CS561, Project 8

1. Parallel Matrix-matrix multiplication using PThreads

Consider the following serial program that multiplies two matrices A and B:

   for (i = 0; i < N; i = i + 1)
      for (j = 0; j < N; j = j + 1)
         C[i*N + j] = 0.0;

   for (i = 0; i < N; i = i + 1)
      for (j = 0; j < N; j = j + 1)
         for (k = 0; k < N; k = k + 1)
            C[i*N + j] = C[i*N + j] + A[i*N + k]*B[k*N + j];

• Write a parallel version of the matrix multiply using pthreads

  HINT: the statement C[i*N + j] = 0.0 can also be part of the parallel section

  HINT: you do NOT need any lock variables !!!

2. Performance comparison

I have written a sequential matrix multiply program and you can obtain a copy of the file here: click here

Compile the program using CC -o pj8 pj8.C

The program is run using the command:

pj8 N print

where N is the size of the matrices. The program constructs 2 random N x N matrices and multiply them.

If the second argument to the program is "print", the result is printed; otherwise, the program simply exits.

The program pj8.C is intended for you to perform some performance experiments by comparing the execution times of your parallel version of matrix multiply and this sequential matrix multiply program.

Your program (should have the name "main" to make grading easy) must accept 3 arguments and is run using the following command:

main N NThreads print

The first argument N is the size of the matrices, the second argument NThreads is the number of worker threads. If the third argument to the program is "print", the result is printed; otherwise, the program simply exits.

You need to make sure that the parellel matrix multiply program is correct; so in the testing phase, run your program as:

main 3 8 print

(Also try 4x4 and 5x5 matrices to be sure).

After you have determined that your program is correct, run performance comparison tests using:

time main 300 8 noprint

Compare the running times with the sequential version.

Run your performance tests on compute.mathcs.emory.edu (more CPUs) !!!

3. Turn in

• Turn in the Makefile file of your program as assignment `pj8':

          /home/cs561000/turnin  Makefile pj8
  

• Turn in your parallel program as assignment `pj8a':

          /home/cs561000/turnin  yourfile.C  pj8a