static int vaapi_encode_wait(AVCodecContext *avctx,
VAAPIEncodePicture *pic)
{
VAAPIEncodeContext *ctx = avctx->priv_data;
VAStatus vas;
av_assert0(pic->encode_issued);
if (pic->encode_complete) {
// Already waited for this picture.
return 0;
}
av_log(avctx, AV_LOG_DEBUG, "Sync to pic %"PRId64"/%"PRId64" "
"(recon surface %#x).\n", pic->display_order,
pic->encode_order, pic->recon_surface);
vas = vaSyncSurface(ctx->hwctx->display, pic->recon_surface);
if (vas != VA_STATUS_SUCCESS) {
av_log(avctx, AV_LOG_ERROR, "Failed to sync to picture completion: "
"%d (%s).\n", vas, vaErrorStr(vas));
return AVERROR(EIO);
}
// Input is definitely finished with now.
av_frame_free(&pic->input_image);
pic->encode_complete = 1;
return 0;
}
bool GLXInteropResource::map(const surface_ptr& surface, GLuint tex, int w, int h, int)
{
Q_UNUSED(w);
Q_UNUSED(h);
surface_glx_ptr glx = glx_surfaces[tex];
if(!surface.isNull() && (surface->width() != m_width || m_height != surface->height())) {
m_width = surface->width();
m_height = surface->height();
m_regenerateGlx = true;
}
if(m_regenerateGlx || !glx) {
if (glx) { glx.clear(); }
glx = surface_glx_ptr(new surface_glx_t(surface->display()));
if (!glx->create(tex))
glx = surface_glx_ptr();
glx_surfaces[tex] = glx;
}
if (!glx) {
qWarning("Fail to create vaapi glx surface");
return false;
}
if (!glx->copy(surface))
return false;
VAWARN(vaSyncSurface(surface->vadisplay(), surface->get()));
return true;
}
bool VaApiMixer::upload(const VideoFrame &frame, bool deint) {
if (!m_glSurface)
return false;
static const int specs[MP_CSP_COUNT] = {
0, //MP_CSP_AUTO,
VA_SRC_BT601, //MP_CSP_BT_601,
VA_SRC_BT709, //MP_CSP_BT_709,
VA_SRC_SMPTE_240, //MP_CSP_SMPTE_240M,
0, //MP_CSP_RGB,
0, //MP_CSP_XYZ,
0, //MP_CSP_YCGCO,
};
static const int field[] = {
// Picture = 0, Top = 1, Bottom = 2
VA_FRAME_PICTURE, VA_TOP_FIELD, VA_BOTTOM_FIELD, VA_FRAME_PICTURE
};
const auto id = (VASurfaceID)(quintptr)frame.data(3);
int flags = specs[frame.format().colorspace()];
if (deint)
flags |= field[frame.field() & VideoFrame::Interlaced];
if (!check(vaCopySurfaceGLX(VaApi::glx(), m_glSurface, id, flags), "Cannot copy OpenGL surface."))
return false;
if (!check(vaSyncSurface(VaApi::glx(), id), "Cannot sync video surface."))
return false;
return true;
}
bool VaapiSurfaceGLXImpl::update(boost::shared_ptr<VaapiSurface> surface)
{
GNASH_REPORT_FUNCTION;
if (!this->surface()) {
return false;
}
VaapiGlobalContext * gvactx = VaapiGlobalContext::get();
if (!gvactx) {
return false;
}
VAStatus status;
status = vaSyncSurface(gvactx->display(), surface->get());
if (!vaapi_check_status(status, "vaSyncSurface()"))
return false;
status = vaCopySurfaceGLX(gvactx->display(), this->surface(),
surface->get(), VA_FRAME_PICTURE);
if (!vaapi_check_status(status, "vaCopySurfaceGLX()")) {
return false;
}
return true;
}
bool VaapiSurface::sync()
{
VAStatus status;
status = vaSyncSurface((VADisplay) m_display->getID(), (VASurfaceID) m_ID);
if (!checkVaapiStatus(status, "vaSyncSurface()"))
return false;
return true;
}
Decode_Status DecodeOutputNull::renderOneFrame(bool drain)
{
m_frame.memoryType = VIDEO_DATA_MEMORY_TYPE_SURFACE_ID;
m_frame.width = 0;
m_frame.height = 0;
m_frame.fourcc = 0;
Decode_Status status = m_decoder->getOutput(&m_frame, drain);
if (status == RENDER_SUCCESS)
vaSyncSurface((void*)(m_frame.handle), m_frame.internalID);
return status;
}
void VAApiWriter::draw( VASurfaceID _id, int _field )
{
if ( _id != VA_INVALID_SURFACE && _field > -1 )
{
if ( id != _id || _field == field )
vaSyncSurface( VADisp, _id );
id = _id;
field = _field;
}
if ( id == VA_INVALID_SURFACE )
return;
bool associated = false;
osd_mutex.lock();
if ( !osd_list.isEmpty() )
{
QRect bounds;
const qreal scaleW = ( qreal )W / outW, scaleH = ( qreal )H / outH;
bool mustRepaint = Functions::mustRepaintOSD( osd_list, osd_checksums, &scaleW, &scaleH, &bounds );
if ( !mustRepaint )
mustRepaint = vaImgSize != bounds.size();
bool canAssociate = !mustRepaint;
if ( mustRepaint )
{
if ( vaImgSize != bounds.size() )
{
clearRGBImage();
vaImgSize = QSize();
if ( vaCreateImage( VADisp, rgbImgFmt, bounds.width(), bounds.height(), &vaImg ) == VA_STATUS_SUCCESS )
{
if ( vaCreateSubpicture( VADisp, vaImg.image_id, &vaSubpicID ) == VA_STATUS_SUCCESS )
vaImgSize = bounds.size();
else
clearRGBImage();
}
}
if ( vaSubpicID )
{
quint8 *buff;
if ( vaMapBuffer( VADisp, vaImg.buf, ( void ** )&buff ) == VA_STATUS_SUCCESS )
{
QImage osdImg( buff += vaImg.offsets[ 0 ], vaImg.pitches[ 0 ] >> 2, bounds.height(), QImage::Format_ARGB32 );
osdImg.fill( 0 );
QPainter p( &osdImg );
p.translate( -bounds.topLeft() );
Functions::paintOSD( osd_list, scaleW, scaleH, p, &osd_checksums );
vaUnmapBuffer( VADisp, vaImg.buf );
canAssociate = true;
}
}
}
uint8_t* VAAPIContext::GetSurfaceIDPointer(void* buf)
{
if (!buf)
return NULL;
const vaapi_surface *surf = (vaapi_surface*)buf;
if (!surf->m_id)
return NULL;
INIT_ST;
va_status = vaSyncSurface(m_ctx.display, surf->m_id);
CHECK_ST;
return (uint8_t*)(uintptr_t)surf->m_id;
}
mp_image *render(const VideoFrame &frame, int flags) {
if (m_rebuild)
update();
auto in = VaApiSurfacePool::getSurface(frame.mpi());
if (!in)
return nullptr;
m_pool.create(5, frame.width(), frame.height(), in->format());
auto ret = m_pool.getMpImage();
if (!ret)
return nullptr;
auto out = VaApiSurfacePool::getSurface(ret);
if (!out)
return nullptr;
VAProcPipelineParameterBuffer *param = nullptr;
VABufferID buffer = BufferMan::create(m_context, VAProcPipelineParameterBufferType, param, 1);
if (buffer == VA_INVALID_ID)
return nullptr;
enum {Begun = 1, Rendered = 2};
int state = 0;
auto pass = [this, out, &ret, &buffer, &state, &frame] () -> mp_image* {
if (state & Begun)
vaEndPicture(VaApi::glx(), m_context);
if (state & Rendered) {
mp_image_copy_attributes(ret, frame.mpi());
} else
mp_image_unrefp(&ret);
vaDestroyBuffer(VaApi::glx(), buffer);
vaSyncSurface(VaApi::glx(), out->id());
return ret;
};
if (!isSuccess(vaBeginPicture(VaApi::glx(), m_context, out->id())))
return pass();
state |= Begun;
if (!BufferMan::map(buffer, param))
return pass();
memset(param, 0, sizeof(*param));
param->surface = in->id();
param->filter_flags = flags;
param->filters = &m_buffers.first();
param->num_filters = m_buffers.size();
param->forward_references = m_forward_refs.data();
param->backward_references = m_backward_refs.data();
param->num_forward_references = m_caps.num_forward_references;
param->num_backward_references = m_caps.num_backward_references;
BufferMan::unmap(buffer);
if (!isSuccess(vaRenderPicture(VaApi::glx(), m_context, &buffer, 1)))
return pass();
state |= Rendered;
return pass();
}
bool GLXInteropResource::map(const surface_ptr& surface, GLuint tex, int w, int h, int)
{
Q_UNUSED(w);
Q_UNUSED(h);
surface_glx_ptr glx = surfaceGLX(surface->display(), tex);
if (!glx) {
qWarning("Fail to create vaapi glx surface");
return false;
}
if (!glx->copy(surface))
return false;
VAWARN(vaSyncSurface(surface->vadisplay(), surface->get()));
return true;
}
quint8 *VAApiWriter::getImage( VAImage &image, VASurfaceID surfaceID, VAImageFormat *img_fmt ) const
{
if ( vaCreateImage( VADisp, img_fmt, outW, outH, &image ) == VA_STATUS_SUCCESS )
{
quint8 *data;
if
(
vaSyncSurface( VADisp, surfaceID ) == VA_STATUS_SUCCESS &&
vaGetImage( VADisp, surfaceID, 0, 0, outW, outH, image.image_id ) == VA_STATUS_SUCCESS &&
vaMapBuffer( VADisp, image.buf, ( void ** )&data ) == VA_STATUS_SUCCESS
) return data;
vaDestroyImage( VADisp, image.image_id );
}
return NULL;
}
/**
* gst_vaapi_surface_sync:
* @surface: a #GstVaapiSurface
*
* Blocks until all pending operations on the @surface have been
* completed.
*
* Return value: %TRUE on success
*/
gboolean
gst_vaapi_surface_sync (GstVaapiSurface * surface)
{
GstVaapiDisplay *display;
VAStatus status;
g_return_val_if_fail (surface != NULL, FALSE);
display = GST_VAAPI_OBJECT_DISPLAY (surface);
if (!display)
return FALSE;
GST_VAAPI_DISPLAY_LOCK (display);
status = vaSyncSurface (GST_VAAPI_DISPLAY_VADISPLAY (display),
GST_VAAPI_OBJECT_ID (surface));
GST_VAAPI_DISPLAY_UNLOCK (display);
if (!vaapi_check_status (status, "vaSyncSurface()"))
return FALSE;
return TRUE;
}
VideoFrame VideoDecoderVAAPI::frame()
{
DPTR_D(VideoDecoderVAAPI);
if (!d.frame->opaque || !d.frame->data[0])
return VideoFrame();
VASurfaceID surface_id = (VASurfaceID)(uintptr_t)d.frame->data[3];
VAStatus status = VA_STATUS_SUCCESS;
if (display() == GLX) {
d.surface_interop->setSurface((va_surface_t*)d.frame->opaque, d.surface_width, d.surface_height);
VideoFrame f(d.surface_width, d.surface_height, VideoFormat::Format_RGB32);
f.setBytesPerLine(d.surface_width*4); //used by gl to compute texture size
f.setSurfaceInterop(d.surface_interop);
return f;
}
#if VA_CHECK_VERSION(0,31,0)
if ((status = vaSyncSurface(d.display, surface_id)) != VA_STATUS_SUCCESS) {
qWarning("vaSyncSurface(VADisplay:%p, VASurfaceID:%#x) == %#x", d.display, surface_id, status);
#else
if (vaSyncSurface(d.display, d.context_id, surface_id)) {
qWarning("vaSyncSurface(VADisplay:%#x, VAContextID:%#x, VASurfaceID:%#x) == %#x", d.display, d.context_id, surface_id, status);
#endif
return VideoFrame();
}
if (!d.disable_derive && d.supports_derive) {
/*
* http://web.archiveorange.com/archive/v/OAywENyq88L319OcRnHI
* vaDeriveImage is faster than vaGetImage. But VAImage is uncached memory and copying from it would be terribly slow
* TODO: copy from USWC, see vlc and https://github.com/OpenELEC/OpenELEC.tv/pull/2937.diff
* https://software.intel.com/en-us/articles/increasing-memory-throughput-with-intel-streaming-simd-extensions-4-intel-sse4-streaming-load
*/
status = vaDeriveImage(d.display, surface_id, &d.image);
if (status != VA_STATUS_SUCCESS) {
qWarning("vaDeriveImage(VADisplay:%p, VASurfaceID:%#x, VAImage*:%p) == %#x", d.display, surface_id, &d.image, status);
return VideoFrame();
}
} else {
status = vaGetImage(d.display, surface_id, 0, 0, d.surface_width, d.surface_height, d.image.image_id);
if (status != VA_STATUS_SUCCESS) {
qWarning("vaGetImage(VADisplay:%p, VASurfaceID:%#x, 0,0, %d, %d, VAImageID:%#x) == %#x", d.display, surface_id, d.surface_width, d.surface_height, d.image.image_id, status);
return VideoFrame();
}
}
void *p_base;
if ((status = vaMapBuffer(d.display, d.image.buf, &p_base)) != VA_STATUS_SUCCESS) {
qWarning("vaMapBuffer(VADisplay:%p, VABufferID:%#x, pBuf:%p) == %#x", d.display, d.image.buf, &p_base, status);
return VideoFrame();
}
VideoFormat::PixelFormat pixfmt = VideoFormat::Format_Invalid;
bool swap_uv = false;
switch (d.image.format.fourcc) {
case VA_FOURCC_YV12:
swap_uv |= d.disable_derive || !d.supports_derive;
pixfmt = VideoFormat::Format_YUV420P;
break;
case VA_FOURCC_IYUV:
swap_uv = true;
pixfmt = VideoFormat::Format_YUV420P;
break;
case VA_FOURCC_NV12:
pixfmt = VideoFormat::Format_NV12;
break;
default:
break;
}
if (pixfmt == VideoFormat::Format_Invalid) {
qWarning("unsupported vaapi pixel format: %#x", d.image.format.fourcc);
return VideoFrame();
}
const VideoFormat fmt(pixfmt);
uint8_t *src[3];
int pitch[3];
for (int i = 0; i < fmt.planeCount(); ++i) {
src[i] = (uint8_t*)p_base + d.image.offsets[i];
pitch[i] = d.image.pitches[i];
}
if (swap_uv) {
std::swap(src[1], src[2]);
std::swap(pitch[1], pitch[2]);
}
VideoFrame frame;
if (d.copy_uswc && d.gpu_mem.isReady()) {
int yuv_size = 0;
if (pixfmt == VideoFormat::Format_NV12)
yuv_size = pitch[0]*d.surface_height*3/2;
else
yuv_size = pitch[0]*d.surface_height + pitch[1]*d.surface_height/2 + pitch[2]*d.surface_height/2;
// 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(fmt.planeCount(), 0);
for (int i = 0; i < dst.size(); ++i) {
dst[i] = plane_ptr;
// TODO: add VideoFormat::planeWidth/Height() ?
const int plane_w = pitch[i];//(i == 0 || pixfmt == VideoFormat::Format_NV12) ? d.surface_width : fmt.chromaWidth(d.surface_width);
const int plane_h = i == 0 ? d.surface_height : fmt.chromaHeight(d.surface_height);
plane_ptr += pitch[i] * plane_h;
d.gpu_mem.copyFrame(src[i], dst[i], plane_w, plane_h, pitch[i]);
}
frame = VideoFrame(buf, d.width, d.height, fmt);
frame.setBits(dst);
frame.setBytesPerLine(pitch);
} else {
frame = VideoFrame(d.width, d.height, fmt);
frame.setBits(src);
frame.setBytesPerLine(pitch);
// TODO: why clone is faster()?
frame = frame.clone();
}
if ((status = vaUnmapBuffer(d.display, d.image.buf)) != VA_STATUS_SUCCESS) {
qWarning("vaUnmapBuffer(VADisplay:%p, VABufferID:%#x) == %#x", d.display, d.image.buf, status);
return VideoFrame();
}
if (!d.disable_derive && d.supports_derive) {
vaDestroyImage(d.display, d.image.image_id);
d.image.image_id = VA_INVALID_ID;
}
return frame;
}
struct display_names_t {
VideoDecoderVAAPI::DisplayType display;
QString name;
};
static const display_names_t display_names[] = {
{ VideoDecoderVAAPI::GLX, "GLX" },
{ VideoDecoderVAAPI::X11, "X11" },
{ VideoDecoderVAAPI::DRM, "DRM" }
};
static VideoDecoderVAAPI::DisplayType displayFromName(QString name) {
for (unsigned int i = 0; i < sizeof(display_names)/sizeof(display_names[0]); ++i) {
if (name.toUpper().contains(display_names[i].name.toUpper())) {
return display_names[i].display;
}
}
return VideoDecoderVAAPI::X11;
}
static QString displayToName(VideoDecoderVAAPI::DisplayType t) {
for (unsigned int i = 0; i < sizeof(display_names)/sizeof(display_names[0]); ++i) {
if (t == display_names[i].display) {
return display_names[i].name;
}
}
return QString();
}
void VideoDecoderVAAPI::setDisplayPriority(const QStringList &priority)
{
DPTR_D(VideoDecoderVAAPI);
d.display_priority.clear();
foreach (QString disp, priority) {
d.display_priority.push_back(displayFromName(disp));
}
void
GtkAggVaapiGlue::render()
{
VaapiGlobalContext * const gvactx = VaapiGlobalContext::get();
if (!gvactx)
return;
if (!_window_is_setup)
return;
if (!_vaapi_image.get() || !_vaapi_surface.get())
return;
if (!_vaapi_image->unmap()) {
printf("ERROR: failed to unmap VA-API image\n");
return;
}
VAStatus status;
status = vaPutSurface(gvactx->display(),
_vaapi_surface->get(),
GDK_DRAWABLE_XID(_drawing_area->window),
0, 0,
_vaapi_surface->width(),
_vaapi_surface->height(),
0, 0,
_window_width,
_window_height,
NULL, 0,
VA_FRAME_PICTURE);
if (!vaapi_check_status(status, "vaPutSurface() canvas"))
return;
Renderer_agg_base::RenderImages::const_iterator img, first_img, last_img;
first_img = _agg_renderer->getFirstRenderImage();
last_img = _agg_renderer->getLastRenderImage();
if (first_img != last_img) {
for (img = first_img; img != last_img; ++img) {
boost::shared_ptr<VaapiSurface> surface = (*img)->surface();
VaapiRectangle src_rect;
src_rect.x = (*img)->x();
src_rect.y = (*img)->y();
src_rect.width = (*img)->width();
src_rect.height = (*img)->height();
VaapiRectangle dst_rect;
const float xscale = _window_width / (float)_vaapi_image_width;
const float yscale = _window_height / (float)_vaapi_image_height;
dst_rect.x = src_rect.x * xscale;
dst_rect.y = src_rect.y * yscale;
dst_rect.width = src_rect.width * xscale;
dst_rect.height = src_rect.height * yscale;
VaapiVideoWindow *videoWindow;
videoWindow = getVideoWindow(surface, _drawing_area->window, dst_rect);
if (!videoWindow) {
log_debug(_("ERROR: failed to setup video window for surface 0x%08x."), surface->get());
continue;
}
videoWindow->moveResize(dst_rect);
VaapiRectangle pic_rect(surface->width(), surface->height());
if (!surface->associateSubpicture(_vaapi_subpicture, src_rect, pic_rect)) {
log_debug(_("ERROR: failed to associate subpicture to surface 0x%08x."), surface->get());
continue;
}
status = vaPutSurface(gvactx->display(),
surface->get(),
videoWindow->xid(),
0, 0, surface->width(), surface->height(),
0, 0, dst_rect.width, dst_rect.height,
NULL, 0,
VA_FRAME_PICTURE);
if (!vaapi_check_status(status, "vaPutSurface() video"))
continue;
surface->deassociateSubpicture(_vaapi_subpicture);
}
for (img = first_img; img != last_img; ++img) {
boost::shared_ptr<VaapiSurface> surface = (*img)->surface();
status = vaSyncSurface(gvactx->display(), surface->get());
if (!vaapi_check_status(status, "vaSyncSurface() video"))
continue;
}
}
}
static void h264_decode_frame(int f_width, int f_height, char *framedata, int framesize, int slice_type)
{
VAStatus va_status;
DebugLog(("%s: called for frame of %d bytes (%dx%d) slice_type=%d\n", __FUNCTION__, framesize, width, height, slice_type));
/* Initialize decode pipeline if necessary */
if ( (f_width > cur_width) || (f_height > cur_height) ) {
if (va_dpy != NULL)
h264_cleanup_decoder();
cur_width = f_width;
cur_height = f_height;
h264_init_decoder(f_width, f_height);
rfbClientLog("%s: decoder initialized\n", __FUNCTION__);
}
/* Decode frame */
static VAPictureH264 va_picture_h264, va_old_picture_h264;
/* The server should always send an I-frame when a new client connects
* or when the resolution of the framebuffer changes, but we check
* just in case.
*/
if ( (slice_type != SLICE_TYPE_I) && (num_frames == 0) ) {
rfbClientLog("First frame is not an I frame !!! Skipping!!!\n");
return;
}
DebugLog(("%s: frame_id=%d va_surface_id[%d]=0x%x field_order_count=%d\n", __FUNCTION__, frame_id, sid, va_surface_id[sid], field_order_count));
va_picture_h264.picture_id = va_surface_id[sid];
va_picture_h264.frame_idx = frame_id;
va_picture_h264.flags = 0;
va_picture_h264.BottomFieldOrderCnt = field_order_count;
va_picture_h264.TopFieldOrderCnt = field_order_count;
/* Set up picture parameter buffer */
if (va_pic_param_buf_id[sid] == VA_INVALID_ID) {
va_status = vaCreateBuffer(va_dpy, va_context_id, VAPictureParameterBufferType, sizeof(VAPictureParameterBufferH264), 1, NULL, &va_pic_param_buf_id[sid]);
CHECK_VASTATUS(va_status, "vaCreateBuffer(PicParam)");
}
CHECK_SURF(va_surface_id[sid]);
VAPictureParameterBufferH264 *pic_param_buf = NULL;
va_status = vaMapBuffer(va_dpy, va_pic_param_buf_id[sid], (void **)&pic_param_buf);
CHECK_VASTATUS(va_status, "vaMapBuffer(PicParam)");
SetVAPictureParameterBufferH264(pic_param_buf, f_width, f_height);
memcpy(&pic_param_buf->CurrPic, &va_picture_h264, sizeof(VAPictureH264));
if (slice_type == SLICE_TYPE_P) {
memcpy(&pic_param_buf->ReferenceFrames[0], &va_old_picture_h264, sizeof(VAPictureH264));
pic_param_buf->ReferenceFrames[0].flags = 0;
}
else if (slice_type != SLICE_TYPE_I) {
rfbClientLog("Frame type %d not supported!!!\n");
return;
}
pic_param_buf->frame_num = frame_id;
va_status = vaUnmapBuffer(va_dpy, va_pic_param_buf_id[sid]);
CHECK_VASTATUS(va_status, "vaUnmapBuffer(PicParam)");
/* Set up IQ matrix buffer */
if (va_mat_param_buf_id[sid] == VA_INVALID_ID) {
va_status = vaCreateBuffer(va_dpy, va_context_id, VAIQMatrixBufferType, sizeof(VAIQMatrixBufferH264), 1, NULL, &va_mat_param_buf_id[sid]);
CHECK_VASTATUS(va_status, "vaCreateBuffer(IQMatrix)");
}
CHECK_SURF(va_surface_id[sid]);
VAIQMatrixBufferH264 *iq_matrix_buf = NULL;
va_status = vaMapBuffer(va_dpy, va_mat_param_buf_id[sid], (void **)&iq_matrix_buf);
CHECK_VASTATUS(va_status, "vaMapBuffer(IQMatrix)");
static const unsigned char m_MatrixBufferH264[]= {
/* ScalingList4x4[6][16] */
0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,
0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,
0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,
0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,
0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,
0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,0x10,
/* ScalingList8x8[2][64] */
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00
};
memcpy(iq_matrix_buf, m_MatrixBufferH264, 224);
va_status = vaUnmapBuffer(va_dpy, va_mat_param_buf_id[sid]);
CHECK_VASTATUS(va_status, "vaUnmapBuffer(IQMatrix)");
VABufferID buffer_ids[2];
buffer_ids[0] = va_pic_param_buf_id[sid];
buffer_ids[1] = va_mat_param_buf_id[sid];
CHECK_SURF(va_surface_id[sid]);
va_status = vaRenderPicture(va_dpy, va_context_id, buffer_ids, 2);
CHECK_VASTATUS(va_status, "vaRenderPicture");
/* Set up slice parameter buffer */
if (va_sp_param_buf_id[sid] == VA_INVALID_ID) {
va_status = vaCreateBuffer(va_dpy, va_context_id, VASliceParameterBufferType, sizeof(VASliceParameterBufferH264), 1, NULL, &va_sp_param_buf_id[sid]);
CHECK_VASTATUS(va_status, "vaCreateBuffer(SliceParam)");
}
CHECK_SURF(va_surface_id[sid]);
VASliceParameterBufferH264 *slice_param_buf = NULL;
va_status = vaMapBuffer(va_dpy, va_sp_param_buf_id[sid], (void **)&slice_param_buf);
CHECK_VASTATUS(va_status, "vaMapBuffer(SliceParam)");
static int t2_first = 1;
if (slice_type == SLICE_TYPE_I) {
SetVASliceParameterBufferH264_Intra(slice_param_buf, t2_first);
t2_first = 0;
} else {
SetVASliceParameterBufferH264(slice_param_buf);
memcpy(&slice_param_buf->RefPicList0[0], &va_old_picture_h264, sizeof(VAPictureH264));
slice_param_buf->RefPicList0[0].flags = 0;
}
slice_param_buf->slice_data_bit_offset = 0;
slice_param_buf->slice_data_size = framesize;
va_status = vaUnmapBuffer(va_dpy, va_sp_param_buf_id[sid]);
CHECK_VASTATUS(va_status, "vaUnmapBuffer(SliceParam)");
CHECK_SURF(va_surface_id[sid]);
/* Set up slice data buffer and copy H.264 encoded data */
if (va_d_param_buf_id[sid] == VA_INVALID_ID) {
/* TODO use estimation matching framebuffer dimensions instead of this large value */
va_status = vaCreateBuffer(va_dpy, va_context_id, VASliceDataBufferType, 4177920, 1, NULL, &va_d_param_buf_id[sid]); /* 1080p size */
CHECK_VASTATUS(va_status, "vaCreateBuffer(SliceData)");
}
char *slice_data_buf;
va_status = vaMapBuffer(va_dpy, va_d_param_buf_id[sid], (void **)&slice_data_buf);
CHECK_VASTATUS(va_status, "vaMapBuffer(SliceData)");
memcpy(slice_data_buf, framedata, framesize);
CHECK_SURF(va_surface_id[sid]);
va_status = vaUnmapBuffer(va_dpy, va_d_param_buf_id[sid]);
CHECK_VASTATUS(va_status, "vaUnmapBuffer(SliceData)");
buffer_ids[0] = va_sp_param_buf_id[sid];
buffer_ids[1] = va_d_param_buf_id[sid];
CHECK_SURF(va_surface_id[sid]);
va_status = vaRenderPicture(va_dpy, va_context_id, buffer_ids, 2);
CHECK_VASTATUS(va_status, "vaRenderPicture");
va_status = vaEndPicture(va_dpy, va_context_id);
CHECK_VASTATUS(va_status, "vaEndPicture");
/* Prepare next one... */
int sid_new = (sid + 1) % SURFACE_NUM;
DebugLog(("%s: new Surface ID = %d\n", __FUNCTION__, sid_new));
va_status = vaBeginPicture(va_dpy, va_context_id, va_surface_id[sid_new]);
CHECK_VASTATUS(va_status, "vaBeginPicture");
/* Get decoded data */
va_status = vaSyncSurface(va_dpy, va_surface_id[sid]);
CHECK_VASTATUS(va_status, "vaSyncSurface");
CHECK_SURF(va_surface_id[sid]);
curr_surface = va_surface_id[sid];
sid = sid_new;
field_order_count += 2;
++frame_id;
if (frame_id > 15) {
frame_id = 0;
}
++num_frames;
memcpy(&va_old_picture_h264, &va_picture_h264, sizeof(VAPictureH264));
}
static int Extract( vlc_va_t *p_external, picture_t *p_picture, AVFrame *p_ff )
{
vlc_va_vaapi_t *p_va = vlc_va_vaapi_Get(p_external);
VASurfaceID i_surface_id = (VASurfaceID)(uintptr_t)p_ff->data[3];
#if VA_CHECK_VERSION(0,31,0)
if( vaSyncSurface( p_va->p_display, i_surface_id ) )
#else
if( vaSyncSurface( p_va->p_display, p_va->i_context_id, i_surface_id ) )
#endif
return VLC_EGENERIC;
if(p_va->b_supports_derive)
{
if(vaDeriveImage(p_va->p_display, i_surface_id, &(p_va->image)) != VA_STATUS_SUCCESS)
return VLC_EGENERIC;
}
else
{
if( vaGetImage( p_va->p_display, i_surface_id,
0, 0, p_va->i_surface_width, p_va->i_surface_height,
p_va->image.image_id) )
return VLC_EGENERIC;
}
void *p_base;
if( vaMapBuffer( p_va->p_display, p_va->image.buf, &p_base ) )
return VLC_EGENERIC;
const uint32_t i_fourcc = p_va->image.format.fourcc;
if( i_fourcc == VA_FOURCC('Y','V','1','2') ||
i_fourcc == VA_FOURCC('I','4','2','0') )
{
bool b_swap_uv = i_fourcc == VA_FOURCC('I','4','2','0');
uint8_t *pp_plane[3];
size_t pi_pitch[3];
for( int i = 0; i < 3; i++ )
{
const int i_src_plane = (b_swap_uv && i != 0) ? (3 - i) : i;
pp_plane[i] = (uint8_t*)p_base + p_va->image.offsets[i_src_plane];
pi_pitch[i] = p_va->image.pitches[i_src_plane];
}
CopyFromYv12( p_picture, pp_plane, pi_pitch,
p_va->i_surface_width,
p_va->i_surface_height,
&p_va->image_cache );
}
else
{
assert( i_fourcc == VA_FOURCC('N','V','1','2') );
uint8_t *pp_plane[2];
size_t pi_pitch[2];
for( int i = 0; i < 2; i++ )
{
pp_plane[i] = (uint8_t*)p_base + p_va->image.offsets[i];
pi_pitch[i] = p_va->image.pitches[i];
}
CopyFromNv12( p_picture, pp_plane, pi_pitch,
p_va->i_surface_width,
p_va->i_surface_height,
&p_va->image_cache );
}
if( vaUnmapBuffer( p_va->p_display, p_va->image.buf ) )
return VLC_EGENERIC;
if(p_va->b_supports_derive)
{
vaDestroyImage( p_va->p_display, p_va->image.image_id );
p_va->image.image_id = VA_INVALID_ID;
}
return VLC_SUCCESS;
}
bool VAAPIContext::CopySurfaceToFrame(VideoFrame *frame, const void *buf)
{
MythXLocker locker(m_display->m_x_disp);
if (!m_deriveSupport && m_image.image_id == VA_INVALID_ID)
InitImage(buf);
if (!frame || !buf || (m_dispType != kVADisplayX11) ||
(!m_deriveSupport && m_image.image_id == VA_INVALID_ID))
return false;
const vaapi_surface *surf = (vaapi_surface*)buf;
INIT_ST;
va_status = vaSyncSurface(m_ctx.display, surf->m_id);
CHECK_ST;
if (m_deriveSupport)
{
va_status = vaDeriveImage(m_ctx.display, surf->m_id, &m_image);
}
else
{
va_status = vaGetImage(m_ctx.display, surf->m_id, 0, 0,
m_size.width(), m_size.height(),
m_image.image_id);
}
CHECK_ST;
if (ok)
{
VideoFrame src;
void* source = NULL;
if (vaMapBuffer(m_ctx.display, m_image.buf, &source))
return false;
if (m_image.format.fourcc == VA_FOURCC_NV12)
{
init(&src, FMT_NV12, (unsigned char*)source, m_image.width,
m_image.height, m_image.data_size, NULL,
NULL, frame->aspect, frame->frame_rate);
for (int i = 0; i < 2; i++)
{
src.pitches[i] = m_image.pitches[i];
src.offsets[i] = m_image.offsets[i];
}
}
else
{
// Our VideoFrame YV12 format, is really YUV420P/IYUV
bool swap = m_image.format.fourcc == VA_FOURCC_YV12;
init(&src, FMT_YV12, (unsigned char*)source, m_image.width,
m_image.height, m_image.data_size, NULL,
NULL, frame->aspect, frame->frame_rate);
src.pitches[0] = m_image.pitches[0];
src.pitches[1] = m_image.pitches[swap ? 2 : 1];
src.pitches[2] = m_image.pitches[swap ? 1 : 2];
src.offsets[0] = m_image.offsets[0];
src.offsets[1] = m_image.offsets[swap ? 2 : 1];
src.offsets[2] = m_image.offsets[swap ? 1 : 2];
}
m_copy->copy(frame, &src);
if (vaUnmapBuffer(m_ctx.display, m_image.buf))
return false;
if (m_deriveSupport)
{
vaDestroyImage(m_ctx.display, m_image.image_id );
m_image.image_id = VA_INVALID_ID;
}
return true;
}
LOG(VB_GENERAL, LOG_ERR, LOC + "Failed to get image");
return false;
}
static int vaapi_map_frame(AVHWFramesContext *hwfc,
AVFrame *dst, const AVFrame *src, int flags)
{
AVVAAPIDeviceContext *hwctx = hwfc->device_ctx->hwctx;
VAAPIFramesContext *ctx = hwfc->internal->priv;
VASurfaceID surface_id;
VAImageFormat *image_format;
VAAPISurfaceMap *map;
VAStatus vas;
void *address = NULL;
int err, i;
surface_id = (VASurfaceID)(uintptr_t)src->data[3];
av_log(hwfc, AV_LOG_DEBUG, "Map surface %#x.\n", surface_id);
if (!ctx->derive_works && (flags & VAAPI_MAP_DIRECT)) {
// Requested direct mapping but it is not possible.
return AVERROR(EINVAL);
}
if (dst->format == AV_PIX_FMT_NONE)
dst->format = hwfc->sw_format;
if (dst->format != hwfc->sw_format && (flags & VAAPI_MAP_DIRECT)) {
// Requested direct mapping but the formats do not match.
return AVERROR(EINVAL);
}
err = vaapi_get_image_format(hwfc->device_ctx, dst->format, &image_format);
if (err < 0) {
// Requested format is not a valid output format.
return AVERROR(EINVAL);
}
map = av_malloc(sizeof(VAAPISurfaceMap));
if (!map)
return AVERROR(ENOMEM);
map->source = src;
map->flags = flags;
map->image.image_id = VA_INVALID_ID;
vas = vaSyncSurface(hwctx->display, surface_id);
if (vas != VA_STATUS_SUCCESS) {
av_log(hwfc, AV_LOG_ERROR, "Failed to sync surface "
"%#x: %d (%s).\n", surface_id, vas, vaErrorStr(vas));
err = AVERROR(EIO);
goto fail;
}
// The memory which we map using derive need not be connected to the CPU
// in a way conducive to fast access. On Gen7-Gen9 Intel graphics, the
// memory is mappable but not cached, so normal memcpy()-like access is
// very slow to read it (but writing is ok). It is possible to read much
// faster with a copy routine which is aware of the limitation, but we
// assume for now that the user is not aware of that and would therefore
// prefer not to be given direct-mapped memory if they request read access.
if (ctx->derive_works &&
((flags & VAAPI_MAP_DIRECT) || !(flags & VAAPI_MAP_READ))) {
vas = vaDeriveImage(hwctx->display, surface_id, &map->image);
if (vas != VA_STATUS_SUCCESS) {
av_log(hwfc, AV_LOG_ERROR, "Failed to derive image from "
"surface %#x: %d (%s).\n",
surface_id, vas, vaErrorStr(vas));
err = AVERROR(EIO);
goto fail;
}
if (map->image.format.fourcc != image_format->fourcc) {
av_log(hwfc, AV_LOG_ERROR, "Derive image of surface %#x "
"is in wrong format: expected %#08x, got %#08x.\n",
surface_id, image_format->fourcc, map->image.format.fourcc);
err = AVERROR(EIO);
goto fail;
}
map->flags |= VAAPI_MAP_DIRECT;
} else {
vas = vaCreateImage(hwctx->display, image_format,
hwfc->width, hwfc->height, &map->image);
if (vas != VA_STATUS_SUCCESS) {
av_log(hwfc, AV_LOG_ERROR, "Failed to create image for "
"surface %#x: %d (%s).\n",
surface_id, vas, vaErrorStr(vas));
err = AVERROR(EIO);
goto fail;
}
if (flags & VAAPI_MAP_READ) {
vas = vaGetImage(hwctx->display, surface_id, 0, 0,
hwfc->width, hwfc->height, map->image.image_id);
if (vas != VA_STATUS_SUCCESS) {
av_log(hwfc, AV_LOG_ERROR, "Failed to read image from "
"surface %#x: %d (%s).\n",
surface_id, vas, vaErrorStr(vas));
err = AVERROR(EIO);
goto fail;
}
}
}
vas = vaMapBuffer(hwctx->display, map->image.buf, &address);
if (vas != VA_STATUS_SUCCESS) {
av_log(hwfc, AV_LOG_ERROR, "Failed to map image from surface "
"%#x: %d (%s).\n", surface_id, vas, vaErrorStr(vas));
err = AVERROR(EIO);
goto fail;
}
dst->width = src->width;
dst->height = src->height;
for (i = 0; i < map->image.num_planes; i++) {
dst->data[i] = (uint8_t*)address + map->image.offsets[i];
dst->linesize[i] = map->image.pitches[i];
}
if (
#ifdef VA_FOURCC_YV16
map->image.format.fourcc == VA_FOURCC_YV16 ||
#endif
map->image.format.fourcc == VA_FOURCC_YV12) {
// Chroma planes are YVU rather than YUV, so swap them.
FFSWAP(uint8_t*, dst->data[1], dst->data[2]);
}
static int Extract( vlc_va_t *va, picture_t *p_picture, uint8_t *data )
{
vlc_va_sys_t *sys = va->sys;
VASurfaceID surface = (VASurfaceID)(uintptr_t)data;
VAImage image;
int ret = VLC_EGENERIC;
#if VA_CHECK_VERSION(0,31,0)
if (vaSyncSurface(sys->hw_ctx.display, surface))
#else
if (vaSyncSurface(sys->hw_ctx.display, sys->hw_ctx.context_id, surface))
#endif
return VLC_EGENERIC;
if (!sys->do_derive || vaDeriveImage(sys->hw_ctx.display, surface, &image))
{ /* Fallback if image derivation is not supported */
if (vaCreateImage(sys->hw_ctx.display, &sys->format, sys->width,
sys->height, &image))
return VLC_EGENERIC;
if (vaGetImage(sys->hw_ctx.display, surface, 0, 0, sys->width,
sys->height, image.image_id))
goto error;
}
void *p_base;
if (vaMapBuffer(sys->hw_ctx.display, image.buf, &p_base))
goto error;
const unsigned i_fourcc = sys->format.fourcc;
if( i_fourcc == VA_FOURCC_YV12 ||
i_fourcc == VA_FOURCC_IYUV )
{
bool b_swap_uv = i_fourcc == VA_FOURCC_IYUV;
uint8_t *pp_plane[3];
size_t pi_pitch[3];
for( int i = 0; i < 3; i++ )
{
const int i_src_plane = (b_swap_uv && i != 0) ? (3 - i) : i;
pp_plane[i] = (uint8_t*)p_base + image.offsets[i_src_plane];
pi_pitch[i] = image.pitches[i_src_plane];
}
CopyFromYv12( p_picture, pp_plane, pi_pitch, sys->width, sys->height,
&sys->image_cache );
}
else
{
assert( i_fourcc == VA_FOURCC_NV12 );
uint8_t *pp_plane[2];
size_t pi_pitch[2];
for( int i = 0; i < 2; i++ )
{
pp_plane[i] = (uint8_t*)p_base + image.offsets[i];
pi_pitch[i] = image.pitches[i];
}
CopyFromNv12( p_picture, pp_plane, pi_pitch, sys->width, sys->height,
&sys->image_cache );
}
vaUnmapBuffer(sys->hw_ctx.display, image.buf);
ret = VLC_SUCCESS;
error:
vaDestroyImage(sys->hw_ctx.display, image.image_id);
return ret;
}
VideoFrame VideoDecoderVAAPI::frame()
{
DPTR_D(VideoDecoderVAAPI);
if (!d.frame->opaque || !d.frame->data[0])
return VideoFrame();
VASurfaceID surface_id = (VASurfaceID)(uintptr_t)d.frame->data[3];
VAStatus status = VA_STATUS_SUCCESS;
if (display() == GLX || (copyMode() == ZeroCopy && display() == X11)) {
surface_ptr p;
std::list<surface_ptr>::iterator it = d.surfaces_used.begin();
for (; it != d.surfaces_used.end() && !p; ++it) {
if((*it)->get() == surface_id) {
p = *it;
break;
}
}
if (!p) {
for (it = d.surfaces_free.begin(); it != d.surfaces_free.end() && !p; ++it) {
if((*it)->get() == surface_id) {
p = *it;
break;
}
}
}
if (!p) {
qWarning("VAAPI - Unable to find surface");
return VideoFrame();
}
((SurfaceInteropVAAPI*)d.surface_interop.data())->setSurface(p);
VideoFrame f(d.width, d.height, VideoFormat::Format_RGB32); //p->width()
f.setBytesPerLine(d.width*4); //used by gl to compute texture size
f.setMetaData("surface_interop", QVariant::fromValue(d.surface_interop));
f.setTimestamp(double(d.frame->pkt_pts)/1000.0);
return f;
}
#if VA_CHECK_VERSION(0,31,0)
if ((status = vaSyncSurface(d.display->get(), surface_id)) != VA_STATUS_SUCCESS) {
qWarning("vaSyncSurface(VADisplay:%p, VASurfaceID:%#x) == %#x", d.display->get(), surface_id, status);
#else
if (vaSyncSurface(d.display->get(), d.context_id, surface_id)) {
qWarning("vaSyncSurface(VADisplay:%#x, VAContextID:%#x, VASurfaceID:%#x) == %#x", d.display, d.context_id, surface_id, status);
#endif
return VideoFrame();
}
if (!d.disable_derive && d.supports_derive) {
/*
* http://web.archiveorange.com/archive/v/OAywENyq88L319OcRnHI
* vaDeriveImage is faster than vaGetImage. But VAImage is uncached memory and copying from it would be terribly slow
* TODO: copy from USWC, see vlc and https://github.com/OpenELEC/OpenELEC.tv/pull/2937.diff
* https://software.intel.com/en-us/articles/increasing-memory-throughput-with-intel-streaming-simd-extensions-4-intel-sse4-streaming-load
*/
VA_ENSURE_TRUE(vaDeriveImage(d.display->get(), surface_id, &d.image), VideoFrame());
} else {
VA_ENSURE_TRUE(vaGetImage(d.display->get(), surface_id, 0, 0, d.width, d.height, d.image.image_id), VideoFrame());
}
void *p_base;
VA_ENSURE_TRUE(vaMapBuffer(d.display->get(), d.image.buf, &p_base), VideoFrame());
VideoFormat::PixelFormat pixfmt = VideoFormat::Format_Invalid;
bool swap_uv = false;
switch (d.image.format.fourcc) {
case VA_FOURCC_YV12:
swap_uv |= d.disable_derive || !d.supports_derive;
pixfmt = VideoFormat::Format_YUV420P;
break;
case VA_FOURCC_IYUV:
swap_uv = true;
pixfmt = VideoFormat::Format_YUV420P;
break;
case VA_FOURCC_NV12:
pixfmt = VideoFormat::Format_NV12;
break;
default:
break;
}
if (pixfmt == VideoFormat::Format_Invalid) {
qWarning("unsupported vaapi pixel format: %#x", d.image.format.fourcc);
return VideoFrame();
}
const VideoFormat fmt(pixfmt);
uint8_t *src[3];
int pitch[3];
for (int i = 0; i < fmt.planeCount(); ++i) {
src[i] = (uint8_t*)p_base + d.image.offsets[i];
pitch[i] = d.image.pitches[i];
}
VideoFrame frame(copyToFrame(fmt, d.surface_height, src, pitch, swap_uv));
VAWARN(vaUnmapBuffer(d.display->get(), d.image.buf));
if (!d.disable_derive && d.supports_derive) {
vaDestroyImage(d.display->get(), d.image.image_id);
d.image.image_id = VA_INVALID_ID;
}
return frame;
}
void VideoDecoderVAAPI::setDisplayPriority(const QStringList &priority)
{
DPTR_D(VideoDecoderVAAPI);
d.display_priority.clear();
int idx = staticMetaObject.indexOfEnumerator("DisplayType");
const QMetaEnum me = staticMetaObject.enumerator(idx);
foreach (const QString& disp, priority) {
d.display_priority.push_back((DisplayType)me.keyToValue(disp.toUtf8().constData()));
}