cloudy: multi_arr< T, d, ALLOC, lgBC > Class Template Reference

multi_arr<double,3> arr; // define a placeholder for the array the first argument is the type of data it holds the second argument is the number of dimensions (between 2 and 6)

arr.alloc(3,4,2); // this will allocate a 3x4x2 block of doubles memory will be allocated by new[] so will be initialized by the constructor, if present

The following is an alternative way of allocating the array. It is very similar to the pre-multi_arr way of allocating arrays. In ARPA_TYPE arrays this will help you save memory since only the data elements that are really needed will be allocated. In C_TYPE allocation, the smallest rectangular block will be allocated that can hold all the data. This will use more memory in return for somewhat improved CPU speed. Tests carried out in 2007 showed that the speed advantage of C_TYPE arrays was only 1 to 2 percent. Hence the memory savings were deemed more important and ARPA_TYPE arrays were made the default. However, C_TYPE arrays are guaranteed to be compatible with C code, so these should be used if they are meant to be passed on to C library routines. The example below allocates a triangular matrix.

arr.reserve(3); for( int i=0, i < 3; ++i ) { arr.reserve( i, i+1 ); // note that size does not need to be constant! for( int j=0, j < i+1; ++j ) arr.reserve( i, j, j+1 ); } arr.alloc();

arr.invalidate(); // this will set float or double arrays to all SNaN it will set any other type array to all 0xff bytes. arr.zero(); // this will set the array to all zero

arr.state_do(io,lgRead); // this will write/read the array to/from a state file depending on the boolean flag lgRead; io should point to a file that is already opened for binary writing/reading the file will remain open after access. this allows you to use state_do() for several arrays in a row

multi_arr<double,2,C_TYPE> a(10,10); // allocate C_TYPE array C_library_routine( a.data(), ... ); // and call C library routine with it

arr.clear(); // this will deallocate the array the destructor will also automatically deallocate

the multi_arr class comes with iterators that allow you to speed up memory access even further. using iterators as shown below will generally speed up the code significantly since it avoids calculating the array index over and over inside the body of the loop. especially in tight loops over arrays with high dimension this can become a significant overhead! a const_iterator is also supplied for read-only access, but no reverse_iterators. you can define and initialize an iterator as follows

all the possible combinations for bool, long, realnum and double multi_arr's are predefined below.

for( int k=0; i < 4; k++ ) { for( md3i p = arr.begin(n,k); p != arr.end(n,k); ++p ) p = 3.; }

however, since many compilers have a hard time figuring out that arr.end() has no side effects, it is better to do the following:

for( int k=0; i < 4; k++ ) { md3i end = arr.end(n,k); for( md3i p = arr.begin(n,k); p != end; ++p ) p = 3.; }

NB NB -- the memory layout may change in future editions, so user code should not make any assumptions about the layout. the only exception is that the user may safely assume that for the default memory layout the last index runs over contiguous memory. this allows for efficient iterator access. the example above was OK since arr[n][k][0] and arr[n][k][1] are guaranteed to be adjacent. however the next example is not OK:

md3i p = arr.begin(n,0); for( int k=0; i < 4; k++ ) for( int i=0; i < 2; i++ ) p++ = 3.; // ERROR, this may segfault. bounds checking will catch this (see below for enabling this)

double sum = 0.; for( int k=0; i < 4; k++ ) { md3ci p = arr.begin(n,k); for( int i=0; i < 2; i++ ) sum += p[i]; }

last, but not least, the multi_arr class supports array bounds checking, both for direct access through the indexing method, as well as iterator access. To enable bounds checking, simply define the preprocessor macro BOUNDS_CHECK during compilation. the resulting code will be MUCH slower, so this should only be used as a debugging tool.

Member Typedef Documentation