Exemple #1
0
int main(int argc, char** argv)
{
    DecodeStreamInput *input;
    int32_t fd = -1;
    int32_t i = 0;
    int32_t ioctlRet = -1;
    char opt;
    YamiMediaCodec::CalcFps calcFps;

    renderMode = 3; // set default render mode to 3

    yamiTraceInit();
    XInitThreads();

    if (!process_cmdline(argc, argv))
        return -1;

    if (!inputFileName) {
        ERROR("no input media file specified\n");
        return -1;
    }
    INFO("input file: %s, renderMode: %d", inputFileName, renderMode);

    if (!dumpOutputDir)
        dumpOutputDir = strdup ("./");

#if !__ENABLE_V4L2_GLX__
    switch (renderMode) {
    case 0:
        memoryType = VIDEO_DATA_MEMORY_TYPE_RAW_COPY;
    break;
    case 3:
        memoryType = VIDEO_DATA_MEMORY_TYPE_DRM_NAME;
    break;
    case 4:
        memoryType = VIDEO_DATA_MEMORY_TYPE_DMA_BUF;
    break;
    default:
        ASSERT(0 && "unsupported render mode, -m [0,3,4] are supported");
    break;
    }
#endif

    input = DecodeStreamInput::create(inputFileName);
    if (input==NULL) {
        ERROR("fail to init input stream\n");
        return -1;
    }

    renderFrameCount = 0;
    calcFps.setAnchor();
    // open device
    fd = YamiV4L2_Open("decoder", 0);
    ASSERT(fd!=-1);

    x11Display = XOpenDisplay(NULL);
    ASSERT(x11Display);
#if __ENABLE_V4L2_GLX__
    ioctlRet = YamiV4L2_SetXDisplay(fd, x11Display);
#endif
    // set output frame memory type
    YamiV4L2_FrameMemoryType(fd, memoryType);

    // query hw capability
    struct v4l2_capability caps;
    memset(&caps, 0, sizeof(caps));
    caps.capabilities = V4L2_CAP_VIDEO_CAPTURE_MPLANE | V4L2_CAP_VIDEO_OUTPUT_MPLANE | V4L2_CAP_STREAMING;
    ioctlRet = YamiV4L2_Ioctl(fd, VIDIOC_QUERYCAP, &caps);
    ASSERT(ioctlRet != -1);

    // set input/output data format
    uint32_t codecFormat = 0;
    const char* mimeType = input->getMimeType();
    if (!strcmp(mimeType, "video/h264"))
        codecFormat = V4L2_PIX_FMT_H264;
    else if (!strcmp(mimeType, "video/x-vnd.on2.vp8"))
        codecFormat = V4L2_PIX_FMT_VP8;
    else if (!strcmp(mimeType, "image/jpeg"))
        codecFormat = V4L2_PIX_FMT_MJPEG;
    else {
        ERROR("unsupported mimetype");
        return -1;
    }

    struct v4l2_format format;
    memset(&format, 0, sizeof(format));
    format.type = V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE;
    format.fmt.pix_mp.pixelformat = codecFormat;
    format.fmt.pix_mp.num_planes = 1;
    format.fmt.pix_mp.plane_fmt[0].sizeimage = k_maxInputBufferSize;
    ioctlRet = YamiV4L2_Ioctl(fd, VIDIOC_S_FMT, &format);
    ASSERT(ioctlRet != -1);

    // set preferred output format
    memset(&format, 0, sizeof(format));
    format.type = V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE;
    format.fmt.pix_mp.pixelformat = V4L2_PIX_FMT_NV12M;
    ioctlRet = YamiV4L2_Ioctl(fd, VIDIOC_S_FMT, &format);
    ASSERT(ioctlRet != -1);

    // start input port
    __u32 type = V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE;
    ioctlRet = YamiV4L2_Ioctl(fd, VIDIOC_STREAMON, &type);
    ASSERT(ioctlRet != -1);

    // start output port
    type = V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE;
    ioctlRet = YamiV4L2_Ioctl(fd, VIDIOC_STREAMON, &type);
    ASSERT(ioctlRet != -1);

    // setup input buffers
    struct v4l2_requestbuffers reqbufs;
    memset(&reqbufs, 0, sizeof(reqbufs));
    reqbufs.type = V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE;
    reqbufs.memory = V4L2_MEMORY_MMAP;
    reqbufs.count = 2;
    ioctlRet = YamiV4L2_Ioctl(fd, VIDIOC_REQBUFS, &reqbufs);
    ASSERT(ioctlRet != -1);
    ASSERT(reqbufs.count>0);
    inputQueueCapacity = reqbufs.count;
    inputFrames.resize(inputQueueCapacity);

    for (i=0; i<inputQueueCapacity; i++) {
        struct v4l2_plane planes[k_inputPlaneCount];
        struct v4l2_buffer buffer;
        memset(&buffer, 0, sizeof(buffer));
        memset(planes, 0, sizeof(planes));
        buffer.index = i;
        buffer.type = V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE;
        buffer.memory = V4L2_MEMORY_MMAP;
        buffer.m.planes = planes;
        buffer.length = k_inputPlaneCount;
        ioctlRet = YamiV4L2_Ioctl(fd, VIDIOC_QUERYBUF, &buffer);
        ASSERT(ioctlRet != -1);

        // length and mem_offset should be filled by VIDIOC_QUERYBUF above
        void* address = YamiV4L2_Mmap(NULL,
                                      buffer.m.planes[0].length,
                                      PROT_READ | PROT_WRITE,
                                      MAP_SHARED, fd,
                                      buffer.m.planes[0].m.mem_offset);
        ASSERT(address);
        inputFrames[i] = static_cast<uint8_t*>(address);
        DEBUG("inputFrames[%d] = %p", i, inputFrames[i]);
    }

    // feed input frames first
    for (i=0; i<inputQueueCapacity; i++) {
        if (!feedOneInputFrame(input, fd, i)) {
            break;
        }
    }

    // query video resolution
    memset(&format, 0, sizeof(format));
    format.type = V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE;
    while (1) {
        if (YamiV4L2_Ioctl(fd, VIDIOC_G_FMT, &format) != 0) {
            if (errno != EINVAL) {
                // EINVAL means we haven't seen sufficient stream to decode the format.
                INFO("ioctl() failed: VIDIOC_G_FMT, haven't get video resolution during start yet, waiting");
            }
        } else {
            break;
        }
        usleep(50);
    }
    outputPlaneCount = format.fmt.pix_mp.num_planes;
    ASSERT(outputPlaneCount == 2);
    videoWidth = format.fmt.pix_mp.width;
    videoHeight = format.fmt.pix_mp.height;
    ASSERT(videoWidth && videoHeight);

    // setup output buffers
    memset(&reqbufs, 0, sizeof(reqbufs));
    reqbufs.type = V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE;
    reqbufs.memory = V4L2_MEMORY_MMAP;
    reqbufs.count = outputPlaneCount;
    ioctlRet = YamiV4L2_Ioctl(fd, VIDIOC_REQBUFS, &reqbufs);
    ASSERT(ioctlRet != -1);
    ASSERT(reqbufs.count>0);
    outputQueueCapacity = reqbufs.count;

    x11Window = XCreateSimpleWindow(x11Display, DefaultRootWindow(x11Display)
        , 0, 0, videoWidth, videoHeight, 0, 0
        , WhitePixel(x11Display, 0));
    XMapWindow(x11Display, x11Window);
#if __ENABLE_V4L2_GLX__
    pixmaps.resize(outputQueueCapacity);
    glxPixmaps.resize(outputQueueCapacity);
    textureIds.resize(outputQueueCapacity);

    if (!glxContext) {
        glxContext = glxInit(x11Display, x11Window);
    }
    ASSERT(glxContext);

    glGenTextures(outputQueueCapacity, &textureIds[0] );
    for (i=0; i<outputQueueCapacity; i++) {
        int ret = createPixmapForTexture(glxContext, textureIds[i], videoWidth, videoHeight, &pixmaps[i], &glxPixmaps[i]);
        DEBUG("textureIds[%d]: 0x%x, pixmaps[%d]=0x%lx, glxPixmaps[%d]: 0x%lx", i, textureIds[i], i, pixmaps[i], i, glxPixmaps[i]);
        ASSERT(ret == 0);
        ret = YamiV4L2_UsePixmap(fd, i, pixmaps[i]);
        ASSERT(ret == 0);
    }
#else
    if (memoryType == VIDEO_DATA_MEMORY_TYPE_RAW_COPY) {
        rawOutputFrames.resize(outputQueueCapacity);
        for (i=0; i<outputQueueCapacity; i++) {
            struct v4l2_plane planes[k_maxOutputPlaneCount];
            struct v4l2_buffer buffer;
            memset(&buffer, 0, sizeof(buffer));
            memset(planes, 0, sizeof(planes));
            buffer.index = i;
            buffer.type = V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE;
            buffer.memory = V4L2_MEMORY_MMAP;
            buffer.m.planes = planes;
            buffer.length = outputPlaneCount;
            ioctlRet = YamiV4L2_Ioctl(fd, VIDIOC_QUERYBUF, &buffer);
            ASSERT(ioctlRet != -1);

            rawOutputFrames[i].width = format.fmt.pix_mp.width;
            rawOutputFrames[i].height = format.fmt.pix_mp.height;
            rawOutputFrames[i].fourcc = format.fmt.pix_mp.pixelformat;

            for (int j=0; j<outputPlaneCount; j++) {
                // length and mem_offset are filled by VIDIOC_QUERYBUF above
                void* address = YamiV4L2_Mmap(NULL,
                                              buffer.m.planes[j].length,
                                              PROT_READ | PROT_WRITE,
                                              MAP_SHARED, fd,
                                              buffer.m.planes[j].m.mem_offset);
                ASSERT(address);
                if (j == 0) {
                    rawOutputFrames[i].data = static_cast<uint8_t*>(address);
                    rawOutputFrames[i].offset[0] = 0;
                } else {
                    rawOutputFrames[i].offset[j] = static_cast<uint8_t*>(address) - rawOutputFrames[i].data;
                }

                rawOutputFrames[i].pitch[j] = format.fmt.pix_mp.plane_fmt[j].bytesperline;
            }
        }
    } else if (memoryType == VIDEO_DATA_MEMORY_TYPE_DMA_BUF || memoryType == VIDEO_DATA_MEMORY_TYPE_DRM_NAME) {
        // setup all textures and eglImages
        eglImages.resize(outputQueueCapacity);
        textureIds.resize(outputQueueCapacity);

        if (!eglContext)
            eglContext = eglInit(x11Display, x11Window, 0 /*VA_FOURCC_RGBA*/);

        glGenTextures(outputQueueCapacity, &textureIds[0] );
        for (i=0; i<outputQueueCapacity; i++) {
             int ret = 0;
             ret = YamiV4L2_UseEglImage(fd, eglContext->eglContext.display, eglContext->eglContext.context, i, &eglImages[i]);
             ASSERT(ret == 0);
             glBindTexture(GL_TEXTURE_2D, textureIds[i]);
             glEGLImageTargetTexture2DOES(GL_TEXTURE_2D, eglImages[i]);

             glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
             glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
             DEBUG("textureIds[%d]: 0x%x, eglImages[%d]: 0x%x", i, textureIds[i], i, eglImages[i]);
        }
    }
#endif

    // feed output frames first
    for (i=0; i<outputQueueCapacity; i++) {
        if (!takeOneOutputFrame(fd, i)) {
            ASSERT(0);
        }
    }

    bool event_pending=true; // try to get video resolution.
    int dqCountAfterEOS = 0;
    do {
        if (event_pending) {
            handleResolutionChange(fd);
        }

        takeOneOutputFrame(fd);
        if (!feedOneInputFrame(input, fd)) {
            if (stagingBufferInDevice == 0)
                break;
            dqCountAfterEOS++;
        }
        if (dqCountAfterEOS == inputQueueCapacity)  // input drain
            break;
    } while (YamiV4L2_Poll(fd, true, &event_pending) == 0);

    // drain output buffer
    int retry = 3;
    while (takeOneOutputFrame(fd) || (--retry)>0) { // output drain
        usleep(10000);
    }

    calcFps.fps(renderFrameCount);
    // YamiV4L2_Munmap(void* addr, size_t length)
    possibleWait(input->getMimeType());

    // release queued input/output buffer
    memset(&reqbufs, 0, sizeof(reqbufs));
    reqbufs.type = V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE;
    reqbufs.memory = V4L2_MEMORY_MMAP;
    reqbufs.count = 0;
    ioctlRet = YamiV4L2_Ioctl(fd, VIDIOC_REQBUFS, &reqbufs);
    ASSERT(ioctlRet != -1);

    memset(&reqbufs, 0, sizeof(reqbufs));
    reqbufs.type = V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE;
    reqbufs.memory = V4L2_MEMORY_MMAP;
    reqbufs.count = 0;
    ioctlRet = YamiV4L2_Ioctl(fd, VIDIOC_REQBUFS, &reqbufs);
    ASSERT(ioctlRet != -1);

    // stop input port
    type = V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE;
    ioctlRet = YamiV4L2_Ioctl(fd, VIDIOC_STREAMOFF, &type);
    ASSERT(ioctlRet != -1);

    // stop output port
    type = V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE;
    ioctlRet = YamiV4L2_Ioctl(fd, VIDIOC_STREAMOFF, &type);
    ASSERT(ioctlRet != -1);

    if(textureIds.size())
        glDeleteTextures(textureIds.size(), &textureIds[0]);
    ASSERT(glGetError() == GL_NO_ERROR);

#if __ENABLE_V4L2_GLX__
    glxRelease(glxContext, &pixmaps[0], &glxPixmaps[0], pixmaps.size());
#else
    for (i=0; i<eglImages.size(); i++) {
        eglDestroyImageKHR(eglContext->eglContext.display, eglImages[i]);
    }
    /*
    there is still randomly fail in mesa; no good idea for it. seems mesa bug
    0  0x00007ffff079c343 in _mesa_symbol_table_dtor () from /usr/lib/x86_64-linux-gnu/libdricore9.2.1.so.1
    1  0x00007ffff073c55d in glsl_symbol_table::~glsl_symbol_table() () from /usr/lib/x86_64-linux-gnu/libdricore9.2.1.so.1
    2  0x00007ffff072a4d5 in ?? () from /usr/lib/x86_64-linux-gnu/libdricore9.2.1.so.1
    3  0x00007ffff072a4bd in ?? () from /usr/lib/x86_64-linux-gnu/libdricore9.2.1.so.1
    4  0x00007ffff064b48f in _mesa_reference_shader () from /usr/lib/x86_64-linux-gnu/libdricore9.2.1.so.1
    5  0x00007ffff0649397 in ?? () from /usr/lib/x86_64-linux-gnu/libdricore9.2.1.so.1
    6  0x000000000040624d in releaseShader (program=0x77cd90) at ./egl/gles2_help.c:158
    7  eglRelease (context=0x615920) at ./egl/gles2_help.c:310
    8  0x0000000000402ca8 in main (argc=<optimized out>, argv=<optimized out>) at v4l2decode.cpp:531
    */
    if (eglContext)
        eglRelease(eglContext);
#endif

    // close device
    ioctlRet = YamiV4L2_Close(fd);
    ASSERT(ioctlRet != -1);

    if(input)
        delete input;

    if (outfp)
        fclose(outfp);

    if (dumpOutputDir)
        free(dumpOutputDir);

    if (x11Display && x11Window)
        XDestroyWindow(x11Display, x11Window);
    if (x11Display)
        XCloseDisplay(x11Display);

    fprintf(stdout, "decode done\n");
}
Exemple #2
0
int main(int argc, char** argv)
{
    DecodeInput *input;
    int32_t fd = -1;
    int32_t i = 0;
    int32_t ioctlRet = -1;
    YamiMediaCodec::CalcFps calcFps;

    renderMode = 3; // set default render mode to 3

    yamiTraceInit();
#if __ENABLE_V4L2_GLX__
    XInitThreads();
#endif

#if __ENABLE_V4L2_OPS__
    // FIXME, use libv4l2codec_hw.so instead
    if (!loadV4l2CodecDevice("libyami_v4l2.so")) {
        ERROR("fail to init v4l2codec device with __ENABLE_V4L2_OPS__\n");
        return -1;
    }
#endif

    if (!process_cmdline(argc, argv))
        return -1;

    if (!inputFileName) {
        ERROR("no input media file specified\n");
        return -1;
    }
    INFO("input file: %s, renderMode: %d", inputFileName, renderMode);

    if (!dumpOutputName)
        dumpOutputName = strdup ("./");

#if !__ENABLE_V4L2_GLX__
    switch (renderMode) {
    case 0:
        memoryType = VIDEO_DATA_MEMORY_TYPE_RAW_COPY;
        memoryTypeStr = typeStrRawData;
    break;
    case 3:
        memoryType = VIDEO_DATA_MEMORY_TYPE_DRM_NAME;
        memoryTypeStr = typeStrDrmName;
    break;
    case 4:
        memoryType = VIDEO_DATA_MEMORY_TYPE_DMA_BUF;
        memoryTypeStr = typeStrDmaBuf;
    break;
    default:
        ASSERT(0 && "unsupported render mode, -m [0,3,4] are supported");
    break;
    }
#endif

    input = DecodeInput::create(inputFileName);
    if (input==NULL) {
        ERROR("fail to init input stream\n");
        return -1;
    }

    renderFrameCount = 0;
    calcFps.setAnchor();
    // open device
    fd = SIMULATE_V4L2_OP(Open)("decoder", 0);
    ASSERT(fd!=-1);

#ifdef ANDROID
#elif __ENABLE_V4L2_GLX__
    x11Display = XOpenDisplay(NULL);
    ASSERT(x11Display);
    ioctlRet = SIMULATE_V4L2_OP(SetXDisplay)(fd, x11Display);
#endif
    // set output frame memory type
#if __ENABLE_V4L2_OPS__
    SIMULATE_V4L2_OP(SetParameter)(fd, "frame-memory-type", memoryTypeStr);
#elif !__ENABLE_V4L2_GLX__
    SIMULATE_V4L2_OP(FrameMemoryType)(fd, memoryType);
#endif

    // query hw capability
    struct v4l2_capability caps;
    memset(&caps, 0, sizeof(caps));
    caps.capabilities = V4L2_CAP_VIDEO_CAPTURE_MPLANE | V4L2_CAP_VIDEO_OUTPUT_MPLANE | V4L2_CAP_STREAMING;
    ioctlRet = SIMULATE_V4L2_OP(Ioctl)(fd, VIDIOC_QUERYCAP, &caps);
    ASSERT(ioctlRet != -1);

    // set input/output data format
    uint32_t codecFormat = v4l2PixelFormatFromMime(input->getMimeType());
    if (!codecFormat) {
        ERROR("unsupported mimetype, %s", input->getMimeType());
        return -1;
    }

    struct v4l2_format format;
    memset(&format, 0, sizeof(format));
    format.type = V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE;
    format.fmt.pix_mp.pixelformat = codecFormat;
    format.fmt.pix_mp.num_planes = 1;
    format.fmt.pix_mp.plane_fmt[0].sizeimage = k_maxInputBufferSize;
    ioctlRet = SIMULATE_V4L2_OP(Ioctl)(fd, VIDIOC_S_FMT, &format);
    ASSERT(ioctlRet != -1);

    // set preferred output format
    memset(&format, 0, sizeof(format));
    format.type = V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE;
    uint8_t* data = (uint8_t*)input->getCodecData().data();
    uint32_t size = input->getCodecData().size();
    //save codecdata, size+data, the type of format.fmt.raw_data is __u8[200]
    //we must make sure enough space (>=sizeof(uint32_t) + size) to store codecdata
    memcpy(format.fmt.raw_data, &size, sizeof(uint32_t));
    if(sizeof(format.fmt.raw_data) >= size + sizeof(uint32_t))
        memcpy(format.fmt.raw_data + sizeof(uint32_t), data, size);
    else {
        ERROR("No enough space to store codec data");
        return -1;
    }
    ioctlRet = SIMULATE_V4L2_OP(Ioctl)(fd, VIDIOC_S_FMT, &format);
    ASSERT(ioctlRet != -1);

    // input port starts as early as possible to decide output frame format
    __u32 type = V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE;
    ioctlRet = SIMULATE_V4L2_OP(Ioctl)(fd, VIDIOC_STREAMON, &type);
    ASSERT(ioctlRet != -1);

    // setup input buffers
    struct v4l2_requestbuffers reqbufs;
    memset(&reqbufs, 0, sizeof(reqbufs));
    reqbufs.type = V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE;
    reqbufs.memory = V4L2_MEMORY_MMAP;
    reqbufs.count = 2;
    ioctlRet = SIMULATE_V4L2_OP(Ioctl)(fd, VIDIOC_REQBUFS, &reqbufs);
    ASSERT(ioctlRet != -1);
    ASSERT(reqbufs.count>0);
    inputQueueCapacity = reqbufs.count;
    inputFrames.resize(inputQueueCapacity);

    for (i=0; i<inputQueueCapacity; i++) {
        struct v4l2_plane planes[k_inputPlaneCount];
        struct v4l2_buffer buffer;
        memset(&buffer, 0, sizeof(buffer));
        memset(planes, 0, sizeof(planes));
        buffer.index = i;
        buffer.type = V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE;
        buffer.memory = V4L2_MEMORY_MMAP;
        buffer.m.planes = planes;
        buffer.length = k_inputPlaneCount;
        ioctlRet = SIMULATE_V4L2_OP(Ioctl)(fd, VIDIOC_QUERYBUF, &buffer);
        ASSERT(ioctlRet != -1);

        // length and mem_offset should be filled by VIDIOC_QUERYBUF above
        void* address = SIMULATE_V4L2_OP(Mmap)(NULL,
                                      buffer.m.planes[0].length,
                                      PROT_READ | PROT_WRITE,
                                      MAP_SHARED, fd,
                                      buffer.m.planes[0].m.mem_offset);
        ASSERT(address);
        inputFrames[i] = static_cast<uint8_t*>(address);
        DEBUG("inputFrames[%d] = %p", i, inputFrames[i]);
    }

    // feed input frames first
    for (i=0; i<inputQueueCapacity; i++) {
        if (!feedOneInputFrame(input, fd, i)) {
            break;
        }
    }

    // query video resolution
    memset(&format, 0, sizeof(format));
    format.type = V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE;
    while (1) {
        if (SIMULATE_V4L2_OP(Ioctl)(fd, VIDIOC_G_FMT, &format) != 0) {
            if (errno != EINVAL) {
                // EINVAL means we haven't seen sufficient stream to decode the format.
                INFO("ioctl() failed: VIDIOC_G_FMT, haven't get video resolution during start yet, waiting");
            }
        } else {
            break;
        }
        usleep(50);
    }
    outputPlaneCount = format.fmt.pix_mp.num_planes;
    ASSERT(outputPlaneCount == 2);
    videoWidth = format.fmt.pix_mp.width;
    videoHeight = format.fmt.pix_mp.height;
    ASSERT(videoWidth && videoHeight);

#ifdef ANDROID
    __u32 pixelformat = format.fmt.pix_mp.pixelformat;
    if (!createNativeWindow(pixelformat)) {
        fprintf(stderr, "create native window error\n");
        return -1;
    }

    int minUndequeuedBuffs = 0;
    status_t err = mNativeWindow->query(mNativeWindow.get(), NATIVE_WINDOW_MIN_UNDEQUEUED_BUFFERS, &minUndequeuedBuffs);
    if (err != 0) {
        fprintf(stderr, "query native window min undequeued buffers error\n");
        return err;
    }
#endif

    // setup output buffers
    // Number of output buffers we need.
    struct v4l2_control ctrl;
    memset(&ctrl, 0, sizeof(ctrl));
    ctrl.id = V4L2_CID_MIN_BUFFERS_FOR_CAPTURE;
    ioctlRet = SIMULATE_V4L2_OP(Ioctl)(fd, VIDIOC_G_CTRL, &ctrl);
#ifndef ANDROID
    uint32_t minOutputFrameCount = ctrl.value + k_extraOutputFrameCount;
#else
    uint32_t minOutputFrameCount = ctrl.value + k_extraOutputFrameCount + minUndequeuedBuffs;
#endif

    memset(&reqbufs, 0, sizeof(reqbufs));
    reqbufs.type = V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE;
    reqbufs.memory = V4L2_MEMORY_MMAP;
    reqbufs.count = minOutputFrameCount;
    ioctlRet = SIMULATE_V4L2_OP(Ioctl)(fd, VIDIOC_REQBUFS, &reqbufs);
    ASSERT(ioctlRet != -1);
    ASSERT(reqbufs.count>0);
    outputQueueCapacity = reqbufs.count;

#ifdef ANDROID
#elif __ENABLE_V4L2_GLX__
    x11Window = XCreateSimpleWindow(x11Display, DefaultRootWindow(x11Display)
        , 0, 0, videoWidth, videoHeight, 0, 0
        , WhitePixel(x11Display, 0));
    XMapWindow(x11Display, x11Window);
    pixmaps.resize(outputQueueCapacity);
    glxPixmaps.resize(outputQueueCapacity);
    textureIds.resize(outputQueueCapacity);

    if (!glxContext) {
        glxContext = glxInit(x11Display, x11Window);
    }
    ASSERT(glxContext);

    glGenTextures(outputQueueCapacity, &textureIds[0] );
    for (i=0; i<outputQueueCapacity; i++) {
        int ret = createPixmapForTexture(glxContext, textureIds[i], videoWidth, videoHeight, &pixmaps[i], &glxPixmaps[i]);
        DEBUG("textureIds[%d]: 0x%x, pixmaps[%d]=0x%lx, glxPixmaps[%d]: 0x%lx", i, textureIds[i], i, pixmaps[i], i, glxPixmaps[i]);
        ASSERT(ret == 0);
        ret = SIMULATE_V4L2_OP(UsePixmap)(fd, i, pixmaps[i]);
        ASSERT(ret == 0);
    }
#else
    if (IS_RAW_DATA()) {
        rawOutputFrames.resize(outputQueueCapacity);
        for (i=0; i<outputQueueCapacity; i++) {
            struct v4l2_plane planes[k_maxOutputPlaneCount];
            struct v4l2_buffer buffer;
            memset(&buffer, 0, sizeof(buffer));
            memset(planes, 0, sizeof(planes));
            buffer.index = i;
            buffer.type = V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE;
            buffer.memory = V4L2_MEMORY_MMAP;
            buffer.m.planes = planes;
            buffer.length = outputPlaneCount;
            ioctlRet = SIMULATE_V4L2_OP(Ioctl)(fd, VIDIOC_QUERYBUF, &buffer);
            ASSERT(ioctlRet != -1);

            rawOutputFrames[i].width = format.fmt.pix_mp.width;
            rawOutputFrames[i].height = format.fmt.pix_mp.height;
            rawOutputFrames[i].fourcc = format.fmt.pix_mp.pixelformat;

            for (int j=0; j<outputPlaneCount; j++) {
                // length and mem_offset are filled by VIDIOC_QUERYBUF above
                void* address = SIMULATE_V4L2_OP(Mmap)(NULL,
                                              buffer.m.planes[j].length,
                                              PROT_READ | PROT_WRITE,
                                              MAP_SHARED, fd,
                                              buffer.m.planes[j].m.mem_offset);
                ASSERT(address);
                if (j == 0) {
                    rawOutputFrames[i].data = static_cast<uint8_t*>(address);
                    rawOutputFrames[i].offset[0] = 0;
                } else {
                    rawOutputFrames[i].offset[j] = static_cast<uint8_t*>(address) - rawOutputFrames[i].data;
                }

                rawOutputFrames[i].pitch[j] = format.fmt.pix_mp.plane_fmt[j].bytesperline;
            }
        }
    } else if (IS_DMA_BUF() || IS_DRM_NAME()) {
        // setup all textures and eglImages
        eglImages.resize(outputQueueCapacity);
        textureIds.resize(outputQueueCapacity);

        if (!eglContext)
            eglContext = eglInit(x11Display, x11Window, 0 /*VA_FOURCC_RGBA*/, IS_DMA_BUF());

        glGenTextures(outputQueueCapacity, &textureIds[0] );
        for (i=0; i<outputQueueCapacity; i++) {
             int ret = 0;
             ret = SIMULATE_V4L2_OP(UseEglImage)(fd, eglContext->eglContext.display, eglContext->eglContext.context, i, &eglImages[i]);
             ASSERT(ret == 0);

             GLenum target = GL_TEXTURE_2D;
             if (IS_DMA_BUF())
                 target = GL_TEXTURE_EXTERNAL_OES;
             glBindTexture(target, textureIds[i]);
             imageTargetTexture2D(target, eglImages[i]);

             glTexParameteri(target, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
             glTexParameteri(target, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
             DEBUG("textureIds[%d]: 0x%x, eglImages[%d]: 0x%p", i, textureIds[i], i, eglImages[i]);
        }
    }
#endif

#ifndef ANDROID
    // feed output frames first
    for (i=0; i<outputQueueCapacity; i++) {
        if (!takeOneOutputFrame(fd, i)) {
            ASSERT(0);
        }
    }
#else
    struct v4l2_buffer buffer;

    err = native_window_set_buffer_count(mNativeWindow.get(), outputQueueCapacity);
    if (err != 0) {
        fprintf(stderr, "native_window_set_buffer_count failed: %s (%d)", strerror(-err), -err);
        return -1;
    }

    //queue buffs
    for (uint32_t i = 0; i < outputQueueCapacity; i++) {
        ANativeWindowBuffer* pbuf = NULL;
        memset(&buffer, 0, sizeof(buffer));

        err = native_window_dequeue_buffer_and_wait(mNativeWindow.get(), &pbuf);
        if (err != 0) {
            fprintf(stderr, "dequeueBuffer failed: %s (%d)\n", strerror(-err), -err);
            return -1;
        }

        buffer.m.userptr = (unsigned long)pbuf;
        buffer.type = V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE;
        buffer.index = i;

        ioctlRet = SIMULATE_V4L2_OP(Ioctl)(fd, VIDIOC_QBUF, &buffer);
        ASSERT(ioctlRet != -1);
        mWindBuff.push_back(pbuf);
    }

    for (uint32_t i = 0; i < minUndequeuedBuffs; i++) {
        memset(&buffer, 0, sizeof(buffer));
        buffer.type = V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE;

        ioctlRet = SIMULATE_V4L2_OP(Ioctl)(fd, VIDIOC_DQBUF, &buffer);
        ASSERT(ioctlRet != -1);

        err = mNativeWindow->cancelBuffer(mNativeWindow.get(), mWindBuff[buffer.index], -1);
        if (err) {
            fprintf(stderr, "queue empty window buffer error\n");
            return -1;
        }
    }
#endif

    // output port starts as late as possible to adopt user provide output buffer
    type = V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE;
    ioctlRet = SIMULATE_V4L2_OP(Ioctl)(fd, VIDIOC_STREAMON, &type);
    ASSERT(ioctlRet != -1);

    bool event_pending=true; // try to get video resolution.
    int dqCountAfterEOS = 0;
    do {
        if (event_pending) {
            handleResolutionChange(fd);
        }

        takeOneOutputFrame(fd);
        if (!feedOneInputFrame(input, fd)) {
            if (stagingBufferInDevice == 0)
                break;
            dqCountAfterEOS++;
        }
        if (dqCountAfterEOS == inputQueueCapacity)  // input drain
            break;
    } while (SIMULATE_V4L2_OP(Poll)(fd, true, &event_pending) == 0);

    // drain output buffer
    int retry = 3;
    while (takeOneOutputFrame(fd) || (--retry)>0) { // output drain
        usleep(10000);
    }

    calcFps.fps(renderFrameCount);
    // SIMULATE_V4L2_OP(Munmap)(void* addr, size_t length)
    possibleWait(input->getMimeType());

    // release queued input/output buffer
    memset(&reqbufs, 0, sizeof(reqbufs));
    reqbufs.type = V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE;
    reqbufs.memory = V4L2_MEMORY_MMAP;
    reqbufs.count = 0;
    ioctlRet = SIMULATE_V4L2_OP(Ioctl)(fd, VIDIOC_REQBUFS, &reqbufs);
    ASSERT(ioctlRet != -1);

    memset(&reqbufs, 0, sizeof(reqbufs));
    reqbufs.type = V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE;
    reqbufs.memory = V4L2_MEMORY_MMAP;
    reqbufs.count = 0;
    ioctlRet = SIMULATE_V4L2_OP(Ioctl)(fd, VIDIOC_REQBUFS, &reqbufs);
    ASSERT(ioctlRet != -1);

    // stop input port
    type = V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE;
    ioctlRet = SIMULATE_V4L2_OP(Ioctl)(fd, VIDIOC_STREAMOFF, &type);
    ASSERT(ioctlRet != -1);

    // stop output port
    type = V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE;
    ioctlRet = SIMULATE_V4L2_OP(Ioctl)(fd, VIDIOC_STREAMOFF, &type);
    ASSERT(ioctlRet != -1);

#ifndef ANDROID
    if(textureIds.size())
        glDeleteTextures(textureIds.size(), &textureIds[0]);
    ASSERT(glGetError() == GL_NO_ERROR);
#endif

#ifdef ANDROID
    //TODO, some resources need to destroy?
#elif __ENABLE_V4L2_GLX__
    glxRelease(glxContext, &pixmaps[0], &glxPixmaps[0], pixmaps.size());
#else
    for (i=0; i<eglImages.size(); i++) {
        destroyImage(eglContext->eglContext.display, eglImages[i]);
    }
    /*
    there is still randomly fail in mesa; no good idea for it. seems mesa bug
    0  0x00007ffff079c343 in _mesa_symbol_table_dtor () from /usr/lib/x86_64-linux-gnu/libdricore9.2.1.so.1
    1  0x00007ffff073c55d in glsl_symbol_table::~glsl_symbol_table() () from /usr/lib/x86_64-linux-gnu/libdricore9.2.1.so.1
    2  0x00007ffff072a4d5 in ?? () from /usr/lib/x86_64-linux-gnu/libdricore9.2.1.so.1
    3  0x00007ffff072a4bd in ?? () from /usr/lib/x86_64-linux-gnu/libdricore9.2.1.so.1
    4  0x00007ffff064b48f in _mesa_reference_shader () from /usr/lib/x86_64-linux-gnu/libdricore9.2.1.so.1
    5  0x00007ffff0649397 in ?? () from /usr/lib/x86_64-linux-gnu/libdricore9.2.1.so.1
    6  0x000000000040624d in releaseShader (program=0x77cd90) at ./egl/gles2_help.c:158
    7  eglRelease (context=0x615920) at ./egl/gles2_help.c:310
    8  0x0000000000402ca8 in main (argc=<optimized out>, argv=<optimized out>) at v4l2decode.cpp:531
    */
    if (eglContext)
        eglRelease(eglContext);
#endif

    // close device
    ioctlRet = SIMULATE_V4L2_OP(Close)(fd);
    ASSERT(ioctlRet != -1);

    if(input)
        delete input;

    if (outfp)
        fclose(outfp);

    if (dumpOutputName)
        free(dumpOutputName);

#if __ENABLE_V4L2_GLX__
    if (x11Display && x11Window)
        XDestroyWindow(x11Display, x11Window);
    if (x11Display)
        XCloseDisplay(x11Display);
#endif

    fprintf(stdout, "decode done\n");
}