1 | /* |
---|
2 | ########################################################## |
---|
3 | # # |
---|
4 | # !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! # |
---|
5 | # !! !! # |
---|
6 | # !! KEEP THIS CODE CONSISTENT WITH KERNELPY.PY !! # |
---|
7 | # !! !! # |
---|
8 | # !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! # |
---|
9 | # # |
---|
10 | ########################################################## |
---|
11 | */ |
---|
12 | |
---|
13 | // NOTE: the following macros are defined in generate.py: |
---|
14 | // |
---|
15 | // MAX_PD : the maximum number of dispersity loops allowed |
---|
16 | // NUM_PARS : the number of parameters in the parameter table |
---|
17 | // NUM_VALUES : the number of values to skip at the start of the |
---|
18 | // values array before you get to the dispersity values. |
---|
19 | // PARAMETER_TABLE : list of parameter declarations used to create the |
---|
20 | // ParameterTable type. |
---|
21 | // KERNEL_NAME : model_Iq, model_Iqxy or model_Imagnetic. This code is |
---|
22 | // included three times, once for each kernel type. |
---|
23 | // MAGNETIC : defined when the magnetic kernel is being instantiated |
---|
24 | // NUM_MAGNETIC : the number of magnetic parameters |
---|
25 | // MAGNETIC_PARS : a comma-separated list of indices to the sld |
---|
26 | // parameters in the parameter table. |
---|
27 | // CALL_VOLUME(table) : call the form volume function |
---|
28 | // CALL_IQ(q, table) : call the Iq function for 1D calcs. |
---|
29 | // CALL_IQ_A(q, table) : call the Iq function with |q| for 2D data. |
---|
30 | // CALL_IQ_AC(qa, qc, table) : call the Iqxy function for symmetric shapes |
---|
31 | // CALL_IQ_ABC(qa, qc, table) : call the Iqxy function for asymmetric shapes |
---|
32 | // INVALID(table) : test if the current point is feesible to calculate. This |
---|
33 | // will be defined in the kernel definition file. |
---|
34 | // PROJECTION : equirectangular=1, sinusoidal=2 |
---|
35 | // see explore/jitter.py for definitions. |
---|
36 | |
---|
37 | #ifndef _PAR_BLOCK_ // protected block so we can include this code twice. |
---|
38 | #define _PAR_BLOCK_ |
---|
39 | |
---|
40 | typedef struct { |
---|
41 | #if MAX_PD > 0 |
---|
42 | int32_t pd_par[MAX_PD]; // id of the nth dispersity variable |
---|
43 | int32_t pd_length[MAX_PD]; // length of the nth dispersity weight vector |
---|
44 | int32_t pd_offset[MAX_PD]; // offset of pd weights in the value & weight vector |
---|
45 | int32_t pd_stride[MAX_PD]; // stride to move to the next index at this level |
---|
46 | #endif // MAX_PD > 0 |
---|
47 | int32_t num_eval; // total number of voxels in hypercube |
---|
48 | int32_t num_weights; // total length of the weights vector |
---|
49 | int32_t num_active; // number of non-trivial pd loops |
---|
50 | int32_t theta_par; // id of first orientation variable |
---|
51 | } ProblemDetails; |
---|
52 | |
---|
53 | // Intel HD 4000 needs private arrays to be a multiple of 4 long |
---|
54 | typedef struct { |
---|
55 | PARAMETER_TABLE |
---|
56 | } ParameterTable; |
---|
57 | typedef union { |
---|
58 | ParameterTable table; |
---|
59 | double vector[4*((NUM_PARS+3)/4)]; |
---|
60 | } ParameterBlock; |
---|
61 | #endif // _PAR_BLOCK_ |
---|
62 | |
---|
63 | #if defined(MAGNETIC) && NUM_MAGNETIC > 0 |
---|
64 | // ===== Helper functions for magnetism ===== |
---|
65 | |
---|
66 | // Return value restricted between low and high |
---|
67 | static double clip(double value, double low, double high) |
---|
68 | { |
---|
69 | return (value < low ? low : (value > high ? high : value)); |
---|
70 | } |
---|
71 | |
---|
72 | // Compute spin cross sections given in_spin and out_spin |
---|
73 | // To convert spin cross sections to sld b: |
---|
74 | // uu * (sld - m_sigma_x); |
---|
75 | // dd * (sld + m_sigma_x); |
---|
76 | // ud * (m_sigma_y - 1j*m_sigma_z); |
---|
77 | // du * (m_sigma_y + 1j*m_sigma_z); |
---|
78 | // weights for spin crosssections: dd du real, ud real, uu, du imag, ud imag |
---|
79 | static void set_spin_weights(double in_spin, double out_spin, double spins[4]) |
---|
80 | { |
---|
81 | in_spin = clip(in_spin, 0.0, 1.0); |
---|
82 | out_spin = clip(out_spin, 0.0, 1.0); |
---|
83 | spins[0] = sqrt(sqrt((1.0-in_spin) * (1.0-out_spin))); // dd |
---|
84 | spins[1] = sqrt(sqrt((1.0-in_spin) * out_spin)); // du real |
---|
85 | spins[2] = sqrt(sqrt(in_spin * (1.0-out_spin))); // ud real |
---|
86 | spins[3] = sqrt(sqrt(in_spin * out_spin)); // uu |
---|
87 | spins[4] = spins[1]; // du imag |
---|
88 | spins[5] = spins[2]; // ud imag |
---|
89 | } |
---|
90 | |
---|
91 | // Compute the magnetic sld |
---|
92 | static double mag_sld( |
---|
93 | const int xs, // 0=dd, 1=du real, 2=ud real, 3=uu, 4=du imag, 5=up imag |
---|
94 | const double qx, const double qy, |
---|
95 | const double px, const double py, |
---|
96 | const double sld, |
---|
97 | const double mx, const double my, const double mz |
---|
98 | ) |
---|
99 | { |
---|
100 | if (xs < 4) { |
---|
101 | const double perp = qy*mx - qx*my; |
---|
102 | switch (xs) { |
---|
103 | case 0: // uu => sld - D M_perpx |
---|
104 | return sld - px*perp; |
---|
105 | case 1: // ud real => -D M_perpy |
---|
106 | return py*perp; |
---|
107 | case 2: // du real => -D M_perpy |
---|
108 | return py*perp; |
---|
109 | case 3: // dd real => sld + D M_perpx |
---|
110 | return sld + px*perp; |
---|
111 | } |
---|
112 | } else { |
---|
113 | if (xs== 4) { |
---|
114 | return -mz; // ud imag => -D M_perpz |
---|
115 | } else { // index == 5 |
---|
116 | return mz; // du imag => D M_perpz |
---|
117 | } |
---|
118 | } |
---|
119 | } |
---|
120 | |
---|
121 | |
---|
122 | #endif |
---|
123 | |
---|
124 | // ===== Helper functions for orientation and jitter ===== |
---|
125 | |
---|
126 | // To change the definition of the angles, run explore/angles.py, which |
---|
127 | // uses sympy to generate the equations. |
---|
128 | |
---|
129 | #if !defined(_QAC_SECTION) && defined(CALL_IQ_AC) |
---|
130 | #define _QAC_SECTION |
---|
131 | |
---|
132 | typedef struct { |
---|
133 | double R31, R32; |
---|
134 | } QACRotation; |
---|
135 | |
---|
136 | // Fill in the rotation matrix R from the view angles (theta, phi) and the |
---|
137 | // jitter angles (dtheta, dphi). This matrix can be applied to all of the |
---|
138 | // (qx, qy) points in the image to produce R*[qx,qy]' = [qa,qc]' |
---|
139 | static void |
---|
140 | qac_rotation( |
---|
141 | QACRotation *rotation, |
---|
142 | double theta, double phi, |
---|
143 | double dtheta, double dphi) |
---|
144 | { |
---|
145 | double sin_theta, cos_theta; |
---|
146 | double sin_phi, cos_phi; |
---|
147 | |
---|
148 | // reverse view matrix |
---|
149 | SINCOS(theta*M_PI_180, sin_theta, cos_theta); |
---|
150 | SINCOS(phi*M_PI_180, sin_phi, cos_phi); |
---|
151 | const double V11 = cos_phi*cos_theta; |
---|
152 | const double V12 = sin_phi*cos_theta; |
---|
153 | const double V21 = -sin_phi; |
---|
154 | const double V22 = cos_phi; |
---|
155 | const double V31 = sin_theta*cos_phi; |
---|
156 | const double V32 = sin_phi*sin_theta; |
---|
157 | |
---|
158 | // reverse jitter matrix |
---|
159 | SINCOS(dtheta*M_PI_180, sin_theta, cos_theta); |
---|
160 | SINCOS(dphi*M_PI_180, sin_phi, cos_phi); |
---|
161 | const double J31 = sin_theta; |
---|
162 | const double J32 = -sin_phi*cos_theta; |
---|
163 | const double J33 = cos_phi*cos_theta; |
---|
164 | |
---|
165 | // reverse matrix |
---|
166 | rotation->R31 = J31*V11 + J32*V21 + J33*V31; |
---|
167 | rotation->R32 = J31*V12 + J32*V22 + J33*V32; |
---|
168 | } |
---|
169 | |
---|
170 | // Apply the rotation matrix returned from qac_rotation to the point (qx,qy), |
---|
171 | // returning R*[qx,qy]' = [qa,qc]' |
---|
172 | static double |
---|
173 | qac_apply( |
---|
174 | QACRotation rotation, |
---|
175 | double qx, double qy, |
---|
176 | double *qa_out, double *qc_out) |
---|
177 | { |
---|
178 | const double dqc = rotation.R31*qx + rotation.R32*qy; |
---|
179 | // Indirect calculation of qab, from qab^2 = |q|^2 - qc^2 |
---|
180 | const double dqa = sqrt(-dqc*dqc + qx*qx + qy*qy); |
---|
181 | |
---|
182 | *qa_out = dqa; |
---|
183 | *qc_out = dqc; |
---|
184 | } |
---|
185 | #endif // _QAC_SECTION |
---|
186 | |
---|
187 | #if !defined(_QABC_SECTION) && defined(CALL_IQ_ABC) |
---|
188 | #define _QABC_SECTION |
---|
189 | |
---|
190 | typedef struct { |
---|
191 | double R11, R12; |
---|
192 | double R21, R22; |
---|
193 | double R31, R32; |
---|
194 | } QABCRotation; |
---|
195 | |
---|
196 | // Fill in the rotation matrix R from the view angles (theta, phi, psi) and the |
---|
197 | // jitter angles (dtheta, dphi, dpsi). This matrix can be applied to all of the |
---|
198 | // (qx, qy) points in the image to produce R*[qx,qy]' = [qa,qb,qc]' |
---|
199 | static void |
---|
200 | qabc_rotation( |
---|
201 | QABCRotation *rotation, |
---|
202 | double theta, double phi, double psi, |
---|
203 | double dtheta, double dphi, double dpsi) |
---|
204 | { |
---|
205 | double sin_theta, cos_theta; |
---|
206 | double sin_phi, cos_phi; |
---|
207 | double sin_psi, cos_psi; |
---|
208 | |
---|
209 | // reverse view matrix |
---|
210 | SINCOS(theta*M_PI_180, sin_theta, cos_theta); |
---|
211 | SINCOS(phi*M_PI_180, sin_phi, cos_phi); |
---|
212 | SINCOS(psi*M_PI_180, sin_psi, cos_psi); |
---|
213 | const double V11 = -sin_phi*sin_psi + cos_phi*cos_psi*cos_theta; |
---|
214 | const double V12 = sin_phi*cos_psi*cos_theta + sin_psi*cos_phi; |
---|
215 | const double V21 = -sin_phi*cos_psi - sin_psi*cos_phi*cos_theta; |
---|
216 | const double V22 = -sin_phi*sin_psi*cos_theta + cos_phi*cos_psi; |
---|
217 | const double V31 = sin_theta*cos_phi; |
---|
218 | const double V32 = sin_phi*sin_theta; |
---|
219 | |
---|
220 | // reverse jitter matrix |
---|
221 | SINCOS(dtheta*M_PI_180, sin_theta, cos_theta); |
---|
222 | SINCOS(dphi*M_PI_180, sin_phi, cos_phi); |
---|
223 | SINCOS(dpsi*M_PI_180, sin_psi, cos_psi); |
---|
224 | const double J11 = cos_psi*cos_theta; |
---|
225 | const double J12 = sin_phi*sin_theta*cos_psi + sin_psi*cos_phi; |
---|
226 | const double J13 = sin_phi*sin_psi - sin_theta*cos_phi*cos_psi; |
---|
227 | const double J21 = -sin_psi*cos_theta; |
---|
228 | const double J22 = -sin_phi*sin_psi*sin_theta + cos_phi*cos_psi; |
---|
229 | const double J23 = sin_phi*cos_psi + sin_psi*sin_theta*cos_phi; |
---|
230 | const double J31 = sin_theta; |
---|
231 | const double J32 = -sin_phi*cos_theta; |
---|
232 | const double J33 = cos_phi*cos_theta; |
---|
233 | |
---|
234 | // reverse matrix |
---|
235 | rotation->R11 = J11*V11 + J12*V21 + J13*V31; |
---|
236 | rotation->R12 = J11*V12 + J12*V22 + J13*V32; |
---|
237 | rotation->R21 = J21*V11 + J22*V21 + J23*V31; |
---|
238 | rotation->R22 = J21*V12 + J22*V22 + J23*V32; |
---|
239 | rotation->R31 = J31*V11 + J32*V21 + J33*V31; |
---|
240 | rotation->R32 = J31*V12 + J32*V22 + J33*V32; |
---|
241 | } |
---|
242 | |
---|
243 | // Apply the rotation matrix returned from qabc_rotation to the point (qx,qy), |
---|
244 | // returning R*[qx,qy]' = [qa,qb,qc]' |
---|
245 | static double |
---|
246 | qabc_apply( |
---|
247 | QABCRotation rotation, |
---|
248 | double qx, double qy, |
---|
249 | double *qa_out, double *qb_out, double *qc_out) |
---|
250 | { |
---|
251 | *qa_out = rotation.R11*qx + rotation.R12*qy; |
---|
252 | *qb_out = rotation.R21*qx + rotation.R22*qy; |
---|
253 | *qc_out = rotation.R31*qx + rotation.R32*qy; |
---|
254 | } |
---|
255 | |
---|
256 | #endif // _QABC_SECTION |
---|
257 | |
---|
258 | |
---|
259 | // ==================== KERNEL CODE ======================== |
---|
260 | |
---|
261 | kernel |
---|
262 | void KERNEL_NAME( |
---|
263 | int32_t nq, // number of q values |
---|
264 | const int32_t pd_start, // where we are in the dispersity loop |
---|
265 | const int32_t pd_stop, // where we are stopping in the dispersity loop |
---|
266 | global const ProblemDetails *details, |
---|
267 | global const double *values, |
---|
268 | global const double *q, // nq q values, with padding to boundary |
---|
269 | global double *result, // nq+1 return values, again with padding |
---|
270 | const double cutoff // cutoff in the dispersity weight product |
---|
271 | ) |
---|
272 | { |
---|
273 | #ifdef USE_OPENCL |
---|
274 | // who we are and what element we are working with |
---|
275 | const int q_index = get_global_id(0); |
---|
276 | if (q_index >= nq) return; |
---|
277 | #else |
---|
278 | // Define q_index here so that debugging statements can be written to work |
---|
279 | // for both OpenCL and DLL using: |
---|
280 | // if (q_index == 0) {printf(...);} |
---|
281 | int q_index = 0; |
---|
282 | #endif |
---|
283 | |
---|
284 | // Storage for the current parameter values. These will be updated as we |
---|
285 | // walk the dispersity mesh. |
---|
286 | ParameterBlock local_values; |
---|
287 | |
---|
288 | #if defined(MAGNETIC) && NUM_MAGNETIC>0 |
---|
289 | // Location of the sld parameters in the parameter vector. |
---|
290 | // These parameters are updated with the effective sld due to magnetism. |
---|
291 | const int32_t slds[] = { MAGNETIC_PARS }; |
---|
292 | |
---|
293 | // Interpret polarization cross section. |
---|
294 | // up_frac_i = values[NUM_PARS+2]; |
---|
295 | // up_frac_f = values[NUM_PARS+3]; |
---|
296 | // up_angle = values[NUM_PARS+4]; |
---|
297 | // TODO: could precompute more magnetism parameters before calling the kernel. |
---|
298 | double spins[8]; // uu, ud real, du real, dd, ud imag, du imag, fill, fill |
---|
299 | double cos_mspin, sin_mspin; |
---|
300 | set_spin_weights(values[NUM_PARS+2], values[NUM_PARS+3], spins); |
---|
301 | SINCOS(-values[NUM_PARS+4]*M_PI_180, sin_mspin, cos_mspin); |
---|
302 | #endif // MAGNETIC |
---|
303 | |
---|
304 | // Fill in the initial variables |
---|
305 | // values[0] is scale |
---|
306 | // values[1] is background |
---|
307 | #ifdef USE_OPENMP |
---|
308 | #pragma omp parallel for |
---|
309 | #endif |
---|
310 | for (int i=0; i < NUM_PARS; i++) { |
---|
311 | local_values.vector[i] = values[2+i]; |
---|
312 | //if (q_index==0) printf("p%d = %g\n",i, local_values.vector[i]); |
---|
313 | } |
---|
314 | //if (q_index==0) printf("NUM_VALUES:%d NUM_PARS:%d MAX_PD:%d\n", NUM_VALUES, NUM_PARS, MAX_PD); |
---|
315 | //if (q_index==0) printf("start:%d stop:%d\n", pd_start, pd_stop); |
---|
316 | |
---|
317 | // If pd_start is zero that means that we are starting a new calculation, |
---|
318 | // and must initialize the result to zero. Otherwise, we are restarting |
---|
319 | // the calculation from somewhere in the middle of the dispersity mesh, |
---|
320 | // and we update the value rather than reset it. Similarly for the |
---|
321 | // normalization factor, which is stored as the final value in the |
---|
322 | // results vector (one past the number of q values). |
---|
323 | // |
---|
324 | // The code differs slightly between opencl and dll since opencl is only |
---|
325 | // seeing one q value (stored in the variable "this_result") while the dll |
---|
326 | // version must loop over all q. |
---|
327 | #ifdef USE_OPENCL |
---|
328 | double pd_norm = (pd_start == 0 ? 0.0 : result[nq]); |
---|
329 | double this_result = (pd_start == 0 ? 0.0 : result[q_index]); |
---|
330 | #else // !USE_OPENCL |
---|
331 | double pd_norm = (pd_start == 0 ? 0.0 : result[nq]); |
---|
332 | if (pd_start == 0) { |
---|
333 | #ifdef USE_OPENMP |
---|
334 | #pragma omp parallel for |
---|
335 | #endif |
---|
336 | for (int q_index=0; q_index < nq; q_index++) result[q_index] = 0.0; |
---|
337 | } |
---|
338 | //if (q_index==0) printf("start %d %g %g\n", pd_start, pd_norm, result[0]); |
---|
339 | #endif // !USE_OPENCL |
---|
340 | |
---|
341 | |
---|
342 | // ====== macros to set up the parts of the loop ======= |
---|
343 | /* |
---|
344 | Based on the level of the loop, uses C preprocessor magic to construct |
---|
345 | level-specific looping variables, including these from loop level 3: |
---|
346 | |
---|
347 | int n3 : length of loop for mesh level 3 |
---|
348 | int i3 : current position in the loop for level 3, which is calculated |
---|
349 | from a combination of pd_start, pd_stride[3] and pd_length[3]. |
---|
350 | int p3 : is the index into the parameter table for mesh level 3 |
---|
351 | double v3[] : pointer into dispersity array to values for loop 3 |
---|
352 | double w3[] : pointer into dispersity array to weights for loop 3 |
---|
353 | double weight3 : the product of weights from levels 3 and up, computed |
---|
354 | as weight5*weight4*w3[i3]. Note that we need an outermost |
---|
355 | value weight5 set to 1.0 for this to work properly. |
---|
356 | |
---|
357 | After expansion, the loop struction will look like the following: |
---|
358 | |
---|
359 | // --- PD_INIT(4) --- |
---|
360 | const int n4 = pd_length[4]; |
---|
361 | const int p4 = pd_par[4]; |
---|
362 | global const double *v4 = pd_value + pd_offset[4]; |
---|
363 | global const double *w4 = pd_weight + pd_offset[4]; |
---|
364 | int i4 = (pd_start/pd_stride[4])%n4; // position in level 4 at pd_start |
---|
365 | |
---|
366 | // --- PD_INIT(3) --- |
---|
367 | const int n3 = pd_length[3]; |
---|
368 | ... |
---|
369 | int i3 = (pd_start/pd_stride[3])%n3; // position in level 3 at pd_start |
---|
370 | |
---|
371 | PD_INIT(2) |
---|
372 | PD_INIT(1) |
---|
373 | PD_INIT(0) |
---|
374 | |
---|
375 | // --- PD_OUTERMOST_WEIGHT(5) --- |
---|
376 | const double weight5 = 1.0; |
---|
377 | |
---|
378 | // --- PD_OPEN(4,5) --- |
---|
379 | while (i4 < n4) { |
---|
380 | parameter[p4] = v4[i4]; // set the value for pd parameter 4 at this mesh point |
---|
381 | const double weight4 = w4[i4] * weight5; |
---|
382 | |
---|
383 | // from PD_OPEN(3,4) |
---|
384 | while (i3 < n3) { |
---|
385 | parameter[p3] = v3[i3]; // set the value for pd parameter 3 at this mesh point |
---|
386 | const double weight3 = w3[i3] * weight4; |
---|
387 | |
---|
388 | PD_OPEN(3,2) |
---|
389 | PD_OPEN(2,1) |
---|
390 | PD_OPEN(0,1) |
---|
391 | |
---|
392 | ... main loop body ... |
---|
393 | |
---|
394 | ++step; // increment counter representing position in dispersity mesh |
---|
395 | |
---|
396 | PD_CLOSE(0) |
---|
397 | PD_CLOSE(1) |
---|
398 | PD_CLOSE(2) |
---|
399 | |
---|
400 | // --- PD_CLOSE(3) --- |
---|
401 | if (step >= pd_stop) break; |
---|
402 | ++i3; |
---|
403 | } |
---|
404 | i3 = 0; // reset loop counter for next round through the loop |
---|
405 | |
---|
406 | // --- PD_CLOSE(4) --- |
---|
407 | if (step >= pd_stop) break; |
---|
408 | ++i4; |
---|
409 | } |
---|
410 | i4 = 0; // reset loop counter even though no more rounds through the loop |
---|
411 | |
---|
412 | */ |
---|
413 | |
---|
414 | // Define looping variables |
---|
415 | #define PD_INIT(_LOOP) \ |
---|
416 | const int n##_LOOP = details->pd_length[_LOOP]; \ |
---|
417 | const int p##_LOOP = details->pd_par[_LOOP]; \ |
---|
418 | global const double *v##_LOOP = pd_value + details->pd_offset[_LOOP]; \ |
---|
419 | global const double *w##_LOOP = pd_weight + details->pd_offset[_LOOP]; \ |
---|
420 | int i##_LOOP = (pd_start/details->pd_stride[_LOOP])%n##_LOOP; |
---|
421 | |
---|
422 | // Jump into the middle of the dispersity loop |
---|
423 | #define PD_OPEN(_LOOP,_OUTER) \ |
---|
424 | while (i##_LOOP < n##_LOOP) { \ |
---|
425 | local_values.vector[p##_LOOP] = v##_LOOP[i##_LOOP]; \ |
---|
426 | const double weight##_LOOP = w##_LOOP[i##_LOOP] * weight##_OUTER; |
---|
427 | |
---|
428 | // create the variable "weight#=1.0" where # is the outermost level+1 (=MAX_PD). |
---|
429 | #define _PD_OUTERMOST_WEIGHT(_n) const double weight##_n = 1.0; |
---|
430 | #define PD_OUTERMOST_WEIGHT(_n) _PD_OUTERMOST_WEIGHT(_n) |
---|
431 | |
---|
432 | // Close out the loop |
---|
433 | #define PD_CLOSE(_LOOP) \ |
---|
434 | if (step >= pd_stop) break; \ |
---|
435 | ++i##_LOOP; \ |
---|
436 | } \ |
---|
437 | i##_LOOP = 0; |
---|
438 | |
---|
439 | // The main loop body is essentially: |
---|
440 | // |
---|
441 | // BUILD_ROTATION // construct the rotation matrix qxy => qabc |
---|
442 | // for each q |
---|
443 | // FETCH_Q // set qx,qy from the q input vector |
---|
444 | // APPLY_ROTATION // convert qx,qy to qa,qb,qc |
---|
445 | // CALL_KERNEL // scattering = Iqxy(qa, qb, qc, p1, p2, ...) |
---|
446 | // |
---|
447 | // Depending on the shape type (radial, axial, triaxial), the variables |
---|
448 | // and calling parameters will be slightly different. These macros |
---|
449 | // capture the differences in one spot so the rest of the code is easier |
---|
450 | // to read. |
---|
451 | |
---|
452 | #if defined(CALL_IQ) |
---|
453 | // unoriented 1D |
---|
454 | double qk; |
---|
455 | #define FETCH_Q() do { qk = q[q_index]; } while (0) |
---|
456 | #define BUILD_ROTATION() do {} while(0) |
---|
457 | #define APPLY_ROTATION() do {} while(0) |
---|
458 | #define CALL_KERNEL() CALL_IQ(qk, local_values.table) |
---|
459 | |
---|
460 | #elif defined(CALL_IQ_A) |
---|
461 | // unoriented 2D |
---|
462 | double qx, qy; |
---|
463 | #define FETCH_Q() do { qx = q[2*q_index]; qy = q[2*q_index+1]; } while (0) |
---|
464 | #define BUILD_ROTATION() do {} while(0) |
---|
465 | #define APPLY_ROTATION() do {} while(0) |
---|
466 | #define CALL_KERNEL() CALL_IQ_A(sqrt(qx*qx+qy*qy), local_values.table) |
---|
467 | |
---|
468 | #elif defined(CALL_IQ_AC) |
---|
469 | // oriented symmetric 2D |
---|
470 | double qx, qy; |
---|
471 | #define FETCH_Q() do { qx = q[2*q_index]; qy = q[2*q_index+1]; } while (0) |
---|
472 | double qa, qc; |
---|
473 | QACRotation rotation; |
---|
474 | // theta, phi, dtheta, dphi are defined below in projection to avoid repeated code. |
---|
475 | #define BUILD_ROTATION() qac_rotation(&rotation, theta, phi, dtheta, dphi); |
---|
476 | #define APPLY_ROTATION() qac_apply(rotation, qx, qy, &qa, &qc) |
---|
477 | #define CALL_KERNEL() CALL_IQ_AC(qa, qc, local_values.table) |
---|
478 | |
---|
479 | #elif defined(CALL_IQ_ABC) |
---|
480 | // oriented asymmetric 2D |
---|
481 | double qx, qy; |
---|
482 | #define FETCH_Q() do { qx = q[2*q_index]; qy = q[2*q_index+1]; } while (0) |
---|
483 | double qa, qb, qc; |
---|
484 | QABCRotation rotation; |
---|
485 | // theta, phi, dtheta, dphi are defined below in projection to avoid repeated code. |
---|
486 | // psi and dpsi are only for IQ_ABC, so they are processed here. |
---|
487 | const double psi = values[details->theta_par+4]; |
---|
488 | #define BUILD_ROTATION() qabc_rotation(&rotation, theta, phi, psi, dtheta, dphi, local_values.table.psi) |
---|
489 | #define APPLY_ROTATION() qabc_apply(rotation, qx, qy, &qa, &qb, &qc) |
---|
490 | #define CALL_KERNEL() CALL_IQ_ABC(qa, qb, qc, local_values.table) |
---|
491 | #endif |
---|
492 | |
---|
493 | // Doing jitter projection code outside the previous if block so that we don't |
---|
494 | // need to repeat the identical logic in the IQ_AC and IQ_ABC branches. This |
---|
495 | // will become more important if we implement more projections, or more |
---|
496 | // complicated projections. |
---|
497 | #if defined(CALL_IQ) || defined(CALL_IQ_A) |
---|
498 | #define APPLY_PROJECTION() const double weight=weight0 |
---|
499 | #else // !spherosymmetric projection |
---|
500 | // Grab the "view" angles (theta, phi, psi) from the initial parameter table. |
---|
501 | const double theta = values[details->theta_par+2]; |
---|
502 | const double phi = values[details->theta_par+3]; |
---|
503 | // The "jitter" angles (dtheta, dphi, dpsi) are stored with the |
---|
504 | // dispersity values and copied to the local parameter table as |
---|
505 | // we go through the mesh. |
---|
506 | double dtheta, dphi, weight; |
---|
507 | #if PROJECTION == 1 |
---|
508 | #define APPLY_PROJECTION() do { \ |
---|
509 | dtheta = local_values.table.theta; \ |
---|
510 | dphi = local_values.table.phi; \ |
---|
511 | weight = fabs(cos(dtheta*M_PI_180)) * weight0; \ |
---|
512 | } while (0) |
---|
513 | #elif PROJECTION == 2 |
---|
514 | #define APPLY_PROJECTION() do { \ |
---|
515 | dtheta = local_values.table.theta; \ |
---|
516 | dphi = local_values.table.phi; \ |
---|
517 | weight = weight0; \ |
---|
518 | if (dtheta != 90.0) dphi /= cos(dtheta*M_PI_180); \ |
---|
519 | else if (dphi != 0.0) weight = 0.; \ |
---|
520 | if (fabs(dphi) >= 180.) weight = 0.; \ |
---|
521 | } while (0) |
---|
522 | #endif |
---|
523 | #endif // !spherosymmetric projection |
---|
524 | |
---|
525 | |
---|
526 | // ====== construct the loops ======= |
---|
527 | |
---|
528 | // Pointers to the start of the dispersity and weight vectors, if needed. |
---|
529 | #if MAX_PD>0 |
---|
530 | global const double *pd_value = values + NUM_VALUES; |
---|
531 | global const double *pd_weight = pd_value + details->num_weights; |
---|
532 | #endif |
---|
533 | |
---|
534 | // The variable "step" is the current position in the dispersity loop. |
---|
535 | // It will be incremented each time a new point in the mesh is accumulated, |
---|
536 | // and used to test whether we have reached pd_stop. |
---|
537 | int step = pd_start; |
---|
538 | |
---|
539 | // define looping variables |
---|
540 | #if MAX_PD>4 |
---|
541 | PD_INIT(4) |
---|
542 | #endif |
---|
543 | #if MAX_PD>3 |
---|
544 | PD_INIT(3) |
---|
545 | #endif |
---|
546 | #if MAX_PD>2 |
---|
547 | PD_INIT(2) |
---|
548 | #endif |
---|
549 | #if MAX_PD>1 |
---|
550 | PD_INIT(1) |
---|
551 | #endif |
---|
552 | #if MAX_PD>0 |
---|
553 | PD_INIT(0) |
---|
554 | #endif |
---|
555 | |
---|
556 | // open nested loops |
---|
557 | PD_OUTERMOST_WEIGHT(MAX_PD) |
---|
558 | #if MAX_PD>4 |
---|
559 | PD_OPEN(4,5) |
---|
560 | #endif |
---|
561 | #if MAX_PD>3 |
---|
562 | PD_OPEN(3,4) |
---|
563 | #endif |
---|
564 | #if MAX_PD>2 |
---|
565 | PD_OPEN(2,3) |
---|
566 | #endif |
---|
567 | #if MAX_PD>1 |
---|
568 | PD_OPEN(1,2) |
---|
569 | #endif |
---|
570 | #if MAX_PD>0 |
---|
571 | PD_OPEN(0,1) |
---|
572 | #endif |
---|
573 | |
---|
574 | |
---|
575 | //if (q_index==0) {printf("step:%d of %d, pars:",step,pd_stop); for (int i=0; i < NUM_PARS; i++) printf("p%d=%g ",i, local_values.vector[i]); printf("\n");} |
---|
576 | |
---|
577 | // ====== loop body ======= |
---|
578 | #ifdef INVALID |
---|
579 | if (!INVALID(local_values.table)) |
---|
580 | #endif |
---|
581 | { |
---|
582 | APPLY_PROJECTION(); |
---|
583 | |
---|
584 | // Accumulate I(q) |
---|
585 | // Note: weight==0 must always be excluded |
---|
586 | if (weight > cutoff) { |
---|
587 | pd_norm += weight * CALL_VOLUME(local_values.table); |
---|
588 | BUILD_ROTATION(); |
---|
589 | |
---|
590 | #ifndef USE_OPENCL |
---|
591 | // DLL needs to explicitly loop over the q values. |
---|
592 | #ifdef USE_OPENMP |
---|
593 | #pragma omp parallel for |
---|
594 | #endif |
---|
595 | for (q_index=0; q_index<nq; q_index++) |
---|
596 | #endif // !USE_OPENCL |
---|
597 | { |
---|
598 | |
---|
599 | FETCH_Q(); |
---|
600 | APPLY_ROTATION(); |
---|
601 | |
---|
602 | // ======= COMPUTE SCATTERING ========== |
---|
603 | #if defined(MAGNETIC) && NUM_MAGNETIC > 0 |
---|
604 | // Compute the scattering from the magnetic cross sections. |
---|
605 | double scattering = 0.0; |
---|
606 | const double qsq = qx*qx + qy*qy; |
---|
607 | if (qsq > 1.e-16) { |
---|
608 | // TODO: what is the magnetic scattering at q=0 |
---|
609 | const double px = (qy*cos_mspin + qx*sin_mspin)/qsq; |
---|
610 | const double py = (qy*sin_mspin - qx*cos_mspin)/qsq; |
---|
611 | |
---|
612 | // loop over uu, ud real, du real, dd, ud imag, du imag |
---|
613 | for (int xs=0; xs<6; xs++) { |
---|
614 | const double xs_weight = spins[xs]; |
---|
615 | if (xs_weight > 1.e-8) { |
---|
616 | // Since the cross section weight is significant, set the slds |
---|
617 | // to the effective slds for this cross section, call the |
---|
618 | // kernel, and add according to weight. |
---|
619 | for (int sk=0; sk<NUM_MAGNETIC; sk++) { |
---|
620 | const int32_t mag_index = NUM_PARS+5 + 3*sk; |
---|
621 | const int32_t sld_index = slds[sk]; |
---|
622 | const double mx = values[mag_index]; |
---|
623 | const double my = values[mag_index+1]; |
---|
624 | const double mz = values[mag_index+2]; |
---|
625 | local_values.vector[sld_index] = |
---|
626 | mag_sld(xs, qx, qy, px, py, values[sld_index+2], mx, my, mz); |
---|
627 | } |
---|
628 | scattering += xs_weight * CALL_KERNEL(); |
---|
629 | } |
---|
630 | } |
---|
631 | } |
---|
632 | #else // !MAGNETIC |
---|
633 | const double scattering = CALL_KERNEL(); |
---|
634 | #endif // !MAGNETIC |
---|
635 | //printf("q_index:%d %g %g %g %g\n", q_index, scattering, weight0); |
---|
636 | |
---|
637 | #ifdef USE_OPENCL |
---|
638 | this_result += weight * scattering; |
---|
639 | #else // !USE_OPENCL |
---|
640 | result[q_index] += weight * scattering; |
---|
641 | #endif // !USE_OPENCL |
---|
642 | } |
---|
643 | } |
---|
644 | } |
---|
645 | |
---|
646 | // close nested loops |
---|
647 | ++step; |
---|
648 | #if MAX_PD>0 |
---|
649 | PD_CLOSE(0) |
---|
650 | #endif |
---|
651 | #if MAX_PD>1 |
---|
652 | PD_CLOSE(1) |
---|
653 | #endif |
---|
654 | #if MAX_PD>2 |
---|
655 | PD_CLOSE(2) |
---|
656 | #endif |
---|
657 | #if MAX_PD>3 |
---|
658 | PD_CLOSE(3) |
---|
659 | #endif |
---|
660 | #if MAX_PD>4 |
---|
661 | PD_CLOSE(4) |
---|
662 | #endif |
---|
663 | |
---|
664 | // Remember the current result and the updated norm. |
---|
665 | #ifdef USE_OPENCL |
---|
666 | result[q_index] = this_result; |
---|
667 | if (q_index == 0) result[nq] = pd_norm; |
---|
668 | //if (q_index == 0) printf("res: %g/%g\n", result[0], pd_norm); |
---|
669 | #else // !USE_OPENCL |
---|
670 | result[nq] = pd_norm; |
---|
671 | //printf("res: %g/%g\n", result[0], pd_norm); |
---|
672 | #endif // !USE_OPENCL |
---|
673 | |
---|
674 | // clear the macros in preparation for the next kernel |
---|
675 | #undef PD_INIT |
---|
676 | #undef PD_OPEN |
---|
677 | #undef PD_CLOSE |
---|
678 | #undef FETCH_Q |
---|
679 | #undef APPLY_PROJECTION |
---|
680 | #undef BUILD_ROTATION |
---|
681 | #undef APPLY_ROTATION |
---|
682 | #undef CALL_KERNEL |
---|
683 | } |
---|