?  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_single / nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_sse2_single.c
blob91a0ab503f99e56ca5a0484c228acc3f4ae9caaa
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_single 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_single.h"
49 #include "kernelutil_x86_sse2_single.h"
50
51 /*
52 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_VF_sse2_single
53 * Electrostatics interaction: ReactionField
54 * VdW interaction: LennardJones
55 * Geometry: Water3-Particle
56 * Calculate force/pot: PotentialAndForce
57 */
58 void
59 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_VF_sse2_single
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,C,D refer to j loop unrolling done with SSE, e.g. for the four 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,jnrC,jnrD;
76 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
77 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
78 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
79 real rcutoff_scalar;
80 real *shiftvec,*fshift,*x,*f;
81 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
82 real scratch[4*DIM];
83 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
84 int vdwioffset0;
85 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
86 int vdwioffset1;
87 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
88 int vdwioffset2;
89 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
90 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
91 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
92 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
93 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
94 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
95 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
96 real *charge;
97 int nvdwtype;
98 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
99 int *vdwtype;
100 real *vdwparam;
101 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
102 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
103 __m128 dummy_mask,cutoff_mask;
104 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
105 __m128 one = _mm_set1_ps(1.0);
106 __m128 two = _mm_set1_ps(2.0);
107 x = xx[0];
108 f = ff[0];
109
110 nri = nlist->nri;
111 iinr = nlist->iinr;
112 jindex = nlist->jindex;
113 jjnr = nlist->jjnr;
114 shiftidx = nlist->shift;
115 gid = nlist->gid;
116 shiftvec = fr->shift_vec[0];
117 fshift = fr->fshift[0];
118 facel = _mm_set1_ps(fr->epsfac);
119 charge = mdatoms->chargeA;
120 krf = _mm_set1_ps(fr->ic->k_rf);
121 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
122 crf = _mm_set1_ps(fr->ic->c_rf);
123 nvdwtype = fr->ntype;
124 vdwparam = fr->nbfp;
125 vdwtype = mdatoms->typeA;
126
127 /* Setup water-specific parameters */
128 inr = nlist->iinr[0];
129 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
130 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
131 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
132 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
133
134 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
135 rcutoff_scalar = fr->rcoulomb;
136 rcutoff = _mm_set1_ps(rcutoff_scalar);
137 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
138
139 sh_vdw_invrcut6 = _mm_set1_ps(fr->ic->sh_invrc6);
140 rvdw = _mm_set1_ps(fr->rvdw);
141
142 /* Avoid stupid compiler warnings */
143 jnrA = jnrB = jnrC = jnrD = 0;
144 j_coord_offsetA = 0;
145 j_coord_offsetB = 0;
146 j_coord_offsetC = 0;
147 j_coord_offsetD = 0;
148
149 outeriter = 0;
150 inneriter = 0;
151
152 for(iidx=0;iidx<4*DIM;iidx++)
153 {
154 scratch[iidx] = 0.0;
155 }
156
157 /* Start outer loop over neighborlists */
158 for(iidx=0; iidx<nri; iidx++)
159 {
160 /* Load shift vector for this list */
161 i_shift_offset = DIM*shiftidx[iidx];
162
163 /* Load limits for loop over neighbors */
164 j_index_start = jindex[iidx];
165 j_index_end = jindex[iidx+1];
166
167 /* Get outer coordinate index */
168 inr = iinr[iidx];
169 i_coord_offset = DIM*inr;
170
171 /* Load i particle coords and add shift vector */
172 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
173 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
174
175 fix0 = _mm_setzero_ps();
176 fiy0 = _mm_setzero_ps();
177 fiz0 = _mm_setzero_ps();
178 fix1 = _mm_setzero_ps();
179 fiy1 = _mm_setzero_ps();
180 fiz1 = _mm_setzero_ps();
181 fix2 = _mm_setzero_ps();
182 fiy2 = _mm_setzero_ps();
183 fiz2 = _mm_setzero_ps();
184
185 /* Reset potential sums */
186 velecsum = _mm_setzero_ps();
187 vvdwsum = _mm_setzero_ps();
188
189 /* Start inner kernel loop */
190 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
191 {
192
193 /* Get j neighbor index, and coordinate index */
194 jnrA = jjnr[jidx];
195 jnrB = jjnr[jidx+1];
196 jnrC = jjnr[jidx+2];
197 jnrD = jjnr[jidx+3];
198 j_coord_offsetA = DIM*jnrA;
199 j_coord_offsetB = DIM*jnrB;
200 j_coord_offsetC = DIM*jnrC;
201 j_coord_offsetD = DIM*jnrD;
202
203 /* load j atom coordinates */
204 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
205 x+j_coord_offsetC,x+j_coord_offsetD,
206 &jx0,&jy0,&jz0);
207
208 /* Calculate displacement vector */
209 dx00 = _mm_sub_ps(ix0,jx0);
210 dy00 = _mm_sub_ps(iy0,jy0);
211 dz00 = _mm_sub_ps(iz0,jz0);
212 dx10 = _mm_sub_ps(ix1,jx0);
213 dy10 = _mm_sub_ps(iy1,jy0);
214 dz10 = _mm_sub_ps(iz1,jz0);
215 dx20 = _mm_sub_ps(ix2,jx0);
216 dy20 = _mm_sub_ps(iy2,jy0);
217 dz20 = _mm_sub_ps(iz2,jz0);
218
219 /* Calculate squared distance and things based on it */
220 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
221 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
222 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
223
224 rinv00 = gmx_mm_invsqrt_ps(rsq00);
225 rinv10 = gmx_mm_invsqrt_ps(rsq10);
226 rinv20 = gmx_mm_invsqrt_ps(rsq20);
227
228 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
229 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
230 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
231
232 /* Load parameters for j particles */
233 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
234 charge+jnrC+0,charge+jnrD+0);
235 vdwjidx0A = 2*vdwtype[jnrA+0];
236 vdwjidx0B = 2*vdwtype[jnrB+0];
237 vdwjidx0C = 2*vdwtype[jnrC+0];
238 vdwjidx0D = 2*vdwtype[jnrD+0];
239
240 fjx0 = _mm_setzero_ps();
241 fjy0 = _mm_setzero_ps();
242 fjz0 = _mm_setzero_ps();
243
244 /**************************
245 * CALCULATE INTERACTIONS *
246 **************************/
247
248 if (gmx_mm_any_lt(rsq00,rcutoff2))
249 {
250
251 /* Compute parameters for interactions between i and j atoms */
252 qq00 = _mm_mul_ps(iq0,jq0);
253 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
254 vdwparam+vdwioffset0+vdwjidx0B,
255 vdwparam+vdwioffset0+vdwjidx0C,
256 vdwparam+vdwioffset0+vdwjidx0D,
257 &c6_00,&c12_00);
258
259 /* REACTION-FIELD ELECTROSTATICS */
260 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
261 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
262
263 /* LENNARD-JONES DISPERSION/REPULSION */
264
265 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
266 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
267 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
268 vvdw = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
269 _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
270 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
271
272 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
273
274 /* Update potential sum for this i atom from the interaction with this j atom. */
275 velec = _mm_and_ps(velec,cutoff_mask);
276 velecsum = _mm_add_ps(velecsum,velec);
277 vvdw = _mm_and_ps(vvdw,cutoff_mask);
278 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
279
280 fscal = _mm_add_ps(felec,fvdw);
281
282 fscal = _mm_and_ps(fscal,cutoff_mask);
283
284 /* Calculate temporary vectorial force */
285 tx = _mm_mul_ps(fscal,dx00);
286 ty = _mm_mul_ps(fscal,dy00);
287 tz = _mm_mul_ps(fscal,dz00);
288
289 /* Update vectorial force */
290 fix0 = _mm_add_ps(fix0,tx);
291 fiy0 = _mm_add_ps(fiy0,ty);
292 fiz0 = _mm_add_ps(fiz0,tz);
293
294 fjx0 = _mm_add_ps(fjx0,tx);
295 fjy0 = _mm_add_ps(fjy0,ty);
296 fjz0 = _mm_add_ps(fjz0,tz);
297
298 }
299
300 /**************************
301 * CALCULATE INTERACTIONS *
302 **************************/
303
304 if (gmx_mm_any_lt(rsq10,rcutoff2))
305 {
306
307 /* Compute parameters for interactions between i and j atoms */
308 qq10 = _mm_mul_ps(iq1,jq0);
309
310 /* REACTION-FIELD ELECTROSTATICS */
311 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
312 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
313
314 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
315
316 /* Update potential sum for this i atom from the interaction with this j atom. */
317 velec = _mm_and_ps(velec,cutoff_mask);
318 velecsum = _mm_add_ps(velecsum,velec);
319
320 fscal = felec;
321
322 fscal = _mm_and_ps(fscal,cutoff_mask);
323
324 /* Calculate temporary vectorial force */
325 tx = _mm_mul_ps(fscal,dx10);
326 ty = _mm_mul_ps(fscal,dy10);
327 tz = _mm_mul_ps(fscal,dz10);
328
329 /* Update vectorial force */
330 fix1 = _mm_add_ps(fix1,tx);
331 fiy1 = _mm_add_ps(fiy1,ty);
332 fiz1 = _mm_add_ps(fiz1,tz);
333
334 fjx0 = _mm_add_ps(fjx0,tx);
335 fjy0 = _mm_add_ps(fjy0,ty);
336 fjz0 = _mm_add_ps(fjz0,tz);
337
338 }
339
340 /**************************
341 * CALCULATE INTERACTIONS *
342 **************************/
343
344 if (gmx_mm_any_lt(rsq20,rcutoff2))
345 {
346
347 /* Compute parameters for interactions between i and j atoms */
348 qq20 = _mm_mul_ps(iq2,jq0);
349
350 /* REACTION-FIELD ELECTROSTATICS */
351 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
352 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
353
354 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
355
356 /* Update potential sum for this i atom from the interaction with this j atom. */
357 velec = _mm_and_ps(velec,cutoff_mask);
358 velecsum = _mm_add_ps(velecsum,velec);
359
360 fscal = felec;
361
362 fscal = _mm_and_ps(fscal,cutoff_mask);
363
364 /* Calculate temporary vectorial force */
365 tx = _mm_mul_ps(fscal,dx20);
366 ty = _mm_mul_ps(fscal,dy20);
367 tz = _mm_mul_ps(fscal,dz20);
368
369 /* Update vectorial force */
370 fix2 = _mm_add_ps(fix2,tx);
371 fiy2 = _mm_add_ps(fiy2,ty);
372 fiz2 = _mm_add_ps(fiz2,tz);
373
374 fjx0 = _mm_add_ps(fjx0,tx);
375 fjy0 = _mm_add_ps(fjy0,ty);
376 fjz0 = _mm_add_ps(fjz0,tz);
377
378 }
379
380 fjptrA = f+j_coord_offsetA;
381 fjptrB = f+j_coord_offsetB;
382 fjptrC = f+j_coord_offsetC;
383 fjptrD = f+j_coord_offsetD;
384
385 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
386
387 /* Inner loop uses 126 flops */
388 }
389
390 if(jidx<j_index_end)
391 {
392
393 /* Get j neighbor index, and coordinate index */
394 jnrlistA = jjnr[jidx];
395 jnrlistB = jjnr[jidx+1];
396 jnrlistC = jjnr[jidx+2];
397 jnrlistD = jjnr[jidx+3];
398 /* Sign of each element will be negative for non-real atoms.
399 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
400 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
401 */
402 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
403 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
404 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
405 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
406 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
407 j_coord_offsetA = DIM*jnrA;
408 j_coord_offsetB = DIM*jnrB;
409 j_coord_offsetC = DIM*jnrC;
410 j_coord_offsetD = DIM*jnrD;
411
412 /* load j atom coordinates */
413 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
414 x+j_coord_offsetC,x+j_coord_offsetD,
415 &jx0,&jy0,&jz0);
416
417 /* Calculate displacement vector */
418 dx00 = _mm_sub_ps(ix0,jx0);
419 dy00 = _mm_sub_ps(iy0,jy0);
420 dz00 = _mm_sub_ps(iz0,jz0);
421 dx10 = _mm_sub_ps(ix1,jx0);
422 dy10 = _mm_sub_ps(iy1,jy0);
423 dz10 = _mm_sub_ps(iz1,jz0);
424 dx20 = _mm_sub_ps(ix2,jx0);
425 dy20 = _mm_sub_ps(iy2,jy0);
426 dz20 = _mm_sub_ps(iz2,jz0);
427
428 /* Calculate squared distance and things based on it */
429 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
430 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
431 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
432
433 rinv00 = gmx_mm_invsqrt_ps(rsq00);
434 rinv10 = gmx_mm_invsqrt_ps(rsq10);
435 rinv20 = gmx_mm_invsqrt_ps(rsq20);
436
437 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
438 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
439 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
440
441 /* Load parameters for j particles */
442 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
443 charge+jnrC+0,charge+jnrD+0);
444 vdwjidx0A = 2*vdwtype[jnrA+0];
445 vdwjidx0B = 2*vdwtype[jnrB+0];
446 vdwjidx0C = 2*vdwtype[jnrC+0];
447 vdwjidx0D = 2*vdwtype[jnrD+0];
448
449 fjx0 = _mm_setzero_ps();
450 fjy0 = _mm_setzero_ps();
451 fjz0 = _mm_setzero_ps();
452
453 /**************************
454 * CALCULATE INTERACTIONS *
455 **************************/
456
457 if (gmx_mm_any_lt(rsq00,rcutoff2))
458 {
459
460 /* Compute parameters for interactions between i and j atoms */
461 qq00 = _mm_mul_ps(iq0,jq0);
462 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
463 vdwparam+vdwioffset0+vdwjidx0B,
464 vdwparam+vdwioffset0+vdwjidx0C,
465 vdwparam+vdwioffset0+vdwjidx0D,
466 &c6_00,&c12_00);
467
468 /* REACTION-FIELD ELECTROSTATICS */
469 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
470 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
471
472 /* LENNARD-JONES DISPERSION/REPULSION */
473
474 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
475 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
476 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
477 vvdw = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
478 _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
479 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
480
481 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
482
483 /* Update potential sum for this i atom from the interaction with this j atom. */
484 velec = _mm_and_ps(velec,cutoff_mask);
485 velec = _mm_andnot_ps(dummy_mask,velec);
486 velecsum = _mm_add_ps(velecsum,velec);
487 vvdw = _mm_and_ps(vvdw,cutoff_mask);
488 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
489 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
490
491 fscal = _mm_add_ps(felec,fvdw);
492
493 fscal = _mm_and_ps(fscal,cutoff_mask);
494
495 fscal = _mm_andnot_ps(dummy_mask,fscal);
496
497 /* Calculate temporary vectorial force */
498 tx = _mm_mul_ps(fscal,dx00);
499 ty = _mm_mul_ps(fscal,dy00);
500 tz = _mm_mul_ps(fscal,dz00);
501
502 /* Update vectorial force */
503 fix0 = _mm_add_ps(fix0,tx);
504 fiy0 = _mm_add_ps(fiy0,ty);
505 fiz0 = _mm_add_ps(fiz0,tz);
506
507 fjx0 = _mm_add_ps(fjx0,tx);
508 fjy0 = _mm_add_ps(fjy0,ty);
509 fjz0 = _mm_add_ps(fjz0,tz);
510
511 }
512
513 /**************************
514 * CALCULATE INTERACTIONS *
515 **************************/
516
517 if (gmx_mm_any_lt(rsq10,rcutoff2))
518 {
519
520 /* Compute parameters for interactions between i and j atoms */
521 qq10 = _mm_mul_ps(iq1,jq0);
522
523 /* REACTION-FIELD ELECTROSTATICS */
524 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
525 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
526
527 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
528
529 /* Update potential sum for this i atom from the interaction with this j atom. */
530 velec = _mm_and_ps(velec,cutoff_mask);
531 velec = _mm_andnot_ps(dummy_mask,velec);
532 velecsum = _mm_add_ps(velecsum,velec);
533
534 fscal = felec;
535
536 fscal = _mm_and_ps(fscal,cutoff_mask);
537
538 fscal = _mm_andnot_ps(dummy_mask,fscal);
539
540 /* Calculate temporary vectorial force */
541 tx = _mm_mul_ps(fscal,dx10);
542 ty = _mm_mul_ps(fscal,dy10);
543 tz = _mm_mul_ps(fscal,dz10);
544
545 /* Update vectorial force */
546 fix1 = _mm_add_ps(fix1,tx);
547 fiy1 = _mm_add_ps(fiy1,ty);
548 fiz1 = _mm_add_ps(fiz1,tz);
549
550 fjx0 = _mm_add_ps(fjx0,tx);
551 fjy0 = _mm_add_ps(fjy0,ty);
552 fjz0 = _mm_add_ps(fjz0,tz);
553
554 }
555
556 /**************************
557 * CALCULATE INTERACTIONS *
558 **************************/
559
560 if (gmx_mm_any_lt(rsq20,rcutoff2))
561 {
562
563 /* Compute parameters for interactions between i and j atoms */
564 qq20 = _mm_mul_ps(iq2,jq0);
565
566 /* REACTION-FIELD ELECTROSTATICS */
567 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
568 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
569
570 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
571
572 /* Update potential sum for this i atom from the interaction with this j atom. */
573 velec = _mm_and_ps(velec,cutoff_mask);
574 velec = _mm_andnot_ps(dummy_mask,velec);
575 velecsum = _mm_add_ps(velecsum,velec);
576
577 fscal = felec;
578
579 fscal = _mm_and_ps(fscal,cutoff_mask);
580
581 fscal = _mm_andnot_ps(dummy_mask,fscal);
582
583 /* Calculate temporary vectorial force */
584 tx = _mm_mul_ps(fscal,dx20);
585 ty = _mm_mul_ps(fscal,dy20);
586 tz = _mm_mul_ps(fscal,dz20);
587
588 /* Update vectorial force */
589 fix2 = _mm_add_ps(fix2,tx);
590 fiy2 = _mm_add_ps(fiy2,ty);
591 fiz2 = _mm_add_ps(fiz2,tz);
592
593 fjx0 = _mm_add_ps(fjx0,tx);
594 fjy0 = _mm_add_ps(fjy0,ty);
595 fjz0 = _mm_add_ps(fjz0,tz);
596
597 }
598
599 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
600 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
601 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
602 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
603
604 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
605
606 /* Inner loop uses 126 flops */
607 }
608
609 /* End of innermost loop */
610
611 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
612 f+i_coord_offset,fshift+i_shift_offset);
613
614 ggid = gid[iidx];
615 /* Update potential energies */
616 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
617 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
618
619 /* Increment number of inner iterations */
620 inneriter += j_index_end - j_index_start;
621
622 /* Outer loop uses 20 flops */
623 }
624
625 /* Increment number of outer iterations */
626 outeriter += nri;
627
628 /* Update outer/inner flops */
629
630 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*126);
631 }
632 /*
633 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_sse2_single
634 * Electrostatics interaction: ReactionField
635 * VdW interaction: LennardJones
636 * Geometry: Water3-Particle
637 * Calculate force/pot: Force
638 */
639 void
640 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_sse2_single
641 (t_nblist * gmx_restrict nlist,
642 rvec * gmx_restrict xx,
643 rvec * gmx_restrict ff,
644 t_forcerec * gmx_restrict fr,
645 t_mdatoms * gmx_restrict mdatoms,
646 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
647 t_nrnb * gmx_restrict nrnb)
648 {
649 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
650 * just 0 for non-waters.
651 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
652 * jnr indices corresponding to data put in the four positions in the SIMD register.
653 */
654 int i_shift_offset,i_coord_offset,outeriter,inneriter;
655 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
656 int jnrA,jnrB,jnrC,jnrD;
657 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
658 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
659 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
660 real rcutoff_scalar;
661 real *shiftvec,*fshift,*x,*f;
662 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
663 real scratch[4*DIM];
664 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
665 int vdwioffset0;
666 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
667 int vdwioffset1;
668 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
669 int vdwioffset2;
670 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
671 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
672 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
673 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
674 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
675 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
676 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
677 real *charge;
678 int nvdwtype;
679 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
680 int *vdwtype;
681 real *vdwparam;
682 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
683 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
684 __m128 dummy_mask,cutoff_mask;
685 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
686 __m128 one = _mm_set1_ps(1.0);
687 __m128 two = _mm_set1_ps(2.0);
688 x = xx[0];
689 f = ff[0];
690
691 nri = nlist->nri;
692 iinr = nlist->iinr;
693 jindex = nlist->jindex;
694 jjnr = nlist->jjnr;
695 shiftidx = nlist->shift;
696 gid = nlist->gid;
697 shiftvec = fr->shift_vec[0];
698 fshift = fr->fshift[0];
699 facel = _mm_set1_ps(fr->epsfac);
700 charge = mdatoms->chargeA;
701 krf = _mm_set1_ps(fr->ic->k_rf);
702 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
703 crf = _mm_set1_ps(fr->ic->c_rf);
704 nvdwtype = fr->ntype;
705 vdwparam = fr->nbfp;
706 vdwtype = mdatoms->typeA;
707
708 /* Setup water-specific parameters */
709 inr = nlist->iinr[0];
710 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
711 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
712 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
713 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
714
715 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
716 rcutoff_scalar = fr->rcoulomb;
717 rcutoff = _mm_set1_ps(rcutoff_scalar);
718 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
719
720 sh_vdw_invrcut6 = _mm_set1_ps(fr->ic->sh_invrc6);
721 rvdw = _mm_set1_ps(fr->rvdw);
722
723 /* Avoid stupid compiler warnings */
724 jnrA = jnrB = jnrC = jnrD = 0;
725 j_coord_offsetA = 0;
726 j_coord_offsetB = 0;
727 j_coord_offsetC = 0;
728 j_coord_offsetD = 0;
729
730 outeriter = 0;
731 inneriter = 0;
732
733 for(iidx=0;iidx<4*DIM;iidx++)
734 {
735 scratch[iidx] = 0.0;
736 }
737
738 /* Start outer loop over neighborlists */
739 for(iidx=0; iidx<nri; iidx++)
740 {
741 /* Load shift vector for this list */
742 i_shift_offset = DIM*shiftidx[iidx];
743
744 /* Load limits for loop over neighbors */
745 j_index_start = jindex[iidx];
746 j_index_end = jindex[iidx+1];
747
748 /* Get outer coordinate index */
749 inr = iinr[iidx];
750 i_coord_offset = DIM*inr;
751
752 /* Load i particle coords and add shift vector */
753 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
754 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
755
756 fix0 = _mm_setzero_ps();
757 fiy0 = _mm_setzero_ps();
758 fiz0 = _mm_setzero_ps();
759 fix1 = _mm_setzero_ps();
760 fiy1 = _mm_setzero_ps();
761 fiz1 = _mm_setzero_ps();
762 fix2 = _mm_setzero_ps();
763 fiy2 = _mm_setzero_ps();
764 fiz2 = _mm_setzero_ps();
765
766 /* Start inner kernel loop */
767 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
768 {
769
770 /* Get j neighbor index, and coordinate index */
771 jnrA = jjnr[jidx];
772 jnrB = jjnr[jidx+1];
773 jnrC = jjnr[jidx+2];
774 jnrD = jjnr[jidx+3];
775 j_coord_offsetA = DIM*jnrA;
776 j_coord_offsetB = DIM*jnrB;
777 j_coord_offsetC = DIM*jnrC;
778 j_coord_offsetD = DIM*jnrD;
779
780 /* load j atom coordinates */
781 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
782 x+j_coord_offsetC,x+j_coord_offsetD,
783 &jx0,&jy0,&jz0);
784
785 /* Calculate displacement vector */
786 dx00 = _mm_sub_ps(ix0,jx0);
787 dy00 = _mm_sub_ps(iy0,jy0);
788 dz00 = _mm_sub_ps(iz0,jz0);
789 dx10 = _mm_sub_ps(ix1,jx0);
790 dy10 = _mm_sub_ps(iy1,jy0);
791 dz10 = _mm_sub_ps(iz1,jz0);
792 dx20 = _mm_sub_ps(ix2,jx0);
793 dy20 = _mm_sub_ps(iy2,jy0);
794 dz20 = _mm_sub_ps(iz2,jz0);
795
796 /* Calculate squared distance and things based on it */
797 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
798 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
799 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
800
801 rinv00 = gmx_mm_invsqrt_ps(rsq00);
802 rinv10 = gmx_mm_invsqrt_ps(rsq10);
803 rinv20 = gmx_mm_invsqrt_ps(rsq20);
804
805 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
806 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
807 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
808
809 /* Load parameters for j particles */
810 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
811 charge+jnrC+0,charge+jnrD+0);
812 vdwjidx0A = 2*vdwtype[jnrA+0];
813 vdwjidx0B = 2*vdwtype[jnrB+0];
814 vdwjidx0C = 2*vdwtype[jnrC+0];
815 vdwjidx0D = 2*vdwtype[jnrD+0];
816
817 fjx0 = _mm_setzero_ps();
818 fjy0 = _mm_setzero_ps();
819 fjz0 = _mm_setzero_ps();
820
821 /**************************
822 * CALCULATE INTERACTIONS *
823 **************************/
824
825 if (gmx_mm_any_lt(rsq00,rcutoff2))
826 {
827
828 /* Compute parameters for interactions between i and j atoms */
829 qq00 = _mm_mul_ps(iq0,jq0);
830 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
831 vdwparam+vdwioffset0+vdwjidx0B,
832 vdwparam+vdwioffset0+vdwjidx0C,
833 vdwparam+vdwioffset0+vdwjidx0D,
834 &c6_00,&c12_00);
835
836 /* REACTION-FIELD ELECTROSTATICS */
837 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
838
839 /* LENNARD-JONES DISPERSION/REPULSION */
840
841 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
842 fvdw = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
843
844 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
845
846 fscal = _mm_add_ps(felec,fvdw);
847
848 fscal = _mm_and_ps(fscal,cutoff_mask);
849
850 /* Calculate temporary vectorial force */
851 tx = _mm_mul_ps(fscal,dx00);
852 ty = _mm_mul_ps(fscal,dy00);
853 tz = _mm_mul_ps(fscal,dz00);
854
855 /* Update vectorial force */
856 fix0 = _mm_add_ps(fix0,tx);
857 fiy0 = _mm_add_ps(fiy0,ty);
858 fiz0 = _mm_add_ps(fiz0,tz);
859
860 fjx0 = _mm_add_ps(fjx0,tx);
861 fjy0 = _mm_add_ps(fjy0,ty);
862 fjz0 = _mm_add_ps(fjz0,tz);
863
864 }
865
866 /**************************
867 * CALCULATE INTERACTIONS *
868 **************************/
869
870 if (gmx_mm_any_lt(rsq10,rcutoff2))
871 {
872
873 /* Compute parameters for interactions between i and j atoms */
874 qq10 = _mm_mul_ps(iq1,jq0);
875
876 /* REACTION-FIELD ELECTROSTATICS */
877 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
878
879 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
880
881 fscal = felec;
882
883 fscal = _mm_and_ps(fscal,cutoff_mask);
884
885 /* Calculate temporary vectorial force */
886 tx = _mm_mul_ps(fscal,dx10);
887 ty = _mm_mul_ps(fscal,dy10);
888 tz = _mm_mul_ps(fscal,dz10);
889
890 /* Update vectorial force */
891 fix1 = _mm_add_ps(fix1,tx);
892 fiy1 = _mm_add_ps(fiy1,ty);
893 fiz1 = _mm_add_ps(fiz1,tz);
894
895 fjx0 = _mm_add_ps(fjx0,tx);
896 fjy0 = _mm_add_ps(fjy0,ty);
897 fjz0 = _mm_add_ps(fjz0,tz);
898
899 }
900
901 /**************************
902 * CALCULATE INTERACTIONS *
903 **************************/
904
905 if (gmx_mm_any_lt(rsq20,rcutoff2))
906 {
907
908 /* Compute parameters for interactions between i and j atoms */
909 qq20 = _mm_mul_ps(iq2,jq0);
910
911 /* REACTION-FIELD ELECTROSTATICS */
912 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
913
914 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
915
916 fscal = felec;
917
918 fscal = _mm_and_ps(fscal,cutoff_mask);
919
920 /* Calculate temporary vectorial force */
921 tx = _mm_mul_ps(fscal,dx20);
922 ty = _mm_mul_ps(fscal,dy20);
923 tz = _mm_mul_ps(fscal,dz20);
924
925 /* Update vectorial force */
926 fix2 = _mm_add_ps(fix2,tx);
927 fiy2 = _mm_add_ps(fiy2,ty);
928 fiz2 = _mm_add_ps(fiz2,tz);
929
930 fjx0 = _mm_add_ps(fjx0,tx);
931 fjy0 = _mm_add_ps(fjy0,ty);
932 fjz0 = _mm_add_ps(fjz0,tz);
933
934 }
935
936 fjptrA = f+j_coord_offsetA;
937 fjptrB = f+j_coord_offsetB;
938 fjptrC = f+j_coord_offsetC;
939 fjptrD = f+j_coord_offsetD;
940
941 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
942
943 /* Inner loop uses 97 flops */
944 }
945
946 if(jidx<j_index_end)
947 {
948
949 /* Get j neighbor index, and coordinate index */
950 jnrlistA = jjnr[jidx];
951 jnrlistB = jjnr[jidx+1];
952 jnrlistC = jjnr[jidx+2];
953 jnrlistD = jjnr[jidx+3];
954 /* Sign of each element will be negative for non-real atoms.
955 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
956 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
957 */
958 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
959 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
960 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
961 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
962 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
963 j_coord_offsetA = DIM*jnrA;
964 j_coord_offsetB = DIM*jnrB;
965 j_coord_offsetC = DIM*jnrC;
966 j_coord_offsetD = DIM*jnrD;
967
968 /* load j atom coordinates */
969 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
970 x+j_coord_offsetC,x+j_coord_offsetD,
971 &jx0,&jy0,&jz0);
972
973 /* Calculate displacement vector */
974 dx00 = _mm_sub_ps(ix0,jx0);
975 dy00 = _mm_sub_ps(iy0,jy0);
976 dz00 = _mm_sub_ps(iz0,jz0);
977 dx10 = _mm_sub_ps(ix1,jx0);
978 dy10 = _mm_sub_ps(iy1,jy0);
979 dz10 = _mm_sub_ps(iz1,jz0);
980 dx20 = _mm_sub_ps(ix2,jx0);
981 dy20 = _mm_sub_ps(iy2,jy0);
982 dz20 = _mm_sub_ps(iz2,jz0);
983
984 /* Calculate squared distance and things based on it */
985 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
986 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
987 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
988
989 rinv00 = gmx_mm_invsqrt_ps(rsq00);
990 rinv10 = gmx_mm_invsqrt_ps(rsq10);
991 rinv20 = gmx_mm_invsqrt_ps(rsq20);
992
993 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
994 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
995 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
996
997 /* Load parameters for j particles */
998 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
999 charge+jnrC+0,charge+jnrD+0);
1000 vdwjidx0A = 2*vdwtype[jnrA+0];
1001 vdwjidx0B = 2*vdwtype[jnrB+0];
1002 vdwjidx0C = 2*vdwtype[jnrC+0];
1003 vdwjidx0D = 2*vdwtype[jnrD+0];
1004
1005 fjx0 = _mm_setzero_ps();
1006 fjy0 = _mm_setzero_ps();
1007 fjz0 = _mm_setzero_ps();
1008
1009 /**************************
1010 * CALCULATE INTERACTIONS *
1011 **************************/
1012
1013 if (gmx_mm_any_lt(rsq00,rcutoff2))
1014 {
1015
1016 /* Compute parameters for interactions between i and j atoms */
1017 qq00 = _mm_mul_ps(iq0,jq0);
1018 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1019 vdwparam+vdwioffset0+vdwjidx0B,
1020 vdwparam+vdwioffset0+vdwjidx0C,
1021 vdwparam+vdwioffset0+vdwjidx0D,
1022 &c6_00,&c12_00);
1023
1024 /* REACTION-FIELD ELECTROSTATICS */
1025 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
1026
1027 /* LENNARD-JONES DISPERSION/REPULSION */
1028
1029 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1030 fvdw = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1031
1032 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
1033
1034 fscal = _mm_add_ps(felec,fvdw);
1035
1036 fscal = _mm_and_ps(fscal,cutoff_mask);
1037
1038 fscal = _mm_andnot_ps(dummy_mask,fscal);
1039
1040 /* Calculate temporary vectorial force */
1041 tx = _mm_mul_ps(fscal,dx00);
1042 ty = _mm_mul_ps(fscal,dy00);
1043 tz = _mm_mul_ps(fscal,dz00);
1044
1045 /* Update vectorial force */
1046 fix0 = _mm_add_ps(fix0,tx);
1047 fiy0 = _mm_add_ps(fiy0,ty);
1048 fiz0 = _mm_add_ps(fiz0,tz);
1049
1050 fjx0 = _mm_add_ps(fjx0,tx);
1051 fjy0 = _mm_add_ps(fjy0,ty);
1052 fjz0 = _mm_add_ps(fjz0,tz);
1053
1054 }
1055
1056 /**************************
1057 * CALCULATE INTERACTIONS *
1058 **************************/
1059
1060 if (gmx_mm_any_lt(rsq10,rcutoff2))
1061 {
1062
1063 /* Compute parameters for interactions between i and j atoms */
1064 qq10 = _mm_mul_ps(iq1,jq0);
1065
1066 /* REACTION-FIELD ELECTROSTATICS */
1067 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1068
1069 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
1070
1071 fscal = felec;
1072
1073 fscal = _mm_and_ps(fscal,cutoff_mask);
1074
1075 fscal = _mm_andnot_ps(dummy_mask,fscal);
1076
1077 /* Calculate temporary vectorial force */
1078 tx = _mm_mul_ps(fscal,dx10);
1079 ty = _mm_mul_ps(fscal,dy10);
1080 tz = _mm_mul_ps(fscal,dz10);
1081
1082 /* Update vectorial force */
1083 fix1 = _mm_add_ps(fix1,tx);
1084 fiy1 = _mm_add_ps(fiy1,ty);
1085 fiz1 = _mm_add_ps(fiz1,tz);
1086
1087 fjx0 = _mm_add_ps(fjx0,tx);
1088 fjy0 = _mm_add_ps(fjy0,ty);
1089 fjz0 = _mm_add_ps(fjz0,tz);
1090
1091 }
1092
1093 /**************************
1094 * CALCULATE INTERACTIONS *
1095 **************************/
1096
1097 if (gmx_mm_any_lt(rsq20,rcutoff2))
1098 {
1099
1100 /* Compute parameters for interactions between i and j atoms */
1101 qq20 = _mm_mul_ps(iq2,jq0);
1102
1103 /* REACTION-FIELD ELECTROSTATICS */
1104 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1105
1106 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
1107
1108 fscal = felec;
1109
1110 fscal = _mm_and_ps(fscal,cutoff_mask);
1111
1112 fscal = _mm_andnot_ps(dummy_mask,fscal);
1113
1114 /* Calculate temporary vectorial force */
1115 tx = _mm_mul_ps(fscal,dx20);
1116 ty = _mm_mul_ps(fscal,dy20);
1117 tz = _mm_mul_ps(fscal,dz20);
1118
1119 /* Update vectorial force */
1120 fix2 = _mm_add_ps(fix2,tx);
1121 fiy2 = _mm_add_ps(fiy2,ty);
1122 fiz2 = _mm_add_ps(fiz2,tz);
1123
1124 fjx0 = _mm_add_ps(fjx0,tx);
1125 fjy0 = _mm_add_ps(fjy0,ty);
1126 fjz0 = _mm_add_ps(fjz0,tz);
1127
1128 }
1129
1130 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1131 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1132 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1133 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1134
1135 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1136
1137 /* Inner loop uses 97 flops */
1138 }
1139
1140 /* End of innermost loop */
1141
1142 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1143 f+i_coord_offset,fshift+i_shift_offset);
1144
1145 /* Increment number of inner iterations */
1146 inneriter += j_index_end - j_index_start;
1147
1148 /* Outer loop uses 18 flops */
1149 }
1150
1151 /* Increment number of outer iterations */
1152 outeriter += nri;
1153
1154 /* Update outer/inner flops */
1155
1156 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*97);
1157 }
The ultimate molecular dynamics simulation package