cse8803pna
Parallel Numerical Algorithms
What? CS/CSE 8803 PNA, Spring 2008
When? T/Th 9:35-10:55a
Where?: College of Computing Building (CCB), Rm 101
Who?: Prof. Richard Vuduc
Contact: richie /at/ cc /dot/ gatech /dot/ edu
Office hours: T/Th 11a-12p
  • “Physical”: Klaus, Rm 1334
  • “Virtual”: AIM VuducOfficeHours
Class mailing list: cse8803-pna-sp08 /at/ t-square /dot/ gatech /dot/ edu
Class web page: CSE-8803-PNA on T-Square; Also: http://vuduc.org/teaching/cse8803-pna-sp08
Course materials: I will post all necessary course materials (links to papers/readings) on the GT T-Square website for this course, CSE-8803-PNA.
 
There are no required textbooks for this course. However, you may find the following references useful:
Description: The goal of this course is to introduce you to parallelization techniques and issues for core numerical algorithms in computational science, including dense and sparse linear algebra computations, numerical solution of ordinary and partial differential equations, signal processing, numerical optimization, and particle simulations. About 2/3 of the course will be devoted to fundamental techniques; the remaining 1/3 will cover current research on topics such as novel algorithms and the interactions between algorithms and machine architectures, among others.
Prerequisites: There are no formal prerequisites, but prior exposure to parallel programming and numerical algorithms (e.g., CS 8803 NMC by Prof. Haesun Park) would be very helpful. The course is open to graduate and advanced undergraduate students, and students in all science and engineering disciplines are encouraged to attend.
Grading and course requirements:
  • 10%: Participation and scribe notes
  • 30%: Homework / mini-projects
  • 60%: Final project
Course policies:
  1. I will maintain all course announcements, materials, and mailing lists on Georgia Tech T-Square (http://t-square.gatech.edu). The course is cross-listed, but I will use the T-square site for “CSE 8803 PNA”
  2. Homework and mini-projects are due before class on the given due date. Late homework will not be accepted without a legitimate excuse and my approval.
  3. Collaboration is encouraged, but don’t cheat! Adhere to the Georgia Tech Academic Honor Code: http://www.honor.gatech.edu
Related courses at other institutions:
Week Day Lec. No. Topic Assigned Due
1 1/8 1 Overview Survey
1/10 2 Hardware trends Survey
2 1/15 3 From problem to parallel algorithm Proposal
1/17 4 Performance models
3 1/22 5 Programming models (1)
1/24 6 Programming models (2)
4 1/29 7 Dense linear algebra (1)
1/31 8 Dense linear algebra (2) HW 1
5 2/5 9 Sparse linear algebra (1)
2/7 10 Sparse linear algebra (2)
6 2/12 11 Fast Fourier transform
2/14 12 Multigrid
7 2/19 13 Parallelism & accuracy tradeoffs
2/21 14 Single-processor performance (1)
8 2/26 15 Guest: Prof. H. Kim on GPGPU HW 1
2/28 16 Single-processor performance (2)
9 3/4 17 Autotuning (1) Checkpoint
3/6 18 Autotuning (2)
10 3/11 19 Autotuning (2.5) Proposal
3/13 No class - Attend SIAM PP
11 3/18 No class - Spring break
3/20 No class - Spring break
12 3/25 20 N-body (1)
3/27 21 N-body (2)
13 4/1 22 Guest: Ryan Riegel on Dual-tree algorithms in statistics
4/3 23 Guest: Dr. Douglass Post of DoD HPC Modernization Program (Room TBD)
14 4/8 (Cancelled) Checkpoint
4/10 24 N-body (3)
15 4/15 25 Multicore tuning for SpMV
4/17 26 Load balancing
16 4/22 27 Graph partitioning
4/24 HPC Day in Klaus Atrium
4/28 Scheduled exam day (8-10:50a) NO EXAM but project reports due Report