src/platform/ffmpeg/ffmpeg-encoder.c (view raw)
1#include "ffmpeg-encoder.h"
2
3#include "gba-video.h"
4
5#include <libavcodec/avcodec.h>
6
7#include <libavutil/imgutils.h>
8#include <libavutil/opt.h>
9
10#include <libavresample/avresample.h>
11#include <libswscale/swscale.h>
12
13static void _ffmpegPostVideoFrame(struct GBAAVStream*, struct GBAVideoRenderer* renderer);
14static void _ffmpegPostAudioFrame(struct GBAAVStream*, int32_t left, int32_t right);
15
16enum {
17 PREFERRED_SAMPLE_RATE = 0x8000
18};
19
20void FFmpegEncoderInit(struct FFmpegEncoder* encoder) {
21 av_register_all();
22
23 encoder->d.postVideoFrame = _ffmpegPostVideoFrame;
24 encoder->d.postAudioFrame = _ffmpegPostAudioFrame;
25
26 encoder->audioCodec = 0;
27 encoder->videoCodec = 0;
28 encoder->containerFormat = 0;
29 FFmpegEncoderSetAudio(encoder, "flac", 0);
30 FFmpegEncoderSetVideo(encoder, "png", 0);
31 FFmpegEncoderSetContainer(encoder, "matroska");
32 FFmpegEncoderSetDimensions(encoder, VIDEO_HORIZONTAL_PIXELS, VIDEO_VERTICAL_PIXELS);
33 encoder->resampleContext = 0;
34 encoder->absf = 0;
35 encoder->context = 0;
36}
37
38bool FFmpegEncoderSetAudio(struct FFmpegEncoder* encoder, const char* acodec, unsigned abr) {
39 static const struct {
40 int format;
41 int priority;
42 } priorities[] = {
43 { AV_SAMPLE_FMT_S16, 0 },
44 { AV_SAMPLE_FMT_S16P, 1 },
45 { AV_SAMPLE_FMT_S32, 2 },
46 { AV_SAMPLE_FMT_S32P, 2 },
47 { AV_SAMPLE_FMT_FLT, 3 },
48 { AV_SAMPLE_FMT_FLTP, 3 },
49 { AV_SAMPLE_FMT_DBL, 4 },
50 { AV_SAMPLE_FMT_DBLP, 4 }
51 };
52
53 if (!acodec) {
54 encoder->audioCodec = 0;
55 return true;
56 }
57
58 AVCodec* codec = avcodec_find_encoder_by_name(acodec);
59 if (!codec) {
60 return false;
61 }
62
63 if (!codec->sample_fmts) {
64 return false;
65 }
66 size_t i;
67 size_t j;
68 int priority = INT_MAX;
69 encoder->sampleFormat = AV_SAMPLE_FMT_NONE;
70 for (i = 0; codec->sample_fmts[i] != AV_SAMPLE_FMT_NONE; ++i) {
71 for (j = 0; j < sizeof(priorities) / sizeof(*priorities); ++j) {
72 if (codec->sample_fmts[i] == priorities[j].format && priority > priorities[j].priority) {
73 priority = priorities[j].priority;
74 encoder->sampleFormat = codec->sample_fmts[i];
75 }
76 }
77 }
78 if (encoder->sampleFormat == AV_SAMPLE_FMT_NONE) {
79 return false;
80 }
81 encoder->sampleRate = PREFERRED_SAMPLE_RATE;
82 if (codec->supported_samplerates) {
83 for (i = 0; codec->supported_samplerates[i]; ++i) {
84 if (codec->supported_samplerates[i] < PREFERRED_SAMPLE_RATE) {
85 continue;
86 }
87 if (encoder->sampleRate == PREFERRED_SAMPLE_RATE || encoder->sampleRate > codec->supported_samplerates[i]) {
88 encoder->sampleRate = codec->supported_samplerates[i];
89 }
90 }
91 } else if (codec->id == AV_CODEC_ID_AAC) {
92 // HACK: AAC doesn't support 32768Hz (it rounds to 32000), but libfaac doesn't tell us that
93 encoder->sampleRate = 44100;
94 }
95 encoder->audioCodec = acodec;
96 encoder->audioBitrate = abr;
97 return true;
98}
99
100bool FFmpegEncoderSetVideo(struct FFmpegEncoder* encoder, const char* vcodec, unsigned vbr) {
101 static const struct {
102 enum AVPixelFormat format;
103 int priority;
104 } priorities[] = {
105 { AV_PIX_FMT_RGB555, 0 },
106 { AV_PIX_FMT_BGR555, 0 },
107 { AV_PIX_FMT_RGB565, 1 },
108 { AV_PIX_FMT_BGR565, 1 },
109 { AV_PIX_FMT_RGB24, 2 },
110 { AV_PIX_FMT_BGR24, 2 },
111 { AV_PIX_FMT_BGR0, 3 },
112 { AV_PIX_FMT_RGB0, 3 },
113 { AV_PIX_FMT_0BGR, 3 },
114 { AV_PIX_FMT_0RGB, 3 },
115 { AV_PIX_FMT_RGB8, 3 },
116 { AV_PIX_FMT_BGR8, 3 },
117 { AV_PIX_FMT_YUV422P, 4 },
118 { AV_PIX_FMT_YUV444P, 5 },
119 { AV_PIX_FMT_YUV420P, 6 }
120 };
121 AVCodec* codec = avcodec_find_encoder_by_name(vcodec);
122 if (!codec) {
123 return false;
124 }
125
126 size_t i;
127 size_t j;
128 int priority = INT_MAX;
129 encoder->pixFormat = AV_PIX_FMT_NONE;
130 for (i = 0; codec->pix_fmts[i] != AV_PIX_FMT_NONE; ++i) {
131 for (j = 0; j < sizeof(priorities) / sizeof(*priorities); ++j) {
132 if (codec->pix_fmts[i] == priorities[j].format && priority > priorities[j].priority) {
133 priority = priorities[j].priority;
134 encoder->pixFormat = codec->pix_fmts[i];
135 }
136 }
137 }
138 if (encoder->pixFormat == AV_PIX_FMT_NONE) {
139 return false;
140 }
141 encoder->videoCodec = vcodec;
142 encoder->videoBitrate = vbr;
143 return true;
144}
145
146bool FFmpegEncoderSetContainer(struct FFmpegEncoder* encoder, const char* container) {
147 AVOutputFormat* oformat = av_guess_format(container, 0, 0);
148 if (!oformat) {
149 return false;
150 }
151 encoder->containerFormat = container;
152 return true;
153}
154
155void FFmpegEncoderSetDimensions(struct FFmpegEncoder* encoder, int width, int height) {
156 encoder->width = width > 0 ? width : VIDEO_HORIZONTAL_PIXELS;
157 encoder->height = height > 0 ? height : VIDEO_VERTICAL_PIXELS;
158}
159
160bool FFmpegEncoderVerifyContainer(struct FFmpegEncoder* encoder) {
161 AVOutputFormat* oformat = av_guess_format(encoder->containerFormat, 0, 0);
162 AVCodec* acodec = avcodec_find_encoder_by_name(encoder->audioCodec);
163 AVCodec* vcodec = avcodec_find_encoder_by_name(encoder->videoCodec);
164 if ((encoder->audioCodec && !acodec) || !vcodec || !oformat) {
165 return false;
166 }
167 if (encoder->audioCodec && !avformat_query_codec(oformat, acodec->id, FF_COMPLIANCE_EXPERIMENTAL)) {
168 return false;
169 }
170 if (!avformat_query_codec(oformat, vcodec->id, FF_COMPLIANCE_EXPERIMENTAL)) {
171 return false;
172 }
173 return true;
174}
175
176bool FFmpegEncoderOpen(struct FFmpegEncoder* encoder, const char* outfile) {
177 AVCodec* acodec = avcodec_find_encoder_by_name(encoder->audioCodec);
178 AVCodec* vcodec = avcodec_find_encoder_by_name(encoder->videoCodec);
179 if ((encoder->audioCodec && !acodec) || !vcodec || !FFmpegEncoderVerifyContainer(encoder)) {
180 return false;
181 }
182
183 encoder->currentAudioSample = 0;
184 encoder->currentAudioFrame = 0;
185 encoder->currentVideoFrame = 0;
186 encoder->nextAudioPts = 0;
187
188 avformat_alloc_output_context2(&encoder->context, 0, 0, outfile);
189
190 encoder->context->oformat = av_guess_format(encoder->containerFormat, 0, 0);
191
192 if (acodec) {
193 encoder->audioStream = avformat_new_stream(encoder->context, acodec);
194 encoder->audio = encoder->audioStream->codec;
195 encoder->audio->bit_rate = encoder->audioBitrate;
196 encoder->audio->channels = 2;
197 encoder->audio->channel_layout = AV_CH_LAYOUT_STEREO;
198 encoder->audio->sample_rate = encoder->sampleRate;
199 encoder->audio->sample_fmt = encoder->sampleFormat;
200 AVDictionary* opts = 0;
201 av_dict_set(&opts, "strict", "-2", 0);
202 if (encoder->context->oformat->flags & AVFMT_GLOBALHEADER) {
203 encoder->audio->flags |= CODEC_FLAG_GLOBAL_HEADER;
204 }
205 avcodec_open2(encoder->audio, acodec, &opts);
206 av_dict_free(&opts);
207 encoder->audioFrame = av_frame_alloc();
208 encoder->audioFrame->nb_samples = encoder->audio->frame_size;
209 encoder->audioFrame->format = encoder->audio->sample_fmt;
210 encoder->audioFrame->pts = 0;
211 encoder->resampleContext = avresample_alloc_context();
212 av_opt_set_int(encoder->resampleContext, "in_channel_layout", AV_CH_LAYOUT_STEREO, 0);
213 av_opt_set_int(encoder->resampleContext, "out_channel_layout", AV_CH_LAYOUT_STEREO, 0);
214 av_opt_set_int(encoder->resampleContext, "in_sample_rate", PREFERRED_SAMPLE_RATE, 0);
215 av_opt_set_int(encoder->resampleContext, "out_sample_rate", encoder->sampleRate, 0);
216 av_opt_set_int(encoder->resampleContext, "in_sample_fmt", AV_SAMPLE_FMT_S16, 0);
217 av_opt_set_int(encoder->resampleContext, "out_sample_fmt", encoder->sampleFormat, 0);
218 avresample_open(encoder->resampleContext);
219 encoder->audioBufferSize = (encoder->audioFrame->nb_samples * PREFERRED_SAMPLE_RATE / encoder->sampleRate) * 4;
220 encoder->audioBuffer = av_malloc(encoder->audioBufferSize);
221 encoder->postaudioBufferSize = av_samples_get_buffer_size(0, encoder->audio->channels, encoder->audio->frame_size, encoder->audio->sample_fmt, 0);
222 encoder->postaudioBuffer = av_malloc(encoder->postaudioBufferSize);
223 avcodec_fill_audio_frame(encoder->audioFrame, encoder->audio->channels, encoder->audio->sample_fmt, (const uint8_t*) encoder->postaudioBuffer, encoder->postaudioBufferSize, 0);
224
225 if (encoder->audio->codec->id == AV_CODEC_ID_AAC &&
226 (strcasecmp(encoder->containerFormat, "mp4") ||
227 strcasecmp(encoder->containerFormat, "m4v") ||
228 strcasecmp(encoder->containerFormat, "mov"))) {
229 // MP4 container doesn't support the raw ADTS AAC format that the encoder spits out
230 encoder->absf = av_bitstream_filter_init("aac_adtstoasc");
231 }
232 }
233
234 encoder->videoStream = avformat_new_stream(encoder->context, vcodec);
235 encoder->video = encoder->videoStream->codec;
236 encoder->video->bit_rate = encoder->videoBitrate;
237 encoder->video->width = encoder->width;
238 encoder->video->height = encoder->height;
239 encoder->video->time_base = (AVRational) { VIDEO_TOTAL_LENGTH, GBA_ARM7TDMI_FREQUENCY };
240 encoder->video->pix_fmt = encoder->pixFormat;
241 encoder->video->gop_size = 60;
242 encoder->video->max_b_frames = 3;
243 if (encoder->context->oformat->flags & AVFMT_GLOBALHEADER) {
244 encoder->video->flags |= CODEC_FLAG_GLOBAL_HEADER;
245 }
246 if (strcmp(vcodec->name, "libx264") == 0) {
247 // Try to adaptively figure out when you can use a slower encoder
248 if (encoder->width * encoder->height > 1000000) {
249 av_opt_set(encoder->video->priv_data, "preset", "superfast", 0);
250 } else if (encoder->width * encoder->height > 500000) {
251 av_opt_set(encoder->video->priv_data, "preset", "veryfast", 0);
252 } else {
253 av_opt_set(encoder->video->priv_data, "preset", "faster", 0);
254 }
255 av_opt_set(encoder->video->priv_data, "tune", "zerolatency", 0);
256 }
257 avcodec_open2(encoder->video, vcodec, 0);
258 encoder->videoFrame = av_frame_alloc();
259 encoder->videoFrame->format = encoder->video->pix_fmt;
260 encoder->videoFrame->width = encoder->video->width;
261 encoder->videoFrame->height = encoder->video->height;
262 encoder->videoFrame->pts = 0;
263 encoder->scaleContext = sws_getContext(VIDEO_HORIZONTAL_PIXELS, VIDEO_VERTICAL_PIXELS, AV_PIX_FMT_0BGR32,
264 encoder->videoFrame->width, encoder->videoFrame->height, encoder->video->pix_fmt,
265 SWS_POINT, 0, 0, 0);
266 av_image_alloc(encoder->videoFrame->data, encoder->videoFrame->linesize, encoder->video->width, encoder->video->height, encoder->video->pix_fmt, 32);
267
268 avio_open(&encoder->context->pb, outfile, AVIO_FLAG_WRITE);
269 avformat_write_header(encoder->context, 0);
270
271 return true;
272}
273
274void FFmpegEncoderClose(struct FFmpegEncoder* encoder) {
275 if (!encoder->context) {
276 return;
277 }
278 av_write_trailer(encoder->context);
279 avio_close(encoder->context->pb);
280
281 if (encoder->audioCodec) {
282 av_free(encoder->postaudioBuffer);
283 if (encoder->audioBuffer) {
284 av_free(encoder->audioBuffer);
285 }
286 av_frame_free(&encoder->audioFrame);
287 avcodec_close(encoder->audio);
288
289 if (encoder->resampleContext) {
290 avresample_close(encoder->resampleContext);
291 }
292
293 if (encoder->absf) {
294 av_bitstream_filter_close(encoder->absf);
295 encoder->absf = 0;
296 }
297 }
298
299 av_frame_free(&encoder->videoFrame);
300 avcodec_close(encoder->video);
301
302 sws_freeContext(encoder->scaleContext);
303
304 avformat_free_context(encoder->context);
305 encoder->context = 0;
306}
307
308bool FFmpegEncoderIsOpen(struct FFmpegEncoder* encoder) {
309 return !!encoder->context;
310}
311
312void _ffmpegPostAudioFrame(struct GBAAVStream* stream, int32_t left, int32_t right) {
313 struct FFmpegEncoder* encoder = (struct FFmpegEncoder*) stream;
314 if (!encoder->context || !encoder->audioCodec) {
315 return;
316 }
317
318 av_frame_make_writable(encoder->audioFrame);
319 encoder->audioBuffer[encoder->currentAudioSample * 2] = left;
320 encoder->audioBuffer[encoder->currentAudioSample * 2 + 1] = right;
321
322 ++encoder->currentAudioFrame;
323 ++encoder->currentAudioSample;
324
325 if ((encoder->currentAudioSample * 4) < encoder->audioBufferSize) {
326 return;
327 }
328 encoder->currentAudioSample = 0;
329
330 int channelSize = 2 * av_get_bytes_per_sample(encoder->audio->sample_fmt);
331 avresample_convert(encoder->resampleContext,
332 0, 0, encoder->postaudioBufferSize / channelSize,
333 (uint8_t**) &encoder->audioBuffer, 0, encoder->audioBufferSize / 4);
334 if ((ssize_t) avresample_available(encoder->resampleContext) < (ssize_t) encoder->postaudioBufferSize / channelSize) {
335 return;
336 }
337 avresample_read(encoder->resampleContext, encoder->audioFrame->data, encoder->postaudioBufferSize / channelSize);
338
339 AVRational timeBase = { 1, PREFERRED_SAMPLE_RATE };
340 encoder->audioFrame->pts = encoder->nextAudioPts;
341 encoder->nextAudioPts = av_rescale_q(encoder->currentAudioFrame, timeBase, encoder->audioStream->time_base);
342
343 AVPacket packet;
344 av_init_packet(&packet);
345 packet.data = 0;
346 packet.size = 0;
347 int gotData;
348 avcodec_encode_audio2(encoder->audio, &packet, encoder->audioFrame, &gotData);
349 if (gotData) {
350 if (encoder->absf) {
351 AVPacket tempPacket = packet;
352 int success = av_bitstream_filter_filter(encoder->absf, encoder->audio, 0,
353 &tempPacket.data, &tempPacket.size,
354 packet.data, packet.size, 0);
355 if (success > 0) {
356 tempPacket.buf = av_buffer_create(tempPacket.data, tempPacket.size, av_buffer_default_free, 0, 0);
357 av_free_packet(&packet);
358 }
359 packet = tempPacket;
360 }
361 packet.stream_index = encoder->audioStream->index;
362 av_interleaved_write_frame(encoder->context, &packet);
363 }
364 av_free_packet(&packet);
365}
366
367void _ffmpegPostVideoFrame(struct GBAAVStream* stream, struct GBAVideoRenderer* renderer) {
368 struct FFmpegEncoder* encoder = (struct FFmpegEncoder*) stream;
369 if (!encoder->context) {
370 return;
371 }
372 uint8_t* pixels;
373 unsigned stride;
374 renderer->getPixels(renderer, &stride, (void**) &pixels);
375 stride *= 4;
376
377 AVPacket packet;
378
379 av_init_packet(&packet);
380 packet.data = 0;
381 packet.size = 0;
382 av_frame_make_writable(encoder->videoFrame);
383 encoder->videoFrame->pts = av_rescale_q(encoder->currentVideoFrame, encoder->video->time_base, encoder->videoStream->time_base);
384 ++encoder->currentVideoFrame;
385
386 sws_scale(encoder->scaleContext, (const uint8_t* const*) &pixels, (const int*) &stride, 0, VIDEO_VERTICAL_PIXELS, encoder->videoFrame->data, encoder->videoFrame->linesize);
387
388 int gotData;
389 avcodec_encode_video2(encoder->video, &packet, encoder->videoFrame, &gotData);
390 if (gotData) {
391 if (encoder->videoStream->codec->coded_frame->key_frame) {
392 packet.flags |= AV_PKT_FLAG_KEY;
393 }
394 packet.stream_index = encoder->videoStream->index;
395 av_interleaved_write_frame(encoder->context, &packet);
396 }
397 av_free_packet(&packet);
398}