
248 CHAPTER 4 Workload-Driven Evaluation
No locks are used in parallel LU factorization. Barriers are used to separate outer-
most loop iterations as well as phases within an iteration (e.g., to ensure that the
perimeter row is computed before the blocks in it are used). Point-to-point synchro-
nization at the block level could have been used to exploit more concurrency, but
barriers make programming much easier.
Radix
The Radix program sorts a series of integers, called keys, using the popular radix
sorting method. Suppose there are n integers to be sorted, each of size b bits. The
algorithm uses a radix of r bits, where r is chosen by the user ...