Virtual-Memory C* Compiler Project

Papers

Overview of ViC*

Alex Colvin and Thomas H. Cormen.
ViC*: A Compiler for Virtual-Memory C*.
Dartmouth College Computer Science Technical Report PCS-TR97-303, November 1997. Submitted to HIPS '98.
A description of why the ViC* compiler is necessary, what it does, and how good the resulting code is.

ViC* and the Parallel Disk Model

Thomas H. Cormen and Melissa Hirschl.
Early Experiences in Evaluating the Parallel Disk Model with the ViC* Implementation.
Parallel Computing, 23:4-5, June 1997, pp. 571-600.
This paper shows that analysis for the PDM is a good indicator of actual performance, with some minor caveats. It describes implementations of PDM algorithms for out-of-core BMMC permutations and external radix sort, and shows how their measured performance matches the predicated performance when PDM parameters vary.

Papers that use the ViC* software

Thomas H. Cormen and David M. Nicol.
Out-of-Core FFTs with Parallel Disks.
ACM SIGMETRICS Performance Evaluation Review, 25:3, December 1997, pp. 3-12.
A compilation of the following two papers.

Thomas H. Cormen and David M. Nicol.
Performing Out-of-Core FFTs on Parallel Disk Systems.
To appear in Parallel Computing. Also Dartmouth College Computer Science Technical Report PCS-TR96-294, August 1996, revised September 1996, and ICASE Report 96-70, December 1996.
We show how to perform out-of-core FFTs on a uniprocessor with multiple disks. We get speedups of up to a factor of 140 compared to running Cooley-Tukey with demand paging. (We cheat slightly to obtain a speedup of 140 times, but the improvement is real.)

Thomas H. Cormen, Jake Wegmann, and David M. Nicol.
Multiprocessor Out-of-Core FFTs with Distributed Memory and Parallel Disks.
IOPADS '97, November 1997, pp. 68-78.
We extend the uniprocessor out-of-core FFT method for multiprocessors in a way that requires no interprocessor communication during the butterfly operations. Communication that would normally occur during butterfly operations is folded into permutations (BMMC, in fact) that are performed anyway. The result on one platform (a network of 8 RS6000's) is about a 10% speedup. Not earthshaking, but it is faster, and it was easy to program.

For other related papers...

...see Tom Cormen's papers page.

Thomas H. Cormen <thc@cs.dartmouth.edu>

Last modified: Fri Dec 5 16:48:50 1997