1 #include "simint/boys/boys.h"
2 #include "simint/ostei/gen/ostei_generated.h"
3 #include "simint/vectorization/vectorization.h"
4 #include <math.h>
5 #include <string.h>
6
7
ostei_s_s_h_s(struct simint_multi_shellpair const P,struct simint_multi_shellpair const Q,double screen_tol,double * const restrict work,double * const restrict INT__s_s_h_s)8 int ostei_s_s_h_s(struct simint_multi_shellpair const P,
9 struct simint_multi_shellpair const Q,
10 double screen_tol,
11 double * const restrict work,
12 double * const restrict INT__s_s_h_s)
13 {
14
15 SIMINT_ASSUME_ALIGN_DBL(work);
16 SIMINT_ASSUME_ALIGN_DBL(INT__s_s_h_s);
17 memset(INT__s_s_h_s, 0, P.nshell12_clip * Q.nshell12_clip * 21 * sizeof(double));
18
19 int ab, cd, abcd;
20 int istart, jstart;
21 int iprimcd, nprim_icd, icd;
22 const int check_screen = (screen_tol > 0.0);
23 int i, j;
24 int n;
25 int not_screened;
26
27 // partition workspace
28 SIMINT_DBLTYPE * const primwork = (SIMINT_DBLTYPE *)(work + SIMINT_NSHELL_SIMD*0);
29 SIMINT_DBLTYPE * const restrict PRIM_INT__s_s_s_s = primwork + 0;
30 SIMINT_DBLTYPE * const restrict PRIM_INT__s_s_p_s = primwork + 6;
31 SIMINT_DBLTYPE * const restrict PRIM_INT__s_s_d_s = primwork + 21;
32 SIMINT_DBLTYPE * const restrict PRIM_INT__s_s_f_s = primwork + 45;
33 SIMINT_DBLTYPE * const restrict PRIM_INT__s_s_g_s = primwork + 75;
34 SIMINT_DBLTYPE * const restrict PRIM_INT__s_s_h_s = primwork + 105;
35 double * const hrrwork = (double *)(primwork + 126);
36
37
38 // Create constants
39 const SIMINT_DBLTYPE const_1 = SIMINT_DBLSET1(1);
40 const SIMINT_DBLTYPE const_2 = SIMINT_DBLSET1(2);
41 const SIMINT_DBLTYPE const_3 = SIMINT_DBLSET1(3);
42 const SIMINT_DBLTYPE const_4 = SIMINT_DBLSET1(4);
43 const SIMINT_DBLTYPE one_half = SIMINT_DBLSET1(0.5);
44
45
46 ////////////////////////////////////////
47 // Loop over shells and primitives
48 ////////////////////////////////////////
49
50 abcd = 0;
51 istart = 0;
52 for(ab = 0; ab < P.nshell12_clip; ++ab)
53 {
54 const int iend = istart + P.nprim12[ab];
55
56 cd = 0;
57 jstart = 0;
58
59 for(cd = 0; cd < Q.nshell12_clip; cd += SIMINT_NSHELL_SIMD)
60 {
61 const int nshellbatch = ((cd + SIMINT_NSHELL_SIMD) > Q.nshell12_clip) ? Q.nshell12_clip - cd : SIMINT_NSHELL_SIMD;
62 int jend = jstart;
63 for(i = 0; i < nshellbatch; i++)
64 jend += Q.nprim12[cd+i];
65
66
67 for(i = istart; i < iend; ++i)
68 {
69 SIMINT_DBLTYPE bra_screen_max; // only used if check_screen
70
71 if(check_screen)
72 {
73 // Skip this whole thing if always insignificant
74 if((P.screen[i] * Q.screen_max) < screen_tol)
75 continue;
76 bra_screen_max = SIMINT_DBLSET1(P.screen[i]);
77 }
78
79 icd = 0;
80 iprimcd = 0;
81 nprim_icd = Q.nprim12[cd];
82 double * restrict PRIM_PTR_INT__s_s_h_s = INT__s_s_h_s + abcd * 21;
83
84
85
86 // Load these one per loop over i
87 const SIMINT_DBLTYPE P_alpha = SIMINT_DBLSET1(P.alpha[i]);
88 const SIMINT_DBLTYPE P_prefac = SIMINT_DBLSET1(P.prefac[i]);
89 const SIMINT_DBLTYPE Pxyz[3] = { SIMINT_DBLSET1(P.x[i]), SIMINT_DBLSET1(P.y[i]), SIMINT_DBLSET1(P.z[i]) };
90
91
92 for(j = jstart; j < jend; j += SIMINT_SIMD_LEN)
93 {
94 // calculate the shell offsets
95 // these are the offset from the shell pointed to by cd
96 // for each element
97 int shelloffsets[SIMINT_SIMD_LEN] = {0};
98 int lastoffset = 0;
99 const int nlane = ( ((j + SIMINT_SIMD_LEN) < jend) ? SIMINT_SIMD_LEN : (jend - j));
100
101 if((iprimcd + SIMINT_SIMD_LEN) >= nprim_icd)
102 {
103 // Handle if the first element of the vector is a new shell
104 if(iprimcd >= nprim_icd && ((icd+1) < nshellbatch))
105 {
106 nprim_icd += Q.nprim12[cd + (++icd)];
107 PRIM_PTR_INT__s_s_h_s += 21;
108 }
109 iprimcd++;
110 for(n = 1; n < SIMINT_SIMD_LEN; ++n)
111 {
112 if(iprimcd >= nprim_icd && ((icd+1) < nshellbatch))
113 {
114 shelloffsets[n] = shelloffsets[n-1] + 1;
115 lastoffset++;
116 nprim_icd += Q.nprim12[cd + (++icd)];
117 }
118 else
119 shelloffsets[n] = shelloffsets[n-1];
120 iprimcd++;
121 }
122 }
123 else
124 iprimcd += SIMINT_SIMD_LEN;
125
126 // Do we have to compute this vector (or has it been screened out)?
127 // (not_screened != 0 means we have to do this vector)
128 if(check_screen)
129 {
130 const double vmax = vector_max(SIMINT_MUL(bra_screen_max, SIMINT_DBLLOAD(Q.screen, j)));
131 if(vmax < screen_tol)
132 {
133 PRIM_PTR_INT__s_s_h_s += lastoffset*21;
134 continue;
135 }
136 }
137
138 const SIMINT_DBLTYPE Q_alpha = SIMINT_DBLLOAD(Q.alpha, j);
139 const SIMINT_DBLTYPE PQalpha_mul = SIMINT_MUL(P_alpha, Q_alpha);
140 const SIMINT_DBLTYPE PQalpha_sum = SIMINT_ADD(P_alpha, Q_alpha);
141 const SIMINT_DBLTYPE one_over_PQalpha_sum = SIMINT_DIV(const_1, PQalpha_sum);
142
143
144 /* construct R2 = (Px - Qx)**2 + (Py - Qy)**2 + (Pz -Qz)**2 */
145 SIMINT_DBLTYPE PQ[3];
146 PQ[0] = SIMINT_SUB(Pxyz[0], SIMINT_DBLLOAD(Q.x, j));
147 PQ[1] = SIMINT_SUB(Pxyz[1], SIMINT_DBLLOAD(Q.y, j));
148 PQ[2] = SIMINT_SUB(Pxyz[2], SIMINT_DBLLOAD(Q.z, j));
149 SIMINT_DBLTYPE R2 = SIMINT_MUL(PQ[0], PQ[0]);
150 R2 = SIMINT_FMADD(PQ[1], PQ[1], R2);
151 R2 = SIMINT_FMADD(PQ[2], PQ[2], R2);
152
153 const SIMINT_DBLTYPE alpha = SIMINT_MUL(PQalpha_mul, one_over_PQalpha_sum); // alpha from MEST
154 const SIMINT_DBLTYPE one_over_p = SIMINT_DIV(const_1, P_alpha);
155 const SIMINT_DBLTYPE one_over_q = SIMINT_DIV(const_1, Q_alpha);
156 const SIMINT_DBLTYPE one_over_2p = SIMINT_MUL(one_half, one_over_p);
157 const SIMINT_DBLTYPE one_over_2q = SIMINT_MUL(one_half, one_over_q);
158 const SIMINT_DBLTYPE one_over_2pq = SIMINT_MUL(one_half, one_over_PQalpha_sum);
159 const SIMINT_DBLTYPE Q_PA[3] = { SIMINT_DBLLOAD(Q.PA_x, j), SIMINT_DBLLOAD(Q.PA_y, j), SIMINT_DBLLOAD(Q.PA_z, j) };
160
161 SIMINT_DBLTYPE a_over_q = SIMINT_MUL(alpha, one_over_q);
162 SIMINT_DBLTYPE aoq_PQ[3];
163 aoq_PQ[0] = SIMINT_MUL(a_over_q, PQ[0]);
164 aoq_PQ[1] = SIMINT_MUL(a_over_q, PQ[1]);
165 aoq_PQ[2] = SIMINT_MUL(a_over_q, PQ[2]);
166 // Put a minus sign here so we don't have to in RR routines
167 a_over_q = SIMINT_NEG(a_over_q);
168
169
170 //////////////////////////////////////////////
171 // Fjt function section
172 // Maximum v value: 5
173 //////////////////////////////////////////////
174 // The parameter to the Fjt function
175 const SIMINT_DBLTYPE F_x = SIMINT_MUL(R2, alpha);
176
177
178 const SIMINT_DBLTYPE Q_prefac = mask_load(nlane, Q.prefac + j);
179
180
181 boys_F_split(PRIM_INT__s_s_s_s, F_x, 5);
182 SIMINT_DBLTYPE prefac = SIMINT_SQRT(one_over_PQalpha_sum);
183 prefac = SIMINT_MUL(SIMINT_MUL(P_prefac, Q_prefac), prefac);
184 for(n = 0; n <= 5; n++)
185 PRIM_INT__s_s_s_s[n] = SIMINT_MUL(PRIM_INT__s_s_s_s[n], prefac);
186
187 //////////////////////////////////////////////
188 // Primitive integrals: Vertical recurrance
189 //////////////////////////////////////////////
190
191 const SIMINT_DBLTYPE vrr_const_1_over_2q = one_over_2q;
192 const SIMINT_DBLTYPE vrr_const_2_over_2q = SIMINT_MUL(const_2, one_over_2q);
193 const SIMINT_DBLTYPE vrr_const_3_over_2q = SIMINT_MUL(const_3, one_over_2q);
194 const SIMINT_DBLTYPE vrr_const_4_over_2q = SIMINT_MUL(const_4, one_over_2q);
195
196
197
198 // Forming PRIM_INT__s_s_p_s[5 * 3];
199 for(n = 0; n < 5; ++n) // loop over orders of auxiliary function
200 {
201
202 PRIM_INT__s_s_p_s[n * 3 + 0] = SIMINT_MUL(Q_PA[0], PRIM_INT__s_s_s_s[n * 1 + 0]);
203 PRIM_INT__s_s_p_s[n * 3 + 0] = SIMINT_FMADD( aoq_PQ[0], PRIM_INT__s_s_s_s[(n+1) * 1 + 0], PRIM_INT__s_s_p_s[n * 3 + 0]);
204
205 PRIM_INT__s_s_p_s[n * 3 + 1] = SIMINT_MUL(Q_PA[1], PRIM_INT__s_s_s_s[n * 1 + 0]);
206 PRIM_INT__s_s_p_s[n * 3 + 1] = SIMINT_FMADD( aoq_PQ[1], PRIM_INT__s_s_s_s[(n+1) * 1 + 0], PRIM_INT__s_s_p_s[n * 3 + 1]);
207
208 PRIM_INT__s_s_p_s[n * 3 + 2] = SIMINT_MUL(Q_PA[2], PRIM_INT__s_s_s_s[n * 1 + 0]);
209 PRIM_INT__s_s_p_s[n * 3 + 2] = SIMINT_FMADD( aoq_PQ[2], PRIM_INT__s_s_s_s[(n+1) * 1 + 0], PRIM_INT__s_s_p_s[n * 3 + 2]);
210
211 }
212
213
214
215 // Forming PRIM_INT__s_s_d_s[4 * 6];
216 for(n = 0; n < 4; ++n) // loop over orders of auxiliary function
217 {
218
219 PRIM_INT__s_s_d_s[n * 6 + 0] = SIMINT_MUL(Q_PA[0], PRIM_INT__s_s_p_s[n * 3 + 0]);
220 PRIM_INT__s_s_d_s[n * 6 + 0] = SIMINT_FMADD( aoq_PQ[0], PRIM_INT__s_s_p_s[(n+1) * 3 + 0], PRIM_INT__s_s_d_s[n * 6 + 0]);
221 PRIM_INT__s_s_d_s[n * 6 + 0] = SIMINT_FMADD( vrr_const_1_over_2q, SIMINT_FMADD(a_over_q, PRIM_INT__s_s_s_s[(n+1) * 1 + 0], PRIM_INT__s_s_s_s[n * 1 + 0]), PRIM_INT__s_s_d_s[n * 6 + 0]);
222
223 PRIM_INT__s_s_d_s[n * 6 + 3] = SIMINT_MUL(Q_PA[1], PRIM_INT__s_s_p_s[n * 3 + 1]);
224 PRIM_INT__s_s_d_s[n * 6 + 3] = SIMINT_FMADD( aoq_PQ[1], PRIM_INT__s_s_p_s[(n+1) * 3 + 1], PRIM_INT__s_s_d_s[n * 6 + 3]);
225 PRIM_INT__s_s_d_s[n * 6 + 3] = SIMINT_FMADD( vrr_const_1_over_2q, SIMINT_FMADD(a_over_q, PRIM_INT__s_s_s_s[(n+1) * 1 + 0], PRIM_INT__s_s_s_s[n * 1 + 0]), PRIM_INT__s_s_d_s[n * 6 + 3]);
226
227 PRIM_INT__s_s_d_s[n * 6 + 5] = SIMINT_MUL(Q_PA[2], PRIM_INT__s_s_p_s[n * 3 + 2]);
228 PRIM_INT__s_s_d_s[n * 6 + 5] = SIMINT_FMADD( aoq_PQ[2], PRIM_INT__s_s_p_s[(n+1) * 3 + 2], PRIM_INT__s_s_d_s[n * 6 + 5]);
229 PRIM_INT__s_s_d_s[n * 6 + 5] = SIMINT_FMADD( vrr_const_1_over_2q, SIMINT_FMADD(a_over_q, PRIM_INT__s_s_s_s[(n+1) * 1 + 0], PRIM_INT__s_s_s_s[n * 1 + 0]), PRIM_INT__s_s_d_s[n * 6 + 5]);
230
231 }
232
233
234
235 // Forming PRIM_INT__s_s_f_s[3 * 10];
236 for(n = 0; n < 3; ++n) // loop over orders of auxiliary function
237 {
238
239 PRIM_INT__s_s_f_s[n * 10 + 0] = SIMINT_MUL(Q_PA[0], PRIM_INT__s_s_d_s[n * 6 + 0]);
240 PRIM_INT__s_s_f_s[n * 10 + 0] = SIMINT_FMADD( aoq_PQ[0], PRIM_INT__s_s_d_s[(n+1) * 6 + 0], PRIM_INT__s_s_f_s[n * 10 + 0]);
241 PRIM_INT__s_s_f_s[n * 10 + 0] = SIMINT_FMADD( vrr_const_2_over_2q, SIMINT_FMADD(a_over_q, PRIM_INT__s_s_p_s[(n+1) * 3 + 0], PRIM_INT__s_s_p_s[n * 3 + 0]), PRIM_INT__s_s_f_s[n * 10 + 0]);
242
243 PRIM_INT__s_s_f_s[n * 10 + 1] = SIMINT_MUL(Q_PA[1], PRIM_INT__s_s_d_s[n * 6 + 0]);
244 PRIM_INT__s_s_f_s[n * 10 + 1] = SIMINT_FMADD( aoq_PQ[1], PRIM_INT__s_s_d_s[(n+1) * 6 + 0], PRIM_INT__s_s_f_s[n * 10 + 1]);
245
246 PRIM_INT__s_s_f_s[n * 10 + 2] = SIMINT_MUL(Q_PA[2], PRIM_INT__s_s_d_s[n * 6 + 0]);
247 PRIM_INT__s_s_f_s[n * 10 + 2] = SIMINT_FMADD( aoq_PQ[2], PRIM_INT__s_s_d_s[(n+1) * 6 + 0], PRIM_INT__s_s_f_s[n * 10 + 2]);
248
249 PRIM_INT__s_s_f_s[n * 10 + 6] = SIMINT_MUL(Q_PA[1], PRIM_INT__s_s_d_s[n * 6 + 3]);
250 PRIM_INT__s_s_f_s[n * 10 + 6] = SIMINT_FMADD( aoq_PQ[1], PRIM_INT__s_s_d_s[(n+1) * 6 + 3], PRIM_INT__s_s_f_s[n * 10 + 6]);
251 PRIM_INT__s_s_f_s[n * 10 + 6] = SIMINT_FMADD( vrr_const_2_over_2q, SIMINT_FMADD(a_over_q, PRIM_INT__s_s_p_s[(n+1) * 3 + 1], PRIM_INT__s_s_p_s[n * 3 + 1]), PRIM_INT__s_s_f_s[n * 10 + 6]);
252
253 PRIM_INT__s_s_f_s[n * 10 + 7] = SIMINT_MUL(Q_PA[2], PRIM_INT__s_s_d_s[n * 6 + 3]);
254 PRIM_INT__s_s_f_s[n * 10 + 7] = SIMINT_FMADD( aoq_PQ[2], PRIM_INT__s_s_d_s[(n+1) * 6 + 3], PRIM_INT__s_s_f_s[n * 10 + 7]);
255
256 PRIM_INT__s_s_f_s[n * 10 + 9] = SIMINT_MUL(Q_PA[2], PRIM_INT__s_s_d_s[n * 6 + 5]);
257 PRIM_INT__s_s_f_s[n * 10 + 9] = SIMINT_FMADD( aoq_PQ[2], PRIM_INT__s_s_d_s[(n+1) * 6 + 5], PRIM_INT__s_s_f_s[n * 10 + 9]);
258 PRIM_INT__s_s_f_s[n * 10 + 9] = SIMINT_FMADD( vrr_const_2_over_2q, SIMINT_FMADD(a_over_q, PRIM_INT__s_s_p_s[(n+1) * 3 + 2], PRIM_INT__s_s_p_s[n * 3 + 2]), PRIM_INT__s_s_f_s[n * 10 + 9]);
259
260 }
261
262
263 VRR_K_s_s_g_s(
264 PRIM_INT__s_s_g_s,
265 PRIM_INT__s_s_f_s,
266 PRIM_INT__s_s_d_s,
267 Q_PA,
268 a_over_q,
269 aoq_PQ,
270 one_over_2q,
271 2);
272
273
274 VRR_K_s_s_h_s(
275 PRIM_INT__s_s_h_s,
276 PRIM_INT__s_s_g_s,
277 PRIM_INT__s_s_f_s,
278 Q_PA,
279 a_over_q,
280 aoq_PQ,
281 one_over_2q,
282 1);
283
284
285
286
287 ////////////////////////////////////
288 // Accumulate contracted integrals
289 ////////////////////////////////////
290 if(lastoffset == 0)
291 {
292 contract_all(21, PRIM_INT__s_s_h_s, PRIM_PTR_INT__s_s_h_s);
293 }
294 else
295 {
296 contract(21, shelloffsets, PRIM_INT__s_s_h_s, PRIM_PTR_INT__s_s_h_s);
297 PRIM_PTR_INT__s_s_h_s += lastoffset*21;
298 }
299
300 } // close loop over j
301 } // close loop over i
302
303 //Advance to the next batch
304 jstart = SIMINT_SIMD_ROUND(jend);
305 abcd += nshellbatch;
306
307 } // close loop cdbatch
308
309 istart = iend;
310 } // close loop over ab
311
312 return P.nshell12_clip * Q.nshell12_clip;
313 }
314
ostei_s_s_s_h(struct simint_multi_shellpair const P,struct simint_multi_shellpair const Q,double screen_tol,double * const restrict work,double * const restrict INT__s_s_s_h)315 int ostei_s_s_s_h(struct simint_multi_shellpair const P,
316 struct simint_multi_shellpair const Q,
317 double screen_tol,
318 double * const restrict work,
319 double * const restrict INT__s_s_s_h)
320 {
321 double Q_AB[3*Q.nshell12];
322 struct simint_multi_shellpair Q_tmp = Q;
323 Q_tmp.PA_x = Q.PB_x; Q_tmp.PA_y = Q.PB_y; Q_tmp.PA_z = Q.PB_z;
324 Q_tmp.PB_x = Q.PA_x; Q_tmp.PB_y = Q.PA_y; Q_tmp.PB_z = Q.PA_z;
325 Q_tmp.AB_x = Q_AB;
326 Q_tmp.AB_y = Q_AB + Q.nshell12;
327 Q_tmp.AB_z = Q_AB + 2*Q.nshell12;
328
329 for(int i = 0; i < Q.nshell12; i++)
330 {
331 Q_tmp.AB_x[i] = -Q.AB_x[i];
332 Q_tmp.AB_y[i] = -Q.AB_y[i];
333 Q_tmp.AB_z[i] = -Q.AB_z[i];
334 }
335
336 int ret = ostei_s_s_h_s(P, Q_tmp, screen_tol, work, INT__s_s_s_h);
337
338 return ret;
339 }
340
341