Abstracting scratch pad memories as distributed arrays

A computing system configured to manage memory therein, the computing system comprising:

• a host configured to direct execution of a program;
• one or more compute devices in communication with the host, each compute device including a first memory and one or more compute units, each compute unit including a second memory and one or more processing elements, each processing element including a third memory and is configured to execute at least a portion of the program;
• wherein the program includes instructions to:
• allocate a segment in a distributed array, wherein the distributed array is a global set of local memory regions;
• bind the segment to a physical memory region, wherein the physical memory region is any one of:
• the first memory, the second memory, or the third memory;
• and
• use the segment by a workgroup in a dispatched data parallel kernel, wherein a workgroup includes one or more work items.

A non-transitory computer-readable storage medium storing a set of instructions for execution by a general purpose computer to manage memory in a computing system, the set of instructions comprising:

• an allocating code segment for allocating a segment in a distributed array, wherein the distributed array is a global set of local memory regions;
• a binding code segment for binding the segment of the distributed array to a physical memory region;
• and
• a using code segment for using the segment of the distributed array by a workgroup in a dispatched data parallel kernel, wherein a workgroup includes one or more work items.