Homework II

Due Wednesday, November 1

Problem 1

In this problem, you will learn about MPI, redoing Homework I in that framework.

• The signature of the function will be the same as before.
• As before, the TA will call the function from his main(). He will time it using MPI's timing functions. He will announce the number of machines to be used as workers.
• The TA will announce the data file to be used.
• Your manager node will read in the file.
• The three fastest groups will receive Extra Credit.
• Submit your code in a file Problem1.c.

Problem 2 (EXTRA CREDIT)

Recall the R zoo package used at the end of Chapter 1. For long vectors, say 50000000, it can have a long run time, suggesting the use of parallel computation. Here you will write the procedure in C/OpenMP, to be called from R.

• The R call form will be

  omp_rollmean(x,k)
  

  with the same meaning is with rollmean() with the default arguments (though here we restrict x to vectors).

• You will use R's .Call() method for interfacing R to C/C++. If you wish, the Rcpp package can be used, which provides convenient wrappers for the basic R macros for interfacing. However, although these are quite powerful, they can be finicky to use, and I suggest working with the SEXPs directly.
• The signature of your C function will be

  NumericVector omp_rollmean(NumericVector x, int k)
  

• The TA will announce the size of x and k that he will use in his speed tests.
• The three fastest groups will receive Extra Credit.
• Put your code in a file Problem2.cpp.

Resources:

• My ECS 256 TA's tutorial on Rcpp.
• My previous ECS 158 TA's tutorial on Rcpp.
• Code examples etc. in the Rcpp Gallery, including some examples with OpenMP.