For teams that need compliance with the Khronos OpenVX™ standard, CoreFlow provides a modern execution runtime with seamless integration into OpenVX – providing an implementation of the OpenVX specification version 1.3.1, with various extensions and features enabled.
Supported Extensions in CoreFlow
The following extensions and features are enabled in the CoreFlow project:
- OPENVX_USE_USER_DATA_OBJECT: Enables the user data object extension, allowing custom data objects to be used within the OpenVX framework.
- OPENVX_USE_IX: Enables the import/export extension, facilitating the import and export of data between different OpenVX contexts.
- OPENVX_USE_XML: Enables XML-based graph serialization and deserialization.
- OPENVX_USE_S16: Enables support for 16-bit signed integer data types.
- OPENVX_USE_OPENCL_INTEROP: Enables interoperability with OpenCL, allowing OpenVX to leverage OpenCL for acceleration.
- OPENVX_USE_NN: Enables the neural network extension, providing support for neural network operations within OpenVX.
- OPENVX_USE_NN_16: Enables half float (float16) support for the neural network extension.
Conditional Extensions for ARM Architectures
The following extensions are conditionally enabled for ARM and ARM64 architectures:
- OPENVX_USE_TILING: Enables the tiling extension, which allows for tiled processing of images.
- OPENVX_KHR_TILING: Enables Khronos tiling extension for efficient image processing.
- EXPERIMENTAL_USE_VENUM: Enables experimental support for VENUM, a vector processing unit.
Conformance and Experimental Features
- OPENVX_CONFORMANCE_VISION: Enables conformance for vision functions per the OpenVX specification.
- OPENVX_USE_ENHANCED_VISION: Enables conforamnce enhanced vision functions per the OpenVX specfication.
- OPENVX_CONFORMANCE_NNEF_IMPORT: Enables conformance for NNEF (Neural Network Exchange Format) import, ensuring compatibility with NNEF models.
- OPENVX_CONFORMANCE_NEURAL_NETWORKS: Enables conformance for neural networks, ensuring that neural network operations meet the OpenVX standard.
- EXPERIMENTAL_PLATFORM_SUPPORTS_16_FLOAT: Enables experimental support for 16-bit floating-point data types.
- EXPERIMENTAL_USE_DOT: Enables experimental support for DOT (Graphviz) output for graph visualization.
- EXPERIMENTAL_USE_OPENCL: Enables experimental support for OpenCL, allowing for further exploration and development of OpenCL-based acceleration within OpenVX.
Conformance Tests
The implementation in this project passes all OpenVX conformance tests to verify the implementation against the standard.