?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Removed simple.h from nb_kernel_sse2_XX
[gromacs.git] / src / gromacs / gmxlib / nonbonded / nb_kernel_sse2_double / nb_kernel_ElecCSTab_VdwNone_GeomW4P1_sse2_double.c
blob2629518e5b69ddfc0d4d7cc926fc8fe92aa1b04d
1 /*
2 * This file is part of the GROMACS molecular simulation package.
3 *
4 * Copyright (c) 2012,2013,2014,2015, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
8 *
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
13 *
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
18 *
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
23 *
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
31 *
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
34 */
35 /*
36 * Note: this file was generated by the GROMACS sse2_double kernel generator.
37 */
38 #include "gmxpre.h"
39
40 #include "config.h"
41
42 #include <math.h>
43
44 #include "../nb_kernel.h"
45 #include "gromacs/math/vec.h"
46 #include "gromacs/legacyheaders/nrnb.h"
47
48 #include "gromacs/simd/math_x86_sse2_double.h"
49 #include "kernelutil_x86_sse2_double.h"
50
51 /*
52 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwNone_GeomW4P1_VF_sse2_double
53 * Electrostatics interaction: CubicSplineTable
54 * VdW interaction: None
55 * Geometry: Water4-Particle
56 * Calculate force/pot: PotentialAndForce
57 */
58 void
59 nb_kernel_ElecCSTab_VdwNone_GeomW4P1_VF_sse2_double
60 (t_nblist * gmx_restrict nlist,
61 rvec * gmx_restrict xx,
62 rvec * gmx_restrict ff,
63 t_forcerec * gmx_restrict fr,
64 t_mdatoms * gmx_restrict mdatoms,
65 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
66 t_nrnb * gmx_restrict nrnb)
67 {
68 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
69 * just 0 for non-waters.
70 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
71 * jnr indices corresponding to data put in the four positions in the SIMD register.
72 */
73 int i_shift_offset,i_coord_offset,outeriter,inneriter;
74 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
75 int jnrA,jnrB;
76 int j_coord_offsetA,j_coord_offsetB;
77 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
78 real rcutoff_scalar;
79 real *shiftvec,*fshift,*x,*f;
80 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
81 int vdwioffset1;
82 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
83 int vdwioffset2;
84 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
85 int vdwioffset3;
86 __m128d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
87 int vdwjidx0A,vdwjidx0B;
88 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
89 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
90 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
91 __m128d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
92 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
93 real *charge;
94 __m128i vfitab;
95 __m128i ifour = _mm_set1_epi32(4);
96 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
97 real *vftab;
98 __m128d dummy_mask,cutoff_mask;
99 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
100 __m128d one = _mm_set1_pd(1.0);
101 __m128d two = _mm_set1_pd(2.0);
102 x = xx[0];
103 f = ff[0];
104
105 nri = nlist->nri;
106 iinr = nlist->iinr;
107 jindex = nlist->jindex;
108 jjnr = nlist->jjnr;
109 shiftidx = nlist->shift;
110 gid = nlist->gid;
111 shiftvec = fr->shift_vec[0];
112 fshift = fr->fshift[0];
113 facel = _mm_set1_pd(fr->epsfac);
114 charge = mdatoms->chargeA;
115
116 vftab = kernel_data->table_elec->data;
117 vftabscale = _mm_set1_pd(kernel_data->table_elec->scale);
118
119 /* Setup water-specific parameters */
120 inr = nlist->iinr[0];
121 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
122 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
123 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
124
125 /* Avoid stupid compiler warnings */
126 jnrA = jnrB = 0;
127 j_coord_offsetA = 0;
128 j_coord_offsetB = 0;
129
130 outeriter = 0;
131 inneriter = 0;
132
133 /* Start outer loop over neighborlists */
134 for(iidx=0; iidx<nri; iidx++)
135 {
136 /* Load shift vector for this list */
137 i_shift_offset = DIM*shiftidx[iidx];
138
139 /* Load limits for loop over neighbors */
140 j_index_start = jindex[iidx];
141 j_index_end = jindex[iidx+1];
142
143 /* Get outer coordinate index */
144 inr = iinr[iidx];
145 i_coord_offset = DIM*inr;
146
147 /* Load i particle coords and add shift vector */
148 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
149 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
150
151 fix1 = _mm_setzero_pd();
152 fiy1 = _mm_setzero_pd();
153 fiz1 = _mm_setzero_pd();
154 fix2 = _mm_setzero_pd();
155 fiy2 = _mm_setzero_pd();
156 fiz2 = _mm_setzero_pd();
157 fix3 = _mm_setzero_pd();
158 fiy3 = _mm_setzero_pd();
159 fiz3 = _mm_setzero_pd();
160
161 /* Reset potential sums */
162 velecsum = _mm_setzero_pd();
163
164 /* Start inner kernel loop */
165 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
166 {
167
168 /* Get j neighbor index, and coordinate index */
169 jnrA = jjnr[jidx];
170 jnrB = jjnr[jidx+1];
171 j_coord_offsetA = DIM*jnrA;
172 j_coord_offsetB = DIM*jnrB;
173
174 /* load j atom coordinates */
175 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
176 &jx0,&jy0,&jz0);
177
178 /* Calculate displacement vector */
179 dx10 = _mm_sub_pd(ix1,jx0);
180 dy10 = _mm_sub_pd(iy1,jy0);
181 dz10 = _mm_sub_pd(iz1,jz0);
182 dx20 = _mm_sub_pd(ix2,jx0);
183 dy20 = _mm_sub_pd(iy2,jy0);
184 dz20 = _mm_sub_pd(iz2,jz0);
185 dx30 = _mm_sub_pd(ix3,jx0);
186 dy30 = _mm_sub_pd(iy3,jy0);
187 dz30 = _mm_sub_pd(iz3,jz0);
188
189 /* Calculate squared distance and things based on it */
190 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
191 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
192 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
193
194 rinv10 = gmx_mm_invsqrt_pd(rsq10);
195 rinv20 = gmx_mm_invsqrt_pd(rsq20);
196 rinv30 = gmx_mm_invsqrt_pd(rsq30);
197
198 /* Load parameters for j particles */
199 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
200
201 fjx0 = _mm_setzero_pd();
202 fjy0 = _mm_setzero_pd();
203 fjz0 = _mm_setzero_pd();
204
205 /**************************
206 * CALCULATE INTERACTIONS *
207 **************************/
208
209 r10 = _mm_mul_pd(rsq10,rinv10);
210
211 /* Compute parameters for interactions between i and j atoms */
212 qq10 = _mm_mul_pd(iq1,jq0);
213
214 /* Calculate table index by multiplying r with table scale and truncate to integer */
215 rt = _mm_mul_pd(r10,vftabscale);
216 vfitab = _mm_cvttpd_epi32(rt);
217 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
218 vfitab = _mm_slli_epi32(vfitab,2);
219
220 /* CUBIC SPLINE TABLE ELECTROSTATICS */
221 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
222 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
223 GMX_MM_TRANSPOSE2_PD(Y,F);
224 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
225 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
226 GMX_MM_TRANSPOSE2_PD(G,H);
227 Heps = _mm_mul_pd(vfeps,H);
228 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
229 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
230 velec = _mm_mul_pd(qq10,VV);
231 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
232 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
233
234 /* Update potential sum for this i atom from the interaction with this j atom. */
235 velecsum = _mm_add_pd(velecsum,velec);
236
237 fscal = felec;
238
239 /* Calculate temporary vectorial force */
240 tx = _mm_mul_pd(fscal,dx10);
241 ty = _mm_mul_pd(fscal,dy10);
242 tz = _mm_mul_pd(fscal,dz10);
243
244 /* Update vectorial force */
245 fix1 = _mm_add_pd(fix1,tx);
246 fiy1 = _mm_add_pd(fiy1,ty);
247 fiz1 = _mm_add_pd(fiz1,tz);
248
249 fjx0 = _mm_add_pd(fjx0,tx);
250 fjy0 = _mm_add_pd(fjy0,ty);
251 fjz0 = _mm_add_pd(fjz0,tz);
252
253 /**************************
254 * CALCULATE INTERACTIONS *
255 **************************/
256
257 r20 = _mm_mul_pd(rsq20,rinv20);
258
259 /* Compute parameters for interactions between i and j atoms */
260 qq20 = _mm_mul_pd(iq2,jq0);
261
262 /* Calculate table index by multiplying r with table scale and truncate to integer */
263 rt = _mm_mul_pd(r20,vftabscale);
264 vfitab = _mm_cvttpd_epi32(rt);
265 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
266 vfitab = _mm_slli_epi32(vfitab,2);
267
268 /* CUBIC SPLINE TABLE ELECTROSTATICS */
269 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
270 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
271 GMX_MM_TRANSPOSE2_PD(Y,F);
272 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
273 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
274 GMX_MM_TRANSPOSE2_PD(G,H);
275 Heps = _mm_mul_pd(vfeps,H);
276 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
277 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
278 velec = _mm_mul_pd(qq20,VV);
279 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
280 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
281
282 /* Update potential sum for this i atom from the interaction with this j atom. */
283 velecsum = _mm_add_pd(velecsum,velec);
284
285 fscal = felec;
286
287 /* Calculate temporary vectorial force */
288 tx = _mm_mul_pd(fscal,dx20);
289 ty = _mm_mul_pd(fscal,dy20);
290 tz = _mm_mul_pd(fscal,dz20);
291
292 /* Update vectorial force */
293 fix2 = _mm_add_pd(fix2,tx);
294 fiy2 = _mm_add_pd(fiy2,ty);
295 fiz2 = _mm_add_pd(fiz2,tz);
296
297 fjx0 = _mm_add_pd(fjx0,tx);
298 fjy0 = _mm_add_pd(fjy0,ty);
299 fjz0 = _mm_add_pd(fjz0,tz);
300
301 /**************************
302 * CALCULATE INTERACTIONS *
303 **************************/
304
305 r30 = _mm_mul_pd(rsq30,rinv30);
306
307 /* Compute parameters for interactions between i and j atoms */
308 qq30 = _mm_mul_pd(iq3,jq0);
309
310 /* Calculate table index by multiplying r with table scale and truncate to integer */
311 rt = _mm_mul_pd(r30,vftabscale);
312 vfitab = _mm_cvttpd_epi32(rt);
313 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
314 vfitab = _mm_slli_epi32(vfitab,2);
315
316 /* CUBIC SPLINE TABLE ELECTROSTATICS */
317 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
318 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
319 GMX_MM_TRANSPOSE2_PD(Y,F);
320 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
321 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
322 GMX_MM_TRANSPOSE2_PD(G,H);
323 Heps = _mm_mul_pd(vfeps,H);
324 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
325 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
326 velec = _mm_mul_pd(qq30,VV);
327 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
328 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
329
330 /* Update potential sum for this i atom from the interaction with this j atom. */
331 velecsum = _mm_add_pd(velecsum,velec);
332
333 fscal = felec;
334
335 /* Calculate temporary vectorial force */
336 tx = _mm_mul_pd(fscal,dx30);
337 ty = _mm_mul_pd(fscal,dy30);
338 tz = _mm_mul_pd(fscal,dz30);
339
340 /* Update vectorial force */
341 fix3 = _mm_add_pd(fix3,tx);
342 fiy3 = _mm_add_pd(fiy3,ty);
343 fiz3 = _mm_add_pd(fiz3,tz);
344
345 fjx0 = _mm_add_pd(fjx0,tx);
346 fjy0 = _mm_add_pd(fjy0,ty);
347 fjz0 = _mm_add_pd(fjz0,tz);
348
349 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
350
351 /* Inner loop uses 132 flops */
352 }
353
354 if(jidx<j_index_end)
355 {
356
357 jnrA = jjnr[jidx];
358 j_coord_offsetA = DIM*jnrA;
359
360 /* load j atom coordinates */
361 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
362 &jx0,&jy0,&jz0);
363
364 /* Calculate displacement vector */
365 dx10 = _mm_sub_pd(ix1,jx0);
366 dy10 = _mm_sub_pd(iy1,jy0);
367 dz10 = _mm_sub_pd(iz1,jz0);
368 dx20 = _mm_sub_pd(ix2,jx0);
369 dy20 = _mm_sub_pd(iy2,jy0);
370 dz20 = _mm_sub_pd(iz2,jz0);
371 dx30 = _mm_sub_pd(ix3,jx0);
372 dy30 = _mm_sub_pd(iy3,jy0);
373 dz30 = _mm_sub_pd(iz3,jz0);
374
375 /* Calculate squared distance and things based on it */
376 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
377 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
378 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
379
380 rinv10 = gmx_mm_invsqrt_pd(rsq10);
381 rinv20 = gmx_mm_invsqrt_pd(rsq20);
382 rinv30 = gmx_mm_invsqrt_pd(rsq30);
383
384 /* Load parameters for j particles */
385 jq0 = _mm_load_sd(charge+jnrA+0);
386
387 fjx0 = _mm_setzero_pd();
388 fjy0 = _mm_setzero_pd();
389 fjz0 = _mm_setzero_pd();
390
391 /**************************
392 * CALCULATE INTERACTIONS *
393 **************************/
394
395 r10 = _mm_mul_pd(rsq10,rinv10);
396
397 /* Compute parameters for interactions between i and j atoms */
398 qq10 = _mm_mul_pd(iq1,jq0);
399
400 /* Calculate table index by multiplying r with table scale and truncate to integer */
401 rt = _mm_mul_pd(r10,vftabscale);
402 vfitab = _mm_cvttpd_epi32(rt);
403 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
404 vfitab = _mm_slli_epi32(vfitab,2);
405
406 /* CUBIC SPLINE TABLE ELECTROSTATICS */
407 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
408 F = _mm_setzero_pd();
409 GMX_MM_TRANSPOSE2_PD(Y,F);
410 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
411 H = _mm_setzero_pd();
412 GMX_MM_TRANSPOSE2_PD(G,H);
413 Heps = _mm_mul_pd(vfeps,H);
414 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
415 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
416 velec = _mm_mul_pd(qq10,VV);
417 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
418 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
419
420 /* Update potential sum for this i atom from the interaction with this j atom. */
421 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
422 velecsum = _mm_add_pd(velecsum,velec);
423
424 fscal = felec;
425
426 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
427
428 /* Calculate temporary vectorial force */
429 tx = _mm_mul_pd(fscal,dx10);
430 ty = _mm_mul_pd(fscal,dy10);
431 tz = _mm_mul_pd(fscal,dz10);
432
433 /* Update vectorial force */
434 fix1 = _mm_add_pd(fix1,tx);
435 fiy1 = _mm_add_pd(fiy1,ty);
436 fiz1 = _mm_add_pd(fiz1,tz);
437
438 fjx0 = _mm_add_pd(fjx0,tx);
439 fjy0 = _mm_add_pd(fjy0,ty);
440 fjz0 = _mm_add_pd(fjz0,tz);
441
442 /**************************
443 * CALCULATE INTERACTIONS *
444 **************************/
445
446 r20 = _mm_mul_pd(rsq20,rinv20);
447
448 /* Compute parameters for interactions between i and j atoms */
449 qq20 = _mm_mul_pd(iq2,jq0);
450
451 /* Calculate table index by multiplying r with table scale and truncate to integer */
452 rt = _mm_mul_pd(r20,vftabscale);
453 vfitab = _mm_cvttpd_epi32(rt);
454 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
455 vfitab = _mm_slli_epi32(vfitab,2);
456
457 /* CUBIC SPLINE TABLE ELECTROSTATICS */
458 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
459 F = _mm_setzero_pd();
460 GMX_MM_TRANSPOSE2_PD(Y,F);
461 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
462 H = _mm_setzero_pd();
463 GMX_MM_TRANSPOSE2_PD(G,H);
464 Heps = _mm_mul_pd(vfeps,H);
465 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
466 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
467 velec = _mm_mul_pd(qq20,VV);
468 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
469 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
470
471 /* Update potential sum for this i atom from the interaction with this j atom. */
472 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
473 velecsum = _mm_add_pd(velecsum,velec);
474
475 fscal = felec;
476
477 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
478
479 /* Calculate temporary vectorial force */
480 tx = _mm_mul_pd(fscal,dx20);
481 ty = _mm_mul_pd(fscal,dy20);
482 tz = _mm_mul_pd(fscal,dz20);
483
484 /* Update vectorial force */
485 fix2 = _mm_add_pd(fix2,tx);
486 fiy2 = _mm_add_pd(fiy2,ty);
487 fiz2 = _mm_add_pd(fiz2,tz);
488
489 fjx0 = _mm_add_pd(fjx0,tx);
490 fjy0 = _mm_add_pd(fjy0,ty);
491 fjz0 = _mm_add_pd(fjz0,tz);
492
493 /**************************
494 * CALCULATE INTERACTIONS *
495 **************************/
496
497 r30 = _mm_mul_pd(rsq30,rinv30);
498
499 /* Compute parameters for interactions between i and j atoms */
500 qq30 = _mm_mul_pd(iq3,jq0);
501
502 /* Calculate table index by multiplying r with table scale and truncate to integer */
503 rt = _mm_mul_pd(r30,vftabscale);
504 vfitab = _mm_cvttpd_epi32(rt);
505 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
506 vfitab = _mm_slli_epi32(vfitab,2);
507
508 /* CUBIC SPLINE TABLE ELECTROSTATICS */
509 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
510 F = _mm_setzero_pd();
511 GMX_MM_TRANSPOSE2_PD(Y,F);
512 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
513 H = _mm_setzero_pd();
514 GMX_MM_TRANSPOSE2_PD(G,H);
515 Heps = _mm_mul_pd(vfeps,H);
516 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
517 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
518 velec = _mm_mul_pd(qq30,VV);
519 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
520 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
521
522 /* Update potential sum for this i atom from the interaction with this j atom. */
523 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
524 velecsum = _mm_add_pd(velecsum,velec);
525
526 fscal = felec;
527
528 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
529
530 /* Calculate temporary vectorial force */
531 tx = _mm_mul_pd(fscal,dx30);
532 ty = _mm_mul_pd(fscal,dy30);
533 tz = _mm_mul_pd(fscal,dz30);
534
535 /* Update vectorial force */
536 fix3 = _mm_add_pd(fix3,tx);
537 fiy3 = _mm_add_pd(fiy3,ty);
538 fiz3 = _mm_add_pd(fiz3,tz);
539
540 fjx0 = _mm_add_pd(fjx0,tx);
541 fjy0 = _mm_add_pd(fjy0,ty);
542 fjz0 = _mm_add_pd(fjz0,tz);
543
544 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
545
546 /* Inner loop uses 132 flops */
547 }
548
549 /* End of innermost loop */
550
551 gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
552 f+i_coord_offset+DIM,fshift+i_shift_offset);
553
554 ggid = gid[iidx];
555 /* Update potential energies */
556 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
557
558 /* Increment number of inner iterations */
559 inneriter += j_index_end - j_index_start;
560
561 /* Outer loop uses 19 flops */
562 }
563
564 /* Increment number of outer iterations */
565 outeriter += nri;
566
567 /* Update outer/inner flops */
568
569 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*132);
570 }
571 /*
572 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwNone_GeomW4P1_F_sse2_double
573 * Electrostatics interaction: CubicSplineTable
574 * VdW interaction: None
575 * Geometry: Water4-Particle
576 * Calculate force/pot: Force
577 */
578 void
579 nb_kernel_ElecCSTab_VdwNone_GeomW4P1_F_sse2_double
580 (t_nblist * gmx_restrict nlist,
581 rvec * gmx_restrict xx,
582 rvec * gmx_restrict ff,
583 t_forcerec * gmx_restrict fr,
584 t_mdatoms * gmx_restrict mdatoms,
585 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
586 t_nrnb * gmx_restrict nrnb)
587 {
588 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
589 * just 0 for non-waters.
590 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
591 * jnr indices corresponding to data put in the four positions in the SIMD register.
592 */
593 int i_shift_offset,i_coord_offset,outeriter,inneriter;
594 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
595 int jnrA,jnrB;
596 int j_coord_offsetA,j_coord_offsetB;
597 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
598 real rcutoff_scalar;
599 real *shiftvec,*fshift,*x,*f;
600 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
601 int vdwioffset1;
602 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
603 int vdwioffset2;
604 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
605 int vdwioffset3;
606 __m128d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
607 int vdwjidx0A,vdwjidx0B;
608 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
609 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
610 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
611 __m128d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
612 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
613 real *charge;
614 __m128i vfitab;
615 __m128i ifour = _mm_set1_epi32(4);
616 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
617 real *vftab;
618 __m128d dummy_mask,cutoff_mask;
619 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
620 __m128d one = _mm_set1_pd(1.0);
621 __m128d two = _mm_set1_pd(2.0);
622 x = xx[0];
623 f = ff[0];
624
625 nri = nlist->nri;
626 iinr = nlist->iinr;
627 jindex = nlist->jindex;
628 jjnr = nlist->jjnr;
629 shiftidx = nlist->shift;
630 gid = nlist->gid;
631 shiftvec = fr->shift_vec[0];
632 fshift = fr->fshift[0];
633 facel = _mm_set1_pd(fr->epsfac);
634 charge = mdatoms->chargeA;
635
636 vftab = kernel_data->table_elec->data;
637 vftabscale = _mm_set1_pd(kernel_data->table_elec->scale);
638
639 /* Setup water-specific parameters */
640 inr = nlist->iinr[0];
641 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
642 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
643 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
644
645 /* Avoid stupid compiler warnings */
646 jnrA = jnrB = 0;
647 j_coord_offsetA = 0;
648 j_coord_offsetB = 0;
649
650 outeriter = 0;
651 inneriter = 0;
652
653 /* Start outer loop over neighborlists */
654 for(iidx=0; iidx<nri; iidx++)
655 {
656 /* Load shift vector for this list */
657 i_shift_offset = DIM*shiftidx[iidx];
658
659 /* Load limits for loop over neighbors */
660 j_index_start = jindex[iidx];
661 j_index_end = jindex[iidx+1];
662
663 /* Get outer coordinate index */
664 inr = iinr[iidx];
665 i_coord_offset = DIM*inr;
666
667 /* Load i particle coords and add shift vector */
668 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
669 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
670
671 fix1 = _mm_setzero_pd();
672 fiy1 = _mm_setzero_pd();
673 fiz1 = _mm_setzero_pd();
674 fix2 = _mm_setzero_pd();
675 fiy2 = _mm_setzero_pd();
676 fiz2 = _mm_setzero_pd();
677 fix3 = _mm_setzero_pd();
678 fiy3 = _mm_setzero_pd();
679 fiz3 = _mm_setzero_pd();
680
681 /* Start inner kernel loop */
682 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
683 {
684
685 /* Get j neighbor index, and coordinate index */
686 jnrA = jjnr[jidx];
687 jnrB = jjnr[jidx+1];
688 j_coord_offsetA = DIM*jnrA;
689 j_coord_offsetB = DIM*jnrB;
690
691 /* load j atom coordinates */
692 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
693 &jx0,&jy0,&jz0);
694
695 /* Calculate displacement vector */
696 dx10 = _mm_sub_pd(ix1,jx0);
697 dy10 = _mm_sub_pd(iy1,jy0);
698 dz10 = _mm_sub_pd(iz1,jz0);
699 dx20 = _mm_sub_pd(ix2,jx0);
700 dy20 = _mm_sub_pd(iy2,jy0);
701 dz20 = _mm_sub_pd(iz2,jz0);
702 dx30 = _mm_sub_pd(ix3,jx0);
703 dy30 = _mm_sub_pd(iy3,jy0);
704 dz30 = _mm_sub_pd(iz3,jz0);
705
706 /* Calculate squared distance and things based on it */
707 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
708 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
709 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
710
711 rinv10 = gmx_mm_invsqrt_pd(rsq10);
712 rinv20 = gmx_mm_invsqrt_pd(rsq20);
713 rinv30 = gmx_mm_invsqrt_pd(rsq30);
714
715 /* Load parameters for j particles */
716 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
717
718 fjx0 = _mm_setzero_pd();
719 fjy0 = _mm_setzero_pd();
720 fjz0 = _mm_setzero_pd();
721
722 /**************************
723 * CALCULATE INTERACTIONS *
724 **************************/
725
726 r10 = _mm_mul_pd(rsq10,rinv10);
727
728 /* Compute parameters for interactions between i and j atoms */
729 qq10 = _mm_mul_pd(iq1,jq0);
730
731 /* Calculate table index by multiplying r with table scale and truncate to integer */
732 rt = _mm_mul_pd(r10,vftabscale);
733 vfitab = _mm_cvttpd_epi32(rt);
734 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
735 vfitab = _mm_slli_epi32(vfitab,2);
736
737 /* CUBIC SPLINE TABLE ELECTROSTATICS */
738 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
739 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
740 GMX_MM_TRANSPOSE2_PD(Y,F);
741 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
742 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
743 GMX_MM_TRANSPOSE2_PD(G,H);
744 Heps = _mm_mul_pd(vfeps,H);
745 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
746 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
747 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
748
749 fscal = felec;
750
751 /* Calculate temporary vectorial force */
752 tx = _mm_mul_pd(fscal,dx10);
753 ty = _mm_mul_pd(fscal,dy10);
754 tz = _mm_mul_pd(fscal,dz10);
755
756 /* Update vectorial force */
757 fix1 = _mm_add_pd(fix1,tx);
758 fiy1 = _mm_add_pd(fiy1,ty);
759 fiz1 = _mm_add_pd(fiz1,tz);
760
761 fjx0 = _mm_add_pd(fjx0,tx);
762 fjy0 = _mm_add_pd(fjy0,ty);
763 fjz0 = _mm_add_pd(fjz0,tz);
764
765 /**************************
766 * CALCULATE INTERACTIONS *
767 **************************/
768
769 r20 = _mm_mul_pd(rsq20,rinv20);
770
771 /* Compute parameters for interactions between i and j atoms */
772 qq20 = _mm_mul_pd(iq2,jq0);
773
774 /* Calculate table index by multiplying r with table scale and truncate to integer */
775 rt = _mm_mul_pd(r20,vftabscale);
776 vfitab = _mm_cvttpd_epi32(rt);
777 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
778 vfitab = _mm_slli_epi32(vfitab,2);
779
780 /* CUBIC SPLINE TABLE ELECTROSTATICS */
781 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
782 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
783 GMX_MM_TRANSPOSE2_PD(Y,F);
784 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
785 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
786 GMX_MM_TRANSPOSE2_PD(G,H);
787 Heps = _mm_mul_pd(vfeps,H);
788 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
789 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
790 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
791
792 fscal = felec;
793
794 /* Calculate temporary vectorial force */
795 tx = _mm_mul_pd(fscal,dx20);
796 ty = _mm_mul_pd(fscal,dy20);
797 tz = _mm_mul_pd(fscal,dz20);
798
799 /* Update vectorial force */
800 fix2 = _mm_add_pd(fix2,tx);
801 fiy2 = _mm_add_pd(fiy2,ty);
802 fiz2 = _mm_add_pd(fiz2,tz);
803
804 fjx0 = _mm_add_pd(fjx0,tx);
805 fjy0 = _mm_add_pd(fjy0,ty);
806 fjz0 = _mm_add_pd(fjz0,tz);
807
808 /**************************
809 * CALCULATE INTERACTIONS *
810 **************************/
811
812 r30 = _mm_mul_pd(rsq30,rinv30);
813
814 /* Compute parameters for interactions between i and j atoms */
815 qq30 = _mm_mul_pd(iq3,jq0);
816
817 /* Calculate table index by multiplying r with table scale and truncate to integer */
818 rt = _mm_mul_pd(r30,vftabscale);
819 vfitab = _mm_cvttpd_epi32(rt);
820 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
821 vfitab = _mm_slli_epi32(vfitab,2);
822
823 /* CUBIC SPLINE TABLE ELECTROSTATICS */
824 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
825 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
826 GMX_MM_TRANSPOSE2_PD(Y,F);
827 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
828 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
829 GMX_MM_TRANSPOSE2_PD(G,H);
830 Heps = _mm_mul_pd(vfeps,H);
831 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
832 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
833 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
834
835 fscal = felec;
836
837 /* Calculate temporary vectorial force */
838 tx = _mm_mul_pd(fscal,dx30);
839 ty = _mm_mul_pd(fscal,dy30);
840 tz = _mm_mul_pd(fscal,dz30);
841
842 /* Update vectorial force */
843 fix3 = _mm_add_pd(fix3,tx);
844 fiy3 = _mm_add_pd(fiy3,ty);
845 fiz3 = _mm_add_pd(fiz3,tz);
846
847 fjx0 = _mm_add_pd(fjx0,tx);
848 fjy0 = _mm_add_pd(fjy0,ty);
849 fjz0 = _mm_add_pd(fjz0,tz);
850
851 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
852
853 /* Inner loop uses 120 flops */
854 }
855
856 if(jidx<j_index_end)
857 {
858
859 jnrA = jjnr[jidx];
860 j_coord_offsetA = DIM*jnrA;
861
862 /* load j atom coordinates */
863 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
864 &jx0,&jy0,&jz0);
865
866 /* Calculate displacement vector */
867 dx10 = _mm_sub_pd(ix1,jx0);
868 dy10 = _mm_sub_pd(iy1,jy0);
869 dz10 = _mm_sub_pd(iz1,jz0);
870 dx20 = _mm_sub_pd(ix2,jx0);
871 dy20 = _mm_sub_pd(iy2,jy0);
872 dz20 = _mm_sub_pd(iz2,jz0);
873 dx30 = _mm_sub_pd(ix3,jx0);
874 dy30 = _mm_sub_pd(iy3,jy0);
875 dz30 = _mm_sub_pd(iz3,jz0);
876
877 /* Calculate squared distance and things based on it */
878 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
879 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
880 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
881
882 rinv10 = gmx_mm_invsqrt_pd(rsq10);
883 rinv20 = gmx_mm_invsqrt_pd(rsq20);
884 rinv30 = gmx_mm_invsqrt_pd(rsq30);
885
886 /* Load parameters for j particles */
887 jq0 = _mm_load_sd(charge+jnrA+0);
888
889 fjx0 = _mm_setzero_pd();
890 fjy0 = _mm_setzero_pd();
891 fjz0 = _mm_setzero_pd();
892
893 /**************************
894 * CALCULATE INTERACTIONS *
895 **************************/
896
897 r10 = _mm_mul_pd(rsq10,rinv10);
898
899 /* Compute parameters for interactions between i and j atoms */
900 qq10 = _mm_mul_pd(iq1,jq0);
901
902 /* Calculate table index by multiplying r with table scale and truncate to integer */
903 rt = _mm_mul_pd(r10,vftabscale);
904 vfitab = _mm_cvttpd_epi32(rt);
905 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
906 vfitab = _mm_slli_epi32(vfitab,2);
907
908 /* CUBIC SPLINE TABLE ELECTROSTATICS */
909 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
910 F = _mm_setzero_pd();
911 GMX_MM_TRANSPOSE2_PD(Y,F);
912 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
913 H = _mm_setzero_pd();
914 GMX_MM_TRANSPOSE2_PD(G,H);
915 Heps = _mm_mul_pd(vfeps,H);
916 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
917 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
918 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
919
920 fscal = felec;
921
922 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
923
924 /* Calculate temporary vectorial force */
925 tx = _mm_mul_pd(fscal,dx10);
926 ty = _mm_mul_pd(fscal,dy10);
927 tz = _mm_mul_pd(fscal,dz10);
928
929 /* Update vectorial force */
930 fix1 = _mm_add_pd(fix1,tx);
931 fiy1 = _mm_add_pd(fiy1,ty);
932 fiz1 = _mm_add_pd(fiz1,tz);
933
934 fjx0 = _mm_add_pd(fjx0,tx);
935 fjy0 = _mm_add_pd(fjy0,ty);
936 fjz0 = _mm_add_pd(fjz0,tz);
937
938 /**************************
939 * CALCULATE INTERACTIONS *
940 **************************/
941
942 r20 = _mm_mul_pd(rsq20,rinv20);
943
944 /* Compute parameters for interactions between i and j atoms */
945 qq20 = _mm_mul_pd(iq2,jq0);
946
947 /* Calculate table index by multiplying r with table scale and truncate to integer */
948 rt = _mm_mul_pd(r20,vftabscale);
949 vfitab = _mm_cvttpd_epi32(rt);
950 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
951 vfitab = _mm_slli_epi32(vfitab,2);
952
953 /* CUBIC SPLINE TABLE ELECTROSTATICS */
954 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
955 F = _mm_setzero_pd();
956 GMX_MM_TRANSPOSE2_PD(Y,F);
957 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
958 H = _mm_setzero_pd();
959 GMX_MM_TRANSPOSE2_PD(G,H);
960 Heps = _mm_mul_pd(vfeps,H);
961 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
962 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
963 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
964
965 fscal = felec;
966
967 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
968
969 /* Calculate temporary vectorial force */
970 tx = _mm_mul_pd(fscal,dx20);
971 ty = _mm_mul_pd(fscal,dy20);
972 tz = _mm_mul_pd(fscal,dz20);
973
974 /* Update vectorial force */
975 fix2 = _mm_add_pd(fix2,tx);
976 fiy2 = _mm_add_pd(fiy2,ty);
977 fiz2 = _mm_add_pd(fiz2,tz);
978
979 fjx0 = _mm_add_pd(fjx0,tx);
980 fjy0 = _mm_add_pd(fjy0,ty);
981 fjz0 = _mm_add_pd(fjz0,tz);
982
983 /**************************
984 * CALCULATE INTERACTIONS *
985 **************************/
986
987 r30 = _mm_mul_pd(rsq30,rinv30);
988
989 /* Compute parameters for interactions between i and j atoms */
990 qq30 = _mm_mul_pd(iq3,jq0);
991
992 /* Calculate table index by multiplying r with table scale and truncate to integer */
993 rt = _mm_mul_pd(r30,vftabscale);
994 vfitab = _mm_cvttpd_epi32(rt);
995 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
996 vfitab = _mm_slli_epi32(vfitab,2);
997
998 /* CUBIC SPLINE TABLE ELECTROSTATICS */
999 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1000 F = _mm_setzero_pd();
1001 GMX_MM_TRANSPOSE2_PD(Y,F);
1002 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1003 H = _mm_setzero_pd();
1004 GMX_MM_TRANSPOSE2_PD(G,H);
1005 Heps = _mm_mul_pd(vfeps,H);
1006 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1007 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1008 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
1009
1010 fscal = felec;
1011
1012 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1013
1014 /* Calculate temporary vectorial force */
1015 tx = _mm_mul_pd(fscal,dx30);
1016 ty = _mm_mul_pd(fscal,dy30);
1017 tz = _mm_mul_pd(fscal,dz30);
1018
1019 /* Update vectorial force */
1020 fix3 = _mm_add_pd(fix3,tx);
1021 fiy3 = _mm_add_pd(fiy3,ty);
1022 fiz3 = _mm_add_pd(fiz3,tz);
1023
1024 fjx0 = _mm_add_pd(fjx0,tx);
1025 fjy0 = _mm_add_pd(fjy0,ty);
1026 fjz0 = _mm_add_pd(fjz0,tz);
1027
1028 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1029
1030 /* Inner loop uses 120 flops */
1031 }
1032
1033 /* End of innermost loop */
1034
1035 gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1036 f+i_coord_offset+DIM,fshift+i_shift_offset);
1037
1038 /* Increment number of inner iterations */
1039 inneriter += j_index_end - j_index_start;
1040
1041 /* Outer loop uses 18 flops */
1042 }
1043
1044 /* Increment number of outer iterations */
1045 outeriter += nri;
1046
1047 /* Update outer/inner flops */
1048
1049 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*120);
1050 }
The ultimate molecular dynamics simulation package