Theme and Laboratory	Outline
Compiler Approaches for Exploring Verious Levels of Parallelism Multi-Processor Computing Fujitsu Laboratory http://www.rwcp.or.jp/activities/achievements/MP/fujitsu/index.html	Multi-Processor Computing Fujitsu Laboratory has been developing a compiler which can exploit both coarse and fine grain parallelism to achieve good performance. On the study of coarse grain parallelism, we propose a set of extensions for OpenMP Fortran API that can be implemented effectively on distributed memory parallel computers (DMP) and workstation clusters. On the digital poster, we present two new concepts, the processor group and the index distribution, which are introduced for applying OpenMP API to DMP environment and show the preliminary results of the performance evaluation which are obtained by the prototype compiler for the extended OpenMP Fortran API. We also show the effects of the optimization of predicated execution which is one of the activities of studying fine grain parallelism. We have developed a compiler for a VLIW processor which partially supports predicated execution. The reduction of branch instruction overhead, the increase of instruction level parallelism, and execution time speed-up are presented. In the best case, about a 60% speed improvement is achieved by predicated execution.
High-Density Optical Bus Optical Interconnection OKI Laboratory http://www.rwcp.or.jp/activities/achievements/OI/oki/hp99-10/achieve-j.html	The Optical Interconnection Oki Laboratory has been developing a high-density parallel optical fiber module technique for device-to-device interconnection and an integrated optical interconnection technology for interconnection within a device (chip-to-chip) to realize a seamless optical interconnection environment. Our digital poster will introduce the following two technologies that we developed during FY1999. (1) For device-to-device optical interconnection, an optical fiber array module prototype with a high-density 125 micrometer interval manufactured using an image fiber plate and experiment data that show high efficiency and low crosstalk characteristics. (2) For chip-to-chip optical interconnection, the basic characteristics of the one-body integrated InP-LED on an Si-LSI device produced using a heterogeneous materials connection process, an optical circuit substrate prototype produced using a deflective optical device, and optical characteristic data that show the feasibility of high-density optical wiring.
GaAsSb long-wavelength VCSEL and Optical Interface Macro for System LSI Optical Interconnection NEC Laboratory http://www.rwcp.or.jp/activities/achievements/OI/nec/index.htm	The Optical Interconnection NEC Laboratory proposed a high-speed optical I/O interface macro, that is, Optical-interconnection Intellectual Property (OIP) for a CMOS-LSI in order to achieve System On Chip (SOC) including an optical interface. The laboratory is also conducting research and development on a long-wavelength surface emitting laser that uses a new material, gallium arsenic antimonide (GaAsSb) as a light-emitting element suitable for OIP. OIP achieves a high-speed optical interconnection as an IP macro of ASIC, enabling a system design that includes optical I/O in the usual design flow by using a hardware description language. RWC 2000 digital poster will exhibit an LSI package for OIP that allows up to 6-channels I/O to be implemented with the transmission speed of 1.25Gbps/ch. GaAsSb long-wavelength vertical-cavity surface emitting laser (VCSEL) oscillates at 1.3 micrometer in which chromatic dispersion of optical fiber becomes zero. Compared to 0.85 micrometer wavelength VCSEL which is commonly used as a transmitter of giga-bit ethernet modules, GaAsSb VCSEL has several advantages such as a wider band-width when used with single-mode-fiber and lower operation voltage. In RWC 2000, we will exhibit a first monolithically grown VCSEL chip, which achieved a room temperature CW oscillating at around 1.3 micrometer band.