Earticle

In cloud computing era, automatic parallelization is still significant for virtualization platform. However, after several decades of development, the overall effect is still to be improved. Summary of the mainstream technology developments will be beneficial to reveal the future direction and trend. This paper reviews the technology of loop parallelization, which is the key issue in automatic parallelization. After introducing the basic models and approaches, we focus on the recent developments, on which we obtain the trend of this field and the conclusions about future.

Recently, computer hardware is evolving toward increasing the number of computing cores, not increasing the clock speed. In order to use the performance of parallelized hardware to the maximum, the running program must also be parallelized. However, software developers are accustomed to sequential programs, and in most cases, write programs that operate sequentially. They also have a lot of difficulty designing and developing software in parallel. We propose a method to automatically convert a sequential C/C++ program into a parallelized program, and develop a parallelization tool that supports it. It supports open multi-processing (OpenMP) and parallel patterns library (PPL) as a parallel framework. Perfect automatic parallelization is difficult due to dynamic features such as pointer operation and polymorphism in C/C++ language. This study focuses on verifying the conditions of parallelization rather than focusing on fully automatic parallelization, and providing advice to developers in detail if parallelization is not possible.

We propose two new approaches that automatically parallelize Java programs at runtime. These approaches, which rely on run-time trace information collected during program execution, dynamically recompile Java byte code that can be executed in parallel. One approach utilizes trace information to improve traditional loop parallelization, and the other parallelizes traces instead of loop iterations. We also describe a cost/benefit model that makes intelligent parallelization decisions, as well as a parallel execution environment to execute parallelized programs. These techniques are based on Jikes RVM. Our approach is evaluated by parallelizing sequential Java programs, and its performance is compared to that of the manually parallelized code. According to the experimental results, our approach has low overheads and achieves competitive speedups compared to the manually parallelizing code. Moreover, trace parallelization can exploit parallelism beyond loop iterations.

프로그램 언어의 수행순서는 종속성으로 인해 결정된다. 병렬 수행을 위해서는 수행 단위 사이의 종속성을 제거해야 한다. 함수 간의 종속성을 발생시키는 주요 요인으로는 전역 변수가 있다. 본 논문의 자동 병렬 수행 시스템은 순차 C언어 프로그램을 병렬 수행하여 순차 C언어 프로그램과 동일한 결과를 내게 한다. 전역 변수를 위한 프레임이 프로세서 내의 지역 메모리에 할당되며 전역 변수의 최종 결정 값을 프로세서 간에 메시지로 전달하고 복사하여 전역변수의 부작용이 발생하지 않도록 한다. 또한 피호출 함수가 수행중인 호출 함수에서는 최종 결정된 전역 변수의 값을 피호출 함수로부터 받아오기까지는 전역 변수를 참조할 수 없고 봉쇄 상태가 되는데 피호출 함수가 복귀하지 않아도 전역 변수에 대해 더 이상의 값 변경이 없음을 알게 되면 곧바로 그 값을 호출 프로세서에 전달함으로써 전역 변수 참조로 인한 수행 지연을 최대한 줄이는 방법을 제안한다.