Virtual Parallelism Allows Relaxing the Synchronization Constraints of SIMD Computing Paradigm
|
Abstract:
In this paper we propose to introduce execution autonomy in the SIMD paradigm to overcome its rigidity while preserving the advantages of its synchronous programming model and we show that Virtual Parallelism support is a necessary condition to the profitable application of execution autonomy. We define execution autonomy as the capability of each processing element of a massively parallel computer to execute the instructions in a block of code of a single common program autonomously and asynchronously. We define virtual parallelism as the capability to emulate a n processors array on a m processor array with n/m performance degradation. In past related works the relaxation of SIMD synchronization has been already proposed, nevertheless its relation with Virtual Parallelism has never been studied.
Citations
| 53 | Polymorphic Processor Arrays – Maresca - 1993 |
| 39 | Steele Jr. Data parallel algorithms – Hillis, Guy - 1986 |
| 38 | The CM-5 Connection Machine: A Scalable Supercomputer – Hillis, Tucker - 1993 |
| 8 | Parallel architectures for vision – Maresca, Lavin, et al. - 1988 |
| 8 | Data Management and Control-Flow Aspects of an SIMD/SPMD Parallel Language/Compiler – Nichols, Siegel, et al. - 1993 |
| 2 | Reissis e Q – Miller, Prasanna-Kumar, et al. - 1993 |
| 1 | Introducing Execution Autonomy – Baglietto, Maresca, et al. |
| 1 | Comparing SFMD and SPMD Computation for On-chip Multiprocessing of Intermediate Level Image Understanding Algorithms – Rehfuss, Hammerstromm - 1997 |
| 1 | Asynchronous SIMD: An Architectural Concept for High Performance Image – Weems - 1997 |
