- Improvements on get method performance

[sergiosvieira]

Hi,

I get some improvements on get access method performance.
In my tests, I used a 1000x1000 matrix and I have made two loops to access each element.

void testInternalAccess(void)
{
SparseMatrix matrix(1000, 1000);
fillRandomMatrix(matrix);
high_resolution_clock::time_point t1 = high_resolution_clock::now();
for (int row = 1; row <= matrix.getRowCount(); ++row)
{
for (int col = 1; col <= matrix.getColumnCount(); ++col)
{
matrix.get(row, col);
}
}
high_resolution_clock::time_point t2 = high_resolution_clock::now();
duration timeSpan = duration_cast<duration>(t2 - t1);
std::cout << "Access matrix values " << timeSpan.count() << " seconds.\n";
}

Test Results

From 5.64532 to 1.22307 seconds.

[uestla]

Thank you, that is an interesting insight!

Looks like the method std::vector::at() is terribly slow and just replacing it with operator[] speeded up your test ~10 times on my machine.

The main difference between those two approaches is that the operator[] does not check for out-of-range indexes - that is no problem in our case because we're always checking that given coordinates are valid (method validateCoordinates()).

And because the slow method at() is being called almost everywhere, replacing it with [] access speeded up every operation!

If you won't mind, I won't merge directly your changes here but commit my own replacement and mention you in the commit message ;-)

[sergiosvieira]

Sure, no problem 😜 Em sex, 16 de fev de 2018 às 20:57, Petr Kessler <notifications@github.com> escreveu:

Thank you, that is an interesting insight! Looks like the method std::vector::at() is terribly slow and just replacing it with operator[] speeded up your test ~10 times on my machine. The main difference between those two approaches is that the operator[] does not check for out-of-range indexes - that is no problem in our case because we're always checking that given coordinates are valid (method validateCoordinates()). And because the slow method at() is being called almost everywhere, replacing it with [] access speeded up every operation! If you won't mind, I won't merge directly your changes here but commit my own replacement and mention you in the commit message ;-) — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#14 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAJuuouNury2rCoNCnl3GxGA3OFEHzr1ks5tVhXegaJpZM4SIvWb> .

-- *Sérgio Vieira* Master in Computer Science State University of Ceará (UECE)