CHAOS/PARTI Runtime Support Library

The CHAOS/PARTI runtime support library is a set of software primitives that are designed to efficiently handle irregular problems on distributed memory systems. The primitives have been designed to ease the implementation of computational problems on parallel architecture machines by relieving users of low-level machine specific issues. The design philosophy has been to leave the original (sequential) source codes essentially unaltered, with the exception of the introduction of various calls to the CHAOS/PARTI primitives which are embedded in the codes at the appropriate locations. CHAOS/PARTI library provides primitives that

• couple partitioners to the application programs,
• remap data and partition work among processors, and
• optimize interprocessor communications.

Parallelizing irregular problems on distributed memory machines using the CHAOS/PARTI runtime support library involves six phases. The first four phases concern mapping data and computations onto processors. The next two steps involve analyzing data access patterns in a loop nest and generating optimized communication calls.

1. Data Partitioning: Assign data array elements to processors
2. Data Remapping: Redistribute data array elements
3. Loop Iteration Partitioning: Allocate iterations to processors
4. Loop Iteration Remapping: Redistribute indirection array elements
5. Inspector:
   • Translate global indices to local indices
   • Identify non-local references
   • Generate communication schedules
6. Executor:
   • Prefetch non-local data using schedules
   • Perform computation

CHAOS/PARTI ftp distribution

[Applications | High Performance I/O | Compilers | Tools]

Questions about the system or webserver: webmaster@cs.umd.edu

Other problems or questions, please mail acha@cs.umd.edu