_cnt[blk][2] / 3) + (mant_cnt[blk][4] >> 1)) * 7;
yading@10:         // bap=3 : 1 mantissa in 3 bits
yading@10:         bits += mant_cnt[blk][3] * 3;
yading@10:         // bap=5 to 15 : get bits per mantissa from table
yading@10:         for (bap = 5; bap < 16; bap++)
yading@10:             bits += mant_cnt[blk][bap] * ff_ac3_bap_bits[bap];
yading@10:     }
yading@10:     return bits;
yading@10: }
yading@10: 
yading@10: static void ac3_extract_exponents_c(uint8_t *exp, int32_t *coef, int nb_coefs)
yading@10: {
yading@10:     int i;
yading@10: 
yading@10:     for (i = 0; i < nb_coefs; i++) {
yading@10:         int v = abs(coef[i]);
yading@10:         exp[i] = v ? 23 - av_log2(v) : 24;
yading@10:     }
yading@10: }
yading@10: 
yading@10: static void ac3_sum_square_butterfly_int32_c(int64_t sum[4],
yading@10:                                              const int32_t *coef0,
yading@10:                                              const int32_t *coef1,
yading@10:                                              int len)
yading@10: {
yading@10:     int i;
yading@10: 
yading@10:     sum[0] = sum[1] = sum[2] = sum[3] = 0;
yading@10: 
yading@10:     for (i = 0; i < len; i++) {
yading@10:         int lt = coef0[i];
yading@10:         int rt = coef1[i];
yading@10:         int md = lt + rt;
yading@10:         int sd = lt - rt;
yading@10:         MAC64(sum[0], lt, lt);
yading@10:         MAC64(sum[1], rt, rt);
yading@10:         MAC64(sum[2], md, md);
yading@10:         MAC64(sum[3], sd, sd);
yading@10:     }
yading@10: }
yading@10: 
yading@10: static void ac3_sum_square_butterfly_float_c(float sum[4],
yading@10:                                              const float *coef0,
yading@10:                                              const float *coef1,
yading@10:                                              int len)
yading@10: {
yading@10:     int i;
yading@10: 
yading@10:     sum[0] = sum[1] = sum[2] = sum[3] = 0;
yading@10: 
yading@10:     for (i = 0; i < len; i++) {
yading@10:         float lt = coef0[i];
yading@10:         float rt = coef1[i];
yading@10:         float md = lt + rt;
yading@10:         float sd = lt - rt;
yading@10:         sum[0] += lt * lt;
yading@10:         sum[1] += rt * rt;
yading@10:         sum[2] += md * md;
yading@10:         sum[3] += sd * sd;
yading@10:     }
yading@10: }
yading@10: 
yading@10: static void ac3_downmix_c(float **samples, float (*matrix)[2],
yading@10:                           int out_ch, int in_ch, int len)
yading@10: {
yading@10:     int i, j;
yading@10:     float v0, v1;
yading@10:     if (out_ch == 2) {
yading@10:         for (i = 0; i < len; i++) {
yading@10:             v0 = v1 = 0.0f;
yading@10:             for (j = 0; j < in_ch; j++) {
yading@10:                 v0 += samples[j][i] * matrix[j][0];
yading@10:                 v1 += samples[j][i] * matrix[j][1];
yading@10:             }
yading@10:             samples[0][i] = v0;
yading@10:             samples[1][i] = v1;
yading@10:         }
yading@10:     } else if (out_ch == 1) {
yading@10:         for (i = 0; i < len; i++) {
yading@10:             v0 = 0.0f;
yading@10:             for (j = 0; j < in_ch; j++)
yading@10:                 v0 += samples[j][i] * matrix[j][0];
yading@10:             samples[0][i] = v0;
yading@10:         }
yading@10:     }
yading@10: }
yading@10: 
yading@10: av_cold void ff_ac3dsp_init(AC3DSPContext *c, int bit_exact)
yading@10: {
yading@10:     c->ac3_exponent_min = ac3_exponent_min_c;
yading@10:     c->ac3_max_msb_abs_int16 = ac3_max_msb_abs_int16_c;
yading@10:     c->ac3_lshift_int16 = ac3_lshift_int16_c;
yading@10:     c->ac3_rshift_int32 = ac3_rshift_int32_c;
yading@10:     c->float_to_fixed24 = float_to_fixed24_c;
yading@10:     c->bit_alloc_calc_bap = ac3_bit_alloc_calc_bap_c;
yading@10:     c->update_bap_counts = ac3_update_bap_counts_c;
yading@10:     c->compute_mantissa_size = ac3_compute_mantissa_size_c;
yading@10:     c->extract_exponents = ac3_extract_exponents_c;
yading@10:     c->sum_square_butterfly_int32 = ac3_sum_square_butterfly_int32_c;
yading@10:     c->sum_square_butterfly_float = ac3_sum_square_butterfly_float_c;
yading@10:     c->downmix = ac3_downmix_c;
yading@10: 
yading@10:     if (ARCH_ARM)
yading@10:         ff_ac3dsp_init_arm(c, bit_exact);
yading@10:     if (ARCH_X86)
yading@10:         ff_ac3dsp_init_x86(c, bit_exact);
yading@10:     if (ARCH_MIPS)
yading@10:         ff_ac3dsp_init_mips(c, bit_exact);
yading@10: }