Intel® oneVPL GPU Runtime

Intel® oneVPL GPU Runtime is a Runtime implementation of oneVPL API for Intel Gen GPUs. Runtime provides access to hardware-accelerated video decode, encode and filtering.

Supported video encoders: HEVC, AVC, MPEG-2, JPEG, VP9
Supported video decoders: HEVC, AVC, VP8, VP9, MPEG-2, VC1, JPEG, AV1
Supported video pre-processing filters: Color Conversion, Deinterlace, Denoise, Resize, Rotate, Composition

Implementation is written in C++ 14 with parts in C-for-Media (CM).

oneVPL GPU Runtime is a part of Intel software stack for graphics:

• Linux Graphics Drivers - General Purpose GPU Drivers for Linux Operating Systems
  • Visit documentation for instructions on installing, deploying, and updating Intel software to enable general purpose GPU (GPGPU) capabilities for Linux-based operating system distributions.

oneVPL Library and oneVPL GPU Runtime are successors for Intel Media SDK.

How to use

This runtime implementation is not self-sufficient. Application should use one of the available frontend dispatcher libraries:

• oneVPL
• Media SDK

We strongly recommend to use oneVPL as a dispatcher frontend (libvpl.so.2). Media SDK dispatcher (libmfx.so.1) can be used as well, but you will get capabilities limited by those which are exposed via Media SDK API and which are actually implemented by oneVPL Runtime. Basically, compatibility with Media SDK is provided to support Media SDK based applications which did not yet migrate to oneVPL on a new HW platforms.

oneVPL Gen Runtime implementation supports the following hardware platforms:

Dependencies

oneVPL GPU Runtime depends on LibVA. This version of oneVPL GPU Runtime is compatible with the open source Intel Media Driver for VAAPI. Dispatcher and Samples code hosted in oneVPL repository.

Table of contents

• License
• How to contribute
• System requirements
• How to build
  • Build steps
  • Enabling Instrumentation and Tracing Technology (ITT)
• Recommendations
• See also

License

oneVPL GPU Runtime is licensed under MIT license. See LICENSE for details.

How to contribute

See CONTRIBUTING for details. Thank you!

System requirements

Operating System:

• Linux x86-64 fully supported
• Linux x86 only build

Software:

• LibVA
• VAAPI backend driver:
  • Intel Media Driver for VAAPI
• Some features require CM Runtime library (part of Intel Media Driver for VAAPI package)

Hardware: Intel platforms supported by the Intel Media Driver for VAAPI starting with Tiger Lake.

How to build

Build steps

Get sources with the following Git* command (pay attention that to get full oneVPL GPU Runtime sources bundle it is required to have Git* with LFS support):

git clone https://github.com/oneapi-src/oneVPL-intel-gpu onevpl-gpu
cd onevpl-gpu

To configure and build oneVPL GPU Runtime install cmake version 3.14 or later and run the following commands:

mkdir build && cd build
cmake ..
make
make install

oneVPL GPU Runtime depends on a number of packages which are identified and checked for the proper version during configuration stage. Please, make sure to install these packages to satisfy oneVPL GPU Runtime requirements. After successful configuration 'make' will build oneVPL GPU Runtime binaries and samples. The following cmake configuration options can be used to customize the build:

The following cmake settings can be used to adjust search path locations for some components oneVPL GPU Runtime build may depend on:

Enabling Instrumentation and Tracing Technology (ITT)

To enable the Instrumentation and Tracing Technology (ITT) API you need to:

• Either install Intel® VTune™ Amplifier
• Or manually build an open source version (see ITT API for details)

and configure oneVPL GPU Runtime with the -DENABLE_ITT=ON. In case of VTune it will be searched in the default location (/opt/intel/vtune_amplifier). You can adjust ITT search path with either CMAKE_ITT_HOME or CMAKE_VTUNE_HOME.

Once oneVPL GPU Runtime was built with ITT support, enable it in a runtime creating per-user configuration file ($HOME/.mfx_trace) or a system wide configuration file (/etc/mfx_trace) with the following content:

Output=0x10

Recommendations

• In case of GCC compiler it is strongly recommended to use GCC version 6 or later since that's the first GCC version which has non-experimental support of C++14 being used in oneVPL GPU Runtime.