?  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_ElecEw_VdwLJ_GeomW4P1_sse2_double.c
blobea51f68105df396f1112cdf9a025caabbb476a11
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_ElecEw_VdwLJ_GeomW4P1_VF_sse2_double
53 * Electrostatics interaction: Ewald
54 * VdW interaction: LennardJones
55 * Geometry: Water4-Particle
56 * Calculate force/pot: PotentialAndForce
57 */
58 void
59 nb_kernel_ElecEw_VdwLJ_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 vdwioffset0;
82 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
83 int vdwioffset1;
84 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
85 int vdwioffset2;
86 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
87 int vdwioffset3;
88 __m128d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
89 int vdwjidx0A,vdwjidx0B;
90 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
91 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
92 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
93 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
94 __m128d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
95 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
96 real *charge;
97 int nvdwtype;
98 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
99 int *vdwtype;
100 real *vdwparam;
101 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
102 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
103 __m128i ewitab;
104 __m128d ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
105 real *ewtab;
106 __m128d dummy_mask,cutoff_mask;
107 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
108 __m128d one = _mm_set1_pd(1.0);
109 __m128d two = _mm_set1_pd(2.0);
110 x = xx[0];
111 f = ff[0];
112
113 nri = nlist->nri;
114 iinr = nlist->iinr;
115 jindex = nlist->jindex;
116 jjnr = nlist->jjnr;
117 shiftidx = nlist->shift;
118 gid = nlist->gid;
119 shiftvec = fr->shift_vec[0];
120 fshift = fr->fshift[0];
121 facel = _mm_set1_pd(fr->epsfac);
122 charge = mdatoms->chargeA;
123 nvdwtype = fr->ntype;
124 vdwparam = fr->nbfp;
125 vdwtype = mdatoms->typeA;
126
127 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
128 ewtab = fr->ic->tabq_coul_FDV0;
129 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
130 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
131
132 /* Setup water-specific parameters */
133 inr = nlist->iinr[0];
134 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
135 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
136 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
137 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
138
139 /* Avoid stupid compiler warnings */
140 jnrA = jnrB = 0;
141 j_coord_offsetA = 0;
142 j_coord_offsetB = 0;
143
144 outeriter = 0;
145 inneriter = 0;
146
147 /* Start outer loop over neighborlists */
148 for(iidx=0; iidx<nri; iidx++)
149 {
150 /* Load shift vector for this list */
151 i_shift_offset = DIM*shiftidx[iidx];
152
153 /* Load limits for loop over neighbors */
154 j_index_start = jindex[iidx];
155 j_index_end = jindex[iidx+1];
156
157 /* Get outer coordinate index */
158 inr = iinr[iidx];
159 i_coord_offset = DIM*inr;
160
161 /* Load i particle coords and add shift vector */
162 gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
163 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
164
165 fix0 = _mm_setzero_pd();
166 fiy0 = _mm_setzero_pd();
167 fiz0 = _mm_setzero_pd();
168 fix1 = _mm_setzero_pd();
169 fiy1 = _mm_setzero_pd();
170 fiz1 = _mm_setzero_pd();
171 fix2 = _mm_setzero_pd();
172 fiy2 = _mm_setzero_pd();
173 fiz2 = _mm_setzero_pd();
174 fix3 = _mm_setzero_pd();
175 fiy3 = _mm_setzero_pd();
176 fiz3 = _mm_setzero_pd();
177
178 /* Reset potential sums */
179 velecsum = _mm_setzero_pd();
180 vvdwsum = _mm_setzero_pd();
181
182 /* Start inner kernel loop */
183 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
184 {
185
186 /* Get j neighbor index, and coordinate index */
187 jnrA = jjnr[jidx];
188 jnrB = jjnr[jidx+1];
189 j_coord_offsetA = DIM*jnrA;
190 j_coord_offsetB = DIM*jnrB;
191
192 /* load j atom coordinates */
193 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
194 &jx0,&jy0,&jz0);
195
196 /* Calculate displacement vector */
197 dx00 = _mm_sub_pd(ix0,jx0);
198 dy00 = _mm_sub_pd(iy0,jy0);
199 dz00 = _mm_sub_pd(iz0,jz0);
200 dx10 = _mm_sub_pd(ix1,jx0);
201 dy10 = _mm_sub_pd(iy1,jy0);
202 dz10 = _mm_sub_pd(iz1,jz0);
203 dx20 = _mm_sub_pd(ix2,jx0);
204 dy20 = _mm_sub_pd(iy2,jy0);
205 dz20 = _mm_sub_pd(iz2,jz0);
206 dx30 = _mm_sub_pd(ix3,jx0);
207 dy30 = _mm_sub_pd(iy3,jy0);
208 dz30 = _mm_sub_pd(iz3,jz0);
209
210 /* Calculate squared distance and things based on it */
211 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
212 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
213 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
214 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
215
216 rinv10 = gmx_mm_invsqrt_pd(rsq10);
217 rinv20 = gmx_mm_invsqrt_pd(rsq20);
218 rinv30 = gmx_mm_invsqrt_pd(rsq30);
219
220 rinvsq00 = gmx_mm_inv_pd(rsq00);
221 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
222 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
223 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
224
225 /* Load parameters for j particles */
226 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
227 vdwjidx0A = 2*vdwtype[jnrA+0];
228 vdwjidx0B = 2*vdwtype[jnrB+0];
229
230 fjx0 = _mm_setzero_pd();
231 fjy0 = _mm_setzero_pd();
232 fjz0 = _mm_setzero_pd();
233
234 /**************************
235 * CALCULATE INTERACTIONS *
236 **************************/
237
238 /* Compute parameters for interactions between i and j atoms */
239 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
240 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
241
242 /* LENNARD-JONES DISPERSION/REPULSION */
243
244 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
245 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
246 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
247 vvdw = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
248 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
249
250 /* Update potential sum for this i atom from the interaction with this j atom. */
251 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
252
253 fscal = fvdw;
254
255 /* Calculate temporary vectorial force */
256 tx = _mm_mul_pd(fscal,dx00);
257 ty = _mm_mul_pd(fscal,dy00);
258 tz = _mm_mul_pd(fscal,dz00);
259
260 /* Update vectorial force */
261 fix0 = _mm_add_pd(fix0,tx);
262 fiy0 = _mm_add_pd(fiy0,ty);
263 fiz0 = _mm_add_pd(fiz0,tz);
264
265 fjx0 = _mm_add_pd(fjx0,tx);
266 fjy0 = _mm_add_pd(fjy0,ty);
267 fjz0 = _mm_add_pd(fjz0,tz);
268
269 /**************************
270 * CALCULATE INTERACTIONS *
271 **************************/
272
273 r10 = _mm_mul_pd(rsq10,rinv10);
274
275 /* Compute parameters for interactions between i and j atoms */
276 qq10 = _mm_mul_pd(iq1,jq0);
277
278 /* EWALD ELECTROSTATICS */
279
280 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
281 ewrt = _mm_mul_pd(r10,ewtabscale);
282 ewitab = _mm_cvttpd_epi32(ewrt);
283 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
284 ewitab = _mm_slli_epi32(ewitab,2);
285 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
286 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
287 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
288 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
289 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
290 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
291 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
292 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
293 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
294 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
295
296 /* Update potential sum for this i atom from the interaction with this j atom. */
297 velecsum = _mm_add_pd(velecsum,velec);
298
299 fscal = felec;
300
301 /* Calculate temporary vectorial force */
302 tx = _mm_mul_pd(fscal,dx10);
303 ty = _mm_mul_pd(fscal,dy10);
304 tz = _mm_mul_pd(fscal,dz10);
305
306 /* Update vectorial force */
307 fix1 = _mm_add_pd(fix1,tx);
308 fiy1 = _mm_add_pd(fiy1,ty);
309 fiz1 = _mm_add_pd(fiz1,tz);
310
311 fjx0 = _mm_add_pd(fjx0,tx);
312 fjy0 = _mm_add_pd(fjy0,ty);
313 fjz0 = _mm_add_pd(fjz0,tz);
314
315 /**************************
316 * CALCULATE INTERACTIONS *
317 **************************/
318
319 r20 = _mm_mul_pd(rsq20,rinv20);
320
321 /* Compute parameters for interactions between i and j atoms */
322 qq20 = _mm_mul_pd(iq2,jq0);
323
324 /* EWALD ELECTROSTATICS */
325
326 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
327 ewrt = _mm_mul_pd(r20,ewtabscale);
328 ewitab = _mm_cvttpd_epi32(ewrt);
329 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
330 ewitab = _mm_slli_epi32(ewitab,2);
331 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
332 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
333 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
334 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
335 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
336 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
337 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
338 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
339 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
340 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
341
342 /* Update potential sum for this i atom from the interaction with this j atom. */
343 velecsum = _mm_add_pd(velecsum,velec);
344
345 fscal = felec;
346
347 /* Calculate temporary vectorial force */
348 tx = _mm_mul_pd(fscal,dx20);
349 ty = _mm_mul_pd(fscal,dy20);
350 tz = _mm_mul_pd(fscal,dz20);
351
352 /* Update vectorial force */
353 fix2 = _mm_add_pd(fix2,tx);
354 fiy2 = _mm_add_pd(fiy2,ty);
355 fiz2 = _mm_add_pd(fiz2,tz);
356
357 fjx0 = _mm_add_pd(fjx0,tx);
358 fjy0 = _mm_add_pd(fjy0,ty);
359 fjz0 = _mm_add_pd(fjz0,tz);
360
361 /**************************
362 * CALCULATE INTERACTIONS *
363 **************************/
364
365 r30 = _mm_mul_pd(rsq30,rinv30);
366
367 /* Compute parameters for interactions between i and j atoms */
368 qq30 = _mm_mul_pd(iq3,jq0);
369
370 /* EWALD ELECTROSTATICS */
371
372 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
373 ewrt = _mm_mul_pd(r30,ewtabscale);
374 ewitab = _mm_cvttpd_epi32(ewrt);
375 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
376 ewitab = _mm_slli_epi32(ewitab,2);
377 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
378 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
379 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
380 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
381 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
382 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
383 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
384 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
385 velec = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
386 felec = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
387
388 /* Update potential sum for this i atom from the interaction with this j atom. */
389 velecsum = _mm_add_pd(velecsum,velec);
390
391 fscal = felec;
392
393 /* Calculate temporary vectorial force */
394 tx = _mm_mul_pd(fscal,dx30);
395 ty = _mm_mul_pd(fscal,dy30);
396 tz = _mm_mul_pd(fscal,dz30);
397
398 /* Update vectorial force */
399 fix3 = _mm_add_pd(fix3,tx);
400 fiy3 = _mm_add_pd(fiy3,ty);
401 fiz3 = _mm_add_pd(fiz3,tz);
402
403 fjx0 = _mm_add_pd(fjx0,tx);
404 fjy0 = _mm_add_pd(fjy0,ty);
405 fjz0 = _mm_add_pd(fjz0,tz);
406
407 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
408
409 /* Inner loop uses 158 flops */
410 }
411
412 if(jidx<j_index_end)
413 {
414
415 jnrA = jjnr[jidx];
416 j_coord_offsetA = DIM*jnrA;
417
418 /* load j atom coordinates */
419 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
420 &jx0,&jy0,&jz0);
421
422 /* Calculate displacement vector */
423 dx00 = _mm_sub_pd(ix0,jx0);
424 dy00 = _mm_sub_pd(iy0,jy0);
425 dz00 = _mm_sub_pd(iz0,jz0);
426 dx10 = _mm_sub_pd(ix1,jx0);
427 dy10 = _mm_sub_pd(iy1,jy0);
428 dz10 = _mm_sub_pd(iz1,jz0);
429 dx20 = _mm_sub_pd(ix2,jx0);
430 dy20 = _mm_sub_pd(iy2,jy0);
431 dz20 = _mm_sub_pd(iz2,jz0);
432 dx30 = _mm_sub_pd(ix3,jx0);
433 dy30 = _mm_sub_pd(iy3,jy0);
434 dz30 = _mm_sub_pd(iz3,jz0);
435
436 /* Calculate squared distance and things based on it */
437 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
438 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
439 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
440 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
441
442 rinv10 = gmx_mm_invsqrt_pd(rsq10);
443 rinv20 = gmx_mm_invsqrt_pd(rsq20);
444 rinv30 = gmx_mm_invsqrt_pd(rsq30);
445
446 rinvsq00 = gmx_mm_inv_pd(rsq00);
447 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
448 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
449 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
450
451 /* Load parameters for j particles */
452 jq0 = _mm_load_sd(charge+jnrA+0);
453 vdwjidx0A = 2*vdwtype[jnrA+0];
454
455 fjx0 = _mm_setzero_pd();
456 fjy0 = _mm_setzero_pd();
457 fjz0 = _mm_setzero_pd();
458
459 /**************************
460 * CALCULATE INTERACTIONS *
461 **************************/
462
463 /* Compute parameters for interactions between i and j atoms */
464 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
465
466 /* LENNARD-JONES DISPERSION/REPULSION */
467
468 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
469 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
470 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
471 vvdw = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
472 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
473
474 /* Update potential sum for this i atom from the interaction with this j atom. */
475 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
476 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
477
478 fscal = fvdw;
479
480 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
481
482 /* Calculate temporary vectorial force */
483 tx = _mm_mul_pd(fscal,dx00);
484 ty = _mm_mul_pd(fscal,dy00);
485 tz = _mm_mul_pd(fscal,dz00);
486
487 /* Update vectorial force */
488 fix0 = _mm_add_pd(fix0,tx);
489 fiy0 = _mm_add_pd(fiy0,ty);
490 fiz0 = _mm_add_pd(fiz0,tz);
491
492 fjx0 = _mm_add_pd(fjx0,tx);
493 fjy0 = _mm_add_pd(fjy0,ty);
494 fjz0 = _mm_add_pd(fjz0,tz);
495
496 /**************************
497 * CALCULATE INTERACTIONS *
498 **************************/
499
500 r10 = _mm_mul_pd(rsq10,rinv10);
501
502 /* Compute parameters for interactions between i and j atoms */
503 qq10 = _mm_mul_pd(iq1,jq0);
504
505 /* EWALD ELECTROSTATICS */
506
507 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
508 ewrt = _mm_mul_pd(r10,ewtabscale);
509 ewitab = _mm_cvttpd_epi32(ewrt);
510 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
511 ewitab = _mm_slli_epi32(ewitab,2);
512 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
513 ewtabD = _mm_setzero_pd();
514 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
515 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
516 ewtabFn = _mm_setzero_pd();
517 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
518 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
519 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
520 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
521 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
522
523 /* Update potential sum for this i atom from the interaction with this j atom. */
524 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
525 velecsum = _mm_add_pd(velecsum,velec);
526
527 fscal = felec;
528
529 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
530
531 /* Calculate temporary vectorial force */
532 tx = _mm_mul_pd(fscal,dx10);
533 ty = _mm_mul_pd(fscal,dy10);
534 tz = _mm_mul_pd(fscal,dz10);
535
536 /* Update vectorial force */
537 fix1 = _mm_add_pd(fix1,tx);
538 fiy1 = _mm_add_pd(fiy1,ty);
539 fiz1 = _mm_add_pd(fiz1,tz);
540
541 fjx0 = _mm_add_pd(fjx0,tx);
542 fjy0 = _mm_add_pd(fjy0,ty);
543 fjz0 = _mm_add_pd(fjz0,tz);
544
545 /**************************
546 * CALCULATE INTERACTIONS *
547 **************************/
548
549 r20 = _mm_mul_pd(rsq20,rinv20);
550
551 /* Compute parameters for interactions between i and j atoms */
552 qq20 = _mm_mul_pd(iq2,jq0);
553
554 /* EWALD ELECTROSTATICS */
555
556 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
557 ewrt = _mm_mul_pd(r20,ewtabscale);
558 ewitab = _mm_cvttpd_epi32(ewrt);
559 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
560 ewitab = _mm_slli_epi32(ewitab,2);
561 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
562 ewtabD = _mm_setzero_pd();
563 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
564 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
565 ewtabFn = _mm_setzero_pd();
566 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
567 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
568 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
569 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
570 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
571
572 /* Update potential sum for this i atom from the interaction with this j atom. */
573 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
574 velecsum = _mm_add_pd(velecsum,velec);
575
576 fscal = felec;
577
578 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
579
580 /* Calculate temporary vectorial force */
581 tx = _mm_mul_pd(fscal,dx20);
582 ty = _mm_mul_pd(fscal,dy20);
583 tz = _mm_mul_pd(fscal,dz20);
584
585 /* Update vectorial force */
586 fix2 = _mm_add_pd(fix2,tx);
587 fiy2 = _mm_add_pd(fiy2,ty);
588 fiz2 = _mm_add_pd(fiz2,tz);
589
590 fjx0 = _mm_add_pd(fjx0,tx);
591 fjy0 = _mm_add_pd(fjy0,ty);
592 fjz0 = _mm_add_pd(fjz0,tz);
593
594 /**************************
595 * CALCULATE INTERACTIONS *
596 **************************/
597
598 r30 = _mm_mul_pd(rsq30,rinv30);
599
600 /* Compute parameters for interactions between i and j atoms */
601 qq30 = _mm_mul_pd(iq3,jq0);
602
603 /* EWALD ELECTROSTATICS */
604
605 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
606 ewrt = _mm_mul_pd(r30,ewtabscale);
607 ewitab = _mm_cvttpd_epi32(ewrt);
608 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
609 ewitab = _mm_slli_epi32(ewitab,2);
610 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
611 ewtabD = _mm_setzero_pd();
612 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
613 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
614 ewtabFn = _mm_setzero_pd();
615 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
616 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
617 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
618 velec = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
619 felec = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
620
621 /* Update potential sum for this i atom from the interaction with this j atom. */
622 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
623 velecsum = _mm_add_pd(velecsum,velec);
624
625 fscal = felec;
626
627 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
628
629 /* Calculate temporary vectorial force */
630 tx = _mm_mul_pd(fscal,dx30);
631 ty = _mm_mul_pd(fscal,dy30);
632 tz = _mm_mul_pd(fscal,dz30);
633
634 /* Update vectorial force */
635 fix3 = _mm_add_pd(fix3,tx);
636 fiy3 = _mm_add_pd(fiy3,ty);
637 fiz3 = _mm_add_pd(fiz3,tz);
638
639 fjx0 = _mm_add_pd(fjx0,tx);
640 fjy0 = _mm_add_pd(fjy0,ty);
641 fjz0 = _mm_add_pd(fjz0,tz);
642
643 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
644
645 /* Inner loop uses 158 flops */
646 }
647
648 /* End of innermost loop */
649
650 gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
651 f+i_coord_offset,fshift+i_shift_offset);
652
653 ggid = gid[iidx];
654 /* Update potential energies */
655 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
656 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
657
658 /* Increment number of inner iterations */
659 inneriter += j_index_end - j_index_start;
660
661 /* Outer loop uses 26 flops */
662 }
663
664 /* Increment number of outer iterations */
665 outeriter += nri;
666
667 /* Update outer/inner flops */
668
669 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*158);
670 }
671 /*
672 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwLJ_GeomW4P1_F_sse2_double
673 * Electrostatics interaction: Ewald
674 * VdW interaction: LennardJones
675 * Geometry: Water4-Particle
676 * Calculate force/pot: Force
677 */
678 void
679 nb_kernel_ElecEw_VdwLJ_GeomW4P1_F_sse2_double
680 (t_nblist * gmx_restrict nlist,
681 rvec * gmx_restrict xx,
682 rvec * gmx_restrict ff,
683 t_forcerec * gmx_restrict fr,
684 t_mdatoms * gmx_restrict mdatoms,
685 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
686 t_nrnb * gmx_restrict nrnb)
687 {
688 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
689 * just 0 for non-waters.
690 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
691 * jnr indices corresponding to data put in the four positions in the SIMD register.
692 */
693 int i_shift_offset,i_coord_offset,outeriter,inneriter;
694 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
695 int jnrA,jnrB;
696 int j_coord_offsetA,j_coord_offsetB;
697 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
698 real rcutoff_scalar;
699 real *shiftvec,*fshift,*x,*f;
700 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
701 int vdwioffset0;
702 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
703 int vdwioffset1;
704 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
705 int vdwioffset2;
706 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
707 int vdwioffset3;
708 __m128d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
709 int vdwjidx0A,vdwjidx0B;
710 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
711 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
712 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
713 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
714 __m128d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
715 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
716 real *charge;
717 int nvdwtype;
718 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
719 int *vdwtype;
720 real *vdwparam;
721 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
722 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
723 __m128i ewitab;
724 __m128d ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
725 real *ewtab;
726 __m128d dummy_mask,cutoff_mask;
727 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
728 __m128d one = _mm_set1_pd(1.0);
729 __m128d two = _mm_set1_pd(2.0);
730 x = xx[0];
731 f = ff[0];
732
733 nri = nlist->nri;
734 iinr = nlist->iinr;
735 jindex = nlist->jindex;
736 jjnr = nlist->jjnr;
737 shiftidx = nlist->shift;
738 gid = nlist->gid;
739 shiftvec = fr->shift_vec[0];
740 fshift = fr->fshift[0];
741 facel = _mm_set1_pd(fr->epsfac);
742 charge = mdatoms->chargeA;
743 nvdwtype = fr->ntype;
744 vdwparam = fr->nbfp;
745 vdwtype = mdatoms->typeA;
746
747 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
748 ewtab = fr->ic->tabq_coul_F;
749 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
750 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
751
752 /* Setup water-specific parameters */
753 inr = nlist->iinr[0];
754 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
755 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
756 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
757 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
758
759 /* Avoid stupid compiler warnings */
760 jnrA = jnrB = 0;
761 j_coord_offsetA = 0;
762 j_coord_offsetB = 0;
763
764 outeriter = 0;
765 inneriter = 0;
766
767 /* Start outer loop over neighborlists */
768 for(iidx=0; iidx<nri; iidx++)
769 {
770 /* Load shift vector for this list */
771 i_shift_offset = DIM*shiftidx[iidx];
772
773 /* Load limits for loop over neighbors */
774 j_index_start = jindex[iidx];
775 j_index_end = jindex[iidx+1];
776
777 /* Get outer coordinate index */
778 inr = iinr[iidx];
779 i_coord_offset = DIM*inr;
780
781 /* Load i particle coords and add shift vector */
782 gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
783 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
784
785 fix0 = _mm_setzero_pd();
786 fiy0 = _mm_setzero_pd();
787 fiz0 = _mm_setzero_pd();
788 fix1 = _mm_setzero_pd();
789 fiy1 = _mm_setzero_pd();
790 fiz1 = _mm_setzero_pd();
791 fix2 = _mm_setzero_pd();
792 fiy2 = _mm_setzero_pd();
793 fiz2 = _mm_setzero_pd();
794 fix3 = _mm_setzero_pd();
795 fiy3 = _mm_setzero_pd();
796 fiz3 = _mm_setzero_pd();
797
798 /* Start inner kernel loop */
799 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
800 {
801
802 /* Get j neighbor index, and coordinate index */
803 jnrA = jjnr[jidx];
804 jnrB = jjnr[jidx+1];
805 j_coord_offsetA = DIM*jnrA;
806 j_coord_offsetB = DIM*jnrB;
807
808 /* load j atom coordinates */
809 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
810 &jx0,&jy0,&jz0);
811
812 /* Calculate displacement vector */
813 dx00 = _mm_sub_pd(ix0,jx0);
814 dy00 = _mm_sub_pd(iy0,jy0);
815 dz00 = _mm_sub_pd(iz0,jz0);
816 dx10 = _mm_sub_pd(ix1,jx0);
817 dy10 = _mm_sub_pd(iy1,jy0);
818 dz10 = _mm_sub_pd(iz1,jz0);
819 dx20 = _mm_sub_pd(ix2,jx0);
820 dy20 = _mm_sub_pd(iy2,jy0);
821 dz20 = _mm_sub_pd(iz2,jz0);
822 dx30 = _mm_sub_pd(ix3,jx0);
823 dy30 = _mm_sub_pd(iy3,jy0);
824 dz30 = _mm_sub_pd(iz3,jz0);
825
826 /* Calculate squared distance and things based on it */
827 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
828 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
829 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
830 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
831
832 rinv10 = gmx_mm_invsqrt_pd(rsq10);
833 rinv20 = gmx_mm_invsqrt_pd(rsq20);
834 rinv30 = gmx_mm_invsqrt_pd(rsq30);
835
836 rinvsq00 = gmx_mm_inv_pd(rsq00);
837 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
838 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
839 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
840
841 /* Load parameters for j particles */
842 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
843 vdwjidx0A = 2*vdwtype[jnrA+0];
844 vdwjidx0B = 2*vdwtype[jnrB+0];
845
846 fjx0 = _mm_setzero_pd();
847 fjy0 = _mm_setzero_pd();
848 fjz0 = _mm_setzero_pd();
849
850 /**************************
851 * CALCULATE INTERACTIONS *
852 **************************/
853
854 /* Compute parameters for interactions between i and j atoms */
855 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
856 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
857
858 /* LENNARD-JONES DISPERSION/REPULSION */
859
860 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
861 fvdw = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
862
863 fscal = fvdw;
864
865 /* Calculate temporary vectorial force */
866 tx = _mm_mul_pd(fscal,dx00);
867 ty = _mm_mul_pd(fscal,dy00);
868 tz = _mm_mul_pd(fscal,dz00);
869
870 /* Update vectorial force */
871 fix0 = _mm_add_pd(fix0,tx);
872 fiy0 = _mm_add_pd(fiy0,ty);
873 fiz0 = _mm_add_pd(fiz0,tz);
874
875 fjx0 = _mm_add_pd(fjx0,tx);
876 fjy0 = _mm_add_pd(fjy0,ty);
877 fjz0 = _mm_add_pd(fjz0,tz);
878
879 /**************************
880 * CALCULATE INTERACTIONS *
881 **************************/
882
883 r10 = _mm_mul_pd(rsq10,rinv10);
884
885 /* Compute parameters for interactions between i and j atoms */
886 qq10 = _mm_mul_pd(iq1,jq0);
887
888 /* EWALD ELECTROSTATICS */
889
890 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
891 ewrt = _mm_mul_pd(r10,ewtabscale);
892 ewitab = _mm_cvttpd_epi32(ewrt);
893 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
894 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
895 &ewtabF,&ewtabFn);
896 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
897 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
898
899 fscal = felec;
900
901 /* Calculate temporary vectorial force */
902 tx = _mm_mul_pd(fscal,dx10);
903 ty = _mm_mul_pd(fscal,dy10);
904 tz = _mm_mul_pd(fscal,dz10);
905
906 /* Update vectorial force */
907 fix1 = _mm_add_pd(fix1,tx);
908 fiy1 = _mm_add_pd(fiy1,ty);
909 fiz1 = _mm_add_pd(fiz1,tz);
910
911 fjx0 = _mm_add_pd(fjx0,tx);
912 fjy0 = _mm_add_pd(fjy0,ty);
913 fjz0 = _mm_add_pd(fjz0,tz);
914
915 /**************************
916 * CALCULATE INTERACTIONS *
917 **************************/
918
919 r20 = _mm_mul_pd(rsq20,rinv20);
920
921 /* Compute parameters for interactions between i and j atoms */
922 qq20 = _mm_mul_pd(iq2,jq0);
923
924 /* EWALD ELECTROSTATICS */
925
926 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
927 ewrt = _mm_mul_pd(r20,ewtabscale);
928 ewitab = _mm_cvttpd_epi32(ewrt);
929 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
930 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
931 &ewtabF,&ewtabFn);
932 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
933 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
934
935 fscal = felec;
936
937 /* Calculate temporary vectorial force */
938 tx = _mm_mul_pd(fscal,dx20);
939 ty = _mm_mul_pd(fscal,dy20);
940 tz = _mm_mul_pd(fscal,dz20);
941
942 /* Update vectorial force */
943 fix2 = _mm_add_pd(fix2,tx);
944 fiy2 = _mm_add_pd(fiy2,ty);
945 fiz2 = _mm_add_pd(fiz2,tz);
946
947 fjx0 = _mm_add_pd(fjx0,tx);
948 fjy0 = _mm_add_pd(fjy0,ty);
949 fjz0 = _mm_add_pd(fjz0,tz);
950
951 /**************************
952 * CALCULATE INTERACTIONS *
953 **************************/
954
955 r30 = _mm_mul_pd(rsq30,rinv30);
956
957 /* Compute parameters for interactions between i and j atoms */
958 qq30 = _mm_mul_pd(iq3,jq0);
959
960 /* EWALD ELECTROSTATICS */
961
962 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
963 ewrt = _mm_mul_pd(r30,ewtabscale);
964 ewitab = _mm_cvttpd_epi32(ewrt);
965 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
966 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
967 &ewtabF,&ewtabFn);
968 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
969 felec = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
970
971 fscal = felec;
972
973 /* Calculate temporary vectorial force */
974 tx = _mm_mul_pd(fscal,dx30);
975 ty = _mm_mul_pd(fscal,dy30);
976 tz = _mm_mul_pd(fscal,dz30);
977
978 /* Update vectorial force */
979 fix3 = _mm_add_pd(fix3,tx);
980 fiy3 = _mm_add_pd(fiy3,ty);
981 fiz3 = _mm_add_pd(fiz3,tz);
982
983 fjx0 = _mm_add_pd(fjx0,tx);
984 fjy0 = _mm_add_pd(fjy0,ty);
985 fjz0 = _mm_add_pd(fjz0,tz);
986
987 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
988
989 /* Inner loop uses 138 flops */
990 }
991
992 if(jidx<j_index_end)
993 {
994
995 jnrA = jjnr[jidx];
996 j_coord_offsetA = DIM*jnrA;
997
998 /* load j atom coordinates */
999 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1000 &jx0,&jy0,&jz0);
1001
1002 /* Calculate displacement vector */
1003 dx00 = _mm_sub_pd(ix0,jx0);
1004 dy00 = _mm_sub_pd(iy0,jy0);
1005 dz00 = _mm_sub_pd(iz0,jz0);
1006 dx10 = _mm_sub_pd(ix1,jx0);
1007 dy10 = _mm_sub_pd(iy1,jy0);
1008 dz10 = _mm_sub_pd(iz1,jz0);
1009 dx20 = _mm_sub_pd(ix2,jx0);
1010 dy20 = _mm_sub_pd(iy2,jy0);
1011 dz20 = _mm_sub_pd(iz2,jz0);
1012 dx30 = _mm_sub_pd(ix3,jx0);
1013 dy30 = _mm_sub_pd(iy3,jy0);
1014 dz30 = _mm_sub_pd(iz3,jz0);
1015
1016 /* Calculate squared distance and things based on it */
1017 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1018 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1019 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1020 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1021
1022 rinv10 = gmx_mm_invsqrt_pd(rsq10);
1023 rinv20 = gmx_mm_invsqrt_pd(rsq20);
1024 rinv30 = gmx_mm_invsqrt_pd(rsq30);
1025
1026 rinvsq00 = gmx_mm_inv_pd(rsq00);
1027 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
1028 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
1029 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
1030
1031 /* Load parameters for j particles */
1032 jq0 = _mm_load_sd(charge+jnrA+0);
1033 vdwjidx0A = 2*vdwtype[jnrA+0];
1034
1035 fjx0 = _mm_setzero_pd();
1036 fjy0 = _mm_setzero_pd();
1037 fjz0 = _mm_setzero_pd();
1038
1039 /**************************
1040 * CALCULATE INTERACTIONS *
1041 **************************/
1042
1043 /* Compute parameters for interactions between i and j atoms */
1044 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1045
1046 /* LENNARD-JONES DISPERSION/REPULSION */
1047
1048 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1049 fvdw = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1050
1051 fscal = fvdw;
1052
1053 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1054
1055 /* Calculate temporary vectorial force */
1056 tx = _mm_mul_pd(fscal,dx00);
1057 ty = _mm_mul_pd(fscal,dy00);
1058 tz = _mm_mul_pd(fscal,dz00);
1059
1060 /* Update vectorial force */
1061 fix0 = _mm_add_pd(fix0,tx);
1062 fiy0 = _mm_add_pd(fiy0,ty);
1063 fiz0 = _mm_add_pd(fiz0,tz);
1064
1065 fjx0 = _mm_add_pd(fjx0,tx);
1066 fjy0 = _mm_add_pd(fjy0,ty);
1067 fjz0 = _mm_add_pd(fjz0,tz);
1068
1069 /**************************
1070 * CALCULATE INTERACTIONS *
1071 **************************/
1072
1073 r10 = _mm_mul_pd(rsq10,rinv10);
1074
1075 /* Compute parameters for interactions between i and j atoms */
1076 qq10 = _mm_mul_pd(iq1,jq0);
1077
1078 /* EWALD ELECTROSTATICS */
1079
1080 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1081 ewrt = _mm_mul_pd(r10,ewtabscale);
1082 ewitab = _mm_cvttpd_epi32(ewrt);
1083 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1084 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1085 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1086 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1087
1088 fscal = felec;
1089
1090 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1091
1092 /* Calculate temporary vectorial force */
1093 tx = _mm_mul_pd(fscal,dx10);
1094 ty = _mm_mul_pd(fscal,dy10);
1095 tz = _mm_mul_pd(fscal,dz10);
1096
1097 /* Update vectorial force */
1098 fix1 = _mm_add_pd(fix1,tx);
1099 fiy1 = _mm_add_pd(fiy1,ty);
1100 fiz1 = _mm_add_pd(fiz1,tz);
1101
1102 fjx0 = _mm_add_pd(fjx0,tx);
1103 fjy0 = _mm_add_pd(fjy0,ty);
1104 fjz0 = _mm_add_pd(fjz0,tz);
1105
1106 /**************************
1107 * CALCULATE INTERACTIONS *
1108 **************************/
1109
1110 r20 = _mm_mul_pd(rsq20,rinv20);
1111
1112 /* Compute parameters for interactions between i and j atoms */
1113 qq20 = _mm_mul_pd(iq2,jq0);
1114
1115 /* EWALD ELECTROSTATICS */
1116
1117 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1118 ewrt = _mm_mul_pd(r20,ewtabscale);
1119 ewitab = _mm_cvttpd_epi32(ewrt);
1120 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1121 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1122 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1123 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1124
1125 fscal = felec;
1126
1127 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1128
1129 /* Calculate temporary vectorial force */
1130 tx = _mm_mul_pd(fscal,dx20);
1131 ty = _mm_mul_pd(fscal,dy20);
1132 tz = _mm_mul_pd(fscal,dz20);
1133
1134 /* Update vectorial force */
1135 fix2 = _mm_add_pd(fix2,tx);
1136 fiy2 = _mm_add_pd(fiy2,ty);
1137 fiz2 = _mm_add_pd(fiz2,tz);
1138
1139 fjx0 = _mm_add_pd(fjx0,tx);
1140 fjy0 = _mm_add_pd(fjy0,ty);
1141 fjz0 = _mm_add_pd(fjz0,tz);
1142
1143 /**************************
1144 * CALCULATE INTERACTIONS *
1145 **************************/
1146
1147 r30 = _mm_mul_pd(rsq30,rinv30);
1148
1149 /* Compute parameters for interactions between i and j atoms */
1150 qq30 = _mm_mul_pd(iq3,jq0);
1151
1152 /* EWALD ELECTROSTATICS */
1153
1154 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1155 ewrt = _mm_mul_pd(r30,ewtabscale);
1156 ewitab = _mm_cvttpd_epi32(ewrt);
1157 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1158 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1159 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1160 felec = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1161
1162 fscal = felec;
1163
1164 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1165
1166 /* Calculate temporary vectorial force */
1167 tx = _mm_mul_pd(fscal,dx30);
1168 ty = _mm_mul_pd(fscal,dy30);
1169 tz = _mm_mul_pd(fscal,dz30);
1170
1171 /* Update vectorial force */
1172 fix3 = _mm_add_pd(fix3,tx);
1173 fiy3 = _mm_add_pd(fiy3,ty);
1174 fiz3 = _mm_add_pd(fiz3,tz);
1175
1176 fjx0 = _mm_add_pd(fjx0,tx);
1177 fjy0 = _mm_add_pd(fjy0,ty);
1178 fjz0 = _mm_add_pd(fjz0,tz);
1179
1180 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1181
1182 /* Inner loop uses 138 flops */
1183 }
1184
1185 /* End of innermost loop */
1186
1187 gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1188 f+i_coord_offset,fshift+i_shift_offset);
1189
1190 /* Increment number of inner iterations */
1191 inneriter += j_index_end - j_index_start;
1192
1193 /* Outer loop uses 24 flops */
1194 }
1195
1196 /* Increment number of outer iterations */
1197 outeriter += nri;
1198
1199 /* Update outer/inner flops */
1200
1201 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*138);
1202 }
The ultimate molecular dynamics simulation package