CS 475/575 -- Spring Quarter 2017 Project #0 Simple OpenMP Experiment 30 Points Due: April 10 This page was last updated: March 26, 2017 Introduction A great use for parallel programming is identical operations on large arrays of numbers. Requirements 1. Pick an array size to do the arithmetic on. Don't pick something too huge, as your machine may not allow you to use that much memory. Don't pick something too small, as the overhead of using threading might dominate the parallelism gains. 2. Using OpenMP, add up two large floating-point arrays, putting the results in another array. 3. Do this for one thread and do this for four threads. (I.e., #define NUMT 1 and #define NUMT 4 4. Time the two runs. Convert the timing results into "Mega-Multiplies per Second". 5. Review the Project Notes pages on "How Reliable is the Timing?" 6. What speedup are you seeing when you move from 1 thread to 4 threads? S = (Execution time with one thread) / (Execution time with four threads) 7. If your 4-thread-to-one-thread speedup is S (S > 1.), compute the parallel fraction: 8. float Fp = (4./3.)*( 1. - (1./S) ); Don't worry what this means just yet. This will become more meaningful soon. 9. Your commentary write-up (turned in as a PDF file) should include: 1. Tell what machine you ran this on 2. What performance results did you get? 3. What was your 4-thread-to-one-thread speedup? 4. Why do you think it is behaving this way? 5. What was your Parallel Fraction, Fp? The main Program Your main program would then look something like this: #include #include #include #define NUMT 4 #define ARRAYSIZE ?? #define NUMTRIES ?? int main( ) { #ifndef _OPENMP fprintf( stderr, "OpenMP is not supported here -- sorry.\n" ); return 1; #endif float *A = new float[ARRAYSIZE]; float *B = new float[ARRAYSIZE]; float *C = new float[ARRAYSIZE]; omp_set_num_threads( NUMT ); fprintf( stderr, "Using %d threads\n", NUMT ); double maxMegaMults = 0.; double sumMegaMults = 0.; for( int t = 0; t < NUMTRIES; t++ ) { double time0 = omp_get_wtime( ); #pragma omp parallel for for( int i = 0; i < ARRAYSIZE; i++ ) { C[i] = A[i] * B[i]; } double time1 = omp_get_wtime( ); double megaMults = (double)ARRAYSIZE/(time1-time0)/1000000.; sumMegaMults += megaMults; if( megaMults > maxMegaMults ) maxMegaMults = megaMults; } double avgMegaMults = sumMegaMults/(double)NUMTRIES ); printf( " Peak Performance = %8.2lf MegaMults/Sec\n", maxMegaMults ); printf( "Average Performance = %8.2lf MegaMults/Sec\n", avgMegaMults ); // note: %lf stands for "long float", which is what printf calls a "double" // %d stands for "decimal integer", not "double" return 0; } Grading: Feature Points Execution time results for 1 thread 5 Execution time results for 4 threads 5 Four-thread-to-one-thread Speedup 5 Parallel Fraction 10 Commentary 5 Potential Total 30 Project Turn-In Procedures  Your project turnins will all be electronic.  Your electronic turnin will be done at http://engr.oregonstate.edu/teach and will consist of: 1. Source files of everything (.cpp, .cl) 2. A Linux or Windows executable file, if needed. 3. A report in PDF format. Electronic submissions are due at 23:59:59 on the listed due date.  Your PDF report will include: 1. A title area, including your name, project number, and project name. 2. Any tables or graphs to show your results. 3. An explanation of what you did and why it worked the way it did. Your submission will not be considered valid unless you at least attempt to explain why it works the way it does. You can turn in all of your files, if you want, in a .zip file except the PDF report. Please leave the PDF report out of the .zip file. (I have a script that vacuums up all your PDF files so I can read them as one big PDF. I can get your project grades back to you faster this way.)  Your project will be graded and the score posted to the class web page. If you want to know why you did not receive full credit, send me