Dynamic wavefront creation for processing units using a hybrid compactor

A non-transitory computer-readable storage medium storing a set of instructions for execution by a general purpose computer to repack dynamic wavefronts during program code execution on a processing unit, each dynamic wavefront including multiple threads, the set of instructions comprising:

• a first determining code segment for determining whether all wavefronts following a same control path in the program code have reached a compaction point, wherein the compaction point is a branch instruction, if a branch instruction is detected in executing the program code;
• a second determining code segment for determining whether all wavefronts following the same control path in the program code have reached a reconvergence point, wherein the reconvergence point is a beginning of a program code segment to be executed by both a taken branch and a not taken branch from a previous branch instruction, if no branch instruction is detected in executing the program code;
• and
• a repacking code segment for repacking the dynamic wavefronts with all threads that follow the same control path in the program code if all wavefronts following the same control path have reached the branch instruction or the reconvergence point.

A processor configured to repack dynamic wavefronts during program code execution on a processing unit, each dynamic wavefront including multiple threads, the processor comprising:

• a compaction table, configured to store compaction point information, wherein the compaction point is a branch instruction;
• a reconvergence stack for each wavefront, each reconvergence stack configured to store compaction point information for the corresponding wavefront;
• a reconvergence table, configured to store reconvergence point information, wherein the reconvergence point is a beginning of a program code segment to be executed by both a taken branch and a not taken branch from a previous branch instruction;
• wherein the processor is configured to:
• determine whether all wavefronts following a same control path in the program code have reached the compaction point, if a branch instruction is detected in executing the program code;
• copy compaction point information from the reconvergence stack for a wavefront to the compaction table;
• wait for all wavefronts following the same control path to reach the compaction point;
• determine whether all wavefronts following the same control path in the program code have reached the reconvergence point, if no branch instruction is detected in executing the program code;
• copy the reconvergence stack entry for the reconvergence point to the reconvergence table;
• wait for all wavefronts following the same control path at the reconvergence point;
• and
• repack the dynamic wavefronts with all threads that follow the same control path in the program code if all wavefronts following the same control path have reached the branch instruction or the reconvergence point.