6569
|
1 @node Dynamically Linked Functions |
|
2 @appendix Dynamically Linked Functions |
|
3 @cindex dynamic-linking |
|
4 |
|
5 Octave has the possibility of including compiled code as dynamically |
|
6 linked extensions and then using these extensions as if they were part |
6571
|
7 of Octave itself. Octave has the option of directly calling C++ code |
6569
|
8 through its native oct-file interface or C code through its mex |
6571
|
9 interface. It can also indirectly call functions written in any other |
|
10 language through a simple wrapper. The reasons to write code in a |
6569
|
11 compiled language might be either to link to an existing piece of code |
|
12 and allow it to be used within Octave, or to allow improved performance |
|
13 for key pieces of code. |
|
14 |
|
15 Before going further, you should first determine if you really need to |
6571
|
16 use dynamically linked functions at all. Before proceeding with writing |
6569
|
17 any dynamically linked function to improve performance you should |
|
18 address ask yourself |
|
19 |
|
20 @itemize @bullet |
|
21 @item |
|
22 Can I get the same functionality using the Octave scripting language only. |
|
23 @item |
6572
|
24 Is it thoroughly optimized Octave code? Vectorization of Octave code, |
6569
|
25 doesn't just make it concise, it generally significantly improves its |
6571
|
26 performance. Above all, if loops must be used, make sure that the |
6569
|
27 allocation of space for variables takes place outside the loops using an |
|
28 assignment to a like matrix or zeros. |
|
29 @item |
|
30 Does it make as much use as possible of existing built-in library |
6572
|
31 routines? These are highly optimized and many do not carry the overhead |
6569
|
32 of being interpreted. |
|
33 @item |
|
34 Does writing a dynamically linked function represent useful investment |
|
35 of your time, relative to staying in Octave? |
|
36 @end itemize |
|
37 |
|
38 Also, as oct- and mex-files are dynamically linked to octave, they |
|
39 introduce to possibility of having Octave abort due to coding errors in |
6571
|
40 the user code. For example a segmentation violation in the users code |
6569
|
41 will cause Octave to abort. |
|
42 |
|
43 @menu |
6572
|
44 * Oct-Files:: |
|
45 * Mex-Files:: |
|
46 * Standalone Programs:: |
6569
|
47 @end menu |
|
48 |
|
49 @node Oct-Files |
|
50 @section Oct-Files |
|
51 @cindex oct-files |
|
52 @cindex mkoctfile |
|
53 @cindex oct |
|
54 |
|
55 @menu |
6572
|
56 * Getting Started with Oct-Files:: |
|
57 * Matrices and Arrays in Oct-Files:: |
|
58 * Character Strings in Oct-Files:: |
|
59 * Cell Arrays in Oct-Files:: |
|
60 * Structures in Oct-Files:: |
|
61 * Sparse Matrices in Oct-Files:: |
|
62 * Accessing Global Variables in Oct-Files:: |
|
63 * Calling Octave Functions from Oct-Files:: |
|
64 * Calling External Code from Oct-Files:: |
|
65 * Allocating Local Memory in Oct-Files:: |
|
66 * Input Parameter Checking in Oct-Files:: |
|
67 * Exception and Error Handling in Oct-Files:: |
|
68 * Documentation and Test of Oct-Files:: |
|
69 * Application Programming Interface for Oct-Files:: |
6569
|
70 @end menu |
|
71 |
|
72 @node Getting Started with Oct-Files |
|
73 @subsection Getting Started with Oct-Files |
|
74 |
|
75 The basic command to build oct-files is @code{mkoctfile} and it can be |
|
76 call from within octave or from the command line. |
|
77 |
|
78 @DOCSTRING(mkoctfile) |
|
79 |
|
80 Consider the short example |
|
81 |
|
82 @example |
|
83 @group |
|
84 #include <octave/oct.h> |
|
85 |
|
86 DEFUN_DLD (helloworld, args, nargout, |
|
87 "Hello World Help String") |
|
88 @{ |
|
89 int nargin = args.length (); |
6571
|
90 octave_stdout << "Hello World has " << nargin |
6569
|
91 << " input arguments and " |
|
92 << nargout << " output arguments.\n"; |
|
93 return octave_value_list (); |
|
94 @} |
|
95 @end group |
|
96 @end example |
|
97 |
|
98 This example although short introduces the basics of writing a C++ |
6571
|
99 function that can be dynamically linked to Octave. The easiest way to |
6569
|
100 make available most of the definitions that might be necessary for an |
|
101 oct-file in Octave is to use the @code{#include <octave/oct.h>} |
6571
|
102 header. |
6569
|
103 |
|
104 The macro that defines the entry point into the dynamically loaded |
6572
|
105 function is @code{DEFUN_DLD}. This macro takes four arguments, these being |
6569
|
106 |
|
107 @enumerate 1 |
6571
|
108 @item The function name as it will be seen in Octave, |
6572
|
109 @item The list of arguments to the function of type @code{octave_value_list}, |
6569
|
110 @item The number of output arguments, which can and often is omitted if |
|
111 not used, and |
|
112 @item The string that will be seen as the help text of the function. |
|
113 @end enumerate |
|
114 |
6572
|
115 The return type of functions defined with @code{DEFUN_DLD} is always |
|
116 @code{octave_value_list}. |
6569
|
117 |
|
118 There are a couple of important considerations in the choice of function |
6571
|
119 name. Firstly, it must be a valid Octave function name and so must be a |
6569
|
120 sequence of letters, digits and underscores, not starting with a |
6571
|
121 digit. Secondly, as Octave uses the function name to define the filename |
6572
|
122 it attempts to find the function in, the function name in the @code{DEFUN_DLD} |
6571
|
123 macro must match the filename of the oct-file. Therefore, the above |
6572
|
124 function should be in a file @file{helloworld.cc}, and it should be |
|
125 compiled to an oct-file using the command |
6569
|
126 |
|
127 @example |
|
128 mkoctfile helloworld.cc |
|
129 @end example |
|
130 |
|
131 This will create a file call helloworld.oct, that is the compiled |
6571
|
132 version of the function. It should be noted that it is perfectly |
6572
|
133 acceptable to have more than one @code{DEFUN_DLD} function in a source |
6571
|
134 file. However, there must either be a symbolic link to the oct-file for |
6572
|
135 each of the functions defined in the source code with the @code{DEFUN_DLD} |
6569
|
136 macro or the autoload (@ref{Function Files}) function should be used. |
|
137 |
|
138 The rest of this function then shows how to find the number of input |
|
139 arguments, how to print through the octave pager, and return from the |
6571
|
140 function. After compiling this function as above, an example of its use |
6569
|
141 is |
|
142 |
|
143 @example |
|
144 @group |
6572
|
145 helloworld (1, 2, 3) |
|
146 @print{} Hello World has 3 input arguments and 0 output arguments. |
6569
|
147 @end group |
|
148 @end example |
|
149 |
|
150 @node Matrices and Arrays in Oct-Files |
|
151 @subsection Matrices and Arrays in Oct-Files |
|
152 |
|
153 Octave supports a number of different array and matrix classes, the |
6571
|
154 majority of which are based on the Array class. The exception is the |
|
155 sparse matrix types discussed separately below. There are three basic |
|
156 matrix types |
6569
|
157 |
6572
|
158 @table @code |
6569
|
159 @item Matrix |
|
160 A double precision matrix class defined in dMatrix.h, |
|
161 @item ComplexMatrix |
|
162 A complex matrix class defined in CMatrix.h, and |
|
163 @item BoolMatrix |
|
164 A boolean matrix class defined in boolMatrix.h. |
|
165 @end table |
|
166 |
6571
|
167 These are the basic two-dimensional matrix types of octave. In |
6569
|
168 additional there are a number of multi-dimensional array types, these |
|
169 being |
|
170 |
6572
|
171 @table @code |
6569
|
172 @item NDArray |
6572
|
173 A double precision array class defined in @file{dNDArray.h} |
6569
|
174 @item ComplexNDarray |
6572
|
175 A complex array class defined in @file{CNDArray.h} |
6569
|
176 @item boolNDArray |
6572
|
177 A boolean array class defined in @file{boolNDArray.h} |
|
178 @item int8NDArray |
|
179 @itemx int16NDArray |
|
180 @itemx int32NDArray |
|
181 @itemx int64NDArray |
|
182 8, 16, 32 and 64-bit signed array classes defined in |
|
183 @file{int8NDArray.h}, @file{int16NDArray.h}, etc. |
|
184 @item uint8NDArray |
|
185 @itemx uint16NDArray |
|
186 @itemx uint32NDArray |
|
187 @itemx uint64NDArray |
|
188 8, 16, 32 and 64-bit unsigned array classes defined in |
|
189 @file{uint8NDArray.h}, @file{uint16NDArray.h}, etc. |
6569
|
190 @end table |
|
191 |
|
192 There are several basic means of constructing matrices of |
6572
|
193 multi-dimensional arrays. Considering the @code{Matrix} type as an |
|
194 example |
6569
|
195 |
|
196 @itemize @bullet |
6571
|
197 @item |
|
198 We can create an empty matrix or array with the empty constructor. For |
6569
|
199 example |
|
200 |
|
201 @example |
|
202 Matrix a; |
|
203 @end example |
|
204 |
|
205 This can be used on all matrix and array types |
6571
|
206 @item |
|
207 Define the dimensions of the matrix or array with a dim_vector. For |
6569
|
208 example |
|
209 |
|
210 @example |
|
211 @group |
6572
|
212 dim_vector dv (2); |
6569
|
213 dv(0) = 2; dv(1) = 2; |
6572
|
214 Matrix a (dv); |
6569
|
215 @end group |
|
216 @end example |
|
217 |
|
218 This can be used on all matrix and array types |
|
219 @item |
6572
|
220 Define the number of rows and columns in the matrix. For example |
6569
|
221 |
|
222 @example |
6572
|
223 Matrix a (2, 2) |
6569
|
224 @end example |
|
225 |
|
226 However, this constructor can only be used with the matrix types. |
|
227 @end itemize |
|
228 |
|
229 These types all share a number of basic methods and operators, a |
|
230 selection of which include |
|
231 |
6572
|
232 @deftypefn Method T& {operator ()} (octave_idx_type) |
|
233 @deftypefnx Method T& elem (octave_idx_type) |
|
234 The @code{()} operator or @code{elem} method allow the values of the |
|
235 matrix or array to be read or set. These can take a single argument, |
|
236 which is of type @code{octave_idx_type}, that is the index into the matrix or |
6571
|
237 array. Additionally, the matrix type allows two argument versions of the |
6572
|
238 @code{()} operator and elem method, giving the row and column index of the |
6569
|
239 value to obtain or set. |
6572
|
240 @end deftypefn |
6569
|
241 |
|
242 Note that these function do significant error checking and so in some |
|
243 circumstances the user might prefer the access the data of the array or |
|
244 matrix directly through the fortran_vec method discussed below. |
6572
|
245 |
|
246 @deftypefn Method octave_idx_type nelem (void) const |
6569
|
247 The total number of elements in the matrix or array. |
6572
|
248 @end deftypefn |
|
249 |
|
250 @deftypefn Method size_t byte_size (void) const |
6569
|
251 The number of bytes used to store the matrix or array. |
6572
|
252 @end deftypefn |
|
253 |
|
254 @deftypefn Method dim_vector dims (void) const |
6569
|
255 The dimensions of the matrix or array in value of type dim_vector. |
6572
|
256 @end deftypefn |
|
257 |
|
258 @deftypefn Method void resize (const dim_vector&) |
|
259 A method taking either an argument of type @code{dim_vector}, or in the |
|
260 case of a matrix two arguments of type @code{octave_idx_type} defining |
|
261 the number of rows and columns in the matrix. |
|
262 @end deftypefn |
|
263 |
|
264 @deftypefn Method T* fortran_vec (void) |
6569
|
265 This method returns a pointer to the underlying data of the matrix or a |
|
266 array so that it can be manipulated directly, either within Octave or by |
|
267 an external library. |
6572
|
268 @end deftypefn |
6569
|
269 |
6572
|
270 Operators such an @code{+}, @code{-}, or @code{*} can be used on the |
|
271 majority of the above types. In addition there are a number of methods |
|
272 that are of interest only for matrices such as @code{transpose}, |
|
273 @code{hermitian}, @code{solve}, etc. |
6569
|
274 |
|
275 The typical way to extract a matrix or array from the input arguments of |
6572
|
276 @code{DEFUN_DLD} function is as follows |
6569
|
277 |
|
278 @example |
|
279 @group |
|
280 #include <octave/oct.h> |
|
281 |
|
282 DEFUN_DLD (addtwomatrices, args, , "Add A to B") |
|
283 @{ |
|
284 int nargin = args.length (); |
|
285 if (nargin != 2) |
|
286 print_usage (); |
|
287 else |
|
288 @{ |
6572
|
289 NDArray A = args(0).array_value (); |
|
290 NDArray B = args(1).array_value (); |
6569
|
291 if (! error_state) |
6572
|
292 return octave_value (A + B); |
6569
|
293 @} |
|
294 return octave_value_list (); |
|
295 @} |
|
296 @end group |
|
297 @end example |
|
298 |
|
299 To avoid segmentation faults causing Octave to abort, this function |
|
300 explicitly checks that there are sufficient arguments available before |
6571
|
301 accessing these arguments. It then obtains two multi-dimensional arrays |
6572
|
302 of type @code{NDArray} and adds these together. Note that the array_value |
|
303 method is called without using the @code{is_matrix_type} type, and instead the |
6571
|
304 error_state is checked before returning @code{A + B}. The reason to |
6569
|
305 prefer this is that the arguments might be a type that is not an |
6572
|
306 @code{NDArray}, but it would make sense to convert it to one. The |
|
307 @code{array_value} method allows this conversion to be performed |
|
308 transparently if possible, and sets @code{error_state} if it is not. |
6569
|
309 |
6572
|
310 @code{A + B}, operating on two @code{NDArray}'s returns an |
|
311 @code{NDArray}, which is cast to an @code{octave_value} on the return |
|
312 from the function. An example of the use of this demonstration function |
|
313 is |
6569
|
314 |
|
315 @example |
|
316 @group |
6572
|
317 addtwomatrices (ones (2, 2), ones (2, 2)) |
6569
|
318 @result{} 2 2 |
|
319 2 2 |
|
320 @end group |
|
321 @end example |
|
322 |
6572
|
323 A list of the basic @code{Matrix} and @code{Array} types, the methods to |
|
324 extract these from an @code{octave_value} and the associated header is |
|
325 listed below. |
6569
|
326 |
|
327 @multitable @columnfractions .3 .4 .3 |
6572
|
328 @item @code{RowVector} @tab @code{row_vector_value} @tab @file{dRowVector.h} |
|
329 @item @code{ComplexRowVector} @tab @code{complex_row_vector_value} @tab @file{CRowVector.h} |
|
330 @item @code{ColumnVector} @tab @code{column_vector_value} @tab @file{dColVector.h} |
|
331 @item @code{ComplexColumnVector} @tab @code{complex_column_vector_value} @tab @file{CColVector.h} |
|
332 @item @code{Matrix} @tab @code{matrix_value} @tab @file{dMatrix.h} |
|
333 @item @code{ComplexMatrix} @tab @code{complex_matrix_value} @tab @file{CMatrix.h} |
|
334 @item @code{boolMatrix} @tab @code{bool_matrix_value} @tab @file{boolMatrix.h} |
|
335 @item @code{charMatrix} @tab @code{char_matrix_value} @tab @file{chMatrix.h} |
|
336 @item @code{NDArray} @tab @code{array_value} @tab @file{dNDArray.h} |
|
337 @item @code{ComplexNDArray} @tab @code{complex_array_value} @tab @file{CNDArray.h} |
|
338 @item @code{boolNDArray} @tab @code{bool_array_value} @tab @file{boolNDArray.h} |
|
339 @item @code{charNDArray} @tab @code{char_array_value} @tab @file{charNDArray.h} |
|
340 @item @code{int8NDArray} @tab @code{int8_array_value} @tab @file{int8NDArray.h} |
|
341 @item @code{int16NDArray} @tab @code{int16_array_value} @tab @file{int16NDArray.h} |
|
342 @item @code{int32NDArray} @tab @code{int32_array_value} @tab @file{int32NDArray.h} |
|
343 @item @code{int64NDArray} @tab @code{int64_array_value} @tab @file{int64NDArray.h} |
|
344 @item @code{uint8NDArray} @tab @code{uint8_array_value} @tab @file{uint8NDArray.h} |
|
345 @item @code{uint16NDArray} @tab @code{uint16_array_value} @tab @file{uint16NDArray.h} |
|
346 @item @code{uint32NDArray} @tab @code{uint32_array_value} @tab @file{uint32NDArray.h} |
|
347 @item @code{uint64NDArray} @tab @code{uint64_array_value} @tab @file{uint64NDArray.h} |
6569
|
348 @end multitable |
|
349 |
6572
|
350 @node Character Strings in Oct-Files |
|
351 @subsection Character Strings in Oct-Files |
|
352 |
|
353 In Octave a character string is just a special @code{Array} class. |
|
354 Consider the example |
|
355 |
|
356 @example |
|
357 @group |
|
358 #include <octave/oct.h> |
|
359 |
|
360 DEFUN_DLD (stringdemo, args, , "String Demo") |
|
361 @{ |
|
362 int nargin = args.length (); |
|
363 octave_value_list retval; |
|
364 |
|
365 if (nargin != 1) |
|
366 print_usage (); |
|
367 else |
|
368 @{ |
|
369 charMatrix ch = args(0).char_matrix_value (); |
|
370 |
|
371 if (! error_state) |
|
372 @{ |
|
373 if (args(0).is_sq_string ()) |
|
374 retval(1) = octave_value (ch, true); |
|
375 else |
|
376 retval(1) = octave_value (ch, true, '\''); |
|
377 |
|
378 octave_idx_type nr = ch.rows(); |
|
379 for (octave_idx_type i = 0; i < nr / 2; i++) |
|
380 @{ |
|
381 std::string tmp = ch.row_as_string (i); |
|
382 ch.insert (ch.row_as_string(nr-i-1).c_str(), i, 0); |
|
383 ch.insert (tmp.c_str(), nr-i-1, 0); |
|
384 @} |
|
385 retval(0) = octave_value (ch, true); |
|
386 @} |
|
387 @} |
|
388 return retval; |
|
389 @} |
|
390 @end group |
|
391 @end example |
|
392 |
|
393 An example of the of the use of this function is |
|
394 |
|
395 @example |
|
396 @group |
|
397 s0 = ["First String"; "Second String"]; |
|
398 [s1,s2] = stringdemo (s0) |
|
399 @result{} s1 = Second String |
|
400 First String |
|
401 |
|
402 @result{} s2 = First String |
|
403 Second String |
|
404 |
|
405 typeinfo (s2) |
|
406 @result{} sq_string |
|
407 typeinfo (s1) |
|
408 @result{} string |
|
409 @end group |
|
410 @end example |
|
411 |
|
412 One additional complication of strings in Octave is the difference |
|
413 between single quoted and double quoted strings. To find out if an |
|
414 @code{octave_value} contains a single or double quoted string an example is |
|
415 |
|
416 @example |
|
417 @group |
|
418 if (args(0).is_sq_string ()) |
|
419 octave_stdout << "First argument is a singularly quoted string\n"; |
|
420 else if (args(0).is_dq_string ()) |
|
421 octave_stdout << "First argument is a doubly quoted string\n"; |
|
422 @end group |
|
423 @end example |
|
424 |
|
425 Note however, that both types of strings are represented by the |
|
426 @code{charNDArray} type, and so when assigning to an |
|
427 @code{octave_value}, the type of string should be specified. For example |
|
428 |
|
429 @example |
|
430 @group |
|
431 octave_value_list retval; |
|
432 charNDArray c; |
|
433 @dots{} |
|
434 retval(0) = octave_value (ch, true); // Create a double quoted string |
|
435 retval(1) = octave_value (ch, true, '\''); // Create single quoted string |
|
436 @end group |
|
437 @end example |
|
438 |
|
439 @node Cell Arrays in Oct-Files |
|
440 @subsection Cell Arrays in Oct-Files |
|
441 |
|
442 Octave's cell type is equally accessible within an oct-files. A cell |
|
443 array is just an array of @code{octave_value}s, and so each element of the cell |
|
444 array can then be treated just like any other @code{octave_value}. A simple |
|
445 example is |
|
446 |
|
447 @example |
|
448 @group |
|
449 #include <octave/oct.h> |
|
450 #include <octave/Cell.h> |
|
451 |
|
452 DEFUN_DLD (celldemo, args, , "Cell Demo") |
|
453 @{ |
|
454 octave_value_list retval; |
|
455 int nargin = args.length(); |
|
456 |
|
457 if (nargin != 1) |
|
458 print_usage (); |
|
459 else |
|
460 @{ |
|
461 Cell c = args (0).cell_value (); |
|
462 if (! error_state) |
|
463 for (octave_idx_type i = 0; i < c.nelem (); i++) |
|
464 retval(i) = c.elem (i); |
|
465 @} |
|
466 return retval; |
|
467 @} |
|
468 @end group |
|
469 @end example |
|
470 |
|
471 Note that cell arrays are used less often in standard oct-files and so |
|
472 the @file{Cell.h} header file must be explicitly included. The rest of this |
|
473 example extracts the @code{octave_value}s one by one from the cell array and |
|
474 returns be as individual return arguments. For example consider |
|
475 |
|
476 @example |
|
477 @group |
|
478 [b1, b2, b3] = celldemo (@{1, [1, 2], "test"@}) |
|
479 @result{} |
|
480 b1 = 1 |
|
481 b2 = |
|
482 |
|
483 1 2 |
|
484 |
|
485 b3 = test |
|
486 @end group |
|
487 @end example |
|
488 |
|
489 @node Structures in Oct-Files |
|
490 @subsection Structures in Oct-Files |
|
491 |
|
492 A structure in Octave is map between a number of fields represented and |
|
493 their values. The Standard Template Library @code{map} class is used, |
|
494 with the pair consisting of a @code{std::string} and an octave |
|
495 @code{Cell} variable. |
|
496 |
|
497 A simple example demonstrating the use of structures within oct-files is |
|
498 |
|
499 @example |
|
500 @group |
|
501 #include <octave/oct.h> |
|
502 #include <octave/ov-struct.h> |
|
503 |
|
504 DEFUN_DLD (structdemo, args, , "Struct demo.") |
|
505 @{ |
|
506 int nargin = args.length (); |
|
507 octave_value retval; |
|
508 |
|
509 if (nargin != 2) |
|
510 print_usage (); |
|
511 else |
|
512 @{ |
|
513 Octave_map arg0 = args(0).map_value (); |
|
514 std::string arg1 = args(1).string_value (); |
|
515 |
|
516 if (! error_state && arg0.contains (arg1)) |
|
517 @{ |
|
518 // The following two lines might be written as |
|
519 // octave_value tmp; |
|
520 // for (Octave_map::iterator p0 = arg0.begin() ; |
|
521 // p0 != arg0.end(); p0++ ) |
|
522 // if (arg0.key (p0) == arg1) |
|
523 // @{ |
|
524 // tmp = arg0.contents (p0) (0); |
|
525 // break; |
|
526 // @} |
|
527 // though using seek is more concise. |
|
528 Octave_map::const_iterator p1 = arg0.seek (arg1); |
|
529 octave_value tmp = arg0.contents( p1 ) (0); |
|
530 Octave_map st; |
|
531 st.assign ("selected", tmp); |
|
532 retval = octave_value (st); |
|
533 @} |
|
534 @} |
|
535 return retval; |
|
536 @} |
|
537 @end group |
|
538 @end example |
|
539 |
|
540 An example of its use is |
|
541 |
|
542 @example |
|
543 @group |
|
544 x.a = 1; x.b = "test"; x.c = [1, 2]; |
|
545 structdemo (x, "b") |
|
546 @result{} selected = test |
|
547 @end group |
|
548 @end example |
|
549 |
|
550 The commented code above demonstrates how to iterated over all of the |
|
551 fields of the structure, where as the following code demonstrates finding |
|
552 a particular field in a more concise manner. |
|
553 |
|
554 As can be seen the @code{contents} method of the @code{Octave_map} class |
|
555 returns a @code{Cell} which allows structure arrays to be represented. |
|
556 Therefore, to obtain the underlying @code{octave_value} we write |
|
557 |
|
558 @example |
|
559 octave_value tmp = arg0.contents (p1) (0); |
|
560 @end example |
|
561 |
|
562 where the trailing (0) is the () operator on the @code{Cell} object. |
|
563 |
|
564 @node Sparse Matrices in Oct-Files |
|
565 @subsection Sparse Matrices in Oct-Files |
6569
|
566 |
|
567 There are three classes of sparse objects that are of interest to the |
|
568 user. |
|
569 |
6572
|
570 @table @code |
6569
|
571 @item SparseMatrix |
|
572 A double precision sparse matrix class |
|
573 @item SparseComplexMatrix |
|
574 A complex sparse matrix class |
|
575 @item SparseBoolMatrix |
|
576 A boolean sparse matrix class |
|
577 @end table |
|
578 |
|
579 All of these classes inherit from the @code{Sparse<T>} template class, |
6571
|
580 and so all have similar capabilities and usage. The @code{Sparse<T>} |
6569
|
581 class was based on Octave @code{Array<T>} class, and so users familiar |
6572
|
582 with Octave's @code{Array} classes will be comfortable with the use of |
6569
|
583 the sparse classes. |
|
584 |
|
585 The sparse classes will not be entirely described in this section, due |
6572
|
586 to their similarity with the existing @code{Array} classes. However, |
|
587 there are a few differences due the different nature of sparse objects, |
|
588 and these will be described. Firstly, although it is fundamentally |
|
589 possible to have N-dimensional sparse objects, the Octave sparse classes do |
6571
|
590 not allow them at this time. So all operations of the sparse classes |
6569
|
591 must be 2-dimensional. This means that in fact @code{SparseMatrix} is |
|
592 similar to Octave's @code{Matrix} class rather than its |
|
593 @code{NDArray} class. |
|
594 |
|
595 @menu |
6572
|
596 * Array and Sparse Differences:: |
|
597 * Creating Sparse Matrices in Oct-Files:: |
|
598 * Using Sparse Matrices in Oct-Files:: |
6569
|
599 @end menu |
|
600 |
6572
|
601 @node Array and Sparse Differences |
6569
|
602 @subsubsection The Differences between the Array and Sparse Classes |
|
603 |
|
604 The number of elements in a sparse matrix is considered to be the number |
6571
|
605 of non-zero elements rather than the product of the dimensions. Therefore |
6569
|
606 |
|
607 @example |
|
608 SparseMatrix sm; |
|
609 @dots{} |
|
610 int nel = sm.nelem (); |
|
611 @end example |
|
612 |
6571
|
613 returns the number of non-zero elements. If the user really requires the |
6569
|
614 number of elements in the matrix, including the non-zero elements, they |
6571
|
615 should use @code{numel} rather than @code{nelem}. Note that for very |
6569
|
616 large matrices, where the product of the two dimensions is large that |
|
617 the representation of the an unsigned int, then @code{numel} can overflow. |
|
618 An example is @code{speye(1e6)} which will create a matrix with a million |
6571
|
619 rows and columns, but only a million non-zero elements. Therefore the |
6569
|
620 number of rows by the number of columns in this case is more than two |
|
621 hundred times the maximum value that can be represented by an unsigned int. |
|
622 The use of @code{numel} should therefore be avoided useless it is known |
|
623 it won't overflow. |
|
624 |
|
625 Extreme care must be take with the elem method and the "()" operator, |
6571
|
626 which perform basically the same function. The reason is that if a |
6569
|
627 sparse object is non-const, then Octave will assume that a |
6571
|
628 request for a zero element in a sparse matrix is in fact a request |
|
629 to create this element so it can be filled. Therefore a piece of |
6569
|
630 code like |
|
631 |
|
632 @example |
|
633 SparseMatrix sm; |
|
634 @dots{} |
|
635 for (int j = 0; j < nc; j++) |
|
636 for (int i = 0; i < nr; i++) |
6571
|
637 std::cerr << " (" << i << "," << j << "): " << sm(i,j) |
6569
|
638 << std::endl; |
|
639 @end example |
|
640 |
|
641 is a great way of turning the sparse matrix into a dense one, and a |
|
642 very slow way at that since it reallocates the sparse object at each |
|
643 zero element in the matrix. |
|
644 |
|
645 An easy way of preventing the above from happening is to create a temporary |
6571
|
646 constant version of the sparse matrix. Note that only the container for |
6569
|
647 the sparse matrix will be copied, while the actual representation of the |
6571
|
648 data will be shared between the two versions of the sparse matrix. So this |
|
649 is not a costly operation. For example, the above would become |
6569
|
650 |
|
651 @example |
|
652 SparseMatrix sm; |
|
653 @dots{} |
|
654 const SparseMatrix tmp (sm); |
|
655 for (int j = 0; j < nc; j++) |
|
656 for (int i = 0; i < nr; i++) |
6571
|
657 std::cerr << " (" << i << "," << j << "): " << tmp(i,j) |
6569
|
658 << std::endl; |
|
659 @end example |
|
660 |
|
661 Finally, as the sparse types aren't just represented as a contiguous |
|
662 block of memory, the @code{fortran_vec} method of the @code{Array<T>} |
6571
|
663 is not available. It is however replaced by three separate methods |
6569
|
664 @code{ridx}, @code{cidx} and @code{data}, that access the raw compressed |
|
665 column format that the Octave sparse matrices are stored in. |
|
666 Additionally, these methods can be used in a manner similar to @code{elem}, |
6571
|
667 to allow the matrix to be accessed or filled. However, in that case it is |
6569
|
668 up to the user to respect the sparse matrix compressed column format |
|
669 discussed previous. |
|
670 |
6572
|
671 @node Creating Sparse Matrices in Oct-Files |
|
672 @subsubsection Creating Sparse Matrices in Oct-Files |
6569
|
673 |
6572
|
674 You have several alternatives for creating a sparse matrix. |
|
675 You can first create the data as three vectors representing the |
6569
|
676 row and column indexes and the data, and from those create the matrix. |
6572
|
677 Or alternatively, you can create a sparse matrix with the appropriate |
6571
|
678 amount of space and then fill in the values. Both techniques have their |
6569
|
679 advantages and disadvantages. |
|
680 |
6572
|
681 Here is an example of how to create a small sparse matrix with the first |
|
682 technique |
6569
|
683 |
|
684 @example |
|
685 int nz = 4, nr = 3, nc = 4; |
|
686 ColumnVector ridx (nz); |
|
687 ColumnVector cidx (nz); |
|
688 ColumnVector data (nz); |
|
689 |
|
690 ridx(0) = 0; ridx(1) = 0; ridx(2) = 1; ridx(3) = 2; |
|
691 cidx(0) = 0; cidx(1) = 1; cidx(2) = 3; cidx(3) = 3; |
|
692 data(0) = 1; data(1) = 2; data(2) = 3; data(3) = 4; |
|
693 |
|
694 SparseMatrix sm (data, ridx, cidx, nr, nc); |
|
695 @end example |
|
696 |
6572
|
697 @noindent |
6571
|
698 which creates the matrix given in section @ref{Storage}. Note that |
6569
|
699 the compressed matrix format is not used at the time of the creation |
6571
|
700 of the matrix itself, however it is used internally. |
6569
|
701 |
|
702 As previously mentioned, the values of the sparse matrix are stored |
6571
|
703 in increasing column-major ordering. Although the data passed by the |
6569
|
704 user does not need to respect this requirement, the pre-sorting the |
|
705 data significantly speeds up the creation of the sparse matrix. |
|
706 |
|
707 The disadvantage of this technique of creating a sparse matrix is |
6571
|
708 that there is a brief time where two copies of the data exists. Therefore |
6569
|
709 for extremely memory constrained problems this might not be the right |
|
710 technique to create the sparse matrix. |
|
711 |
|
712 The alternative is to first create the sparse matrix with the desired |
6571
|
713 number of non-zero elements and then later fill those elements in. The |
|
714 easiest way to do this is |
6569
|
715 |
6571
|
716 @example |
6569
|
717 int nz = 4, nr = 3, nc = 4; |
|
718 SparseMatrix sm (nr, nc, nz); |
|
719 sm(0,0) = 1; sm(0,1) = 2; sm(1,3) = 3; sm(2,3) = 4; |
|
720 @end example |
|
721 |
6571
|
722 That creates the same matrix as previously. Again, although it is not |
6569
|
723 strictly necessary, it is significantly faster if the sparse matrix is |
|
724 created in this manner that the elements are added in column-major |
6571
|
725 ordering. The reason for this is that if the elements are inserted |
6569
|
726 at the end of the current list of known elements then no element |
|
727 in the matrix needs to be moved to allow the new element to be |
6571
|
728 inserted. Only the column indexes need to be updated. |
6569
|
729 |
|
730 There are a few further points to note about this technique of creating |
6572
|
731 a sparse matrix. Firstly, it is possible to create a sparse matrix |
6571
|
732 with fewer elements than are actually inserted in the matrix. Therefore |
6569
|
733 |
6571
|
734 @example |
6569
|
735 int nz = 4, nr = 3, nc = 4; |
|
736 SparseMatrix sm (nr, nc, 0); |
|
737 sm(0,0) = 1; sm(0,1) = 2; sm(1,3) = 3; sm(2,3) = 4; |
|
738 @end example |
|
739 |
6572
|
740 @noindent |
|
741 is perfectly valid. However it is a very bad idea. The reason is that |
6569
|
742 as each new element is added to the sparse matrix the space allocated |
6571
|
743 to it is increased by reallocating the memory. This is an expensive |
6569
|
744 operation, that will significantly slow this means of creating a sparse |
6572
|
745 matrix. Furthermore, it is possible to create a sparse matrix with |
|
746 too much storage, so having @var{nz} above equaling 6 is also valid. |
6569
|
747 The disadvantage is that the matrix occupies more memory than strictly |
|
748 needed. |
|
749 |
6572
|
750 It is not always easy to know the number of non-zero elements prior |
6571
|
751 to filling a matrix. For this reason the additional storage for the |
6569
|
752 sparse matrix can be removed after its creation with the |
6571
|
753 @dfn{maybe_compress} function. Furthermore, the maybe_compress can |
6569
|
754 deallocate the unused storage, but it can equally remove zero elements |
|
755 from the matrix. The removal of zero elements from the matrix is |
|
756 controlled by setting the argument of the @dfn{maybe_compress} function |
6572
|
757 to be @samp{true}. However, the cost of removing the zeros is high because it |
6571
|
758 implies resorting the elements. Therefore, if possible it is better |
|
759 is the user doesn't add the zeros in the first place. An example of |
6569
|
760 the use of @dfn{maybe_compress} is |
|
761 |
|
762 @example |
|
763 int nz = 6, nr = 3, nc = 4; |
|
764 SparseMatrix sm1 (nr, nc, nz); |
|
765 sm1(0,0) = 1; sm1(0,1) = 2; sm1(1,3) = 3; sm1(2,3) = 4; |
|
766 sm1.maybe_compress (); // No zero elements were added |
|
767 |
|
768 SparseMatrix sm2 (nr, nc, nz); |
6571
|
769 sm2(0,0) = 1; sm2(0,1) = 2; sm(0,2) = 0; sm(1,2) = 0; |
6569
|
770 sm1(1,3) = 3; sm1(2,3) = 4; |
|
771 sm2.maybe_compress (true); // Zero elements were added |
|
772 @end example |
|
773 |
|
774 The use of the @dfn{maybe_compress} function should be avoided if |
|
775 possible, as it will slow the creation of the matrices. |
|
776 |
|
777 A third means of creating a sparse matrix is to work directly with |
6571
|
778 the data in compressed row format. An example of this technique might |
6569
|
779 be |
|
780 |
|
781 @c Note the @verbatim environment is a relatively new addition to texinfo. |
6571
|
782 @c Therefore use the @example environment and replace @, with @@, |
6569
|
783 @c { with @{, etc |
|
784 |
|
785 @example |
|
786 octave_value arg; |
6571
|
787 |
6569
|
788 @dots{} |
|
789 |
6571
|
790 int nz = 6, nr = 3, nc = 4; // Assume we know the max no nz |
6569
|
791 SparseMatrix sm (nr, nc, nz); |
|
792 Matrix m = arg.matrix_value (); |
|
793 |
|
794 int ii = 0; |
|
795 sm.cidx (0) = 0; |
|
796 for (int j = 1; j < nc; j++) |
|
797 @{ |
|
798 for (int i = 0; i < nr; i++) |
|
799 @{ |
|
800 double tmp = foo (m(i,j)); |
|
801 if (tmp != 0.) |
|
802 @{ |
|
803 sm.data(ii) = tmp; |
|
804 sm.ridx(ii) = i; |
|
805 ii++; |
|
806 @} |
|
807 @} |
|
808 sm.cidx(j+1) = ii; |
|
809 @} |
|
810 sm.maybe_compress (); // If don't know a-priori the final no of nz. |
|
811 @end example |
|
812 |
6572
|
813 @noindent |
6569
|
814 which is probably the most efficient means of creating the sparse matrix. |
|
815 |
|
816 Finally, it might sometimes arise that the amount of storage initially |
6571
|
817 created is insufficient to completely store the sparse matrix. Therefore, |
6569
|
818 the method @code{change_capacity} exists to reallocate the sparse memory. |
6571
|
819 The above example would then be modified as |
6569
|
820 |
|
821 @example |
|
822 octave_value arg; |
6571
|
823 |
6569
|
824 @dots{} |
|
825 |
6571
|
826 int nz = 6, nr = 3, nc = 4; // Assume we know the max no nz |
6569
|
827 SparseMatrix sm (nr, nc, nz); |
|
828 Matrix m = arg.matrix_value (); |
|
829 |
|
830 int ii = 0; |
|
831 sm.cidx (0) = 0; |
|
832 for (int j = 1; j < nc; j++) |
|
833 @{ |
|
834 for (int i = 0; i < nr; i++) |
|
835 @{ |
|
836 double tmp = foo (m(i,j)); |
|
837 if (tmp != 0.) |
|
838 @{ |
|
839 if (ii == nz) |
|
840 @{ |
|
841 nz += 2; // Add 2 more elements |
|
842 sm.change_capacity (nz); |
|
843 @} |
|
844 sm.data(ii) = tmp; |
|
845 sm.ridx(ii) = i; |
|
846 ii++; |
|
847 @} |
|
848 @} |
|
849 sm.cidx(j+1) = ii; |
|
850 @} |
|
851 sm.maybe_mutate (); // If don't know a-priori the final no of nz. |
|
852 @end example |
|
853 |
|
854 Note that both increasing and decreasing the number of non-zero elements in |
6571
|
855 a sparse matrix is expensive, as it involves memory reallocation. Also as |
6569
|
856 parts of the matrix, though not its entirety, exist as the old and new copy |
6571
|
857 at the same time, additional memory is needed. Therefore if possible this |
6569
|
858 should be avoided. |
|
859 |
6572
|
860 @node Using Sparse Matrices in Oct-Files |
6569
|
861 @subsubsection Using Sparse Matrices in Oct-Files |
|
862 |
|
863 Most of the same operators and functions on sparse matrices that are |
|
864 available from the Octave are equally available with oct-files. |
|
865 The basic means of extracting a sparse matrix from an @code{octave_value} |
|
866 and returning them as an @code{octave_value}, can be seen in the |
|
867 following example |
|
868 |
|
869 @example |
|
870 octave_value_list retval; |
|
871 |
|
872 SparseMatrix sm = args(0).sparse_matrix_value (); |
|
873 SparseComplexMatrix scm = args(1).sparse_complex_matrix_value (); |
|
874 SparseBoolMatrix sbm = args(2).sparse_bool_matrix_value (); |
|
875 |
|
876 @dots{} |
|
877 |
|
878 retval(2) = sbm; |
|
879 retval(1) = scm; |
|
880 retval(0) = sm; |
|
881 @end example |
|
882 |
|
883 The conversion to an octave-value is handled by the sparse |
|
884 @code{octave_value} constructors, and so no special care is needed. |
|
885 |
|
886 @node Accessing Global Variables in Oct-Files |
|
887 @subsection Accessing Global Variables in Oct-Files |
|
888 |
|
889 Global variables allow variables in the global scope to be |
6571
|
890 accessed. Global variables can easily be accessed with oct-files using |
6569
|
891 the support functions @code{get_global_value} and |
6571
|
892 @code{set_global_value}. @code{get_global_value} takes two arguments, |
|
893 the first is a string representing the variable name to obtain. The |
6569
|
894 second argument is a boolean argument specifying what to do in the case |
6571
|
895 that no global variable of the desired name is found. An example of the |
6569
|
896 use of these two functions is |
|
897 |
|
898 @example |
|
899 @group |
|
900 #include <octave/oct.h> |
|
901 |
|
902 DEFUN_DLD (globaldemo, args, , "Global demo.") |
|
903 @{ |
|
904 int nargin = args.length(); |
|
905 octave_value retval; |
|
906 |
|
907 if (nargin != 1) |
|
908 print_usage (); |
|
909 else |
|
910 @{ |
|
911 std::string s = args(0).string_value (); |
|
912 if (! error_state) |
|
913 @{ |
|
914 octave_value tmp = get_global_value (s, true); |
|
915 if (tmp.is_defined ()) |
|
916 retval = tmp; |
|
917 else |
|
918 retval = "Global variable not found"; |
|
919 |
|
920 set_global_value ("a", 42.0); |
|
921 @} |
|
922 @} |
|
923 return retval; |
|
924 @} |
|
925 @end group |
6571
|
926 @end example |
6569
|
927 |
|
928 An example of its use is |
|
929 |
|
930 @example |
|
931 @group |
|
932 global a b |
|
933 b = 10; |
|
934 globaldemo ("b") |
|
935 @result{} 10 |
|
936 globaldemo ("c") |
|
937 @result{} "Global variable not found" |
|
938 num2str (a) |
|
939 @result{} 42 |
|
940 @end group |
|
941 @end example |
|
942 |
|
943 @node Calling Octave Functions from Oct-Files |
|
944 @subsection Calling Octave Functions from Oct-Files |
|
945 |
|
946 There is often a need to be able to call another octave function from |
|
947 within an oct-file, and there are many examples of such within octave |
6571
|
948 itself. For example the @code{quad} function is an oct-file that |
6569
|
949 calculates the definite integral by quadrature over a user supplied |
|
950 function. |
|
951 |
6571
|
952 There are also many ways in which a function might be passed. It might |
|
953 be passed as one of |
6569
|
954 |
|
955 @enumerate 1 |
|
956 @item Function Handle |
|
957 @item Anonymous Function Handle |
|
958 @item Inline Function |
|
959 @item String |
|
960 @end enumerate |
|
961 |
|
962 The example below demonstrates an example that accepts all four means of |
6571
|
963 passing a function to an oct-file. |
6569
|
964 |
|
965 @example |
|
966 @group |
|
967 #include <octave/oct.h> |
|
968 #include <octave/parse.h> |
|
969 |
|
970 DEFUN_DLD (funcdemo, args, nargout, "Function Demo") |
|
971 @{ |
|
972 int nargin = args.length(); |
|
973 octave_value_list retval; |
|
974 |
|
975 if (nargin < 2) |
|
976 print_usage (); |
|
977 else |
|
978 @{ |
|
979 octave_value_list newargs; |
|
980 for (octave_idx_type i = nargin - 1; i > 0; i--) |
|
981 newargs (i - 1) = args(i); |
6571
|
982 if (args(0).is_function_handle () || |
6569
|
983 args(0).is_inline_function ()) |
|
984 @{ |
|
985 octave_function *fcn = args(0).function_value (); |
|
986 if (! error_state) |
|
987 retval = feval (fcn, newargs, nargout); |
|
988 @} |
|
989 else if (args(0).is_string ()) |
|
990 @{ |
|
991 std::string fcn = args (0).string_value (); |
|
992 if (! error_state) |
|
993 retval = feval (fcn, newargs, nargout); |
|
994 @} |
|
995 else |
|
996 error ("funcdemo: expected string, inline or function handle"); |
|
997 @} |
|
998 return retval; |
|
999 @} |
|
1000 @end group |
|
1001 @end example |
|
1002 |
|
1003 The first argument to this demonstration is the user supplied function |
|
1004 and the following arguments are all passed to the user function. |
|
1005 |
|
1006 @example |
|
1007 @group |
6572
|
1008 funcdemo (@@sin,1) |
6569
|
1009 @result{} 0.84147 |
6572
|
1010 funcdemo (@@(x) sin(x), 1) |
6569
|
1011 @result{} 0.84147 |
6572
|
1012 funcdemo (inline ("sin(x)"), 1) |
6569
|
1013 @result{} 0.84147 |
6572
|
1014 funcdemo ("sin",1) |
6569
|
1015 @result{} 0.84147 |
|
1016 funcdemo (@@atan2, 1, 1) |
|
1017 @result{} 0.78540 |
|
1018 @end group |
|
1019 @end example |
|
1020 |
|
1021 When the user function is passed as a string, the treatment of the |
6571
|
1022 function is different. In some cases it is necessary to always have the |
6572
|
1023 user supplied function as an @code{octave_function} object. In that |
|
1024 case the string argument can be used to create a temporary function like |
6569
|
1025 |
|
1026 @example |
|
1027 @group |
|
1028 std::octave fcn_name = unique_symbol_name ("__fcn__"); |
|
1029 std::string fname = "function y = "; |
|
1030 fname.append (fcn_name); |
|
1031 fname.append ("(x) y = "); |
6571
|
1032 fcn = extract_function (args(0), "funcdemo", fcn_name, |
6569
|
1033 fname, "; endfunction"); |
|
1034 @dots{} |
6572
|
1035 if (fcn_name.length ()) |
6569
|
1036 clear_function (fcn_name); |
|
1037 @end group |
|
1038 @end example |
|
1039 |
6571
|
1040 There are two important things to know in this case. The number of input |
6569
|
1041 arguments to the user function is fixed, and in the above is a single |
|
1042 argument, and secondly to avoid leaving the temporary function in the |
|
1043 Octave symbol table it should be cleared after use. |
|
1044 |
|
1045 @node Calling External Code from Oct-Files |
|
1046 @subsection Calling External Code from Oct-Files |
|
1047 |
|
1048 Linking external C code to Octave is relatively simple, as the C |
6571
|
1049 functions can easily be called directly from C++. One possible issue is |
6569
|
1050 the declarations of the external C functions might need to be explicitly |
6571
|
1051 defined as C functions to the compiler. If the declarations of the |
6569
|
1052 external C functions are in the header @code{foo.h}, then the manner in |
|
1053 which to ensure that the C++ compiler treats these declarations as C |
|
1054 code is |
|
1055 |
|
1056 @example |
|
1057 @group |
|
1058 #ifdef __cplusplus |
6571
|
1059 extern "C" |
6569
|
1060 @{ |
|
1061 #endif |
|
1062 #include "foo.h" |
|
1063 #ifdef __cplusplus |
|
1064 @} /* end extern "C" */ |
|
1065 #endif |
|
1066 @end group |
|
1067 @end example |
|
1068 |
6571
|
1069 Calling Fortran code however can pose some difficulties. This is due to |
6569
|
1070 differences in the manner in compilers treat the linking of Fortran code |
6571
|
1071 with C or C++ code. Octave supplies a number of macros that allow |
6569
|
1072 consistent behavior across a number of compilers. |
|
1073 |
|
1074 The underlying Fortran code should use the @code{XSTOPX} function to |
6571
|
1075 replace the Fortran @code{STOP} function. @code{XSTOPX} uses the Octave |
6569
|
1076 exception handler to treat failing cases in the fortran code |
6571
|
1077 explicitly. Note that Octave supplies its own replacement blas |
6569
|
1078 @code{XERBLA} function, which uses @code{XSTOPX}. |
|
1079 |
6572
|
1080 If the underlying code calls @code{XSTOPX}, then the @code{F77_XFCN} |
6571
|
1081 macro should be used to call the underlying fortran function. The Fortran |
6569
|
1082 exception state can then be checked with the global variable |
6572
|
1083 @code{f77_exception_encountered}. If @code{XSTOPX} will not be called, |
6569
|
1084 then the @code{F77_FCN} macro should be used instead to call the Fortran |
|
1085 code. |
|
1086 |
|
1087 There is no harm in using @code{F77_XFCN} in all cases, except that for |
|
1088 Fortran code that is short running and executes a large number of times, |
6571
|
1089 there is potentially an overhead in doing so. However, if @code{F77_FCN} |
6569
|
1090 is used with code that calls @code{XSTOP}, Octave can generate a |
|
1091 segmentation fault. |
|
1092 |
|
1093 An example of the inclusion of a Fortran function in an oct-file is |
|
1094 given in the following example, where the C++ wrapper is |
|
1095 |
|
1096 @example |
|
1097 @group |
|
1098 #include <octave/oct.h> |
|
1099 #include <octave/f77-fcn.h> |
|
1100 |
6571
|
1101 extern "C" |
6569
|
1102 @{ |
6571
|
1103 F77_RET_T |
|
1104 F77_FUNC (fortsub, FORTSUB) |
|
1105 (const int&, double*, F77_CHAR_ARG_DECL |
6569
|
1106 F77_CHAR_ARG_LEN_DECL); |
|
1107 @} |
|
1108 |
6572
|
1109 DEFUN_DLD (fortdemo, args, , "Fortran Demo.") |
6569
|
1110 @{ |
6571
|
1111 octave_value_list retval; |
6569
|
1112 int nargin = args.length(); |
|
1113 if (nargin != 1) |
|
1114 print_usage (); |
|
1115 else |
|
1116 @{ |
|
1117 NDArray a = args(0).array_value (); |
|
1118 if (! error_state) |
|
1119 @{ |
|
1120 double *av = a.fortran_vec (); |
|
1121 octave_idx_type na = a.nelem (); |
|
1122 OCTAVE_LOCAL_BUFFER (char, ctmp, 128); |
|
1123 |
6572
|
1124 F77_XFCN (fortsub, FORTSUB, (na, av, ctmp F77_CHAR_ARG_LEN (128))); |
6569
|
1125 |
6572
|
1126 if (f77_exception_encountered) |
6569
|
1127 error ("fortdemo: error in fortran"); |
|
1128 else |
|
1129 @{ |
|
1130 retval(1) = std::string (ctmp); |
|
1131 retval(0) = a; |
|
1132 @} |
|
1133 @} |
|
1134 @} |
|
1135 return retval; |
|
1136 @} |
|
1137 @end group |
|
1138 @end example |
|
1139 |
6572
|
1140 @noindent |
6569
|
1141 and the fortran function is |
|
1142 |
|
1143 @example |
|
1144 @group |
|
1145 subroutine fortsub (n, a, s) |
|
1146 implicit none |
|
1147 character*(*) s |
|
1148 real*8 a(*) |
|
1149 integer*4 i, n, ioerr |
|
1150 do i = 1, n |
|
1151 if (a (i) .eq. 0d0) then |
6572
|
1152 call xstopx ('fortsub:divide by zero') |
6569
|
1153 else |
|
1154 a (i) = 1d0 / a (i) |
|
1155 end if |
|
1156 end do |
6572
|
1157 write (unit = s, fmt = '(a,i3,a,a)', iostat = ioerr) |
|
1158 $ 'There are ', n, ' values in the input vector', |
|
1159 $ char(0) |
6569
|
1160 if (ioerr .ne. 0) then |
6572
|
1161 call xstopx ('fortsub: error writing string') |
6569
|
1162 end if |
|
1163 return |
|
1164 end |
|
1165 @end group |
|
1166 @end example |
|
1167 |
|
1168 This example demonstrates most of the features needed to link to an |
|
1169 external Fortran function, including passing arrays and strings, as well |
6571
|
1170 as exception handling. An example of the behavior of this function is |
6569
|
1171 |
|
1172 @example |
|
1173 @group |
6572
|
1174 [b, s] = fortdemo (1:3) |
6569
|
1175 @result{} |
|
1176 b = 1.00000 0.50000 0.33333 |
|
1177 s = There are 3 values in the input vector |
|
1178 [b, s] = fortdemo(0:3) |
|
1179 error: fortsub:divide by zero |
|
1180 error: exception encountered in Fortran subroutine fortsub_ |
|
1181 error: fortdemo: error in fortran |
|
1182 @end group |
|
1183 @end example |
|
1184 |
|
1185 @node Allocating Local Memory in Oct-Files |
|
1186 @subsection Allocating Local Memory in Oct-Files |
|
1187 |
|
1188 Allocating memory within an oct-file might seem easy as the C++ |
6571
|
1189 new/delete operators can be used. However, in that case care must be |
|
1190 taken to avoid memory leaks. The preferred manner in which to allocate |
6572
|
1191 memory for use locally is to use the @code{OCTAVE_LOCAL_BUFFER} macro. |
|
1192 An example of its use is |
6569
|
1193 |
|
1194 @example |
|
1195 OCTAVE_LOCAL_BUFFER (double, tmp, len) |
|
1196 @end example |
|
1197 |
|
1198 that returns a pointer @code{tmp} of type @code{double *} of length |
|
1199 @code{len}. |
|
1200 |
|
1201 @node Input Parameter Checking in Oct-Files |
|
1202 @subsection Input Parameter Checking in Oct-Files |
|
1203 |
|
1204 WRITE ME |
|
1205 |
|
1206 @node Exception and Error Handling in Oct-Files |
|
1207 @subsection Exception and Error Handling in Oct-Files |
|
1208 |
|
1209 Another important feature of Octave is its ability to react to the user |
6571
|
1210 typing @kbd{Control-C} even during calculations. This ability is based on the |
6569
|
1211 C++ exception handler, where memory allocated by the C++ new/delete |
6571
|
1212 methods are automatically released when the exception is treated. When |
6569
|
1213 writing an oct-file, to allow Octave to treat the user typing @kbd{Control-C}, |
6571
|
1214 the @code{OCTAVE_QUIT} macro is supplied. For example |
6569
|
1215 |
|
1216 @example |
|
1217 @group |
|
1218 for (octave_idx_type i = 0; i < a.nelem (); i++) |
|
1219 @{ |
|
1220 OCTAVE_QUIT; |
|
1221 b.elem(i) = 2. * a.elem(i); |
|
1222 @} |
|
1223 @end group |
|
1224 @end example |
|
1225 |
6572
|
1226 The presence of the @code{OCTAVE_QUIT} macro in the inner loop allows Octave to |
6571
|
1227 treat the user request with the @kbd{Control-C}. Without this macro, the user |
6569
|
1228 must either wait for the function to return before the interrupt is |
|
1229 processed, or press @kbd{Control-C} three times to force Octave to exit. |
|
1230 |
6572
|
1231 The @code{OCTAVE_QUIT} macro does impose a very small speed penalty, and so for |
6569
|
1232 loops that are known to be small it might not make sense to include |
6572
|
1233 @code{OCTAVE_QUIT}. |
6569
|
1234 |
|
1235 When creating an oct-file that uses an external libraries, the function |
|
1236 might spend a significant portion of its time in the external |
6572
|
1237 library. It is not generally possible to use the @code{OCTAVE_QUIT} macro in |
6571
|
1238 this case. The alternative in this case is |
6569
|
1239 |
|
1240 @example |
|
1241 @group |
|
1242 BEGIN_INTERRUPT_IMMEDIATELY_IN_FOREIGN_CODE; |
|
1243 @dots{} some code that calls a "foreign" function @dots{} |
|
1244 END_INTERRUPT_IMMEDIATELY_IN_FOREIGN_CODE; |
|
1245 @end group |
|
1246 @end example |
|
1247 |
|
1248 The disadvantage of this is that if the foreign code allocates any |
|
1249 memory internally, then this memory might be lost during an interrupt, |
6571
|
1250 without being deallocated. Therefore, ideally Octave itself should |
6569
|
1251 allocate any memory that is needed by the foreign code, with either the |
6572
|
1252 fortran_vec method or the @code{OCTAVE_LOCAL_BUFFER} macro. |
6569
|
1253 |
6571
|
1254 The Octave unwind_protect mechanism (@ref{The unwind_protect Statement}) |
|
1255 can also be used in oct-files. In conjunction with the exception |
6569
|
1256 handling of Octave, it is important to enforce that certain code is run |
6571
|
1257 to allow variables, etc to be restored even if an exception occurs. An |
6569
|
1258 example of the use of this mechanism is |
|
1259 |
|
1260 @example |
|
1261 @group |
|
1262 #include <octave/oct.h> |
|
1263 #include <octave/unwind-prot.h> |
|
1264 |
|
1265 void |
|
1266 err_hand (const char *fmt, ...) |
|
1267 @{ |
|
1268 // Do nothing!! |
|
1269 @} |
|
1270 |
|
1271 DEFUN_DLD (unwinddemo, args, nargout, "Unwind Demo") |
|
1272 @{ |
|
1273 int nargin = args.length(); |
|
1274 octave_value retval; |
|
1275 if (nargin < 2) |
|
1276 print_usage (); |
|
1277 else |
|
1278 @{ |
|
1279 NDArray a = args(0).array_value (); |
|
1280 NDArray b = args(1).array_value (); |
|
1281 |
|
1282 if (! error_state) |
|
1283 @{ |
6572
|
1284 unwind_protect::begin_frame ("Funwinddemo"); |
|
1285 unwind_protect_ptr (current_liboctave_warning_handler); |
|
1286 set_liboctave_warning_handler (err_hand); |
|
1287 retval = octave_value (quotient (a, b)); |
|
1288 unwind_protect::run_frame ("Funwinddemo"); |
6569
|
1289 @} |
|
1290 @} |
|
1291 return retval; |
|
1292 @} |
|
1293 @end group |
|
1294 @end example |
|
1295 |
|
1296 As can be seen in the example |
|
1297 |
|
1298 @example |
|
1299 @group |
6572
|
1300 unwinddemo (1, 0) |
6569
|
1301 @result{} Inf |
|
1302 1 / 0 |
|
1303 @result{} warning: division by zero |
|
1304 Inf |
|
1305 @end group |
|
1306 @end example |
|
1307 |
|
1308 The division by zero (and in fact all warnings) is disabled in the |
|
1309 @code{unwinddemo} function. |
|
1310 |
|
1311 @node Documentation and Test of Oct-Files |
|
1312 @subsection Documentation and Test of Oct-Files |
|
1313 |
6572
|
1314 WRITE ME, reference how to use Texinfo in oct-file and embed test code. |
6569
|
1315 |
|
1316 @node Application Programming Interface for Oct-Files |
|
1317 @subsection Application Programming Interface for Oct-Files |
|
1318 |
|
1319 WRITE ME, using Coda section 1.3 as a starting point. |
|
1320 |
|
1321 @node Mex-Files |
|
1322 @section Mex-Files |
|
1323 @cindex mex-files |
|
1324 @cindex mex |
|
1325 |
|
1326 Octave includes an interface to allow legacy mex-files to be compiled |
6571
|
1327 and used with Octave. This interface can also be used to share code |
|
1328 between Octave and non Octave users. However, as mex-files expose the |
6569
|
1329 intern API of a product alternative to Octave, and the internal |
|
1330 structure of Octave is different to this product, a mex-file can never |
6571
|
1331 have the same performance in Octave as the equivalent oct-file. In |
6569
|
1332 particular to support the manner in which mex-files access the variables |
|
1333 passed to mex functions, there are a significant number of additional |
6571
|
1334 copies of memory when calling or returning from a mex function. For this |
6569
|
1335 reason, new code should be written using the oct-file interface |
|
1336 discussed above if possible. |
|
1337 |
|
1338 @menu |
6572
|
1339 * Getting Started with Mex-Files:: |
|
1340 * Structures with Mex-Files:: |
|
1341 * Sparse Matrices with Mex-Files:: |
|
1342 * Calling External Functions in Mex-Files:: |
6569
|
1343 @end menu |
|
1344 |
|
1345 @node Getting Started with Mex-Files |
|
1346 @subsection Getting Started with Mex-Files |
|
1347 |
|
1348 The basic command to build a mex-file is either @code{mkoctfile --mex} or |
6571
|
1349 @code{mex}. The first can either be used from within Octave or from the |
|
1350 commandline. However, to avoid issues with the installation of other |
6569
|
1351 products, the use of the command @code{mex} is limited to within Octave. |
|
1352 |
|
1353 @DOCSTRING(mex) |
|
1354 |
|
1355 @DOCSTRING(mexext) |
|
1356 |
|
1357 One important difference between the use of mex with other products and |
|
1358 with Octave is that the header file "matrix.h" is implicitly included |
6571
|
1359 through the inclusion of "mex.h". This is to avoid a conflict with the |
6569
|
1360 Octave file "Matrix.h" with operating systems and compilers that don't |
|
1361 distinguish between filenames in upper and lower case |
|
1362 |
|
1363 Consider the short example |
|
1364 |
|
1365 @example |
|
1366 @group |
|
1367 #include "mex.h" |
|
1368 |
|
1369 void |
|
1370 mexFunction (int nlhs, mxArray *plhs[], int nrhs, const mxArray *prhs[]) |
|
1371 @{ |
|
1372 mxArray *v = mxCreateDoubleMatrix (1, 1, mxREAL); |
|
1373 double *data = mxGetPr (v); |
|
1374 *data = 1.23456789; |
|
1375 plhs[0] = v; |
|
1376 @} |
|
1377 @end group |
|
1378 @end example |
|
1379 |
|
1380 This simple example demonstrates the basics of writing a mex-file. |
|
1381 |
|
1382 WRITE ME |
|
1383 |
6572
|
1384 @node Structures with Mex-Files |
|
1385 @subsection Structures with Mex-Files |
6569
|
1386 |
|
1387 WRITE ME |
|
1388 |
|
1389 @node Sparse Matrices with Mex-Files |
|
1390 @subsection Sparse Matrices with Mex-Files |
|
1391 |
|
1392 WRITE ME |
|
1393 |
|
1394 @node Calling External Functions in Mex-Files |
|
1395 @subsection Calling External Functions in Mex-Files |
|
1396 |
|
1397 WRITE ME |
|
1398 |
|
1399 @node Standalone Programs |
|
1400 @section Standalone Programs |
|
1401 |
|
1402 The libraries Octave itself uses, can be utilized in standalone |
6571
|
1403 applications. These applications then have access, for example, to the |
6569
|
1404 array and matrix classes as well as to all the Octave algorithms. The |
|
1405 following C++ program, uses class Matrix from liboctave.a or |
|
1406 liboctave.so. |
|
1407 |
|
1408 @example |
|
1409 @group |
|
1410 #include <iostream> |
|
1411 #include <octave/oct.h> |
|
1412 int |
|
1413 main (void) |
|
1414 @{ |
|
1415 std::cout << "Hello Octave world!\n"; |
6572
|
1416 int n = 2; |
|
1417 Matrix a_matrix = Matrix (n, n); |
|
1418 for (octave_idx_type i = 0; i < n; i++) |
6569
|
1419 @{ |
6572
|
1420 for (octave_idx_type j = 0; j < n; j++) |
6569
|
1421 @{ |
6572
|
1422 a_matrix(row,column) = (i+1)*10 + (j+1); |
6569
|
1423 @} |
|
1424 @} |
|
1425 std::cout << a_matrix; |
|
1426 return 0; |
|
1427 @} |
|
1428 @end group |
|
1429 @end example |
|
1430 |
|
1431 mkoctfile can then be used to build a standalone application with a |
|
1432 command like |
|
1433 |
|
1434 @example |
|
1435 @group |
|
1436 $ mkoctfile --link-stand-alone hello.cc -o hello |
|
1437 $ ./hello |
|
1438 Hello Octave world! |
|
1439 11 12 |
|
1440 21 22 |
|
1441 $ |
|
1442 @end group |
|
1443 @end example |
|
1444 |
|
1445 Note that the application @code{hello} will be dynamically linked |
|
1446 against the octave libraries and any octave support libraries. |