Parallel computing architecture for use with a non-greedy scheduling algorithm

A method of coordinating the operation of a plurality of processing units in a parallel computing architecture, the method comprising:

• requesting a resource from a main storage, for storage in a temporary storage, the resource being associated with a work group of one or more work groups in a first pool of pending work associated with a first processing unit of the plurality of processing units, each work group of the one or more work groups comprising one or more work elements to be processed by the first processing unit using the associated resource;
• transmitting a notification message to scheduling logic associated with a second processing unit of the plurality of processing units, the notification message comprising an indication of the requested resource;
• in response to receiving the notification message at the scheduling logic associated with the second processing unit, determining whether a second pool of pending work associated with the second processing unit comprises a work group associated with the resource;
• and
• in response to determining that the second pool of pending work comprises a work group associated with the resource, prioritizing processing of the work group associated with the resource by the second processing unit so as to obtain the resource from the temporary storage.

A non-transitory computer readable storage medium having stored thereon a computer readable description of an integrated circuit that, when processed in an integrated circuit manufacturing system, causes the integrated circuit manufacturing system to manufacture a parallel computing architecture comprising, for each of a plurality of processing units:

• a pool of pending work configured to store one or more work groups, each work group being associated with a resource and comprising one or more work elements to be processed by the processing unit using that resource;
• scheduling logic configured to:
• select a work group of the one or more work groups for processing;
• request a particular resource associated with the selected work group from a main storage, for storage in a temporary storage;
• transmit a notification message to one or more other processing units of the plurality of processing units, the notification message comprising an indication of the requested particular resource;
• and
• in response to receiving a notification message from another processing unit of the plurality of processing units which comprises an indication of a specific resource that has been requested, determine whether the pool of pending work comprises a work group associated with the specific resource;
• and
• in response to determining that the pool of pending work comprises a work group associated with the specific resource, prioritize processing of the work group associated with the specific resource so as to obtain the specific resource from the temporary storage.