encoders/generic: Implement video encoding and always use frame queue

This implements video encoding for the generic encoder by using frame queue and a set lag amount (in frames). The lag amount determines for how many frames no repeated attempts at submitting or retrieving a frame should be made, and after that many frames have been submitted the generic encoder will try to retrieve packets if at all possible. If not possible, it will delay for as long as possible to show an Encoder overloaded message.
This commit is contained in:
Michael Fabian 'Xaymar' Dirks
2019-07-06 14:22:15 +02:00
parent 9a58a6a5ff
commit 0a20b750b8
2 changed files with 218 additions and 106 deletions
+195 -87
View File
@@ -47,7 +47,6 @@ extern "C" {
#define P_MULTITHREADING "MultiThreading"
#define P_MULTITHREADING_MODEL "MultiThreading.Model"
#define P_MULTITHREADING_THREADCOUNT "MultiThreading.ThreadCount"
#define P_MULTITHREADING_FRAMEQUEUE "MultiThreading.FrameQueue"
// FFmpeg
#define P_FFMPEG "FFmpeg"
@@ -171,51 +170,6 @@ void encoder::generic_factory::register_encoder()
throw;
}
};
this->info.oei.encode = [](void* ptr, struct encoder_frame* frame, struct encoder_packet* packet,
bool* received_packet) {
try {
return reinterpret_cast<generic*>(ptr)->encode(frame, packet, received_packet);
} catch (std::exception const& e) {
PLOG_ERROR("exception: %s", e.what());
throw e;
} catch (...) {
PLOG_ERROR("unknown exception");
throw;
}
};
this->info.oei.get_audio_info = [](void* ptr, struct audio_convert_info* info) {
try {
reinterpret_cast<generic*>(ptr)->get_audio_info(info);
} catch (std::exception const& e) {
PLOG_ERROR("exception: %s", e.what());
throw e;
} catch (...) {
PLOG_ERROR("unknown exception");
throw;
}
};
this->info.oei.get_frame_size = [](void* ptr) {
try {
return reinterpret_cast<generic*>(ptr)->get_frame_size();
} catch (std::exception const& e) {
PLOG_ERROR("exception: %s", e.what());
throw e;
} catch (...) {
PLOG_ERROR("unknown exception");
throw;
}
};
this->info.oei.get_video_info = [](void* ptr, struct video_scale_info* info) {
try {
reinterpret_cast<generic*>(ptr)->get_video_info(info);
} catch (std::exception const& e) {
PLOG_ERROR("exception: %s", e.what());
throw e;
} catch (...) {
PLOG_ERROR("unknown exception");
throw;
}
};
this->info.oei.get_sei_data = [](void* ptr, uint8_t** sei_data, size_t* size) {
try {
return reinterpret_cast<generic*>(ptr)->get_sei_data(sei_data, size);
@@ -238,11 +192,11 @@ void encoder::generic_factory::register_encoder()
throw;
}
};
this->info.oei.encode_texture = [](void* ptr, uint32_t handle, int64_t pts, uint64_t lock_key,
uint64_t* next_key, struct encoder_packet* packet, bool* received_packet) {
if (this->avcodec_ptr->type == AVMediaType::AVMEDIA_TYPE_VIDEO) {
this->info.oei.get_video_info = [](void* ptr, struct video_scale_info* info) {
try {
return reinterpret_cast<generic*>(ptr)->encode_texture(handle, pts, lock_key, next_key, packet,
received_packet);
reinterpret_cast<generic*>(ptr)->get_video_info(info);
} catch (std::exception const& e) {
PLOG_ERROR("exception: %s", e.what());
throw e;
@@ -251,6 +205,69 @@ void encoder::generic_factory::register_encoder()
throw;
}
};
this->info.oei.encode = [](void* ptr, struct encoder_frame* frame, struct encoder_packet* packet,
bool* received_packet) {
try {
return reinterpret_cast<generic*>(ptr)->video_encode(frame, packet, received_packet);
} catch (std::exception const& e) {
PLOG_ERROR("exception: %s", e.what());
throw e;
} catch (...) {
PLOG_ERROR("unknown exception");
throw;
}
};
this->info.oei.encode_texture = [](void* ptr, uint32_t handle, int64_t pts, uint64_t lock_key,
uint64_t* next_key, struct encoder_packet* packet,
bool* received_packet) {
try {
return reinterpret_cast<generic*>(ptr)->video_encode_texture(
handle, pts, lock_key, next_key, packet, received_packet);
} catch (std::exception const& e) {
PLOG_ERROR("exception: %s", e.what());
throw e;
} catch (...) {
PLOG_ERROR("unknown exception");
throw;
}
};
} else if (this->avcodec_ptr->type == AVMediaType::AVMEDIA_TYPE_AUDIO) {
this->info.oei.get_audio_info = [](void* ptr, struct audio_convert_info* info) {
try {
reinterpret_cast<generic*>(ptr)->get_audio_info(info);
} catch (std::exception const& e) {
PLOG_ERROR("exception: %s", e.what());
throw e;
} catch (...) {
PLOG_ERROR("unknown exception");
throw;
}
};
this->info.oei.get_frame_size = [](void* ptr) {
try {
return reinterpret_cast<generic*>(ptr)->get_frame_size();
} catch (std::exception const& e) {
PLOG_ERROR("exception: %s", e.what());
throw e;
} catch (...) {
PLOG_ERROR("unknown exception");
throw;
}
};
this->info.oei.encode = [](void* ptr, struct encoder_frame* frame, struct encoder_packet* packet,
bool* received_packet) {
try {
return reinterpret_cast<generic*>(ptr)->audio_encode(frame, packet, received_packet);
} catch (std::exception const& e) {
PLOG_ERROR("exception: %s", e.what());
throw e;
} catch (...) {
PLOG_ERROR("unknown exception");
throw;
}
};
}
// Finally store ourself as type data.
this->info.oei.type_data = this;
@@ -305,7 +322,6 @@ void encoder::generic_factory::get_defaults(obs_data_t* settings)
// Threading
obs_data_set_default_int(settings, P_MULTITHREADING_MODEL, -1);
obs_data_set_default_bool(settings, P_MULTITHREADING_FRAMEQUEUE, true);
obs_data_set_default_int(settings, P_MULTITHREADING_THREADCOUNT, 0);
// FFmpeg
@@ -490,12 +506,6 @@ void encoder::generic_factory::get_properties(obs_properties_t* props)
std::thread::hardware_concurrency() * 4, 1);
obs_property_set_long_description(p, TRANSLATE(DESC(P_MULTITHREADING_THREADCOUNT)));
};
{
auto p = obs_properties_add_bool(prs, P_MULTITHREADING_FRAMEQUEUE,
TRANSLATE(P_MULTITHREADING_FRAMEQUEUE));
obs_property_set_long_description(p, TRANSLATE(DESC(P_MULTITHREADING_FRAMEQUEUE)));
};
obs_properties_add_group(props, P_MULTITHREADING, TRANSLATE(P_MULTITHREADING), OBS_GROUP_NORMAL, prs);
};
{
@@ -574,13 +584,13 @@ encoder::generic::generic(obs_data_t* settings, obs_encoder_t* encoder) : self(e
this->context->bit_rate = obs_data_get_int(settings, P_RATECONTROL_BITRATE);
this->context->strict_std_compliance = obs_data_get_int(settings, P_FFMPEG_STANDARDCOMPLIANCE);
this->context->debug = 0;
/// Threading
{
int64_t tt = obs_data_get_int(settings, P_MULTITHREADING_MODEL);
if (tt == 0) {
this->context->thread_count = 1;
this->context->thread_type = 0;
this->lag_in_frames = 0;
} else if (tt == -1) {
if (this->codec->capabilities & AV_CODEC_CAP_SLICE_THREADS) {
this->context->thread_type = FF_THREAD_SLICE;
@@ -594,9 +604,11 @@ encoder::generic::generic(obs_data_t* settings, obs_encoder_t* encoder) : self(e
}
if (tt != 0) {
this->context->thread_count = obs_data_get_int(settings, P_MULTITHREADING_THREADCOUNT);
this->lag_in_frames = this->context->thread_count;
if (!(this->codec->capabilities & AV_CODEC_CAP_AUTO_THREADS)) {
if (this->context->thread_count == 0) {
this->context->thread_count = std::thread::hardware_concurrency();
this->lag_in_frames = std::thread::hardware_concurrency();
}
}
}
@@ -607,12 +619,42 @@ encoder::generic::generic(obs_data_t* settings, obs_encoder_t* encoder) : self(e
// FFmpeg Video Settings
auto encvideo = obs_encoder_video(this->self);
auto voi = video_output_get_info(encvideo);
/// Resolution and framerate.
// Resolution
this->context->width = voi->width;
this->context->height = voi->height;
this->swscale.set_source_size(this->context->width, this->context->height);
this->swscale.set_target_size(this->context->width, this->context->height);
// Color
this->context->colorspace = ffmpeg::tools::obs_videocolorspace_to_avcolorspace(voi->colorspace);
this->context->color_range = ffmpeg::tools::obs_videorangetype_to_avcolorrange(voi->range);
this->context->field_order = AV_FIELD_PROGRESSIVE;
this->swscale.set_source_color(this->context->color_range, this->context->colorspace);
this->swscale.set_target_color(this->context->color_range, this->context->colorspace);
// Pixel Format
{
// Due to unsupported color formats and ffmpeg not automatically converting formats from A to B,
// we have to detect the closest format that we can still use and initialize our swscale instance
// these formats. This has a massive cost attached unfortunately.
AVPixelFormat source = ffmpeg::tools::obs_videoformat_to_avpixelformat(voi->format);
AVPixelFormat target = AV_PIX_FMT_NONE;
int loss = 0;
target = avcodec_find_best_pix_fmt_of_list(this->codec->pix_fmts, source, false, &loss);
this->context->pix_fmt = target;
this->swscale.set_source_format(source);
this->swscale.set_target_format(this->context->pix_fmt);
}
// Framerate
this->context->time_base.num = voi->fps_num;
this->context->time_base.den = voi->fps_den;
this->context->ticks_per_frame = 1;
/// Group of Pictures
if (static_cast<keyframe_type>(obs_data_get_int(settings, P_RATECONTROL_KEYFRAME_TYPE))
== keyframe_type::Frames) {
@@ -621,11 +663,7 @@ encoder::generic::generic(obs_data_t* settings, obs_encoder_t* encoder) : self(e
double_t real_gop = obs_data_get_double(settings, P_RATECONTROL_KEYFRAME_INTERVAL ".Seconds");
this->context->gop_size = static_cast<int64_t>(real_gop * voi->fps_num / voi->fps_den);
}
/// Color, Profile
this->context->colorspace = ffmpeg::tools::obs_videocolorspace_to_avcolorspace(voi->colorspace);
this->context->color_range = ffmpeg::tools::obs_videorangetype_to_avcolorrange(voi->range);
this->context->pix_fmt = ffmpeg::tools::obs_videoformat_to_avpixelformat(voi->format);
this->context->field_order = AV_FIELD_PROGRESSIVE;
} else if (this->codec->type == AVMEDIA_TYPE_AUDIO) {
}
@@ -643,12 +681,6 @@ encoder::generic::generic(obs_data_t* settings, obs_encoder_t* encoder) : self(e
// Video/Audio exclusive setup part 2.
if (this->codec->type == AVMEDIA_TYPE_VIDEO) {
// Create Scaler
swscale.set_source_size(this->context->width, this->context->height);
swscale.set_target_size(this->context->width, this->context->height);
swscale.set_source_format(this->context->pix_fmt);
swscale.set_target_format(this->context->pix_fmt);
swscale.set_source_color(this->context->color_range, this->context->colorspace);
swscale.set_target_color(this->context->color_range, this->context->colorspace);
if (!swscale.initialize(SWS_FAST_BILINEAR)) {
PLOG_ERROR(
@@ -656,26 +688,14 @@ encoder::generic::generic(obs_data_t* settings, obs_encoder_t* encoder) : self(e
"range '%s'.",
ffmpeg::tools::get_pixel_format_name(this->context->pix_fmt),
ffmpeg::tools::get_color_space_name(this->context->colorspace),
swscale.is_source_full_range() ? "Full" : "Partial");
this->swscale.is_source_full_range() ? "Full" : "Partial");
throw std::runtime_error("failed to initialize swscaler.");
}
// Create Frame queue
frame_queue.set_pixel_format(this->context->pix_fmt);
frame_queue.set_resolution(this->context->width, this->context->height);
if (obs_data_get_bool(settings, P_MULTITHREADING_FRAMEQUEUE)) {
if (this->context->thread_count > 0) {
this->frame_queue.precache(this->context->thread_count);
} else {
if (this->context->thread_type != 0) {
this->frame_queue.set_pixel_format(this->context->pix_fmt);
this->frame_queue.set_resolution(this->context->width, this->context->height);
this->frame_queue.precache(std::thread::hardware_concurrency());
} else {
this->frame_queue.precache(1);
}
}
} else {
this->frame_queue.precache(1);
}
} else if (this->codec->type == AVMEDIA_TYPE_AUDIO) {
}
@@ -706,10 +726,93 @@ bool encoder::generic::update(obs_data_t* settings)
return false;
}
bool encoder::generic::encode(encoder_frame* frame, encoder_packet* packet, bool* received_packet)
bool encoder::generic::video_encode(encoder_frame* frame, encoder_packet* packet, bool* received_packet)
{
int res = 0;
bool have_pushed_frame = false;
bool have_pulled_frame = false;
auto loop_begin = std::chrono::high_resolution_clock::now();
auto loop_time = std::chrono::nanoseconds(
static_cast<int64_t>(1000.0 * this->context->time_base.den / this->context->time_base.num) * lag_in_frames);
if ((lag_in_frames - frame_count) > 0) {
loop_time = std::chrono::nanoseconds(1);
}
while (!have_pulled_frame || !have_pushed_frame) {
if (!have_pushed_frame) {
AVFrame* vframe = frame_queue.pop();
vframe->pts = frame->pts;
vframe->color_range = this->context->color_range;
vframe->colorspace = this->context->colorspace;
{
res = swscale.convert(reinterpret_cast<uint8_t**>(frame->data),
reinterpret_cast<int*>(frame->linesize), 0, this->context->height,
vframe->data, vframe->linesize);
if (res <= 0) {
PLOG_ERROR("Failed to convert frame: %s (%ld).",
ffmpeg::tools::get_error_description(res), res);
return false;
}
}
{
res = avcodec_send_frame(this->context, vframe);
if (res < 0) {
PLOG_ERROR("Failed to encode frame: %s (%ld).",
ffmpeg::tools::get_error_description(res), res);
return false;
}
}
frame_queue_used.push(vframe);
have_pushed_frame = true;
frame_count++;
}
if (!have_pulled_frame) {
res = avcodec_receive_packet(this->context, this->current_packet);
if (res < 0) {
if (res == AVERROR(EAGAIN)) {
*received_packet = false;
have_pulled_frame = true;
} else if (res == AVERROR(EOF)) {
*received_packet = false;
have_pulled_frame = true;
} else {
PLOG_ERROR("Failed to receive packet: %s (%ld).",
ffmpeg::tools::get_error_description(res), res);
return false;
}
} else {
AVFrame* uframe = frame_queue_used.pop_only();
if (uframe) {
frame_queue.push(uframe);
}
packet->type = OBS_ENCODER_VIDEO;
packet->pts = this->current_packet->pts;
packet->dts = this->current_packet->pts;
packet->data = this->current_packet->data;
packet->size = this->current_packet->size;
packet->keyframe = !!(this->current_packet->flags & AV_PKT_FLAG_KEY);
packet->drop_priority = 0;
*received_packet = true;
have_pulled_frame = true;
}
}
if ((std::chrono::high_resolution_clock::now() - loop_begin) >= loop_time) {
break;
} else {
std::this_thread::sleep_for(std::chrono::milliseconds(1));
}
}
return true;
}
void encoder::generic::get_audio_info(audio_convert_info* info) {}
@@ -718,6 +821,11 @@ size_t encoder::generic::get_frame_size()
return size_t();
}
bool encoder::generic::audio_encode(encoder_frame* frame, encoder_packet* packet, bool* received_packet)
{
return false;
}
void encoder::generic::get_video_info(video_scale_info* info) {}
bool encoder::generic::get_sei_data(uint8_t** sei_data, size_t* size)
@@ -735,7 +843,7 @@ bool encoder::generic::get_extra_data(uint8_t** extra_data, size_t* size)
return true;
}
bool encoder::generic::encode_texture(uint32_t handle, int64_t pts, uint64_t lock_key, uint64_t* next_key,
bool encoder::generic::video_encode_texture(uint32_t handle, int64_t pts, uint64_t lock_key, uint64_t* next_key,
encoder_packet* packet, bool* received_packet)
{
return false;
+8 -4
View File
@@ -64,6 +64,9 @@ namespace encoder {
ffmpeg::swscale swscale;
AVPacket* current_packet = nullptr;
int64_t lag_in_frames;
int64_t frame_count;
public:
generic(obs_data_t* settings, obs_encoder_t* encoder);
virtual ~generic();
@@ -74,13 +77,13 @@ namespace encoder {
bool update(obs_data_t* settings);
bool encode(struct encoder_frame* frame, struct encoder_packet* packet, bool* received_packet);
// Audio only
void get_audio_info(struct audio_convert_info* info);
size_t get_frame_size();
bool audio_encode(struct encoder_frame* frame, struct encoder_packet* packet, bool* received_packet);
// Video only
void get_video_info(struct video_scale_info* info);
@@ -88,8 +91,9 @@ namespace encoder {
bool get_extra_data(uint8_t** extra_data, size_t* size);
// GPU Video only
bool encode_texture(uint32_t handle, int64_t pts, uint64_t lock_key, uint64_t* next_key,
bool video_encode(struct encoder_frame* frame, struct encoder_packet* packet, bool* received_packet);
bool video_encode_texture(uint32_t handle, int64_t pts, uint64_t lock_key, uint64_t* next_key,
struct encoder_packet* packet, bool* received_packet);
};
} // namespace encoder