A new kind of parallelism and its programming in the Explicitly Many-Processor Approach

TL;DR

This paper introduces EMPA, a novel parallel architecture that enhances parallelism and simplifies programming by integrating synchronization, OS duties, and machine instructions, supported by a dedicated toolchain including assembler and simulator.

Contribution

It presents EMPA, a new parallel architecture with unique features, and provides a toolchain for programming and simulating EMPA-based systems.

Findings

EMPA architecture enables higher parallelism.

A toolchain with assembler and simulator supports EMPA programming.

The system allows conventional-style programming with advanced features.

Abstract

The processor accelerators are effective because they are working not (completely) on principles of stored program computers. They use some kind of parallelism, and it is rather hard to program them effectively: a parallel architecture by means of (and thinking in) sequential programming. The recently introduced EMPA architecture uses a new kind of parallelism, which offers the potential of reaching higher degree of parallelism, and also provides extra possibilities and challenges. It not only provides synchronization and inherent parallelization, but also takes over some duties typically offered by the OS, and even opens the till now closed machine instructions for the end-user. A toolchain for EMPA architecture with Y86 cores has been prepared, including an assembler and a cycle-accurate simulator. The assembler is equipped with some meta-instructions, which allow to use all advanced possibilities of the EMPA architecture, and at the same time provide a (nearly) conventional-style programming. The cycle accurate simulator is able to execute the EMPA-aware object code, and is a good tool for developing algorithms for EMPA