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_p_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_p_s_p)8 int ostei_s_p_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_p_s_p)
13 {
14
15 SIMINT_ASSUME_ALIGN_DBL(work);
16 SIMINT_ASSUME_ALIGN_DBL(INT__s_p_s_p);
17 memset(INT__s_p_s_p, 0, P.nshell12_clip * Q.nshell12_clip * 9 * 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_p_s_s = primwork + 3;
31 SIMINT_DBLTYPE * const restrict PRIM_INT__s_p_s_p = primwork + 9;
32 double * const hrrwork = (double *)(primwork + 18);
33
34
35 // Create constants
36 const SIMINT_DBLTYPE const_1 = SIMINT_DBLSET1(1);
37 const SIMINT_DBLTYPE one_half = SIMINT_DBLSET1(0.5);
38
39
40 ////////////////////////////////////////
41 // Loop over shells and primitives
42 ////////////////////////////////////////
43
44 abcd = 0;
45 istart = 0;
46 for(ab = 0; ab < P.nshell12_clip; ++ab)
47 {
48 const int iend = istart + P.nprim12[ab];
49
50 cd = 0;
51 jstart = 0;
52
53 for(cd = 0; cd < Q.nshell12_clip; cd += SIMINT_NSHELL_SIMD)
54 {
55 const int nshellbatch = ((cd + SIMINT_NSHELL_SIMD) > Q.nshell12_clip) ? Q.nshell12_clip - cd : SIMINT_NSHELL_SIMD;
56 int jend = jstart;
57 for(i = 0; i < nshellbatch; i++)
58 jend += Q.nprim12[cd+i];
59
60
61 for(i = istart; i < iend; ++i)
62 {
63 SIMINT_DBLTYPE bra_screen_max; // only used if check_screen
64
65 if(check_screen)
66 {
67 // Skip this whole thing if always insignificant
68 if((P.screen[i] * Q.screen_max) < screen_tol)
69 continue;
70 bra_screen_max = SIMINT_DBLSET1(P.screen[i]);
71 }
72
73 icd = 0;
74 iprimcd = 0;
75 nprim_icd = Q.nprim12[cd];
76 double * restrict PRIM_PTR_INT__s_p_s_p = INT__s_p_s_p + abcd * 9;
77
78
79
80 // Load these one per loop over i
81 const SIMINT_DBLTYPE P_alpha = SIMINT_DBLSET1(P.alpha[i]);
82 const SIMINT_DBLTYPE P_prefac = SIMINT_DBLSET1(P.prefac[i]);
83 const SIMINT_DBLTYPE Pxyz[3] = { SIMINT_DBLSET1(P.x[i]), SIMINT_DBLSET1(P.y[i]), SIMINT_DBLSET1(P.z[i]) };
84
85 const SIMINT_DBLTYPE P_PB[3] = { SIMINT_DBLSET1(P.PB_x[i]), SIMINT_DBLSET1(P.PB_y[i]), SIMINT_DBLSET1(P.PB_z[i]) };
86
87 for(j = jstart; j < jend; j += SIMINT_SIMD_LEN)
88 {
89 // calculate the shell offsets
90 // these are the offset from the shell pointed to by cd
91 // for each element
92 int shelloffsets[SIMINT_SIMD_LEN] = {0};
93 int lastoffset = 0;
94 const int nlane = ( ((j + SIMINT_SIMD_LEN) < jend) ? SIMINT_SIMD_LEN : (jend - j));
95
96 if((iprimcd + SIMINT_SIMD_LEN) >= nprim_icd)
97 {
98 // Handle if the first element of the vector is a new shell
99 if(iprimcd >= nprim_icd && ((icd+1) < nshellbatch))
100 {
101 nprim_icd += Q.nprim12[cd + (++icd)];
102 PRIM_PTR_INT__s_p_s_p += 9;
103 }
104 iprimcd++;
105 for(n = 1; n < SIMINT_SIMD_LEN; ++n)
106 {
107 if(iprimcd >= nprim_icd && ((icd+1) < nshellbatch))
108 {
109 shelloffsets[n] = shelloffsets[n-1] + 1;
110 lastoffset++;
111 nprim_icd += Q.nprim12[cd + (++icd)];
112 }
113 else
114 shelloffsets[n] = shelloffsets[n-1];
115 iprimcd++;
116 }
117 }
118 else
119 iprimcd += SIMINT_SIMD_LEN;
120
121 // Do we have to compute this vector (or has it been screened out)?
122 // (not_screened != 0 means we have to do this vector)
123 if(check_screen)
124 {
125 const double vmax = vector_max(SIMINT_MUL(bra_screen_max, SIMINT_DBLLOAD(Q.screen, j)));
126 if(vmax < screen_tol)
127 {
128 PRIM_PTR_INT__s_p_s_p += lastoffset*9;
129 continue;
130 }
131 }
132
133 const SIMINT_DBLTYPE Q_alpha = SIMINT_DBLLOAD(Q.alpha, j);
134 const SIMINT_DBLTYPE PQalpha_mul = SIMINT_MUL(P_alpha, Q_alpha);
135 const SIMINT_DBLTYPE PQalpha_sum = SIMINT_ADD(P_alpha, Q_alpha);
136 const SIMINT_DBLTYPE one_over_PQalpha_sum = SIMINT_DIV(const_1, PQalpha_sum);
137
138
139 /* construct R2 = (Px - Qx)**2 + (Py - Qy)**2 + (Pz -Qz)**2 */
140 SIMINT_DBLTYPE PQ[3];
141 PQ[0] = SIMINT_SUB(Pxyz[0], SIMINT_DBLLOAD(Q.x, j));
142 PQ[1] = SIMINT_SUB(Pxyz[1], SIMINT_DBLLOAD(Q.y, j));
143 PQ[2] = SIMINT_SUB(Pxyz[2], SIMINT_DBLLOAD(Q.z, j));
144 SIMINT_DBLTYPE R2 = SIMINT_MUL(PQ[0], PQ[0]);
145 R2 = SIMINT_FMADD(PQ[1], PQ[1], R2);
146 R2 = SIMINT_FMADD(PQ[2], PQ[2], R2);
147
148 const SIMINT_DBLTYPE alpha = SIMINT_MUL(PQalpha_mul, one_over_PQalpha_sum); // alpha from MEST
149 const SIMINT_DBLTYPE one_over_p = SIMINT_DIV(const_1, P_alpha);
150 const SIMINT_DBLTYPE one_over_q = SIMINT_DIV(const_1, Q_alpha);
151 const SIMINT_DBLTYPE one_over_2p = SIMINT_MUL(one_half, one_over_p);
152 const SIMINT_DBLTYPE one_over_2q = SIMINT_MUL(one_half, one_over_q);
153 const SIMINT_DBLTYPE one_over_2pq = SIMINT_MUL(one_half, one_over_PQalpha_sum);
154 const SIMINT_DBLTYPE Q_PB[3] = { SIMINT_DBLLOAD(Q.PB_x, j), SIMINT_DBLLOAD(Q.PB_y, j), SIMINT_DBLLOAD(Q.PB_z, j) };
155
156 // NOTE: Minus sign!
157 const SIMINT_DBLTYPE a_over_p = SIMINT_MUL(SIMINT_NEG(alpha), one_over_p);
158 SIMINT_DBLTYPE aop_PQ[3];
159 aop_PQ[0] = SIMINT_MUL(a_over_p, PQ[0]);
160 aop_PQ[1] = SIMINT_MUL(a_over_p, PQ[1]);
161 aop_PQ[2] = SIMINT_MUL(a_over_p, PQ[2]);
162
163 SIMINT_DBLTYPE a_over_q = SIMINT_MUL(alpha, one_over_q);
164 SIMINT_DBLTYPE aoq_PQ[3];
165 aoq_PQ[0] = SIMINT_MUL(a_over_q, PQ[0]);
166 aoq_PQ[1] = SIMINT_MUL(a_over_q, PQ[1]);
167 aoq_PQ[2] = SIMINT_MUL(a_over_q, PQ[2]);
168 // Put a minus sign here so we don't have to in RR routines
169 a_over_q = SIMINT_NEG(a_over_q);
170
171
172 //////////////////////////////////////////////
173 // Fjt function section
174 // Maximum v value: 2
175 //////////////////////////////////////////////
176 // The parameter to the Fjt function
177 const SIMINT_DBLTYPE F_x = SIMINT_MUL(R2, alpha);
178
179
180 const SIMINT_DBLTYPE Q_prefac = mask_load(nlane, Q.prefac + j);
181
182
183 boys_F_split(PRIM_INT__s_s_s_s, F_x, 2);
184 SIMINT_DBLTYPE prefac = SIMINT_SQRT(one_over_PQalpha_sum);
185 prefac = SIMINT_MUL(SIMINT_MUL(P_prefac, Q_prefac), prefac);
186 for(n = 0; n <= 2; n++)
187 PRIM_INT__s_s_s_s[n] = SIMINT_MUL(PRIM_INT__s_s_s_s[n], prefac);
188
189 //////////////////////////////////////////////
190 // Primitive integrals: Vertical recurrance
191 //////////////////////////////////////////////
192
193 const SIMINT_DBLTYPE vrr_const_1_over_2pq = one_over_2pq;
194
195
196
197 // Forming PRIM_INT__s_p_s_s[2 * 3];
198 for(n = 0; n < 2; ++n) // loop over orders of auxiliary function
199 {
200
201 PRIM_INT__s_p_s_s[n * 3 + 0] = SIMINT_MUL(P_PB[0], PRIM_INT__s_s_s_s[n * 1 + 0]);
202 PRIM_INT__s_p_s_s[n * 3 + 0] = SIMINT_FMADD( aop_PQ[0], PRIM_INT__s_s_s_s[(n+1) * 1 + 0], PRIM_INT__s_p_s_s[n * 3 + 0]);
203
204 PRIM_INT__s_p_s_s[n * 3 + 1] = SIMINT_MUL(P_PB[1], PRIM_INT__s_s_s_s[n * 1 + 0]);
205 PRIM_INT__s_p_s_s[n * 3 + 1] = SIMINT_FMADD( aop_PQ[1], PRIM_INT__s_s_s_s[(n+1) * 1 + 0], PRIM_INT__s_p_s_s[n * 3 + 1]);
206
207 PRIM_INT__s_p_s_s[n * 3 + 2] = SIMINT_MUL(P_PB[2], PRIM_INT__s_s_s_s[n * 1 + 0]);
208 PRIM_INT__s_p_s_s[n * 3 + 2] = SIMINT_FMADD( aop_PQ[2], PRIM_INT__s_s_s_s[(n+1) * 1 + 0], PRIM_INT__s_p_s_s[n * 3 + 2]);
209
210 }
211
212
213
214 // Forming PRIM_INT__s_p_s_p[1 * 9];
215 for(n = 0; n < 1; ++n) // loop over orders of auxiliary function
216 {
217
218 PRIM_INT__s_p_s_p[n * 9 + 0] = SIMINT_MUL(Q_PB[0], PRIM_INT__s_p_s_s[n * 3 + 0]);
219 PRIM_INT__s_p_s_p[n * 9 + 0] = SIMINT_FMADD( aoq_PQ[0], PRIM_INT__s_p_s_s[(n+1) * 3 + 0], PRIM_INT__s_p_s_p[n * 9 + 0]);
220 PRIM_INT__s_p_s_p[n * 9 + 0] = SIMINT_FMADD( vrr_const_1_over_2pq, PRIM_INT__s_s_s_s[(n+1) * 1 + 0], PRIM_INT__s_p_s_p[n * 9 + 0]);
221
222 PRIM_INT__s_p_s_p[n * 9 + 1] = SIMINT_MUL(Q_PB[1], PRIM_INT__s_p_s_s[n * 3 + 0]);
223 PRIM_INT__s_p_s_p[n * 9 + 1] = SIMINT_FMADD( aoq_PQ[1], PRIM_INT__s_p_s_s[(n+1) * 3 + 0], PRIM_INT__s_p_s_p[n * 9 + 1]);
224
225 PRIM_INT__s_p_s_p[n * 9 + 2] = SIMINT_MUL(Q_PB[2], PRIM_INT__s_p_s_s[n * 3 + 0]);
226 PRIM_INT__s_p_s_p[n * 9 + 2] = SIMINT_FMADD( aoq_PQ[2], PRIM_INT__s_p_s_s[(n+1) * 3 + 0], PRIM_INT__s_p_s_p[n * 9 + 2]);
227
228 PRIM_INT__s_p_s_p[n * 9 + 3] = SIMINT_MUL(Q_PB[0], PRIM_INT__s_p_s_s[n * 3 + 1]);
229 PRIM_INT__s_p_s_p[n * 9 + 3] = SIMINT_FMADD( aoq_PQ[0], PRIM_INT__s_p_s_s[(n+1) * 3 + 1], PRIM_INT__s_p_s_p[n * 9 + 3]);
230
231 PRIM_INT__s_p_s_p[n * 9 + 4] = SIMINT_MUL(Q_PB[1], PRIM_INT__s_p_s_s[n * 3 + 1]);
232 PRIM_INT__s_p_s_p[n * 9 + 4] = SIMINT_FMADD( aoq_PQ[1], PRIM_INT__s_p_s_s[(n+1) * 3 + 1], PRIM_INT__s_p_s_p[n * 9 + 4]);
233 PRIM_INT__s_p_s_p[n * 9 + 4] = SIMINT_FMADD( vrr_const_1_over_2pq, PRIM_INT__s_s_s_s[(n+1) * 1 + 0], PRIM_INT__s_p_s_p[n * 9 + 4]);
234
235 PRIM_INT__s_p_s_p[n * 9 + 5] = SIMINT_MUL(Q_PB[2], PRIM_INT__s_p_s_s[n * 3 + 1]);
236 PRIM_INT__s_p_s_p[n * 9 + 5] = SIMINT_FMADD( aoq_PQ[2], PRIM_INT__s_p_s_s[(n+1) * 3 + 1], PRIM_INT__s_p_s_p[n * 9 + 5]);
237
238 PRIM_INT__s_p_s_p[n * 9 + 6] = SIMINT_MUL(Q_PB[0], PRIM_INT__s_p_s_s[n * 3 + 2]);
239 PRIM_INT__s_p_s_p[n * 9 + 6] = SIMINT_FMADD( aoq_PQ[0], PRIM_INT__s_p_s_s[(n+1) * 3 + 2], PRIM_INT__s_p_s_p[n * 9 + 6]);
240
241 PRIM_INT__s_p_s_p[n * 9 + 7] = SIMINT_MUL(Q_PB[1], PRIM_INT__s_p_s_s[n * 3 + 2]);
242 PRIM_INT__s_p_s_p[n * 9 + 7] = SIMINT_FMADD( aoq_PQ[1], PRIM_INT__s_p_s_s[(n+1) * 3 + 2], PRIM_INT__s_p_s_p[n * 9 + 7]);
243
244 PRIM_INT__s_p_s_p[n * 9 + 8] = SIMINT_MUL(Q_PB[2], PRIM_INT__s_p_s_s[n * 3 + 2]);
245 PRIM_INT__s_p_s_p[n * 9 + 8] = SIMINT_FMADD( aoq_PQ[2], PRIM_INT__s_p_s_s[(n+1) * 3 + 2], PRIM_INT__s_p_s_p[n * 9 + 8]);
246 PRIM_INT__s_p_s_p[n * 9 + 8] = SIMINT_FMADD( vrr_const_1_over_2pq, PRIM_INT__s_s_s_s[(n+1) * 1 + 0], PRIM_INT__s_p_s_p[n * 9 + 8]);
247
248 }
249
250
251
252
253 ////////////////////////////////////
254 // Accumulate contracted integrals
255 ////////////////////////////////////
256 if(lastoffset == 0)
257 {
258 contract_all(9, PRIM_INT__s_p_s_p, PRIM_PTR_INT__s_p_s_p);
259 }
260 else
261 {
262 contract(9, shelloffsets, PRIM_INT__s_p_s_p, PRIM_PTR_INT__s_p_s_p);
263 PRIM_PTR_INT__s_p_s_p += lastoffset*9;
264 }
265
266 } // close loop over j
267 } // close loop over i
268
269 //Advance to the next batch
270 jstart = SIMINT_SIMD_ROUND(jend);
271 abcd += nshellbatch;
272
273 } // close loop cdbatch
274
275 istart = iend;
276 } // close loop over ab
277
278 return P.nshell12_clip * Q.nshell12_clip;
279 }
280
281