Currently the only transport method demonstrated is XSCOPE_FILEIO. The XSCOPE_FILEIO method has one drawback. It requires the user to interpret the files on the XCORE device. Code for this may not be readily available or easy to port to XCORE. For example, HDF5 is a common array/tensor data file format, as are Python pickle archives. Libraries to read these file formats are readily available on x86.
A new transport method (XSCOPE_STREAMIO) could be easily added so raw data arrays can be streamed from the host to the device. With this, data stored in complex file formats can be read on the host and just processed on the XCORE.
The example application would be extended to support both methods with a configuration-time switch to select the desired method.
Currently the only transport method demonstrated is XSCOPE_FILEIO. The XSCOPE_FILEIO method has one drawback. It requires the user to interpret the files on the XCORE device. Code for this may not be readily available or easy to port to XCORE. For example, HDF5 is a common array/tensor data file format, as are Python pickle archives. Libraries to read these file formats are readily available on x86.
A new transport method (XSCOPE_STREAMIO) could be easily added so raw data arrays can be streamed from the host to the device. With this, data stored in complex file formats can be read on the host and just processed on the XCORE.
The example application would be extended to support both methods with a configuration-time switch to select the desired method.