X-Video Motion Compensation (XvMC), is an extension of the X video extension (Xv) for the X Window System (a.k.a. X11) under Linux operating-systems. The XvMC API allows video programs to offload portions of the video decoding process to the GPU video-hardware. In theory this process should also reduce bus bandwidth requirements. Currently, the supported portions to be offloaded by XvMC onto the GPU are motion compensation (mo comp) and inverse discrete cosine transform (iDCT) for MPEG-2 video. XvMC also supports offloading decoding of mo comp, iDCT, and Variable Length Decoding|VLD]] ("Variable Length Decoding", more commonly known as "slice-level acceleration") for not only MPEG-2 but also MPEG-4 ASP (H.263) and MPEG-4 AVC (H.264) video on VIA Unichrome (S3 Graphics Chrome Series) hardware. Popular software applications known to take advantage of XvMC include MPlayer, MythTV, and xine.
XvMC is the UNIX equivalent of the Microsoft Windows DirectX Video Acceleration (DxVA) API. Mac OS X also includes MPEG-2 acceleration capabilities, but Apple has chosen not to expose that API for use outside their own DVD-Video player application.
- 1 Device drivers
- 2 XvMC API technical overview
- 3 Software support
- 4 Processes that could be accelerated
- 5 Current limitations
- 6 See also
- 7 References
- 8 External links
Each hardware video GPU capable of XvMC video acceleration requires a X Window System (a.k.a. X11) software device-driver to enable these features.
There are currently three X11 Nvidia drivers available: One open source developed by the Linux community called Nouveau, one 2D only open source driver maintained by NVIDIA called nv, and one proprietary binary device driver by NVIDIA. Nouveau is currently working on XvMC support , the 2D nv driver does not support XvMC, and the official proprietary binary device driver by NVIDIA only supports MPEG-2 offloading (mo comp and iDCT).
VIA provides open source device drivers for its VIA Unichrome (S3 Graphics Chrome Series) hardware, supporting offloading of MPEG-2, MPEG-4 ASP (H.263) and MPEG-4 AVC (H.264) video. Thanks to VLD level of decoding VIA offloads much more decoding tasks from CPU than GPUs supporting iDCT or mo comp levels only.
Intel provides official open source device drivers which supports MPEG-2 offloading (mo comp and iDCT) on Intel's 8xx/9xx range of integrated graphics chips. 
There are no device drivers which support XvMC on ATI/AMD hardware, (although all ATI/AMD Radeon GPU hardware has support for MPEG-2 acceleration on iDCT and mo comp levels).
There are no device drivers which support XvMC on Matrox hardware, (although the Matrox Parhelia GPU hardware has support for MPEG-2 acceleration on mo comp level).
Binary device driver by S3 only supports MPEG-2 offloading in initial 2.0.16 driver on Chrome 20 and up GPUs.
XvMC API technical overview
XvMC extends the X video extension (Xv) and makes use of the familiar concept of the XvPort. Ports have attributes that can be set and queried through Xv. In XvMC ports can also have hardware motion compensation contexts created for use with them. Ports which support XvImages (ie. they have an "XV_IMAGE" port encoding as described in the Xv version 2.2 API addendum) can be queried for the list of XvMCSurface types they support. If they support any XvMCSurface types an XvMCContext can be created for that port.
XvMCContext describes the state of the motion compensation pipeline. An individual XvMCContext can be created for use with a single port, surface type, motion compensation type, width and height combination. For example, a context might be created for a particular port that does MPEG-2 motion compensation on 720 x 480 4:2:0 surfaces. Once the context is created, referencing it implies the port, surface type, size and the motion compensation type. Contexts may be "direct" or "indirect". For indirect contexts the X server renders all video using the data passed to it by the client. For direct contexts the client libraries render the video with little or no interaction with the X server.
XvMCSurfaces are buffers into which the motion compensation hardware can render. The data in the buffers themselves are not client accessible and may be stored in a hardware-specific format. Any number of buffers can be created for use with a particular context (resources permitting).
XvMC provides video acceleration starting at one of two places in the video pipeline. Acceleration starting at the first point, which we shall call the "Motion Compensation" level, begins after the the inverse quantization and IDCT at the place where motion compensation is to be applied. The second point, which we shall call the "IDCT" level, begins before the IDCT just after the inverse quantization.
Rendering is done by presenting the library with a target XvMCSurface and up to two reference XvMCSurfaces for the motion compensation, a buffer of 8x8 blocks and a command buffer which describes how to use the 8x8 blocks along with motion compensation vectors to construct the data in the target XvMCSurface. When the pipeline starts at the iDCT level, Xv will perform the IDCT on the blocks before performing the motion compensation. A function is provided to copy/overlay a portion of the XvMCSurface to a drawable with arbitrary scaling.
XvMCSubpictures are separate surfaces that may be blended with the target surface. Any number of XvMCSubpictures may be created for use with a context (resources permitting). Both "backend" and "frontend" subpicture behavior are supported.
XvMC acceleration is supported in:
Processes that could be accelerated
Even though XvMC currently only support hardware acceleration of motion compensation (mo comp) and inverse discrete cosine transform (iDCT), (and Variable-Length Decoding for VIA Unichrome GPU), additional video decoding processes could be passed on to modern GPUs which on Microsoft Windows can accelerate many more via Microsoft's DirectX Video Acceleration (DxVA) API. XvMC could be extended in the future to support the same processes:
- Motion compensation (mocomp)
- Inverse discrete cosine transform (iDCT)
- Inverse telecine 3:2 and 2:2 pull-down correction
- Inverse modified discrete cosine transform (iMDCT)
- In-loop deblocking filter
- Intra-frame prediction
- Inverse quantization (IQ)
- Variable-Length Decoding (VLD), more commonly known as slice-level acceleration
- Spatial-temporal deinterlacing and automatic interlace/progressive source detection
- Bitstream processing (CAVLC/CABAC)
Besides not matching all of the features and function of DxVA (which is the Microsoft equivalent API for Windows), and lacking support for other video formats than MPEG-2 in Linux device drivers from ATI and NVIDIA, the XvMC API specification version 1.0 currently also has these other limitations:
- BOB and onefield are the only deinterlacing methods that work with XvMC.
- Picture in Picture (PiP) does work with XvMC.
- Stepping though the video frame by frame can cause artifacts with XvMC.
- Editing a video is difficult to impossible with XvMC activated.
- The client application must run as root
- Lacking network transparency in the XvMC API
- Lacking the ability for the client application to be completely independent of which client side library needs to be used (ie. a unified wrapper library).
A project called Video Acceleration API (VA API) is underway to develop a more modern video acceleration API which will support the video acceleration features of modern GPUs.
- Video Acceleration API (VA API)
- DirectX Video Acceleration (DxVA) API - Microsoft Windows equivalent to VA API and XvMC