March 4: classes normal
March 11: I have to be in Washington DC. Replacement classes will be held
on March 12, usual times.
*****************************
High preformance computing 3:30-4
Sobolev Spaces 4:00-4:45
Jan
******************

Projects for "High Performance Computing"

Some existing benchmark projects are in ~jmandel/test
Feel free to peruse.

Every project starts with "write a simple test program to time the
following" and ends with "plot the results if applicable and analyze
and interpret the results".

All tests should run on our three architectures:
- IBM RS/6000 (tiger) (single cpu only)
- SUN Sparcenter (math) 
- SGI Power Challenge (dracula)

You will be surprised at the results!

1. effect of if-then, subroutine call, I/O statement on ddot loop
(Chris Linden)
2,3 dependence of the speed of dot loop on 
 -data size
 -number of processors and loop scheduling
 plot results as x=log of data size, y=MFLOPS
 - simple Fortran loop or BLAS routine
 - same for daxpy loop
(Marian Brezina)

4,5. same for matrix*vector multiply (matrix is square)
 same for transpose(matrix)*vector multiply (matrix is square)
(John Wilson)

6. speed of square matrix * vector multiply on unrolling depth and type
(Petr Vanek)

7. speed of n by n matrix * n by 2 matrix multiply on unrolling depth and type
(Radek Tezaur)

These all require the use of FORTRAN 77. Extra food for thought: try 
compare with array primitives in FORTRAN 90.

9. Speed of Gauss-Seidel iteration (John Weber)

10. FFT on Power Challenge (R. Sweet)

Look in ~jmandel/test/choke for inspiration and timing routines