Roofline and Performance Modeling

Memory hierarchy

Matrix Multiplication

Shared Memory Parallelism

Sources of Parallelism and Locality

Communication-avoiding matrix multiplication

Data Parallel Algorithms

CUDA (Compute Unified Device Architecture)

Graphics Processor (GPU)

Distributed Memory Machines and Programming

Advanced MPI

Collective Communication Algorithms

UPC++: Partitioned Global Address Space Languages

Domain Specific Languages (Halide)

Distributed Data Structures (BCL)

Parallel Matrix Multiply

Dense Linear Algebra

Sparse-Matrix-Vector-Multiplication

Iterative Solvers

Structured Grids

Machine learning

Supervised learning

Deep learning

Unsupervised learning

Graph Partitioning

Fast Fourier transform

Graph Algorithms

Cloud Computing

Big Data Processing

High Performance Computing (HPC)

Dynamic Load Balancing

Hierarchical Methods for the N-Body Problem

Sorting algorithm

Search algorithm

Computational Biology

Quantum Computing

CS 267: Applications of Parallel Computers UC Berkeley Spring 2020 The course addresses programming parallel computers to solve complex scientific and engineering problems. It covers an array of parallelization strategies for numerical simulation, data analysis, and machine learning, and provides experience with popular parallel programming tools. CS267 was originally designed to teach students how to program parallel computers to efficiently solve challenging problems in science and engineering, where very fast computers are required either to perform complex simulations or to analyze enormous datasets. CS267 is intended to be useful for students from many departments and with different backgrounds, although we will assume reasonable programming skills in a conventional (non-parallel) language, as well as enough mathematical skills to understand the problems and algorithmic solutions presented. CS267 satisfies part of the course requirements for the Designated Emphasis ("graduate minor") in [Computational Science and Engineering](http://www.google.com/url?q=http%3A%2F%2Fcitris-uc.org%2Finitiatives%2Fdecse%2F&sa=D&sntz=1&usg=AOvVaw1fRFGvIaIYhkwE3kp4J1nf).

While this general outline remains, a large change in the computing world started in the mid 2000's: not only are the [fastest computers](http://www.google.com/url?q=http%3A%2F%2Fwww.top500.org%2F&sa=D&sntz=1&usg=AOvVaw1SiByfUjOpTR6ctURAV2Ef) parallel, but nearly all computers are becoming parallel, because the physics of semiconductor manufacturing will no longer let conventional sequential processors get faster year after year, as they have for so long (roughly doubling in speed every 18 months for many years). So all programs that need to run faster will have to become parallel programs. (It is considered very unlikely that compilers will be able to automatically find enough parallelism in most sequential programs to solve this problem.) For background on this trend toward parallelism, click [here](http://www.google.com/url?q=http%3A%2F%2Fwww.eecs.berkeley.edu%2FPubs%2FTechRpts%2F2006%2FEECS-2006-183.html&sa=D&sntz=1&usg=AOvVaw2XoPTheTzKVlyjp1G_YcM5).

Students in CS267 will get an overview of the parallel architecture space, gain experience using some of the most popular parallel programming tools, and be exposed to a number of open research questions. The lectures will also cover a broad set of parallelization strategies for applications covering numerical simulation and data analysis to machine learning.   

CS 267: Applications of Parallel Computers

Parallelism and Distributed Systems

Roofline model

Roofline and Performance Modeling

1 courses cover this concept

CS 267: Applications of Parallel Computers