Timeline for How do I declare a 2d array in C++ using new?
Current License: CC BY-SA 3.0
14 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jun 3, 2021 at 13:32 | comment | added | Caleth |
You can even keep the array[x][y] syntax if you are willing to write a row proxy for Matrix::operator[] to return
|
|
| Feb 3, 2020 at 18:44 | comment | added | Dronz | @BorisDalstein Oh, that's interesting and useful, thanks very much for the explanation! | |
| Feb 3, 2020 at 11:54 | comment | added | Boris Dalstein |
@Dronz With a[x][y], what you are actually doing is *(*(a + x) + y): two additions and two memory fetches. With a[index(x, y)], what you are actually doing is *(a + x + w*y): two additions, one multiplication, and one memory fetch. The latter is often preferable, for the reasons exposed in this answer (i.e., trading the extra memory fetch with a multiplication is worth it, especially because the data isn't fragmented and therefore you don't cache-miss).
|
|
| Jun 17, 2019 at 19:14 | comment | added | user1741137 | Anyone who has done some Cuda programming will recognize this as a good solution; | |
| May 26, 2018 at 4:22 | comment | added | Dronz | @AshKetchum Inlining (or maybe macro substitution) makes sense to optimize, but how is the compiled computation more complex than what needs to be done to resolve the address of a[x][y] ? | |
| May 8, 2018 at 16:12 | comment | added | rampion |
You can now have lambdas return lvalue references, which can make this a little more user friendly: auto one_dim_array = new int[num_rows * num_cols]; auto two_dim_array = [num_cols, &one_dim_array](size_t row, size_t col) -> int& { return one_dim_array[row*num_cols + col]; } and use it like two_dim_array(i, j) = 5
|
|
| May 23, 2017 at 11:33 | history | edited | URL Rewriter Bot |
replaced http://stackoverflow.com/ with https://stackoverflow.com/
|
|
| Apr 27, 2017 at 16:49 | comment | added | Ash Ketchum |
The major problem with this solution is that there is extra computation for every index. It becomes worse if you put the index calculation in a function which adds extra overhead. At least, consider using macros or inline functions to reduce overhead. An example macro for C++: #define ROW_COL_TO_INDEX(row, col, num_cols) (row*num_cols + col) Then you can use it as int COLS = 4; A[ ROW_COL_TO_INDEX(r, c, COLS) ] = 75; The overhead really affects when we do matrix multiplications which are of complexity O(n^3) or O(n^2.81) for Strassen's algorithm.
|
|
| Nov 26, 2016 at 20:57 | comment | added | Ulrar | I like this solution a lot, is it applicable to 3 dimensions array too ? I'm thinking something like this : (x + m_width * y) + (m_width * m_height * z) | |
| May 26, 2015 at 19:51 | comment | added | Jeff Wofford |
@Noein, I sort of sketched a solution without prescribing a particular one. More detail might look like: class Matrix { int* array; int m_width; public: Matrix( int w, int h ) : m_width( w ), array( new int[ w * h ] ) {} ~Matrix() { delete[] array; } int at( int x, int y ) const { return array[ index( x, y ) ]; } protected: int index( int x, int y ) const { return x + m_width * y; } }; If you straighten out that code it might make sense, and might shed light on the answer above.
|
|
| May 22, 2015 at 11:35 | comment | added | Noein |
I don't quite get this. So array is something raw, not of the Matrix type? Then, how does m_width work here? And shouldn't it be static? Also, if we're wrapping arrays anyway, why not implement an .at(j, i) member function instead? Great answer otherwise (especially the 'don't ptr to ptr please' bit; been having trouble with that in assignments so I agree).
|
|
| Mar 18, 2015 at 15:42 | history | edited | Jeff Wofford | CC BY-SA 3.0 |
edited body
|
| Mar 3, 2015 at 20:48 | history | edited | Jeff Wofford | CC BY-SA 3.0 |
deleted 3 characters in body
|
| Mar 3, 2015 at 20:43 | history | answered | Jeff Wofford | CC BY-SA 3.0 |