Multiple Byte Processing with Full-Word Instructions (1975)
The article discusses advancements in parallel computing techniques using multiple byte processing with full-word instructions. Leslie Lamport highlights the efficiency improvements achieved by packing data into single words for simultaneous manipulation. This method is particularly relevant today due to larger word sizes and the shift towards masking in conditional operations.
- ▪Leslie Lamport's algorithms for parallelizing loops were initially inefficient but could be improved with parallel execution on array processors.
- ▪Packing multiple data into a single word allows for simultaneous manipulation, enhancing computation speed.
- ▪Modern computers benefit from larger word sizes and the use of masking for conditional operations, making this technique more applicable today.
Opening excerpt (first ~120 words) tap to expand
Multiple Byte Processing with Full-Word Instructions Leslie Lamport Communications of the ACM | August 1975 , Vol 18(8): pp. 471-475 Download BibTex My algorithms for parallelizing loops, described in papers starting with [9], were rather inefficient. They could be sped up with parallel execution on an array processor like the Illiac-IV. But I realized one could do even better than the 64-times speedup provided by the Illiac’s 64 processors. Each datum being manipulated was just a few bits, so I had the idea of packing several of the data into a single word and manipulating them simultaneously. Not only could this speed computation on the Illiac, but it allowed one to do array processing on an ordinary uniprocessor.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at Microsoft Research.
Discussion
0 commentsMore from Microsoft Research
