scar-chat7/client/media/video_decoder.cpp

#include "video_decoder.h"
#include <iostream>
#include <cstring>

extern "C" {
#include <libavutil/imgutils.h>
}

namespace scar {

VideoDecoder::VideoDecoder()
    : codec_ctx_(nullptr),
      frame_(nullptr),
      sws_ctx_(nullptr),
      width_(0),
      height_(0),
      initialized_(false) {
}

VideoDecoder::~VideoDecoder() {
    cleanup();
}

bool VideoDecoder::initialize() {
    cleanup(); // Clean up any previous state
    
    // Find H.264 decoder
    const AVCodec* codec = avcodec_find_decoder(AV_CODEC_ID_H264);
    if (!codec) {
        std::cerr << "H.264 decoder not found" << std::endl;
        return false;
    }
    
    std::cout << "Using H.264 decoder" << std::endl;
    
    // Create codec context
    codec_ctx_ = avcodec_alloc_context3(codec);
    if (!codec_ctx_) {
        std::cerr << "Failed to allocate decoder context" << std::endl;
        return false;
    }
    
    // Open codec
    if (avcodec_open2(codec_ctx_, codec, nullptr) < 0) {
        std::cerr << "Failed to open decoder" << std::endl;
        cleanup();
        return false;
    }
    
    // Allocate frame
    frame_ = av_frame_alloc();
    if (!frame_) {
        std::cerr << "Failed to allocate frame" << std::endl;
        cleanup();
        return false;
    }
    
    initialized_ = true;
    
    std::cout << "Video decoder initialized" << std::endl;
    
    return true;
}

std::vector<uint8_t> VideoDecoder::decode(const std::vector<uint8_t>& encoded_data, 
                                           int& out_width, int& out_height, int& out_linesize) {
    std::lock_guard<std::mutex> lock(decode_mutex_);
    
    if (!initialized_) {
        std::cerr << "Decoder not initialized" << std::endl;
        return {};
    }
    
    if (encoded_data.empty()) {
        std::cerr << "Empty encoded data" << std::endl;
        return {};
    }
    
    static int decode_count = 0;
    decode_count++;
    
    // Create packet and copy data (FFmpeg will manage the memory)
    AVPacket* pkt = av_packet_alloc();
    if (!pkt) {
        std::cerr << "Failed to allocate packet" << std::endl;
        return {};
    }
    
    // Allocate buffer and copy data - FFmpeg will own and free this memory
    if (av_new_packet(pkt, encoded_data.size()) < 0) {
        std::cerr << "Failed to allocate packet buffer" << std::endl;
        av_packet_free(&pkt);
        return {};
    }
    
    memcpy(pkt->data, encoded_data.data(), encoded_data.size());
    
    // Send packet to decoder
    int ret = avcodec_send_packet(codec_ctx_, pkt);
    if (ret < 0) {
        char errbuf[256];
        av_strerror(ret, errbuf, sizeof(errbuf));
        std::cerr << "Error sending packet to decoder: " << errbuf << std::endl;
        av_packet_free(&pkt);
        return {};
    }
    
    // Receive decoded frame
    ret = avcodec_receive_frame(codec_ctx_, frame_);
    
    // Free packet immediately after sending
    av_packet_free(&pkt);
    
    if (ret == AVERROR(EAGAIN)) {
        // Decoder needs more data
        return {};
    } else if (ret < 0) {
        char errbuf[256];
        av_strerror(ret, errbuf, sizeof(errbuf));
        std::cerr << "Error receiving frame from decoder: " << errbuf << std::endl;
        return {};
    }
    
    std::cout << "Decode #" << decode_count << ": frame " << frame_->width << "x" << frame_->height << std::endl;
    
    // Update dimensions if this is the first frame or they changed
    if (width_ != frame_->width || height_ != frame_->height) {
        width_ = frame_->width;
        height_ = frame_->height;
        
        // Recreate swscale context for new dimensions
        if (sws_ctx_) {
            sws_freeContext(sws_ctx_);
        }
        
        sws_ctx_ = sws_getContext(
            width_, height_, AV_PIX_FMT_YUV420P,
            width_, height_, AV_PIX_FMT_RGB24,
            SWS_FAST_BILINEAR, nullptr, nullptr, nullptr
        );
        
        if (!sws_ctx_) {
            std::cerr << "Failed to create swscale context" << std::endl;
            return {};
        }
        
        std::cout << "Decoder dimensions: " << width_ << "x" << height_ << std::endl;
    }
    
    // CRITICAL: Use FFmpeg's default alignment (32 bytes) for proper SIMD operations
    // Alignment=1 was causing heap corruption because sws_scale expects aligned buffers
    int buffer_size = av_image_get_buffer_size(AV_PIX_FMT_RGB24, width_, height_, 32);
    if (buffer_size < 0) {
        std::cerr << "Failed to get buffer size" << std::endl;
        av_frame_unref(frame_);
        return {};
    }
    
    std::cout << "Allocating aligned buffer: " << buffer_size << " bytes" << std::endl;
    
    // Allocate buffer with extra space for alignment
    std::vector<uint8_t> rgb_buffer(buffer_size);
    
    // Use av_image_fill_arrays with same alignment as buffer allocation
    uint8_t* dest[4] = {nullptr};
    int dest_linesize[4] = {0};
    
    int fill_ret = av_image_fill_arrays(dest, dest_linesize, rgb_buffer.data(), 
                         AV_PIX_FMT_RGB24, width_, height_, 32);
    
    if (fill_ret < 0 || fill_ret > buffer_size) {
        std::cerr << "av_image_fill_arrays failed or returned size > buffer_size: "
                  << fill_ret << " vs " << buffer_size << std::endl;
        av_frame_unref(frame_);
        return {};
    }
    
    std::cout << "Allocated buffer: " << buffer_size << " bytes, "
              << "linesize[0]=" << dest_linesize[0] 
              << ", width=" << width_ << ", height=" << height_ << std::endl;
    
    std::cout << "Calling sws_scale..." << std::endl;
    int converted_height = sws_scale(
        sws_ctx_,
        frame_->data, frame_->linesize, 0, height_,
        dest, dest_linesize
    );
    std::cout << "sws_scale returned: " << converted_height << std::endl;
    
    if (converted_height != height_) {
        std::cerr << "sws_scale failed: converted " << converted_height << " lines, expected " << height_ << std::endl;
        av_frame_unref(frame_);
        return {};
    }
    
    // Verify sws_scale didn't write beyond our buffer
    size_t bytes_written = dest_linesize[0] * converted_height;
    if (bytes_written > rgb_buffer.size()) {
        std::cerr << "BUFFER OVERFLOW: sws_scale wrote " << bytes_written 
                  << " bytes but buffer is only " << rgb_buffer.size() << " bytes!" << std::endl;
        av_frame_unref(frame_);
        return {};
    }
    
    // Clean up frame
    av_frame_unref(frame_);
    
    // Set output dimensions
    out_width = width_;
    out_height = height_;
    out_linesize = dest_linesize[0];  // Use FFmpeg's actual linesize
    
    std::cout << "About to return buffer, size=" << rgb_buffer.size() 
              << ", linesize=" << out_linesize << std::endl;
    
    // Return by value - compiler will use RVO (Return Value Optimization)
    return rgb_buffer;
}

void VideoDecoder::cleanup() {
    if (sws_ctx_) {
        sws_freeContext(sws_ctx_);
        sws_ctx_ = nullptr;
    }
    
    if (frame_) {
        av_frame_free(&frame_);
        frame_ = nullptr;
    }
    
    if (codec_ctx_) {
        avcodec_free_context(&codec_ctx_);
        codec_ctx_ = nullptr;
    }
    
    initialized_ = false;
    width_ = 0;
    height_ = 0;
}

} // namespace scar