1 /*
2  * Interface to libmp3lame for mp3 encoding
3  * Copyright (c) 2002 Lennert Buytenhek <buytenh@gnu.org>
4  *
5  * This file is part of FFmpeg.
6  *
7  * FFmpeg is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU Lesser General Public
9  * License as published by the Free Software Foundation; either
10  * version 2.1 of the License, or (at your option) any later version.
11  *
12  * FFmpeg is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15  * Lesser General Public License for more details.
16  *
17  * You should have received a copy of the GNU Lesser General Public
18  * License along with FFmpeg; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21 
22 /**
23  * @file
24  * Interface to libmp3lame for mp3 encoding.
25  */
26 
27 #include <lame/lame.h>
28 
29 #include "libavutil/channel_layout.h"
30 #include "libavutil/common.h"
31 #include "libavutil/float_dsp.h"
32 #include "libavutil/intreadwrite.h"
33 #include "libavutil/log.h"
34 #include "libavutil/opt.h"
35 #include "avcodec.h"
36 #include "audio_frame_queue.h"
37 #include "codec_internal.h"
38 #include "encode.h"
39 #include "mpegaudio.h"
40 #include "mpegaudiodecheader.h"
41 
42 #define BUFFER_SIZE (7200 + 2 * MPA_FRAME_SIZE + MPA_FRAME_SIZE / 4+1000) // FIXME: Buffer size to small? Adding 1000 to make up for it.
43 
44 typedef struct LAMEContext {
45     AVClass *class;
46     AVCodecContext *avctx;
47     lame_global_flags *gfp;
48     uint8_t *buffer;
49     int buffer_index;
50     int buffer_size;
51     int reservoir;
52     int joint_stereo;
53     int abr;
54     int delay_sent;
55     float *samples_flt[2];
56     AudioFrameQueue afq;
57     AVFloatDSPContext *fdsp;
58 } LAMEContext;
59 
60 
realloc_buffer(LAMEContext *s)61 static int realloc_buffer(LAMEContext *s)
62 {
63     if (!s->buffer || s->buffer_size - s->buffer_index < BUFFER_SIZE) {
64         int new_size = s->buffer_index + 2 * BUFFER_SIZE, err;
65 
66         ff_dlog(s->avctx, "resizing output buffer: %d -> %d\n", s->buffer_size,
67                 new_size);
68         if ((err = av_reallocp(&s->buffer, new_size)) < 0) {
69             s->buffer_size = s->buffer_index = 0;
70             return err;
71         }
72         s->buffer_size = new_size;
73     }
74     return 0;
75 }
76 
mp3lame_encode_close(AVCodecContext *avctx)77 static av_cold int mp3lame_encode_close(AVCodecContext *avctx)
78 {
79     LAMEContext *s = avctx->priv_data;
80 
81     av_freep(&s->samples_flt[0]);
82     av_freep(&s->samples_flt[1]);
83     av_freep(&s->buffer);
84     av_freep(&s->fdsp);
85 
86     ff_af_queue_close(&s->afq);
87 
88     lame_close(s->gfp);
89     return 0;
90 }
91 
mp3lame_encode_init(AVCodecContext *avctx)92 static av_cold int mp3lame_encode_init(AVCodecContext *avctx)
93 {
94     LAMEContext *s = avctx->priv_data;
95     int ret;
96 
97     s->avctx = avctx;
98 
99     /* initialize LAME and get defaults */
100     if (!(s->gfp = lame_init()))
101         return AVERROR(ENOMEM);
102 
103 
104     lame_set_num_channels(s->gfp, avctx->ch_layout.nb_channels);
105     lame_set_mode(s->gfp, avctx->ch_layout.nb_channels > 1 ?
106                           s->joint_stereo ? JOINT_STEREO : STEREO : MONO);
107 
108     /* sample rate */
109     lame_set_in_samplerate (s->gfp, avctx->sample_rate);
110     lame_set_out_samplerate(s->gfp, avctx->sample_rate);
111 
112     /* algorithmic quality */
113     if (avctx->compression_level != FF_COMPRESSION_DEFAULT)
114         lame_set_quality(s->gfp, avctx->compression_level);
115 
116     /* rate control */
117     if (avctx->flags & AV_CODEC_FLAG_QSCALE) { // VBR
118         lame_set_VBR(s->gfp, vbr_default);
119         lame_set_VBR_quality(s->gfp, avctx->global_quality / (float)FF_QP2LAMBDA);
120     } else {
121         if (avctx->bit_rate) {
122             if (s->abr) {                   // ABR
123                 lame_set_VBR(s->gfp, vbr_abr);
124                 lame_set_VBR_mean_bitrate_kbps(s->gfp, avctx->bit_rate / 1000);
125             } else                          // CBR
126                 lame_set_brate(s->gfp, avctx->bit_rate / 1000);
127         }
128     }
129 
130     /* lowpass cutoff frequency */
131     if (avctx->cutoff)
132         lame_set_lowpassfreq(s->gfp, avctx->cutoff);
133 
134     /* do not get a Xing VBR header frame from LAME */
135     lame_set_bWriteVbrTag(s->gfp,0);
136 
137     /* bit reservoir usage */
138     lame_set_disable_reservoir(s->gfp, !s->reservoir);
139 
140     /* set specified parameters */
141     if (lame_init_params(s->gfp) < 0) {
142         ret = AVERROR_EXTERNAL;
143         goto error;
144     }
145 
146     /* get encoder delay */
147     avctx->initial_padding = lame_get_encoder_delay(s->gfp) + 528 + 1;
148     ff_af_queue_init(avctx, &s->afq);
149 
150     avctx->frame_size  = lame_get_framesize(s->gfp);
151 
152     /* allocate float sample buffers */
153     if (avctx->sample_fmt == AV_SAMPLE_FMT_FLTP) {
154         int ch;
155         for (ch = 0; ch < avctx->ch_layout.nb_channels; ch++) {
156             s->samples_flt[ch] = av_malloc_array(avctx->frame_size,
157                                            sizeof(*s->samples_flt[ch]));
158             if (!s->samples_flt[ch]) {
159                 ret = AVERROR(ENOMEM);
160                 goto error;
161             }
162         }
163     }
164 
165     ret = realloc_buffer(s);
166     if (ret < 0)
167         goto error;
168 
169     s->fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
170     if (!s->fdsp) {
171         ret = AVERROR(ENOMEM);
172         goto error;
173     }
174 
175 
176     return 0;
177 error:
178     mp3lame_encode_close(avctx);
179     return ret;
180 }
181 
182 #define ENCODE_BUFFER(func, buf_type, buf_name) do {                        \
183     lame_result = func(s->gfp,                                              \
184                        (const buf_type *)buf_name[0],                       \
185                        (const buf_type *)buf_name[1], frame->nb_samples,    \
186                        s->buffer + s->buffer_index,                         \
187                        s->buffer_size - s->buffer_index);                   \
188 } while (0)
189 
mp3lame_encode_frame(AVCodecContext *avctx, AVPacket *avpkt, const AVFrame *frame, int *got_packet_ptr)190 static int mp3lame_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
191                                 const AVFrame *frame, int *got_packet_ptr)
192 {
193     LAMEContext *s = avctx->priv_data;
194     MPADecodeHeader hdr;
195     int len, ret, ch, discard_padding;
196     int lame_result;
197     uint32_t h;
198 
199     if (frame) {
200         switch (avctx->sample_fmt) {
201         case AV_SAMPLE_FMT_S16P:
202             ENCODE_BUFFER(lame_encode_buffer, int16_t, frame->data);
203             break;
204         case AV_SAMPLE_FMT_S32P:
205             ENCODE_BUFFER(lame_encode_buffer_int, int32_t, frame->data);
206             break;
207         case AV_SAMPLE_FMT_FLTP:
208             if (frame->linesize[0] < 4 * FFALIGN(frame->nb_samples, 8)) {
209                 av_log(avctx, AV_LOG_ERROR, "inadequate AVFrame plane padding\n");
210                 return AVERROR(EINVAL);
211             }
212             for (ch = 0; ch < avctx->ch_layout.nb_channels; ch++) {
213                 s->fdsp->vector_fmul_scalar(s->samples_flt[ch],
214                                            (const float *)frame->data[ch],
215                                            32768.0f,
216                                            FFALIGN(frame->nb_samples, 8));
217             }
218             ENCODE_BUFFER(lame_encode_buffer_float, float, s->samples_flt);
219             break;
220         default:
221             return AVERROR_BUG;
222         }
223     } else if (!s->afq.frame_alloc) {
224         lame_result = 0;
225     } else {
226         lame_result = lame_encode_flush(s->gfp, s->buffer + s->buffer_index,
227                                         s->buffer_size - s->buffer_index);
228     }
229     if (lame_result < 0) {
230         if (lame_result == -1) {
231             av_log(avctx, AV_LOG_ERROR,
232                    "lame: output buffer too small (buffer index: %d, free bytes: %d)\n",
233                    s->buffer_index, s->buffer_size - s->buffer_index);
234         }
235         return AVERROR(ENOMEM);
236     }
237     s->buffer_index += lame_result;
238     ret = realloc_buffer(s);
239     if (ret < 0) {
240         av_log(avctx, AV_LOG_ERROR, "error reallocating output buffer\n");
241         return ret;
242     }
243 
244     /* add current frame to the queue */
245     if (frame) {
246         if ((ret = ff_af_queue_add(&s->afq, frame)) < 0)
247             return ret;
248     }
249 
250     /* Move 1 frame from the LAME buffer to the output packet, if available.
251        We have to parse the first frame header in the output buffer to
252        determine the frame size. */
253     if (s->buffer_index < 4)
254         return 0;
255     h = AV_RB32(s->buffer);
256 
257     ret = avpriv_mpegaudio_decode_header(&hdr, h);
258     if (ret < 0) {
259         av_log(avctx, AV_LOG_ERROR, "Invalid mp3 header at start of buffer\n");
260         return AVERROR_BUG;
261     } else if (ret) {
262         av_log(avctx, AV_LOG_ERROR, "free format output not supported\n");
263         return AVERROR_INVALIDDATA;
264     }
265     len = hdr.frame_size;
266     ff_dlog(avctx, "in:%d packet-len:%d index:%d\n", avctx->frame_size, len,
267             s->buffer_index);
268     if (len <= s->buffer_index) {
269         if ((ret = ff_get_encode_buffer(avctx, avpkt, len, 0)) < 0)
270             return ret;
271         memcpy(avpkt->data, s->buffer, len);
272         s->buffer_index -= len;
273         memmove(s->buffer, s->buffer + len, s->buffer_index);
274 
275         /* Get the next frame pts/duration */
276         ff_af_queue_remove(&s->afq, avctx->frame_size, &avpkt->pts,
277                            &avpkt->duration);
278 
279         discard_padding = avctx->frame_size - avpkt->duration;
280         // Check if subtraction resulted in an overflow
281         if ((discard_padding < avctx->frame_size) != (avpkt->duration > 0)) {
282             av_log(avctx, AV_LOG_ERROR, "discard padding overflow\n");
283             av_packet_unref(avpkt);
284             return AVERROR(EINVAL);
285         }
286         if ((!s->delay_sent && avctx->initial_padding > 0) || discard_padding > 0) {
287             uint8_t* side_data = av_packet_new_side_data(avpkt,
288                                                          AV_PKT_DATA_SKIP_SAMPLES,
289                                                          10);
290             if(!side_data) {
291                 av_packet_unref(avpkt);
292                 return AVERROR(ENOMEM);
293             }
294             if (!s->delay_sent) {
295                 AV_WL32(side_data, avctx->initial_padding);
296                 s->delay_sent = 1;
297             }
298             AV_WL32(side_data + 4, discard_padding);
299         }
300 
301         *got_packet_ptr = 1;
302     }
303     return 0;
304 }
305 
306 #define OFFSET(x) offsetof(LAMEContext, x)
307 #define AE AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
308 static const AVOption options[] = {
309     { "reservoir",    "use bit reservoir", OFFSET(reservoir),    AV_OPT_TYPE_BOOL, { .i64 = 1 }, 0, 1, AE },
310     { "joint_stereo", "use joint stereo",  OFFSET(joint_stereo), AV_OPT_TYPE_BOOL, { .i64 = 1 }, 0, 1, AE },
311     { "abr",          "use ABR",           OFFSET(abr),          AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, AE },
312     { NULL },
313 };
314 
315 static const AVClass libmp3lame_class = {
316     .class_name = "libmp3lame encoder",
317     .item_name  = av_default_item_name,
318     .option     = options,
319     .version    = LIBAVUTIL_VERSION_INT,
320 };
321 
322 static const FFCodecDefault libmp3lame_defaults[] = {
323     { "b",          "0" },
324     { NULL },
325 };
326 
327 static const int libmp3lame_sample_rates[] = {
328     44100, 48000,  32000, 22050, 24000, 16000, 11025, 12000, 8000, 0
329 };
330 
331 const FFCodec ff_libmp3lame_encoder = {
332     .p.name                = "libmp3lame",
333     .p.long_name           = NULL_IF_CONFIG_SMALL("libmp3lame MP3 (MPEG audio layer 3)"),
334     .p.type                = AVMEDIA_TYPE_AUDIO,
335     .p.id                  = AV_CODEC_ID_MP3,
336     .p.capabilities        = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DELAY |
337                              AV_CODEC_CAP_SMALL_LAST_FRAME,
338     .priv_data_size        = sizeof(LAMEContext),
339     .init                  = mp3lame_encode_init,
340     FF_CODEC_ENCODE_CB(mp3lame_encode_frame),
341     .close                 = mp3lame_encode_close,
342     .p.sample_fmts         = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S32P,
343                                                              AV_SAMPLE_FMT_FLTP,
344                                                              AV_SAMPLE_FMT_S16P,
345                                                              AV_SAMPLE_FMT_NONE },
346     .p.supported_samplerates = libmp3lame_sample_rates,
347 #if FF_API_OLD_CHANNEL_LAYOUT
348     .p.channel_layouts     = (const uint64_t[]) { AV_CH_LAYOUT_MONO,
349                                                   AV_CH_LAYOUT_STEREO,
350                                                   0 },
351 #endif
352     .p.ch_layouts          = (const AVChannelLayout[]) { AV_CHANNEL_LAYOUT_MONO,
353                                                          AV_CHANNEL_LAYOUT_STEREO,
354                                                          { 0 },
355     },
356     .p.priv_class          = &libmp3lame_class,
357     .defaults              = libmp3lame_defaults,
358     .p.wrapper_name        = "libmp3lame",
359 };
360