I recently started working with C++ for numerical computations where I want to use a Struct Operators to store 3D Fields over the course of the simulation.
I create the 3D arrays on the heap with
const unsigned int RES = 256;
auto arr3D = new double [RES][RES][RES];
because from what I've tested this approach is faster than using either Boost_multiarr, Eigen Tensor or nested Vectors.
So far this worked fine with my minimalistic Test.cpp but when I try to implement these same 3D Arrays as Members of my Struct Operators, I cannot use the auto command anymore:
const unsigned int RES = 256;
struct Operators {
public:
std::complex<double>*** wfc; // edited, see 'spitconsumers' comment
Operators(Settings &set) { // just another structure used by Operators
wfc = new std::complex<double> [RES][RES][RES];
// ...Initializing wfc using Settings
};
In this case I found no way of declaring wfc, such that I do not get errors of the type
error: cannot convert 'std::complex (*)[256][256]' to 'std::complex***' in assignment
So my question is how to correctly declare the 3D Array wfc and whether maintaining this Structure approach is at all possible/useful. Would it generally be faster to access the wfc[i][j][k] if wfc was not member of a structure? (I will have to do this ~10^6 times)
Thanks in advance!
The error message returns the correct declaration, std::complex<double>(*wfc)[RES][RES];.
const unsigned int RES = 256;
struct Settings {};
struct Operators {
public:
std::complex<double>(*wfc)[RES][RES]; // edited, see 'spitconsumers' comment
Operators(Settings& set) { // just another structure used by Operators
wfc = new std::complex<double>[RES][RES][RES];
// ...Initializing wfc using Settings
// Setting the last element
wfc[254][254[254] = 42;
};
}
In your error message, the compiler is telling you that new complex [RES][RES][RES] is returning the type std::complex (*)[256][256].
double wfc should be std::complex*** wfc.
Class member types must be known, and since you are not initialising the member variable while it is being defined, the compiler cannot deduce the type of the variable, which is why auto doesn't work as a class member type, when you aren't immediately assigning something to it. But you are assigning the value directly to the variable in your test program, allowing the compiler to deduce the type there.
Also, the new operator is returning a pointer to an array of uninitialised std::complex<double>** pointers. You need to call new for all of the uninitialised pointers.
std::complex<double>*** wfc;
wfc = new std::complex<double>** [RES];
for(int i = 0; i < RES; i++) {
wfc[i] = new std::complex<double>* [RES];
for(int j = 0; j < RES; j++) {
wfc[i][j] = new std::complex<double> [RES];
for(int k = 0; k < RES; k++) {
wfc[i][j][k] = /* insert default value here */;
}
}
}
Related
I am still perfecting the art of posting here so bear with me, I will edit and fix anything suggested!
I have a homework that requires me to create functions that manipulate vectors. The "catch" is that all the data passed to the function is passed by reference to this struct:
struct Vector { // this struct must stay as is
int sze = 0; // "size" took out i for compatability
int capacity = 0;
int * data = nullptr ;
}a,b,c;
i.e.
void construct_vector ( Vector& v, int size= 0, int initVal= 0);
The problem that I am having is in the function construct_vector() I have to, you guessed it, construct a vector and use int* data to point to the vector on the heap? I am not positive about that last part). I just know I have to use the int pointer to point to the vector created within the construct function, and cannot for the life of me figure out how to do that.
An example of what I am trying:
void construct_vector ( Vector &v, int size, int initVal){
std::vector<int> t(size,initVal);
*v.data = &t ; // ERROR: Assigning to 'int' from incompatible type 'std::vector<int> *'
v.capacity = size; //
v.sze = size;
for (int i=0; i < t.size(); i++){
/* I originally tried to implement a
dynamic int pointer here but I cannot change int* data
to int*data[sze] within the struct*/
}
}
The reason int * data must point to the vector is because the data is passed to the subsequent functions by reference to struct member v:
void destroy_vector ( Vector & v );
void copy_data ( Vector & v );
Edit: My problem was that I misunderstood the objective of my assignment but I think the answers I received can really help people understand dynamic memory and how it should be used within functions. So I am going to leave everything as is!
You have two problems here:
std::vector<int> t(size,initVal);
*v.data = &t ; // ERROR: Assigning to 'int' from
First, *v.data and &t are different types, one is int, the other is a pointer to a vector of ints.
You can get it compile with (but you SHOULD NOT, see the second problem)
v.data = t.data();
The other problem is the vector is local to the function. As soon as the function returns, your pointer will dangle.
So the right solution for your problem is using a dynamic array:
v.data = new int[size];
Don't forget to delete[] it in the struct's destructor when you are done using it:
delete [] data;
Instead of
std::vector<int> t(size,initVal);
*v.data = &t ;
You need
v.data = new int[size];
To fill up the object with the input value, use
for ( int i = 0; i < size; ++i )
{
v.data[i] = initVal;
}
You can use std::fill to make your code a bit simpler.
std::fill(v.data, v.data+size, initVal);
Make sure to follow The Rule of Three when you manage dynamic memory yourself.
I was wondering if it was possible to create an array of objects when the object needs things passed into it for the constructor. I want something like this:
MyClass *myVar;
myVar = new MyClass[num]; // I would like to specify the array size after declaration
int i = 0;
for(i = 0;i < num;i++)
myVar[i] = new MyClass(0,0); // I would also like to populate the array with new objects
I know that this works:
MyClass *myVar;
myVar = new MyClass[num];
but this only works when the constructor has nothing passed into it. Is what I am trying to do possible? If so, how do I do it?
EDIT: I found out how to do it with using arrays. Here is how I did it:
MyClass **myVar;
myVar = new MyClass *[num];
for(i = 0;i < num;i++)
myVar[0] = new MyClass(0,0);
I would use vectors and such but my teacher has told us to use basic arrays whenever possible. The above solution I actually got from some code my teacher wrote. Thank you all for your help!
MyClass *myVar;
myVar = new MyClass[num];
Actually in this form you cannot invoke constructor which takes parameter(s). It is not allowed by the language specification.
However, if you use std::vector, which I recommend you to use, then you can create a vector calling non-default constructor as:
#include <vector> //header file where std::vector is defined
std::vector<MyClass> arr(num, MyClass(10,20));
It creates a vector of num elements, each element is created by calling copy-constructor of the class, passing MyClass(10,20) as argument to it.
The vector is also good because now you dont need to manage memory yourself. Neither manual allocation, nor manual deallocation. Plus, you can know the number of elements by calling arr.size() anytime. You always know how many elements the vector contains. You can also add elements anytime, just by calling .push_back() member function as:
arr.push_back(MyClass(20,30));
And now you can access elements, just like you access array, i.e by using index:
f(arr[i]); // 0 <= i < arr.size();
Additionally, you can use iterators which facilitate idiomatic programming, enabling you to use various algorithmic functions from <algorithm> header as:
#include <algorithm> //header file where std::for_each is defined
std::for_each(arr.begin(), arr.end(), f);
where f is function which takes one argument of type MyClass& (or MyClass const &) depending on what you want to do in f.
In C++11, you can use lambda as:
std::for_each(arr.begin(), arr.end(), [](const MyClass & m)
{
//working with m
});
In C++0x, this grammar works, which can call the non-default constructor in new expression:
MyClass *myVar;
myVar = new MyClass[2]{{10, 20},{20, 30}};
But I doubt if it works when the number of elements in available only at run time.
The vector approach would be better, as shown in Nawaz's answer.
Pointer to pointer is equivalent to 1. array of pointers, and 2. vector<T*> vector of pointers. One way I've done this in the past is using a double pointer. This approach eliminates the overhead of vector data structure and preferred memory efficient is needed.
MyClass ** myvar;
myvar = new Myclass*[num]
for(int i = 0; i < num; i++){
*(myvar+i) = new Myclass(i);}
Works with pretty much any control structure you can imagine. The drawback is that the allocation of memory is not contiguous and my affect speed for large number of num.
You can do something like this too:
MyClass *myVar[num];
for(int i = 0; i < num; i += 1)
{
myVar[i] = new MyClass(0, 0);
}
Actually, you can use a placement new to handle this:
MyClass * myVar;
myVar = reinterpret_cast<MyClass *>(new char[num * sizeof(MyClass)]);
int i = 0;
for (i = 0; i < num; i++) {
new(&myVar[i]) MyClass(0,0);
}
#Nawaz answer is really good about using vectors, but didn't work for me because it create vector of the same objects (all of them reference to the same object)
class Graph
{
public:
Graph(long V); // none default Constructor
}
std::vector<Graph> myGraph;
for (int i = 0; i < T; i++) // read all graphs
{
Graph newGraph(N);
myGraph.push_back(newGraph);
}
I would like to allocate memory for 3D array inside a function.
void*** myFunc(int myType){
double ***p2DArray;
// Allocate memory
p2DArray = new double**[HEIGHT];
for (int i = 0; i < HEIGHT; ++i) {
p2DArray[i] = new double*[WIDTH];
for (int j = 0; j < WIDTH; ++j) {
p2DArray[i][j] = new double[DEPTH];
for (int k = 0; k < LENGTH; ++k)
p2DArray[i][j][k] = 0;
}
}
return p2DArray;
}
In the code above, double type is created. My question is that how to create different data types according to the myType parameter? I don't want to copy the above code twice by the way.
When you find yourself wanting to pass a type to a function in C++ you need to use templates:
template <typename T>
T*** func() {
T*** array;
array = new T**[HEIGHT];
...
return array
}
You can then call this function like so:
double*** var = func<double>();
When the compiler sees a call to func<double>() it will look up the template for func() and generate a regular (non-template) function that simply replaces all the T's with double.
For this design you should consider using a idiomatic C++ container such as a vector. This provides a number of features that will make your code safer and more readable such as iterators.
One option for vectors is simply to replace your array with vector<vector<vector<T>>>. Another option is to use a single vector<T> with a size of WIDTH * HEIGHT * DEPTH.
const int ADJ_MATRIX[VERTEX_NUM][VERTEX_NUM]={
{0,1,1,0,0,0,0,0},
{1,0,0,1,1,0,0,0},
{1,0,0,0,0,1,1,0},
{0,1,0,0,0,0,0,1},
{0,1,0,0,0,0,0,1},
{0,0,1,0,0,0,1,0},
{0,0,1,0,0,1,0,0},
{0,0,0,1,1,0,0,0}
};
typedef struct {
int vertex;
int matrix[VERTEX_NUM][VERTEX_NUM];
int vNum;
int eNum;
}Graph;
void buildGraph(Graph *graph){
graph->vNum = VERTEX_NUM;
graph->eNum = EDGE_NUM;
graph->matrix = ADJ_MATRIX;
}
The error occurs in this sentence:
graph->matrix = ADJ_MATRIX;
I am new to c++. please tell me why this problem occur and how to solve it?
I want to assign ADJ_MATRIX to the matrix in struct.
As was said, you can't assign arrays in C++. This is due to the compiler being a meanie, because the compiler can. It just won't let you do it...
... unless you trick it ;)
template <typename T, int N>
struct square_matrix {
T data[N][N];
};
square_matrix<int, 10> a;
square_matrix<int, 10> b;
a = b; // fine, and actually assigns the .data arrays
a.data = b.data; // not allowed, compiler won't let you assign arrays
The catch? Now the code needs some little things:
const square_matrix<int, VERTEX_NUM> ADJ_MATRIX={{
// blah blah
}}; // extra set of braces
typedef struct {
int vertex;
square_matrix<int, VERTEX_NUM> matrix;
int vNum;
int eNum;
}Graph;
void buildGraph(Graph *graph){
graph->vNum = VERTEX_NUM;
graph->eNum = EDGE_NUM;
graph->matrix = ADJ_MATRIX; // no change
}
And to access the cells, now we need to use graph->matrix.data[1][2]. This can be mitigated by overloading operator[] or operator() for square_matrix. However, this is now getting terribly close to the new std::array class, or the Boost equivalent boost::array, so it might be wise to consider those instead.
Unfortunately (or maybe fortunately, who knows...) you can't just assign one array to another in C++.
If you want to copy an array, you will need to either copy each of it's elements into a new array one by one, or use the memcpy() function:
for( int i = 0; i < VERTEX_NUM; i++ )
for( int j = 0; j < VERTEX_NUM; j++ )
graph->matrix[i][j] = ADJ_MATRIX[i][j];
or
memcpy( graph->matrix, ADJ_MATRIX, VERTEX_NUM * VERTEX_NUM * sizeof(int) );
Arrays are not assignable. You can use memcpy:
memcpy(graph->matrix, ADJ_MATRIX, sizeof(graph->matrix));
You cannot assign an array to another array. You will need to copy the elements from the source to the destination index by index, or use memcpy to copy the data. Array assignment like this is not allowed
You are trying to assign your variable address of a constant data,
try using
memcpy(graph->matrix,ADJ_MATRIX,sizeof(ADJ_MATRIX));//using sizeof(graph->matrix) is safer.
You can't use an array in assignments. You may use cycles or memcpy instead
memcpy(graph->matrix, ADJ_MATRIX, VERTEX_NUM * VERTEX_NUM * sizeof(int));
or
for(int i = 0; i < VERTEX_NUM; ++i){
for(int j = 0; j < VERTEX_NUM; ++j){
graph->matrix[i][j] = ADJ_MATRIX[i][j];
}
}
The error is thrown, because int matrix[VERTEX_NUM][VERTEX_NUM] in a structure definition means that each structure will have a 2D array of integers of the predefined size and matrix is going to be pointing to its first element. The thing is that matrix cannot be assigned to an arbitrary address, because it's a const pointer i.e. its value (the address it's pointing to) cannot change.
You have 2 options here: you can either use memcpy or some stl algorithms to copy the ADJ_MATRIX into matrix directly or you can declare matrix as a pointer and do the assignment that is currently produces an error.
The latter can be done in the following way:
typedef struct {
int vertex;
const int (*matrix)[VERTEX_NUM];
int vNum;
int eNum;
}Graph;
Thus you can do graph->matrix = ADJ_MATRIX assignment, but you won't be able to modify the individual items in matrix due to constness. This means, graph->matrix[0][1] = 3; is not allowed, while you can read the elements freely.
I am trying to return an array Data Member from one smaller 2D Array Object, and trying to insert the array into a larger 2D array object. But when attempting this, I came into two problems.
First problem is that I want to return the name of the 2D array, but I do not know how to properly syntax to return 2D Array name.
This is what my 2D Array data member looks like
private:
int pieceArray[4][4];
// 2D Smaller Array
and I want to return this array into a function, but this one causes a compiler error:
int Piece::returnPiece()
{
return pieceArray; //not vaild
// return the 2D array name
}
I tired using this return type and it worked:
int Piece::returnPiece()
{
return pieceArray[4][4];
}
But I am unsure if this is what I want, as I want to return the array and all of it's content.
The other problem is the InsertArray() function, where I would put the returnPiece() function in the InsertArray()'s argument.
The problem with the InsertArray() is the argument, heres the code for it:
void Grid::InsertArray( int arr[4][4] ) //Compiler accepts, but does not work
{
for(int i = 0; i < x_ROWS ; ++i)
{
for (int j = 0; j < y_COLUMNS ; ++j)
{
squares[i][j] = arr[i][j];
}
}
}
The problem with this is that it does not accept my returnPiece(), and if i remove the "[4][4]", my compiler does not accept.
Mostly all these are syntax errors, but how do I solve these problems?
Returning the whole pieceArray in returnPiece()
The correct syntax for the argument in InsertArray()
The argument of InsertArray() accepting the returnPiece()
These 3 are the major problems that I need help with, and had the same problem when I attempt to use the pointer pointer method. Does anyone know how to solve these 3 problems?
When passing your array around, you have to decide whether or not you want to make a copy of the array, or if you just want to return a pointer to the array. For returning arrays, you can't (easily) return a copy - you can only return a pointer (or reference in C++). For example:
// Piece::returnPiece is a function taking no arguments and returning a pointer to a
// 4x4 array of integers
int (*Piece::returnPiece(void))[4][4]
{
// return pointer to the array
return &pieceArray;
}
To use it, call it like so:
int (*arrayPtr)[4][4] = myPiece->returnPiece();
int cell = (*arrayPtr)[i][j]; // cell now stores the contents of the (i,j)th element
Note the similarity between the type declaration and using it - the parentheses, dereferencing operator *, and brackets are in the same places.
Your declaration for Grid::InsertArray is correct - it takes one argument, which is a 4x4 array of integers. This is call-by-value: whenever you call it, you make a copy of your 4x4 array, so any modification you make are not reflected in the array passed in. If you instead wanted to use call-by-reference, you could pass a pointer to an array instead:
// InsertArray takes one argument which is a pointer to a 4x4 array of integers
void Grid::InsertArray(int (*arr)[4][4])
{
for(int i = 0; i < x_ROWS; i++)
{
for(int j = 0; j < y_COLUMNS ; j++)
squares[i][j] = (*arr)[i][j];
}
}
These type declarations with pointers to multidimensional arrays can get really confusing fast. I recommend making a typedef for it like so:
// Declare IntArray4x4Ptr to be a pointer to a 4x4 array of ints
typedef int (*IntArray4x4Ptr)[4][4];
Then you can declare your functions much more readable:
IntArray4x4Ptr Piece::returnPiece(void) { ... }
void Grid::InsertArray(IntArray4x4Ptr arr) { ... }
You can also use the cdecl program to help decipher complicated C/C++ types.
It seems like you need to read up more on pointers in C++ and on pass by reference vs. pass by value.
Your returnPiece method is defined as returning the value of a single cell. Given the index (e.g., [4][4]) you return a copy of the contents of that cell, so you won't be able to change it, or more correctly, changing it would change the copy.
I'm sure someone will give you the correct syntax, but I would really recommend learning this stuff since otherwise you may use the code that you do get incorrectly.
Here is how I would do it:
class Array {
public:
Array() {
for (int i = 0; i < 4; ++i)
{
for (int j = 0; j < 4; ++j)
{
(*this)(i, j) = 0;
}
}
}
int &operator()(int i, int j)
{
return pieceArray[i][j];
}
private:
int pieceArray[4][4];
};
You can then do something like:
Array x; // create 4x4 array
x(1, 2) = 3; // modify element
Trouble with Adam Rosenfield's arrayPtr in that Piece::pieceArray can change out from under you. (Copy-by-reference vs Copy-by-value.)
Copy-by-value is inefficient. But if you really want to do it, just cheat:
struct FOO { int piece [4][4]; };
FOO Piece::returnPiece()
{
FOO f;
memcpy( f.piece, pieceArray, sizeof(pieceArray) );
return f;
}
void Grid::InsertArray( const FOO & theFoo )
{
// use theFoo.piece[i][j]
}
Of course, a better, more Object-Oriented solution, would be to have returnPiece() create and return a Piece or Array object. (As Juan suggested...)