This preview shows pages 1–3. Sign up to view the full content.
This preview has intentionally blurred sections. Sign up to view the full version.View Full Document
Unformatted text preview: 1 Atms 502 Numerical Fluid Dynamics Thu., Dec. 07, 2006 12/7/06 Atms 502 - Fall 2006 - Jewett 2 Some thoughts: extra credit • OpenMP Don’t use automatic parallelization! Introduce your own OpenMP directives It’s easy to do the basic steps. !$OMP PARALLEL DO PRIVATE (i,j,k) do k = 2,nz-1 do j = 1,ny do i = 1,nx u3(i,j,k) = u3(i,j,k) + tstep*diff*( u1(i+1,j,k)... enddo enddo enddo !$OMP END PARALLEL DO llnl.gov 12/7/06 Atms 502 - Fall 2006 - Jewett 3 Parallel performance • Benchmarks Wallclock time Total user time Zone cycles per second Speedup Parallel efficiency Architectures Resources: • High Performance Computing (Kuck) • Parallel Computing - LLNL www. llnl.gov/computing • Concepts in Parallel Computing - Alf Wachsmann www.slac.stanford.edu/~alfw/talks 2 12/7/06 Atms 502 - Fall 2006 - Jewett 5 Flynn’s Taxonomy MIMD Multiple Instruction, Multiple Data Clusters MISD Multiple Instruction, Single Data Not generally used SIMD Single Instruction, Multiple Data Vector computing SISD Single Instruction, Single Data Sequential computing 12/7/06 Atms 502 - Fall 2006 - Jewett 6 Flynn’s Taxonomy • SISD Single instruction, single data Serial computer SI: One instruction per clock cycle SD: one data stream per cycle Most PCs 12/7/06 Atms 502 - Fall 2006 - Jewett 7 Flynn’s Taxonomy • SIMD Single instruction, multiple data Parallel SI: all cpu execute same instruction per cycle MD: each cpu can process different data Best suited for problems “with high degree of regularity” 12/7/06 Atms 502 - Fall 2006 - Jewett 8 Flynn’s Taxonomy • MIMD Multiple instruction, multiple data Parallel MI: cpus may execute...
View Full Document
This note was uploaded on 10/06/2009 for the course ATMOSPHERI 502 taught by Professor Jewett during the Fall '09 term at University of Illinois at Urbana–Champaign.
- Fall '09