yading@10: /* yading@10: * Atrac 3 compatible decoder yading@10: * Copyright (c) 2006-2008 Maxim Poliakovski yading@10: * Copyright (c) 2006-2008 Benjamin Larsson 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: /** yading@10: * @file yading@10: * Atrac 3 compatible decoder. yading@10: * This decoder handles Sony's ATRAC3 data. yading@10: * yading@10: * Container formats used to store atrac 3 data: yading@10: * RealMedia (.rm), RIFF WAV (.wav, .at3), Sony OpenMG (.oma, .aa3). yading@10: * yading@10: * To use this decoder, a calling application must supply the extradata yading@10: * bytes provided in the containers above. yading@10: */ yading@10: yading@10: #include yading@10: #include yading@10: #include yading@10: yading@10: #include "libavutil/float_dsp.h" yading@10: #include "libavutil/libm.h" yading@10: #include "avcodec.h" yading@10: #include "bytestream.h" yading@10: #include "fft.h" yading@10: #include "fmtconvert.h" yading@10: #include "get_bits.h" yading@10: #include "internal.h" yading@10: yading@10: #include "atrac.h" yading@10: #include "atrac3data.h" yading@10: yading@10: #define JOINT_STEREO 0x12 yading@10: #define STEREO 0x2 yading@10: yading@10: #define SAMPLES_PER_FRAME 1024 yading@10: #define MDCT_SIZE 512 yading@10: yading@10: typedef struct GainInfo { yading@10: int num_gain_data; yading@10: int lev_code[8]; yading@10: int loc_code[8]; yading@10: } GainInfo; yading@10: yading@10: typedef struct GainBlock { yading@10: GainInfo g_block[4]; yading@10: } GainBlock; yading@10: yading@10: typedef struct TonalComponent { yading@10: int pos; yading@10: int num_coefs; yading@10: float coef[8]; yading@10: } TonalComponent; yading@10: yading@10: typedef struct ChannelUnit { yading@10: int bands_coded; yading@10: int num_components; yading@10: float prev_frame[SAMPLES_PER_FRAME]; yading@10: int gc_blk_switch; yading@10: TonalComponent components[64]; yading@10: GainBlock gain_block[2]; yading@10: yading@10: DECLARE_ALIGNED(32, float, spectrum)[SAMPLES_PER_FRAME]; yading@10: DECLARE_ALIGNED(32, float, imdct_buf)[SAMPLES_PER_FRAME]; yading@10: yading@10: float delay_buf1[46]; ///mdct_ctx.imdct_calc(&q->mdct_ctx, output, input); yading@10: yading@10: /* Perform windowing on the output. */ yading@10: q->fdsp.vector_fmul(output, output, mdct_window, MDCT_SIZE); yading@10: } yading@10: yading@10: /* yading@10: * indata descrambling, only used for data coming from the rm container yading@10: */ yading@10: static int decode_bytes(const uint8_t *input, uint8_t *out, int bytes) yading@10: { yading@10: int i, off; yading@10: uint32_t c; yading@10: const uint32_t *buf; yading@10: uint32_t *output = (uint32_t *)out; yading@10: yading@10: off = (intptr_t)input & 3; yading@10: buf = (const uint32_t *)(input - off); yading@10: if (off) yading@10: c = av_be2ne32((0x537F6103U >> (off * 8)) | (0x537F6103U << (32 - (off * 8)))); yading@10: else yading@10: c = av_be2ne32(0x537F6103U); yading@10: bytes += 3 + off; yading@10: for (i = 0; i < bytes / 4; i++) yading@10: output[i] = c ^ buf[i]; yading@10: yading@10: if (off) yading@10: avpriv_request_sample(NULL, "Offset of %d", off); yading@10: yading@10: return off; yading@10: } yading@10: yading@10: static av_cold void init_atrac3_window(void) yading@10: { yading@10: int i, j; yading@10: yading@10: /* generate the mdct window, for details see yading@10: * http://wiki.multimedia.cx/index.php?title=RealAudio_atrc#Windows */ yading@10: for (i = 0, j = 255; i < 128; i++, j--) { yading@10: float wi = sin(((i + 0.5) / 256.0 - 0.5) * M_PI) + 1.0; yading@10: float wj = sin(((j + 0.5) / 256.0 - 0.5) * M_PI) + 1.0; yading@10: float w = 0.5 * (wi * wi + wj * wj); yading@10: mdct_window[i] = mdct_window[511 - i] = wi / w; yading@10: mdct_window[j] = mdct_window[511 - j] = wj / w; yading@10: } yading@10: } yading@10: yading@10: static av_cold int atrac3_decode_close(AVCodecContext *avctx) yading@10: { yading@10: ATRAC3Context *q = avctx->priv_data; yading@10: yading@10: av_free(q->units); yading@10: av_free(q->decoded_bytes_buffer); yading@10: yading@10: ff_mdct_end(&q->mdct_ctx); yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: /** yading@10: * Mantissa decoding yading@10: * yading@10: * @param selector which table the output values are coded with yading@10: * @param coding_flag constant length coding or variable length coding yading@10: * @param mantissas mantissa output table yading@10: * @param num_codes number of values to get yading@10: */ yading@10: static void read_quant_spectral_coeffs(GetBitContext *gb, int selector, yading@10: int coding_flag, int *mantissas, yading@10: int num_codes) yading@10: { yading@10: int i, code, huff_symb; yading@10: yading@10: if (selector == 1) yading@10: num_codes /= 2; yading@10: yading@10: if (coding_flag != 0) { yading@10: /* constant length coding (CLC) */ yading@10: int num_bits = clc_length_tab[selector]; yading@10: yading@10: if (selector > 1) { yading@10: for (i = 0; i < num_codes; i++) { yading@10: if (num_bits) yading@10: code = get_sbits(gb, num_bits); yading@10: else yading@10: code = 0; yading@10: mantissas[i] = code; yading@10: } yading@10: } else { yading@10: for (i = 0; i < num_codes; i++) { yading@10: if (num_bits) yading@10: code = get_bits(gb, num_bits); // num_bits is always 4 in this case yading@10: else yading@10: code = 0; yading@10: mantissas[i * 2 ] = mantissa_clc_tab[code >> 2]; yading@10: mantissas[i * 2 + 1] = mantissa_clc_tab[code & 3]; yading@10: } yading@10: } yading@10: } else { yading@10: /* variable length coding (VLC) */ yading@10: if (selector != 1) { yading@10: for (i = 0; i < num_codes; i++) { yading@10: huff_symb = get_vlc2(gb, spectral_coeff_tab[selector-1].table, yading@10: spectral_coeff_tab[selector-1].bits, 3); yading@10: huff_symb += 1; yading@10: code = huff_symb >> 1; yading@10: if (huff_symb & 1) yading@10: code = -code; yading@10: mantissas[i] = code; yading@10: } yading@10: } else { yading@10: for (i = 0; i < num_codes; i++) { yading@10: huff_symb = get_vlc2(gb, spectral_coeff_tab[selector - 1].table, yading@10: spectral_coeff_tab[selector - 1].bits, 3); yading@10: mantissas[i * 2 ] = mantissa_vlc_tab[huff_symb * 2 ]; yading@10: mantissas[i * 2 + 1] = mantissa_vlc_tab[huff_symb * 2 + 1]; yading@10: } yading@10: } yading@10: } yading@10: } yading@10: yading@10: /** yading@10: * Restore the quantized band spectrum coefficients yading@10: * yading@10: * @return subband count, fix for broken specification/files yading@10: */ yading@10: static int decode_spectrum(GetBitContext *gb, float *output) yading@10: { yading@10: int num_subbands, coding_mode, i, j, first, last, subband_size; yading@10: int subband_vlc_index[32], sf_index[32]; yading@10: int mantissas[128]; yading@10: float scale_factor; yading@10: yading@10: num_subbands = get_bits(gb, 5); // number of coded subbands yading@10: coding_mode = get_bits1(gb); // coding Mode: 0 - VLC/ 1-CLC yading@10: yading@10: /* get the VLC selector table for the subbands, 0 means not coded */ yading@10: for (i = 0; i <= num_subbands; i++) yading@10: subband_vlc_index[i] = get_bits(gb, 3); yading@10: yading@10: /* read the scale factor indexes from the stream */ yading@10: for (i = 0; i <= num_subbands; i++) { yading@10: if (subband_vlc_index[i] != 0) yading@10: sf_index[i] = get_bits(gb, 6); yading@10: } yading@10: yading@10: for (i = 0; i <= num_subbands; i++) { yading@10: first = subband_tab[i ]; yading@10: last = subband_tab[i + 1]; yading@10: yading@10: subband_size = last - first; yading@10: yading@10: if (subband_vlc_index[i] != 0) { yading@10: /* decode spectral coefficients for this subband */ yading@10: /* TODO: This can be done faster is several blocks share the yading@10: * same VLC selector (subband_vlc_index) */ yading@10: read_quant_spectral_coeffs(gb, subband_vlc_index[i], coding_mode, yading@10: mantissas, subband_size); yading@10: yading@10: /* decode the scale factor for this subband */ yading@10: scale_factor = ff_atrac_sf_table[sf_index[i]] * yading@10: inv_max_quant[subband_vlc_index[i]]; yading@10: yading@10: /* inverse quantize the coefficients */ yading@10: for (j = 0; first < last; first++, j++) yading@10: output[first] = mantissas[j] * scale_factor; yading@10: } else { yading@10: /* this subband was not coded, so zero the entire subband */ yading@10: memset(output + first, 0, subband_size * sizeof(*output)); yading@10: } yading@10: } yading@10: yading@10: /* clear the subbands that were not coded */ yading@10: first = subband_tab[i]; yading@10: memset(output + first, 0, (SAMPLES_PER_FRAME - first) * sizeof(*output)); yading@10: return num_subbands; yading@10: } yading@10: yading@10: /** yading@10: * Restore the quantized tonal components yading@10: * yading@10: * @param components tonal components yading@10: * @param num_bands number of coded bands yading@10: */ yading@10: static int decode_tonal_components(GetBitContext *gb, yading@10: TonalComponent *components, int num_bands) yading@10: { yading@10: int i, b, c, m; yading@10: int nb_components, coding_mode_selector, coding_mode; yading@10: int band_flags[4], mantissa[8]; yading@10: int component_count = 0; yading@10: yading@10: nb_components = get_bits(gb, 5); yading@10: yading@10: /* no tonal components */ yading@10: if (nb_components == 0) yading@10: return 0; yading@10: yading@10: coding_mode_selector = get_bits(gb, 2); yading@10: if (coding_mode_selector == 2) yading@10: return AVERROR_INVALIDDATA; yading@10: yading@10: coding_mode = coding_mode_selector & 1; yading@10: yading@10: for (i = 0; i < nb_components; i++) { yading@10: int coded_values_per_component, quant_step_index; yading@10: yading@10: for (b = 0; b <= num_bands; b++) yading@10: band_flags[b] = get_bits1(gb); yading@10: yading@10: coded_values_per_component = get_bits(gb, 3); yading@10: yading@10: quant_step_index = get_bits(gb, 3); yading@10: if (quant_step_index <= 1) yading@10: return AVERROR_INVALIDDATA; yading@10: yading@10: if (coding_mode_selector == 3) yading@10: coding_mode = get_bits1(gb); yading@10: yading@10: for (b = 0; b < (num_bands + 1) * 4; b++) { yading@10: int coded_components; yading@10: yading@10: if (band_flags[b >> 2] == 0) yading@10: continue; yading@10: yading@10: coded_components = get_bits(gb, 3); yading@10: yading@10: for (c = 0; c < coded_components; c++) { yading@10: TonalComponent *cmp = &components[component_count]; yading@10: int sf_index, coded_values, max_coded_values; yading@10: float scale_factor; yading@10: yading@10: sf_index = get_bits(gb, 6); yading@10: if (component_count >= 64) yading@10: return AVERROR_INVALIDDATA; yading@10: yading@10: cmp->pos = b * 64 + get_bits(gb, 6); yading@10: yading@10: max_coded_values = SAMPLES_PER_FRAME - cmp->pos; yading@10: coded_values = coded_values_per_component + 1; yading@10: coded_values = FFMIN(max_coded_values, coded_values); yading@10: yading@10: scale_factor = ff_atrac_sf_table[sf_index] * yading@10: inv_max_quant[quant_step_index]; yading@10: yading@10: read_quant_spectral_coeffs(gb, quant_step_index, coding_mode, yading@10: mantissa, coded_values); yading@10: yading@10: cmp->num_coefs = coded_values; yading@10: yading@10: /* inverse quant */ yading@10: for (m = 0; m < coded_values; m++) yading@10: cmp->coef[m] = mantissa[m] * scale_factor; yading@10: yading@10: component_count++; yading@10: } yading@10: } yading@10: } yading@10: yading@10: return component_count; yading@10: } yading@10: yading@10: /** yading@10: * Decode gain parameters for the coded bands yading@10: * yading@10: * @param block the gainblock for the current band yading@10: * @param num_bands amount of coded bands yading@10: */ yading@10: static int decode_gain_control(GetBitContext *gb, GainBlock *block, yading@10: int num_bands) yading@10: { yading@10: int i, cf, num_data; yading@10: int *level, *loc; yading@10: yading@10: GainInfo *gain = block->g_block; yading@10: yading@10: for (i = 0; i <= num_bands; i++) { yading@10: num_data = get_bits(gb, 3); yading@10: gain[i].num_gain_data = num_data; yading@10: level = gain[i].lev_code; yading@10: loc = gain[i].loc_code; yading@10: yading@10: for (cf = 0; cf < gain[i].num_gain_data; cf++) { yading@10: level[cf] = get_bits(gb, 4); yading@10: loc [cf] = get_bits(gb, 5); yading@10: if (cf && loc[cf] <= loc[cf - 1]) yading@10: return AVERROR_INVALIDDATA; yading@10: } yading@10: } yading@10: yading@10: /* Clear the unused blocks. */ yading@10: for (; i < 4 ; i++) yading@10: gain[i].num_gain_data = 0; yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: /** yading@10: * Apply gain parameters and perform the MDCT overlapping part yading@10: * yading@10: * @param input input buffer yading@10: * @param prev previous buffer to perform overlap against yading@10: * @param output output buffer yading@10: * @param gain1 current band gain info yading@10: * @param gain2 next band gain info yading@10: */ yading@10: static void gain_compensate_and_overlap(float *input, float *prev, yading@10: float *output, GainInfo *gain1, yading@10: GainInfo *gain2) yading@10: { yading@10: float g1, g2, gain_inc; yading@10: int i, j, num_data, start_loc, end_loc; yading@10: yading@10: yading@10: if (gain2->num_gain_data == 0) yading@10: g1 = 1.0; yading@10: else yading@10: g1 = gain_tab1[gain2->lev_code[0]]; yading@10: yading@10: if (gain1->num_gain_data == 0) { yading@10: for (i = 0; i < 256; i++) yading@10: output[i] = input[i] * g1 + prev[i]; yading@10: } else { yading@10: num_data = gain1->num_gain_data; yading@10: gain1->loc_code[num_data] = 32; yading@10: gain1->lev_code[num_data] = 4; yading@10: yading@10: for (i = 0, j = 0; i < num_data; i++) { yading@10: start_loc = gain1->loc_code[i] * 8; yading@10: end_loc = start_loc + 8; yading@10: yading@10: g2 = gain_tab1[gain1->lev_code[i]]; yading@10: gain_inc = gain_tab2[gain1->lev_code[i + 1] - yading@10: gain1->lev_code[i ] + 15]; yading@10: yading@10: /* interpolate */ yading@10: for (; j < start_loc; j++) yading@10: output[j] = (input[j] * g1 + prev[j]) * g2; yading@10: yading@10: /* interpolation is done over eight samples */ yading@10: for (; j < end_loc; j++) { yading@10: output[j] = (input[j] * g1 + prev[j]) * g2; yading@10: g2 *= gain_inc; yading@10: } yading@10: } yading@10: yading@10: for (; j < 256; j++) yading@10: output[j] = input[j] * g1 + prev[j]; yading@10: } yading@10: yading@10: /* Delay for the overlapping part. */ yading@10: memcpy(prev, &input[256], 256 * sizeof(*prev)); yading@10: } yading@10: yading@10: /** yading@10: * Combine the tonal band spectrum and regular band spectrum yading@10: * yading@10: * @param spectrum output spectrum buffer yading@10: * @param num_components number of tonal components yading@10: * @param components tonal components for this band yading@10: * @return position of the last tonal coefficient yading@10: */ yading@10: static int add_tonal_components(float *spectrum, int num_components, yading@10: TonalComponent *components) yading@10: { yading@10: int i, j, last_pos = -1; yading@10: float *input, *output; yading@10: yading@10: for (i = 0; i < num_components; i++) { yading@10: last_pos = FFMAX(components[i].pos + components[i].num_coefs, last_pos); yading@10: input = components[i].coef; yading@10: output = &spectrum[components[i].pos]; yading@10: yading@10: for (j = 0; j < components[i].num_coefs; j++) yading@10: output[j] += input[j]; yading@10: } yading@10: yading@10: return last_pos; yading@10: } yading@10: yading@10: #define INTERPOLATE(old, new, nsample) \ yading@10: ((old) + (nsample) * 0.125 * ((new) - (old))) yading@10: yading@10: static void reverse_matrixing(float *su1, float *su2, int *prev_code, yading@10: int *curr_code) yading@10: { yading@10: int i, nsample, band; yading@10: float mc1_l, mc1_r, mc2_l, mc2_r; yading@10: yading@10: for (i = 0, band = 0; band < 4 * 256; band += 256, i++) { yading@10: int s1 = prev_code[i]; yading@10: int s2 = curr_code[i]; yading@10: nsample = band; yading@10: yading@10: if (s1 != s2) { yading@10: /* Selector value changed, interpolation needed. */ yading@10: mc1_l = matrix_coeffs[s1 * 2 ]; yading@10: mc1_r = matrix_coeffs[s1 * 2 + 1]; yading@10: mc2_l = matrix_coeffs[s2 * 2 ]; yading@10: mc2_r = matrix_coeffs[s2 * 2 + 1]; yading@10: yading@10: /* Interpolation is done over the first eight samples. */ yading@10: for (; nsample < band + 8; nsample++) { yading@10: float c1 = su1[nsample]; yading@10: float c2 = su2[nsample]; yading@10: c2 = c1 * INTERPOLATE(mc1_l, mc2_l, nsample - band) + yading@10: c2 * INTERPOLATE(mc1_r, mc2_r, nsample - band); yading@10: su1[nsample] = c2; yading@10: su2[nsample] = c1 * 2.0 - c2; yading@10: } yading@10: } yading@10: yading@10: /* Apply the matrix without interpolation. */ yading@10: switch (s2) { yading@10: case 0: /* M/S decoding */ yading@10: for (; nsample < band + 256; nsample++) { yading@10: float c1 = su1[nsample]; yading@10: float c2 = su2[nsample]; yading@10: su1[nsample] = c2 * 2.0; yading@10: su2[nsample] = (c1 - c2) * 2.0; yading@10: } yading@10: break; yading@10: case 1: yading@10: for (; nsample < band + 256; nsample++) { yading@10: float c1 = su1[nsample]; yading@10: float c2 = su2[nsample]; yading@10: su1[nsample] = (c1 + c2) * 2.0; yading@10: su2[nsample] = c2 * -2.0; yading@10: } yading@10: break; yading@10: case 2: yading@10: case 3: yading@10: for (; nsample < band + 256; nsample++) { yading@10: float c1 = su1[nsample]; yading@10: float c2 = su2[nsample]; yading@10: su1[nsample] = c1 + c2; yading@10: su2[nsample] = c1 - c2; yading@10: } yading@10: break; yading@10: default: yading@10: av_assert1(0); yading@10: } yading@10: } yading@10: } yading@10: yading@10: static void get_channel_weights(int index, int flag, float ch[2]) yading@10: { yading@10: if (index == 7) { yading@10: ch[0] = 1.0; yading@10: ch[1] = 1.0; yading@10: } else { yading@10: ch[0] = (index & 7) / 7.0; yading@10: ch[1] = sqrt(2 - ch[0] * ch[0]); yading@10: if (flag) yading@10: FFSWAP(float, ch[0], ch[1]); yading@10: } yading@10: } yading@10: yading@10: static void channel_weighting(float *su1, float *su2, int *p3) yading@10: { yading@10: int band, nsample; yading@10: /* w[x][y] y=0 is left y=1 is right */ yading@10: float w[2][2]; yading@10: yading@10: if (p3[1] != 7 || p3[3] != 7) { yading@10: get_channel_weights(p3[1], p3[0], w[0]); yading@10: get_channel_weights(p3[3], p3[2], w[1]); yading@10: yading@10: for (band = 256; band < 4 * 256; band += 256) { yading@10: for (nsample = band; nsample < band + 8; nsample++) { yading@10: su1[nsample] *= INTERPOLATE(w[0][0], w[0][1], nsample - band); yading@10: su2[nsample] *= INTERPOLATE(w[1][0], w[1][1], nsample - band); yading@10: } yading@10: for(; nsample < band + 256; nsample++) { yading@10: su1[nsample] *= w[1][0]; yading@10: su2[nsample] *= w[1][1]; yading@10: } yading@10: } yading@10: } yading@10: } yading@10: yading@10: /** yading@10: * Decode a Sound Unit yading@10: * yading@10: * @param snd the channel unit to be used yading@10: * @param output the decoded samples before IQMF in float representation yading@10: * @param channel_num channel number yading@10: * @param coding_mode the coding mode (JOINT_STEREO or regular stereo/mono) yading@10: */ yading@10: static int decode_channel_sound_unit(ATRAC3Context *q, GetBitContext *gb, yading@10: ChannelUnit *snd, float *output, yading@10: int channel_num, int coding_mode) yading@10: { yading@10: int band, ret, num_subbands, last_tonal, num_bands; yading@10: GainBlock *gain1 = &snd->gain_block[ snd->gc_blk_switch]; yading@10: GainBlock *gain2 = &snd->gain_block[1 - snd->gc_blk_switch]; yading@10: yading@10: if (coding_mode == JOINT_STEREO && channel_num == 1) { yading@10: if (get_bits(gb, 2) != 3) { yading@10: av_log(NULL,AV_LOG_ERROR,"JS mono Sound Unit id != 3.\n"); yading@10: return AVERROR_INVALIDDATA; yading@10: } yading@10: } else { yading@10: if (get_bits(gb, 6) != 0x28) { yading@10: av_log(NULL,AV_LOG_ERROR,"Sound Unit id != 0x28.\n"); yading@10: return AVERROR_INVALIDDATA; yading@10: } yading@10: } yading@10: yading@10: /* number of coded QMF bands */ yading@10: snd->bands_coded = get_bits(gb, 2); yading@10: yading@10: ret = decode_gain_control(gb, gain2, snd->bands_coded); yading@10: if (ret) yading@10: return ret; yading@10: yading@10: snd->num_components = decode_tonal_components(gb, snd->components, yading@10: snd->bands_coded); yading@10: if (snd->num_components == -1) yading@10: return -1; yading@10: yading@10: num_subbands = decode_spectrum(gb, snd->spectrum); yading@10: yading@10: /* Merge the decoded spectrum and tonal components. */ yading@10: last_tonal = add_tonal_components(snd->spectrum, snd->num_components, yading@10: snd->components); yading@10: yading@10: yading@10: /* calculate number of used MLT/QMF bands according to the amount of coded yading@10: spectral lines */ yading@10: num_bands = (subband_tab[num_subbands] - 1) >> 8; yading@10: if (last_tonal >= 0) yading@10: num_bands = FFMAX((last_tonal + 256) >> 8, num_bands); yading@10: yading@10: yading@10: /* Reconstruct time domain samples. */ yading@10: for (band = 0; band < 4; band++) { yading@10: /* Perform the IMDCT step without overlapping. */ yading@10: if (band <= num_bands) yading@10: imlt(q, &snd->spectrum[band * 256], snd->imdct_buf, band & 1); yading@10: else yading@10: memset(snd->imdct_buf, 0, 512 * sizeof(*snd->imdct_buf)); yading@10: yading@10: /* gain compensation and overlapping */ yading@10: gain_compensate_and_overlap(snd->imdct_buf, yading@10: &snd->prev_frame[band * 256], yading@10: &output[band * 256], yading@10: &gain1->g_block[band], yading@10: &gain2->g_block[band]); yading@10: } yading@10: yading@10: /* Swap the gain control buffers for the next frame. */ yading@10: snd->gc_blk_switch ^= 1; yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: static int decode_frame(AVCodecContext *avctx, const uint8_t *databuf, yading@10: float **out_samples) yading@10: { yading@10: ATRAC3Context *q = avctx->priv_data; yading@10: int ret, i; yading@10: uint8_t *ptr1; yading@10: yading@10: if (q->coding_mode == JOINT_STEREO) { yading@10: /* channel coupling mode */ yading@10: /* decode Sound Unit 1 */ yading@10: init_get_bits(&q->gb, databuf, avctx->block_align * 8); yading@10: yading@10: ret = decode_channel_sound_unit(q, &q->gb, q->units, out_samples[0], 0, yading@10: JOINT_STEREO); yading@10: if (ret != 0) yading@10: return ret; yading@10: yading@10: /* Framedata of the su2 in the joint-stereo mode is encoded in yading@10: * reverse byte order so we need to swap it first. */ yading@10: if (databuf == q->decoded_bytes_buffer) { yading@10: uint8_t *ptr2 = q->decoded_bytes_buffer + avctx->block_align - 1; yading@10: ptr1 = q->decoded_bytes_buffer; yading@10: for (i = 0; i < avctx->block_align / 2; i++, ptr1++, ptr2--) yading@10: FFSWAP(uint8_t, *ptr1, *ptr2); yading@10: } else { yading@10: const uint8_t *ptr2 = databuf + avctx->block_align - 1; yading@10: for (i = 0; i < avctx->block_align; i++) yading@10: q->decoded_bytes_buffer[i] = *ptr2--; yading@10: } yading@10: yading@10: /* Skip the sync codes (0xF8). */ yading@10: ptr1 = q->decoded_bytes_buffer; yading@10: for (i = 4; *ptr1 == 0xF8; i++, ptr1++) { yading@10: if (i >= avctx->block_align) yading@10: return AVERROR_INVALIDDATA; yading@10: } yading@10: yading@10: yading@10: /* set the bitstream reader at the start of the second Sound Unit*/ yading@10: init_get_bits8(&q->gb, ptr1, q->decoded_bytes_buffer + avctx->block_align - ptr1); yading@10: yading@10: /* Fill the Weighting coeffs delay buffer */ yading@10: memmove(q->weighting_delay, &q->weighting_delay[2], yading@10: 4 * sizeof(*q->weighting_delay)); yading@10: q->weighting_delay[4] = get_bits1(&q->gb); yading@10: q->weighting_delay[5] = get_bits(&q->gb, 3); yading@10: yading@10: for (i = 0; i < 4; i++) { yading@10: q->matrix_coeff_index_prev[i] = q->matrix_coeff_index_now[i]; yading@10: q->matrix_coeff_index_now[i] = q->matrix_coeff_index_next[i]; yading@10: q->matrix_coeff_index_next[i] = get_bits(&q->gb, 2); yading@10: } yading@10: yading@10: /* Decode Sound Unit 2. */ yading@10: ret = decode_channel_sound_unit(q, &q->gb, &q->units[1], yading@10: out_samples[1], 1, JOINT_STEREO); yading@10: if (ret != 0) yading@10: return ret; yading@10: yading@10: /* Reconstruct the channel coefficients. */ yading@10: reverse_matrixing(out_samples[0], out_samples[1], yading@10: q->matrix_coeff_index_prev, yading@10: q->matrix_coeff_index_now); yading@10: yading@10: channel_weighting(out_samples[0], out_samples[1], q->weighting_delay); yading@10: } else { yading@10: /* normal stereo mode or mono */ yading@10: /* Decode the channel sound units. */ yading@10: for (i = 0; i < avctx->channels; i++) { yading@10: /* Set the bitstream reader at the start of a channel sound unit. */ yading@10: init_get_bits(&q->gb, yading@10: databuf + i * avctx->block_align / avctx->channels, yading@10: avctx->block_align * 8 / avctx->channels); yading@10: yading@10: ret = decode_channel_sound_unit(q, &q->gb, &q->units[i], yading@10: out_samples[i], i, q->coding_mode); yading@10: if (ret != 0) yading@10: return ret; yading@10: } yading@10: } yading@10: yading@10: /* Apply the iQMF synthesis filter. */ yading@10: for (i = 0; i < avctx->channels; i++) { yading@10: float *p1 = out_samples[i]; yading@10: float *p2 = p1 + 256; yading@10: float *p3 = p2 + 256; yading@10: float *p4 = p3 + 256; yading@10: ff_atrac_iqmf(p1, p2, 256, p1, q->units[i].delay_buf1, q->temp_buf); yading@10: ff_atrac_iqmf(p4, p3, 256, p3, q->units[i].delay_buf2, q->temp_buf); yading@10: ff_atrac_iqmf(p1, p3, 512, p1, q->units[i].delay_buf3, q->temp_buf); yading@10: } yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: static int atrac3_decode_frame(AVCodecContext *avctx, void *data, yading@10: int *got_frame_ptr, AVPacket *avpkt) yading@10: { yading@10: AVFrame *frame = data; yading@10: const uint8_t *buf = avpkt->data; yading@10: int buf_size = avpkt->size; yading@10: ATRAC3Context *q = avctx->priv_data; yading@10: int ret; yading@10: const uint8_t *databuf; yading@10: yading@10: if (buf_size < avctx->block_align) { yading@10: av_log(avctx, AV_LOG_ERROR, yading@10: "Frame too small (%d bytes). Truncated file?\n", buf_size); yading@10: return AVERROR_INVALIDDATA; yading@10: } yading@10: yading@10: /* get output buffer */ yading@10: frame->nb_samples = SAMPLES_PER_FRAME; yading@10: if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) yading@10: return ret; yading@10: yading@10: /* Check if we need to descramble and what buffer to pass on. */ yading@10: if (q->scrambled_stream) { yading@10: decode_bytes(buf, q->decoded_bytes_buffer, avctx->block_align); yading@10: databuf = q->decoded_bytes_buffer; yading@10: } else { yading@10: databuf = buf; yading@10: } yading@10: yading@10: ret = decode_frame(avctx, databuf, (float **)frame->extended_data); yading@10: if (ret) { yading@10: av_log(NULL, AV_LOG_ERROR, "Frame decoding error!\n"); yading@10: return ret; yading@10: } yading@10: yading@10: *got_frame_ptr = 1; yading@10: yading@10: return avctx->block_align; yading@10: } yading@10: yading@10: static void atrac3_init_static_data(void) yading@10: { yading@10: int i; yading@10: yading@10: init_atrac3_window(); yading@10: ff_atrac_generate_tables(); yading@10: yading@10: /* Initialize the VLC tables. */ yading@10: for (i = 0; i < 7; i++) { yading@10: spectral_coeff_tab[i].table = &atrac3_vlc_table[atrac3_vlc_offs[i]]; yading@10: spectral_coeff_tab[i].table_allocated = atrac3_vlc_offs[i + 1] - yading@10: atrac3_vlc_offs[i ]; yading@10: init_vlc(&spectral_coeff_tab[i], 9, huff_tab_sizes[i], yading@10: huff_bits[i], 1, 1, yading@10: huff_codes[i], 1, 1, INIT_VLC_USE_NEW_STATIC); yading@10: } yading@10: yading@10: /* Generate gain tables */ yading@10: for (i = 0; i < 16; i++) yading@10: gain_tab1[i] = exp2f (4 - i); yading@10: yading@10: for (i = -15; i < 16; i++) yading@10: gain_tab2[i + 15] = exp2f (i * -0.125); yading@10: } yading@10: yading@10: static av_cold int atrac3_decode_init(AVCodecContext *avctx) yading@10: { yading@10: static int static_init_done; yading@10: int i, ret; yading@10: int version, delay, samples_per_frame, frame_factor; yading@10: const uint8_t *edata_ptr = avctx->extradata; yading@10: ATRAC3Context *q = avctx->priv_data; yading@10: yading@10: if (avctx->channels <= 0 || avctx->channels > 2) { yading@10: av_log(avctx, AV_LOG_ERROR, "Channel configuration error!\n"); yading@10: return AVERROR(EINVAL); yading@10: } yading@10: yading@10: if (!static_init_done) yading@10: atrac3_init_static_data(); yading@10: static_init_done = 1; yading@10: yading@10: /* Take care of the codec-specific extradata. */ yading@10: if (avctx->extradata_size == 14) { yading@10: /* Parse the extradata, WAV format */ yading@10: av_log(avctx, AV_LOG_DEBUG, "[0-1] %d\n", yading@10: bytestream_get_le16(&edata_ptr)); // Unknown value always 1 yading@10: edata_ptr += 4; // samples per channel yading@10: q->coding_mode = bytestream_get_le16(&edata_ptr); yading@10: av_log(avctx, AV_LOG_DEBUG,"[8-9] %d\n", yading@10: bytestream_get_le16(&edata_ptr)); //Dupe of coding mode yading@10: frame_factor = bytestream_get_le16(&edata_ptr); // Unknown always 1 yading@10: av_log(avctx, AV_LOG_DEBUG,"[12-13] %d\n", yading@10: bytestream_get_le16(&edata_ptr)); // Unknown always 0 yading@10: yading@10: /* setup */ yading@10: samples_per_frame = SAMPLES_PER_FRAME * avctx->channels; yading@10: version = 4; yading@10: delay = 0x88E; yading@10: q->coding_mode = q->coding_mode ? JOINT_STEREO : STEREO; yading@10: q->scrambled_stream = 0; yading@10: yading@10: if (avctx->block_align != 96 * avctx->channels * frame_factor && yading@10: avctx->block_align != 152 * avctx->channels * frame_factor && yading@10: avctx->block_align != 192 * avctx->channels * frame_factor) { yading@10: av_log(avctx, AV_LOG_ERROR, "Unknown frame/channel/frame_factor " yading@10: "configuration %d/%d/%d\n", avctx->block_align, yading@10: avctx->channels, frame_factor); yading@10: return AVERROR_INVALIDDATA; yading@10: } yading@10: } else if (avctx->extradata_size == 12 || avctx->extradata_size == 10) { yading@10: /* Parse the extradata, RM format. */ yading@10: version = bytestream_get_be32(&edata_ptr); yading@10: samples_per_frame = bytestream_get_be16(&edata_ptr); yading@10: delay = bytestream_get_be16(&edata_ptr); yading@10: q->coding_mode = bytestream_get_be16(&edata_ptr); yading@10: q->scrambled_stream = 1; yading@10: yading@10: } else { yading@10: av_log(NULL, AV_LOG_ERROR, "Unknown extradata size %d.\n", yading@10: avctx->extradata_size); yading@10: return AVERROR(EINVAL); yading@10: } yading@10: yading@10: if (q->coding_mode == JOINT_STEREO && avctx->channels < 2) { yading@10: av_log(avctx, AV_LOG_ERROR, "Invalid coding mode\n"); yading@10: return AVERROR_INVALIDDATA; yading@10: } yading@10: yading@10: /* Check the extradata */ yading@10: yading@10: if (version != 4) { yading@10: av_log(avctx, AV_LOG_ERROR, "Version %d != 4.\n", version); yading@10: return AVERROR_INVALIDDATA; yading@10: } yading@10: yading@10: if (samples_per_frame != SAMPLES_PER_FRAME && yading@10: samples_per_frame != SAMPLES_PER_FRAME * 2) { yading@10: av_log(avctx, AV_LOG_ERROR, "Unknown amount of samples per frame %d.\n", yading@10: samples_per_frame); yading@10: return AVERROR_INVALIDDATA; yading@10: } yading@10: yading@10: if (delay != 0x88E) { yading@10: av_log(avctx, AV_LOG_ERROR, "Unknown amount of delay %x != 0x88E.\n", yading@10: delay); yading@10: return AVERROR_INVALIDDATA; yading@10: } yading@10: yading@10: if (q->coding_mode == STEREO) yading@10: av_log(avctx, AV_LOG_DEBUG, "Normal stereo detected.\n"); yading@10: else if (q->coding_mode == JOINT_STEREO) yading@10: av_log(avctx, AV_LOG_DEBUG, "Joint stereo detected.\n"); yading@10: else { yading@10: av_log(avctx, AV_LOG_ERROR, "Unknown channel coding mode %x!\n", yading@10: q->coding_mode); yading@10: return AVERROR_INVALIDDATA; yading@10: } yading@10: yading@10: if (avctx->block_align >= UINT_MAX / 2) yading@10: return AVERROR(EINVAL); yading@10: yading@10: q->decoded_bytes_buffer = av_mallocz(FFALIGN(avctx->block_align, 4) + yading@10: FF_INPUT_BUFFER_PADDING_SIZE); yading@10: if (q->decoded_bytes_buffer == NULL) yading@10: return AVERROR(ENOMEM); yading@10: yading@10: avctx->sample_fmt = AV_SAMPLE_FMT_FLTP; yading@10: yading@10: /* initialize the MDCT transform */ yading@10: if ((ret = ff_mdct_init(&q->mdct_ctx, 9, 1, 1.0 / 32768)) < 0) { yading@10: av_log(avctx, AV_LOG_ERROR, "Error initializing MDCT\n"); yading@10: av_freep(&q->decoded_bytes_buffer); yading@10: return ret; yading@10: } yading@10: yading@10: /* init the joint-stereo decoding data */ yading@10: q->weighting_delay[0] = 0; yading@10: q->weighting_delay[1] = 7; yading@10: q->weighting_delay[2] = 0; yading@10: q->weighting_delay[3] = 7; yading@10: q->weighting_delay[4] = 0; yading@10: q->weighting_delay[5] = 7; yading@10: yading@10: for (i = 0; i < 4; i++) { yading@10: q->matrix_coeff_index_prev[i] = 3; yading@10: q->matrix_coeff_index_now[i] = 3; yading@10: q->matrix_coeff_index_next[i] = 3; yading@10: } yading@10: yading@10: avpriv_float_dsp_init(&q->fdsp, avctx->flags & CODEC_FLAG_BITEXACT); yading@10: ff_fmt_convert_init(&q->fmt_conv, avctx); yading@10: yading@10: q->units = av_mallocz(sizeof(*q->units) * avctx->channels); yading@10: if (!q->units) { yading@10: atrac3_decode_close(avctx); yading@10: return AVERROR(ENOMEM); yading@10: } yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: AVCodec ff_atrac3_decoder = { yading@10: .name = "atrac3", yading@10: .type = AVMEDIA_TYPE_AUDIO, yading@10: .id = AV_CODEC_ID_ATRAC3, yading@10: .priv_data_size = sizeof(ATRAC3Context), yading@10: .init = atrac3_decode_init, yading@10: .close = atrac3_decode_close, yading@10: .decode = atrac3_decode_frame, yading@10: .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_DR1, yading@10: .long_name = NULL_IF_CONFIG_SMALL("Atrac 3 (Adaptive TRansform Acoustic Coding 3)"), yading@10: .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP, yading@10: AV_SAMPLE_FMT_NONE }, yading@10: };