Controlling operation of multiple computational engines

A computer-implemented method for controlling operation of multiple computational engines of a physical computing device, the computer-implemented method comprising:

• providing a multiplexer module in the device, the multiplexer module comprising a first and second memory region;
• receiving from a first driver at the multiplexer module a data processing request to be processed by a first set of one or more computational engines of the computational engines;
• subsequent to receiving the data processing request, assigning a request sub-region of the first region and a response sub-region of the second region to the first driver;
• communicating data indicative of the request sub-region and the response sub-region to the first driver, thereby causing the first driver to exclusively write data of the request in the request sub-region until the writing of data is completed;
• subsequent to receiving at the multiplexer module from the first driver information indicating that the first driver has completed the writing of data of the request, dispatching the data of the request to the first set of engines;
• receiving results of processing the request at the response sub-region;
• and
• causing the first driver to exclusively access the results in the response sub-region, the communicating of data comprising sending an acknowledgement of request acceptance to the first driver, wherein the acknowledgment comprises acknowledgment data indicative of the request sub-region and the response sub-region and an interrupt-generating address, which corresponds to the last bytes of the response sub-region.

A multiplexer module for a physical computing device, the multiplexer module comprising a first and second memory region, the multiplexer module being configured for:

• receiving from a first driver a data processing request to be processed by a first set of one or more computational engines of multiple computational engines of the physical computing device;
• subsequent to receiving the data processing request, assigning a request sub-region of the first region and a response sub-region of the second region to the first driver;
• communicating data indicative of the request sub-region and the response sub-region to the first driver, thereby causing the first driver to exclusively write data of the request in the request sub-region until the writing of data is completed;
• subsequent to receiving from the first driver information indicating that the first driver has completed the writing of data of the request, dispatching the data of the request to the first set of engines;
• receiving results of processing the request at the response sub-region;
• and
• causing the first driver to exclusively access the results in the response sub-region wherein causing the first driver to exclusively access the results in the response sub-region comprises sending an interrupt signal to the first driver by writing in a host memory region of the first driver an interrupt-generating address, which corresponds to the last bytes of the response sub-region, thereby waking up the first driver.