vcmi/client/media/CVideoHandler.cpp

/*
 * CVideoHandler.cpp, part of VCMI engine
 *
 * Authors: listed in file AUTHORS in main folder
 *
 * License: GNU General Public License v2.0 or later
 * Full text of license available in license.txt file, in main folder
 *
 */
#include "StdInc.h"
#include "CVideoHandler.h"

#ifndef DISABLE_VIDEO

#include "ISoundPlayer.h"

#include "../CGameInfo.h"
#include "../CMT.h"
#include "../CPlayerInterface.h"
#include "../eventsSDL/InputHandler.h"
#include "../gui/CGuiHandler.h"
#include "../gui/FramerateManager.h"
#include "../render/Canvas.h"
#include "../renderSDL/SDL_Extensions.h"

#include "../../lib/filesystem/CInputStream.h"
#include "../../lib/filesystem/Filesystem.h"

#include <SDL_render.h>

extern "C" {
#include <libavformat/avformat.h>
#include <libavcodec/avcodec.h>
#include <libavutil/imgutils.h>
#include <libswscale/swscale.h>
}

// Define a set of functions to read data
static int lodRead(void * opaque, uint8_t * buf, int size)
{
	auto * data = static_cast<CInputStream *>(opaque);
	int bytes = static_cast<int>(data->read(buf, size));
	if(bytes == 0)
		return AVERROR_EOF;

	return bytes;
}

static si64 lodSeek(void * opaque, si64 pos, int whence)
{
	auto * data = static_cast<CInputStream *>(opaque);

	if(whence & AVSEEK_SIZE)
		return data->getSize();

	return data->seek(pos);
}

[[noreturn]] static void throwFFmpegError(int errorCode)
{
	std::array<char, AV_ERROR_MAX_STRING_SIZE> errorMessage{};
	av_strerror(errorCode, errorMessage.data(), errorMessage.size());

	throw std::runtime_error(errorMessage.data());
}

static std::unique_ptr<CInputStream> findVideoData(const VideoPath & videoToOpen)
{
	if(CResourceHandler::get()->existsResource(videoToOpen))
		return CResourceHandler::get()->load(videoToOpen);

	auto highQualityVideoToOpenWithDir = videoToOpen.addPrefix("VIDEO/");
	auto lowQualityVideo = videoToOpen.toType<EResType::VIDEO_LOW_QUALITY>();
	auto lowQualityVideoWithDir = highQualityVideoToOpenWithDir.toType<EResType::VIDEO_LOW_QUALITY>();

	if(CResourceHandler::get()->existsResource(highQualityVideoToOpenWithDir))
		return CResourceHandler::get()->load(highQualityVideoToOpenWithDir);

	if(CResourceHandler::get()->existsResource(lowQualityVideo))
		return CResourceHandler::get()->load(lowQualityVideo);

	return CResourceHandler::get()->load(lowQualityVideoWithDir);
}

void FFMpegStream::openInput(const VideoPath & videoToOpen)
{
	input = findVideoData(videoToOpen);
}

void FFMpegStream::openContext()
{
	static const int BUFFER_SIZE = 4096;
	input->seek(0);

	auto * buffer = static_cast<unsigned char *>(av_malloc(BUFFER_SIZE)); // will be freed by ffmpeg
	context = avio_alloc_context(buffer, BUFFER_SIZE, 0, input.get(), lodRead, nullptr, lodSeek);

	formatContext = avformat_alloc_context();
	formatContext->pb = context;
	// filename is not needed - file was already open and stored in this->data;
	int avfopen = avformat_open_input(&formatContext, "dummyFilename", nullptr, nullptr);

	if(avfopen != 0)
		throwFFmpegError(avfopen);

	// Retrieve stream information
	int findStreamInfo = avformat_find_stream_info(formatContext, nullptr);

	if(avfopen < 0)
		throwFFmpegError(findStreamInfo);
}

void FFMpegStream::openCodec(int desiredStreamIndex)
{
	streamIndex = desiredStreamIndex;

	// Find the decoder for the stream
	codec = avcodec_find_decoder(formatContext->streams[streamIndex]->codecpar->codec_id);

	if(codec == nullptr)
		throw std::runtime_error("Unsupported codec");

	codecContext = avcodec_alloc_context3(codec);
	if(codecContext == nullptr)
		throw std::runtime_error("Failed to create codec context");

	// Get a pointer to the codec context for the video stream
	int ret = avcodec_parameters_to_context(codecContext, formatContext->streams[streamIndex]->codecpar);
	if(ret < 0)
	{
		//We cannot get codec from parameters
		avcodec_free_context(&codecContext);
		throwFFmpegError(ret);
	}

	// Open codec
	ret = avcodec_open2(codecContext, codec, nullptr);
	if(ret < 0)
	{
		// Could not open codec
		codec = nullptr;
		throwFFmpegError(ret);
	}

	// Allocate video frame
	frame = av_frame_alloc();
}

const AVCodecParameters * FFMpegStream::getCodecParameters()
{
	return formatContext->streams[streamIndex]->codecpar;
}

const AVCodecContext * FFMpegStream::getCodecContext()
{
	return codecContext;
}

const AVFrame * FFMpegStream::getCurrentFrame()
{
	return frame;
}

void CVideoInstance::openVideo()
{
	openContext();
	openCodec(findVideoStream());
}

void CVideoInstance::prepareOutput(bool scaleToScreenSize, bool useTextureOutput)
{
	//setup scaling
	if(scaleToScreenSize)
	{
		dimensions.x = screen->w;
		dimensions.y = screen->h;
	}
	else
	{
		dimensions.x  = getCodecContext()->width;
		dimensions.y = getCodecContext()->height;
	}

	// Allocate a place to put our YUV image on that screen
	if (useTextureOutput)
	{
		std::array potentialFormats = {
			AV_PIX_FMT_YUV420P, // -> SDL_PIXELFORMAT_IYUV - most of H3 videos use YUV format, so it is preferred to save some space & conversion time
			AV_PIX_FMT_RGB32,   // -> SDL_PIXELFORMAT_ARGB8888 - some .smk videos actually use palette, so RGB > YUV. This is also our screen texture format
			AV_PIX_FMT_NONE
		};

		auto preferredFormat = avcodec_find_best_pix_fmt_of_list(potentialFormats.data(), getCodecContext()->pix_fmt, false, nullptr);

		if (preferredFormat == AV_PIX_FMT_YUV420P)
			textureYUV = SDL_CreateTexture( mainRenderer, SDL_PIXELFORMAT_IYUV, SDL_TEXTUREACCESS_STREAMING, dimensions.x, dimensions.y);
		else
			textureRGB = SDL_CreateTexture( mainRenderer, SDL_PIXELFORMAT_ARGB8888, SDL_TEXTUREACCESS_STREAMING, dimensions.x, dimensions.y);
		sws = sws_getContext(getCodecContext()->width, getCodecContext()->height, getCodecContext()->pix_fmt,
							dimensions.x, dimensions.y, preferredFormat,
							 SWS_BICUBIC, nullptr, nullptr, nullptr);
	}
	else
	{
		surface = CSDL_Ext::newSurface(dimensions.x, dimensions.y);
		sws = sws_getContext(getCodecContext()->width, getCodecContext()->height, getCodecContext()->pix_fmt,
							 dimensions.x, dimensions.y, AV_PIX_FMT_RGB32,
							 SWS_BICUBIC, nullptr, nullptr, nullptr);
	}

	if (sws == nullptr)
		throw std::runtime_error("Failed to create sws");
}

void FFMpegStream::decodeNextFrame()
{
	AVPacket packet;

	for(;;)
	{
		int rc = avcodec_receive_frame(codecContext, frame);
		if(rc == AVERROR(EAGAIN))
			break;

		if(rc < 0)
			throwFFmpegError(rc);

		return;
	}

	for(;;)
	{
		int ret = av_read_frame(formatContext, &packet);
		if(ret < 0)
		{
			if(ret == AVERROR_EOF)
			{
				av_packet_unref(&packet);
				av_frame_free(&frame);
				frame = nullptr;
				return;
			}
			throwFFmpegError(ret);
		}

		// Is this a packet from the video stream?
		if(packet.stream_index == streamIndex)
		{
			// Decode video frame
			int rc = avcodec_send_packet(codecContext, &packet);
			if(rc < 0 && rc != AVERROR(EAGAIN))
				throwFFmpegError(rc);

			rc = avcodec_receive_frame(codecContext, frame);
			if(rc == AVERROR(EAGAIN))
			{
				av_packet_unref(&packet);
				continue;
			}
			if(rc < 0)
				throwFFmpegError(rc);

			av_packet_unref(&packet);
			return;
		}
		av_packet_unref(&packet);
	}
}

bool CVideoInstance::loadNextFrame()
{
	decodeNextFrame();
	const AVFrame * frame = getCurrentFrame();

	if(!frame)
		return false;

	uint8_t * data[4] = {};
	int linesize[4] = {};

	if(textureYUV)
	{
		av_image_alloc(data, linesize, dimensions.x, dimensions.y, AV_PIX_FMT_YUV420P, 1);
		sws_scale(sws, frame->data, frame->linesize, 0, getCodecContext()->height, data, linesize);
		SDL_UpdateYUVTexture(textureYUV, nullptr, data[0], linesize[0], data[1], linesize[1], data[2], linesize[2]);
		av_freep(&data[0]);
	}
	if(textureRGB)
	{
		av_image_alloc(data, linesize, dimensions.x, dimensions.y, AV_PIX_FMT_RGB32, 1);
		sws_scale(sws, frame->data, frame->linesize, 0, getCodecContext()->height, data, linesize);
		SDL_UpdateTexture(textureRGB, nullptr, data[0], linesize[0]);
		av_freep(&data[0]);
	}
	if(surface)
	{
		// Avoid buffer overflow caused by sws_scale():
		// http://trac.ffmpeg.org/ticket/9254

		size_t pic_bytes = surface->pitch * surface->h;
		size_t ffmped_pad = 1024; /* a few bytes of overflow will go here */
		void * for_sws = av_malloc(pic_bytes + ffmped_pad);
		data[0] = (ui8 *)for_sws;
		linesize[0] = surface->pitch;

		sws_scale(sws, frame->data, frame->linesize, 0, getCodecContext()->height, data, linesize);
		memcpy(surface->pixels, for_sws, pic_bytes);
		av_free(for_sws);
	}
	return true;
}

bool CVideoInstance::videoEnded()
{
	return getCurrentFrame() == nullptr;
}

CVideoInstance::~CVideoInstance()
{
	sws_freeContext(sws);
	SDL_DestroyTexture(textureYUV);
	SDL_DestroyTexture(textureRGB);
	SDL_FreeSurface(surface);
}

FFMpegStream::~FFMpegStream()
{
	// state.videoStream.codec???
	// state.audioStream.codec???
	av_frame_free(&frame);

	avcodec_close(codecContext);
	avcodec_free_context(&codecContext);

	avcodec_close(codecContext);
	avcodec_free_context(&codecContext);

	avformat_close_input(&formatContext);
	av_free(context);
}

Point CVideoInstance::size()
{
	if(!getCurrentFrame())
		throw std::runtime_error("Invalid video frame!");

	return Point(getCurrentFrame()->width, getCurrentFrame()->height);
}

void CVideoInstance::show(const Point & position, Canvas & canvas)
{
	if(sws == nullptr)
		throw std::runtime_error("No video to show!");

	CSDL_Ext::blitSurface(surface, canvas.getInternalSurface(), position);
}

double FFMpegStream::getCurrentFrameEndTime()
{
#if(LIBAVUTIL_VERSION_MAJOR < 58)
	auto packet_duration = frame->pkt_duration;
#else
	auto packet_duration = frame->duration;
#endif
	return (frame->pts + packet_duration) * av_q2d(formatContext->streams[streamIndex]->time_base);
}

double FFMpegStream::getCurrentFrameDuration()
{
#if(LIBAVUTIL_VERSION_MAJOR < 58)
	auto packet_duration = frame->pkt_duration;
#else
	auto packet_duration = frame->duration;
#endif
	return (packet_duration) * av_q2d(formatContext->streams[streamIndex]->time_base);
}

void CVideoInstance::tick(uint32_t msPassed)
{
	if(sws == nullptr)
		throw std::runtime_error("No video to show!");

	if(videoEnded())
		throw std::runtime_error("Video already ended!");

	frameTime += msPassed / 1000.0;

	if(frameTime >= getCurrentFrameEndTime())
		loadNextFrame();
}

struct FFMpegFormatDescription
{
	uint8_t sampleSizeBytes;
	uint8_t wavFormatID;
	bool isPlanar;
};

static FFMpegFormatDescription getAudioFormatProperties(int audioFormat)
{
	switch (audioFormat)
	{
		case AV_SAMPLE_FMT_U8:   return { 1, 1, false};
		case AV_SAMPLE_FMT_U8P:  return { 1, 1, true};
		case AV_SAMPLE_FMT_S16:  return { 2, 1, false};
		case AV_SAMPLE_FMT_S16P: return { 2, 1, true};
		case AV_SAMPLE_FMT_S32:  return { 4, 1, false};
		case AV_SAMPLE_FMT_S32P: return { 4, 1, true};
		case AV_SAMPLE_FMT_S64:  return { 8, 1, false};
		case AV_SAMPLE_FMT_S64P: return { 8, 1, true};
		case AV_SAMPLE_FMT_FLT:  return { 4, 3, false};
		case AV_SAMPLE_FMT_FLTP: return { 4, 3, true};
		case AV_SAMPLE_FMT_DBL:  return { 8, 3, false};
		case AV_SAMPLE_FMT_DBLP: return { 8, 3, true};
	}
	throw std::runtime_error("Invalid audio format");
}

int FFMpegStream::findAudioStream()
{
	for(int i = 0; i < formatContext->nb_streams; i++)
		if(formatContext->streams[i]->codecpar->codec_type == AVMEDIA_TYPE_AUDIO)
			return i;

	return -1;
}

int FFMpegStream::findVideoStream()
{
	for(int i = 0; i < formatContext->nb_streams; i++)
		if(formatContext->streams[i]->codecpar->codec_type == AVMEDIA_TYPE_VIDEO)
			return i;

	return -1;
}

std::pair<std::unique_ptr<ui8 []>, si64> CAudioInstance::extractAudio(const VideoPath & videoToOpen)
{
	openInput(videoToOpen);
	openContext();
	openCodec(findAudioStream());

	const auto * codecpar = getCodecParameters();

	std::vector<ui8> samples;

	auto formatProperties = getAudioFormatProperties(codecpar->format);
#if(LIBAVUTIL_VERSION_MAJOR < 58)
	int numChannels = codecpar->channels;
#else
	int numChannels = codecpar->ch_layout.nb_channels;
#endif

	samples.reserve(44100 * 5); // arbitrary 5-second buffer

	for (;;)
	{
		decodeNextFrame();
		const AVFrame * frame = getCurrentFrame();

		if (!frame)
			break;

		int samplesToRead = frame->nb_samples * numChannels;
		int bytesToRead = samplesToRead * formatProperties.sampleSizeBytes;

		if (formatProperties.isPlanar && numChannels > 1)
		{
			// Workaround for lack of resampler
			// Currently, ffmpeg on conan systems is built without sws resampler
			// Because of that, and because wav format does not supports 'planar' formats from ffmpeg
			// we need to de-planarize it and convert to "normal" (non-planar / interleaved) steram
			samples.reserve(samples.size() + bytesToRead);
			for (int sm = 0; sm < frame->nb_samples; ++sm)
				for (int ch = 0; ch < numChannels; ++ch)
					samples.insert(samples.end(), frame->data[ch] + sm * formatProperties.sampleSizeBytes, frame->data[ch] + (sm+1) * formatProperties.sampleSizeBytes );
		}
		else
		{
			samples.insert(samples.end(), frame->data[0], frame->data[0] + bytesToRead);
		}
	}

	typedef struct WAV_HEADER {
		ui8 RIFF[4] = {'R', 'I', 'F', 'F'};
		ui32 ChunkSize;
		ui8 WAVE[4] = {'W', 'A', 'V', 'E'};
		ui8 fmt[4] = {'f', 'm', 't', ' '};
		ui32 Subchunk1Size = 16;
		ui16 AudioFormat = 1;
		ui16 NumOfChan = 2;
		ui32 SamplesPerSec = 22050;
		ui32 bytesPerSec = 22050 * 2;
		ui16 blockAlign = 2;
		ui16 bitsPerSample = 32;
		ui8 Subchunk2ID[4] = {'d', 'a', 't', 'a'};
		ui32 Subchunk2Size;
	} wav_hdr;

	wav_hdr wav;
	wav.ChunkSize = samples.size() + sizeof(wav_hdr) - 8;
	wav.AudioFormat = formatProperties.wavFormatID; // 1 = PCM, 3 = IEEE float
	wav.NumOfChan = numChannels;
	wav.SamplesPerSec = codecpar->sample_rate;
	wav.bytesPerSec = codecpar->sample_rate * formatProperties.sampleSizeBytes;
	wav.bitsPerSample = formatProperties.sampleSizeBytes * 8;
	wav.Subchunk2Size = samples.size() + sizeof(wav_hdr) - 44;
	auto wavPtr = reinterpret_cast<ui8*>(&wav);

	auto dat = std::make_pair(std::make_unique<ui8[]>(samples.size() + sizeof(wav_hdr)), samples.size() + sizeof(wav_hdr));
	std::copy(wavPtr, wavPtr + sizeof(wav_hdr), dat.first.get());
	std::copy(samples.begin(), samples.end(), dat.first.get() + sizeof(wav_hdr));

	return dat;
	//CCS->soundh->playSound(dat);
}

bool CVideoPlayer::openAndPlayVideoImpl(const VideoPath & name, const Point & position, bool useOverlay, bool scale, bool stopOnKey)
{
	CVideoInstance instance;
	CAudioInstance audio;

	auto extractedAudio = audio.extractAudio(name);
	int audioHandle = CCS->soundh->playSound(extractedAudio);

	instance.openInput(name);
	instance.openVideo();
	instance.prepareOutput(scale, useOverlay);

	auto lastTimePoint = boost::chrono::steady_clock::now();

	while(instance.loadNextFrame())
	{
		if(stopOnKey)
		{
			GH.input().fetchEvents();
			if(GH.input().ignoreEventsUntilInput())
			{
				CCS->soundh->stopSound(audioHandle);
				return false;
			}
		}

		SDL_Rect rect;
		rect.x = position.x;
		rect.y = position.y;
		rect.w = instance.dimensions.x;
		rect.h = instance.dimensions.y;

		if(useOverlay)
			SDL_RenderFillRect(mainRenderer, &rect);
		else
			SDL_RenderClear(mainRenderer);

		if(instance.textureYUV)
			SDL_RenderCopy(mainRenderer, instance.textureYUV, nullptr, &rect);
		else
			SDL_RenderCopy(mainRenderer, instance.textureRGB, nullptr, &rect);

		SDL_RenderPresent(mainRenderer);

		// Framerate delay
		double targetFrameTimeSeconds = instance.getCurrentFrameDuration();
		auto targetFrameTime = boost::chrono::milliseconds(static_cast<int>(1000 * (targetFrameTimeSeconds)));

		auto timePointAfterPresent = boost::chrono::steady_clock::now();
		auto timeSpentBusy = boost::chrono::duration_cast<boost::chrono::milliseconds>(timePointAfterPresent - lastTimePoint);

		logGlobal->info("Sleeping for %d", (targetFrameTime - timeSpentBusy).count());
		if(targetFrameTime > timeSpentBusy)
			boost::this_thread::sleep_for(targetFrameTime - timeSpentBusy);

		lastTimePoint = boost::chrono::steady_clock::now();
	}
	return true;
}

bool CVideoPlayer::playIntroVideo(const VideoPath & name)
{
	return openAndPlayVideoImpl(name, Point(0, 0), true, true, true);
}

void CVideoPlayer::playSpellbookAnimation(const VideoPath & name, const Point & position)
{
	openAndPlayVideoImpl(name, position, false, false, false);
}

std::unique_ptr<IVideoInstance> CVideoPlayer::open(const VideoPath & name, bool scaleToScreen)
{
	auto result = std::make_unique<CVideoInstance>();

	result->openInput(name);
	result->openVideo();
	result->prepareOutput(scaleToScreen, false);

	return result;
}

std::pair<std::unique_ptr<ui8[]>, si64> CVideoPlayer::getAudio(const VideoPath & videoToOpen)
{
	CAudioInstance audio;
	return audio.extractAudio(videoToOpen);
}

#endif