Configurable stream switch with virtual channels for the sharing of I/O ports in stream-based architectures

A hardware accelerator, comprising:

• a plurality of processing elements;
• a plurality of streaming engines;
• and
• a stream switch coupled to the plurality of processing elements and to the plurality of streaming engines, wherein the stream switch, in operation, streams data tensors between the plurality of streaming engines and the plurality of processing elements, the stream switch including:
• a data router having:
• a plurality of input ports, each input port having a plurality of associated virtual input channels;
• and
• a plurality of output ports, each output port having a plurality of associated virtual output channels, wherein the data router, in operation, transmits data tensors from input ports of the plurality of input ports to one or more output ports of the plurality of output ports;
• configuration registers coupled to the data router, wherein the configuration registers, in operation, store configuration data associated with the virtual output channels of the respective output ports of the plurality of output ports, the stored configuration data associated with a virtual output channel identifying a source input port and virtual input channel ID associated with the virtual output channel of the output port;
• and
• arbitration logic that performs output-level arbitration using a plurality of arbitrators corresponding to respective output ports of the plurality of output ports and comprises a master arbitrator that resolves conflicts between the plurality of arbitrators, wherein the arbitration logic is coupled to the configuration registers and the data router, wherein the arbitration logic, in operation, allocates bandwidth of the data router based on request signals associated with virtual input channels of the input ports and the configuration data associated with the virtual output channels.

A method, comprising:

• storing configuration data associated with virtual output channels of output ports of a data router of a stream switch, the data router having a plurality of input ports, each having an associated plurality of virtual input channels, a plurality of output ports, each having a plurality of associated virtual output channels, and arbitration logic comprising a plurality of arbitrators corresponding to respective output ports of the plurality of output ports and a master arbitrator that resolves conflicts between the plurality of arbitrators;
• receiving requests to transmit data tensors associated with respective virtual input channels of the plurality of input ports;
• for each output port of the plurality of output ports, selecting, based on the request signals associated with virtual input channels of the input ports and the configuration data associated with the virtual output channels, a virtual output channel of the plurality of virtual output channels associated with the output port;
• allocating, using the arbitration logic, bandwidth of the data router to virtual input channels of the input ports based on the selected virtual output channels;
• and
• streaming data tensors via the data router based on the allocated bandwidth.

A non-transitory computer-readable medium having contents which cause a stream switch to perform a method, the method comprising:

• storing configuration data associated with virtual output channels of output ports of a data router of the stream switch, the data router having a plurality of input ports, each having an associated plurality of virtual input channels, a plurality of output ports, each having a plurality of associated virtual output channels, and arbitration logic comprising a plurality of arbitrators corresponding to respective output ports of the plurality of output ports and a master arbitrator that resolves conflicts between the plurality of arbitrators;
• receiving requests to transmit data tensors associated with respective virtual input channels of the plurality of input ports;
• for each output port of the plurality of output ports, selecting, based on the request signals associated with virtual input channels of the input ports and the configuration data associated with the virtual output channels, a virtual output channel of the plurality of virtual output channels associated with the output port;
• allocating, using the arbitration logic, bandwidth of the data router to virtual input channels of the input ports based on the selected virtual output channels;
• and
• streaming data tensors via the data router based on the allocated bandwidth.