yading@10: /*
yading@10:  * Nellymoser encoder
yading@10:  * This code is developed as part of Google Summer of Code 2008 Program.
yading@10:  *
yading@10:  * Copyright (c) 2008 Bartlomiej Wolowiec
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:  * Nellymoser encoder
yading@10:  * by Bartlomiej Wolowiec
yading@10:  *
yading@10:  * Generic codec information: libavcodec/nellymoserdec.c
yading@10:  *
yading@10:  * Some information also from: http://samples.mplayerhq.hu/A-codecs/Nelly_Moser/ASAO/ASAO.zip
yading@10:  *                             (Copyright Joseph Artsimovich and UAB "DKD")
yading@10:  *
yading@10:  * for more information about nellymoser format, visit:
yading@10:  * http://wiki.multimedia.cx/index.php?title=Nellymoser
yading@10:  */
yading@10: 
yading@10: #include "libavutil/float_dsp.h"
yading@10: #include "libavutil/mathematics.h"
yading@10: #include "nellymoser.h"
yading@10: #include "avcodec.h"
yading@10: #include "audio_frame_queue.h"
yading@10: #include "fft.h"
yading@10: #include "internal.h"
yading@10: #include "sinewin.h"
yading@10: 
yading@10: #define BITSTREAM_WRITER_LE
yading@10: #include "put_bits.h"
yading@10: 
yading@10: #define POW_TABLE_SIZE (1<<11)
yading@10: #define POW_TABLE_OFFSET 3
yading@10: #define OPT_SIZE ((1<<15) + 3000)
yading@10: 
yading@10: typedef struct NellyMoserEncodeContext {
yading@10:     AVCodecContext  *avctx;
yading@10:     int             last_frame;
yading@10:     AVFloatDSPContext fdsp;
yading@10:     FFTContext      mdct_ctx;
yading@10:     AudioFrameQueue afq;
yading@10:     DECLARE_ALIGNED(32, float, mdct_out)[NELLY_SAMPLES];
yading@10:     DECLARE_ALIGNED(32, float, in_buff)[NELLY_SAMPLES];
yading@10:     DECLARE_ALIGNED(32, float, buf)[3 * NELLY_BUF_LEN];     ///< sample buffer
yading@10:     float           (*opt )[OPT_SIZE];
yading@10:     uint8_t         (*path)[OPT_SIZE];
yading@10: } NellyMoserEncodeContext;
yading@10: 
yading@10: static float pow_table[POW_TABLE_SIZE];     ///< -pow(2, -i / 2048.0 - 3.0);
yading@10: 
yading@10: static const uint8_t sf_lut[96] = {
yading@10:      0,  1,  1,  1,  1,  1,  1,  2,  2,  2,  2,  3,  3,  3,  4,  4,
yading@10:      5,  5,  5,  6,  7,  7,  8,  8,  9, 10, 11, 11, 12, 13, 13, 14,
yading@10:     15, 15, 16, 17, 17, 18, 19, 19, 20, 21, 22, 22, 23, 24, 25, 26,
yading@10:     27, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 37, 38, 39, 40,
yading@10:     41, 41, 42, 43, 44, 45, 45, 46, 47, 48, 49, 50, 51, 52, 52, 53,
yading@10:     54, 55, 55, 56, 57, 57, 58, 59, 59, 60, 60, 60, 61, 61, 61, 62,
yading@10: };
yading@10: 
yading@10: static const uint8_t sf_delta_lut[78] = {
yading@10:      0,  1,  1,  1,  1,  1,  1,  2,  2,  2,  2,  3,  3,  3,  4,  4,
yading@10:      4,  5,  5,  5,  6,  6,  7,  7,  8,  8,  9, 10, 10, 11, 11, 12,
yading@10:     13, 13, 14, 15, 16, 17, 17, 18, 19, 19, 20, 21, 21, 22, 22, 23,
yading@10:     23, 24, 24, 25, 25, 25, 26, 26, 26, 26, 27, 27, 27, 27, 27, 28,
yading@10:     28, 28, 28, 28, 28, 29, 29, 29, 29, 29, 29, 29, 29, 30,
yading@10: };
yading@10: 
yading@10: static const uint8_t quant_lut[230] = {
yading@10:      0,
yading@10: 
yading@10:      0,  1,  2,
yading@10: 
yading@10:      0,  1,  2,  3,  4,  5,  6,
yading@10: 
yading@10:      0,  1,  1,  2,  2,  3,  3,  4,  5,  6,  7,  8,  9, 10, 11, 11,
yading@10:     12, 13, 13, 13, 14,
yading@10: 
yading@10:      0,  1,  1,  2,  2,  2,  3,  3,  4,  4,  5,  5,  6,  6,  7,  8,
yading@10:      8,  9, 10, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22,
yading@10:     22, 23, 23, 24, 24, 25, 25, 26, 26, 27, 27, 28, 28, 29, 29, 29,
yading@10:     30,
yading@10: 
yading@10:      0,  1,  1,  1,  1,  1,  1,  2,  2,  2,  2,  2,  3,  3,  3,  3,
yading@10:      4,  4,  4,  5,  5,  5,  6,  6,  7,  7,  7,  8,  8,  9,  9,  9,
yading@10:     10, 10, 11, 11, 11, 12, 12, 13, 13, 13, 13, 14, 14, 14, 15, 15,
yading@10:     15, 15, 16, 16, 16, 17, 17, 17, 18, 18, 18, 19, 19, 20, 20, 20,
yading@10:     21, 21, 22, 22, 23, 23, 24, 25, 26, 26, 27, 28, 29, 30, 31, 32,
yading@10:     33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 42, 43, 44, 44, 45, 45,
yading@10:     46, 47, 47, 48, 48, 49, 49, 50, 50, 50, 51, 51, 51, 52, 52, 52,
yading@10:     53, 53, 53, 54, 54, 54, 55, 55, 55, 56, 56, 56, 57, 57, 57, 57,
yading@10:     58, 58, 58, 58, 59, 59, 59, 59, 60, 60, 60, 60, 60, 61, 61, 61,
yading@10:     61, 61, 61, 61, 62,
yading@10: };
yading@10: 
yading@10: static const float quant_lut_mul[7] = { 0.0,  0.0,  2.0,  2.0,  5.0, 12.0,  36.6 };
yading@10: static const float quant_lut_add[7] = { 0.0,  0.0,  2.0,  7.0, 21.0, 56.0, 157.0 };
yading@10: static const uint8_t quant_lut_offset[8] = { 0, 0, 1, 4, 11, 32, 81, 230 };
yading@10: 
yading@10: static void apply_mdct(NellyMoserEncodeContext *s)
yading@10: {
yading@10:     float *in0 = s->buf;
yading@10:     float *in1 = s->buf + NELLY_BUF_LEN;
yading@10:     float *in2 = s->buf + 2 * NELLY_BUF_LEN;
yading@10: 
yading@10:     s->fdsp.vector_fmul        (s->in_buff,                 in0, ff_sine_128, NELLY_BUF_LEN);
yading@10:     s->fdsp.vector_fmul_reverse(s->in_buff + NELLY_BUF_LEN, in1, ff_sine_128, NELLY_BUF_LEN);
yading@10:     s->mdct_ctx.mdct_calc(&s->mdct_ctx, s->mdct_out, s->in_buff);
yading@10: 
yading@10:     s->fdsp.vector_fmul        (s->in_buff,                 in1, ff_sine_128, NELLY_BUF_LEN);
yading@10:     s->fdsp.vector_fmul_reverse(s->in_buff + NELLY_BUF_LEN, in2, ff_sine_128, NELLY_BUF_LEN);
yading@10:     s->mdct_ctx.mdct_calc(&s->mdct_ctx, s->mdct_out + NELLY_BUF_LEN, s->in_buff);
yading@10: }
yading@10: 
yading@10: static av_cold int encode_end(AVCodecContext *avctx)
yading@10: {
yading@10:     NellyMoserEncodeContext *s = avctx->priv_data;
yading@10: 
yading@10:     ff_mdct_end(&s->mdct_ctx);
yading@10: 
yading@10:     if (s->avctx->trellis) {
yading@10:         av_free(s->opt);
yading@10:         av_free(s->path);
yading@10:     }
yading@10:     ff_af_queue_close(&s->afq);
yading@10: 
yading@10:     return 0;
yading@10: }
yading@10: 
yading@10: static av_cold int encode_init(AVCodecContext *avctx)
yading@10: {
yading@10:     NellyMoserEncodeContext *s = avctx->priv_data;
yading@10:     int i, ret;
yading@10: 
yading@10:     if (avctx->channels != 1) {
yading@10:         av_log(avctx, AV_LOG_ERROR, "Nellymoser supports only 1 channel\n");
yading@10:         return AVERROR(EINVAL);
yading@10:     }
yading@10: 
yading@10:     if (avctx->sample_rate != 8000 && avctx->sample_rate != 16000 &&
yading@10:         avctx->sample_rate != 11025 &&
yading@10:         avctx->sample_rate != 22050 && avctx->sample_rate != 44100 &&
yading@10:         avctx->strict_std_compliance >= FF_COMPLIANCE_NORMAL) {
yading@10:         av_log(avctx, AV_LOG_ERROR, "Nellymoser works only with 8000, 16000, 11025, 22050 and 44100 sample rate\n");
yading@10:         return AVERROR(EINVAL);
yading@10:     }
yading@10: 
yading@10:     avctx->frame_size = NELLY_SAMPLES;
yading@10:     avctx->delay      = NELLY_BUF_LEN;
yading@10:     ff_af_queue_init(avctx, &s->afq);
yading@10:     s->avctx = avctx;
yading@10:     if ((ret = ff_mdct_init(&s->mdct_ctx, 8, 0, 32768.0)) < 0)
yading@10:         goto error;
yading@10:     avpriv_float_dsp_init(&s->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
yading@10: 
yading@10:     /* Generate overlap window */
yading@10:     ff_init_ff_sine_windows(7);
yading@10:     for (i = 0; i < POW_TABLE_SIZE; i++)
yading@10:         pow_table[i] = -pow(2, -i / 2048.0 - 3.0 + POW_TABLE_OFFSET);
yading@10: 
yading@10:     if (s->avctx->trellis) {
yading@10:         s->opt  = av_malloc(NELLY_BANDS * OPT_SIZE * sizeof(float  ));
yading@10:         s->path = av_malloc(NELLY_BANDS * OPT_SIZE * sizeof(uint8_t));
yading@10:         if (!s->opt || !s->path) {
yading@10:             ret = AVERROR(ENOMEM);
yading@10:             goto error;
yading@10:         }
yading@10:     }
yading@10: 
yading@10:     return 0;
yading@10: error:
yading@10:     encode_end(avctx);
yading@10:     return ret;
yading@10: }
yading@10: 
yading@10: #define find_best(val, table, LUT, LUT_add, LUT_size) \
yading@10:     best_idx = \
yading@10:         LUT[av_clip ((lrintf(val) >> 8) + LUT_add, 0, LUT_size - 1)]; \
yading@10:     if (fabs(val - table[best_idx]) > fabs(val - table[best_idx + 1])) \
yading@10:         best_idx++;
yading@10: 
yading@10: static void get_exponent_greedy(NellyMoserEncodeContext *s, float *cand, int *idx_table)
yading@10: {
yading@10:     int band, best_idx, power_idx = 0;
yading@10:     float power_candidate;
yading@10: 
yading@10:     //base exponent
yading@10:     find_best(cand[0], ff_nelly_init_table, sf_lut, -20, 96);
yading@10:     idx_table[0] = best_idx;
yading@10:     power_idx = ff_nelly_init_table[best_idx];
yading@10: 
yading@10:     for (band = 1; band < NELLY_BANDS; band++) {
yading@10:         power_candidate = cand[band] - power_idx;
yading@10:         find_best(power_candidate, ff_nelly_delta_table, sf_delta_lut, 37, 78);
yading@10:         idx_table[band] = best_idx;
yading@10:         power_idx += ff_nelly_delta_table[best_idx];
yading@10:     }
yading@10: }
yading@10: 
yading@10: static inline float distance(float x, float y, int band)
yading@10: {
yading@10:     //return pow(fabs(x-y), 2.0);
yading@10:     float tmp = x - y;
yading@10:     return tmp * tmp;
yading@10: }
yading@10: 
yading@10: static void get_exponent_dynamic(NellyMoserEncodeContext *s, float *cand, int *idx_table)
yading@10: {
yading@10:     int i, j, band, best_idx;
yading@10:     float power_candidate, best_val;
yading@10: 
yading@10:     float  (*opt )[OPT_SIZE] = s->opt ;
yading@10:     uint8_t(*path)[OPT_SIZE] = s->path;
yading@10: 
yading@10:     for (i = 0; i < NELLY_BANDS * OPT_SIZE; i++) {
yading@10:         opt[0][i] = INFINITY;
yading@10:     }
yading@10: 
yading@10:     for (i = 0; i < 64; i++) {
yading@10:         opt[0][ff_nelly_init_table[i]] = distance(cand[0], ff_nelly_init_table[i], 0);
yading@10:         path[0][ff_nelly_init_table[i]] = i;
yading@10:     }
yading@10: 
yading@10:     for (band = 1; band < NELLY_BANDS; band++) {
yading@10:         int q, c = 0;
yading@10:         float tmp;
yading@10:         int idx_min, idx_max, idx;
yading@10:         power_candidate = cand[band];
yading@10:         for (q = 1000; !c && q < OPT_SIZE; q <<= 2) {
yading@10:             idx_min = FFMAX(0, cand[band] - q);
yading@10:             idx_max = FFMIN(OPT_SIZE, cand[band - 1] + q);
yading@10:             for (i = FFMAX(0, cand[band - 1] - q); i < FFMIN(OPT_SIZE, cand[band - 1] + q); i++) {
yading@10:                 if ( isinf(opt[band - 1][i]) )
yading@10:                     continue;
yading@10:                 for (j = 0; j < 32; j++) {
yading@10:                     idx = i + ff_nelly_delta_table[j];
yading@10:                     if (idx > idx_max)
yading@10:                         break;
yading@10:                     if (idx >= idx_min) {
yading@10:                         tmp = opt[band - 1][i] + distance(idx, power_candidate, band);
yading@10:                         if (opt[band][idx] > tmp) {
yading@10:                             opt[band][idx] = tmp;
yading@10:                             path[band][idx] = j;
yading@10:                             c = 1;
yading@10:                         }
yading@10:                     }
yading@10:                 }
yading@10:             }
yading@10:         }
yading@10:         assert(c); //FIXME
yading@10:     }
yading@10: 
yading@10:     best_val = INFINITY;
yading@10:     best_idx = -1;
yading@10:     band = NELLY_BANDS - 1;
yading@10:     for (i = 0; i < OPT_SIZE; i++) {
yading@10:         if (best_val > opt[band][i]) {
yading@10:             best_val = opt[band][i];
yading@10:             best_idx = i;
yading@10:         }
yading@10:     }
yading@10:     for (band = NELLY_BANDS - 1; band >= 0; band--) {
yading@10:         idx_table[band] = path[band][best_idx];
yading@10:         if (band) {
yading@10:             best_idx -= ff_nelly_delta_table[path[band][best_idx]];
yading@10:         }
yading@10:     }
yading@10: }
yading@10: 
yading@10: /**
yading@10:  * Encode NELLY_SAMPLES samples. It assumes, that samples contains 3 * NELLY_BUF_LEN values
yading@10:  *  @param s               encoder context
yading@10:  *  @param output          output buffer
yading@10:  *  @param output_size     size of output buffer
yading@10:  */
yading@10: static void encode_block(NellyMoserEncodeContext *s, unsigned char *output, int output_size)
yading@10: {
yading@10:     PutBitContext pb;
yading@10:     int i, j, band, block, best_idx, power_idx = 0;
yading@10:     float power_val, coeff, coeff_sum;
yading@10:     float pows[NELLY_FILL_LEN];
yading@10:     int bits[NELLY_BUF_LEN], idx_table[NELLY_BANDS];
yading@10:     float cand[NELLY_BANDS];
yading@10: 
yading@10:     apply_mdct(s);
yading@10: 
yading@10:     init_put_bits(&pb, output, output_size * 8);
yading@10: 
yading@10:     i = 0;
yading@10:     for (band = 0; band < NELLY_BANDS; band++) {
yading@10:         coeff_sum = 0;
yading@10:         for (j = 0; j < ff_nelly_band_sizes_table[band]; i++, j++) {
yading@10:             coeff_sum += s->mdct_out[i                ] * s->mdct_out[i                ]
yading@10:                        + s->mdct_out[i + NELLY_BUF_LEN] * s->mdct_out[i + NELLY_BUF_LEN];
yading@10:         }
yading@10:         cand[band] =
yading@10:             log(FFMAX(1.0, coeff_sum / (ff_nelly_band_sizes_table[band] << 7))) * 1024.0 / M_LN2;
yading@10:     }
yading@10: 
yading@10:     if (s->avctx->trellis) {
yading@10:         get_exponent_dynamic(s, cand, idx_table);
yading@10:     } else {
yading@10:         get_exponent_greedy(s, cand, idx_table);
yading@10:     }
yading@10: 
yading@10:     i = 0;
yading@10:     for (band = 0; band < NELLY_BANDS; band++) {
yading@10:         if (band) {
yading@10:             power_idx += ff_nelly_delta_table[idx_table[band]];
yading@10:             put_bits(&pb, 5, idx_table[band]);
yading@10:         } else {
yading@10:             power_idx = ff_nelly_init_table[idx_table[0]];
yading@10:             put_bits(&pb, 6, idx_table[0]);
yading@10:         }
yading@10:         power_val = pow_table[power_idx & 0x7FF] / (1 << ((power_idx >> 11) + POW_TABLE_OFFSET));
yading@10:         for (j = 0; j < ff_nelly_band_sizes_table[band]; i++, j++) {
yading@10:             s->mdct_out[i] *= power_val;
yading@10:             s->mdct_out[i + NELLY_BUF_LEN] *= power_val;
yading@10:             pows[i] = power_idx;
yading@10:         }
yading@10:     }
yading@10: 
yading@10:     ff_nelly_get_sample_bits(pows, bits);
yading@10: 
yading@10:     for (block = 0; block < 2; block++) {
yading@10:         for (i = 0; i < NELLY_FILL_LEN; i++) {
yading@10:             if (bits[i] > 0) {
yading@10:                 const float *table = ff_nelly_dequantization_table + (1 << bits[i]) - 1;
yading@10:                 coeff = s->mdct_out[block * NELLY_BUF_LEN + i];
yading@10:                 best_idx =
yading@10:                     quant_lut[av_clip (
yading@10:                             coeff * quant_lut_mul[bits[i]] + quant_lut_add[bits[i]],
yading@10:                             quant_lut_offset[bits[i]],
yading@10:                             quant_lut_offset[bits[i]+1] - 1
yading@10:                             )];
yading@10:                 if (fabs(coeff - table[best_idx]) > fabs(coeff - table[best_idx + 1]))
yading@10:                     best_idx++;
yading@10: 
yading@10:                 put_bits(&pb, bits[i], best_idx);
yading@10:             }
yading@10:         }
yading@10:         if (!block)
yading@10:             put_bits(&pb, NELLY_HEADER_BITS + NELLY_DETAIL_BITS - put_bits_count(&pb), 0);
yading@10:     }
yading@10: 
yading@10:     flush_put_bits(&pb);
yading@10:     memset(put_bits_ptr(&pb), 0, output + output_size - put_bits_ptr(&pb));
yading@10: }
yading@10: 
yading@10: static int encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
yading@10:                         const AVFrame *frame, int *got_packet_ptr)
yading@10: {
yading@10:     NellyMoserEncodeContext *s = avctx->priv_data;
yading@10:     int ret;
yading@10: 
yading@10:     if (s->last_frame)
yading@10:         return 0;
yading@10: 
yading@10:     memcpy(s->buf, s->buf + NELLY_SAMPLES, NELLY_BUF_LEN * sizeof(*s->buf));
yading@10:     if (frame) {
yading@10:         memcpy(s->buf + NELLY_BUF_LEN, frame->data[0],
yading@10:                frame->nb_samples * sizeof(*s->buf));
yading@10:         if (frame->nb_samples < NELLY_SAMPLES) {
yading@10:             memset(s->buf + NELLY_BUF_LEN + frame->nb_samples, 0,
yading@10:                    (NELLY_SAMPLES - frame->nb_samples) * sizeof(*s->buf));
yading@10:             if (frame->nb_samples >= NELLY_BUF_LEN)
yading@10:                 s->last_frame = 1;
yading@10:         }
yading@10:         if ((ret = ff_af_queue_add(&s->afq, frame)) < 0)
yading@10:             return ret;
yading@10:     } else {
yading@10:         memset(s->buf + NELLY_BUF_LEN, 0, NELLY_SAMPLES * sizeof(*s->buf));
yading@10:         s->last_frame = 1;
yading@10:     }
yading@10: 
yading@10:     if ((ret = ff_alloc_packet2(avctx, avpkt, NELLY_BLOCK_LEN)) < 0)
yading@10:         return ret;
yading@10:     encode_block(s, avpkt->data, avpkt->size);
yading@10: 
yading@10:     /* Get the next frame pts/duration */
yading@10:     ff_af_queue_remove(&s->afq, avctx->frame_size, &avpkt->pts,
yading@10:                        &avpkt->duration);
yading@10: 
yading@10:     *got_packet_ptr = 1;
yading@10:     return 0;
yading@10: }
yading@10: 
yading@10: AVCodec ff_nellymoser_encoder = {
yading@10:     .name           = "nellymoser",
yading@10:     .type           = AVMEDIA_TYPE_AUDIO,
yading@10:     .id             = AV_CODEC_ID_NELLYMOSER,
yading@10:     .priv_data_size = sizeof(NellyMoserEncodeContext),
yading@10:     .init           = encode_init,
yading@10:     .encode2        = encode_frame,
yading@10:     .close          = encode_end,
yading@10:     .capabilities   = CODEC_CAP_SMALL_LAST_FRAME | CODEC_CAP_DELAY,
yading@10:     .long_name      = NULL_IF_CONFIG_SMALL("Nellymoser Asao"),
yading@10:     .sample_fmts    = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_FLT,
yading@10:                                                      AV_SAMPLE_FMT_NONE },
yading@10: };