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_p_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_p_f)8 int ostei_s_s_p_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_p_f)
13 {
14
15 SIMINT_ASSUME_ALIGN_DBL(work);
16 SIMINT_ASSUME_ALIGN_DBL(INT__s_s_p_f);
17 int ab, cd, abcd;
18 int istart, jstart;
19 int iprimcd, nprim_icd, icd;
20 const int check_screen = (screen_tol > 0.0);
21 int i, j;
22 int n;
23 int not_screened;
24 int real_abcd;
25 int ibra;
26
27 // partition workspace
28 double * const INT__s_s_s_f = work + (SIMINT_NSHELL_SIMD * 0);
29 double * const INT__s_s_s_g = work + (SIMINT_NSHELL_SIMD * 10);
30 SIMINT_DBLTYPE * const primwork = (SIMINT_DBLTYPE *)(work + SIMINT_NSHELL_SIMD*25);
31 SIMINT_DBLTYPE * const restrict PRIM_INT__s_s_s_s = primwork + 0;
32 SIMINT_DBLTYPE * const restrict PRIM_INT__s_s_s_p = primwork + 5;
33 SIMINT_DBLTYPE * const restrict PRIM_INT__s_s_s_d = primwork + 17;
34 SIMINT_DBLTYPE * const restrict PRIM_INT__s_s_s_f = primwork + 35;
35 SIMINT_DBLTYPE * const restrict PRIM_INT__s_s_s_g = primwork + 55;
36 double * const hrrwork = (double *)(primwork + 70);
37
38
39 // Create constants
40 const SIMINT_DBLTYPE const_1 = SIMINT_DBLSET1(1);
41 const SIMINT_DBLTYPE const_2 = SIMINT_DBLSET1(2);
42 const SIMINT_DBLTYPE const_3 = SIMINT_DBLSET1(3);
43 const SIMINT_DBLTYPE one_half = SIMINT_DBLSET1(0.5);
44
45
46 ////////////////////////////////////////
47 // Loop over shells and primitives
48 ////////////////////////////////////////
49
50 real_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 // Clear the beginning of the workspace (where we are accumulating integrals)
67 memset(work, 0, SIMINT_NSHELL_SIMD * 25 * sizeof(double));
68 abcd = 0;
69
70
71 for(i = istart; i < iend; ++i)
72 {
73 SIMINT_DBLTYPE bra_screen_max; // only used if check_screen
74
75 if(check_screen)
76 {
77 // Skip this whole thing if always insignificant
78 if((P.screen[i] * Q.screen_max) < screen_tol)
79 continue;
80 bra_screen_max = SIMINT_DBLSET1(P.screen[i]);
81 }
82
83 icd = 0;
84 iprimcd = 0;
85 nprim_icd = Q.nprim12[cd];
86 double * restrict PRIM_PTR_INT__s_s_s_f = INT__s_s_s_f + abcd * 10;
87 double * restrict PRIM_PTR_INT__s_s_s_g = INT__s_s_s_g + abcd * 15;
88
89
90
91 // Load these one per loop over i
92 const SIMINT_DBLTYPE P_alpha = SIMINT_DBLSET1(P.alpha[i]);
93 const SIMINT_DBLTYPE P_prefac = SIMINT_DBLSET1(P.prefac[i]);
94 const SIMINT_DBLTYPE Pxyz[3] = { SIMINT_DBLSET1(P.x[i]), SIMINT_DBLSET1(P.y[i]), SIMINT_DBLSET1(P.z[i]) };
95
96
97 for(j = jstart; j < jend; j += SIMINT_SIMD_LEN)
98 {
99 // calculate the shell offsets
100 // these are the offset from the shell pointed to by cd
101 // for each element
102 int shelloffsets[SIMINT_SIMD_LEN] = {0};
103 int lastoffset = 0;
104 const int nlane = ( ((j + SIMINT_SIMD_LEN) < jend) ? SIMINT_SIMD_LEN : (jend - j));
105
106 if((iprimcd + SIMINT_SIMD_LEN) >= nprim_icd)
107 {
108 // Handle if the first element of the vector is a new shell
109 if(iprimcd >= nprim_icd && ((icd+1) < nshellbatch))
110 {
111 nprim_icd += Q.nprim12[cd + (++icd)];
112 PRIM_PTR_INT__s_s_s_f += 10;
113 PRIM_PTR_INT__s_s_s_g += 15;
114 }
115 iprimcd++;
116 for(n = 1; n < SIMINT_SIMD_LEN; ++n)
117 {
118 if(iprimcd >= nprim_icd && ((icd+1) < nshellbatch))
119 {
120 shelloffsets[n] = shelloffsets[n-1] + 1;
121 lastoffset++;
122 nprim_icd += Q.nprim12[cd + (++icd)];
123 }
124 else
125 shelloffsets[n] = shelloffsets[n-1];
126 iprimcd++;
127 }
128 }
129 else
130 iprimcd += SIMINT_SIMD_LEN;
131
132 // Do we have to compute this vector (or has it been screened out)?
133 // (not_screened != 0 means we have to do this vector)
134 if(check_screen)
135 {
136 const double vmax = vector_max(SIMINT_MUL(bra_screen_max, SIMINT_DBLLOAD(Q.screen, j)));
137 if(vmax < screen_tol)
138 {
139 PRIM_PTR_INT__s_s_s_f += lastoffset*10;
140 PRIM_PTR_INT__s_s_s_g += lastoffset*15;
141 continue;
142 }
143 }
144
145 const SIMINT_DBLTYPE Q_alpha = SIMINT_DBLLOAD(Q.alpha, j);
146 const SIMINT_DBLTYPE PQalpha_mul = SIMINT_MUL(P_alpha, Q_alpha);
147 const SIMINT_DBLTYPE PQalpha_sum = SIMINT_ADD(P_alpha, Q_alpha);
148 const SIMINT_DBLTYPE one_over_PQalpha_sum = SIMINT_DIV(const_1, PQalpha_sum);
149
150
151 /* construct R2 = (Px - Qx)**2 + (Py - Qy)**2 + (Pz -Qz)**2 */
152 SIMINT_DBLTYPE PQ[3];
153 PQ[0] = SIMINT_SUB(Pxyz[0], SIMINT_DBLLOAD(Q.x, j));
154 PQ[1] = SIMINT_SUB(Pxyz[1], SIMINT_DBLLOAD(Q.y, j));
155 PQ[2] = SIMINT_SUB(Pxyz[2], SIMINT_DBLLOAD(Q.z, j));
156 SIMINT_DBLTYPE R2 = SIMINT_MUL(PQ[0], PQ[0]);
157 R2 = SIMINT_FMADD(PQ[1], PQ[1], R2);
158 R2 = SIMINT_FMADD(PQ[2], PQ[2], R2);
159
160 const SIMINT_DBLTYPE alpha = SIMINT_MUL(PQalpha_mul, one_over_PQalpha_sum); // alpha from MEST
161 const SIMINT_DBLTYPE one_over_p = SIMINT_DIV(const_1, P_alpha);
162 const SIMINT_DBLTYPE one_over_q = SIMINT_DIV(const_1, Q_alpha);
163 const SIMINT_DBLTYPE one_over_2p = SIMINT_MUL(one_half, one_over_p);
164 const SIMINT_DBLTYPE one_over_2q = SIMINT_MUL(one_half, one_over_q);
165 const SIMINT_DBLTYPE one_over_2pq = SIMINT_MUL(one_half, one_over_PQalpha_sum);
166 const SIMINT_DBLTYPE Q_PB[3] = { SIMINT_DBLLOAD(Q.PB_x, j), SIMINT_DBLLOAD(Q.PB_y, j), SIMINT_DBLLOAD(Q.PB_z, j) };
167
168 SIMINT_DBLTYPE a_over_q = SIMINT_MUL(alpha, one_over_q);
169 SIMINT_DBLTYPE aoq_PQ[3];
170 aoq_PQ[0] = SIMINT_MUL(a_over_q, PQ[0]);
171 aoq_PQ[1] = SIMINT_MUL(a_over_q, PQ[1]);
172 aoq_PQ[2] = SIMINT_MUL(a_over_q, PQ[2]);
173 // Put a minus sign here so we don't have to in RR routines
174 a_over_q = SIMINT_NEG(a_over_q);
175
176
177 //////////////////////////////////////////////
178 // Fjt function section
179 // Maximum v value: 4
180 //////////////////////////////////////////////
181 // The parameter to the Fjt function
182 const SIMINT_DBLTYPE F_x = SIMINT_MUL(R2, alpha);
183
184
185 const SIMINT_DBLTYPE Q_prefac = mask_load(nlane, Q.prefac + j);
186
187
188 boys_F_split(PRIM_INT__s_s_s_s, F_x, 4);
189 SIMINT_DBLTYPE prefac = SIMINT_SQRT(one_over_PQalpha_sum);
190 prefac = SIMINT_MUL(SIMINT_MUL(P_prefac, Q_prefac), prefac);
191 for(n = 0; n <= 4; n++)
192 PRIM_INT__s_s_s_s[n] = SIMINT_MUL(PRIM_INT__s_s_s_s[n], prefac);
193
194 //////////////////////////////////////////////
195 // Primitive integrals: Vertical recurrance
196 //////////////////////////////////////////////
197
198 const SIMINT_DBLTYPE vrr_const_1_over_2q = one_over_2q;
199 const SIMINT_DBLTYPE vrr_const_2_over_2q = SIMINT_MUL(const_2, one_over_2q);
200 const SIMINT_DBLTYPE vrr_const_3_over_2q = SIMINT_MUL(const_3, one_over_2q);
201
202
203
204 // Forming PRIM_INT__s_s_s_p[4 * 3];
205 for(n = 0; n < 4; ++n) // loop over orders of auxiliary function
206 {
207
208 PRIM_INT__s_s_s_p[n * 3 + 0] = SIMINT_MUL(Q_PB[0], PRIM_INT__s_s_s_s[n * 1 + 0]);
209 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]);
210
211 PRIM_INT__s_s_s_p[n * 3 + 1] = SIMINT_MUL(Q_PB[1], PRIM_INT__s_s_s_s[n * 1 + 0]);
212 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]);
213
214 PRIM_INT__s_s_s_p[n * 3 + 2] = SIMINT_MUL(Q_PB[2], PRIM_INT__s_s_s_s[n * 1 + 0]);
215 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]);
216
217 }
218
219
220
221 // Forming PRIM_INT__s_s_s_d[3 * 6];
222 for(n = 0; n < 3; ++n) // loop over orders of auxiliary function
223 {
224
225 PRIM_INT__s_s_s_d[n * 6 + 0] = SIMINT_MUL(Q_PB[0], PRIM_INT__s_s_s_p[n * 3 + 0]);
226 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]);
227 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]);
228
229 PRIM_INT__s_s_s_d[n * 6 + 1] = SIMINT_MUL(Q_PB[1], PRIM_INT__s_s_s_p[n * 3 + 0]);
230 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]);
231
232 PRIM_INT__s_s_s_d[n * 6 + 3] = SIMINT_MUL(Q_PB[1], PRIM_INT__s_s_s_p[n * 3 + 1]);
233 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]);
234 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]);
235
236 PRIM_INT__s_s_s_d[n * 6 + 5] = SIMINT_MUL(Q_PB[2], PRIM_INT__s_s_s_p[n * 3 + 2]);
237 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]);
238 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]);
239
240 }
241
242
243
244 // Forming PRIM_INT__s_s_s_f[2 * 10];
245 for(n = 0; n < 2; ++n) // loop over orders of auxiliary function
246 {
247
248 PRIM_INT__s_s_s_f[n * 10 + 0] = SIMINT_MUL(Q_PB[0], PRIM_INT__s_s_s_d[n * 6 + 0]);
249 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]);
250 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]);
251
252 PRIM_INT__s_s_s_f[n * 10 + 1] = SIMINT_MUL(Q_PB[1], PRIM_INT__s_s_s_d[n * 6 + 0]);
253 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]);
254
255 PRIM_INT__s_s_s_f[n * 10 + 2] = SIMINT_MUL(Q_PB[2], PRIM_INT__s_s_s_d[n * 6 + 0]);
256 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]);
257
258 PRIM_INT__s_s_s_f[n * 10 + 3] = SIMINT_MUL(Q_PB[0], PRIM_INT__s_s_s_d[n * 6 + 3]);
259 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]);
260
261 PRIM_INT__s_s_s_f[n * 10 + 4] = SIMINT_MUL(Q_PB[2], PRIM_INT__s_s_s_d[n * 6 + 1]);
262 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]);
263
264 PRIM_INT__s_s_s_f[n * 10 + 5] = SIMINT_MUL(Q_PB[0], PRIM_INT__s_s_s_d[n * 6 + 5]);
265 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]);
266
267 PRIM_INT__s_s_s_f[n * 10 + 6] = SIMINT_MUL(Q_PB[1], PRIM_INT__s_s_s_d[n * 6 + 3]);
268 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]);
269 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]);
270
271 PRIM_INT__s_s_s_f[n * 10 + 7] = SIMINT_MUL(Q_PB[2], PRIM_INT__s_s_s_d[n * 6 + 3]);
272 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]);
273
274 PRIM_INT__s_s_s_f[n * 10 + 8] = SIMINT_MUL(Q_PB[1], PRIM_INT__s_s_s_d[n * 6 + 5]);
275 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]);
276
277 PRIM_INT__s_s_s_f[n * 10 + 9] = SIMINT_MUL(Q_PB[2], PRIM_INT__s_s_s_d[n * 6 + 5]);
278 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]);
279 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]);
280
281 }
282
283
284 VRR_L_s_s_s_g(
285 PRIM_INT__s_s_s_g,
286 PRIM_INT__s_s_s_f,
287 PRIM_INT__s_s_s_d,
288 Q_PB,
289 a_over_q,
290 aoq_PQ,
291 one_over_2q,
292 1);
293
294
295
296
297 ////////////////////////////////////
298 // Accumulate contracted integrals
299 ////////////////////////////////////
300 if(lastoffset == 0)
301 {
302 contract_all(10, PRIM_INT__s_s_s_f, PRIM_PTR_INT__s_s_s_f);
303 contract_all(15, PRIM_INT__s_s_s_g, PRIM_PTR_INT__s_s_s_g);
304 }
305 else
306 {
307 contract(10, shelloffsets, PRIM_INT__s_s_s_f, PRIM_PTR_INT__s_s_s_f);
308 contract(15, shelloffsets, PRIM_INT__s_s_s_g, PRIM_PTR_INT__s_s_s_g);
309 PRIM_PTR_INT__s_s_s_f += lastoffset*10;
310 PRIM_PTR_INT__s_s_s_g += lastoffset*15;
311 }
312
313 } // close loop over j
314 } // close loop over i
315
316 //Advance to the next batch
317 jstart = SIMINT_SIMD_ROUND(jend);
318
319 //////////////////////////////////////////////
320 // Contracted integrals: Horizontal recurrance
321 //////////////////////////////////////////////
322
323
324
325
326 for(abcd = 0; abcd < nshellbatch; ++abcd, ++real_abcd)
327 {
328 const double hCD[3] = { Q.AB_x[cd+abcd], Q.AB_y[cd+abcd], Q.AB_z[cd+abcd] };
329
330 // set up HRR pointers
331 double const * restrict HRR_INT__s_s_s_f = INT__s_s_s_f + abcd * 10;
332 double const * restrict HRR_INT__s_s_s_g = INT__s_s_s_g + abcd * 15;
333 double * restrict HRR_INT__s_s_p_f = INT__s_s_p_f + real_abcd * 30;
334
335 // form INT__s_s_p_f
336 HRR_K_p_f(
337 HRR_INT__s_s_p_f,
338 HRR_INT__s_s_s_f,
339 HRR_INT__s_s_s_g,
340 hCD, 1);
341
342
343 } // close HRR loop
344
345
346 } // close loop cdbatch
347
348 istart = iend;
349 } // close loop over ab
350
351 return P.nshell12_clip * Q.nshell12_clip;
352 }
353
354