Toward a dataflow/von Neumann hybrid architecture
ISCA '88 Proceedings of the 15th Annual International Symposium on Computer architecture
The Epsilon dataflow processor
ISCA '89 Proceedings of the 16th annual international symposium on Computer architecture
Can dataflow subsume von Neumann computing?
ISCA '89 Proceedings of the 16th annual international symposium on Computer architecture
I-structures: Data structures for parallel computing
Proceedings of the Workshop on Graph Reduction
Hi-index | 0.00 |
Future parallel computing systems will require high speed communication, efficient synchronization primitives, and low overhead context switches for communicating and synchronizing instruction streams. These issues are developed with respect to I-storage, a global memory that can synchronize requests for data values issued before the data is produced. Methods to implement these I-stores are described for the transputer architecture, with programs and timing results. This description, for both single and multiprocessors, includes a discussion of run-time systems requiring no extra hardware. However, simple hardware extensions are also proposed, which will provide faster implementations. This hardware is expected to be much less expensive than a custom dataflow machine, allowing a larger research community to experiment with dataflow languages and applications.