Communication optimizations for global multi-threaded instruction scheduling

  • Authors:

  • Guilherme Ottoni;David I. August

  • Affiliations:

  • Princeton University, Princeton, NJ;Princeton University, Princeton, NJ

  • Venue:
  • Proceedings of the 13th international conference on Architectural support for programming languages and operating systems
  • Year:
  • 2008

Quantified Score

Hi-index 0.00

Visualization

Abstract

The recent shift in the industry towards chip multiprocessor (CMP) designs has brought the need for multi-threaded applications to mainstream computing. As observed in several limit studies, most of the parallelization opportunities require looking for parallelism beyond local regions of code. To exploit these opportunities, especially for sequential applications, researchers have recently proposed global multi-threaded instruction scheduling techniques, including DSWP and GREMIO. These techniques simultaneously schedule instructions from large regions of code, such as arbitrary loop nests or whole procedures, and have been shown to be effective at extracting threads for many applications. A key enabler of these global instruction scheduling techniques is the Multi-Threaded Code Generation (MTCG) algorithm proposed in [16], which generates multi-threaded code for any partition of the instructions into threads. This algorithm inserts communication and synchronization instructions in order to satisfy all inter-thread dependences. In this paper, we present a general compiler framework, COCO, to optimize the communication and synchronization instructions inserted by the MTCG algorithm. This framework, based on thread-aware data-flow analyses and graph min-cut algorithms, appropriately models andoptimizes all kinds of inter-thread dependences, including register, memory, and control dependences. Our experiments, using a fully automatic compiler implementation of these techniques, demonstrate significant reductions (about 30% on average) in the number of dynamic communication instructions in code parallelized with DSWP and GREMIO. This reduction in communication translates to performance gains of up to 40%.