STRUCTURED PROGRAMMING FOR MULTIPROCESSORS PPT

STRUCTURED PROGRAMMING FOR MULTIPROCESSORS PPT

Instructor:John Thornley

Overview of multiprocessor architectures with an emphasis on shared-memory architectures.
Slides: html or PowerPoint 97.

Overview of multithreaded programming for shared-memory multiprocessors.
Slides: html or PowerPoint 97.

Overview of structured multithreaded programming and Sthreads. Introduction to multithreadable blocks and the sthread_block function.
Slides: html or PowerPoint 97.

Multithreaded recursive mergesort as an example of multithreadable blocks and the sthread_block function. Discussion of Homework One.
Slides: html or PowerPoint 97.

Overview of conditions for using pragma multithreadable: restrictions on shared variables and control flow. Introduction to multithreadable regular for loops and the sthread_regular_for_loop function.
Slides: html or PowerPoint 97.

Multithreaded matrix multiplication as an example of multithreadable regular for loops and the sthread_regular_for_loop function. Chunk size and thread mappings for multithreadable regular for loops. Cache effects in matrix multiplication. Discussion of Homework Two.
Slides: html or PowerPoint 97.

Introduction to barrier synchronization and the sthread_barrier functions.
Slides: html or PowerPoint 97.

Review of caching and memory access on quad-processor Pentium Pros. Advice on developing an efficient multithreaded matrix multiply. Description of the PSRS (Parallel Sorting by Regular Sampling) algorithm. Discussion of Homework Three.
Slides: html or PowerPoint 97.

Performance issues of the PSRS algorithm. Discussion of general multithreaded programming performance issues: multithreading overheads, load balancing, granularity, memory contention, underlying algorithm.
Slides: html or PowerPoint 97.

Timing the execution overheads of multithreaded regular for loops and barriers. Multithreaded All-Pairs Shortest-Paths (the Floyd-Warshall algorithm) as an example of the use of barriers. Discussion of Homework Four.
Slides: html or PowerPoint 97.

Introduction to synchronization flags and the sthread_flag functions.
Slides: html or PowerPoint 97.

Mergesort of a stream as an example of using synchronization flags. Discussion of Homework Five.
Slides: html or PowerPoint 97.

Introduction to synchronization counters and the sthread_counter functions. LU factorization as an example of the use of counters.
Slides: html or PowerPoint 97.

Discussion of Homework Six - All-Pairs Shortest-Paths revisted.
Slides: html or PowerPoint 97.

Guest lecture by Dr. Rajesh Bordawekar on distributed shared-memory architectures, in particular the SGI/Cray Origin and HP/Convex Exemplar.
No slides for this lecture.

Discussion of Homework Seven - the Skyline Matrix Problem.
Slides: html or PowerPoint 97.