## Vector/Parallel Scientific Computing

Department:
MATH
Course Number:
4777
Hours - Lecture:
3
Hours - Lab:
0
Hours - Recitation:
0
Hours - Total Credit:
3
Typical Scheduling:
Not currently offered

Scientific computational algorithms on vector and parallel computers. Speedup, algorithm complexity, interprocesses communication, synchronization, modern algorithms for linear systems, programming techniques, code optimization.

Prerequisites:

MATH 1502 OR MATH 1512 OR MATH 1555 OR MATH 1504 ((MATH 1552 OR MATH 15X2 OR MATH 1X52) AND (MATH 1522 OR MATH 1553 OR MATH 1554 OR MATH 1564 OR MATH 1X53))

Course Text:

No text

Topic Outline:
• Need for Faster and Faster Computations Weather prediction Computational fluid dynamics Monte Carlo methods Other
• Vector, Parallel Computation Increase cycle time -- advances in solid-state physics Operation of a von Neumann computer Cycle time =1/instruction rate, MPS, FLOPS, MFLOPS Remote computer usage notes Parallelize the computations SIMD, MIMD Amdahl's law a. Profilers Vector computers -- single CPU, pipelining (SIMD) a. Operation calculations -- asymptotic rate, Hockney's Law, equiv. CPU's b. Cray programming notes (1) Recognizing and treating loop dependencies Parallel computers -- memory classification a. Shared memory -- bus, switch (1) Lock and barrier solutions to contention and race conditions (2) Shared memory programming notes (a) Recognizing and treating data dependencies b. Distributed memory -- topology (1) Linear array, ring (2) Mesh (3) Hypercube -- nodes, connectivity, diameter (4) Massively parallel -- SIMD (5) CM Programming Notes (6) LAN Parallel operation equations a. Speedup, efficiency, communication delay
• Basic Linear Computational Tasks Task description Scalar merge -- e.g., sum a vector-fan-in Vector addition, scalar multiplication Matrix vector product -- i,j forms Matrix matrix product Matrix transposition Banded matrix products Partial sums, partial products -- recursive doubling a. pdf to cdf problem b. Polynomial evaluation problem