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