Computer Systems Architecture at Yale. The Enormous Longword Instruction (ELI) Machine Progress and Research Plans.

The overall goal of the Yale Computer Science Departments Attached Processor Project systems group is to improve dramatically the practical state of the art in CPU-bound scientific computing. Specifically, we are building a very long probably over 500 bits instruction word machine, the ELI-512. A machine with this much irregular parallelism can reasonably be coded only in high-level languages this requires state-of-the-art techniques in compiling horizontal microcode. An effective approach to this problem, trace scheduling, has been developed at Yale over the past three years. Longword instruction machines are now quite popular and may offer the best alternative for obtaining supercomputer power at a fraction of its current cost. Unfortunately, they are already being built too wide for people or todays compilers to generate significant quantities of good code for them. Without this or similar work, there is little chance that more usable wide-word architectures will be commercially developed. It is an aim of this project to cause such commercial development to occur. The following are key steps in building the ELI-512 1 Parallelism measurements, 2 Compiling for existing machines, 3 Compiling for, designing, and building the highly parallel ELI-512, 4 Allowing the compiler to design highly parallel customized procesors.