CMSC 732

Deprecated: The each() function is deprecated. This message will be suppressed on further calls in /home/zhenxiangba/zhenxiangba.com/public_html/phproxy-improved-master/index.php on line 456
CMSC 732

CMSC 732/838T - Compiling for High Performance Architectures

Catalog Description

Compilation techniques for exploiting parallelism on a variety of advanced multiprocessors. Analysis of scientific programs (dependence testing, dependence graphs, pointers). Program transformations (loop interchange, loop fusion, tiling, renaming). Exploiting parallelism for vector, parallel, and message-passing architectures. Memory hierarchy management (register/cache reuse, prefetching, multithreading). Programming environments and performance debugging.

Course Objectives

This course focuses on compilation techniques needed to obtain high performance on parallel architectures. Compiler analysis and optimization techniques are presented in class lectures. Students will gain experience in reading and presenting research papers. A programming project provides experience with actual implementations and allow students to explore research issues in greater detail.

Prerequisites

CMSC 430 or permission of instructor.

Class Times and Location

Monday & Wednesday 11am-12:15pm
CLB 104

Staff

Chau-Wen Tseng (tseng@cs.umd.edu)

Syllabus

Lectures

September 4, 1996	Introduction
September 9,11 1996	Data dependence
September 16, 1996	Dependence tests (Rice)
September 19, 1996	Dependence tests (Stanford)
September 23, 1996	Dependence tests (Maryland)
September 25, 1996	Dependence tests (Banerjee)
September 30, 1996	Loop transformations
October 7, 1996	Unimodular transformations
October 9, 1996	Data locality
October 14, 16 1996	Tiling and prefetching
October 21, 1996	Vectorization and parallelization
October 23, 1996	Midterm (solutions)
October 28, 30 1996	Distributed-memory compilers
November 4, 1996	NUMA compilers, prog environments
November 6, 1996	Detecting Coarse-Grain Parallelism
November 11, 1996	Eliminating Barrier Synchronization
November 13, 1996	Global Communication Optimization
November 18, 1996	Reducing False Sharing
November 20, 1996	Supercomputing'96
November 22, 1996	Automatic I/O Prefetching
November 25, 1996	Compiler-Directed Page Coloring
November 27, 1996	Optimization for Efficient Threads
December 2, 1996	Shasta: Fine-Grain Shared Memory
December 4, 1996	Data and Computation Transformations
December 6, 1996	Efficient Pointer Analysis for C
December 9, 1996	Is it a Tree, a DAG, or a Cyclic Graph?
December 11, 1996	Project summary presentations
December 13, 1996	Final

Papers

Projects

Topics

Introduction
Dependence analysis
Loop transformations
Memory hierarchy management
Parallelization
Vectorization
Shared-memory machines
Message-passing machines
Interfaces

Course Text

M. Wolfe, High Performance Compilers for Parallel Computing, Addison-Wesley

Grading

15% - Midterm
30% - Final
30% - Project and paper
25% - Presentations

SUIF

Newsgroups

Students

Miscellaneous

NullStone Suite

Acknowledgments

Monica Lam, Mary Hall, and Kathryn McKinley provided the original lecture notes. The SUIF compiler system is from Monica Lam's SUIF compiler group at Stanford.