\documentclass[twocolumn]{article}
% \documentclass{article}

\setlength{\oddsidemargin}{-0.5in}
\setlength{\evensidemargin}{-0.5in}
\setlength{\topmargin}{0.0in}
\setlength{\headheight}{0in}
\setlength{\headsep}{0in}
\setlength{\textwidth}{7.0in}
\setlength{\textheight}{9.5in}
\setlength{\parindent}{0in}
\setlength{\parskip}{0.05in}
\setlength{\columnseprule}{0.3pt}

\setlength\columnsep{25pt}

\usepackage{fancyvrb}
\usepackage{relsize}
\usepackage{hyperref}

\usepackage{listings}
\usepackage{amsmath}

\begin{document}

\lstset{xleftmargin=5mm,framexleftmargin=10mm,basicstyle=\footnotesize}

Name: \_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_

Directions: MAKE SURE TO COPY YOUR ANSWERS TO A SEPARATE SHEET FOR SENDING
ME AN ELECTRONIC COPY LATER.

{\bf 1.} (15) The online help for the {\bf clusterApply()} function in R's
{\bf parallel} package says, 

\begin{quote}
‘clusterApplyLB’ is a load balancing version of ‘clusterApply’.
If the length ‘p’ of ‘seq’ is not greater than the number of nodes
‘n’, then a job is sent to ‘p’ nodes.  Otherwise the first ‘n’
jobs are placed in order on the ‘n’ nodes.  When the first job
completes, the next job is placed on the node that has become
free; this continues until all jobs are complete.  Using
‘clusterApplyLB’ can result in better cluster utilization than
using ‘clusterApply’, but increased communication can reduce
performance.  Furthermore, the node that executes a particular job
is non-deterministic.
\end{quote}

Fill in the blanks:  This is similar to the 
\_\_\_\_\_\_\_\_\_\_\_\_
option in 
\_\_\_\_\_\_\_\_\_\_\_\_ programming, 
with chunk size
\_\_\_\_\_\_\_\_\_\_\_\_.

{\bf 2.} (65)  Here you will work on a Thrust version of the CUDA code
in our last quiz, which solved a problem similar to the root finding
example in Section 4.11.  It finds the root of a user-supplied function
{\bf f()}, which is increasing on (0,1) and has a root somewhere inside.
The initial search interval is (0,1), but the interval gets smaller with
each iteration.  At any iteration, the current interval is divided in
subintervals, with each thread handling one subinterval.  Fill in the
blanks.


\begin{lstlisting}
// Thrust example:  find the root of an 
// increasing function on (0,1); not
// assumed efficient

#include <stdio.h>
#include <thrust/device_vector.h>
#include <thrust/remove.h>
#include <thrust/sequence.h>

__host__ __device__ float f(float x) {
   return x*x - 0.5;
}

struct signchange {
   float width;
   thrust::device_vector<float>::iterator ab;
   signchange(
      ________,  // blank (a)
      float _width):
         ab(_dab),width(_width) {}
   __host__ __device__
   bool operator()(int i)
   {  if (_________)  // blank (b)
            return true;
      else return false;
   }
};

// do niters iterations, with nsubintervals 
// checked each time; typically would want 
// nsubintervals = number of threads
float throot(int niters, int nsubintervals)
{  int iter;
   thrust::host_vector<float> hab(2);
   hab[0] = 0.0;
   hab[1] = 1.0;
   float width;  // subinterval width
   thrust::device_vector<float> dab(hab);
   thrust::host_vector<int> hfoundit(1);
   thrust::device_vector<int> dfoundit(1);
   thrust::device_vector<int> 
      seq(nsubintervals);
   thrust::sequence(seq.begin(),seq.end(),0);
   for (iter = 0; iter < niters; iter++) {
      width = 
         (hab[1] - hab[0]) / nsubintervals;
      __________(  // blank (c)
         ____________  // blank(d), contains 
                       // .begin(), .end()
         ________________  // blank (e)
         signchange(dab.begin(),width));
      thrust::copy(dfoundit.begin(),
         dfoundit.end(),hfoundit.begin());
      hab[0] = _____________  // blank (f)
      hab[1] = hab[0] + width;
      thrust::copy(hab.begin(),hab.end(),
         dab.begin());
   }
   return _________; // blank (g)
}

// test case
int main(int argc, char **argv)
{  float root;
   int niters = atoi(argv[1]),
       nsubintervals = atoi(argv[2]);
   root = throot(niters,nsubintervals);
   printf("%f\n",root);
}
\end{lstlisting}

{\bf 3.} Suppose we wish to use Thrust to compress an upper-triangular
matrix, storing only the upper-triangular portion, column by column.
For instance, the matrix

$$
\left (
\begin{array}{rrr}
5 & 12 & 13\\
0 & 168 & 8 \\
0 & 0 & 1 \\
\end{array}
\right )
$$

would be stored as (5,12,168,13,8,1).

\begin{itemize}

\item [(a)] (10) Which would be appropriate here, a Thrust scatter or gather
operation?

\item [(b)] (10) For a $4 \times 4$ input matrix, what would be the
appropriate map vector, given your answer in (a)?  Assume row-major
order.  Answer in vector form, e.g. (8,88,-2,-6).

\end{itemize}

\onecolumn

\newpage

{\bf Solutions:}

{\bf 1.} dynamic; OpenMP; 1

{\bf 2.}

\begin{lstlisting}
// Thrust example:  find the root of an increasing function on (0,1)

#include <stdio.h>
#include <thrust/device_vector.h>
#include <thrust/remove.h>
#include <thrust/sequence.h>

__host__ __device__ float f(float x) {
   return x*x - 0.5;
}

struct signchange {
   float width;
   thrust::device_vector<float>::iterator ab;
   signchange(thrust::device_vector<float>::iterator _dab, 
              float _width):
         ab(_dab),width(_width) {}
   __host__ __device__
   bool operator()(int i)
   {  if (f(ab[0]+i*width) < 0 &&
         f(ab[0]+(i+1)*width) > 0)
            return true;
      else return false;
   }
};

// do niters iterations, with nsubintervals checked each time; typically
// would want nsubintervals = number of threads
float throot(int niters, int nsubintervals)
{  int iter;
   thrust::host_vector<float> hab(2);
   hab[0] = 0.0;
   hab[1] = 1.0;
   float width;  // subinterval width
   thrust::device_vector<float> dab(hab);
   // index of subinterval where sign change is found
   thrust::host_vector<int> hfoundit(1);
   thrust::device_vector<int> dfoundit(1);
   thrust::device_vector<int> seq(nsubintervals);
   thrust::sequence(seq.begin(),seq.end(),0);
   for (iter = 0; iter < niters; iter++) {
      width = (hab[1] - hab[0]) / nsubintervals;
      thrust::copy_if(seq.begin(),seq.end(),
         dfoundit.begin(),
            signchange(dab.begin(),width));
      thrust::copy(dfoundit.begin(),dfoundit.end(),
         hfoundit.begin()); 
      hab[0] = hab[0] + hfoundit[0] * width;
      hab[1] = hab[0] + width;
      thrust::copy(hab.begin(),hab.end(),dab.begin());
   }
   return hab[0];
}

// test case
int main(int argc, char **argv)
{  float root;
   int niters = atoi(argv[1]),
       nsubintervals = atoi(argv[2]);
   root = throot(niters,nsubintervals);
   printf("%f\n",root);
}
\end{lstlisting}

{\bf 3a.} gather

{\bf 3b.} (0,1,5,2,6,10,3,7,11,15)

\end{document}