VideoFrame VideoDecoderFFmpegHW::copyToFrame(const VideoFormat& fmt, int surface_h, quint8 *src[], int pitch[], bool swapUV)
{
DPTR_D(VideoDecoderFFmpegHW);
Q_ASSERT_X(src[0] && pitch[0] > 0, "VideoDecoderFFmpegHW::copyToFrame", "src[0] and pitch[0] must be set");
const int nb_planes = fmt.planeCount();
const int chroma_pitch = nb_planes > 1 ? fmt.bytesPerLine(pitch[0], 1) : 0;
const int chroma_h = fmt.chromaHeight(surface_h);
int h[] = { surface_h, 0, 0};
for (int i = 1; i < nb_planes; ++i) {
h[i] = chroma_h;
// set chroma address and pitch if not set
if (pitch[i] <= 0)
pitch[i] = chroma_pitch;
if (!src[i])
src[i] = src[i-1] + pitch[i-1]*h[i-1];
}
if (swapUV) {
std::swap(src[1], src[2]);
std::swap(pitch[1], pitch[2]);
}
VideoFrame frame;
if (copyMode() == VideoDecoderFFmpegHW::OptimizedCopy && d.gpu_mem.isReady()) {
int yuv_size = 0;
for (int i = 0; i < nb_planes; ++i) {
yuv_size += pitch[i]*h[i];
}
// additional 15 bytes to ensure 16 bytes aligned
QByteArray buf(15 + yuv_size, 0);
const int offset_16 = (16 - ((uintptr_t)buf.data() & 0x0f)) & 0x0f;
// plane 1, 2... is aligned?
uchar* plane_ptr = (uchar*)buf.data() + offset_16;
QVector<uchar*> dst(nb_planes, 0);
for (int i = 0; i < nb_planes; ++i) {
dst[i] = plane_ptr;
// TODO: add VideoFormat::planeWidth/Height() ?
// pitch instead of surface_width
plane_ptr += pitch[i] * h[i];
d.gpu_mem.copyFrame(src[i], dst[i], pitch[i], h[i], pitch[i]);
}
frame = VideoFrame(buf, width(), height(), fmt);
frame.setBits(dst);
frame.setBytesPerLine(pitch);
} else {
frame = VideoFrame(width(), height(), fmt);
frame.setBits(src);
frame.setBytesPerLine(pitch);
// TODO: why clone is faster()?
// TODO: buffer pool and create VideoFrame when needed to avoid copy? also for other va
frame = frame.clone();
}
frame.setTimestamp(double(d.frame->pkt_pts)/1000.0);
frame.setDisplayAspectRatio(d.getDAR(d.frame));
d.updateColorDetails(&frame);
return frame;
}
void GLWidgetRendererPrivate::uploadPlane(int p, GLint internalFormat, GLenum format, const QRect& roi)
{
// FIXME: why happens on win?
if (video_frame.bytesPerLine(p) <= 0)
return;
glActiveTexture(GL_TEXTURE0 + p); //xbmc: only for es, not for desktop?
glBindTexture(GL_TEXTURE_2D, textures[p]);
////nv12: 2
//glPixelStorei(GL_UNPACK_ALIGNMENT, 1);//GetAlign(video_frame.bytesPerLine(p)));
#if defined(GL_UNPACK_ROW_LENGTH)
// glPixelStorei(GL_UNPACK_ROW_LENGTH, video_frame.bytesPerLine(p));
#endif
setupQuality();
//qDebug("bpl[%d]=%d width=%d", p, video_frame.bytesPerLine(p), video_frame.planeWidth(p));
// This is necessary for non-power-of-two textures
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
//uploading part of image eats less gpu memory, but may be more cpu(gles)
//FIXME: more cpu usage then qpainter. FBO, VBO?
//roi for planes?
if (ROI_TEXCOORDS || roi.size() == video_frame.size()) {
glTexSubImage2D(GL_TEXTURE_2D
, 0 //level
, 0 // xoffset
, 0 // yoffset
, texture_size[p].width()
, texture_size[p].height()
, format //format, must the same as internal format?
, data_type[p]
, video_frame.bits(p));
} else {
int roi_x = roi.x();
int roi_y = roi.y();
int roi_w = roi.width();
int roi_h = roi.height();
int plane_w = video_frame.planeWidth(p);
VideoFormat fmt = video_frame.format();
if (p == 0) {
plane0Size = QSize(roi_w, roi_h);
} else {
roi_x = fmt.chromaWidth(roi_x);
roi_y = fmt.chromaHeight(roi_y);
roi_w = fmt.chromaWidth(roi_w);
roi_h = fmt.chromaHeight(roi_h);
}
qDebug("roi: %d, %d %dx%d", roi_x, roi_y, roi_w, roi_h);
#if 0// defined(GL_UNPACK_ROW_LENGTH) && defined(GL_UNPACK_SKIP_PIXELS)
// http://stackoverflow.com/questions/205522/opengl-subtexturing
glPixelStorei(GL_UNPACK_ROW_LENGTH, plane_w);
//glPixelStorei or compute pointer
glPixelStorei(GL_UNPACK_SKIP_PIXELS, roi_x);
glPixelStorei(GL_UNPACK_SKIP_ROWS, roi_y);
glTexImage2D(GL_TEXTURE_2D, 0, internalFormat, roi_w, roi_h, 0, format, GL_UNSIGNED_BYTE, video_frame.bits(p));
//glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, roi_w, roi_h, format, GL_UNSIGNED_BYTE, video_frame.bits(p));
glPixelStorei(GL_UNPACK_ROW_LENGTH, 0);
glPixelStorei(GL_UNPACK_SKIP_PIXELS, 0);
glPixelStorei(GL_UNPACK_SKIP_ROWS, 0);
#else // GL ES
//define it? or any efficient way?
//glPixelStorei(GL_UNPACK_ALIGNMENT, 1);
glTexImage2D(GL_TEXTURE_2D, 0, internalFormat, roi_w, roi_h, 0, format, GL_UNSIGNED_BYTE, NULL);
const char *src = (char*)video_frame.bits(p) + roi_y*plane_w + roi_x*fmt.bytesPerPixel(p);
#define UPLOAD_LINE 1
#if UPLOAD_LINE
for (int y = 0; y < roi_h; y++) {
glTexSubImage2D(GL_TEXTURE_2D, 0, 0, y, roi_w, 1, format, GL_UNSIGNED_BYTE, src);
}
#else
int line_size = roi_w*fmt.bytesPerPixel(p);
char *sub = (char*)malloc(roi_h*line_size);
char *dst = sub;
for (int y = 0; y < roi_h; y++) {
memcpy(dst, src, line_size);
src += video_frame.bytesPerLine(p);
dst += line_size;
}
// FIXME: crash
glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, roi_w, roi_h, format, GL_UNSIGNED_BYTE, sub);
free(sub);
#endif //UPLOAD_LINE
#endif //GL_UNPACK_ROW_LENGTH
}
#if defined(GL_UNPACK_ROW_LENGTH)
// glPixelStorei(GL_UNPACK_ROW_LENGTH, 0);
#endif
//glPixelStorei(GL_UNPACK_ALIGNMENT, 4);
glBindTexture(GL_TEXTURE_2D, 0);
}
