You should consider to use a package like boost::multi_array which does
a lot of array magic for you.
There are several other array packages (like Blitz++) that could be
used.
Cheers,
Ger
Daniel Cervantes 04/14/11 10:16 PM >>>
Oh wow!!! really thanks for your answer It is very clear and complete,
it will be very useful for my app.
I'm programming using OOP technique and I think that It would be nice
to use overload the operator ()
(based in your MACROS) to acces to elements of dynamic array of my
class.
Best regards.
Daniel.
Daniel,
There are lots of granular ways this can be done (using
sophisticated rank, offset, and stride parameters) but user defined
values in memory consumed by HDF5 (and HDF4) I/O functions generally
need to be in contiguous, adjacent addresses. The common stack memory
style examples accomplish this by default, but they suffer a lot from
not being extensible to real world problems due to the severe size
restrictions one can expect when using stack memory.
So -- most of the intro examples in the HDF5 docs illustrate the
use of HDF5 I/O functions using fixed sized arrays built off stack
memory, not heap memory. The lack of a range of good HDF5 I/O
examples highlighting the use of dynamic (heap) memory to implement
multidimensional arrays (IMHO) is persistant and quite unfortunate,
because almost all real world applications will tend to rely on
dynamic memory off the heap; consequently, a disproportionate fraction
of problems encountered by new users tend to arise from confusion
about this exact issue.
One simple way around this that guarantees that array slots are
contiguous and adjacent in memory, is to declare a given
multidimensional array as a simple dynamic 1D array, pass this to/from
HDF5 I/O functions as a simple 1D pointer, but make sure to reference
these in your own logic using appropriate 2D and 3D (or higher)
macros.
As an example, I've included here several macros below (one for a
2D case, one for a 3D case) that show how this is typically done. In
this case, we assume normal base-0 offsets, where:
"x" refers to a "column",
"y" refers to a "row", and
"z" refers to a 3rd dimension,
"n_y" refers to the number of rows in the array,
"n_x" refers to the number of columns, and
"n_z" refers to the number of 3rd dimension all assuming
row-major-order (RMO) indexing.
#define ARY_2D_OFFSET_B0(y,x,n_x) (((y)*n_x)+(x))
#define ARY_3D_OFFSET_B0(z,y,x,n_y,n_x)
(((z)*((n_y)*(n_x)))+((y)*(n_x))+(x))
Example ANSI C Snippet
double *myArray = NULL; /* model the multidimensional array as a ptr
to a 1D array */
long n_x = 1024;
long n_y = 768;
long n_z = 3;
long nElem = (n_y * n_x * n_z);
long thisRow, thisCol,thisDepth;
long offset = 0 ;
hid_t datasetID = 0 ; /* assign to an appropriate dataset ID somewhere
*/
datasetID = fetchDatasetID(....);
/* always use calloc() not malloc() since calloc() allocates a
initialized (zero filled) pointer to the array */
myArray = (double *)calloc(sizeof(double), nElem);
/* populate it with some values */
for(thisRow = 0; thisRow < n_y; thisRow++)
for(thisCol = 0; thisCol < n_x ; thisCol++)
for(thisDepth =0; thisDepth < n_z; thisDepth)
{
offset = ARY_3D_OFFSET(thisDepth,thisRow,thisCol,n_y,n_x);
myArray[offset] = someAssignmentFunction();
}
/* later, write this 3D array to a HDF5 dataset, or read from it, etc
etc */
h5_status = H5Dwrite(dtasetID,
H5T_IEEE_F64LE,H5S_ALL,H5S_ALL,H5P_DEFAULT,
myArray) ;
/* later, release heap memory. its nice to not have to partition out
the free() call
individuallly by rank, since we simply use the single ptr to point
to the
head of allocated contiguous memory to free */
free(myArray);
Ok, this is just a very rough snippet, and I encourage you to
NOT "cut-and-paste" these sort of snippets directly into your code,
but instead use them to think about how you would like to do this, fit
your own use-context.
These should however give you the idea, where serialized 1D
dynamic arrays may be generalized to represent arrays of higher
dimensionality, using the same design pattern shown for the 2D and 3D
macros. While cumbersome, here are the 4D and 5D equivalents of these
macros, as further examples. Of course, you could also implement any
of these macros as real C functions instead, which can help debugging
and make them more transparent in how they show up in static analysis
tools like cxref, ctags, etc.
#define ARY_4D_OFFSET_B0(a,b,c,d,n_b,n_c,n_d) \
(((a)*(n_b*n_c*n_d))+((b)*(n_c*n_d))+((c)*(n_d))+(d))
#define ARY_5D_OFFSET_B0(a,b,c,d,e,n_b,n_c,n_d,n_e) \
(((a)*(n_b*n_c*n_d))+((b)*(n_c*n_d))+((c)*(n_d))+((d)*(n_e))+(e))
There may be other, simpler ways to do all of this, as others may
submit... ....HTH.
Joe
Hello everybody I am new using HDF5 and I have a problem using dynamic
memory array reading a data set.
When I read my data set using static arrays (like in the all examples
in
documentation ) all it is right
data_out[n][m][l];
dataSet.read(data_out, PredType::NATIVE_DOUBLE,dataSpace);
the array dada_out have all numbers in the data set well.
However if I define data_out using dynamic memory, i.e.
double *** data_out;
....
data_out = new double**[n];
for(int i=0; i < n; ++i)
data_out[i] = new double *[m];
for(int i=0; i < n; ++i)
for(int j=0; j < m; ++j)
data_out[i][j] = new double [l];
dataSet.read(data_out[0][0], PredType::NATIVE_DOUBLE,dataSpace);
only the first row (data_out[0][0][1..l]) have the correct numbers,
and the
···
2011/4/14 J Glassy <um.glassy@gmail.com>
On Thu, Apr 14, 2011 at 10:38 AM, Daniel Cervantes <dcchivela@gmail.com> wrote:
another elements are in a wrong order or they are 0.
Anyone know what is the problem ?
Anyone have a simple example using a dynamic memory arrays ?
Best regards.
Daniel.
_______________________________________________
Hdf-forum is for HDF software users discussion.
Hdf-forum@hdfgroup.org
http://mail.hdfgroup.org/mailman/listinfo/hdf-forum_hdfgroup.org
------------------------------------------------------------
Joseph Glassy
Lead Software Engineer (contractor)
NASA Measures (Freeze/Thaw),Rm CFC 424
College of Forestry and Conservation
Univ. Montana, Missoula, MT 59812
Lupine Logic Inc.
www.lupinelogic.com
Scientific and Technical Programming
_______________________________________________
Hdf-forum is for HDF software users discussion.
Hdf-forum@hdfgroup.org
http://mail.hdfgroup.org/mailman/listinfo/hdf-forum_hdfgroup.org
