I agree with other comments that using vectors is a much safer way to solve your problem: using arrays directly is definitely error-prone. Of course, if your exercise requires using arrays, then you should use arrays.
Regarding the performance, I have written a small test using g++ on Ubuntu 10.04. Running
g++ --version
I get
g++ (Ubuntu 4.4.3-4ubuntu5) 4.4.3
My test program creates a 100x100 matrix and sets each element to some value. It first has a few declarations:
#include <vector>
#include <iostream>
#include "util.h" // Timer utilities.
#define LINE_COUNT (100) // The number of lines.
#define COL_COUNT (100) // The number of columns.
#define REPETITIONS (100000) // Number of repetitions for each test.
using namespace std;
Then I have the test using vectors:
void use_vectors()
{
int line = LINE_COUNT;
int member = COL_COUNT;
vector<vector<int> > matrix(line, vector<int>(member));
// Set data.
for (int i = 0; i < line; i++)
{
for (int j = 0; j < member; j++)
{
matrix[i][j] = -5;
}
}
}
Then I have a function to perform the same test (create matrix and set values) using arrays:
void use_arrays()
{
int line = LINE_COUNT;
int member = COL_COUNT;
int **matrix;
matrix = new int * [line];
for (int i = 0; i < line; i++)
{
matrix[i] = new int[member];
}
// Set data.
for (int i = 0; i < line; i++)
{
for (int j = 0; j < member; j++)
{
matrix[i][j] = -5;
}
}
for (int i = 0; i < line; ++i)
{
delete [] matrix[i];
}
delete [] matrix;
}
The main program repeats both tests, and records the time needed for each of them. Here is the main program:
main()
{
long int es = 0;
long int eu = 0;
start_timer();
for (int i = 0; i < REPETITIONS; i++)
{
use_vectors();
}
stop_timer();
es = elapsed_sec();
eu = elapsed_usec();
cout << "Vectors needed: " << es << " sec, " << eu << " usec" << endl;
start_timer();
for (int i = 0; i < REPETITIONS; i++)
{
use_arrays();
}
stop_timer();
es = elapsed_sec();
eu = elapsed_usec();
cout << "Arrays needed: " << es << " sec, " << eu << " usec" << endl;
}
The timer functions are based on the library function gettimeofday() (see e.g. http://linux.die.net/man/2/gettimeofday).
The result is the following:
Vectors needed: 24 sec, 624416 usec
Arrays needed: 10 sec, 16970 usec
So it seems that vectors do have some overhead wrt to arrays. Or can I do something to improve the performance of vectors? I checked my benchmark code a few times and it seems to me I got it right.
Anyway, I would by no means advise using arrays just to gain performance unless it really makes a big difference in your application.