DATA PARALLEL ALGORITHMS (CM-2)

Speaker :

Guy L. Steele Jr., Thinking Machines

Sponsor: Thinking Machines

US Price: $5.00 / Int'l Price: $5.00
Order #: 03-Steele

Tape Description

The data parallel programming style emphasizes the exploitation of parallelism in the data, and opportunities to perform similar processing on many data elements at once. The assumption that there are enough processors to provide one for each data item. This is like array processing, but data structures more general than arrays of floating-point numbers are considered. Many interesting algorithms use pointers. Recurring themes in parallel programming include elementwise operations, conditionals, replication (broadcasting, spreading, and fetching), reduction (full reduction, row/column reduction, and combining store), permutation, and scanning (parallel prefix). Parallel algorithms for matrix multiplication and region labeling in an image illustrate the use of these building blocks in real applications.

Tape Length: 55 minutes

Date Recorded:

Usage Suggestions: Upper-division/Graduate-level/Industry & Professional Use

Speaker Biography

Guy L. Steele Jr. received his BA in applied mathematics from Harvard College in 1975, and his MS and PhD in computer science and artificial intelligence in 1977 and 1980 respectively, from the Massachusetts Institute of Technology. He currently is a senior scientist at Thinking Machines Corporation where he is responsible for the design and implementation of parallel programming languages and other systems software for the Connection Machine computer system.