sample_embedding_folder2/0614776.txt

Ticket Name: TDA2HG: OpenVX

Query Text:
Part Number: TDA2HG Other Parts Discussed in Thread: TDA2 What accelerator does the OpenVX implementation supports? I know that your demos won’t be moved over to OpenVX but how stable is the OpenVX implementation if we wanted to implement our algorithms with OpenVX for the TDA2 family?

Responses:
Aaron, Thanks for your questions. 1. Supported targets: As of today, all of the OpenVX kernels are implemented on C66 DSPs, so they are certainly supported as available targets. Connectivity to user kernels on EVE target has been verified using a TI extension "Harris Corners" kernel. And the ARM M4s and A15s are also supported targets. 2. OpenVX is currently in beta for TDA2x family. What we mean by this is that it passes the OpenVX conformance tests, as well as additional robustness tests, but it hasn't been internally tested in non-trivial algorithm demos/applications yet. We intend to do this internally by the time we make our end of year release, and in the mean time we also intend to continue to add features (such as graph pipelining) and support bug fixes on TDA2x platform. Algorithms which could benefit from graph pipelining will function, but not as optimally as will be the case when graph pipelinining is supported. Adding custom kernels can be done by following the pattern from existing OpenVX kernels, and one of the tutorials shows how to do this. We are also planning on adding scripts and specific app notes for this in the next few months which facilitate this effort. Please let me know if you have any more questions. Jesse