Methods for scalably exploiting parallelism in a parallel processing system

A method for operating a computer system to generate an array of output data elements using program code that, when executed, generates at least one output data element for the array of output data elements, the method comprising:

• defining, by the computer system, based at least in part on an array size parameter for the array of output data elements, a first number (n1) of blocks of output data elements, wherein each block includes a number (ne) of the output data elements;
• defining, by the computer system, a thread array corresponding to each block, each thread array including a number (n0) of threads of execution of the program code sufficient to generate each of the ne output data elements in the corresponding block;
• distributing, by the computer system, the n1 thread arrays among one or more of a number Ctot of processing cores;
• and
• operating each of the one or more processing cores to execute each of the thread arrays distributed thereto,
• wherein executing a thread array includes executing the n0 threads of the program code to generate each of the ne output data elements in the corresponding block and storing the ne output data elements in a memory.

A method for operating a computer system to generate an array of output data elements using program code that, when executed, generates at least one output data element for the array of output data elements, the method comprising:

• defining, by the computer system, a number (n2) of grids, each grid including a subset of the output data elements;
• defining, by the computer system, within each grid, a number (n1) of blocks, each block including a number (ne) of the plurality of output data elements, the number n1 being determined based at least in part on an array size parameter for the array of output data elements;
• defining, by the computer system, a thread array corresponding to each block within each grid, each thread array including a number (n0) of threads of execution of the program code sufficient to generate each of the ne output data elements in the corresponding block;
• distributing, by the computer system, the n1 thread arrays for each of the n2 grids across one or more processing cores;
• and
• operating each of the one or more processing cores to execute each of the thread arrays distributed thereto,
• wherein executing a thread array includes executing the n0 threads of the program code to generate each of the ne output data elements in the corresponding block and storing the ne output data elements in a memory.