====================
Added
-----
- Adds an event system for better control of the processing.
- Adds a hashing method to the CLEnvironment.
- Adds use_host_ptr flag to the Array class to enable/disable that flag in the buffer allocation.
- Adds a flag to the Zeros kerneldata to enable non host accessible buffers (does not use_host_ptr)
- Adds inline keyword to the Scalar kernel data.
- Adds ctype property to the kernel data classes.
- Adds a do_data_transfer, is_blocking and wait_for flag to the evaluate method of a CLFunction.
Changed
-------
- Changed the get_subset signature to also accept a batch_range instead of batch_start and batch_end.
- Changed the get_subset method to check for consecutive problem indices and if so, use the batch range method instead.
- Cached the CLEnvironment creation to ensure same CLEnvironments have the same OpenCL queue.
- Removed the 'COAW' requirements from the array conversions, it was not needed.
- The CL function now caches the compiled kernels for faster consecutive execution.
Other
-----
- Set the ``enqueue_{host,device}_access`` methods of the KernelData to accept a list of CLEnvironments.
- Buffer allocation is now done per context instead of per CLEnvironment.