difference of POD struct array and flat array

[Peter Steiner]

hi there!

my problem: i need arrays of float-triples (math vector). i need these
available in a flat float* array for batch processing (opengl). is an
array of POD structs with three float members each equivalent to a flat
array of floats in terms of memory layout?

ie. is the following code legal?


struct Vector {
float v[3];
};

Vector vectors[3];
float flatVectors[3][3];

memcpy((const void*)flatVectors, (void*)vectors, sizeof(float)*9);
// vectors now contains the same values in Vector::v as flatVectors


i suspect that struct member alignment, as applied by compilers,
forbids that kind of usage. is that true, and if so, is there a way to
circumvent that problem in a portable fashion?

i am interested in such a solution because i would like to implement a
convenient math vector class in a platform-independend, purely object
oriented program.

-- peter

 

[Valentin.Samko]

struct Vector {
float v[3];
};

Vector vectors[3];
float flatVectors[3][3];

memcpy((const void*)flatVectors, (void*)vectors, sizeof(float)*9);

i suspect that struct member alignment, as applied by compilers,
forbids that kind of usage.

Yes. sizeof(Vector) does not have to be equal to sizeof(float)*3.

is that true, and if so, is there a way to
circumvent that problem in a portable fashion?

Write a wrapper around the raw float array. Something like

template<class T, size_t vector_size>
class Matrix
{
typedef std::vector<T> container;

template<class U, class SizeType, class Iterator>
struct VectorT
{
typedef SizeType size_type;
typedef U value_type;
VectorT() {}
value_type& operator[] (size_type pos) const { return * (ptr_ + pos);
}
private:
VectorT(const VectorT&);
VectorT& operator=(const VectorT&);
VectorT(Iterator ptr) : ptr_(ptr) {}
friend class Matrix;
Iterator ptr_;
};
public:

typedef typename container::size_type size_type;
typedef VectorT<T, size_type, typename container::iterator>
vector_proxy;
typedef VectorT<const T, size_type, typename
container::const_iterator> const_vector_proxy;

vector_proxy operator[](size_t pos) { return vector_proxy(c_.begin() +
(vector_size * pos)); }
const_vector_proxy operator[](size_t pos) const { return
const_vector_proxy(c_.begin() + (vector_size * pos)); }

T* get_raw_data_for_opengl() { return &c_[0]; }

private:
container c_;
};

Or you could even implement a full blown STL compatible matrix like
container with iterators.

 

[Peter Steiner]

thanks for the elegant solution!

i'll go for this approach, which gives me the comfort with the proxy
class and the safety by wrapping the container of raw values. win-win.


-- peter