Graphics processor with non-blocking concurrent architecture

A method of performing concurrent graphics computation using a plurality of computation elements, each computation element comprising a local memory, and a plurality of Single Instruction Multiple Data (SIMD) Arithmetic Logic Units (ALUs), the method comprising:

• receiving inputs defining computation instances to be performed on the plurality of computation elements, each of the instances associated with a respective data element stored in a local memory of one of the plurality of computation elements;
• sorting the instances into packets;
• distributing the instances of a packet among the computation elements, for execution thereon, based on where data elements associated with the instances are located among the local memories of the computation elements;
• and
• each of the computation elements:
• (i) combining instances for concurrent execution by the ALUs thereof, for which the same type of computation is to be performed and which refer to different data elements stored in the local memory thereof, and (ii) executing the computation for the combined instances of computation.

Computation apparatus comprising:

• a plurality of computation elements, each computation element comprising a local memory, and a plurality of Arithmetic Logic Units (ALUs);
• scheduling logic configured to receive indications of computation instances to be performed on the plurality of computation elements, each of the instances associated with a respective data element stored in a local memory of one of the plurality of computation elements, the scheduling logic further configured to sort the instances into packets;
• and
• distribution logic configured to distribute the instances of a packet among the computation elements, for execution thereon, based on where data elements associated with the instances are located among the local memories of the computation elements, wherein
• each of the computation elements is operable to combine instances for concurrent execution by the ALUs thereof, for which the same type of computation is to be performed and which refer to different data elements stored in the local memory thereof.