1 | |
---|
2 | /* |
---|
3 | ########################################################## |
---|
4 | # # |
---|
5 | # !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! # |
---|
6 | # !! !! # |
---|
7 | # !! KEEP THIS CODE CONSISTENT WITH KERNELPY.PY !! # |
---|
8 | # !! !! # |
---|
9 | # !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! # |
---|
10 | # # |
---|
11 | ########################################################## |
---|
12 | */ |
---|
13 | |
---|
14 | #ifndef _PAR_BLOCK_ // protected block so we can include this code twice. |
---|
15 | #define _PAR_BLOCK_ |
---|
16 | |
---|
17 | typedef struct { |
---|
18 | #if MAX_PD > 0 |
---|
19 | int32_t pd_par[MAX_PD]; // id of the nth polydispersity variable |
---|
20 | int32_t pd_length[MAX_PD]; // length of the nth polydispersity weight vector |
---|
21 | int32_t pd_offset[MAX_PD]; // offset of pd weights in the value & weight vector |
---|
22 | int32_t pd_stride[MAX_PD]; // stride to move to the next index at this level |
---|
23 | #endif // MAX_PD > 0 |
---|
24 | int32_t num_eval; // total number of voxels in hypercube |
---|
25 | int32_t num_weights; // total length of the weights vector |
---|
26 | int32_t num_active; // number of non-trivial pd loops |
---|
27 | int32_t theta_par; // id of spherical correction variable |
---|
28 | } ProblemDetails; |
---|
29 | |
---|
30 | // Intel HD 4000 needs private arrays to be a multiple of 4 long |
---|
31 | typedef struct { |
---|
32 | PARAMETER_TABLE |
---|
33 | } ParameterTable; |
---|
34 | typedef union { |
---|
35 | ParameterTable table; |
---|
36 | double vector[4*((NUM_PARS+3)/4)]; |
---|
37 | } ParameterBlock; |
---|
38 | #endif // _PAR_BLOCK_ |
---|
39 | |
---|
40 | |
---|
41 | #if defined(MAGNETIC) && NUM_MAGNETIC>0 |
---|
42 | |
---|
43 | // Return value restricted between low and high |
---|
44 | static double clip(double value, double low, double high) |
---|
45 | { |
---|
46 | return (value < low ? low : (value > high ? high : value)); |
---|
47 | } |
---|
48 | |
---|
49 | // Compute spin cross sections given in_spin and out_spin |
---|
50 | // To convert spin cross sections to sld b: |
---|
51 | // uu * (sld - m_sigma_x); |
---|
52 | // dd * (sld + m_sigma_x); |
---|
53 | // ud * (m_sigma_y + 1j*m_sigma_z); |
---|
54 | // du * (m_sigma_y - 1j*m_sigma_z); |
---|
55 | static void set_spins(double in_spin, double out_spin, double spins[4]) |
---|
56 | { |
---|
57 | in_spin = clip(in_spin, 0.0, 1.0); |
---|
58 | out_spin = clip(out_spin, 0.0, 1.0); |
---|
59 | spins[0] = sqrt(sqrt((1.0-in_spin) * (1.0-out_spin))); // dd |
---|
60 | spins[1] = sqrt(sqrt((1.0-in_spin) * out_spin)); // du |
---|
61 | spins[2] = sqrt(sqrt(in_spin * (1.0-out_spin))); // ud |
---|
62 | spins[3] = sqrt(sqrt(in_spin * out_spin)); // uu |
---|
63 | } |
---|
64 | |
---|
65 | static double mag_sld(double qx, double qy, double p, |
---|
66 | double mx, double my, double sld) |
---|
67 | { |
---|
68 | const double perp = qy*mx - qx*my; |
---|
69 | return sld + perp*p; |
---|
70 | } |
---|
71 | |
---|
72 | #endif // MAGNETIC |
---|
73 | |
---|
74 | kernel |
---|
75 | void KERNEL_NAME( |
---|
76 | int32_t nq, // number of q values |
---|
77 | const int32_t pd_start, // where we are in the polydispersity loop |
---|
78 | const int32_t pd_stop, // where we are stopping in the polydispersity loop |
---|
79 | global const ProblemDetails *details, |
---|
80 | global const double *values, |
---|
81 | global const double *q, // nq q values, with padding to boundary |
---|
82 | global double *result, // nq+1 return values, again with padding |
---|
83 | const double cutoff // cutoff in the polydispersity weight product |
---|
84 | ) |
---|
85 | { |
---|
86 | |
---|
87 | // who we are and what element we are working with |
---|
88 | const int q_index = get_global_id(0); |
---|
89 | if (q_index >= nq) return; |
---|
90 | |
---|
91 | // Storage for the current parameter values. These will be updated as we |
---|
92 | // walk the polydispersity cube. |
---|
93 | ParameterBlock local_values; |
---|
94 | |
---|
95 | #if defined(MAGNETIC) && NUM_MAGNETIC>0 |
---|
96 | // Location of the sld parameters in the parameter vector. |
---|
97 | // These parameters are updated with the effective sld due to magnetism. |
---|
98 | #if NUM_MAGNETIC > 3 |
---|
99 | const int32_t slds[] = { MAGNETIC_PARS }; |
---|
100 | #endif |
---|
101 | |
---|
102 | // TODO: could precompute these outside of the kernel. |
---|
103 | // Interpret polarization cross section. |
---|
104 | // up_frac_i = values[NUM_PARS+2]; |
---|
105 | // up_frac_f = values[NUM_PARS+3]; |
---|
106 | // up_angle = values[NUM_PARS+4]; |
---|
107 | double spins[4]; |
---|
108 | double cos_mspin, sin_mspin; |
---|
109 | set_spins(values[NUM_PARS+2], values[NUM_PARS+3], spins); |
---|
110 | SINCOS(-values[NUM_PARS+4]*M_PI_180, sin_mspin, cos_mspin); |
---|
111 | #endif // MAGNETIC |
---|
112 | |
---|
113 | // Fill in the initial variables |
---|
114 | // values[0] is scale |
---|
115 | // values[1] is background |
---|
116 | for (int i=0; i < NUM_PARS; i++) { |
---|
117 | local_values.vector[i] = values[2+i]; |
---|
118 | //if (q_index==0) printf("p%d = %g\n",i, local_values.vector[i]); |
---|
119 | } |
---|
120 | //if (q_index==0) printf("NUM_VALUES:%d NUM_PARS:%d MAX_PD:%d\n", NUM_VALUES, NUM_PARS, MAX_PD); |
---|
121 | //if (q_index==0) printf("start:%d stop:%d\n", pd_start, pd_stop); |
---|
122 | |
---|
123 | double pd_norm = (pd_start == 0 ? 0.0 : result[nq]); |
---|
124 | double this_result = (pd_start == 0 ? 0.0 : result[q_index]); |
---|
125 | //if (q_index==0) printf("start %d %g %g\n", pd_start, pd_norm, this_result); |
---|
126 | |
---|
127 | #if MAX_PD>0 |
---|
128 | global const double *pd_value = values + NUM_VALUES; |
---|
129 | global const double *pd_weight = pd_value + details->num_weights; |
---|
130 | #endif |
---|
131 | |
---|
132 | // Jump into the middle of the polydispersity loop |
---|
133 | #if MAX_PD>4 |
---|
134 | int n4=details->pd_length[4]; |
---|
135 | int i4=(pd_start/details->pd_stride[4])%n4; |
---|
136 | const int p4=details->pd_par[4]; |
---|
137 | global const double *v4 = pd_value + details->pd_offset[4]; |
---|
138 | global const double *w4 = pd_weight + details->pd_offset[4]; |
---|
139 | #endif |
---|
140 | #if MAX_PD>3 |
---|
141 | int n3=details->pd_length[3]; |
---|
142 | int i3=(pd_start/details->pd_stride[3])%n3; |
---|
143 | const int p3=details->pd_par[3]; |
---|
144 | global const double *v3 = pd_value + details->pd_offset[3]; |
---|
145 | global const double *w3 = pd_weight + details->pd_offset[3]; |
---|
146 | //if (q_index==0) printf("offset %d: %d %d\n", 3, details->pd_offset[3], NUM_VALUES); |
---|
147 | #endif |
---|
148 | #if MAX_PD>2 |
---|
149 | int n2=details->pd_length[2]; |
---|
150 | int i2=(pd_start/details->pd_stride[2])%n2; |
---|
151 | const int p2=details->pd_par[2]; |
---|
152 | global const double *v2 = pd_value + details->pd_offset[2]; |
---|
153 | global const double *w2 = pd_weight + details->pd_offset[2]; |
---|
154 | #endif |
---|
155 | #if MAX_PD>1 |
---|
156 | int n1=details->pd_length[1]; |
---|
157 | int i1=(pd_start/details->pd_stride[1])%n1; |
---|
158 | const int p1=details->pd_par[1]; |
---|
159 | global const double *v1 = pd_value + details->pd_offset[1]; |
---|
160 | global const double *w1 = pd_weight + details->pd_offset[1]; |
---|
161 | #endif |
---|
162 | #if MAX_PD>0 |
---|
163 | int n0=details->pd_length[0]; |
---|
164 | int i0=(pd_start/details->pd_stride[0])%n0; |
---|
165 | const int p0=details->pd_par[0]; |
---|
166 | global const double *v0 = pd_value + details->pd_offset[0]; |
---|
167 | global const double *w0 = pd_weight + details->pd_offset[0]; |
---|
168 | #endif |
---|
169 | |
---|
170 | |
---|
171 | #if MAX_PD>0 |
---|
172 | const int theta_par = details->theta_par; |
---|
173 | const bool fast_theta = (theta_par == p0); |
---|
174 | const bool slow_theta = (theta_par >= 0 && !fast_theta); |
---|
175 | double spherical_correction = 1.0; |
---|
176 | #else |
---|
177 | // Note: if not polydisperse the weights cancel and we don't need the |
---|
178 | // spherical correction. |
---|
179 | const double spherical_correction = 1.0; |
---|
180 | #endif |
---|
181 | |
---|
182 | int step = pd_start; |
---|
183 | |
---|
184 | |
---|
185 | #if MAX_PD>4 |
---|
186 | const double weight5 = 1.0; |
---|
187 | while (i4 < n4) { |
---|
188 | local_values.vector[p4] = v4[i4]; |
---|
189 | double weight4 = w4[i4] * weight5; |
---|
190 | //if (q_index == 0) printf("step:%d level %d: p:%d i:%d n:%d value:%g weight:%g\n", step, 4, p4, i4, n4, local_values.vector[p4], weight4); |
---|
191 | #elif MAX_PD>3 |
---|
192 | const double weight4 = 1.0; |
---|
193 | #endif |
---|
194 | #if MAX_PD>3 |
---|
195 | while (i3 < n3) { |
---|
196 | local_values.vector[p3] = v3[i3]; |
---|
197 | double weight3 = w3[i3] * weight4; |
---|
198 | //if (q_index == 0) printf("step:%d level %d: p:%d i:%d n:%d value:%g weight:%g\n", step, 3, p3, i3, n3, local_values.vector[p3], weight3); |
---|
199 | #elif MAX_PD>2 |
---|
200 | const double weight3 = 1.0; |
---|
201 | #endif |
---|
202 | #if MAX_PD>2 |
---|
203 | while (i2 < n2) { |
---|
204 | local_values.vector[p2] = v2[i2]; |
---|
205 | double weight2 = w2[i2] * weight3; |
---|
206 | //if (q_index == 0) printf("step:%d level %d: p:%d i:%d n:%d value:%g weight:%g\n", step, 2, p2, i2, n2, local_values.vector[p2], weight2); |
---|
207 | #elif MAX_PD>1 |
---|
208 | const double weight2 = 1.0; |
---|
209 | #endif |
---|
210 | #if MAX_PD>1 |
---|
211 | while (i1 < n1) { |
---|
212 | local_values.vector[p1] = v1[i1]; |
---|
213 | double weight1 = w1[i1] * weight2; |
---|
214 | //if (q_index == 0) printf("step:%d level %d: p:%d i:%d n:%d value:%g weight:%g\n", step, 1, p1, i1, n1, local_values.vector[p1], weight1); |
---|
215 | #elif MAX_PD>0 |
---|
216 | const double weight1 = 1.0; |
---|
217 | #endif |
---|
218 | #if MAX_PD>0 |
---|
219 | if (slow_theta) { // Theta is not in inner loop |
---|
220 | spherical_correction = fmax(fabs(cos(M_PI_180*local_values.vector[theta_par])), 1.e-6); |
---|
221 | } |
---|
222 | while(i0 < n0) { |
---|
223 | local_values.vector[p0] = v0[i0]; |
---|
224 | double weight0 = w0[i0] * weight1; |
---|
225 | //if (q_index == 0) printf("step:%d level %d: p:%d i:%d n:%d value:%g weight:%g\n", step, 0, p0, i0, n0, local_values.vector[p0], weight0); |
---|
226 | if (fast_theta) { // Theta is in inner loop |
---|
227 | spherical_correction = fmax(fabs(cos(M_PI_180*local_values.vector[p0])), 1.e-6); |
---|
228 | } |
---|
229 | #else |
---|
230 | const double weight0 = 1.0; |
---|
231 | #endif |
---|
232 | |
---|
233 | //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"); } |
---|
234 | //if (q_index == 0) printf("sphcor: %g\n", spherical_correction); |
---|
235 | |
---|
236 | #ifdef INVALID |
---|
237 | if (!INVALID(local_values.table)) |
---|
238 | #endif |
---|
239 | { |
---|
240 | // Accumulate I(q) |
---|
241 | // Note: weight==0 must always be excluded |
---|
242 | if (weight0 > cutoff) { |
---|
243 | // spherical correction is set at a minimum of 1e-6, otherwise there |
---|
244 | // would be problems looking at models with theta=90. |
---|
245 | const double weight = weight0 * spherical_correction; |
---|
246 | pd_norm += weight * CALL_VOLUME(local_values.table); |
---|
247 | |
---|
248 | #if defined(MAGNETIC) && NUM_MAGNETIC > 0 |
---|
249 | const double qx = q[2*q_index]; |
---|
250 | const double qy = q[2*q_index+1]; |
---|
251 | const double qsq = qx*qx + qy*qy; |
---|
252 | |
---|
253 | // Constant across orientation, polydispersity for given qx, qy |
---|
254 | double scattering = 0.0; |
---|
255 | // TODO: what is the magnetic scattering at q=0 |
---|
256 | if (qsq > 1.e-16) { |
---|
257 | double p[4]; // dd, du, ud, uu |
---|
258 | p[0] = (qy*cos_mspin + qx*sin_mspin)/qsq; |
---|
259 | p[3] = -p[0]; |
---|
260 | p[1] = p[2] = (qy*sin_mspin - qx*cos_mspin)/qsq; |
---|
261 | |
---|
262 | for (int index=0; index<4; index++) { |
---|
263 | const double xs = spins[index]; |
---|
264 | if (xs > 1.e-8) { |
---|
265 | const int spin_flip = (index==1) || (index==2); |
---|
266 | const double pk = p[index]; |
---|
267 | for (int axis=0; axis<=spin_flip; axis++) { |
---|
268 | #define M1 NUM_PARS+5 |
---|
269 | #define M2 NUM_PARS+8 |
---|
270 | #define M3 NUM_PARS+13 |
---|
271 | #define SLD(_M_offset, _sld_offset) \ |
---|
272 | local_values.vector[_sld_offset] = xs * (axis \ |
---|
273 | ? (index==1 ? -values[_M_offset+2] : values[_M_offset+2]) \ |
---|
274 | : mag_sld(qx, qy, pk, values[_M_offset], values[_M_offset+1], \ |
---|
275 | (spin_flip ? 0.0 : values[_sld_offset+2]))) |
---|
276 | #if NUM_MAGNETIC==1 |
---|
277 | SLD(M1, MAGNETIC_PAR1); |
---|
278 | #elif NUM_MAGNETIC==2 |
---|
279 | SLD(M1, MAGNETIC_PAR1); |
---|
280 | SLD(M2, MAGNETIC_PAR2); |
---|
281 | #elif NUM_MAGNETIC==3 |
---|
282 | SLD(M1, MAGNETIC_PAR1); |
---|
283 | SLD(M2, MAGNETIC_PAR2); |
---|
284 | SLD(M3, MAGNETIC_PAR3); |
---|
285 | #else |
---|
286 | for (int sk=0; sk<NUM_MAGNETIC; sk++) { |
---|
287 | SLD(M1+3*sk, slds[sk]); |
---|
288 | } |
---|
289 | #endif |
---|
290 | scattering += CALL_IQ(q, q_index, local_values.table); |
---|
291 | } |
---|
292 | } |
---|
293 | } |
---|
294 | } |
---|
295 | #else // !MAGNETIC |
---|
296 | const double scattering = CALL_IQ(q, q_index, local_values.table); |
---|
297 | #endif // !MAGNETIC |
---|
298 | this_result += weight * scattering; |
---|
299 | } |
---|
300 | } |
---|
301 | ++step; |
---|
302 | #if MAX_PD>0 |
---|
303 | if (step >= pd_stop) break; |
---|
304 | ++i0; |
---|
305 | } |
---|
306 | i0 = 0; |
---|
307 | #endif |
---|
308 | #if MAX_PD>1 |
---|
309 | if (step >= pd_stop) break; |
---|
310 | ++i1; |
---|
311 | } |
---|
312 | i1 = 0; |
---|
313 | #endif |
---|
314 | #if MAX_PD>2 |
---|
315 | if (step >= pd_stop) break; |
---|
316 | ++i2; |
---|
317 | } |
---|
318 | i2 = 0; |
---|
319 | #endif |
---|
320 | #if MAX_PD>3 |
---|
321 | if (step >= pd_stop) break; |
---|
322 | ++i3; |
---|
323 | } |
---|
324 | i3 = 0; |
---|
325 | #endif |
---|
326 | #if MAX_PD>4 |
---|
327 | if (step >= pd_stop) break; |
---|
328 | ++i4; |
---|
329 | } |
---|
330 | i4 = 0; |
---|
331 | #endif |
---|
332 | |
---|
333 | //if (q_index==0) printf("res: %g/%g\n", this_result, pd_norm); |
---|
334 | // Remember the current result and the updated norm. |
---|
335 | result[q_index] = this_result; |
---|
336 | if (q_index == 0) result[nq] = pd_norm; |
---|
337 | } |
---|