void test()
{
VAEntrypoint entrypoints[5];
int num_entrypoints,slice_entrypoint;
VAConfigAttrib attrib[2];
VAConfigID config_id;
unsigned int fourcc, luma_stride, chroma_u_stride, chroma_v_stride, luma_offset, chroma_u_offset;
unsigned int chroma_v_offset, buffer_name;
unsigned int *p_buffer;
int i, ret;
char c;
int frame_width=640, frame_height=480;
VASurfaceID surface_id;
VAImage image_id;
unsigned char *usrbuf;
usrbuf = (unsigned char*)malloc(frame_width * frame_height * 2);
ASSERT ( usrbuf != NULL);
VASurfaceAttribExternalBuffers vaSurfaceExternBuf;
VASurfaceAttrib attrib_list[2];
va_status = vaQueryConfigEntrypoints(va_dpy, VAProfileH264Baseline, entrypoints, &num_entrypoints);
ASSERT( VA_STATUS_SUCCESS == va_status );
for (slice_entrypoint = 0; slice_entrypoint < num_entrypoints; slice_entrypoint++) {
if (entrypoints[slice_entrypoint] == VAEntrypointEncSlice)
break;
}
if (slice_entrypoint == num_entrypoints) {
/* not find Slice entry point */
status("VAEntrypointEncSlice doesn't support, exit.\n");
ASSERT(0);
}
/* find out the format for the render target, and rate control mode */
attrib[0].type = VAConfigAttribRTFormat;
attrib[1].type = VAConfigAttribRateControl;
va_status = vaGetConfigAttributes(va_dpy, VAProfileH264Baseline, VAEntrypointEncSlice, &attrib[0], 2);
ASSERT( VA_STATUS_SUCCESS == va_status );
if ((attrib[0].value & VA_RT_FORMAT_YUV420) == 0) {
/* not find desired YUV420 RT format */
status("VA_RT_FORMAT_YUV420 doesn't support, exit\n");
ASSERT(0);
}
if ((attrib[1].value & VA_RC_VBR) == 0) {
/* Can't find matched RC mode */
status("VBR mode doesn't found, exit\n");
ASSERT(0);
}
attrib[0].value = VA_RT_FORMAT_YUV420; /* set to desired RT format */
attrib[1].value = VA_RC_VBR; /* set to desired RC mode */
va_status = vaCreateConfig(va_dpy, VAProfileH264Baseline, VAEntrypointEncSlice, &attrib[0], 2, &config_id);
ASSERT( VA_STATUS_SUCCESS == va_status );
attrib_list[1].type = (VASurfaceAttribType)VASurfaceAttribExternalBufferDescriptor;
attrib_list[0].type = (VASurfaceAttribType)VASurfaceAttribMemoryType;
va_status = vaGetSurfaceAttributes(va_dpy, config_id, attrib_list, 2);
ASSERT( VA_STATUS_SUCCESS == va_status );
if (attrib_list[0].flags != VA_SURFACE_ATTRIB_NOT_SUPPORTED) {
status("supported memory type:\n");
if (attrib_list[0].value.value.i & VA_SURFACE_ATTRIB_MEM_TYPE_VA)
status("\tVA_SURFACE_ATTRIB_MEM_TYPE_VA\n");
if (attrib_list[0].value.value.i & VA_SURFACE_ATTRIB_MEM_TYPE_V4L2)
status("\tVA_SURFACE_ATTRIB_MEM_TYPE_V4L2\n");
if (attrib_list[0].value.value.i & VA_SURFACE_ATTRIB_MEM_TYPE_USER_PTR)
status("\tVA_SURFACE_ATTRIB_MEM_TYPE_USER_PTR\n");
if (attrib_list[0].value.value.i & VA_SURFACE_ATTRIB_MEM_TYPE_ANDROID_GRALLOC)
status("\tVA_SURFACE_ATTRIB_MEM_TYPE_ANDROID_GRALLOC\n");
if (attrib_list[0].value.value.i & VA_SURFACE_ATTRIB_MEM_TYPE_ANDROID_ION)
status("\tVA_SURFACE_ATTRIB_MEM_TYPE_ANDROID_ION\n");
if (attrib_list[0].value.value.i & VA_SURFACE_ATTRIB_MEM_TYPE_KERNEL_DRM)
status("\tVA_SURFACE_ATTRIB_MEM_TYPE_KERNEL_DRM\n");
}
if ((attrib_list[1].flags != VA_SURFACE_ATTRIB_NOT_SUPPORTED) &&
(attrib_list[0].value.value.i & VA_SURFACE_ATTRIB_MEM_TYPE_USER_PTR)) {
status("vaCreateSurfaces from external usr pointer\n");
vaSurfaceExternBuf.buffers = (unsigned long*)malloc(sizeof(unsigned int));
vaSurfaceExternBuf.buffers[0] = usrbuf;
vaSurfaceExternBuf.num_buffers = 1;
vaSurfaceExternBuf.width = frame_width;
vaSurfaceExternBuf.height = frame_height;
vaSurfaceExternBuf.pitches[0] = vaSurfaceExternBuf.pitches[1] = vaSurfaceExternBuf.pitches[2] = frame_width;
//vaSurfaceExternBuf.flags = VA_SURFACE_ATTRIB_MEM_TYPE_ANDROID_GRALLOC;
vaSurfaceExternBuf.pixel_format = VA_FOURCC_NV12;
//vaSurfaceExternBuf.pitches[0] = attribute_tpi->luma_stride;
attrib_list[1].flags = VA_SURFACE_ATTRIB_SETTABLE;
attrib_list[1].value.type = VAGenericValueTypePointer;
attrib_list[1].value.value.p = (void *)&vaSurfaceExternBuf;
attrib_list[0].flags = VA_SURFACE_ATTRIB_SETTABLE;
attrib_list[0].value.type = VAGenericValueTypeInteger;
attrib_list[0].value.value.i = VA_SURFACE_ATTRIB_MEM_TYPE_USER_PTR;
va_status = vaCreateSurfaces(va_dpy, VA_RT_FORMAT_YUV420, frame_width, frame_height, &surface_id, 1, attrib_list, 2);
ASSERT( VA_STATUS_SUCCESS == va_status );
va_status = vaDeriveImage(va_dpy, surface_id, &image_id);
ASSERT( VA_STATUS_SUCCESS == va_status );
va_status = vaMapBuffer(va_dpy, image_id.buf, (void**)&p_buffer);
ASSERT( VA_STATUS_SUCCESS == va_status );
memset(p_buffer, 0x80, image_id.width * image_id.height);
va_status = vaUnmapBuffer(va_dpy, image_id.buf);
ASSERT( VA_STATUS_SUCCESS == va_status );
va_status = vaDestroyImage(va_dpy, image_id.image_id);
ASSERT( VA_STATUS_SUCCESS == va_status );
va_status = vaDestroySurfaces(va_dpy, &surface_id, 1);
ASSERT( VA_STATUS_SUCCESS == va_status );
}
va_status = vaDestroyConfig(va_dpy, config_id);
ASSERT( VA_STATUS_SUCCESS == va_status );
free(usrbuf);
}
ImageRawPtr VaapiImageRaw::create(const DisplayPtr& display, const ImagePtr& image, VideoDataMemoryType memoryType)
{
ImageRawPtr raw;
RealeaseCallback release;
uintptr_t handle;
VAStatus status;
VAImagePtr& vaImage = image->m_image;
if (memoryType == VIDEO_DATA_MEMORY_TYPE_RAW_POINTER || memoryType == VIDEO_DATA_MEMORY_TYPE_RAW_COPY) {
void* data;
status = vaMapBuffer(display->getID(), vaImage->buf, &data);
release = vaUnmapBuffer;
handle = (uintptr_t)data;
} else {
VABufferInfo bufferInfo;
if (memoryType == VIDEO_DATA_MEMORY_TYPE_DRM_NAME)
bufferInfo.mem_type = VA_SURFACE_ATTRIB_MEM_TYPE_KERNEL_DRM;
else if (memoryType == VIDEO_DATA_MEMORY_TYPE_DMA_BUF)
bufferInfo.mem_type = VA_SURFACE_ATTRIB_MEM_TYPE_DRM_PRIME;
else
ASSERT(0);
status = vaAcquireBufferHandle(display->getID(), vaImage->buf, &bufferInfo);
release = vaReleaseBufferHandle;
handle = (uintptr_t)bufferInfo.handle;
}
if (!checkVaapiStatus(status, "VaapiImageRaw::create()"))
return raw;
raw.reset(new VaapiImageRaw(display, image, memoryType, handle, release));
return raw;
}
/* Maps VA buffer */
gpointer
vaapi_map_buffer (VADisplay dpy, VABufferID buf_id)
{
VAStatus status;
gpointer data = NULL;
status = vaMapBuffer (dpy, buf_id, &data);
if (!vaapi_check_status (status, "vaMapBuffer()"))
return NULL;
return data;
}
static void *alloc_buffer(struct vaapi_context *vactx, int type, unsigned int size, uint32_t *buf_id)
{
void *data = NULL;
*buf_id = 0;
if (vaCreateBuffer(vactx->display, vactx->context_id,
type, size, 1, NULL, buf_id) == VA_STATUS_SUCCESS)
vaMapBuffer(vactx->display, *buf_id, &data);
return data;
}
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;
}
}
}
/* Maps VA buffer */
void *vaapiMapBuffer(VADisplay dpy, VABufferID bufId)
{
VAStatus status;
void *data = NULL;
status = vaMapBuffer(dpy, bufId, &data);
if (!checkVaapiStatus(status, "vaMapBuffer()")) {
ERROR("fail to map bufId = %x", bufId);
return NULL;
}
return data;
}
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;
}
VaapiImageRaw *VaapiImage::map(VideoDataMemoryType memoryType)
{
uint32_t i;
void *data;
VAStatus status;
if (m_isMapped) {
return &m_rawImage;
}
m_rawImage.format = m_image->format.fourcc;
m_rawImage.width = m_image->width;
m_rawImage.height = m_image->height;
m_rawImage.numPlanes = m_image->num_planes;
m_rawImage.size = m_image->data_size;
for (i = 0; i < m_image->num_planes; i++) {
m_rawImage.strides[i] = m_image->pitches[i];
}
if (memoryType == VIDEO_DATA_MEMORY_TYPE_RAW_POINTER || memoryType == VIDEO_DATA_MEMORY_TYPE_RAW_COPY) {
status = vaMapBuffer(m_display->getID(), m_image->buf, &data);
if (!checkVaapiStatus(status, "vaMapBuffer()"))
return NULL;
for (i = 0; i < m_image->num_planes; i++) {
m_rawImage.handles[i] = (intptr_t) data + m_image->offsets[i];
}
} else {
VABufferInfo bufferInfo;
if (memoryType == VIDEO_DATA_MEMORY_TYPE_DRM_NAME)
bufferInfo.mem_type = VA_SURFACE_ATTRIB_MEM_TYPE_KERNEL_DRM;
else if (memoryType == VIDEO_DATA_MEMORY_TYPE_DMA_BUF)
bufferInfo.mem_type = VA_SURFACE_ATTRIB_MEM_TYPE_DRM_PRIME;
else
ASSERT(0);
status = vaAcquireBufferHandle(m_display->getID(), m_image->buf, &bufferInfo);
if (!checkVaapiStatus(status, "vaAcquireBufferHandle()"))
return NULL;
for (i = 0; i < m_image->num_planes; i++) {
m_rawImage.handles[i] = (intptr_t) bufferInfo.handle;
}
}
m_rawImage.memoryType = memoryType;
m_isMapped = true;
return &m_rawImage;
}
static void *alloc_buffer(VAAPIContext *vaapi, int type, unsigned int size, VABufferID *buf_id)
{
VAStatus status;
void *data = NULL;
*buf_id = 0;
status = vaCreateBuffer(vaapi->display, vaapi->context_id,
type, size, 1, NULL, buf_id);
if (!vaapi_check_status(status, "vaCreateBuffer()"))
return NULL;
status = vaMapBuffer(vaapi->display, *buf_id, &data);
if (!vaapi_check_status(status, "vaMapBuffer()"))
return NULL;
return data;
}
Encode_Status VideoEncoderH263::renderSliceParams(EncodeTask *task) {
VAStatus vaStatus = VA_STATUS_SUCCESS;
uint32_t sliceHeight;
uint32_t sliceHeightInMB;
LOG_V("Begin\n\n");
sliceHeight = mComParams.resolution.height;
sliceHeight += 15;
sliceHeight &= (~15);
sliceHeightInMB = sliceHeight / 16;
vaStatus = vaCreateBuffer(
mVADisplay, mVAContext,
VAEncSliceParameterBufferType,
sizeof(VAEncSliceParameterBuffer),
1, NULL, &mSliceParamBuf);
CHECK_VA_STATUS_RETURN("vaCreateBuffer");
VAEncSliceParameterBuffer *sliceParams;
vaStatus = vaMapBuffer(mVADisplay, mSliceParamBuf, (void **)&sliceParams);
CHECK_VA_STATUS_RETURN("vaMapBuffer");
// starting MB row number for this slice
sliceParams->start_row_number = 0;
// slice height measured in MB
sliceParams->slice_height = sliceHeightInMB;
sliceParams->slice_flags.bits.is_intra = (task->type == FTYPE_I)?1:0;
sliceParams->slice_flags.bits.disable_deblocking_filter_idc = 0;
LOG_V("======h263 slice params======\n");
LOG_I("start_row_number = %d\n", (int) sliceParams->start_row_number);
LOG_I("slice_height_in_mb = %d\n", (int) sliceParams->slice_height);
LOG_I("slice.is_intra = %d\n", (int) sliceParams->slice_flags.bits.is_intra);
vaStatus = vaUnmapBuffer(mVADisplay, mSliceParamBuf);
CHECK_VA_STATUS_RETURN("vaUnmapBuffer");
vaStatus = vaRenderPicture(mVADisplay, mVAContext, &mSliceParamBuf, 1);
CHECK_VA_STATUS_RETURN("vaRenderPicture");
LOG_V("end\n");
return ENCODE_SUCCESS;
}
gboolean
_gst_vaapi_image_map(GstVaapiImage *image, GstVaapiImageRaw *raw_image)
{
GstVaapiImagePrivate * const priv = image->priv;
GstVaapiDisplay *display;
void *image_data;
VAStatus status;
guint i;
if (_gst_vaapi_image_is_mapped(image))
return TRUE;
display = GST_VAAPI_OBJECT_DISPLAY(image);
if (!display)
return FALSE;
GST_VAAPI_DISPLAY_LOCK(display);
status = vaMapBuffer(
GST_VAAPI_DISPLAY_VADISPLAY(display),
image->priv->image.buf,
&image_data
);
GST_VAAPI_DISPLAY_UNLOCK(display);
if (!vaapi_check_status(status, "vaMapBuffer()"))
return FALSE;
image->priv->image_data = image_data;
if (raw_image) {
const VAImage * const va_image = &priv->image;
raw_image->format = priv->format;
raw_image->width = va_image->width;
raw_image->height = va_image->height;
raw_image->num_planes = va_image->num_planes;
for (i = 0; i < raw_image->num_planes; i++) {
raw_image->pixels[i] = (guchar *)image_data + va_image->offsets[i];
raw_image->stride[i] = va_image->pitches[i];
}
}
return TRUE;
}
static int bind_image(VAImage *va_image, Image *image)
{
VAAPIContext * const vaapi = vaapi_get_context();
VAImageFormat * const va_format = &va_image->format;
VAStatus status;
void *va_image_data;
unsigned int i;
if (va_image->num_planes > MAX_IMAGE_PLANES)
return -1;
status = vaMapBuffer(vaapi->display, va_image->buf, &va_image_data);
if (!vaapi_check_status(status, "vaMapBuffer()"))
return -1;
memset(image, 0, sizeof(*image));
image->format = va_format->fourcc;
if (is_vaapi_rgb_format(va_format)) {
image->format = image_rgba_format(
va_format->bits_per_pixel,
va_format->byte_order == VA_MSB_FIRST,
va_format->red_mask,
va_format->green_mask,
va_format->blue_mask,
va_format->alpha_mask
);
if (!image->format)
return -1;
}
image->width = va_image->width;
image->height = va_image->height;
image->num_planes = va_image->num_planes;
for (i = 0; i < va_image->num_planes; i++) {
image->pixels[i] = (uint8_t *)va_image_data + va_image->offsets[i];
image->pitches[i] = va_image->pitches[i];
}
return 0;
}
void* SurfaceInteropVAAPI::mapToHost(const VideoFormat &format, void *handle, int plane)
{
Q_UNUSED(plane);
VAImage image;
static const unsigned int fcc[] = { VA_FOURCC_NV12, VA_FOURCC_YV12, VA_FOURCC_IYUV, 0};
va_new_image(m_surface->vadisplay(), fcc, &image, m_surface->width(), m_surface->height());
if (image.image_id == VA_INVALID_ID)
return NULL;
void *p_base;
VA_ENSURE(vaGetImage(m_surface->vadisplay(), m_surface->get(), 0, 0, m_surface->width(), m_surface->height(), image.image_id), NULL);
VA_ENSURE(vaMapBuffer(m_surface->vadisplay(), image.buf, &p_base), NULL); //TODO: destroy image before return
VideoFormat::PixelFormat pixfmt = pixelFormatFromVA(image.format.fourcc);
bool swap_uv = image.format.fourcc != VA_FOURCC_NV12;
if (pixfmt == VideoFormat::Format_Invalid) {
qWarning("unsupported vaapi pixel format: %#x", image.format.fourcc);
VA_ENSURE(vaDestroyImage(m_surface->vadisplay(), image.image_id), NULL);
return NULL;
}
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 + image.offsets[i];
pitch[i] = image.pitches[i];
}
VideoFrame frame = VideoFrame::fromGPU(fmt, frame_width, frame_height, m_surface->height(), src, pitch, true, swap_uv);
if (format != fmt)
frame = frame.to(format);
VAWARN(vaUnmapBuffer(m_surface->vadisplay(), image.buf));
VAWARN(vaDestroyImage(m_surface->vadisplay(), image.image_id));
image.image_id = VA_INVALID_ID;
VideoFrame *f = reinterpret_cast<VideoFrame*>(handle);
frame.setTimestamp(f->timestamp());
*f = frame;
return f;
}
static bool map(VABufferID id, T *&buf) {
return id != VA_INVALID_ID && vaMapBuffer(VaApi::glx(), id, (void**)&buf) == VA_STATUS_SUCCESS;
}
mfxStatus vaapiFrameAllocator::LockFrame(mfxMemId mid, mfxFrameData *ptr)
{
mfxStatus mfx_res = MFX_ERR_NONE;
VAStatus va_res = VA_STATUS_SUCCESS;
vaapiMemId* vaapi_mid = (vaapiMemId*)mid;
mfxU8* pBuffer = 0;
if (!vaapi_mid || !(vaapi_mid->m_surface)) return MFX_ERR_INVALID_HANDLE;
if (MFX_FOURCC_P8 == vaapi_mid->m_fourcc) // bitstream processing
{
VACodedBufferSegment *coded_buffer_segment;
va_res = vaMapBuffer(m_dpy, *(vaapi_mid->m_surface), (void **)(&coded_buffer_segment));
mfx_res = va_to_mfx_status(va_res);
ptr->Y = (mfxU8*)coded_buffer_segment->buf; // !!! bug
}
else // Image processing
{
va_res = vaDeriveImage(m_dpy, *(vaapi_mid->m_surface), &(vaapi_mid->m_image));
mfx_res = va_to_mfx_status(va_res);
if (MFX_ERR_NONE == mfx_res)
{
va_res = vaMapBuffer(m_dpy, vaapi_mid->m_image.buf, (void **) &pBuffer);
mfx_res = va_to_mfx_status(va_res);
}
if (MFX_ERR_NONE == mfx_res)
{
switch (vaapi_mid->m_image.format.fourcc)
{
case VA_FOURCC_NV12:
if (vaapi_mid->m_fourcc == MFX_FOURCC_NV12)
{
ptr->Pitch = (mfxU16)vaapi_mid->m_image.pitches[0];
ptr->Y = pBuffer + vaapi_mid->m_image.offsets[0];
ptr->U = pBuffer + vaapi_mid->m_image.offsets[1];
ptr->V = ptr->U + 1;
}
else mfx_res = MFX_ERR_LOCK_MEMORY;
break;
case VA_FOURCC_YV12:
if (vaapi_mid->m_fourcc == MFX_FOURCC_YV12)
{
ptr->Pitch = (mfxU16)vaapi_mid->m_image.pitches[0];
ptr->Y = pBuffer + vaapi_mid->m_image.offsets[0];
ptr->V = pBuffer + vaapi_mid->m_image.offsets[1];
ptr->U = pBuffer + vaapi_mid->m_image.offsets[2];
}
else mfx_res = MFX_ERR_LOCK_MEMORY;
break;
case VA_FOURCC_YUY2:
if (vaapi_mid->m_fourcc == MFX_FOURCC_YUY2)
{
ptr->Pitch = (mfxU16)vaapi_mid->m_image.pitches[0];
ptr->Y = pBuffer + vaapi_mid->m_image.offsets[0];
ptr->U = ptr->Y + 1;
ptr->V = ptr->Y + 3;
}
else mfx_res = MFX_ERR_LOCK_MEMORY;
break;
case VA_FOURCC_ARGB:
if (vaapi_mid->m_fourcc == MFX_FOURCC_RGB4)
{
ptr->Pitch = (mfxU16)vaapi_mid->m_image.pitches[0];
ptr->B = pBuffer + vaapi_mid->m_image.offsets[0];
ptr->G = ptr->B + 1;
ptr->R = ptr->B + 2;
ptr->A = ptr->B + 3;
}
else mfx_res = MFX_ERR_LOCK_MEMORY;
break;
default:
mfx_res = MFX_ERR_LOCK_MEMORY;
break;
}
}
}
return mfx_res;
}
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_encode_output(AVCodecContext *avctx,
VAAPIEncodePicture *pic, AVPacket *pkt)
{
VAAPIEncodeContext *ctx = avctx->priv_data;
VACodedBufferSegment *buf_list, *buf;
VAStatus vas;
int err;
err = vaapi_encode_wait(avctx, pic);
if (err < 0)
return err;
buf_list = NULL;
vas = vaMapBuffer(ctx->hwctx->display, pic->output_buffer,
(void**)&buf_list);
if (vas != VA_STATUS_SUCCESS) {
av_log(avctx, AV_LOG_ERROR, "Failed to map output buffers: "
"%d (%s).\n", vas, vaErrorStr(vas));
err = AVERROR(EIO);
goto fail;
}
for (buf = buf_list; buf; buf = buf->next) {
av_log(avctx, AV_LOG_DEBUG, "Output buffer: %u bytes "
"(status %08x).\n", buf->size, buf->status);
err = av_new_packet(pkt, buf->size);
if (err < 0)
goto fail;
memcpy(pkt->data, buf->buf, buf->size);
}
if (pic->type == PICTURE_TYPE_IDR)
pkt->flags |= AV_PKT_FLAG_KEY;
pkt->pts = pic->pts;
vas = vaUnmapBuffer(ctx->hwctx->display, pic->output_buffer);
if (vas != VA_STATUS_SUCCESS) {
av_log(avctx, AV_LOG_ERROR, "Failed to unmap output buffers: "
"%d (%s).\n", vas, vaErrorStr(vas));
err = AVERROR(EIO);
goto fail;
}
vaDestroyBuffer(ctx->hwctx->display, pic->output_buffer);
pic->output_buffer = VA_INVALID_ID;
av_log(avctx, AV_LOG_DEBUG, "Output read for pic %"PRId64"/%"PRId64".\n",
pic->display_order, pic->encode_order);
return 0;
fail:
if (pic->output_buffer != VA_INVALID_ID) {
vaUnmapBuffer(ctx->hwctx->display, pic->output_buffer);
vaDestroyBuffer(ctx->hwctx->display, pic->output_buffer);
pic->output_buffer = VA_INVALID_ID;
}
return err;
}
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()));
}
static int deint_vaapi_filter_frame(AVFilterLink *inlink, AVFrame *input_frame)
{
AVFilterContext *avctx = inlink->dst;
AVFilterLink *outlink = avctx->outputs[0];
DeintVAAPIContext *ctx = avctx->priv;
AVFrame *output_frame = NULL;
VASurfaceID input_surface, output_surface;
VASurfaceID backward_references[MAX_REFERENCES];
VASurfaceID forward_references[MAX_REFERENCES];
VAProcPipelineParameterBuffer params;
VAProcFilterParameterBufferDeinterlacing *filter_params;
VARectangle input_region;
VABufferID params_id;
VAStatus vas;
void *filter_params_addr = NULL;
int err, i, field, current_frame_index;
av_log(avctx, AV_LOG_DEBUG, "Filter input: %s, %ux%u (%"PRId64").\n",
av_get_pix_fmt_name(input_frame->format),
input_frame->width, input_frame->height, input_frame->pts);
if (ctx->queue_count < ctx->queue_depth) {
ctx->frame_queue[ctx->queue_count++] = input_frame;
if (ctx->queue_count < ctx->queue_depth) {
// Need more reference surfaces before we can continue.
return 0;
}
} else {
av_frame_free(&ctx->frame_queue[0]);
for (i = 0; i + 1 < ctx->queue_count; i++)
ctx->frame_queue[i] = ctx->frame_queue[i + 1];
ctx->frame_queue[i] = input_frame;
}
current_frame_index = ctx->pipeline_caps.num_forward_references;
input_frame = ctx->frame_queue[current_frame_index];
input_surface = (VASurfaceID)(uintptr_t)input_frame->data[3];
for (i = 0; i < ctx->pipeline_caps.num_forward_references; i++)
forward_references[i] = (VASurfaceID)(uintptr_t)
ctx->frame_queue[current_frame_index - i - 1]->data[3];
for (i = 0; i < ctx->pipeline_caps.num_backward_references; i++)
backward_references[i] = (VASurfaceID)(uintptr_t)
ctx->frame_queue[current_frame_index + i + 1]->data[3];
av_log(avctx, AV_LOG_DEBUG, "Using surface %#x for "
"deinterlace input.\n", input_surface);
av_log(avctx, AV_LOG_DEBUG, "Backward references:");
for (i = 0; i < ctx->pipeline_caps.num_backward_references; i++)
av_log(avctx, AV_LOG_DEBUG, " %#x", backward_references[i]);
av_log(avctx, AV_LOG_DEBUG, "\n");
av_log(avctx, AV_LOG_DEBUG, "Forward references:");
for (i = 0; i < ctx->pipeline_caps.num_forward_references; i++)
av_log(avctx, AV_LOG_DEBUG, " %#x", forward_references[i]);
av_log(avctx, AV_LOG_DEBUG, "\n");
for (field = 0; field < ctx->field_rate; field++) {
output_frame = ff_get_video_buffer(outlink, ctx->output_width,
ctx->output_height);
if (!output_frame) {
err = AVERROR(ENOMEM);
goto fail;
}
output_surface = (VASurfaceID)(uintptr_t)output_frame->data[3];
av_log(avctx, AV_LOG_DEBUG, "Using surface %#x for "
"deinterlace output.\n", output_surface);
memset(¶ms, 0, sizeof(params));
input_region = (VARectangle) {
.x = 0,
.y = 0,
.width = input_frame->width,
.height = input_frame->height,
};
params.surface = input_surface;
params.surface_region = &input_region;
params.surface_color_standard =
vaapi_proc_colour_standard(input_frame->colorspace);
params.output_region = NULL;
params.output_background_color = 0xff000000;
params.output_color_standard = params.surface_color_standard;
params.pipeline_flags = 0;
params.filter_flags = VA_FRAME_PICTURE;
if (!ctx->auto_enable || input_frame->interlaced_frame) {
vas = vaMapBuffer(ctx->hwctx->display, ctx->filter_buffer,
&filter_params_addr);
if (vas != VA_STATUS_SUCCESS) {
av_log(avctx, AV_LOG_ERROR, "Failed to map filter parameter "
"buffer: %d (%s).\n", vas, vaErrorStr(vas));
err = AVERROR(EIO);
goto fail;
}
filter_params = filter_params_addr;
filter_params->flags = 0;
if (input_frame->top_field_first) {
filter_params->flags |= field ? VA_DEINTERLACING_BOTTOM_FIELD : 0;
} else {
filter_params->flags |= VA_DEINTERLACING_BOTTOM_FIELD_FIRST;
filter_params->flags |= field ? 0 : VA_DEINTERLACING_BOTTOM_FIELD;
}
filter_params_addr = NULL;
vas = vaUnmapBuffer(ctx->hwctx->display, ctx->filter_buffer);
if (vas != VA_STATUS_SUCCESS)
av_log(avctx, AV_LOG_ERROR, "Failed to unmap filter parameter "
"buffer: %d (%s).\n", vas, vaErrorStr(vas));
params.filters = &ctx->filter_buffer;
params.num_filters = 1;
params.forward_references = forward_references;
params.num_forward_references =
ctx->pipeline_caps.num_forward_references;
params.backward_references = backward_references;
params.num_backward_references =
ctx->pipeline_caps.num_backward_references;
} else {
params.filters = NULL;
params.num_filters = 0;
}
vas = vaBeginPicture(ctx->hwctx->display,
ctx->va_context, output_surface);
if (vas != VA_STATUS_SUCCESS) {
av_log(avctx, AV_LOG_ERROR, "Failed to attach new picture: "
"%d (%s).\n", vas, vaErrorStr(vas));
err = AVERROR(EIO);
goto fail;
}
vas = vaCreateBuffer(ctx->hwctx->display, ctx->va_context,
VAProcPipelineParameterBufferType,
sizeof(params), 1, ¶ms, ¶ms_id);
if (vas != VA_STATUS_SUCCESS) {
av_log(avctx, AV_LOG_ERROR, "Failed to create parameter buffer: "
"%d (%s).\n", vas, vaErrorStr(vas));
err = AVERROR(EIO);
goto fail_after_begin;
}
av_log(avctx, AV_LOG_DEBUG, "Pipeline parameter buffer is %#x.\n",
params_id);
vas = vaRenderPicture(ctx->hwctx->display, ctx->va_context,
¶ms_id, 1);
if (vas != VA_STATUS_SUCCESS) {
av_log(avctx, AV_LOG_ERROR, "Failed to render parameter buffer: "
"%d (%s).\n", vas, vaErrorStr(vas));
err = AVERROR(EIO);
goto fail_after_begin;
}
vas = vaEndPicture(ctx->hwctx->display, ctx->va_context);
if (vas != VA_STATUS_SUCCESS) {
av_log(avctx, AV_LOG_ERROR, "Failed to start picture processing: "
"%d (%s).\n", vas, vaErrorStr(vas));
err = AVERROR(EIO);
goto fail_after_render;
}
if (ctx->hwctx->driver_quirks &
AV_VAAPI_DRIVER_QUIRK_RENDER_PARAM_BUFFERS) {
vas = vaDestroyBuffer(ctx->hwctx->display, params_id);
if (vas != VA_STATUS_SUCCESS) {
av_log(avctx, AV_LOG_ERROR, "Failed to free parameter buffer: "
"%d (%s).\n", vas, vaErrorStr(vas));
// And ignore.
}
}
err = av_frame_copy_props(output_frame, input_frame);
if (err < 0)
goto fail;
if (ctx->field_rate == 2) {
if (field == 0)
output_frame->pts = 2 * input_frame->pts;
else
output_frame->pts = input_frame->pts +
ctx->frame_queue[current_frame_index + 1]->pts;
}
output_frame->interlaced_frame = 0;
av_log(avctx, AV_LOG_DEBUG, "Filter output: %s, %ux%u (%"PRId64").\n",
av_get_pix_fmt_name(output_frame->format),
output_frame->width, output_frame->height, output_frame->pts);
err = ff_filter_frame(outlink, output_frame);
if (err < 0)
break;
}
return err;
fail_after_begin:
vaRenderPicture(ctx->hwctx->display, ctx->va_context, ¶ms_id, 1);
fail_after_render:
vaEndPicture(ctx->hwctx->display, ctx->va_context);
fail:
if (filter_params_addr)
vaUnmapBuffer(ctx->hwctx->display, ctx->filter_buffer);
av_frame_free(&output_frame);
return err;
}
static av_cold int deint_vaapi_init(AVFilterContext *avctx)
{
DeintVAAPIContext *ctx = avctx->priv;
ctx->va_config = VA_INVALID_ID;
ctx->va_context = VA_INVALID_ID;
ctx->filter_buffer = VA_INVALID_ID;
ctx->valid_ids = 1;
return 0;
}
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]);
}
qint64 VAApiWriter::write( const QByteArray &data )
{
VideoFrame *videoFrame = ( VideoFrame * )data.data();
const VASurfaceID curr_id = ( unsigned long )videoFrame->data[ 3 ];
const int field = FFCommon::getField( videoFrame, deinterlace, 0, VA_TOP_FIELD, VA_BOTTOM_FIELD );
#ifdef HAVE_VPP
if ( use_vpp )
{
const bool do_vpp_deint = field != 0 && vpp_deint != VA_INVALID_ID;
bool vpp_ok = false;
if ( !do_vpp_deint )
{
forward_reference = VA_INVALID_SURFACE;
vpp_second = false;
}
if ( do_vpp_deint && forward_reference == VA_INVALID_SURFACE )
forward_reference = curr_id;
if ( !vpp_second && forward_reference == curr_id )
return data.size();
if ( do_vpp_deint && !vpp_second )
{
VAProcFilterParameterBufferDeinterlacing *deint_params = NULL;
if ( vaMapBuffer( VADisp, vpp_deint, ( void ** )&deint_params ) == VA_STATUS_SUCCESS )
{
deint_params->flags = field == VA_TOP_FIELD ? VPP_TFF : VPP_BFF;
vaUnmapBuffer( VADisp, vpp_deint );
}
}
VABufferID pipeline_buf;
if ( vaCreateBuffer( VADisp, context_vpp, VAProcPipelineParameterBufferType, sizeof( VAProcPipelineParameterBuffer ), 1, NULL, &pipeline_buf ) == VA_STATUS_SUCCESS )
{
VAProcPipelineParameterBuffer *pipeline_param = NULL;
if ( vaMapBuffer( VADisp, pipeline_buf, ( void ** )&pipeline_param ) == VA_STATUS_SUCCESS )
{
memset( pipeline_param, 0, sizeof *pipeline_param );
pipeline_param->surface = curr_id;
pipeline_param->output_background_color = 0xFF000000;
if ( do_vpp_deint )
{
pipeline_param->num_filters = 1;
pipeline_param->filters = &vpp_deint;
pipeline_param->num_forward_references = 1;
pipeline_param->forward_references = &forward_reference;
}
vaUnmapBuffer( VADisp, pipeline_buf );
if ( vaBeginPicture( VADisp, context_vpp, id_vpp ) == VA_STATUS_SUCCESS )
{
vpp_ok = vaRenderPicture( VADisp, context_vpp, &pipeline_buf, 1 ) == VA_STATUS_SUCCESS;
vaEndPicture( VADisp, context_vpp );
}
}
if ( !vpp_ok )
vaDestroyBuffer( VADisp, pipeline_buf );
}
if ( vpp_second )
forward_reference = curr_id;
if ( do_vpp_deint )
vpp_second = !vpp_second;
if ( ( ok = vpp_ok ) )
draw( id_vpp, do_vpp_deint ? 0 : field );
}
else
#endif
draw( curr_id, field );
paused = false;
return data.size();
}
OMX_ERRORTYPE OMXVideoDecoderVP9HWR::ProcessorInit(void)
{
unsigned int i = 0;
for (i = 0; i < MAX_NATIVE_BUFFER_COUNT; i++) {
extMIDs[i] = (vaapiMemId*)malloc(sizeof(vaapiMemId));
extMIDs[i]->m_usrAddr = NULL;
extMIDs[i]->m_surface = new VASurfaceID;
}
initDecoder();
if (RAWDATA_MODE == mWorkingMode) {
OMX_PARAM_PORTDEFINITIONTYPE paramPortDefinitionInput;
memcpy(¶mPortDefinitionInput,
this->ports[INPORT_INDEX]->GetPortDefinition(),
sizeof(paramPortDefinitionInput));
extNativeBufferSize = INTERNAL_MAX_FRAME_WIDTH *
INTERNAL_MAX_FRAME_HEIGHT * 1.5;
extActualBufferStride = INTERNAL_MAX_FRAME_WIDTH;
extActualBufferHeightStride = INTERNAL_MAX_FRAME_HEIGHT;
for (i = 0; i < OUTPORT_ACTUAL_BUFFER_COUNT; i++ ) {
extMIDs[i]->m_usrAddr = (unsigned char*)malloc(sizeof(unsigned char) *
extNativeBufferSize);
extMIDs[i]->m_render_done = true;
extMIDs[i]->m_released = true;
}
extMappedNativeBufferCount = OUTPORT_ACTUAL_BUFFER_COUNT;
return OMX_ErrorNone;
}
#ifdef DECODE_WITH_GRALLOC_BUFFER
if (mOMXBufferHeaderTypePtrNum > MAX_NATIVE_BUFFER_COUNT) {
LOGE("Actual OMX outport buffer header type num (%d) > MAX_NATIVE_BUFFER_COUNT (%d)",
mOMXBufferHeaderTypePtrNum, MAX_NATIVE_BUFFER_COUNT);
return OMX_ErrorOverflow;
}
OMX_PARAM_PORTDEFINITIONTYPE paramPortDefinitionInput;
memcpy(¶mPortDefinitionInput,
this->ports[INPORT_INDEX]->GetPortDefinition(),
sizeof(paramPortDefinitionInput));
int surfaceWidth = mGraphicBufferParam.graphicBufferWidth;
int surfaceHeight = mGraphicBufferParam.graphicBufferHeight;
int surfaceStride = mGraphicBufferParam.graphicBufferStride;
extNativeBufferSize = mGraphicBufferParam.graphicBufferStride *
mGraphicBufferParam.graphicBufferHeight * 1.5;
extActualBufferStride = surfaceStride;
extActualBufferHeightStride = surfaceHeight;
for (i = 0; i < mOMXBufferHeaderTypePtrNum; i++) {
OMX_BUFFERHEADERTYPE *buf_hdr = mOMXBufferHeaderTypePtrArray[i];
extMIDs[i]->m_key = (unsigned int)(buf_hdr->pBuffer);
extMIDs[i]->m_render_done = false;
extMIDs[i]->m_released = true;
VAStatus va_res;
unsigned int buffer;
VASurfaceAttrib attribs[2];
VASurfaceAttribExternalBuffers* surfExtBuf = new VASurfaceAttribExternalBuffers;
int32_t format = VA_RT_FORMAT_YUV420;
surfExtBuf->buffers= (unsigned long *)&buffer;
surfExtBuf->num_buffers = 1;
surfExtBuf->pixel_format = VA_FOURCC_NV12;
surfExtBuf->width = surfaceWidth;
surfExtBuf->height = surfaceHeight;
surfExtBuf->data_size = surfaceStride * surfaceHeight * 1.5;
surfExtBuf->num_planes = 2;
surfExtBuf->pitches[0] = surfaceStride;
surfExtBuf->pitches[1] = surfaceStride;
surfExtBuf->pitches[2] = 0;
surfExtBuf->pitches[3] = 0;
surfExtBuf->offsets[0] = 0;
surfExtBuf->offsets[1] = surfaceStride * surfaceHeight;
surfExtBuf->offsets[2] = 0;
surfExtBuf->offsets[3] = 0;
surfExtBuf->flags = VA_SURFACE_ATTRIB_MEM_TYPE_ANDROID_GRALLOC;
surfExtBuf->buffers[0] = (unsigned int)buf_hdr->pBuffer;
attribs[0].type = (VASurfaceAttribType)VASurfaceAttribMemoryType;
attribs[0].flags = VA_SURFACE_ATTRIB_SETTABLE;
attribs[0].value.type = VAGenericValueTypeInteger;
attribs[0].value.value.i = VA_SURFACE_ATTRIB_MEM_TYPE_ANDROID_GRALLOC;
attribs[1].type = (VASurfaceAttribType)VASurfaceAttribExternalBufferDescriptor;
attribs[1].flags = VA_SURFACE_ATTRIB_SETTABLE;
attribs[1].value.type = VAGenericValueTypePointer;
attribs[1].value.value.p = (void *)surfExtBuf;
va_res = vaCreateSurfaces(mVADisplay,
format,
surfaceWidth,
surfaceHeight,
extMIDs[i]->m_surface,
1,
attribs,
2);
if (va_res != VA_STATUS_SUCCESS) {
LOGE("Failed to create vaSurface!");
return OMX_ErrorUndefined;
}
delete surfExtBuf;
VAImage image;
unsigned char* usrptr;
va_res = vaDeriveImage(mVADisplay, *(extMIDs[i]->m_surface), &image);
if (VA_STATUS_SUCCESS == va_res) {
va_res = vaMapBuffer(mVADisplay, image.buf, (void **) &usrptr);
if (VA_STATUS_SUCCESS == va_res) {
extMIDs[i]->m_usrAddr = usrptr;
vaUnmapBuffer(mVADisplay, image.buf);
}
vaDestroyImage(mVADisplay, image.image_id);
}
extMappedNativeBufferCount++;
}
return OMX_ErrorNone;
#endif
}
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;
}
static void nv12_to_rgba(const VAImage vaImage, rfbClient *client, int ch_x, int ch_y, int ch_w, int ch_h)
{
DebugLog(("%s: converting region (%d, %d)-(%d, %d) from NV12->RGBA\n", __FUNCTION__, ch_x, ch_y, ch_w, ch_h));
VAStatus va_status;
uint8_t *nv12_buf;
va_status = vaMapBuffer(va_dpy, vaImage.buf, (void **)&nv12_buf);
CHECK_VASTATUS(va_status, "vaMapBuffer(DecodedData)");
/* adjust x, y, width, height of the affected area so
* x, y, width and height are always even.
*/
if (ch_x % 2) { --ch_x; ++ch_w; }
if (ch_y % 2) { --ch_y; ++ch_h; }
if ((ch_x + ch_w) % 2) { ++ch_w; }
if ((ch_y + ch_h) % 2) { ++ch_h; }
/* point nv12_buf and dst to upper left corner of changed area */
uint8_t *nv12_y = &nv12_buf[vaImage.offsets[0] + vaImage.pitches[0] * ch_y + ch_x];
uint8_t *nv12_uv = &nv12_buf[vaImage.offsets[1] + vaImage.pitches[1] * (ch_y / 2) + ch_x];
uint32_t *dst = &((uint32_t*)client->frameBuffer)[client->width * ch_y + ch_x];
/* TODO: optimize R, G, B calculation. Possible ways to do this:
* - use lookup tables
* - convert from floating point to integer arithmetic
* - use MMX/SSE to vectorize calculations
* - use GPU (VA VPP, shader...)
*/
int src_x, src_y;
for (src_y = 0; src_y < ch_h; src_y += 2) {
for (src_x = 0; src_x < ch_w; src_x += 2) {
uint8_t nv_u = nv12_uv[src_x];
uint8_t nv_v = nv12_uv[src_x + 1];
uint8_t nv_y[4] = { nv12_y[ src_x], nv12_y[ src_x + 1],
nv12_y[vaImage.pitches[0] + src_x], nv12_y[vaImage.pitches[0] + src_x + 1] };
int i;
for (i = 0; i < 4; ++i) {
double R = 1.164 * (nv_y[i] - 16) + 1.596 * (nv_v - 128);
double G = 1.164 * (nv_y[i] - 16) - 0.391 * (nv_u - 128) - 0.813 * (nv_v - 128);
double B = 1.164 * (nv_y[i] - 16) + 2.018 * (nv_u - 128);
/* clamp R, G, B values. For some Y, U, V combinations,
* the results of the above calculations fall outside of
* the range 0-255.
*/
if (R < 0.0) R = 0.0;
if (G < 0.0) G = 0.0;
if (B < 0.0) B = 0.0;
if (R > 255.0) R = 255.0;
if (G > 255.0) G = 255.0;
if (B > 255.0) B = 255.0;
dst[client->width * (i / 2) + src_x + (i % 2)] = 0
| ((unsigned int)(R + 0.5) << client->format.redShift)
| ((unsigned int)(G + 0.5) << client->format.greenShift)
| ((unsigned int)(B + 0.5) << client->format.blueShift);
}
}
nv12_y += 2 * vaImage.pitches[0];
nv12_uv += vaImage.pitches[1];
dst += 2 * client->width;
}
CHECK_SURF(va_surface_id[sid]);
va_status = vaUnmapBuffer(va_dpy, vaImage.buf);
CHECK_VASTATUS(va_status, "vaUnmapBuffer(DecodedData)");
}
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));
}
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));
}
VdpStatus
vdpVideoSurfaceGetBitsYCbCr(VdpVideoSurface surface, VdpYCbCrFormat destination_ycbcr_format,
void *const *destination_data, uint32_t const *destination_pitches)
{
VdpStatus err_code;
if (!destination_data || !destination_pitches)
return VDP_STATUS_INVALID_POINTER;
VdpVideoSurfaceData *srcSurfData = handle_acquire(surface, HANDLETYPE_VIDEO_SURFACE);
if (NULL == srcSurfData)
return VDP_STATUS_INVALID_HANDLE;
VdpDeviceData *deviceData = srcSurfData->deviceData;
VADisplay va_dpy = deviceData->va_dpy;
if (deviceData->va_available) {
VAImage q;
vaDeriveImage(va_dpy, srcSurfData->va_surf, &q);
if (VA_FOURCC('N', 'V', '1', '2') == q.format.fourcc &&
VDP_YCBCR_FORMAT_NV12 == destination_ycbcr_format)
{
uint8_t *img_data;
vaMapBuffer(va_dpy, q.buf, (void **)&img_data);
if (destination_pitches[0] == q.pitches[0] &&
destination_pitches[1] == q.pitches[1])
{
const uint32_t sz = (uint32_t)q.width * (uint32_t)q.height;
memcpy(destination_data[0], img_data + q.offsets[0], sz);
memcpy(destination_data[1], img_data + q.offsets[1], sz / 2);
} else {
uint8_t *src = img_data + q.offsets[0];
uint8_t *dst = destination_data[0];
for (unsigned int y = 0; y < q.height; y ++) { // Y plane
memcpy (dst, src, q.width);
src += q.pitches[0];
dst += destination_pitches[0];
}
src = img_data + q.offsets[1];
dst = destination_data[1];
for (unsigned int y = 0; y < q.height / 2; y ++) { // UV plane
memcpy(dst, src, q.width); // q.width/2 samples of U and V each, hence q.width
src += q.pitches[1];
dst += destination_pitches[1];
}
}
vaUnmapBuffer(va_dpy, q.buf);
} else if (VA_FOURCC('N', 'V', '1', '2') == q.format.fourcc &&
VDP_YCBCR_FORMAT_YV12 == destination_ycbcr_format)
{
uint8_t *img_data;
vaMapBuffer(va_dpy, q.buf, (void **)&img_data);
// Y plane
if (destination_pitches[0] == q.pitches[0]) {
const uint32_t sz = (uint32_t)q.width * (uint32_t)q.height;
memcpy(destination_data[0], img_data + q.offsets[0], sz);
} else {
uint8_t *src = img_data + q.offsets[0];
uint8_t *dst = destination_data[0];
for (unsigned int y = 0; y < q.height; y ++) {
memcpy (dst, src, q.width);
src += q.pitches[0];
dst += destination_pitches[0];
}
}
// unpack mixed UV to separate planes
for (unsigned int y = 0; y < q.height/2; y ++) {
uint8_t *src = img_data + q.offsets[1] + y * q.pitches[1];
uint8_t *dst_u = destination_data[1] + y * destination_pitches[1];
uint8_t *dst_v = destination_data[2] + y * destination_pitches[2];
for (unsigned int x = 0; x < q.width/2; x++) {
*dst_v++ = *src++;
*dst_u++ = *src++;
}
}
vaUnmapBuffer(va_dpy, q.buf);
} else {
const char *c = (const char *)&q.format.fourcc;
traceError("error (%s): not implemented conversion VA FOURCC %c%c%c%c -> %s\n",
__func__, *c, *(c+1), *(c+2), *(c+3),
reverse_ycbcr_format(destination_ycbcr_format));
vaDestroyImage(va_dpy, q.image_id);
err_code = VDP_STATUS_INVALID_Y_CB_CR_FORMAT;
goto quit;
}
vaDestroyImage(va_dpy, q.image_id);
} else {
// software fallback
traceError("error (%s): not implemented software fallback\n", __func__);
err_code = VDP_STATUS_ERROR;
goto quit;
}
GLenum gl_error = glGetError();
if (GL_NO_ERROR != gl_error) {
traceError("error (%s): gl error %d\n", __func__, gl_error);
err_code = VDP_STATUS_ERROR;
goto quit;
}
err_code = VDP_STATUS_OK;
quit:
handle_release(surface);
return err_code;
}