Search

Supervisor:篠田 陽一, 先端科学技術研究科, 修士（情報科学）

We are developing the FUCE processor based on the dataflow computing model. FUCE means FUsion of Communication and Execution. In order to execute many threads with multiple thread execution units efficiently, the FUCE processor executes multiple threads using the exclusive multi-thread execution model. The core concept of the exclusive multi-thread execution model is continuation based multi-thread execution, which is derived from dataflow computing. The FUCE processor unifies processing inside the processor and communication with processors outside as events, and executes the event as a thread. In this paper, we introduce the thread-programming model and the architecture of the FUCE processor and evaluate the hardware cost of a FUCE processor and the concurrency performance of a FUCE processor which we described in VHDL. As a result, we understood that the processor has concurrency capability when there is sufficient thread level parallelism.

ソフトウェアシステムは, 複数のサブシステムから構成するのが一般的である. しかし, それぞれのサブシステムが独立に進化するような状況においては, サブシステムを組み合わせたときにシステムが正しく動作するかどうかを確認するのは難しく, ソフトウェア開発において問題となっている. そこで, Java言語を対象として, オブジェクトのインタフェース情報と, データフロー解析を利用して, サブシステムの互換性を計算する手法を提案する., It is a common practice to develop a software system as a collection of sub-systems. In case of each sub-system being independently evolving, the developers have to confirm it will work well in company with the other sub-systems, which is a difficult and troublesome task in a development process. This paper presents an automatic analysis for Java program to calculate compatibility among sub-systems from interface and data flow of objects.

Fine-grain non-strict data structures such as I-structure provide high level abstraction to easily write programs with potentially high parallelism due to the eager evaluation (lenient evaluation) of non-strict functions and non-strict structured-data such as arrays. Non-strict data structures require frequent dynamic scheduling at fine-grain level, which offsets the gain of latency hiding. Not only the dynamic scheduling but also asynchronous accesses to structured-data using non-strict data structures cause heavy overhead on stock machines. Mutual exclusion of structured-data on shared memory systems also causes overhead. In order to reduce overhead of fine-grain non-strict structured-data, we propose a compilation technique to analyze dependencies between the structured-data and to schedule producers and consumers of the structured-data. The performance evaluation results indicate that the technique is effective to improve the performance of fine-grain non-strict programs with structured-data on shared memory systems.