Segmentation fault when printing copy of array - c++

The following program segfaults when v2 is printed but not during the array copy. Does anyone know why?
#include <stdio.h>
#include <stdlib.h>
void cpyarray (void* dst, void* src, size_t memsize, size_t size) {
for (size_t i = 0; i < size; i++) {
*(((char*) dst) + i*memsize) = *(((char*) src) + i*memsize);
}
}
int main () {
size_t N = 10;
double* v1 = (double *) malloc(N * sizeof(double));
double* v2 = (double *) malloc(N * sizeof(double));
for (size_t i = 0; i < N; i++) *(v1+i) = i;
printf("\n\nv1: ");
for (size_t i = 0; i < N; i++) printf("%g ", v1[i]);
cpyarray(&v2, &v1, sizeof(double), N);
printf("\n\nv2: ");
for (size_t i = 0; i < N; i++) printf("%g ", v2[i]); // program crashes here
return 0;
}
EDIT: the code does not crash if I copy arrays of ints instead of doubles.

v1, v2 are pointers to memory blocks that you want to operate on. But you're passing their addresses to your cpyarray function.
So you're operating on the wrong memory blocks, stepping on memory around the v2 variable, and changing what v2 points to.
cpyarray(v2, v1, sizeof(double), N);

You are passing &v2 to cpyarray(), which means that your function is changing where v2 is pointing! Since it's now pointing to an invalid location, you get a seg fault when you dereference it for printf().
Instead, just pass v2 to cpyarray().

You are passing wrong pointers to cpyarray function.
Your code sends pointer to array (pointer to pointer to double)
cpyarray(&v2, &v1, sizeof(double), N);
You should send array instead (pointer to double)
cpyarray(v2, v1, sizeof(double), N);

In addition to the problem that the current answers have pointed out, cpyarray doesn't copy the array, just some parts of it. Instead of trying to imitate types with i*memsize, just treat the memory block as a bunch of bytes and copy them one by one. Now all you have to do is figure out how many bytes to copy, which isn't hard.

Related

Allreduce with user defined function and MPI_BOTTOM

Consider the following program that is supposed to do some stupid addition of doubles:
#include <iostream>
#include <vector>
#include <mpi.h>
void add(void* invec, void* inoutvec, int* len, MPI_Datatype*)
{
double* a = reinterpret_cast <double*> (inoutvec);
double* b = reinterpret_cast <double*> (invec);
for (int i = 0; i != *len; ++i)
{
a[i] += b[i];
}
}
int main(int argc, char* argv[])
{
MPI_Init(&argc, &argv);
std::vector<double> buffer = { 2.0, 3.0 };
MPI_Op operation;
MPI_Op_create(add, 1, &operation);
MPI_Datatype types[1];
MPI_Aint addresses[1];
int lengths[1];
int count = 1;
MPI_Get_address(buffer.data(), &addresses[0]);
lengths[0] = buffer.size();
types[0] = MPI_DOUBLE;
MPI_Datatype type;
MPI_Type_create_struct(count, lengths, addresses, types, &type);
MPI_Type_commit(&type);
MPI_Allreduce(MPI_IN_PLACE, MPI_BOTTOM, 1, type, operation, MPI_COMM_WORLD);
MPI_Type_free(&type);
MPI_Op_free(&operation);
MPI_Finalize();
std::cout << buffer[0] << " " << buffer[1] << "\n";
}
Because this is part of larger program where the data I want to send is 1) on the heap and 2) consists of different types I think I have to use a user-defined type.
Now something must be wrong here because the program crashes when run with mpirun -n 2 ./a.out. The backtrace from gdb is:
#0 __memcpy_sse2_unaligned () at ../sysdeps/x86_64/multiarch/memcpy-sse2-unaligned.S:158
#1 0x00007ffff65de460 in non_overlap_copy_content_same_ddt () from /usr/local/lib/libopen-pal.so.6
#2 0x00007ffff180a69b in ompi_coll_tuned_allreduce_intra_recursivedoubling () from /usr/local/lib/openmpi/mca_coll_tuned.so
#3 0x00007ffff793bb8b in PMPI_Allreduce () from /usr/local/lib/libmpi.so.1
#4 0x00000000004088b6 in main (argc=1, argv=0x7fffffffd708) at mpi_test.cpp:39
Line 39 is the MPI_Allreduce call. This is probably a dumb mistake, but after staring on it for hours I still don't see it. Does anyone spot the mistake? Thanks!
Edit: There is a bug in how Open MPI handles types with non-zero lower bounds (such as the one that you create when using absolute addresses) while performing in-place reduce-to-all. It seems to exist in all versions, including the development branch. The status can be tracked by following the issue on GitHub.
Your add operator is wrong as you fail to account for the datatype's lower bound. A proper solution would be something like:
void add(void* invec, void* inoutvec, int* len, MPI_Datatype* datatype)
{
MPI_Aint lb, extent;
MPI_Type_get_true_extent(*datatype, &lb, &extent);
double* a = reinterpret_cast <double*> (reinterpret_cast <char*>(inoutvec) + lb);
double* b = reinterpret_cast <double*> (reinterpret_cast <char*>(invec) + lb);
for (int i = 0; i != *len; ++i)
{
a[i] += b[i];
}
}
This will access the data correctly but is still wrong. *len will be 1 as that is what you pass to MPI_Allreduce but there are two doubles behind each element. The correctly written operator will either use the type introspection mechanism to obtain the length of the block of doubles and multiply *len by it or simply hardcode the vector length to be two:
for (int i = 0; i < 2*(*len); i++)
{
a[i] += b[i];
}

Declaring and allocating a 2d array in C++

I am a Fortran user and do not know C++ well enough. I need to make some additions into an existing C++ code. I need to create a 2d matrix (say A) of type double whose size (say m x n) is known only during the run. With Fortran this can be done as follows
real*8, allocatable :: A(:,:)
integer :: m, n
read(*,*) m
read(*,*) n
allocate(a(m,n))
A(:,:) = 0.0d0
How do I create a matrix A(m,n), in C++, when m and n are not known at the time of compilation? I believe the operator new in C++ can be useful but not not sure how to implement it with doubles. Also, when I use following in C++
int * x;
x = new int [10];
and check the size of x using sizeof(x)/sizeof(x[0]), I do not have 10, any comments why?
To allocate dynamically a construction similar to 2D array use the following template.
#include <iostream>
int main()
{
int m, n;
std::cout << "Enter the number of rows: ";
std::cin >> m;
std::cout << "Enter the number of columns: ";
std::cin >> n;
double **a = new double * [m];
for ( int i = 0; i < m; i++ ) a[i] = new double[n]();
//...
for ( int i = 0; i < m; i++ ) delete []a[i];
delete []a;
}
Also you can use class std::vector instead of the manually allocated pointers.
#include <iostream>
#include <vector>
int main()
{
int m, n;
std::cout << "Enter the number of rows: ";
std::cin >> m;
std::cout << "Enter the number of columns: ";
std::cin >> n;
std::vector<std::vector<double>> v( m, std::vector<double>( n ) );
//...
}
As for this code snippet
int * x;
x = new int [10];
then x has type int * and x[0] has type int. So if the size of the pointer is equal to 4 and the size of an object of type int is equal also to 4 then sizeof( x ) / sizeof( x[0] ) will yields 1. Pointers do not keep the information whether they point to only a single object or the first object pf some sequence of objects.
I would recommend using std::vector and avoid all the headache of manually allocating and deallocating memory.
Here's an example program:
#include <iostream>
#include <vector>
typedef std::vector<double> Row;
typedef std::vector<Row> Matrix;
void testMatrix(int M, int N)
{
// Create a row with all elements set to 0.0
Row row(N, 0.0);
// Create a matrix with all elements set to 0.0
Matrix matrix(M, row);
// Test accessing the matrix.
for ( int i = 0; i < M; ++i )
{
for ( int j = 0; j < N; ++j )
{
matrix[i][j] = i+j;
std::cout << matrix[i][j] << " ";
}
std::cout << std::endl;
}
}
int main()
{
testMatrix(10, 20);
}
The formal C++ way of doing it would be this:
std::vector<std::vector<int>> a;
This creates container which contains a zero size set of sub-containers. C++11/C++13 provide std::array for fixed-sized containers, but you specified runtime sizing.
We now have to impart our dimensions on this and, unfortunately. Lets assign the top-level:
a.resize(10);
(you can also push or insert elements)
What we now have is a vector of 10 vectors. Unfortunately, they are all independent, so you would need to:
for (size_t i = 0; i < a.size(); ++i) {
a[i].resize(10);
}
We now have a 10x10. We can also use vectors constructor:
std::vector<std::vector<int>> a(xSize, std::vector<int>(ySize)); // assuming you want a[x][y]
Note that vectors are fully dynamic, so we can resize elements as we need:
a[1].push_back(10); // push value '10' onto a[1], creating an 11th element in a[1]
a[2].erase(2); // remove element 2 from a[2], reducing a[2]s size to 9
To get the size of a particular slot:
a.size(); // returns 10
a[1].size(); // returns 11 after the above
a[2].size(); // returns 9 after teh above.
Unfortunately C++ doesn't provide a strong, first-class way to allocate an array that retains size information. But you can always create a simple C-style array on the stack:
int a[10][10];
std::cout << "sizeof a is " << sizeof(a) <<'\n';
But using an allocator, that is placing the data onto the heap, requires /you/ to track size.
int* pointer = new int[10];
At this point, "pointer" is a numeric value, zero to indicate not enough memory was available or the location in memory where the first of your 10 consecutive integer storage spaces are located.
The use of the pointer decorator syntax tells the compiler that this integer value will be used as a pointer to store addresses and so allow pointer operations via the variable.
The important thing here is that all we have is an address, and the original C standard didn't specify how the memory allocator would track size information, and so there is no way to retrieve the size information. (OK, technically there is, but it requires using compiler/os/implementation specific information that is subject to frequent change)
These integers must be treated as a single object when interfacing with the memory allocation system -- you can't, for example:
delete pointer + 5;
to delete the 5th integer. They are a single allocation unit; this notion allows the system to track blocks rather than individual elements.
To delete an array, the C++ syntax is
delete[] pointer;
To allocate a 2-dimensional array, you will need to either:
Flatten the array and handle sizing/offsets yourself:
static const size_t x = 10, y = 10;
int* pointer = new int[x * y];
pointer[0] = 0; // position 0, the 1st element.
pointer[x * 1] = 0; // pointer[1][0]
or you could use
int access_2d_array_element(int* pointer, const size_t xSize, const size_t ySize, size_t x, size_t y)
{
assert(x < xSize && y < ySize);
return pointer[y * xSize + x];
}
That's kind of a pain, so you would probably be steered towards encapsulation:
class Array2D
{
int* m_pointer;
const size_t m_xSize, m_ySize;
public:
Array2D(size_t xSize, size_t ySize)
: m_pointer(new int[xSize * ySize])
, m_xSize(xSize)
, m_ySize(ySize)
{}
int& at(size_t x, size_t y)
{
assert(x < m_xSize && y < m_ySize);
return m_pointer[y * m_xSize + x];
}
// total number of elements.
size_t arrsizeof() const
{
return m_xSize * m_ySize;
}
// total size of all data elements.
size_t sizeof() const
{
// this sizeof syntax makes the code more generic.
return arrsizeof() * sizeof(*m_pointer);
}
~Array2D()
{
delete[] m_pointer;
}
};
Array2D a(10, 10);
a.at(1, 3) = 13;
int x = a.at(1, 3);
Or,
For each Nth dimension (N < dimensions) allocate an array of pointers-to-pointers, only allocating actual ints for the final dimension.
const size_t xSize = 10, ySize = 10;
int* pointer = new int*(x); // the first level of indirection.
for (size_t i = 0; i < x; ++i) {
pointer[i] = new int(y);
}
pointer[0][0] = 0;
for (size_t i = 0; i < x; ++i) {
delete[] pointer[i];
}
delete[] pointer;
This last is more-or-less doing the same work, it just creates more memory fragmentation than the former.
-----------EDIT-----------
To answer the question "why do I not have 10" you're probably compiling in 64-bit mode, which means that "x" is an array of 10 pointers-to-int, and because you're in 64-bit mode, pointers are 64-bits long, while ints are 32 bits.
The C++ equivalent of your Fortran code is:
int cols, rows;
if ( !(std::cin >> cols >> rows) )
// error handling...
std::vector<double> A(cols * rows);
To access an element of this array you would need to write A[r * rows + c] (or you could do it in a column-major fashion, that's up to you).
The element access is a bit clunky, so you could write a class that wraps up holding this vector and provides a 2-D accessor method.
In fact your best bet is to find a free library that already does this, instead of reinventing the wheel. There isn't a standard Matrix class in C++, because somebody would always want a different option (e.g. some would want row-major storage, some column-major, particular operations provided, etc. etc.)
Someone suggested boost::multi_array; that stores all its data contiguously in row-major order and is probably suitable. If you want standard matrix operations consider something like Eigen, again there are a lot of alternatives out there.
If you want to roll your own then it could look like:
struct FortranArray2D // actually easily extensible to any number of dimensions
{
FortranArray2D(size_t n_cols, size_t n_rows)
: n_cols(n_cols), n_rows(n_rows), content(n_cols * n_rows) { }
double &operator()(size_t col, size_t row)
{ return content.at(row * n_rows + col); }
void resize(size_t new_cols, size_t new_rows)
{
FortranArray2D temp(new_cols, new_rows);
// insert some logic to move values from old to new...
*this = std::move(temp);
}
private:
size_t n_rows, n_cols;
std::vector<double> content;
};
Note in particular that by avoiding new you avoid the thousand and one headaches that come with manual memory management. Your class is copyable and movable by default. You could add further methods to replicate any functionality that the Fortran array has which you need.
int ** x;
x = new int* [10];
for(int i = 0; i < 10; i++)
x[i] = new int[5];
Unfortunately you'll have to store the size of matrix somewhere else.
C/C++ won't do it for you. sizeof() works only when compiler knows the size, which is not true in dynamic arrays.
And if you wan to achieve it with something more safe than dynamic arrays:
#include <vector>
// ...
std::vector<std::vector<int>> vect(10, std::vector<int>(5));
vect[3][2] = 1;

Passing 2D array of const size

How should an array of constant size:
const int m = 5, n = 3;
int arr[m][n];
be passed to a function in a way which is both C89 and C++-compatible?
void func(const int m, const int n, int arr[][n]) { }
isn't valid C++ (giving errors such as "A parameter is not allowed" and "Variable 'n' was not declared in this scope"), even though the size of arr is determinate at compile-time. (It is valid C, however.) #defineing m and n works but is not preferred due to scope issues. Passing a pointer to the first element of the array leads to ugly code in the function body.
Feel free to take a look at this FAQ for context.
In C++, you can pass an array to a function with full type information intact by utilizing a template and an array reference function argument:
template <unsigned M, unsigned N>
void func (int (&arr)[M][N]) {
//...
}
The function prototype you are using is using a C99 feature called VLA to provide a dynamic binding of the array dimension. This is not a C++ feature, although some C++ compilers will allow it as an extension to the C++ language.
The C-FAQ was written before C99 was ratified, so the variable length array feature was not yet a standard feature of C. With a modern C compiler with VLA support, the function prototype you provided works just fine.
There is another alternative to use if you have an older compiler for which VLA support is not available. That is to treat the 2-D array as a flattened 1-D array, and use manual calculations to index the correct integer:
void func(const int m, const int n, void *p) {
int *a = p;
int i, j;
for (i = 0; i < m; ++i) {
for (j = 0; j < n; ++j) {
printf(" %d", a[i*n + j]);
}
puts("");
}
}
Then you call func(m, n, arr). In side the function, the expression
a[i*n + j]
steps over n ints i times, then steps over j ints. Since each row is n ints long, the calculation returns the ith row and the jth column, which corresponds precisely to arr[i][j].
I have tried this code:
void func(const int m, const int n, int arr[][n])
{
printf("%d\n", arr[4][2]);
}
int main()
{
const int m = 5, n = 3;
int arr[m][n];
arr[4][2] = 10;
func(m, n, arr);
}
and this work with no warnings
Your array arr[m][n] is not constant. However you have constant variables M and N. You should also define the arr[m][n] as a constant and not just an int array.
You may want to consider dynamicaly allocating your array so that you can just pass the pointer address down.
const int m = 5, n = 3;
int i = 0;
int* *arr; //Pointer to an integer pointer (Note can also be int **arr or int** arr)
arr = malloc(sizeof(int*)*(m+1)); //I add one because I am assuming that 'm' does not account for the terminating null character. But if you do not need a terminating null then you can remove this and the perantheses around the 'm'.
for(i = 0; i < m; i++)
{
arr[i] = malloc(sizeof(int*)*(n+1)); //Same as before
}
The inital malloc() call allocates memory for an array of integer arrays or said in another way, it allocates a pointer to a series of other pointers. The for loop will allocate an integer array of 'm' size for each element of the original array or said another way it will allocate space for every pointer address pointed to by the original pointer address. I left out error checking in order to simplfy my example but here is the same example with error checking.
const int m = 5, n = 3;
int i = 0;
int* *arr = NULL;
if((arr = malloc(sizeof(int*)*(m+1))) == NULL)
{
perror("ERROR(1): Failed to allocate memory for the initial pointer address ");
return 1;
}
for(i = 0; i < m; i++)
{
if((arr = malloc(sizeof(int*)*(m+1))) == NULL)
{
perror("ERROR(2): Failed to allocate memory for a subsequent pointer address ");
return 2;
}
}
Now that you have dynamicaly allocated your array you can just pass the pointer address.
int* *arr in the following the way.
void fun(const int n, const int m, int* *arr) {}
Also you don't necessarily have to keep track of the size of your arrays if the sizes are constant and if you use null terminated arrays. You just have to malloc the array using the constant integer variable's actual value and then check for the terminating null byte when iterating threw the array.
int* *arr = NULL;
if((arr = malloc(sizeof(int*)*6)) == NULL)'m'+1 = 6;
{
perror("ERROR(1): Failed to allocate memory for the initial pointer address ");
return 1;
}
for(i = 0; i < m; i++)
{
if((arr = malloc(sizeof(int*)*4) == NULL)//'n'+1 = 4
{
perror("ERROR(2): Failed to allocate memory for a subsequent pointer address ");
return 2;
}
}
You can then display the entire two dimensional array in the following way. Note that '\000' is the octagonal value for a null byte(00000000).
int i, j;
for(i = 0; arr[i] != '\000'; i++)
{
for(j = 0; arr[i][j] != '\000'; j++)
{
printf("%i ", arr[i][j]); //Prints the current element of the current array
}
printf("\n"); //This just ends the line so that each of the arrays is printed on it's own line.
}
Of course the above mentioned loops would have the same result as the following.
int i, j;
int m = 5;
int n = 3;
for(i = 0; i < m; i++)
{
for(j = 0; i < n; j++)
{
printf("%i ", arr[i][j]); //Prints the current element of the current array
}
printf("\n"); //This just ends the line so that each of the arrays is printed on it's own line.
}
Which means, in most situations there is no need for keeping track of an array's size but there are situations in which it is necessary. For example if one your arrays could possible contain a null byte other than the terminating null byte. The new null byte would shorten the array's size to the index of the new null byte. If you have any questions or comments feel free to comment below or message me.
The problem here is the "missing" support for dynamic arrays in C++.
const int m = 5, n = 3;
int arr[m][n];
Works since m and n are compile time constant and accessible directly at the declaration of the array.
void func(const int m, const int n, int arr[][n]) { }
The compiler handles your function regardless of where it is called in first place.
Therefore n is unknown/variable and thus prohibited as a array dimensionality.
The following example won't work too because of the very same reason:
void foo (const int n)
{
int arr[n]; // error, n is const but not compile time constant
}
int main (void)
{
foo(4);
}
jxh answered what to do about it.

arr is used before its value is set

Hello I have a generate array method:
void generateArray(double *data, int count) {
for (int i = 0; i < count; i++)
data[i] = rand() / ((rand() + rand()) / 2.0 + 1);
}
and I have below main
int main(void) {
double *arr;
generateArray(arr,40000);
cout << arr[0];
return 0;
}
it says arr is used before its value is set why? Why does not genearate array fills it? How can i fix this problem?
double *arr;
This only gives you a pointer. You haven't allocated any doubles for it to point at and you haven't initialised the pointer to actually point anywhere. Here, you'll find it easier to declare an array of the appropriate size instead:
double arr[40000];
Passing arr as you are currently doing will perform array-to-pointer conversion and it will work as expected.
However, an array of 40000 doubles is a pretty large object to have on the stack. You may prefer to dynamically allocate your array with double* arr = new double[40000];, but you would need to make sure you do delete[] arr; when you are done with it to avoid memory leaks. Instead, you might prefer to use a std::vector:
std::vector<double> arr(40000);
And simply change the first argument of generateArray to a std::vector<double>&.
You simply give it a pointer, but never allocate any memory for the array. You want a data = new double[count] in your generateArray, or to simply allocate it beforehand with double arr[40000] (or better yet, use a vector).
This would be preferable as:
std::vector<double> generateArray(int count) {
std::vector<double> data(count);
for (int i = 0; i < count; i++)
data[i] = rand() / ((rand() + rand()) / 2.0 + 1);
}
return data;
}
If you need to get array data back somehow, vector is guaranteed to utilize a contiguous array internally, accessible with &v[0] to &v[v.size()-1].
You never initialize the array with a size. You need to do
arr = new double[count];
inside the function. Like below. Don't forget to do delete[] arr; when you're done!
void generateArray(double *data, int count) {
arr = new double[count];
for (int i = 0; i < count; i++)
data[i] = rand() / ((rand() + rand()) / 2.0 + 1);
}
Another option would be doing double *arr = new double[40000] outside the function.

How do I treat a pointer as a multi array?

I have this loop which gives seg. fault.
s->c = malloc(width * height * sizeof(double));
if (s->c == NULL) { puts("malloc failed"); exit(1); }
for (int n = 0; n < width; n++) {
for (int m = 0; m < height; m++) {
d = (&s->c)[m][n];
printf("d %f\n", d);
printf("m %i\n", m);
printf("n %i\n", n);
}
}
Inside s->c is:
double* c;
When executed it just outputs:
d 27.000000
m 0
n 0
and then seg. fault.
It worked when I treated the s->c as a 1D array, but I would really like to treat it as a 2D array.
Is that possible, when the c pointer is in a struct?
If so, is (&s->c)[m][n] then the correct way to access the elements?
Sandra
The problem is that the compiler doesn't know the dimensions of your matrix.
When you have: double tab[m][n] you can access the element tab[row][col] as *(tab + (row * n) + col)
In your case you only have double *tab; that can be considered as the pointer to the element tab[0][0] with no information on the matrix dimensions and the compiler can't compute the right address.
You could compute the address yourself (for example using a macro) but would lose the nice tab[x][y] syntax.
I`m surprised it compiles. You should have received at least a warning about implicitly casting a double to a pointer.
I'm very surprised it even compiles. Apparently c is a double*, so (&s->c)[m] is the m'th double. Now, double doesn't have an operator[], so I don't see how the [n] part in (&s->c)[m][n] can be legal.
Presumably, you have declared c differently. There are different solutions: a pointer to a pointer, an pointer to an array of doubles, an array of pointers to doubles, etcetera. All might work, if the allocations match the declaration. In your case, the allocation will not match the declaration.
The correct way to access the array elements is
d = s->c[m * width + n];
Is that what you mean by treating it as a 1D array?
Access the elements using
double d = s->c[m*width+n];
Perhaps through an inline function, to avoid unexpected behaviour.
The compiler does not know about the width of your intended 2D array. It might possibly interpret (&s->c)[m][n] as s->c[m+n], or as something quite different.
Short answer: you can't treat it as a 2D array, at least not in the way you expect.
The reason writing
(&s->c)[m][n]
doesn't work can be shown as follows. Assume the address of s->c is 0x00080004, and the address of the dynamically allocated memory pointed to by s->c is 0x00001000.
The expression (&s->c)[m][n] is evaluated as *(*(&s->c + m) + n);
The expression &s->c evaluates to 0x00080004;
The expression (&s->c + m) evaluates to 0x00080004+m;
The expression *(&s->c + m) evaluates to the value of whatever is pointed to by 0x00080004+m. If m is 0, then 0x00080004+m points to 0x00001000, which is the address of your dynamically allocated memory (*(&x) == x). If m is any other value, then 0x00080004+m points somewhere random;
The expression (*(&s->c + m) + n) evaluates to whatever 0x00080004+m points to offset by n. If m is 0, then the value is 0x00001000+n, or an offset into your dynamically allocated memory. If m is not 0, then the value is something random;
The expression *(*(&s->c) + m) + n) attempts to dereference the above value. If m is 0, then the result is the value of an element in the dynamically allocated array. If m is not 0, then the result is ... something else. In your case, a segfault.
If you want to dynamically allocate a 2D array, you have to use a pointer to a pointer and allocate it in steps, like so:
struct {
...
double **c;
...
} *s;
...
/**
* Notes: type of s->c is double **
* type of *(s->c) and s->c[i] is double *
* type of *(s->c[i]) and s->c[i][j] is double
*/
s->c = malloc(sizeof *(s->c) * rows);
if (s->c)
{
for (i = 0; i < rows; i++)
{
s->c[i] = malloc(sizeof *(s->c[i]) * columns);
if (s->c[i])
{
// initialize s->c[i][0] through s->c[i][columns-1]
}
}
}
If you want to allocate s->c as a 1D array, then you can define a macro that does the job for you (but you need to know the second dimension):
#define AR(M, X, Y) ((M)[(Y) + dimy * (X)])
I'm surprised nobody mentioned boost::multi_array_ref:
#include <iostream>
#include <boost/multi_array.hpp>
int main()
{
int rows = 4, cols = 3;
// Allocate one big block of contiguous data for entire multi-array
double* arrayData = new double[rows*cols];
if (arrayData)
{
boost::multi_array_ref<double, 2>
arrayRef(arrayData, boost::extents[rows][cols]);
for (int row = 0; row < rows; ++row)
{
for (int col = 0; col < cols; ++col)
{
arrayRef[row][col] = row*cols + col;
std::cout << arrayRef[row][col] << " ";
}
std::cout << std::endl;
}
}
delete [] arrayData;
}
You can also just use boost::multi_array and resize it dynamically:
boost::multi_array_ref<double, 2> marray;
marray.resize(boost::extents[rows][cols]);