Advanced Computing Tools for Applied Research
Lab Practice CH7: Performance
Test the performance of the following codes
C code
In the Lab you can use MinGW Development Environment, installed under Inicio > Programas >
Desarrollo de software.
Exercise C1: “compact” expressions
sum += x * y; vs sum = x * y + sum;
sum+=*p++; vs sum += p[i];
Exercise C2: Use of cache
Compare the performance of the following sorting algorithms
//Sort1: Bubble method (close exchanges)
for(x=0; x<n-1; x++) {
for(y=0; y<n-x-1; y++) {
if(array[y]>array[y+1]) {
temp = array[y+1];
array[y+1] = array[y];
array[y] = temp;
}
}
}
//Sort2: Long exchanges
for(x=0; x<n-1; x++) {
for(y=x+1; y<n; y++) {
if(array[x]>array[y]) {
temp = array[x];
array[x] = array[y];
array[y] = temp;
}
}
}
Exercise C3: Memory allocation
Analyze the degradation of performance due to memory reallocation.
numbers=NULL;
//numbers = (int*) realloc (numbers, N * sizeof(int)); //Good 1.8 s
for (i=1; i<N; i++) {
numbers = (int*) realloc (numbers, i * sizeof(int)); //Bad 17 s
if (numbers!=NULL) {
numbers[i-1]=rand();
}
else {
//free (numbers);
puts ("Error (re)allocating memory");
exit (1);
}
}
free(numbers);
Matlab code
Use tic and toc to make execution-‐time measurements.
You may need to use a loop to make accurate measurement. For example:
>> tic, for t=1:100, prueba2, end, toc
Elapsed time is 10.846000 seconds
Exercise M1: Use of cache
Compare the performance of the following sorting algorithms
//Add by columns
suma=0;
for i=1:1000
for j=1:1000
suma=suma+z(i,j);
end
end
//Add by rows
suma=0;
for i=1:1000
for j=1:1000
suma=suma+z(j,i);
end
end
//Add internal loops. How many cores does it use?
suma=sum(z(:));
Exercise M2: Virtual memory
Try the following code for different values of sz. Use Task Manager to monitor memory use.
(Matlab may stop responding when you hit the memory limit)
sz=100000000;
x=rand(sz,1); tic, x=sort(x); toc
tic, x=rand(sz,1); x=sort(x); toc
Exercise M3: Cache and Virtual memory conflicts
Run two or more matlab sessions in parallel (or C programs in parallel) compare the performance with
running the same programs one after another.
Programs block each other while trying to access memory. More remarkable (but more difficult to
simulate) is while accessing the hard drive.
Another way to test disk access conflicts is executing to copy commands in parallel.
Exercise M4: Memory allocation
Measure execution time in the following cases.
Use Mlint to detect pre-‐allocation suggestions in Matlab
//Without memory allocation. Matlab uses realloc
for i=1:1000
for j=1:1000
z2(j,i)=z(j,i);
end
end
//With memory allocation
z2=ones(size(z));
for i=1:1000
for j=1:1000
z2(j,i)=z(j,i);
end
end
Webpage analysis
Measure loading times with Google Chrome by activating developer tools.
-‐Try loading different pages, ex. www.iit.upcomillas.es and www.upcomillas.es
-‐Try reloading, and reloading without cache.
-‐Try setting mobile behavior: Resolution and Network speed.
Browser comparison
Run tests in different browsers (may include mobile phone) and compare results.
Exercise B1-‐ Javascript
https://www.webkit.org/perf/sunspider/sunspider.html
Exercise B2-‐ CSS performance
http://ie.microsoft.com/TEStdrive/Performance/MazeSolver/Default.html
Laboratory report
• Program listings
• Time measurements
