cl_mem
clCreateFromDX9MediaSurfaceKHR
(
| cl_context context, |
cl_mem_flags flags, | |
cl_dx9_media_adapter_type_khr adapter_type, | |
void *surface_info, | |
cl_uint plane, | |
cl_int
*errcode_ret
) |
context
A valid OpenCL context created from a media adapter.
flags
A bit-field that is used to specify usage information. Refer
to the table of allowed values for flags
for
clCreateBuffer. Only
CL_MEM_READ_ONLY
, CL_MEM_WRITE_ONLY
,
and CL_MEM_READ_WRITE
values specified
in the table of allowed values for flags
for
clCreateBuffer
can be used.
adapter_type
A value from enumeration of supported adapters described in the
table of cl_dx9_media_adapter_type_khr
values for
clGetDeviceIDsFromDX9MediaAdapterKHR.
The type of surface_info
is determined by the adapter
type. The implementation does not need to support all adapter types. This
approach provides flexibility to support additional adapter types in the
future. Supported adapter types are CL_ADAPTER_D3D9_KHR
,
CL_ADAPTER_D3D9EX_KHR
, and
CL_ADAPTER_DXVA_KHR
.
If adapter_type
is CL_ADAPTER_D3D9_KHR
,
CL_ADAPTER_D3D9EX_KHR
, or
CL_ADAPTER_DXVA_KHR
, the surface_info
points to the following structure:
typedef struct _cl_dx9_surface_info_khr
{
IDirect3DSurface9 *resource; HANDLE shared_handle;
} cl_dx9_surface_info_khr;
For D3D9 surfaces, we need both the handle to the resource and the resource
itself to have a sufficient amount of information to eliminate a copy of the
surface for sharing in cases where this is possible. Elimination of the copy
is driver dependent. shared_handle
may be NULL and this
may result in sub-optimal performance.
surface_info
A pointer to one of the structures defined in the adapter_type
description above passed in as a void *.
plane
The plane of resource to share for planar surface formats. For planar formats,
we use the plane parameter to obtain a handle to thie specific plane (Y, U or
V for example). For nonplanar formats used by media, plane
must be 0.
errcode_ret
An appropriate error code. If errcode_ret
is NULL, no
error code is returned.
The width and height of the returned OpenCL 2D image object are determined by the width and height of the plane of resource. The channel type and order of the returned image object is determined by the format and plane of resource and are described in tables 9.10.3 and 9.10.4.
This call will increment the internal media surface count on
resource
. The internal media surface reference count on
resource
will be decremented when the OpenCL reference count on
the returned OpenCL memory object drops to zero.
The goal of this extension is to allow applications to use media surfaces as OpenCL memory objects. This allows efficient sharing of data between OpenCL and selected adapter APIs (only DX9 for now). If this extension is supported, an OpenCL image object can be created from a media surface and the OpenCL API can be used to execute kernels that read and/or write memory objects that are media surfaces. Note that OpenCL memory objects may be created from the adapter media surface if and only if the OpenCL context has been created from that adapter.
If this extension is supported by
an implementation, the string "cl_khr_media_sharing" will be present in the
CL_PLATFORM_EXTENSIONS
or CL_DEVICE_EXTENSIONS
string described in the table of allowed values for param_name
for
clGetDeviceInfo.
As currently proposed the interfaces for this
extension would be provided in cl_dx9_media_sharing.h
.
This section includes the D3D surface formats that are supported when the adapter type is one of the Direct 3D lineage . Using a D3D surface format not listed here is an error. To extend the use of this extension to support media adapters beyond DirectX9 tables similar to the ones in this section will need to be defined for the surface formats supported by the new media adapter. All implementations that support this extension are required to support the NV12 surface format, the other surface formats supported are the same surface formats that the adapter you are sharing with supports as long as they are listed in the two tables below.
FOUR CC code | CL image format (channel order, channel data type) |
---|---|
FOURCC('N', 'V', '1', '2'), Plane 0 |
CL_R, CL_UNORM_INT8
|
FOURCC('N', 'V', '1', '2'), Plane 1 |
CL_RG, CL_UNORM_INT8
|
FOURCC('Y', 'V', '1', '2'), Plane 0 |
CL_R, CL_UNORM_INT8
|
FOURCC('Y', 'V', '1', '2'), Plane 1 |
CL_R, CL_UNORM_INT8
|
FOURCC('Y', 'V', '1', '2'), Plane 2 |
CL_R, CL_UNORM_INT8
|
In the table above, NV12 Plane 0 corresponds to the luminance (Y) channel and Plane 1
corresponds to the UV channels. The YV12 Plane 0 corresponds to the Y channel, Plane
1 to the U channel and Plane 2 to the V channel. Note that the YUV formats map to
CL_R
and CL_RG
but do not perform any YUV to
RGB conversion and vice-versa.
D3D format | CL image format (channel order, channel data type) |
---|---|
D3DFMT_R32F |
CL_R, CL_FLOAT
|
D3DFMT_R16F |
CL_R, CL_HALF_FLOAT
|
D3DFMT_L16 |
CL_R, CL_UNORM_INT16
|
D3DFMT_A8 |
CL_A, CL_UNORM_INT8
|
D3DFMT_L8 |
CL_R, CL_UNORM_INT8
|
D3DFMT_G32R32F |
CL_RG, CL_FLOAT
|
D3DFMT_G16R16F |
CL_RG, CL_HALF_FLOAT
|
D3DFMT_G16R16 |
CL_RG, CL_UNORM_INT16
|
D3DFMT_A8L8 |
CL_RG, CL_UNORM_INT8
|
D3DFMT_A32B32G32R32F |
CL_RGBA, CL_FLOAT
|
D3DFMT_A16B16G16R16F |
CL_RGBA, CL_HALF_FLOAT
|
D3DFMT_A16B16G16R16 |
CL_RGBA, CL_UNORM_INT16
|
D3DFMT_A8B8G8R8 |
CL_RGBA, CL_UNORM_INT8
|
D3DFMT_X8B8G8R8 |
CL_RGBA, CL_UNORM_INT8
|
D3DFMT_A8R8G8B8 |
CL_BGRA, CL_UNORM_INT8
|
D3DFMT_X8R8G8B8 |
CL_BGRA, CL_UNORM_INT8
|
Note that D3D9 format names seem to imply that the order of the color channels are switched
relative to OpenCL but this is not the case. For example, layout of channels for each pixel
for D3DFMT_A32FB32FG32FR32F is the same as CL_RGBA
, CL_FLOAT
.
Properties of media surface objects may be queried using
clGetMemObjectInfo
and clGetImageInfo with
param_name
CL_MEM_MEDIA_ADAPTER_TYPE_KHR
,
CL_MEM_MEDIA_SURFACE_INFO_KHR
and
CL_IMAGE_MEDIA_SURFACE_PLANE_KHR
as described for
clGetImageInfo and in the
information below about accessing mapped regions of a memory object.
This section describes the behavior of OpenCL commands that access mapped regions of a memory object.
The contents of the region of a memory object and associated memory objects
(sub-buffer objects or 1D image buffer objects that overlap this region) mapped
for writing (i.e. CL_MAP_WRITE
or
CL_MAP_WRITE_INVALIDATE_REGION
is set in map_flags
argument to clEnqueueMapBuffer
or clEnqueueMapImage) are
considered to be undefined until this region is unmapped.
Multiple commands in command-queues can map a region or overlapping regions of a memory
object and associated memory objects (sub-buffer objects or 1D image buffer objects that
overlap this region) for reading (i.e. map_flags
= CL_MAP_READ
).
The contents of the regions of a memory object mapped for reading can also be read by kernels
and other OpenCL commands (such as
clEnqueueCopyBuffer) executing on a device(s).
Mapping (and unmapping) overlapped regions in a memory object and/or associated memory
objects (sub-buffer objects or 1D image buffer objects that overlap this region) for writing
is an error and will result in CL_INVALID_OPERATION
error returned by
clEnqueueMapBuffer
or clEnqueueMapImage.
If a memory object is currently mapped for writing, the application must ensure that the memory object is unmapped before any enqueued kernels or commands that read from or write to this memory object or any of its associated memory objects (sub-buffer or 1D image buffer objects) or its parent object (if the memory object is a sub-buffer or 1D image buffer object) begin execution; otherwise the behavior is undefined.
If a memory object is currently mapped for reading, the application must ensure that the memory object is unmapped before any enqueued kernels or commands that write to this memory object or any of its associated memory objects (sub-buffer or 1D image buffer objects) or its parent object (if the memory object is a sub-buffer or 1D image buffer object) begin execution; otherwise the behavior is undefined.
Accessing the contents of the memory region referred to by the mapped pointer that has been unmapped is undefined.
The mapped pointer returned by
clEnqueueMapBuffer
or clEnqueueMapImage can be used as
ptr
argument value to
clEnqueueReadBuffer,
clEnqueueWriteBuffer,
clEnqueueReadBufferRect,
clEnqueueWriteBufferRect,
clEnqueueReadImage, and
clEnqueueWriteImage,
provided the rules described above are adhered to.
Returns CL_SUCCESS returns a valid non-zero 2D image object and
errcode_ret
is set to CL_SUCCESS if the 2D
image object is created successfully. Otherwise, it returns a NULL value with one
of the following error values returned in errcode_ret
:
context
is not
a valid context.
flags
are not valid or if plane is not a valid plane of
resource
specified in surface_info
..
resource
specified in surface_info
is not
a valid resource or is not associated with adapter_type
(e.g.,
adapter_type
is set to CL_ADAPTER_D3D9_KHR
and resource
is not a Direct3D 9 surface created in
D3DPOOL_DEFAULT
).
shared_handle
specified in surface_info
is not NULL or a valid handle value.
resource
is not listed in tables 9.10.3 and 9.10.4.
context
that support adapter_type
.
cl_khr_dx9_media_sharing, clGetDeviceIDsFromDX9MediaAdapterKHR, clEnqueueAcquireDX9MediaSurfacesKHR, clEnqueueReleaseDX9MediaSurfacesKHR