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_f_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_f_s_p)8 int ostei_s_f_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_f_s_p)
13 {
14
15 SIMINT_ASSUME_ALIGN_DBL(work);
16 SIMINT_ASSUME_ALIGN_DBL(INT__s_f_s_p);
17 memset(INT__s_f_s_p, 0, P.nshell12_clip * Q.nshell12_clip * 30 * 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 + 5;
31 SIMINT_DBLTYPE * const restrict PRIM_INT__s_d_s_s = primwork + 17;
32 SIMINT_DBLTYPE * const restrict PRIM_INT__s_f_s_s = primwork + 35;
33 SIMINT_DBLTYPE * const restrict PRIM_INT__s_f_s_p = primwork + 55;
34 double * const hrrwork = (double *)(primwork + 85);
35
36
37 // Create constants
38 const SIMINT_DBLTYPE const_1 = SIMINT_DBLSET1(1);
39 const SIMINT_DBLTYPE const_2 = SIMINT_DBLSET1(2);
40 const SIMINT_DBLTYPE const_3 = SIMINT_DBLSET1(3);
41 const SIMINT_DBLTYPE one_half = SIMINT_DBLSET1(0.5);
42
43
44 ////////////////////////////////////////
45 // Loop over shells and primitives
46 ////////////////////////////////////////
47
48 abcd = 0;
49 istart = 0;
50 for(ab = 0; ab < P.nshell12_clip; ++ab)
51 {
52 const int iend = istart + P.nprim12[ab];
53
54 cd = 0;
55 jstart = 0;
56
57 for(cd = 0; cd < Q.nshell12_clip; cd += SIMINT_NSHELL_SIMD)
58 {
59 const int nshellbatch = ((cd + SIMINT_NSHELL_SIMD) > Q.nshell12_clip) ? Q.nshell12_clip - cd : SIMINT_NSHELL_SIMD;
60 int jend = jstart;
61 for(i = 0; i < nshellbatch; i++)
62 jend += Q.nprim12[cd+i];
63
64
65 for(i = istart; i < iend; ++i)
66 {
67 SIMINT_DBLTYPE bra_screen_max; // only used if check_screen
68
69 if(check_screen)
70 {
71 // Skip this whole thing if always insignificant
72 if((P.screen[i] * Q.screen_max) < screen_tol)
73 continue;
74 bra_screen_max = SIMINT_DBLSET1(P.screen[i]);
75 }
76
77 icd = 0;
78 iprimcd = 0;
79 nprim_icd = Q.nprim12[cd];
80 double * restrict PRIM_PTR_INT__s_f_s_p = INT__s_f_s_p + abcd * 30;
81
82
83
84 // Load these one per loop over i
85 const SIMINT_DBLTYPE P_alpha = SIMINT_DBLSET1(P.alpha[i]);
86 const SIMINT_DBLTYPE P_prefac = SIMINT_DBLSET1(P.prefac[i]);
87 const SIMINT_DBLTYPE Pxyz[3] = { SIMINT_DBLSET1(P.x[i]), SIMINT_DBLSET1(P.y[i]), SIMINT_DBLSET1(P.z[i]) };
88
89 const SIMINT_DBLTYPE P_PB[3] = { SIMINT_DBLSET1(P.PB_x[i]), SIMINT_DBLSET1(P.PB_y[i]), SIMINT_DBLSET1(P.PB_z[i]) };
90
91 for(j = jstart; j < jend; j += SIMINT_SIMD_LEN)
92 {
93 // calculate the shell offsets
94 // these are the offset from the shell pointed to by cd
95 // for each element
96 int shelloffsets[SIMINT_SIMD_LEN] = {0};
97 int lastoffset = 0;
98 const int nlane = ( ((j + SIMINT_SIMD_LEN) < jend) ? SIMINT_SIMD_LEN : (jend - j));
99
100 if((iprimcd + SIMINT_SIMD_LEN) >= nprim_icd)
101 {
102 // Handle if the first element of the vector is a new shell
103 if(iprimcd >= nprim_icd && ((icd+1) < nshellbatch))
104 {
105 nprim_icd += Q.nprim12[cd + (++icd)];
106 PRIM_PTR_INT__s_f_s_p += 30;
107 }
108 iprimcd++;
109 for(n = 1; n < SIMINT_SIMD_LEN; ++n)
110 {
111 if(iprimcd >= nprim_icd && ((icd+1) < nshellbatch))
112 {
113 shelloffsets[n] = shelloffsets[n-1] + 1;
114 lastoffset++;
115 nprim_icd += Q.nprim12[cd + (++icd)];
116 }
117 else
118 shelloffsets[n] = shelloffsets[n-1];
119 iprimcd++;
120 }
121 }
122 else
123 iprimcd += SIMINT_SIMD_LEN;
124
125 // Do we have to compute this vector (or has it been screened out)?
126 // (not_screened != 0 means we have to do this vector)
127 if(check_screen)
128 {
129 const double vmax = vector_max(SIMINT_MUL(bra_screen_max, SIMINT_DBLLOAD(Q.screen, j)));
130 if(vmax < screen_tol)
131 {
132 PRIM_PTR_INT__s_f_s_p += lastoffset*30;
133 continue;
134 }
135 }
136
137 const SIMINT_DBLTYPE Q_alpha = SIMINT_DBLLOAD(Q.alpha, j);
138 const SIMINT_DBLTYPE PQalpha_mul = SIMINT_MUL(P_alpha, Q_alpha);
139 const SIMINT_DBLTYPE PQalpha_sum = SIMINT_ADD(P_alpha, Q_alpha);
140 const SIMINT_DBLTYPE one_over_PQalpha_sum = SIMINT_DIV(const_1, PQalpha_sum);
141
142
143 /* construct R2 = (Px - Qx)**2 + (Py - Qy)**2 + (Pz -Qz)**2 */
144 SIMINT_DBLTYPE PQ[3];
145 PQ[0] = SIMINT_SUB(Pxyz[0], SIMINT_DBLLOAD(Q.x, j));
146 PQ[1] = SIMINT_SUB(Pxyz[1], SIMINT_DBLLOAD(Q.y, j));
147 PQ[2] = SIMINT_SUB(Pxyz[2], SIMINT_DBLLOAD(Q.z, j));
148 SIMINT_DBLTYPE R2 = SIMINT_MUL(PQ[0], PQ[0]);
149 R2 = SIMINT_FMADD(PQ[1], PQ[1], R2);
150 R2 = SIMINT_FMADD(PQ[2], PQ[2], R2);
151
152 const SIMINT_DBLTYPE alpha = SIMINT_MUL(PQalpha_mul, one_over_PQalpha_sum); // alpha from MEST
153 const SIMINT_DBLTYPE one_over_p = SIMINT_DIV(const_1, P_alpha);
154 const SIMINT_DBLTYPE one_over_q = SIMINT_DIV(const_1, Q_alpha);
155 const SIMINT_DBLTYPE one_over_2p = SIMINT_MUL(one_half, one_over_p);
156 const SIMINT_DBLTYPE one_over_2q = SIMINT_MUL(one_half, one_over_q);
157 const SIMINT_DBLTYPE one_over_2pq = SIMINT_MUL(one_half, one_over_PQalpha_sum);
158 const SIMINT_DBLTYPE Q_PB[3] = { SIMINT_DBLLOAD(Q.PB_x, j), SIMINT_DBLLOAD(Q.PB_y, j), SIMINT_DBLLOAD(Q.PB_z, j) };
159
160 // NOTE: Minus sign!
161 const SIMINT_DBLTYPE a_over_p = SIMINT_MUL(SIMINT_NEG(alpha), one_over_p);
162 SIMINT_DBLTYPE aop_PQ[3];
163 aop_PQ[0] = SIMINT_MUL(a_over_p, PQ[0]);
164 aop_PQ[1] = SIMINT_MUL(a_over_p, PQ[1]);
165 aop_PQ[2] = SIMINT_MUL(a_over_p, PQ[2]);
166
167 SIMINT_DBLTYPE a_over_q = SIMINT_MUL(alpha, one_over_q);
168 SIMINT_DBLTYPE aoq_PQ[3];
169 aoq_PQ[0] = SIMINT_MUL(a_over_q, PQ[0]);
170 aoq_PQ[1] = SIMINT_MUL(a_over_q, PQ[1]);
171 aoq_PQ[2] = SIMINT_MUL(a_over_q, PQ[2]);
172 // Put a minus sign here so we don't have to in RR routines
173 a_over_q = SIMINT_NEG(a_over_q);
174
175
176 //////////////////////////////////////////////
177 // Fjt function section
178 // Maximum v value: 4
179 //////////////////////////////////////////////
180 // The parameter to the Fjt function
181 const SIMINT_DBLTYPE F_x = SIMINT_MUL(R2, alpha);
182
183
184 const SIMINT_DBLTYPE Q_prefac = mask_load(nlane, Q.prefac + j);
185
186
187 boys_F_split(PRIM_INT__s_s_s_s, F_x, 4);
188 SIMINT_DBLTYPE prefac = SIMINT_SQRT(one_over_PQalpha_sum);
189 prefac = SIMINT_MUL(SIMINT_MUL(P_prefac, Q_prefac), prefac);
190 for(n = 0; n <= 4; n++)
191 PRIM_INT__s_s_s_s[n] = SIMINT_MUL(PRIM_INT__s_s_s_s[n], prefac);
192
193 //////////////////////////////////////////////
194 // Primitive integrals: Vertical recurrance
195 //////////////////////////////////////////////
196
197 const SIMINT_DBLTYPE vrr_const_1_over_2p = one_over_2p;
198 const SIMINT_DBLTYPE vrr_const_2_over_2p = SIMINT_MUL(const_2, one_over_2p);
199 const SIMINT_DBLTYPE vrr_const_1_over_2pq = one_over_2pq;
200 const SIMINT_DBLTYPE vrr_const_2_over_2pq = SIMINT_MUL(const_2, one_over_2pq);
201 const SIMINT_DBLTYPE vrr_const_3_over_2pq = SIMINT_MUL(const_3, one_over_2pq);
202
203
204
205 // Forming PRIM_INT__s_p_s_s[4 * 3];
206 for(n = 0; n < 4; ++n) // loop over orders of auxiliary function
207 {
208
209 PRIM_INT__s_p_s_s[n * 3 + 0] = SIMINT_MUL(P_PB[0], PRIM_INT__s_s_s_s[n * 1 + 0]);
210 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]);
211
212 PRIM_INT__s_p_s_s[n * 3 + 1] = SIMINT_MUL(P_PB[1], PRIM_INT__s_s_s_s[n * 1 + 0]);
213 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]);
214
215 PRIM_INT__s_p_s_s[n * 3 + 2] = SIMINT_MUL(P_PB[2], PRIM_INT__s_s_s_s[n * 1 + 0]);
216 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]);
217
218 }
219
220
221
222 // Forming PRIM_INT__s_d_s_s[3 * 6];
223 for(n = 0; n < 3; ++n) // loop over orders of auxiliary function
224 {
225
226 PRIM_INT__s_d_s_s[n * 6 + 0] = SIMINT_MUL(P_PB[0], PRIM_INT__s_p_s_s[n * 3 + 0]);
227 PRIM_INT__s_d_s_s[n * 6 + 0] = SIMINT_FMADD( aop_PQ[0], PRIM_INT__s_p_s_s[(n+1) * 3 + 0], PRIM_INT__s_d_s_s[n * 6 + 0]);
228 PRIM_INT__s_d_s_s[n * 6 + 0] = SIMINT_FMADD( vrr_const_1_over_2p, SIMINT_FMADD(a_over_p, PRIM_INT__s_s_s_s[(n+1) * 1 + 0], PRIM_INT__s_s_s_s[n * 1 + 0]), PRIM_INT__s_d_s_s[n * 6 + 0]);
229
230 PRIM_INT__s_d_s_s[n * 6 + 1] = SIMINT_MUL(P_PB[1], PRIM_INT__s_p_s_s[n * 3 + 0]);
231 PRIM_INT__s_d_s_s[n * 6 + 1] = SIMINT_FMADD( aop_PQ[1], PRIM_INT__s_p_s_s[(n+1) * 3 + 0], PRIM_INT__s_d_s_s[n * 6 + 1]);
232
233 PRIM_INT__s_d_s_s[n * 6 + 2] = SIMINT_MUL(P_PB[2], PRIM_INT__s_p_s_s[n * 3 + 0]);
234 PRIM_INT__s_d_s_s[n * 6 + 2] = SIMINT_FMADD( aop_PQ[2], PRIM_INT__s_p_s_s[(n+1) * 3 + 0], PRIM_INT__s_d_s_s[n * 6 + 2]);
235
236 PRIM_INT__s_d_s_s[n * 6 + 3] = SIMINT_MUL(P_PB[1], PRIM_INT__s_p_s_s[n * 3 + 1]);
237 PRIM_INT__s_d_s_s[n * 6 + 3] = SIMINT_FMADD( aop_PQ[1], PRIM_INT__s_p_s_s[(n+1) * 3 + 1], PRIM_INT__s_d_s_s[n * 6 + 3]);
238 PRIM_INT__s_d_s_s[n * 6 + 3] = SIMINT_FMADD( vrr_const_1_over_2p, SIMINT_FMADD(a_over_p, PRIM_INT__s_s_s_s[(n+1) * 1 + 0], PRIM_INT__s_s_s_s[n * 1 + 0]), PRIM_INT__s_d_s_s[n * 6 + 3]);
239
240 PRIM_INT__s_d_s_s[n * 6 + 4] = SIMINT_MUL(P_PB[2], PRIM_INT__s_p_s_s[n * 3 + 1]);
241 PRIM_INT__s_d_s_s[n * 6 + 4] = SIMINT_FMADD( aop_PQ[2], PRIM_INT__s_p_s_s[(n+1) * 3 + 1], PRIM_INT__s_d_s_s[n * 6 + 4]);
242
243 PRIM_INT__s_d_s_s[n * 6 + 5] = SIMINT_MUL(P_PB[2], PRIM_INT__s_p_s_s[n * 3 + 2]);
244 PRIM_INT__s_d_s_s[n * 6 + 5] = SIMINT_FMADD( aop_PQ[2], PRIM_INT__s_p_s_s[(n+1) * 3 + 2], PRIM_INT__s_d_s_s[n * 6 + 5]);
245 PRIM_INT__s_d_s_s[n * 6 + 5] = SIMINT_FMADD( vrr_const_1_over_2p, SIMINT_FMADD(a_over_p, PRIM_INT__s_s_s_s[(n+1) * 1 + 0], PRIM_INT__s_s_s_s[n * 1 + 0]), PRIM_INT__s_d_s_s[n * 6 + 5]);
246
247 }
248
249
250
251 // Forming PRIM_INT__s_f_s_s[2 * 10];
252 for(n = 0; n < 2; ++n) // loop over orders of auxiliary function
253 {
254
255 PRIM_INT__s_f_s_s[n * 10 + 0] = SIMINT_MUL(P_PB[0], PRIM_INT__s_d_s_s[n * 6 + 0]);
256 PRIM_INT__s_f_s_s[n * 10 + 0] = SIMINT_FMADD( aop_PQ[0], PRIM_INT__s_d_s_s[(n+1) * 6 + 0], PRIM_INT__s_f_s_s[n * 10 + 0]);
257 PRIM_INT__s_f_s_s[n * 10 + 0] = SIMINT_FMADD( vrr_const_2_over_2p, SIMINT_FMADD(a_over_p, PRIM_INT__s_p_s_s[(n+1) * 3 + 0], PRIM_INT__s_p_s_s[n * 3 + 0]), PRIM_INT__s_f_s_s[n * 10 + 0]);
258
259 PRIM_INT__s_f_s_s[n * 10 + 1] = SIMINT_MUL(P_PB[1], PRIM_INT__s_d_s_s[n * 6 + 0]);
260 PRIM_INT__s_f_s_s[n * 10 + 1] = SIMINT_FMADD( aop_PQ[1], PRIM_INT__s_d_s_s[(n+1) * 6 + 0], PRIM_INT__s_f_s_s[n * 10 + 1]);
261
262 PRIM_INT__s_f_s_s[n * 10 + 2] = SIMINT_MUL(P_PB[2], PRIM_INT__s_d_s_s[n * 6 + 0]);
263 PRIM_INT__s_f_s_s[n * 10 + 2] = SIMINT_FMADD( aop_PQ[2], PRIM_INT__s_d_s_s[(n+1) * 6 + 0], PRIM_INT__s_f_s_s[n * 10 + 2]);
264
265 PRIM_INT__s_f_s_s[n * 10 + 3] = SIMINT_MUL(P_PB[0], PRIM_INT__s_d_s_s[n * 6 + 3]);
266 PRIM_INT__s_f_s_s[n * 10 + 3] = SIMINT_FMADD( aop_PQ[0], PRIM_INT__s_d_s_s[(n+1) * 6 + 3], PRIM_INT__s_f_s_s[n * 10 + 3]);
267
268 PRIM_INT__s_f_s_s[n * 10 + 4] = SIMINT_MUL(P_PB[2], PRIM_INT__s_d_s_s[n * 6 + 1]);
269 PRIM_INT__s_f_s_s[n * 10 + 4] = SIMINT_FMADD( aop_PQ[2], PRIM_INT__s_d_s_s[(n+1) * 6 + 1], PRIM_INT__s_f_s_s[n * 10 + 4]);
270
271 PRIM_INT__s_f_s_s[n * 10 + 5] = SIMINT_MUL(P_PB[0], PRIM_INT__s_d_s_s[n * 6 + 5]);
272 PRIM_INT__s_f_s_s[n * 10 + 5] = SIMINT_FMADD( aop_PQ[0], PRIM_INT__s_d_s_s[(n+1) * 6 + 5], PRIM_INT__s_f_s_s[n * 10 + 5]);
273
274 PRIM_INT__s_f_s_s[n * 10 + 6] = SIMINT_MUL(P_PB[1], PRIM_INT__s_d_s_s[n * 6 + 3]);
275 PRIM_INT__s_f_s_s[n * 10 + 6] = SIMINT_FMADD( aop_PQ[1], PRIM_INT__s_d_s_s[(n+1) * 6 + 3], PRIM_INT__s_f_s_s[n * 10 + 6]);
276 PRIM_INT__s_f_s_s[n * 10 + 6] = SIMINT_FMADD( vrr_const_2_over_2p, SIMINT_FMADD(a_over_p, PRIM_INT__s_p_s_s[(n+1) * 3 + 1], PRIM_INT__s_p_s_s[n * 3 + 1]), PRIM_INT__s_f_s_s[n * 10 + 6]);
277
278 PRIM_INT__s_f_s_s[n * 10 + 7] = SIMINT_MUL(P_PB[2], PRIM_INT__s_d_s_s[n * 6 + 3]);
279 PRIM_INT__s_f_s_s[n * 10 + 7] = SIMINT_FMADD( aop_PQ[2], PRIM_INT__s_d_s_s[(n+1) * 6 + 3], PRIM_INT__s_f_s_s[n * 10 + 7]);
280
281 PRIM_INT__s_f_s_s[n * 10 + 8] = SIMINT_MUL(P_PB[1], PRIM_INT__s_d_s_s[n * 6 + 5]);
282 PRIM_INT__s_f_s_s[n * 10 + 8] = SIMINT_FMADD( aop_PQ[1], PRIM_INT__s_d_s_s[(n+1) * 6 + 5], PRIM_INT__s_f_s_s[n * 10 + 8]);
283
284 PRIM_INT__s_f_s_s[n * 10 + 9] = SIMINT_MUL(P_PB[2], PRIM_INT__s_d_s_s[n * 6 + 5]);
285 PRIM_INT__s_f_s_s[n * 10 + 9] = SIMINT_FMADD( aop_PQ[2], PRIM_INT__s_d_s_s[(n+1) * 6 + 5], PRIM_INT__s_f_s_s[n * 10 + 9]);
286 PRIM_INT__s_f_s_s[n * 10 + 9] = SIMINT_FMADD( vrr_const_2_over_2p, SIMINT_FMADD(a_over_p, PRIM_INT__s_p_s_s[(n+1) * 3 + 2], PRIM_INT__s_p_s_s[n * 3 + 2]), PRIM_INT__s_f_s_s[n * 10 + 9]);
287
288 }
289
290
291 VRR_L_s_f_s_p(
292 PRIM_INT__s_f_s_p,
293 PRIM_INT__s_f_s_s,
294 PRIM_INT__s_d_s_s,
295 Q_PB,
296 aoq_PQ,
297 one_over_2pq,
298 1);
299
300
301
302
303 ////////////////////////////////////
304 // Accumulate contracted integrals
305 ////////////////////////////////////
306 if(lastoffset == 0)
307 {
308 contract_all(30, PRIM_INT__s_f_s_p, PRIM_PTR_INT__s_f_s_p);
309 }
310 else
311 {
312 contract(30, shelloffsets, PRIM_INT__s_f_s_p, PRIM_PTR_INT__s_f_s_p);
313 PRIM_PTR_INT__s_f_s_p += lastoffset*30;
314 }
315
316 } // close loop over j
317 } // close loop over i
318
319 //Advance to the next batch
320 jstart = SIMINT_SIMD_ROUND(jend);
321 abcd += nshellbatch;
322
323 } // close loop cdbatch
324
325 istart = iend;
326 } // close loop over ab
327
328 return P.nshell12_clip * Q.nshell12_clip;
329 }
330
331