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