blob: fe64d73a95fbd5ad44938207ca48150daff913cb (
plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
|
Libva Buffer Sharing HowTo
==========================
Beignet has extensions (clCreateBufferFromLibvaIntel/clCreateImageFromLibvaIntel)
to share gpu buffer object with libva. So users can utilize OpenCL to do processing
on VASurface or VABuffer without buffer copy.
Prerequisite
------------
Libva api version >= 0.36.0. Please check your libva api version by command
`pkg-config --modversion libva`.
Steps
-----
In order to use the extension clCreateBufferFromLibvaIntel/clCreateImageFromLibvaIntel
in your program, please follow the steps as below (We have added an example of using
clCreateImageFromLibvaIntel, please read next section for details):
- Get the address of this extension by the function:
clGetExtensionFunctionAddress("clCreateBufferFromLibvaIntel")
or clGetExtensionFunctionAddress("clCreateImageFromLibvaIntel")
- Invoke vaAcquireBufferHandle to get VASurface/VABuffer's handle, which you want
to do processing by OpenCL.
- Use clCreateBufferFromLibvaIntel/clCreateImageFromLibvaIntel to create corresponding
cl memory object from VASurface/VABuffer's handle. If you create a normal VASurface,
you should use clCreateImageFromLibvaIntel to create a cl image object from VASurface,
because VASurface is a tiling gpu buffer object. The only case you should use
clCreateBufferFromLibvaIntel to create a cl buffer object from VASurface is that: The
VASurface is created from an external untiling gpu buffer object. And You should use
clCreateBufferFromLibvaIntel to create a cl buffer object from VABuffer.
- Use OpenCL to do post-processing.
- Release this cl buffer object by clReleaseMemObject.
- Unlock this VABuffer by vaReleaseBufferHandle.
Sample code
-----------
We have developed an example of using clCreateImageFromLibvaIntel in examples/libva_buffer_sharing
directory. This example read a source nv12 file to a VASurface, and create a target VASurface.
Then create corresponding cl image objects from them. After using ocl to do a mirror effect
post-processing on this source surface, target VASurface is shown on screen by default.
- Add option -DBUILD_EXAMPLES=ON to enable building examples when running cmake, such as:
`> mkdir build`
`> cd build`
`> cmake -DBUILD_EXAMPLES=ON ../`
- Build source code:
`> make`
- Run:
`> cd examples`
`> . ../utests/setenv.sh`
`> ./example-libva_buffer_sharing`
In addition, you can choose to save the result as a nv12 file. You can use gst-launch-1.0 to see
the result. Just install gstreamer1.0-plugins-base, gstreamer1.0-plugins-bad and
gstreamer1.0-x by apt on Ubuntu. Then running the following command:
`> gst-launch-1.0 filesrc location=file_name ! videoparse format=nv12 width=xxx height=xxx \
! imagefreeze ! videoconvert ! video/x-raw, format=BGRx ! ximagesink`
|
