I'm trying to code a function that, given an index as the input argument, extract the corresponding layer from a 3D matrix, returning a 2D matrix.
My default 3D matrix constructor looks something like this:
Matrice3D(unsigned int depth, unsigned int height, unsigned int width, const T &value) : _3D_matrix(0), _height(0), _width(0), _depth(0) {
try {
_3D_matrix = new T[height * width * depth];
for (int z = 0; z < depth; z++) {
for (int y = 0; y < height; y++) {
for (int x = 0; x < width; x++) {
_3D_matrix[y * width * depth + x * depth + z] = value;
}
}
}
}
catch(...) {
delete[] _3D_matrix;
throw;
}
_height = height;
_width = width;
_depth = depth;
}
(the try/catch is still a wip, I know it's not a thing to do).
So far, I've coded this:
void slice(int index) {
try{
_2D_matrix = new T[_height * _width];
}catch(...){
delete[] _2D_matrix;
_height = 0;
_width = 0;
_depth = 0;
throw;
}
_depth = 1;
for (int k = 0; k< _depth; k++) {
for (int j = 0; j< _height; j++) {
for (int i = 0; i< _width; i++) {
_2D_matrix[j * _width + i] =
_3D_matrix[j * _width * _depth + i * _depth + index];
}
}
}
}
I guess the logic behind the assignment done with the nested for cycles is right, but I don't really know how to return the new matrix.
From the main, used to test the code, I'm calling
std::cout << "--------TEST SLICE------------" << std::endl;
Matrice3D<int> m1(3,3,3,17);
std::cout << "Setter su (0,0,2,99)" << std::endl;
m1(0,0,2,91); //setting a value
m1.slice(2);
std::cout << "Matrix obtained from slicing the layer 2: " <<std::endl;
std::cout << m1;
but I keep getting the first layer of the matrix, whatever index I choose for the input.
Create a new class Matrice2D and return that in slice().
The reason why your get total garbage in your code is that you destroy the _depth of the 3D matrix. It's not even the first layer but really just garbage.
The only time new and delete should appear is in classes named something_ptr. You don't need raw pointers here, and you should be returning a 2DMatrix from slice.
template <typename T>
class 2DMatrix;
template <typename T>
class 3DMatrix {
std::vector<T> data;
std::size_t height, width, depth;
public:
3DMatrix(std::size_t height, std::size_t width, std::size_t depth, T value = {})
: data(height * width * depth, value),
height(height),
width(width),
depth(depth)
{}
2DMatrix<T> slice(std::size_t index) {
2DMatrix<T> result(height, width);
for (std::size_t i = index; i < data.size(); i += depth) {
result.data[i / depth] = data[i];
}
return result;
}
// other members
}
template <typename T>
class 2DMatrix {
std::vector<T> data;
std::size_t height, width;
friend class 3DMatrix<T>;
public:
2DMatrix(std::size_t height, std::size_t width, T value = {})
: data(height * width, value),
height(height),
width(width)
{}
// other members
}
Related
I'm trying to do task in C++. I need create this function:
void fillArray(std::array<std::array<int, maxColumns>, maxRows> array, size_t rows, size_t columns) {
}
Right now my example code looks like this:
#include <iostream>
#include <array>
constexpr int maxColumns = 42;
constexpr int maxRows = 334;
void fillArray(std::array<std::array<int, maxColumns>, maxRows> array, size_t rows, size_t columns) {
}
int main()
{
}
I need to fill the array with numbers from 1 to rows*columns starting from [0][0] and diagonally. How to declare and initialize the function with array in this example and then fill it diagonally? Any help would be greatly appreciated!
It should be
template <std::size_t maxColumns, std::size_t maxRows>
void fillArray(std::array<std::array<int, maxColumns>, maxRows>& array) {
// ...
}
Demo
Let's suppose you use a simple one-dimensional valarray (or array if you insist) of the size width * height wrapped in a class:
class Matrix
{
private:
std::valarray<int> _data;
int _width, _height;
public:
Matrix(int width, int height) : _width(width), _height(height), _data(width * height)
{
}
}
Then you can add a member function that maps x, y coordinates to an item reference:
int& item(int x, int y) { return _data[x + _width * y]; }
... and another one for filling it diagonally like this:
void fillDiagonally(int value = 0, int step = 1)
{
for (int i = 0; i < _height + _width; ++i) {
// calculate starting coordinates (along left edge and then bottom edge)
int row = i < _height ? i : _height - 1;
int col = i < _height ? 0 : i - _height + 1;
// move diagonally up and right until you reach margins, while filling-in values
for (int j = 0; j < _width - col && j <= row; ++j) {
item(col + j, row - j) = value;
value += step;
}
}
}
and use it like this:
int main()
{
Matrix m(8, 5);
m.fillDiagonally(1);
}
This way, you don't need to pass the array as an argument, because it's a part of the class. Otherwise you would have to pass it by reference, like you were suggested above.
I created a matrix class and for example a simple cleanup function
template<typename T>
class matrix
{
public:
matrix(int Lenght, int Height)
{
this->container = new T[this->lenght * this->height];
this->lenght = Lenght;
this->height = Height;
}
void Nulling()
{
size_t A = sizeof(T);
for (int i = 0; i < (this->height * this->lenght); i++)
{
this->container)[i] = 0;
//((T*)this->container)[i] = 0; // same result
}
}
T* container = nullptr;
int lenght;
int height;
};
when i == 100226, debugging stops and an error occurs while writing ( i have 512*512 matrix so its 262144 elements)
I think problem is incorrect calculating adress with templates, i try this construction
size_t A = sizeof(T);
for (int i = 0; i < (this->height * this->lenght); i++)
{
this->container[0] = 0;
this->container += A;
}
its cause error on i == 12529
So i dont know what to do now.
In your constructor, the statement
this->container = new T[this->lenght * this->height];
Has undefined behavior, because you are using this->lenght and this->height before they have been assigned values, thus the size of the allocated array is indeterminate. You need to do those assignments first:
matrix(int Lenght, int Height)
{
this->lenght = Lenght;
this->height = Height;
this->container = new T[this->lenght * this->height];
}
Alternatively, use the input values instead of the class members when allocating the array:
matrix(int Lenght, int Height)
{
this->container = new T[Lenght * Height];
this->lenght = Lenght;
this->height = Height;
}
I have created a program which creates a Mandelbrot set. Now I'm trying to make it multithreaded.
// mandelbrot.cpp
// compile with: g++ -std=c++11 mandelbrot.cpp -o mandelbrot
// view output with: eog mandelbrot.ppm
#include <fstream>
#include <complex> // if you make use of complex number facilities in C++
#include <iostream>
#include <cstdlib>
#include <thread>
#include <mutex>
#include <vector>
using namespace std;
template <class T> struct RGB { T r, g, b; };
template <class T>
class Matrix {
public:
Matrix(const size_t rows, const size_t cols) : _rows(rows), _cols(cols) {
_matrix = new T*[rows];
for (size_t i = 0; i < rows; ++i) {
_matrix[i] = new T[cols];
}
}
Matrix(const Matrix &m) : _rows(m._rows), _cols(m._cols) {
_matrix = new T*[m._rows];
for (size_t i = 0; i < m._rows; ++i) {
_matrix[i] = new T[m._cols];
for (size_t j = 0; j < m._cols; ++j) {
_matrix[i][j] = m._matrix[i][j];
}
}
}
~Matrix() {
for (size_t i = 0; i < _rows; ++i) {
delete [] _matrix[i];
}
delete [] _matrix;
}
T *operator[] (const size_t nIndex)
{
return _matrix[nIndex];
}
size_t width() const { return _cols; }
size_t height() const { return _rows; }
protected:
size_t _rows, _cols;
T **_matrix;
};
// Portable PixMap image
class PPMImage : public Matrix<RGB<unsigned char> >
{
public:
unsigned int size;
PPMImage(const size_t height, const size_t width) : Matrix(height, width) { }
void save(const std::string &filename)
{
std::ofstream out(filename, std::ios_base::binary);
out <<"P6" << std::endl << _cols << " " << _rows << std::endl << 255 << std::endl;
for (size_t y=0; y<_rows; y++)
for (size_t x=0; x<_cols; x++)
out << _matrix[y][x].r << _matrix[y][x].g << _matrix[y][x].b;
}
};
/*Draw mandelbrot according to the provided parameters*/
void draw_Mandelbrot(PPMImage & image, const unsigned width, const unsigned height, double cxmin, double cxmax, double cymin, double cymax,unsigned int max_iterations)
{
for (std::size_t ix = 0; ix < width; ++ix)
for (std::size_t iy = 0; iy < height; ++iy)
{
std::complex<double> c(cxmin + ix / (width - 1.0)*(cxmax - cxmin), cymin + iy / (height - 1.0)*(cymax - cymin));
std::complex<double> z = 0;
unsigned int iterations;
for (iterations = 0; iterations < max_iterations && std::abs(z) < 2.0; ++iterations)
z = z*z + c;
image[iy][ix].r = image[iy][ix].g = image[iy][ix].b = iterations;
}
}
int main()
{
const unsigned width = 1600;
const unsigned height = 1600;
PPMImage image(height, width);
int parts = 8;
std::vector<int>bnd (parts, image.size);
std::thread *tt = new std::thread[parts - 1];
time_t start, end;
time(&start);
//Lauch parts-1 threads
for (int i = 0; i < parts - 1; ++i) {
tt[i] = std::thread(draw_Mandelbrot,ref(image), width, height, -2.0, 0.5, -1.0, 1.0, 10);
}
//Use the main thread to do part of the work !!!
for (int i = parts - 1; i < parts; ++i) {
draw_Mandelbrot(ref(image), width, height, -2.0, 0.5, -1.0, 1.0, 10);
}
//Join parts-1 threads
for (int i = 0; i < parts - 1; ++i)
tt[i].join();
time(&end);
std::cout << difftime(end, start) << " seconds" << std::endl;
image.save("mandelbrot.ppm");
delete[] tt;
return 0;
}
Now every thread draws the complete fractal (look in main()). How can I let the threads draw different parts of the fractal?
You're making this (quite a lot) harder than it needs to be. This is the sort of task to which OpenMP is almost perfectly suited. For this task it gives almost perfect scaling with a bare minimum of effort.
I modified your draw_mandelbrot by inserting a pragma before the outer for loop:
#pragma omp parallel for
for (int ix = 0; ix < width; ++ix)
for (int iy = 0; iy < height; ++iy)
Then I simplified your main down to:
int main() {
const unsigned width = 1600;
const unsigned height = 1600;
PPMImage image(height, width);
clock_t start = clock();
draw_Mandelbrot(image, width, height, -2.0, 0.5, -1.0, 1.0, 10);
clock_t stop = clock();
std::cout << (double(stop - start) / CLOCKS_PER_SEC) << " seconds\n";
image.save("mandelbrot.ppm");
return 0;
}
On my (fairly slow) machine, your original code ran in 4.73 seconds. My modified code ran in 1.38 seconds. That's an improvement of 3.4x out of code that's nearly indistinguishable from a trivial single-threaded version.
Just for what it's worth, I did a bit more rewriting to get this:
// mandelbrot.cpp
// compile with: g++ -std=c++11 mandelbrot.cpp -o mandelbrot
// view output with: eog mandelbrot.ppm
#include <fstream>
#include <complex> // if you make use of complex number facilities in C++
#include <iostream>
#include <cstdlib>
#include <thread>
#include <mutex>
#include <vector>
using namespace std;
template <class T> struct RGB { T r, g, b; };
template <class T>
struct Matrix
{
std::vector<T> data;
size_t rows;
size_t cols;
class proxy {
Matrix &m;
size_t index_1;
public:
proxy(Matrix &m, size_t index_1) : m(m), index_1(index_1) { }
T &operator[](size_t index) { return m.data[index * m.rows + index_1]; }
};
class const_proxy {
Matrix const &m;
size_t index_1;
public:
const_proxy(Matrix const &m, size_t index_1) : m(m), index_1(index_1) { }
T const &operator[](size_t index) const { return m.data[index * m.rows + index_1]; }
};
public:
Matrix(size_t rows, size_t cols) : data(rows * cols), rows(rows), cols(cols) { }
proxy operator[](size_t index) { return proxy(*this, index); }
const_proxy operator[](size_t index) const { return const_proxy(*this, index); }
};
template <class T>
std::ostream &operator<<(std::ostream &out, Matrix<T> const &m) {
out << "P6" << std::endl << m.cols << " " << m.rows << std::endl << 255 << std::endl;
for (size_t y = 0; y < m.rows; y++)
for (size_t x = 0; x < m.cols; x++) {
T pixel = m[y][x];
out << pixel.r << pixel.g << pixel.b;
}
return out;
}
/*Draw Mandelbrot according to the provided parameters*/
template <class T>
void draw_Mandelbrot(T & image, const unsigned width, const unsigned height, double cxmin, double cxmax, double cymin, double cymax, unsigned int max_iterations) {
#pragma omp parallel for
for (int ix = 0; ix < width; ++ix)
for (int iy = 0; iy < height; ++iy)
{
std::complex<double> c(cxmin + ix / (width - 1.0)*(cxmax - cxmin), cymin + iy / (height - 1.0)*(cymax - cymin));
std::complex<double> z = 0;
unsigned int iterations;
for (iterations = 0; iterations < max_iterations && std::abs(z) < 2.0; ++iterations)
z = z*z + c;
image[iy][ix].r = image[iy][ix].g = image[iy][ix].b = iterations;
}
}
int main() {
const unsigned width = 1600;
const unsigned height = 1600;
Matrix<RGB<unsigned char>> image(height, width);
clock_t start = clock();
draw_Mandelbrot(image, width, height, -2.0, 0.5, -1.0, 1.0, 255);
clock_t stop = clock();
std::cout << (double(stop - start) / CLOCKS_PER_SEC) << " seconds\n";
std::ofstream out("mandelbrot.ppm", std::ios::binary);
out << image;
return 0;
}
On my machine, this code runs in about 0.5 to 0.6 seconds.
As to why I made these changes: mostly to make it faster, cleaner, and simpler. Your Matrix class allocated a separate block of memory for each row (or perhaps column--didn't pay very close of attention). This allocates one contiguous block of the entire matrix instead. This eliminates a level of indirection to get to the data, and increases locality of reference, thus improving cache usage. It also reduces the total amount of data used.
Changing from using time to using clock to do the timing was to measure CPU time instead of wall time (and typically improve precision substantially as well).
Getting rid of the PPMImage class was done simply because (IMO) having a PPImage class that derives from a Matrix class just doesn't make much (if any) sense. I suppose it works (for a sufficiently loose definition of "work") but it doesn't strike me as good design. If you insist on doing it at all, it should at least be private derivation, because you're just using the Matrix as a way of implementing your PPMImage class, not (at least I certainly hope not) trying to make assertions about properties of PPM images.
If, for whatever, reason, you decide to handle the threading manually, the obvious way of dividing the work up between threads would still be by looking at the loops inside of draw_mandelbrot. The obvious one would be to leave your outer loop alone, but send the computation for each iteration off to a thread pool:
for (int ix = 0; ix < width; ++ix)
compute_thread(ix);
where the body of compute_thread is basically this chunk of code:
for (int iy = 0; iy < height; ++iy)
{
std::complex<double> c(cxmin + ix / (width - 1.0)*(cxmax - cxmin), cymin + iy / (height - 1.0)*(cymax - cymin));
std::complex<double> z = 0;
unsigned int iterations;
for (iterations = 0; iterations < max_iterations && std::abs(z) < 2.0; ++iterations)
z = z*z + c;
image[iy][ix].r = image[iy][ix].g = image[iy][ix].b = iterations;
}
There would obviously be a little work involved in passing the correct data to the compute thread (each thread should be pass a reference to a slice of the resulting picture), but that would be an obvious and fairly clean place to divide things up. In particular it divides the job up into enough tasks that you semi-automatically get pretty good load balancing (i.e., you can keep all the cores busy) but large enough that you don't waste massive amounts of time on communication and synchronization between the threads.
As to the result, with the number of iterations set to 255, I get the following (scaled to 25%):
...which is pretty much as I'd expect.
One of the big issues with this approach is that different regions take different amounts of time to calculate.
A more general approach is.
Start 1 source thread.
Start N worker threads.
Start 1 sink thread.
Create 2 thread safe queues (call them the source queue and the sink queue).
Divide the image into M (many more than N) pieces.
The source thread pushes pieces into the source queue
The workers pull piecse from the source queue, convert the pieces into result fragments, and pushes those fragments into the sink queue.
The sink thread takes fragments from the sink queue and combines them into the final image.
By dividing up the work this way, all the worker threads will be busy all the time.
You can divide the fractal into pieces by divide the start and end of the fractal with the screen dimension:
$this->stepsRe = (double)((($this->startRe * -1) + ($this->endeRe)) / ($this->size_x-1));
$this->stepsIm = (double)((($this->startIm * -1) + ($this->endeIm)) / ($this->size_y-1));
I have the assignment to finish a mandelbrot program in C++. I'm not that good in C++, I prefer Java or C# but this has to be done in C++. I got some sample code which I have to finish. I'm trying to put the drawing code in the main (between the works comments) into a method (draw_Mandelbrot). The code in the main method works and gives me a nice mandelbrot image but when I use the draw_Mandelbrot method (and comment the draw code in main) I get a grey rectangle image as output. How can I make the draw_Mandelbrot method work? The code above the draw_Mandelbrot method is all sample code and not created by myself.
// mandelbrot.cpp
// compile with: g++ -std=c++11 mandelbrot.cpp -o mandelbrot
// view output with: eog mandelbrot.ppm
#include <fstream>
#include <complex> // if you make use of complex number facilities in C++
#include <iostream>
#include <cstdlib>
#include <complex>
using namespace std;
template <class T> struct RGB { T r, g, b; };
template <class T>
class Matrix {
public:
Matrix(const size_t rows, const size_t cols) : _rows(rows), _cols(cols) {
_matrix = new T*[rows];
for (size_t i = 0; i < rows; ++i) {
_matrix[i] = new T[cols];
}
}
Matrix(const Matrix &m) : _rows(m._rows), _cols(m._cols) {
_matrix = new T*[m._rows];
for (size_t i = 0; i < m._rows; ++i) {
_matrix[i] = new T[m._cols];
for (size_t j = 0; j < m._cols; ++j) {
_matrix[i][j] = m._matrix[i][j];
}
}
}
~Matrix() {
for (size_t i = 0; i < _rows; ++i) {
delete [] _matrix[i];
}
delete [] _matrix;
}
T *operator[] (const size_t nIndex)
{
return _matrix[nIndex];
}
size_t width() const { return _cols; }
size_t height() const { return _rows; }
protected:
size_t _rows, _cols;
T **_matrix;
};
// Portable PixMap image
class PPMImage : public Matrix<RGB<unsigned char> >
{
public:
PPMImage(const size_t height, const size_t width) : Matrix(height, width) { }
void save(const std::string &filename)
{
std::ofstream out(filename, std::ios_base::binary);
out <<"P6" << std::endl << _cols << " " << _rows << std::endl << 255 << std::endl;
for (size_t y=0; y<_rows; y++)
for (size_t x=0; x<_cols; x++)
out << _matrix[y][x].r << _matrix[y][x].g << _matrix[y][x].b;
}
};
void draw_Mandelbrot(PPMImage image, const unsigned width, const unsigned height, double cxmin, double cxmax, double cymin, double cymax,unsigned int max_iterations)
{
for (std::size_t ix = 0; ix < width; ++ix)
for (std::size_t iy = 0; iy < height; ++iy)
{
std::complex<double> c(cxmin + ix / (width - 1.0)*(cxmax - cxmin), cymin + iy / (height - 1.0)*(cymax - cymin));
std::complex<double> z = 0;
unsigned int iterations;
for (iterations = 0; iterations < max_iterations && std::abs(z) < 2.0; ++iterations)
z = z*z + c;
image[iy][ix].r = image[iy][ix].g = image[iy][ix].b = iterations;
}
}
int main()
{
const unsigned width = 1600;
const unsigned height = 1600;
PPMImage image(height, width);
//image[y][x].r = image[y][x].g = image[y][x].b = 255; // white pixel
//image[y][x].r = image[y][x].g = image[y][x][b] = 0; // black pixel
//image[y][x].r = image[y][x].g = image[y][x].b = 0; // black pixel
//// red pixel
//image[y][x].r = 255;
//image[y][x].g = 0;
//image[y][x].b = 0;
draw_Mandelbrot(image, width, height, -2.0, 0.5, -1.0, 1.0, 10);
//works
//double cymin = -1.0;
//double cymax = 1.0;
//double cxmin = -2.0;
//double cxmax = 0.5;
//unsigned int max_iterations = 100;
//for (std::size_t ix = 0; ix < width; ++ix)
// for (std::size_t iy = 0; iy < height; ++iy)
// {
// std::complex<double> c(cxmin + ix / (width - 1.0)*(cxmax - cxmin), cymin + iy / (height - 1.0)*(cymax - cymin));
// std::complex<double> z = 0;
// unsigned int iterations;
// for (iterations = 0; iterations < max_iterations && std::abs(z) < 2.0; ++iterations)
// z = z*z + c;
// image[iy][ix].r = image[iy][ix].g = image[iy][ix].b = iterations;
// }
//works
image.save("mandelbrot.ppm");
return 0;
}
Output image when using the code in the main method
You're passing the image by value, so the function works on a separate image to the one in main, which is left in its initial state.
Either pass by reference:
void draw_Mandelbrot(PPMImage & image, ...)
or return a value:
PPMImage draw_Mandelbrot(...) {
PPMImage image(height, width);
// your code here
return image;
}
// in main
PPMImage image = draw_Mandelbrot(...);
I had a previous question about a stack overflow error and switch to vectors for my arrays of objects. That question can be referenced here if needed: How to get rid of stack overflow error
My current question is however, how do I speed up the initialization of the vectors. My current method currently takes ~15 seconds. Using arrays instead of vectors it took like a second with a size of arrays small enough that didn't throw the stack overflow error.
Here is how I am initializing it:
in main.cpp I initialize my dungeon object:
dungeon = Dungeon(0, &textureHandler, MIN_X, MAX_Y);
in my dungeon(...) constructor, I initialize my 5x5 vector of rooms and call loadDungeon:
Dungeon::Dungeon(int dungeonID, TextureHandler* textureHandler, int topLeftX, int topLeftY)
{
currentRoomRow = 0;
currentRoomCol = 0;
for (int r = 0; r < MAX_RM_ROWS; ++r)
{
rooms.push_back(vector<Room>());
for (int c = 0; c < MAX_RM_COLS; ++c)
{
rooms[r].push_back(Room());
}
}
loadDungeon(dungeonID, textureHandler, topLeftX, topLeftY);
}
my Room constructor populates my 30x50 vector of cells (so I can set them up in the loadDungeon function):
Room::Room()
{
for (int r = 0; r < MAX_ROWS; ++r)
{
cells.push_back(vector<Cell>());
for (int c = 0; c < MAX_COLS; ++c)
{
cells[r].push_back(Cell());
}
}
}
My default cell constructor is simple and isn't doing much but I'll post it anyway:
Cell::Cell()
{
x = 0;
y = 0;
width = 16;
height = 16;
solid = false;
texCoords.push_back(0);
texCoords.push_back(0);
texCoords.push_back(1);
texCoords.push_back(0);
texCoords.push_back(1);
texCoords.push_back(1);
texCoords.push_back(0);
texCoords.push_back(1);
}
And lastly my loadDungeon() function will set up the cells. Eventually this will read from a file and load the cells up but for now I would like to optimize this a bit if possible.
void Dungeon::loadDungeon(int dungeonID, TextureHandler* textureHandler, int topLeftX, int topLeftY)
{
int startX = topLeftX + (textureHandler->getSpriteWidth()/2);
int startY = topLeftY - (textureHandler->getSpriteHeight()/2);
int xOffset = 0;
int yOffset = 0;
for (int r = 0; r < MAX_RM_ROWS; ++r)
{
for (int c = 0; c < MAX_RM_COLS; ++c)
{
for (int cellRow = 0; cellRow < rooms[r][c].getMaxRows(); ++cellRow)
{
xOffset = 0;
for (int cellCol = 0; cellCol < rooms[r][c].getMaxCols(); ++cellCol)
{
rooms[r][c].setupCell(cellRow, cellCol, startX + xOffset, startY - yOffset, textureHandler->getSpriteWidth(), textureHandler->getSpriteHeight(), false, textureHandler->getSpriteTexCoords("grass"));
xOffset += textureHandler->getSpriteWidth();
}
yOffset += textureHandler->getSpriteHeight();
}
}
}
currentDungeon = dungeonID;
currentRoomRow = 0;
currentRoomCol = 0;
}
So how can I speed this up so it doesn't take ~15 seconds to load up every time. I feel like it shouldn't take 15 seconds to load a simple 2D game.
SOLUTION
Well my solution was to use std::vector::reserve call (rooms.reserve in my code and it ended up working well. I changed my function Dungeon::loadDungeon to Dungeon::loadDefaultDungeon because it now loads off a save file.
Anyway here is the code (I got it down to about 4-5 seconds from ~15+ seconds in debug mode):
Dungeon::Dungeon()
{
rooms.reserve(MAX_RM_ROWS * MAX_RM_COLS);
currentDungeon = 0;
currentRoomRow = 0;
currentRoomCol = 0;
}
void Dungeon::loadDefaultDungeon(TextureHandler* textureHandler, int topLeftX, int topLeftY)
{
int startX = topLeftX + (textureHandler->getSpriteWidth()/2);
int startY = topLeftY - (textureHandler->getSpriteHeight()/2);
int xOffset = 0;
int yOffset = 0;
cerr << "Loading default dungeon..." << endl;
for (int roomRow = 0; roomRow < MAX_RM_ROWS; ++roomRow)
{
for (int roomCol = 0; roomCol < MAX_RM_COLS; ++roomCol)
{
rooms.push_back(Room());
int curRoom = roomRow * MAX_RM_COLS + roomCol;
for (int cellRow = 0; cellRow < rooms[curRoom].getMaxRows(); ++cellRow)
{
for (int cellCol = 0; cellCol < rooms[curRoom].getMaxCols(); ++cellCol)
{
rooms[curRoom].setupCell(cellRow, cellCol, startX + xOffset, startY - yOffset, textureHandler->getSpriteWidth(), textureHandler->getSpriteHeight(), false, textureHandler->getSpriteTexCoords("default"), "default");
xOffset += textureHandler->getSpriteWidth();
}
yOffset += textureHandler->getSpriteHeight();
xOffset = 0;
}
cerr << " room " << curRoom << " complete" << endl;
}
}
cerr << "default dungeon loaded" << endl;
}
Room::Room()
{
cells.reserve(MAX_ROWS * MAX_COLS);
for (int r = 0; r < MAX_ROWS; ++r)
{
for (int c = 0; c < MAX_COLS; ++c)
{
cells.push_back(Cell());
}
}
}
void Room::setupCell(int row, int col, float x, float y, float width, float height, bool solid, /*std::array<float, 8>*/ vector<float> texCoords, string texName)
{
cells[row * MAX_COLS + col].setup(x, y, width, height, solid, texCoords, texName);
}
void Cell::setup(float x, float y, float width, float height, bool solid, /*std::array<float,8>*/ vector<float> t, string texName)
{
this->x = x;
this->y = y;
this->width = width;
this->height = height;
this->solid = solid;
for (int i = 0; i < t.size(); ++i)
this->texCoords.push_back(t[i]);
this->texName = texName;
}
It seems wasteful to have so many dynamic allocations. You can get away with one single allocation by flattening out your vector and accessing it in strides:
std::vector<Room> rooms;
rooms.resize(MAX_RM_ROWS * MAX_RM_COLS);
for (unsigned int i = 0; i != MAX_RM_ROWS; ++i)
{
for (unsigned int j = 0; j != MAX_RM_COLS; ++j)
{
Room & r = rooms[i * MAX_RM_COLS + j];
// use `r` ^^^^^^^^^^^^^^^^^^^-----<< strides!
}
}
Note how resize is performed exactly once, incurring only one single allocation, as well as default-constructing each element. If you'd rather construct each element specifically, use rooms.reserve(MAX_RM_ROWS * MAX_RM_COLS); instead and populate the vector in the loop.
You may also wish to profile with rows and columns swapped and see which is faster.
Since it seems that your vectors have their size defined at compile time, if you can use C++11, you may consider using std::array instead of std::vector. std::array cannot be resized and lacks many of the operations in std::vector, but is much more lightweight and it seems a good fit for what you are doing.
As an example, you could declare cells as:
#include <array>
/* ... */
std::array<std::array<Cell, MAX_COLS>, MAX_ROWS> cells;
UPDATE: since a locally defined std::array allocates its internal array on the stack, the OP will experience a stack overflow due to the considerably large size of the arrays. Still, it is possible to use an std::array (and its benefits compared to using std::vector), by allocating the array on the heap. That can be done by doing something like:
typedef std::array<std::array<Cell, MAX_COLS>, MAX_ROWS> Map;
Map* cells;
/* ... */
cells = new Map();
Even better, smart pointers can be used:
#include <memory>
/* ... */
std::unique_ptr<Map> cells;
cells = std::unique_ptr(new Map());