Parallel programmers must deal with a host of issues peculiar to parallel programs such as synchronization, protecting shared variables, and finding thread safe versions of common math routines (such as random number generation). One of the most subtle problems faced by the parallel programmer, however, arises from the properties of floating point numbers. Floating point numbers are the same in serial and parallel computations, of course, but when a program executes in parallel, special features of these numbers are more likely to impact your results. Tim Mattson et al. of Intel writes about floating point numbers and parallelization.
Parallelization and Floating Point Numbers
June 30th, 2008 · No Comments
Tags: MulticoreInfo · Programming
Like what you're reading? Come back every day for multicore news, or subscribe to RSS updates.
