Abstract:
The article is devoted to the analysis of the quick sort (QS) program execution in a vector dataflow processor (VDP), which uses the pointer vectors method to store arrays. The deficiency of pointer vectors method revealed in the QS program is analyzed and a method for solving this disadvantage is proposed by introducing split and fuse commands into the processor instruction set. Despite the significant complication of the graph and increase the overall number of commands executed in the QS program, the introduction of new split and fuse commands into VDP command system made it possible to achieve up to 7.4$\times$ faster performance than Intel Skylake processor core.
Key words and phrases:vector processor, dataflow architecture, sorting program, instruction level parallelism, fine-grained parallelism, vector performance.