2 * Copyright (c) CMU 1993 Computer Science, Speech Group
3 * Chengxiang Lu and Alex Hauptmann
4 * Copyright (c) 2005 Steve Underwood <steveu at coppice.org>
5 * Copyright (c) 2009 Kenan Gillet
6 * Copyright (c) 2010 Martin Storsjo
8 * This file is part of FFmpeg.
10 * FFmpeg is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU Lesser General Public
12 * License as published by the Free Software Foundation; either
13 * version 2.1 of the License, or (at your option) any later version.
15 * FFmpeg is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * Lesser General Public License for more details.
20 * You should have received a copy of the GNU Lesser General Public
21 * License along with FFmpeg; if not, write to the Free Software
22 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
27 * G.722 ADPCM audio encoder
30 #include "libavutil/avassert.h"
34 #include "libavutil/common.h"
36 #define FREEZE_INTERVAL 128
38 /* This is an arbitrary value. Allowing insanely large values leads to strange
39 problems, so we limit it to a reasonable value */
40 #define MAX_FRAME_SIZE 32768
42 /* We clip the value of avctx->trellis to prevent data type overflows and
43 undefined behavior. Using larger values is insanely slow anyway. */
45 #define MAX_TRELLIS 16
47 static av_cold
int g722_encode_close(AVCodecContext
*avctx
)
49 G722Context
*c
= avctx
->priv_data
;
51 for (i
= 0; i
< 2; i
++) {
52 av_freep(&c
->paths
[i
]);
53 av_freep(&c
->node_buf
[i
]);
54 av_freep(&c
->nodep_buf
[i
]);
59 static av_cold
int g722_encode_init(AVCodecContext
* avctx
)
61 G722Context
*c
= avctx
->priv_data
;
64 if (avctx
->channels
!= 1) {
65 av_log(avctx
, AV_LOG_ERROR
, "Only mono tracks are allowed.\n");
66 return AVERROR_INVALIDDATA
;
69 c
->band
[0].scale_factor
= 8;
70 c
->band
[1].scale_factor
= 2;
71 c
->prev_samples_pos
= 22;
74 int frontier
= 1 << avctx
->trellis
;
75 int max_paths
= frontier
* FREEZE_INTERVAL
;
77 for (i
= 0; i
< 2; i
++) {
78 c
->paths
[i
] = av_mallocz(max_paths
* sizeof(**c
->paths
));
79 c
->node_buf
[i
] = av_mallocz(2 * frontier
* sizeof(**c
->node_buf
));
80 c
->nodep_buf
[i
] = av_mallocz(2 * frontier
* sizeof(**c
->nodep_buf
));
81 if (!c
->paths
[i
] || !c
->node_buf
[i
] || !c
->nodep_buf
[i
]) {
82 ret
= AVERROR(ENOMEM
);
88 if (avctx
->frame_size
) {
89 /* validate frame size */
90 if (avctx
->frame_size
& 1 || avctx
->frame_size
> MAX_FRAME_SIZE
) {
93 if (avctx
->frame_size
== 1)
95 else if (avctx
->frame_size
> MAX_FRAME_SIZE
)
96 new_frame_size
= MAX_FRAME_SIZE
;
98 new_frame_size
= avctx
->frame_size
- 1;
100 av_log(avctx
, AV_LOG_WARNING
, "Requested frame size is not "
101 "allowed. Using %d instead of %d\n", new_frame_size
,
103 avctx
->frame_size
= new_frame_size
;
106 /* This is arbitrary. We use 320 because it's 20ms @ 16kHz, which is
107 a common packet size for VoIP applications */
108 avctx
->frame_size
= 320;
110 avctx
->initial_padding
= 22;
112 if (avctx
->trellis
) {
113 /* validate trellis */
114 if (avctx
->trellis
< MIN_TRELLIS
|| avctx
->trellis
> MAX_TRELLIS
) {
115 int new_trellis
= av_clip(avctx
->trellis
, MIN_TRELLIS
, MAX_TRELLIS
);
116 av_log(avctx
, AV_LOG_WARNING
, "Requested trellis value is not "
117 "allowed. Using %d instead of %d\n", new_trellis
,
119 avctx
->trellis
= new_trellis
;
125 g722_encode_close(avctx
);
129 static const int16_t low_quant
[33] = {
130 35, 72, 110, 150, 190, 233, 276, 323,
131 370, 422, 473, 530, 587, 650, 714, 786,
132 858, 940, 1023, 1121, 1219, 1339, 1458, 1612,
133 1765, 1980, 2195, 2557, 2919
136 static inline void filter_samples(G722Context
*c
, const int16_t *samples
,
137 int *xlow
, int *xhigh
)
140 c
->prev_samples
[c
->prev_samples_pos
++] = samples
[0];
141 c
->prev_samples
[c
->prev_samples_pos
++] = samples
[1];
142 ff_g722_apply_qmf(c
->prev_samples
+ c
->prev_samples_pos
- 24, &xout1
, &xout2
);
143 *xlow
= xout1
+ xout2
>> 14;
144 *xhigh
= xout1
- xout2
>> 14;
145 if (c
->prev_samples_pos
>= PREV_SAMPLES_BUF_SIZE
) {
146 memmove(c
->prev_samples
,
147 c
->prev_samples
+ c
->prev_samples_pos
- 22,
148 22 * sizeof(c
->prev_samples
[0]));
149 c
->prev_samples_pos
= 22;
153 static inline int encode_high(const struct G722Band
*state
, int xhigh
)
155 int diff
= av_clip_int16(xhigh
- state
->s_predictor
);
156 int pred
= 141 * state
->scale_factor
>> 8;
157 /* = diff >= 0 ? (diff < pred) + 2 : diff >= -pred */
158 return ((diff
^ (diff
>> (sizeof(diff
)*8-1))) < pred
) + 2*(diff
>= 0);
161 static inline int encode_low(const struct G722Band
* state
, int xlow
)
163 int diff
= av_clip_int16(xlow
- state
->s_predictor
);
164 /* = diff >= 0 ? diff : -(diff + 1) */
165 int limit
= diff
^ (diff
>> (sizeof(diff
)*8-1));
167 limit
= limit
+ 1 << 10;
168 if (limit
> low_quant
[8] * state
->scale_factor
)
170 while (i
< 29 && limit
> low_quant
[i
] * state
->scale_factor
)
172 return (diff
< 0 ? (i
< 2 ? 63 : 33) : 61) - i
;
175 static void g722_encode_trellis(G722Context
*c
, int trellis
,
176 uint8_t *dst
, int nb_samples
,
177 const int16_t *samples
)
180 int frontier
= 1 << trellis
;
181 struct TrellisNode
**nodes
[2];
182 struct TrellisNode
**nodes_next
[2];
183 int pathn
[2] = {0, 0}, froze
= -1;
184 struct TrellisPath
*p
[2];
186 for (i
= 0; i
< 2; i
++) {
187 nodes
[i
] = c
->nodep_buf
[i
];
188 nodes_next
[i
] = c
->nodep_buf
[i
] + frontier
;
189 memset(c
->nodep_buf
[i
], 0, 2 * frontier
* sizeof(*c
->nodep_buf
[i
]));
190 nodes
[i
][0] = c
->node_buf
[i
] + frontier
;
191 nodes
[i
][0]->ssd
= 0;
192 nodes
[i
][0]->path
= 0;
193 nodes
[i
][0]->state
= c
->band
[i
];
196 for (i
= 0; i
< nb_samples
>> 1; i
++) {
198 struct TrellisNode
*next
[2];
199 int heap_pos
[2] = {0, 0};
201 for (j
= 0; j
< 2; j
++) {
202 next
[j
] = c
->node_buf
[j
] + frontier
*(i
& 1);
203 memset(nodes_next
[j
], 0, frontier
* sizeof(**nodes_next
));
206 filter_samples(c
, &samples
[2*i
], &xlow
, &xhigh
);
208 for (j
= 0; j
< frontier
&& nodes
[0][j
]; j
++) {
209 /* Only k >> 2 affects the future adaptive state, therefore testing
210 * small steps that don't change k >> 2 is useless, the original
211 * value from encode_low is better than them. Since we step k
212 * in steps of 4, make sure range is a multiple of 4, so that
213 * we don't miss the original value from encode_low. */
214 int range
= j
< frontier
/2 ? 4 : 0;
215 struct TrellisNode
*cur_node
= nodes
[0][j
];
217 int ilow
= encode_low(&cur_node
->state
, xlow
);
219 for (k
= ilow
- range
; k
<= ilow
+ range
&& k
<= 63; k
+= 4) {
220 int decoded
, dec_diff
, pos
;
222 struct TrellisNode
* node
;
227 decoded
= av_clip((cur_node
->state
.scale_factor
*
228 ff_g722_low_inv_quant6
[k
] >> 10)
229 + cur_node
->state
.s_predictor
, -16384, 16383);
230 dec_diff
= xlow
- decoded
;
232 #define STORE_NODE(index, UPDATE, VALUE)\
233 ssd = cur_node->ssd + dec_diff*dec_diff;\
234 /* Check for wraparound. Using 64 bit ssd counters would \
235 * be simpler, but is slower on x86 32 bit. */\
236 if (ssd < cur_node->ssd)\
238 if (heap_pos[index] < frontier) {\
239 pos = heap_pos[index]++;\
240 av_assert2(pathn[index] < FREEZE_INTERVAL * frontier);\
241 node = nodes_next[index][pos] = next[index]++;\
242 node->path = pathn[index]++;\
244 /* Try to replace one of the leaf nodes with the new \
245 * one, but not always testing the same leaf position */\
246 pos = (frontier>>1) + (heap_pos[index] & ((frontier>>1) - 1));\
247 if (ssd >= nodes_next[index][pos]->ssd)\
250 node = nodes_next[index][pos];\
253 node->state = cur_node->state;\
255 c->paths[index][node->path].value = VALUE;\
256 c->paths[index][node->path].prev = cur_node->path;\
257 /* Sift the newly inserted node up in the heap to restore \
258 * the heap property */\
260 int parent = (pos - 1) >> 1;\
261 if (nodes_next[index][parent]->ssd <= ssd)\
263 FFSWAP(struct TrellisNode*, nodes_next[index][parent],\
264 nodes_next[index][pos]);\
267 STORE_NODE(0, ff_g722_update_low_predictor(&node
->state
, k
>> 2), k
);
271 for (j
= 0; j
< frontier
&& nodes
[1][j
]; j
++) {
273 struct TrellisNode
*cur_node
= nodes
[1][j
];
275 /* We don't try to get any initial guess for ihigh via
276 * encode_high - since there's only 4 possible values, test
277 * them all. Testing all of these gives a much, much larger
278 * gain than testing a larger range around ilow. */
279 for (ihigh
= 0; ihigh
< 4; ihigh
++) {
280 int dhigh
, decoded
, dec_diff
, pos
;
282 struct TrellisNode
* node
;
284 dhigh
= cur_node
->state
.scale_factor
*
285 ff_g722_high_inv_quant
[ihigh
] >> 10;
286 decoded
= av_clip(dhigh
+ cur_node
->state
.s_predictor
,
288 dec_diff
= xhigh
- decoded
;
290 STORE_NODE(1, ff_g722_update_high_predictor(&node
->state
, dhigh
, ihigh
), ihigh
);
294 for (j
= 0; j
< 2; j
++) {
295 FFSWAP(struct TrellisNode
**, nodes
[j
], nodes_next
[j
]);
297 if (nodes
[j
][0]->ssd
> (1 << 16)) {
298 for (k
= 1; k
< frontier
&& nodes
[j
][k
]; k
++)
299 nodes
[j
][k
]->ssd
-= nodes
[j
][0]->ssd
;
300 nodes
[j
][0]->ssd
= 0;
304 if (i
== froze
+ FREEZE_INTERVAL
) {
305 p
[0] = &c
->paths
[0][nodes
[0][0]->path
];
306 p
[1] = &c
->paths
[1][nodes
[1][0]->path
];
307 for (j
= i
; j
> froze
; j
--) {
308 dst
[j
] = p
[1]->value
<< 6 | p
[0]->value
;
309 p
[0] = &c
->paths
[0][p
[0]->prev
];
310 p
[1] = &c
->paths
[1][p
[1]->prev
];
313 pathn
[0] = pathn
[1] = 0;
314 memset(nodes
[0] + 1, 0, (frontier
- 1)*sizeof(**nodes
));
315 memset(nodes
[1] + 1, 0, (frontier
- 1)*sizeof(**nodes
));
319 p
[0] = &c
->paths
[0][nodes
[0][0]->path
];
320 p
[1] = &c
->paths
[1][nodes
[1][0]->path
];
321 for (j
= i
; j
> froze
; j
--) {
322 dst
[j
] = p
[1]->value
<< 6 | p
[0]->value
;
323 p
[0] = &c
->paths
[0][p
[0]->prev
];
324 p
[1] = &c
->paths
[1][p
[1]->prev
];
326 c
->band
[0] = nodes
[0][0]->state
;
327 c
->band
[1] = nodes
[1][0]->state
;
330 static av_always_inline
void encode_byte(G722Context
*c
, uint8_t *dst
,
331 const int16_t *samples
)
333 int xlow
, xhigh
, ilow
, ihigh
;
334 filter_samples(c
, samples
, &xlow
, &xhigh
);
335 ihigh
= encode_high(&c
->band
[1], xhigh
);
336 ilow
= encode_low (&c
->band
[0], xlow
);
337 ff_g722_update_high_predictor(&c
->band
[1], c
->band
[1].scale_factor
*
338 ff_g722_high_inv_quant
[ihigh
] >> 10, ihigh
);
339 ff_g722_update_low_predictor(&c
->band
[0], ilow
>> 2);
340 *dst
= ihigh
<< 6 | ilow
;
343 static void g722_encode_no_trellis(G722Context
*c
,
344 uint8_t *dst
, int nb_samples
,
345 const int16_t *samples
)
348 for (i
= 0; i
< nb_samples
; i
+= 2)
349 encode_byte(c
, dst
++, &samples
[i
]);
352 static int g722_encode_frame(AVCodecContext
*avctx
, AVPacket
*avpkt
,
353 const AVFrame
*frame
, int *got_packet_ptr
)
355 G722Context
*c
= avctx
->priv_data
;
356 const int16_t *samples
= (const int16_t *)frame
->data
[0];
357 int nb_samples
, out_size
, ret
;
359 out_size
= (frame
->nb_samples
+ 1) / 2;
360 if ((ret
= ff_alloc_packet2(avctx
, avpkt
, out_size
)) < 0)
363 nb_samples
= frame
->nb_samples
- (frame
->nb_samples
& 1);
366 g722_encode_trellis(c
, avctx
->trellis
, avpkt
->data
, nb_samples
, samples
);
368 g722_encode_no_trellis(c
, avpkt
->data
, nb_samples
, samples
);
370 /* handle last frame with odd frame_size */
371 if (nb_samples
< frame
->nb_samples
) {
372 int16_t last_samples
[2] = { samples
[nb_samples
], samples
[nb_samples
] };
373 encode_byte(c
, &avpkt
->data
[nb_samples
>> 1], last_samples
);
376 if (frame
->pts
!= AV_NOPTS_VALUE
)
377 avpkt
->pts
= frame
->pts
- ff_samples_to_time_base(avctx
, avctx
->initial_padding
);
382 AVCodec ff_adpcm_g722_encoder
= {
384 .long_name
= NULL_IF_CONFIG_SMALL("G.722 ADPCM"),
385 .type
= AVMEDIA_TYPE_AUDIO
,
386 .id
= AV_CODEC_ID_ADPCM_G722
,
387 .priv_data_size
= sizeof(G722Context
),
388 .init
= g722_encode_init
,
389 .close
= g722_encode_close
,
390 .encode2
= g722_encode_frame
,
391 .capabilities
= CODEC_CAP_SMALL_LAST_FRAME
,
392 .sample_fmts
= (const enum AVSampleFormat
[]){ AV_SAMPLE_FMT_S16
,
393 AV_SAMPLE_FMT_NONE
},