1 | /* DebyeSpheres.c |
---|
2 | |
---|
3 | |
---|
4 | */ |
---|
5 | |
---|
6 | #include "XOPStandardHeaders.h" // Include ANSI headers, Mac headers, IgorXOP.h, XOP.h and XOPSupport.h |
---|
7 | #include "DebyeSpheres.h" |
---|
8 | |
---|
9 | //#pragma XOP_SET_STRUCT_PACKING // All structures are 2-byte-aligned. |
---|
10 | |
---|
11 | // Prototypes |
---|
12 | HOST_IMPORT int main(IORecHandle ioRecHandle); |
---|
13 | |
---|
14 | // Custom error codes |
---|
15 | //#define REQUIRES_IGOR_200 1 + FIRST_XOP_ERR |
---|
16 | #define NON_EXISTENT_WAVE 2 + FIRST_XOP_ERR |
---|
17 | #define REQUIRES_SP_OR_DP_WAVE 3 + FIRST_XOP_ERR |
---|
18 | |
---|
19 | |
---|
20 | |
---|
21 | /* Calculates the scattered intensity from a collection of spheres |
---|
22 | using Debye's method. |
---|
23 | |
---|
24 | everything was previously declared as float, some inputs were as double. |
---|
25 | |
---|
26 | -- to avoid confusion, calculate everything in double, pass everything in as double |
---|
27 | -- this is more precision than is necessary, but avoids strange results from type casting |
---|
28 | behind the scenes |
---|
29 | */ |
---|
30 | int |
---|
31 | DebyeSpheresX(AltiParamsPtr p) |
---|
32 | { |
---|
33 | double qv; // input q-value |
---|
34 | double ival; //output intensity value |
---|
35 | double *xv,*yv,*zv,*rv; //pointers to input xyz-rho coordinates |
---|
36 | int i,j; |
---|
37 | int npt; |
---|
38 | double rval,grid,vol,fQR,dum,dij; |
---|
39 | |
---|
40 | |
---|
41 | |
---|
42 | // check for all of the required waves |
---|
43 | if (p->rhowavH == NIL) { |
---|
44 | SetNaN64(&p->result); |
---|
45 | return NON_EXISTENT_WAVE; |
---|
46 | } |
---|
47 | if (p->zwavH == NIL) { |
---|
48 | SetNaN64(&p->result); |
---|
49 | return NON_EXISTENT_WAVE; |
---|
50 | } |
---|
51 | if (p->ywavH == NIL) { |
---|
52 | SetNaN64(&p->result); |
---|
53 | return NON_EXISTENT_WAVE; |
---|
54 | } |
---|
55 | if (p->xwavH == NIL) { |
---|
56 | SetNaN64(&p->result); |
---|
57 | return NON_EXISTENT_WAVE; |
---|
58 | } |
---|
59 | |
---|
60 | |
---|
61 | //check to see that all are float, not double |
---|
62 | /* |
---|
63 | if(WaveType(p->rhowavH) != NT_FP32 ) { |
---|
64 | SetNaN64(&p->result); |
---|
65 | return kExpectedNT_FP32; |
---|
66 | } |
---|
67 | if(WaveType(p->zwavH) != NT_FP32 ) { |
---|
68 | SetNaN64(&p->result); |
---|
69 | return kExpectedNT_FP32; |
---|
70 | } |
---|
71 | if(WaveType(p->ywavH) != NT_FP32 ) { |
---|
72 | SetNaN64(&p->result); |
---|
73 | return kExpectedNT_FP32; |
---|
74 | } |
---|
75 | if(WaveType(p->xwavH) != NT_FP32 ) { |
---|
76 | SetNaN64(&p->result); |
---|
77 | return kExpectedNT_FP32; |
---|
78 | } |
---|
79 | */ |
---|
80 | |
---|
81 | //check to see that all are double |
---|
82 | if(WaveType(p->rhowavH) != NT_FP64 ) { |
---|
83 | SetNaN64(&p->result); |
---|
84 | return kExpectedNT_FP64; |
---|
85 | } |
---|
86 | if(WaveType(p->zwavH) != NT_FP64 ) { |
---|
87 | SetNaN64(&p->result); |
---|
88 | return kExpectedNT_FP64; |
---|
89 | } |
---|
90 | if(WaveType(p->ywavH) != NT_FP64 ) { |
---|
91 | SetNaN64(&p->result); |
---|
92 | return kExpectedNT_FP64; |
---|
93 | } |
---|
94 | if(WaveType(p->xwavH) != NT_FP64 ) { |
---|
95 | SetNaN64(&p->result); |
---|
96 | return kExpectedNT_FP64; |
---|
97 | } |
---|
98 | |
---|
99 | |
---|
100 | // (NO) -- convert the input to float. Do all calculations as float. |
---|
101 | // do everything in double. no reason to do in float, except for the previous Altivec incarnation |
---|
102 | |
---|
103 | rval = p->Rprimary; // primary sphere radius |
---|
104 | grid = p->grid; // calling program should set this to 0.62*Rprimary |
---|
105 | qv = p->qval; |
---|
106 | |
---|
107 | // |
---|
108 | npt = (int) WavePoints(p->xwavH); //wavePoints returns long, number of XYZ points |
---|
109 | xv = (double*)WaveData(p->xwavH); //xyz locations |
---|
110 | yv = (double*)WaveData(p->ywavH); |
---|
111 | zv = (double*)WaveData(p->zwavH); |
---|
112 | rv = (double*)WaveData(p->rhowavH); |
---|
113 | |
---|
114 | |
---|
115 | vol = 4.0*3.1415927/3.0*rval*rval*rval; |
---|
116 | ival = 0.0; |
---|
117 | fQR = PhiQR(qv,rval); |
---|
118 | //do the i=j sum |
---|
119 | for(i=0;i<npt;i+=1) { |
---|
120 | dum = rv[i]*vol*fQR; |
---|
121 | ival += dum*dum; |
---|
122 | } |
---|
123 | //do the i!=j double sum |
---|
124 | dum = vol*vol*fQR*fQR; |
---|
125 | for(i=0;i<npt;i+=1) { |
---|
126 | for(j=(i+1);j<npt;j+=1) { |
---|
127 | dij=XYZDistance(xv[i],xv[j],yv[i],yv[j],zv[i],zv[j]) * grid; |
---|
128 | ival += 2.0*rv[i]*rv[j]*dum*sin(dij*qv)/dij/qv; |
---|
129 | } |
---|
130 | } |
---|
131 | |
---|
132 | p->result = ival; |
---|
133 | |
---|
134 | return 0; |
---|
135 | |
---|
136 | } |
---|
137 | |
---|
138 | |
---|
139 | double PhiQR(double qval, double rval) |
---|
140 | { |
---|
141 | double retval,qr; |
---|
142 | |
---|
143 | qr = qval*rval; |
---|
144 | retval = ( 3*(sin(qr) - qr*cos(qr))/qr/qr/qr ); |
---|
145 | return(retval); |
---|
146 | } |
---|
147 | |
---|
148 | double XYZDistance(double x1, double x2,double y1, double y2,double z1, double z2) |
---|
149 | { |
---|
150 | double retval,dx,dy,dz; |
---|
151 | |
---|
152 | dx = (x1-x2); |
---|
153 | dy = (y1-y2); |
---|
154 | dz = (z1-z2); |
---|
155 | retval = sqrt( dx*dx + dy*dy + dz*dz ); |
---|
156 | return(retval); |
---|
157 | } |
---|
158 | |
---|
159 | |
---|
160 | /* |
---|
161 | |
---|
162 | given the distances XYZ as a triplet (on a unit grid) |
---|
163 | return the maximum distance. The calling program must multiply by |
---|
164 | the grid dimension to get real distance |
---|
165 | |
---|
166 | */ |
---|
167 | int |
---|
168 | maxDistanceX(DistParamPtr p) |
---|
169 | { |
---|
170 | double dmax,dij; //output dmax value, dij |
---|
171 | double *xv,*yv,*zv; //pointers to input xyz coordinates |
---|
172 | int i,j; |
---|
173 | int npt; |
---|
174 | int p1,p2; |
---|
175 | |
---|
176 | // check for all of the required waves |
---|
177 | if (p->zwavH == NIL) { |
---|
178 | SetNaN64(&p->result); |
---|
179 | return NON_EXISTENT_WAVE; |
---|
180 | } |
---|
181 | if (p->ywavH == NIL) { |
---|
182 | SetNaN64(&p->result); |
---|
183 | return NON_EXISTENT_WAVE; |
---|
184 | } |
---|
185 | if (p->xwavH == NIL) { |
---|
186 | SetNaN64(&p->result); |
---|
187 | return NON_EXISTENT_WAVE; |
---|
188 | } |
---|
189 | |
---|
190 | //check to see that all are double |
---|
191 | if(WaveType(p->zwavH) != NT_FP64 ) { |
---|
192 | SetNaN64(&p->result); |
---|
193 | return kExpectedNT_FP64; |
---|
194 | } |
---|
195 | if(WaveType(p->ywavH) != NT_FP64 ) { |
---|
196 | SetNaN64(&p->result); |
---|
197 | return kExpectedNT_FP64; |
---|
198 | } |
---|
199 | if(WaveType(p->xwavH) != NT_FP64 ) { |
---|
200 | SetNaN64(&p->result); |
---|
201 | return kExpectedNT_FP64; |
---|
202 | } |
---|
203 | |
---|
204 | // |
---|
205 | npt = (int) WavePoints(p->xwavH); //wavePoints returns long, number of XYZ points |
---|
206 | xv = (double*)WaveData(p->xwavH); //xyz locations |
---|
207 | yv = (double*)WaveData(p->ywavH); |
---|
208 | zv = (double*)WaveData(p->zwavH); |
---|
209 | |
---|
210 | p1 = (int) p->p1; |
---|
211 | p2 = (int) p->p2; |
---|
212 | |
---|
213 | dmax = 0; |
---|
214 | //do the i!=j double loop, keeping the maximum distance |
---|
215 | |
---|
216 | for(i=p1;i<p2;i+=1) { |
---|
217 | for(j=(i+1);j<npt;j+=1) { |
---|
218 | // dij=XYZDistance(xv[i],xv[j],yv[i],yv[j],zv[i],zv[j]); |
---|
219 | dij = (xv[i]-xv[j])*(xv[i]-xv[j]) + (yv[i]-yv[j])*(yv[i]-yv[j]) + (zv[i]-zv[j])*(zv[i]-zv[j]); |
---|
220 | if(dij > dmax) { |
---|
221 | dmax = dij; |
---|
222 | } |
---|
223 | } |
---|
224 | } |
---|
225 | |
---|
226 | p->result = dmax; |
---|
227 | |
---|
228 | return 0; |
---|
229 | |
---|
230 | } |
---|
231 | |
---|
232 | /* |
---|
233 | |
---|
234 | given the distances XYZ as a triplet (on a unit grid) |
---|
235 | return the binned histogram of distances |
---|
236 | |
---|
237 | */ |
---|
238 | int |
---|
239 | binDistanceX(BinParamPtr p) |
---|
240 | { |
---|
241 | double *xv,*yv,*zv,*bv; //pointers to input xyz coordinates |
---|
242 | int i,j; |
---|
243 | int npt,numBins,binIndex; |
---|
244 | double grid,binWidth,val; |
---|
245 | int p1,p2; |
---|
246 | |
---|
247 | |
---|
248 | |
---|
249 | // check for all of the required waves |
---|
250 | if (p->bwavH == NIL) { |
---|
251 | SetNaN64(&p->result); |
---|
252 | return NON_EXISTENT_WAVE; |
---|
253 | } |
---|
254 | if (p->zwavH == NIL) { |
---|
255 | SetNaN64(&p->result); |
---|
256 | return NON_EXISTENT_WAVE; |
---|
257 | } |
---|
258 | if (p->ywavH == NIL) { |
---|
259 | SetNaN64(&p->result); |
---|
260 | return NON_EXISTENT_WAVE; |
---|
261 | } |
---|
262 | if (p->xwavH == NIL) { |
---|
263 | SetNaN64(&p->result); |
---|
264 | return NON_EXISTENT_WAVE; |
---|
265 | } |
---|
266 | |
---|
267 | //check to see that all are double |
---|
268 | if(WaveType(p->bwavH) != NT_FP64 ) { |
---|
269 | SetNaN64(&p->result); |
---|
270 | return kExpectedNT_FP64; |
---|
271 | } |
---|
272 | if(WaveType(p->zwavH) != NT_FP64 ) { |
---|
273 | SetNaN64(&p->result); |
---|
274 | return kExpectedNT_FP64; |
---|
275 | } |
---|
276 | if(WaveType(p->ywavH) != NT_FP64 ) { |
---|
277 | SetNaN64(&p->result); |
---|
278 | return kExpectedNT_FP64; |
---|
279 | } |
---|
280 | if(WaveType(p->xwavH) != NT_FP64 ) { |
---|
281 | SetNaN64(&p->result); |
---|
282 | return kExpectedNT_FP64; |
---|
283 | } |
---|
284 | |
---|
285 | // |
---|
286 | npt = (int) WavePoints(p->xwavH); //wavePoints returns long, number of XYZ points |
---|
287 | numBins = (int) WavePoints(p->bwavH); //wavePoints returns long, number of points in bin wave |
---|
288 | p1 = (int) p->p1; |
---|
289 | p2 = (int) p->p2; |
---|
290 | |
---|
291 | |
---|
292 | xv = (double*)WaveData(p->xwavH); //xyz locations |
---|
293 | yv = (double*)WaveData(p->ywavH); |
---|
294 | zv = (double*)WaveData(p->zwavH); |
---|
295 | bv = (double*)WaveData(p->bwavH); |
---|
296 | |
---|
297 | grid = p->grid; |
---|
298 | binWidth = p->binWidth; |
---|
299 | |
---|
300 | //do the i!=j double loop, |
---|
301 | for(i=p1;i<p2;i+=1) { |
---|
302 | for(j=(i+1);j<npt;j+=1) { |
---|
303 | val = XYZDistance(xv[i],xv[j],yv[i],yv[j],zv[i],zv[j])*grid; |
---|
304 | binIndex = (int)(val/binWidth-0.5); |
---|
305 | if(binIndex > numBins -1 ) { |
---|
306 | //Print "bad index" |
---|
307 | } else { |
---|
308 | bv[binIndex] += 1; |
---|
309 | } |
---|
310 | |
---|
311 | } |
---|
312 | } |
---|
313 | |
---|
314 | p->result = 0; |
---|
315 | |
---|
316 | return 0; |
---|
317 | |
---|
318 | } |
---|
319 | |
---|
320 | |
---|
321 | /* |
---|
322 | |
---|
323 | given the distances XYZ as a triplet (on a unit grid) and the SLD at each point, |
---|
324 | return the binned histogram of distances for each of the parwise interactions |
---|
325 | |
---|
326 | The returned binning is a matrix, and has to be assigned as such |
---|
327 | |
---|
328 | */ |
---|
329 | int |
---|
330 | binSLDDistanceX(BinSLDParamPtr p) |
---|
331 | { |
---|
332 | double *xv,*yv,*zv; //pointers to input xyz coordinates |
---|
333 | double *rho,*SLDLook,*PSFid; // rho and the SLD lookup vector |
---|
334 | int i,j; |
---|
335 | int npt,numBins,binIndex; |
---|
336 | double grid,binWidth,val,retVal; |
---|
337 | int p1,p2; |
---|
338 | int intSLD; |
---|
339 | |
---|
340 | |
---|
341 | // for accessing the 2D wave data to write the results |
---|
342 | waveHndl wavH,PSFwavH; |
---|
343 | // long numDimensions; |
---|
344 | // long dimensionSizes[MAX_DIMENSIONS+1]; |
---|
345 | double value[2]; // Pointers used for double data. |
---|
346 | long indices[MAX_DIMENSIONS]; |
---|
347 | // |
---|
348 | long rhoi,rhoj,rii,rji,PSFIndex; |
---|
349 | |
---|
350 | |
---|
351 | // check for all of the required waves |
---|
352 | if (p->PSFidH == NIL) { |
---|
353 | SetNaN64(&p->result); |
---|
354 | return NON_EXISTENT_WAVE; |
---|
355 | } |
---|
356 | if (p->SLDLookH == NIL) { |
---|
357 | SetNaN64(&p->result); |
---|
358 | return NON_EXISTENT_WAVE; |
---|
359 | } |
---|
360 | if (p->bwavH == NIL) { |
---|
361 | SetNaN64(&p->result); |
---|
362 | return NON_EXISTENT_WAVE; |
---|
363 | } |
---|
364 | if (p->rhowavH == NIL) { |
---|
365 | SetNaN64(&p->result); |
---|
366 | return NON_EXISTENT_WAVE; |
---|
367 | } |
---|
368 | if (p->zwavH == NIL) { |
---|
369 | SetNaN64(&p->result); |
---|
370 | return NON_EXISTENT_WAVE; |
---|
371 | } |
---|
372 | if (p->ywavH == NIL) { |
---|
373 | SetNaN64(&p->result); |
---|
374 | return NON_EXISTENT_WAVE; |
---|
375 | } |
---|
376 | if (p->xwavH == NIL) { |
---|
377 | SetNaN64(&p->result); |
---|
378 | return NON_EXISTENT_WAVE; |
---|
379 | } |
---|
380 | |
---|
381 | //check to see that all are double |
---|
382 | if(WaveType(p->PSFidH) != NT_FP64 ) { |
---|
383 | SetNaN64(&p->result); |
---|
384 | return kExpectedNT_FP64; |
---|
385 | } |
---|
386 | if(WaveType(p->SLDLookH) != NT_FP64 ) { |
---|
387 | SetNaN64(&p->result); |
---|
388 | return kExpectedNT_FP64; |
---|
389 | } |
---|
390 | if(WaveType(p->bwavH) != NT_FP64 ) { |
---|
391 | SetNaN64(&p->result); |
---|
392 | return kExpectedNT_FP64; |
---|
393 | } |
---|
394 | if(WaveType(p->rhowavH) != NT_FP64 ) { |
---|
395 | SetNaN64(&p->result); |
---|
396 | return kExpectedNT_FP64; |
---|
397 | } |
---|
398 | if(WaveType(p->zwavH) != NT_FP64 ) { |
---|
399 | SetNaN64(&p->result); |
---|
400 | return kExpectedNT_FP64; |
---|
401 | } |
---|
402 | if(WaveType(p->ywavH) != NT_FP64 ) { |
---|
403 | SetNaN64(&p->result); |
---|
404 | return kExpectedNT_FP64; |
---|
405 | } |
---|
406 | if(WaveType(p->xwavH) != NT_FP64 ) { |
---|
407 | SetNaN64(&p->result); |
---|
408 | return kExpectedNT_FP64; |
---|
409 | } |
---|
410 | |
---|
411 | |
---|
412 | // access the 2D wave data for writing using the direct method |
---|
413 | wavH = p->bwavH; |
---|
414 | if (wavH == NIL) |
---|
415 | return NOWAV; |
---|
416 | // |
---|
417 | PSFwavH = p->PSFidH; |
---|
418 | |
---|
419 | npt = (int) WavePoints(p->xwavH); //wavePoints returns long, number of XYZ points |
---|
420 | numBins = (int) WavePoints(p->bwavH); //wavePoints returns long, number of points in bin wave |
---|
421 | |
---|
422 | xv = (double*)WaveData(p->xwavH); //xyz locations |
---|
423 | yv = (double*)WaveData(p->ywavH); |
---|
424 | zv = (double*)WaveData(p->zwavH); |
---|
425 | rho = (double*)WaveData(p->rhowavH); |
---|
426 | SLDLook = (double*)WaveData(p->SLDLookH); |
---|
427 | PSFid = (double*)WaveData(p->PSFidH); //this one is 2D |
---|
428 | |
---|
429 | p1 = (int) p->p1; |
---|
430 | p2 = (int) p->p2; |
---|
431 | |
---|
432 | intSLD = (int) p->minSLD; //convert to int for use as index |
---|
433 | |
---|
434 | grid = p->grid; |
---|
435 | binWidth = p->binWidth; |
---|
436 | |
---|
437 | //do the i!=j double loop, |
---|
438 | for(i=p1;i<p2;i+=1) { |
---|
439 | for(j=(i+1);j<npt;j+=1) { |
---|
440 | val = XYZDistance(xv[i],xv[j],yv[i],yv[j],zv[i],zv[j])*grid; |
---|
441 | binIndex = (int)(val/binWidth-0.5); |
---|
442 | if(binIndex > numBins -1 ) { |
---|
443 | //Print "bad index" |
---|
444 | } else { |
---|
445 | rhoi = (long) rho[i]; //get the rho value at i and j |
---|
446 | rhoj = (long) rho[j]; |
---|
447 | rii = (long) SLDLook[rhoi+intSLD]; //rho i index |
---|
448 | rji = (long) SLDLook[rhoj+intSLD]; //rho j index |
---|
449 | MemClear(indices, sizeof(indices)); // Must be 0 for unused dimensions. |
---|
450 | indices[0] = rii; |
---|
451 | indices[1] = rji; |
---|
452 | if (retVal = MDGetNumericWavePointValue(PSFwavH, indices, value)) |
---|
453 | return retVal; |
---|
454 | //PSFIndex = (long) PSFid[rii][rji]; //doesn't work |
---|
455 | PSFIndex = (long) value[0]; |
---|
456 | |
---|
457 | //now do the assignment to the 2D |
---|
458 | // equivalent to binMatrix[binIndex][PSFIndex] |
---|
459 | |
---|
460 | MemClear(indices, sizeof(indices)); // Must be 0 for unused dimensions. |
---|
461 | indices[0] = binIndex; |
---|
462 | indices[1] = PSFIndex; |
---|
463 | if (retVal = MDGetNumericWavePointValue(wavH, indices, value)) |
---|
464 | return retVal; |
---|
465 | value[0] += 1; // Real part |
---|
466 | if (retVal = MDSetNumericWavePointValue(wavH, indices, value)) |
---|
467 | return retVal; |
---|
468 | |
---|
469 | } |
---|
470 | |
---|
471 | } |
---|
472 | } |
---|
473 | |
---|
474 | p->result = 0; |
---|
475 | |
---|
476 | return 0; |
---|
477 | |
---|
478 | } |
---|
479 | |
---|
480 | |
---|
481 | ///// this is directly from Numerical Recipes |
---|
482 | // -- I did change the float to double, since Igor treats all as double |
---|
483 | // and n is an int, not a pointer (seemed unnecessary) |
---|
484 | // |
---|
485 | #define MAXBIT 30 |
---|
486 | #define MAXDIM 6 |
---|
487 | static int iminarg1,iminarg2; |
---|
488 | #define IMIN(a,b) (iminarg1=(a),iminarg2=(b),(iminarg1) < (iminarg2) ? (iminarg1) : (iminarg2)) |
---|
489 | |
---|
490 | int |
---|
491 | SobolX(SobolParamPtr p) |
---|
492 | { |
---|
493 | int j,k,l; |
---|
494 | unsigned long i,im,ipp; |
---|
495 | static double fac; |
---|
496 | static unsigned long in,ix[MAXDIM+1],*iu[MAXBIT+1]; |
---|
497 | static unsigned long mdeg[MAXDIM+1]={0,1,2,3,3,4,4}; |
---|
498 | static unsigned long ip[MAXDIM+1]={0,0,1,1,2,1,4}; |
---|
499 | static unsigned long iv[MAXDIM*MAXBIT+1]={ |
---|
500 | 0,1,1,1,1,1,1,3,1,3,3,1,1,5,7,7,3,3,5,15,11,5,15,13,9}; |
---|
501 | |
---|
502 | static int initDone=0; |
---|
503 | char buf[256]; |
---|
504 | |
---|
505 | int n=0; |
---|
506 | double *x; //output x vector |
---|
507 | |
---|
508 | // check for all of the required waves |
---|
509 | if (p->bwavH == NIL) { |
---|
510 | SetNaN64(&p->result); |
---|
511 | return NON_EXISTENT_WAVE; |
---|
512 | } |
---|
513 | |
---|
514 | //check to see that all are double |
---|
515 | if(WaveType(p->bwavH) != NT_FP64 ) { |
---|
516 | SetNaN64(&p->result); |
---|
517 | return kExpectedNT_FP64; |
---|
518 | } |
---|
519 | x = (double*)WaveData(p->bwavH); |
---|
520 | n = (int)(p->nIn); // not sure that the negative input will be properly cast to int |
---|
521 | |
---|
522 | // sprintf(buf, "input, recast n = %g %d\r",p->nIn, n); |
---|
523 | // XOPNotice(buf); |
---|
524 | |
---|
525 | if (n < 0) { |
---|
526 | |
---|
527 | if(initDone) { |
---|
528 | sprintf(buf, "Don't re-initialize\r"); |
---|
529 | XOPNotice(buf); |
---|
530 | return 0; |
---|
531 | } |
---|
532 | |
---|
533 | for (j=1,k=0;j<=MAXBIT;j++,k+=MAXDIM) iu[j] = &iv[k]; |
---|
534 | for (k=1;k<=MAXDIM;k++) { |
---|
535 | for (j=1;j<=mdeg[k];j++) iu[j][k] <<= (MAXBIT-j); |
---|
536 | for (j=mdeg[k]+1;j<=MAXBIT;j++) { |
---|
537 | ipp=ip[k]; |
---|
538 | i=iu[j-mdeg[k]][k]; |
---|
539 | i ^= (i >> mdeg[k]); |
---|
540 | for (l=mdeg[k]-1;l>=1;l--) { |
---|
541 | if (ipp & 1) i ^= iu[j-l][k]; |
---|
542 | ipp >>= 1; |
---|
543 | } |
---|
544 | iu[j][k]=i; |
---|
545 | } |
---|
546 | } |
---|
547 | fac=1.0/(1L << MAXBIT); |
---|
548 | in=0; |
---|
549 | |
---|
550 | initDone=1; |
---|
551 | |
---|
552 | sprintf(buf, "Initialization loop done\r"); |
---|
553 | XOPNotice(buf); |
---|
554 | |
---|
555 | } else { |
---|
556 | im=in; |
---|
557 | for (j=1;j<=MAXBIT;j++) { |
---|
558 | if (!(im & 1)) break; |
---|
559 | im >>= 1; |
---|
560 | } |
---|
561 | if (j > MAXBIT) { |
---|
562 | sprintf(buf, "MAXBIT too small in sobseq\r"); |
---|
563 | XOPNotice(buf); |
---|
564 | } |
---|
565 | im=(j-1)*MAXDIM; |
---|
566 | for (k=1;k<=IMIN(n,MAXDIM);k++) { |
---|
567 | ix[k] ^= iv[im+k]; |
---|
568 | x[k-1]=ix[k]*fac; /// this is a real array to send back, count this one from zero |
---|
569 | //sprintf(buf, "calculate x[%d] = %g\r",k,ix[k]*fac); |
---|
570 | //XOPNotice(buf); |
---|
571 | } |
---|
572 | in++; |
---|
573 | } |
---|
574 | |
---|
575 | // sprintf(buf, "x[0],x[1] = %g %g\r",x[0],x[1]); |
---|
576 | // XOPNotice(buf); |
---|
577 | |
---|
578 | |
---|
579 | p->result = 0; |
---|
580 | |
---|
581 | return 0; |
---|
582 | } |
---|
583 | |
---|
584 | #undef MAXBIT |
---|
585 | #undef MAXDIM |
---|
586 | |
---|
587 | |
---|
588 | //#pragma XOP_RESET_STRUCT_PACKING // All structures are 2-byte-aligned. |
---|