blob: 85008a8881b412ba51b34ac7d9eab0eb5b55267a (
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
68
69
70
71
|
V4l2 Buffer Sharing HowTo
=========================
Beignet has extension clGetMemObjectFdIntel to share gpu buffer object with v4l2. So users
can utilize OpenCL to do processing on input/ouput buffers of v4l2 device without buffer copy.
Prerequisite
------------
Linux kernel supports DMABUF buffer sharing for v4l2 from version 3.8. DMABUF buffer
sharing runs well for V4L2_PIX_FMT_MJPEG format on this version, but there is a bug
for V4L2_PIX_FMT_YUYV format. Linux kernel 3.19.0-rc1 fix this bug, so please use kernel
version 3.19.0-rc1 at least if you want to utilize this feature for V4L2_PIX_FMT_YUYV
format.
Steps
-----
The below official v4l2 document describes the details of sharing DMA buffers between
v4l devices and other devices using v4l2 as a DMABUF importer:
[http://linuxtv.org/downloads/v4l-dvb-apis/dmabuf.html](http://linuxtv.org/downloads/v4l-dvb-apis/dmabuf.html)
Beignet has added extension clGetMemObjectFdIntel to support this mechanism. Please follow
the steps as below to utilize DMABUF buffer sharing between v4l devices and Beignet:
- Get the address of this extension by the function:
clGetExtensionFunctionAddress("clGetMemObjectFdIntel")
- Create a number of cl buffer objects, invoke clGetMemObjectFdIntel to get these buffer
objects' file descriptors.
- Initiating streaming I/O with DMABUF buffer sharing by calling the VIDIOC_REQBUFS v4l2 ioctl.
- Enqueue these buffers by calling the VIDIOC_QBUF, dequeue a buffer by calling VIDIOC_DQBUF,
use OpenCL to do processing on this buffer and re-enqueue...
- Close file descriptors of these buffers by close if your program doesn't need DMABUF buffer
sharing anymore.
Sample code
-----------
We have developed an example showing how to share DMA buffers between webcam and Beignet in
examples/v4l2_buffer_sharing directory. The webcam directly captures V4L2_PIX_FMT_YUYV frames
into cl buffer objects by the way of DMABUF buffer sharing, then frames are got mirror effect
by OpenCL kernel, and finally show on screen by libva.
Steps to build and run this example:
- Update your linux kernel to at least 3.19.0-rc1.
- Make sure there is a webcam connected to your pc.
- 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-v4l2_buffer_sharing`
|
