?  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_VdwLJSw_GeomW4P1_sse2_single.c
blob42bb1e09e81f80b4e8b50c3f444fee7085e95f40
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_VdwLJSw_GeomW4P1_VF_sse2_single
53 * Electrostatics interaction: ReactionField
54 * VdW interaction: LennardJones
55 * Geometry: Water4-Particle
56 * Calculate force/pot: PotentialAndForce
57 */
58 void
59 nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_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 vdwioffset3;
91 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
92 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
93 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
94 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
95 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
96 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
97 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
98 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
99 real *charge;
100 int nvdwtype;
101 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
102 int *vdwtype;
103 real *vdwparam;
104 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
105 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
106 __m128 rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
107 real rswitch_scalar,d_scalar;
108 __m128 dummy_mask,cutoff_mask;
109 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
110 __m128 one = _mm_set1_ps(1.0);
111 __m128 two = _mm_set1_ps(2.0);
112 x = xx[0];
113 f = ff[0];
114
115 nri = nlist->nri;
116 iinr = nlist->iinr;
117 jindex = nlist->jindex;
118 jjnr = nlist->jjnr;
119 shiftidx = nlist->shift;
120 gid = nlist->gid;
121 shiftvec = fr->shift_vec[0];
122 fshift = fr->fshift[0];
123 facel = _mm_set1_ps(fr->epsfac);
124 charge = mdatoms->chargeA;
125 krf = _mm_set1_ps(fr->ic->k_rf);
126 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
127 crf = _mm_set1_ps(fr->ic->c_rf);
128 nvdwtype = fr->ntype;
129 vdwparam = fr->nbfp;
130 vdwtype = mdatoms->typeA;
131
132 /* Setup water-specific parameters */
133 inr = nlist->iinr[0];
134 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
135 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
136 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
137 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
138
139 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
140 rcutoff_scalar = fr->rcoulomb;
141 rcutoff = _mm_set1_ps(rcutoff_scalar);
142 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
143
144 rswitch_scalar = fr->rvdw_switch;
145 rswitch = _mm_set1_ps(rswitch_scalar);
146 /* Setup switch parameters */
147 d_scalar = rcutoff_scalar-rswitch_scalar;
148 d = _mm_set1_ps(d_scalar);
149 swV3 = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
150 swV4 = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
151 swV5 = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
152 swF2 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
153 swF3 = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
154 swF4 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
155
156 /* Avoid stupid compiler warnings */
157 jnrA = jnrB = jnrC = jnrD = 0;
158 j_coord_offsetA = 0;
159 j_coord_offsetB = 0;
160 j_coord_offsetC = 0;
161 j_coord_offsetD = 0;
162
163 outeriter = 0;
164 inneriter = 0;
165
166 for(iidx=0;iidx<4*DIM;iidx++)
167 {
168 scratch[iidx] = 0.0;
169 }
170
171 /* Start outer loop over neighborlists */
172 for(iidx=0; iidx<nri; iidx++)
173 {
174 /* Load shift vector for this list */
175 i_shift_offset = DIM*shiftidx[iidx];
176
177 /* Load limits for loop over neighbors */
178 j_index_start = jindex[iidx];
179 j_index_end = jindex[iidx+1];
180
181 /* Get outer coordinate index */
182 inr = iinr[iidx];
183 i_coord_offset = DIM*inr;
184
185 /* Load i particle coords and add shift vector */
186 gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
187 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
188
189 fix0 = _mm_setzero_ps();
190 fiy0 = _mm_setzero_ps();
191 fiz0 = _mm_setzero_ps();
192 fix1 = _mm_setzero_ps();
193 fiy1 = _mm_setzero_ps();
194 fiz1 = _mm_setzero_ps();
195 fix2 = _mm_setzero_ps();
196 fiy2 = _mm_setzero_ps();
197 fiz2 = _mm_setzero_ps();
198 fix3 = _mm_setzero_ps();
199 fiy3 = _mm_setzero_ps();
200 fiz3 = _mm_setzero_ps();
201
202 /* Reset potential sums */
203 velecsum = _mm_setzero_ps();
204 vvdwsum = _mm_setzero_ps();
205
206 /* Start inner kernel loop */
207 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
208 {
209
210 /* Get j neighbor index, and coordinate index */
211 jnrA = jjnr[jidx];
212 jnrB = jjnr[jidx+1];
213 jnrC = jjnr[jidx+2];
214 jnrD = jjnr[jidx+3];
215 j_coord_offsetA = DIM*jnrA;
216 j_coord_offsetB = DIM*jnrB;
217 j_coord_offsetC = DIM*jnrC;
218 j_coord_offsetD = DIM*jnrD;
219
220 /* load j atom coordinates */
221 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
222 x+j_coord_offsetC,x+j_coord_offsetD,
223 &jx0,&jy0,&jz0);
224
225 /* Calculate displacement vector */
226 dx00 = _mm_sub_ps(ix0,jx0);
227 dy00 = _mm_sub_ps(iy0,jy0);
228 dz00 = _mm_sub_ps(iz0,jz0);
229 dx10 = _mm_sub_ps(ix1,jx0);
230 dy10 = _mm_sub_ps(iy1,jy0);
231 dz10 = _mm_sub_ps(iz1,jz0);
232 dx20 = _mm_sub_ps(ix2,jx0);
233 dy20 = _mm_sub_ps(iy2,jy0);
234 dz20 = _mm_sub_ps(iz2,jz0);
235 dx30 = _mm_sub_ps(ix3,jx0);
236 dy30 = _mm_sub_ps(iy3,jy0);
237 dz30 = _mm_sub_ps(iz3,jz0);
238
239 /* Calculate squared distance and things based on it */
240 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
241 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
242 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
243 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
244
245 rinv00 = gmx_mm_invsqrt_ps(rsq00);
246 rinv10 = gmx_mm_invsqrt_ps(rsq10);
247 rinv20 = gmx_mm_invsqrt_ps(rsq20);
248 rinv30 = gmx_mm_invsqrt_ps(rsq30);
249
250 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
251 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
252 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
253 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
254
255 /* Load parameters for j particles */
256 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
257 charge+jnrC+0,charge+jnrD+0);
258 vdwjidx0A = 2*vdwtype[jnrA+0];
259 vdwjidx0B = 2*vdwtype[jnrB+0];
260 vdwjidx0C = 2*vdwtype[jnrC+0];
261 vdwjidx0D = 2*vdwtype[jnrD+0];
262
263 fjx0 = _mm_setzero_ps();
264 fjy0 = _mm_setzero_ps();
265 fjz0 = _mm_setzero_ps();
266
267 /**************************
268 * CALCULATE INTERACTIONS *
269 **************************/
270
271 if (gmx_mm_any_lt(rsq00,rcutoff2))
272 {
273
274 r00 = _mm_mul_ps(rsq00,rinv00);
275
276 /* Compute parameters for interactions between i and j atoms */
277 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
278 vdwparam+vdwioffset0+vdwjidx0B,
279 vdwparam+vdwioffset0+vdwjidx0C,
280 vdwparam+vdwioffset0+vdwjidx0D,
281 &c6_00,&c12_00);
282
283 /* LENNARD-JONES DISPERSION/REPULSION */
284
285 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
286 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
287 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
288 vvdw = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
289 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
290
291 d = _mm_sub_ps(r00,rswitch);
292 d = _mm_max_ps(d,_mm_setzero_ps());
293 d2 = _mm_mul_ps(d,d);
294 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
295
296 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
297
298 /* Evaluate switch function */
299 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
300 fvdw = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
301 vvdw = _mm_mul_ps(vvdw,sw);
302 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
303
304 /* Update potential sum for this i atom from the interaction with this j atom. */
305 vvdw = _mm_and_ps(vvdw,cutoff_mask);
306 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
307
308 fscal = fvdw;
309
310 fscal = _mm_and_ps(fscal,cutoff_mask);
311
312 /* Calculate temporary vectorial force */
313 tx = _mm_mul_ps(fscal,dx00);
314 ty = _mm_mul_ps(fscal,dy00);
315 tz = _mm_mul_ps(fscal,dz00);
316
317 /* Update vectorial force */
318 fix0 = _mm_add_ps(fix0,tx);
319 fiy0 = _mm_add_ps(fiy0,ty);
320 fiz0 = _mm_add_ps(fiz0,tz);
321
322 fjx0 = _mm_add_ps(fjx0,tx);
323 fjy0 = _mm_add_ps(fjy0,ty);
324 fjz0 = _mm_add_ps(fjz0,tz);
325
326 }
327
328 /**************************
329 * CALCULATE INTERACTIONS *
330 **************************/
331
332 if (gmx_mm_any_lt(rsq10,rcutoff2))
333 {
334
335 /* Compute parameters for interactions between i and j atoms */
336 qq10 = _mm_mul_ps(iq1,jq0);
337
338 /* REACTION-FIELD ELECTROSTATICS */
339 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
340 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
341
342 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
343
344 /* Update potential sum for this i atom from the interaction with this j atom. */
345 velec = _mm_and_ps(velec,cutoff_mask);
346 velecsum = _mm_add_ps(velecsum,velec);
347
348 fscal = felec;
349
350 fscal = _mm_and_ps(fscal,cutoff_mask);
351
352 /* Calculate temporary vectorial force */
353 tx = _mm_mul_ps(fscal,dx10);
354 ty = _mm_mul_ps(fscal,dy10);
355 tz = _mm_mul_ps(fscal,dz10);
356
357 /* Update vectorial force */
358 fix1 = _mm_add_ps(fix1,tx);
359 fiy1 = _mm_add_ps(fiy1,ty);
360 fiz1 = _mm_add_ps(fiz1,tz);
361
362 fjx0 = _mm_add_ps(fjx0,tx);
363 fjy0 = _mm_add_ps(fjy0,ty);
364 fjz0 = _mm_add_ps(fjz0,tz);
365
366 }
367
368 /**************************
369 * CALCULATE INTERACTIONS *
370 **************************/
371
372 if (gmx_mm_any_lt(rsq20,rcutoff2))
373 {
374
375 /* Compute parameters for interactions between i and j atoms */
376 qq20 = _mm_mul_ps(iq2,jq0);
377
378 /* REACTION-FIELD ELECTROSTATICS */
379 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
380 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
381
382 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
383
384 /* Update potential sum for this i atom from the interaction with this j atom. */
385 velec = _mm_and_ps(velec,cutoff_mask);
386 velecsum = _mm_add_ps(velecsum,velec);
387
388 fscal = felec;
389
390 fscal = _mm_and_ps(fscal,cutoff_mask);
391
392 /* Calculate temporary vectorial force */
393 tx = _mm_mul_ps(fscal,dx20);
394 ty = _mm_mul_ps(fscal,dy20);
395 tz = _mm_mul_ps(fscal,dz20);
396
397 /* Update vectorial force */
398 fix2 = _mm_add_ps(fix2,tx);
399 fiy2 = _mm_add_ps(fiy2,ty);
400 fiz2 = _mm_add_ps(fiz2,tz);
401
402 fjx0 = _mm_add_ps(fjx0,tx);
403 fjy0 = _mm_add_ps(fjy0,ty);
404 fjz0 = _mm_add_ps(fjz0,tz);
405
406 }
407
408 /**************************
409 * CALCULATE INTERACTIONS *
410 **************************/
411
412 if (gmx_mm_any_lt(rsq30,rcutoff2))
413 {
414
415 /* Compute parameters for interactions between i and j atoms */
416 qq30 = _mm_mul_ps(iq3,jq0);
417
418 /* REACTION-FIELD ELECTROSTATICS */
419 velec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
420 felec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
421
422 cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2);
423
424 /* Update potential sum for this i atom from the interaction with this j atom. */
425 velec = _mm_and_ps(velec,cutoff_mask);
426 velecsum = _mm_add_ps(velecsum,velec);
427
428 fscal = felec;
429
430 fscal = _mm_and_ps(fscal,cutoff_mask);
431
432 /* Calculate temporary vectorial force */
433 tx = _mm_mul_ps(fscal,dx30);
434 ty = _mm_mul_ps(fscal,dy30);
435 tz = _mm_mul_ps(fscal,dz30);
436
437 /* Update vectorial force */
438 fix3 = _mm_add_ps(fix3,tx);
439 fiy3 = _mm_add_ps(fiy3,ty);
440 fiz3 = _mm_add_ps(fiz3,tz);
441
442 fjx0 = _mm_add_ps(fjx0,tx);
443 fjy0 = _mm_add_ps(fjy0,ty);
444 fjz0 = _mm_add_ps(fjz0,tz);
445
446 }
447
448 fjptrA = f+j_coord_offsetA;
449 fjptrB = f+j_coord_offsetB;
450 fjptrC = f+j_coord_offsetC;
451 fjptrD = f+j_coord_offsetD;
452
453 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
454
455 /* Inner loop uses 167 flops */
456 }
457
458 if(jidx<j_index_end)
459 {
460
461 /* Get j neighbor index, and coordinate index */
462 jnrlistA = jjnr[jidx];
463 jnrlistB = jjnr[jidx+1];
464 jnrlistC = jjnr[jidx+2];
465 jnrlistD = jjnr[jidx+3];
466 /* Sign of each element will be negative for non-real atoms.
467 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
468 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
469 */
470 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
471 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
472 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
473 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
474 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
475 j_coord_offsetA = DIM*jnrA;
476 j_coord_offsetB = DIM*jnrB;
477 j_coord_offsetC = DIM*jnrC;
478 j_coord_offsetD = DIM*jnrD;
479
480 /* load j atom coordinates */
481 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
482 x+j_coord_offsetC,x+j_coord_offsetD,
483 &jx0,&jy0,&jz0);
484
485 /* Calculate displacement vector */
486 dx00 = _mm_sub_ps(ix0,jx0);
487 dy00 = _mm_sub_ps(iy0,jy0);
488 dz00 = _mm_sub_ps(iz0,jz0);
489 dx10 = _mm_sub_ps(ix1,jx0);
490 dy10 = _mm_sub_ps(iy1,jy0);
491 dz10 = _mm_sub_ps(iz1,jz0);
492 dx20 = _mm_sub_ps(ix2,jx0);
493 dy20 = _mm_sub_ps(iy2,jy0);
494 dz20 = _mm_sub_ps(iz2,jz0);
495 dx30 = _mm_sub_ps(ix3,jx0);
496 dy30 = _mm_sub_ps(iy3,jy0);
497 dz30 = _mm_sub_ps(iz3,jz0);
498
499 /* Calculate squared distance and things based on it */
500 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
501 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
502 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
503 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
504
505 rinv00 = gmx_mm_invsqrt_ps(rsq00);
506 rinv10 = gmx_mm_invsqrt_ps(rsq10);
507 rinv20 = gmx_mm_invsqrt_ps(rsq20);
508 rinv30 = gmx_mm_invsqrt_ps(rsq30);
509
510 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
511 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
512 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
513 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
514
515 /* Load parameters for j particles */
516 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
517 charge+jnrC+0,charge+jnrD+0);
518 vdwjidx0A = 2*vdwtype[jnrA+0];
519 vdwjidx0B = 2*vdwtype[jnrB+0];
520 vdwjidx0C = 2*vdwtype[jnrC+0];
521 vdwjidx0D = 2*vdwtype[jnrD+0];
522
523 fjx0 = _mm_setzero_ps();
524 fjy0 = _mm_setzero_ps();
525 fjz0 = _mm_setzero_ps();
526
527 /**************************
528 * CALCULATE INTERACTIONS *
529 **************************/
530
531 if (gmx_mm_any_lt(rsq00,rcutoff2))
532 {
533
534 r00 = _mm_mul_ps(rsq00,rinv00);
535 r00 = _mm_andnot_ps(dummy_mask,r00);
536
537 /* Compute parameters for interactions between i and j atoms */
538 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
539 vdwparam+vdwioffset0+vdwjidx0B,
540 vdwparam+vdwioffset0+vdwjidx0C,
541 vdwparam+vdwioffset0+vdwjidx0D,
542 &c6_00,&c12_00);
543
544 /* LENNARD-JONES DISPERSION/REPULSION */
545
546 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
547 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
548 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
549 vvdw = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
550 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
551
552 d = _mm_sub_ps(r00,rswitch);
553 d = _mm_max_ps(d,_mm_setzero_ps());
554 d2 = _mm_mul_ps(d,d);
555 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
556
557 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
558
559 /* Evaluate switch function */
560 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
561 fvdw = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
562 vvdw = _mm_mul_ps(vvdw,sw);
563 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
564
565 /* Update potential sum for this i atom from the interaction with this j atom. */
566 vvdw = _mm_and_ps(vvdw,cutoff_mask);
567 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
568 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
569
570 fscal = fvdw;
571
572 fscal = _mm_and_ps(fscal,cutoff_mask);
573
574 fscal = _mm_andnot_ps(dummy_mask,fscal);
575
576 /* Calculate temporary vectorial force */
577 tx = _mm_mul_ps(fscal,dx00);
578 ty = _mm_mul_ps(fscal,dy00);
579 tz = _mm_mul_ps(fscal,dz00);
580
581 /* Update vectorial force */
582 fix0 = _mm_add_ps(fix0,tx);
583 fiy0 = _mm_add_ps(fiy0,ty);
584 fiz0 = _mm_add_ps(fiz0,tz);
585
586 fjx0 = _mm_add_ps(fjx0,tx);
587 fjy0 = _mm_add_ps(fjy0,ty);
588 fjz0 = _mm_add_ps(fjz0,tz);
589
590 }
591
592 /**************************
593 * CALCULATE INTERACTIONS *
594 **************************/
595
596 if (gmx_mm_any_lt(rsq10,rcutoff2))
597 {
598
599 /* Compute parameters for interactions between i and j atoms */
600 qq10 = _mm_mul_ps(iq1,jq0);
601
602 /* REACTION-FIELD ELECTROSTATICS */
603 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
604 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
605
606 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
607
608 /* Update potential sum for this i atom from the interaction with this j atom. */
609 velec = _mm_and_ps(velec,cutoff_mask);
610 velec = _mm_andnot_ps(dummy_mask,velec);
611 velecsum = _mm_add_ps(velecsum,velec);
612
613 fscal = felec;
614
615 fscal = _mm_and_ps(fscal,cutoff_mask);
616
617 fscal = _mm_andnot_ps(dummy_mask,fscal);
618
619 /* Calculate temporary vectorial force */
620 tx = _mm_mul_ps(fscal,dx10);
621 ty = _mm_mul_ps(fscal,dy10);
622 tz = _mm_mul_ps(fscal,dz10);
623
624 /* Update vectorial force */
625 fix1 = _mm_add_ps(fix1,tx);
626 fiy1 = _mm_add_ps(fiy1,ty);
627 fiz1 = _mm_add_ps(fiz1,tz);
628
629 fjx0 = _mm_add_ps(fjx0,tx);
630 fjy0 = _mm_add_ps(fjy0,ty);
631 fjz0 = _mm_add_ps(fjz0,tz);
632
633 }
634
635 /**************************
636 * CALCULATE INTERACTIONS *
637 **************************/
638
639 if (gmx_mm_any_lt(rsq20,rcutoff2))
640 {
641
642 /* Compute parameters for interactions between i and j atoms */
643 qq20 = _mm_mul_ps(iq2,jq0);
644
645 /* REACTION-FIELD ELECTROSTATICS */
646 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
647 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
648
649 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
650
651 /* Update potential sum for this i atom from the interaction with this j atom. */
652 velec = _mm_and_ps(velec,cutoff_mask);
653 velec = _mm_andnot_ps(dummy_mask,velec);
654 velecsum = _mm_add_ps(velecsum,velec);
655
656 fscal = felec;
657
658 fscal = _mm_and_ps(fscal,cutoff_mask);
659
660 fscal = _mm_andnot_ps(dummy_mask,fscal);
661
662 /* Calculate temporary vectorial force */
663 tx = _mm_mul_ps(fscal,dx20);
664 ty = _mm_mul_ps(fscal,dy20);
665 tz = _mm_mul_ps(fscal,dz20);
666
667 /* Update vectorial force */
668 fix2 = _mm_add_ps(fix2,tx);
669 fiy2 = _mm_add_ps(fiy2,ty);
670 fiz2 = _mm_add_ps(fiz2,tz);
671
672 fjx0 = _mm_add_ps(fjx0,tx);
673 fjy0 = _mm_add_ps(fjy0,ty);
674 fjz0 = _mm_add_ps(fjz0,tz);
675
676 }
677
678 /**************************
679 * CALCULATE INTERACTIONS *
680 **************************/
681
682 if (gmx_mm_any_lt(rsq30,rcutoff2))
683 {
684
685 /* Compute parameters for interactions between i and j atoms */
686 qq30 = _mm_mul_ps(iq3,jq0);
687
688 /* REACTION-FIELD ELECTROSTATICS */
689 velec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
690 felec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
691
692 cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2);
693
694 /* Update potential sum for this i atom from the interaction with this j atom. */
695 velec = _mm_and_ps(velec,cutoff_mask);
696 velec = _mm_andnot_ps(dummy_mask,velec);
697 velecsum = _mm_add_ps(velecsum,velec);
698
699 fscal = felec;
700
701 fscal = _mm_and_ps(fscal,cutoff_mask);
702
703 fscal = _mm_andnot_ps(dummy_mask,fscal);
704
705 /* Calculate temporary vectorial force */
706 tx = _mm_mul_ps(fscal,dx30);
707 ty = _mm_mul_ps(fscal,dy30);
708 tz = _mm_mul_ps(fscal,dz30);
709
710 /* Update vectorial force */
711 fix3 = _mm_add_ps(fix3,tx);
712 fiy3 = _mm_add_ps(fiy3,ty);
713 fiz3 = _mm_add_ps(fiz3,tz);
714
715 fjx0 = _mm_add_ps(fjx0,tx);
716 fjy0 = _mm_add_ps(fjy0,ty);
717 fjz0 = _mm_add_ps(fjz0,tz);
718
719 }
720
721 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
722 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
723 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
724 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
725
726 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
727
728 /* Inner loop uses 168 flops */
729 }
730
731 /* End of innermost loop */
732
733 gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
734 f+i_coord_offset,fshift+i_shift_offset);
735
736 ggid = gid[iidx];
737 /* Update potential energies */
738 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
739 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
740
741 /* Increment number of inner iterations */
742 inneriter += j_index_end - j_index_start;
743
744 /* Outer loop uses 26 flops */
745 }
746
747 /* Increment number of outer iterations */
748 outeriter += nri;
749
750 /* Update outer/inner flops */
751
752 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*168);
753 }
754 /*
755 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_F_sse2_single
756 * Electrostatics interaction: ReactionField
757 * VdW interaction: LennardJones
758 * Geometry: Water4-Particle
759 * Calculate force/pot: Force
760 */
761 void
762 nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_F_sse2_single
763 (t_nblist * gmx_restrict nlist,
764 rvec * gmx_restrict xx,
765 rvec * gmx_restrict ff,
766 t_forcerec * gmx_restrict fr,
767 t_mdatoms * gmx_restrict mdatoms,
768 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
769 t_nrnb * gmx_restrict nrnb)
770 {
771 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
772 * just 0 for non-waters.
773 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
774 * jnr indices corresponding to data put in the four positions in the SIMD register.
775 */
776 int i_shift_offset,i_coord_offset,outeriter,inneriter;
777 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
778 int jnrA,jnrB,jnrC,jnrD;
779 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
780 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
781 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
782 real rcutoff_scalar;
783 real *shiftvec,*fshift,*x,*f;
784 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
785 real scratch[4*DIM];
786 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
787 int vdwioffset0;
788 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
789 int vdwioffset1;
790 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
791 int vdwioffset2;
792 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
793 int vdwioffset3;
794 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
795 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
796 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
797 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
798 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
799 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
800 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
801 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
802 real *charge;
803 int nvdwtype;
804 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
805 int *vdwtype;
806 real *vdwparam;
807 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
808 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
809 __m128 rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
810 real rswitch_scalar,d_scalar;
811 __m128 dummy_mask,cutoff_mask;
812 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
813 __m128 one = _mm_set1_ps(1.0);
814 __m128 two = _mm_set1_ps(2.0);
815 x = xx[0];
816 f = ff[0];
817
818 nri = nlist->nri;
819 iinr = nlist->iinr;
820 jindex = nlist->jindex;
821 jjnr = nlist->jjnr;
822 shiftidx = nlist->shift;
823 gid = nlist->gid;
824 shiftvec = fr->shift_vec[0];
825 fshift = fr->fshift[0];
826 facel = _mm_set1_ps(fr->epsfac);
827 charge = mdatoms->chargeA;
828 krf = _mm_set1_ps(fr->ic->k_rf);
829 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
830 crf = _mm_set1_ps(fr->ic->c_rf);
831 nvdwtype = fr->ntype;
832 vdwparam = fr->nbfp;
833 vdwtype = mdatoms->typeA;
834
835 /* Setup water-specific parameters */
836 inr = nlist->iinr[0];
837 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
838 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
839 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
840 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
841
842 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
843 rcutoff_scalar = fr->rcoulomb;
844 rcutoff = _mm_set1_ps(rcutoff_scalar);
845 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
846
847 rswitch_scalar = fr->rvdw_switch;
848 rswitch = _mm_set1_ps(rswitch_scalar);
849 /* Setup switch parameters */
850 d_scalar = rcutoff_scalar-rswitch_scalar;
851 d = _mm_set1_ps(d_scalar);
852 swV3 = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
853 swV4 = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
854 swV5 = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
855 swF2 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
856 swF3 = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
857 swF4 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
858
859 /* Avoid stupid compiler warnings */
860 jnrA = jnrB = jnrC = jnrD = 0;
861 j_coord_offsetA = 0;
862 j_coord_offsetB = 0;
863 j_coord_offsetC = 0;
864 j_coord_offsetD = 0;
865
866 outeriter = 0;
867 inneriter = 0;
868
869 for(iidx=0;iidx<4*DIM;iidx++)
870 {
871 scratch[iidx] = 0.0;
872 }
873
874 /* Start outer loop over neighborlists */
875 for(iidx=0; iidx<nri; iidx++)
876 {
877 /* Load shift vector for this list */
878 i_shift_offset = DIM*shiftidx[iidx];
879
880 /* Load limits for loop over neighbors */
881 j_index_start = jindex[iidx];
882 j_index_end = jindex[iidx+1];
883
884 /* Get outer coordinate index */
885 inr = iinr[iidx];
886 i_coord_offset = DIM*inr;
887
888 /* Load i particle coords and add shift vector */
889 gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
890 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
891
892 fix0 = _mm_setzero_ps();
893 fiy0 = _mm_setzero_ps();
894 fiz0 = _mm_setzero_ps();
895 fix1 = _mm_setzero_ps();
896 fiy1 = _mm_setzero_ps();
897 fiz1 = _mm_setzero_ps();
898 fix2 = _mm_setzero_ps();
899 fiy2 = _mm_setzero_ps();
900 fiz2 = _mm_setzero_ps();
901 fix3 = _mm_setzero_ps();
902 fiy3 = _mm_setzero_ps();
903 fiz3 = _mm_setzero_ps();
904
905 /* Start inner kernel loop */
906 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
907 {
908
909 /* Get j neighbor index, and coordinate index */
910 jnrA = jjnr[jidx];
911 jnrB = jjnr[jidx+1];
912 jnrC = jjnr[jidx+2];
913 jnrD = jjnr[jidx+3];
914 j_coord_offsetA = DIM*jnrA;
915 j_coord_offsetB = DIM*jnrB;
916 j_coord_offsetC = DIM*jnrC;
917 j_coord_offsetD = DIM*jnrD;
918
919 /* load j atom coordinates */
920 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
921 x+j_coord_offsetC,x+j_coord_offsetD,
922 &jx0,&jy0,&jz0);
923
924 /* Calculate displacement vector */
925 dx00 = _mm_sub_ps(ix0,jx0);
926 dy00 = _mm_sub_ps(iy0,jy0);
927 dz00 = _mm_sub_ps(iz0,jz0);
928 dx10 = _mm_sub_ps(ix1,jx0);
929 dy10 = _mm_sub_ps(iy1,jy0);
930 dz10 = _mm_sub_ps(iz1,jz0);
931 dx20 = _mm_sub_ps(ix2,jx0);
932 dy20 = _mm_sub_ps(iy2,jy0);
933 dz20 = _mm_sub_ps(iz2,jz0);
934 dx30 = _mm_sub_ps(ix3,jx0);
935 dy30 = _mm_sub_ps(iy3,jy0);
936 dz30 = _mm_sub_ps(iz3,jz0);
937
938 /* Calculate squared distance and things based on it */
939 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
940 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
941 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
942 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
943
944 rinv00 = gmx_mm_invsqrt_ps(rsq00);
945 rinv10 = gmx_mm_invsqrt_ps(rsq10);
946 rinv20 = gmx_mm_invsqrt_ps(rsq20);
947 rinv30 = gmx_mm_invsqrt_ps(rsq30);
948
949 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
950 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
951 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
952 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
953
954 /* Load parameters for j particles */
955 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
956 charge+jnrC+0,charge+jnrD+0);
957 vdwjidx0A = 2*vdwtype[jnrA+0];
958 vdwjidx0B = 2*vdwtype[jnrB+0];
959 vdwjidx0C = 2*vdwtype[jnrC+0];
960 vdwjidx0D = 2*vdwtype[jnrD+0];
961
962 fjx0 = _mm_setzero_ps();
963 fjy0 = _mm_setzero_ps();
964 fjz0 = _mm_setzero_ps();
965
966 /**************************
967 * CALCULATE INTERACTIONS *
968 **************************/
969
970 if (gmx_mm_any_lt(rsq00,rcutoff2))
971 {
972
973 r00 = _mm_mul_ps(rsq00,rinv00);
974
975 /* Compute parameters for interactions between i and j atoms */
976 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
977 vdwparam+vdwioffset0+vdwjidx0B,
978 vdwparam+vdwioffset0+vdwjidx0C,
979 vdwparam+vdwioffset0+vdwjidx0D,
980 &c6_00,&c12_00);
981
982 /* LENNARD-JONES DISPERSION/REPULSION */
983
984 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
985 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
986 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
987 vvdw = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
988 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
989
990 d = _mm_sub_ps(r00,rswitch);
991 d = _mm_max_ps(d,_mm_setzero_ps());
992 d2 = _mm_mul_ps(d,d);
993 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
994
995 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
996
997 /* Evaluate switch function */
998 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
999 fvdw = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1000 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
1001
1002 fscal = fvdw;
1003
1004 fscal = _mm_and_ps(fscal,cutoff_mask);
1005
1006 /* Calculate temporary vectorial force */
1007 tx = _mm_mul_ps(fscal,dx00);
1008 ty = _mm_mul_ps(fscal,dy00);
1009 tz = _mm_mul_ps(fscal,dz00);
1010
1011 /* Update vectorial force */
1012 fix0 = _mm_add_ps(fix0,tx);
1013 fiy0 = _mm_add_ps(fiy0,ty);
1014 fiz0 = _mm_add_ps(fiz0,tz);
1015
1016 fjx0 = _mm_add_ps(fjx0,tx);
1017 fjy0 = _mm_add_ps(fjy0,ty);
1018 fjz0 = _mm_add_ps(fjz0,tz);
1019
1020 }
1021
1022 /**************************
1023 * CALCULATE INTERACTIONS *
1024 **************************/
1025
1026 if (gmx_mm_any_lt(rsq10,rcutoff2))
1027 {
1028
1029 /* Compute parameters for interactions between i and j atoms */
1030 qq10 = _mm_mul_ps(iq1,jq0);
1031
1032 /* REACTION-FIELD ELECTROSTATICS */
1033 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1034
1035 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
1036
1037 fscal = felec;
1038
1039 fscal = _mm_and_ps(fscal,cutoff_mask);
1040
1041 /* Calculate temporary vectorial force */
1042 tx = _mm_mul_ps(fscal,dx10);
1043 ty = _mm_mul_ps(fscal,dy10);
1044 tz = _mm_mul_ps(fscal,dz10);
1045
1046 /* Update vectorial force */
1047 fix1 = _mm_add_ps(fix1,tx);
1048 fiy1 = _mm_add_ps(fiy1,ty);
1049 fiz1 = _mm_add_ps(fiz1,tz);
1050
1051 fjx0 = _mm_add_ps(fjx0,tx);
1052 fjy0 = _mm_add_ps(fjy0,ty);
1053 fjz0 = _mm_add_ps(fjz0,tz);
1054
1055 }
1056
1057 /**************************
1058 * CALCULATE INTERACTIONS *
1059 **************************/
1060
1061 if (gmx_mm_any_lt(rsq20,rcutoff2))
1062 {
1063
1064 /* Compute parameters for interactions between i and j atoms */
1065 qq20 = _mm_mul_ps(iq2,jq0);
1066
1067 /* REACTION-FIELD ELECTROSTATICS */
1068 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1069
1070 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
1071
1072 fscal = felec;
1073
1074 fscal = _mm_and_ps(fscal,cutoff_mask);
1075
1076 /* Calculate temporary vectorial force */
1077 tx = _mm_mul_ps(fscal,dx20);
1078 ty = _mm_mul_ps(fscal,dy20);
1079 tz = _mm_mul_ps(fscal,dz20);
1080
1081 /* Update vectorial force */
1082 fix2 = _mm_add_ps(fix2,tx);
1083 fiy2 = _mm_add_ps(fiy2,ty);
1084 fiz2 = _mm_add_ps(fiz2,tz);
1085
1086 fjx0 = _mm_add_ps(fjx0,tx);
1087 fjy0 = _mm_add_ps(fjy0,ty);
1088 fjz0 = _mm_add_ps(fjz0,tz);
1089
1090 }
1091
1092 /**************************
1093 * CALCULATE INTERACTIONS *
1094 **************************/
1095
1096 if (gmx_mm_any_lt(rsq30,rcutoff2))
1097 {
1098
1099 /* Compute parameters for interactions between i and j atoms */
1100 qq30 = _mm_mul_ps(iq3,jq0);
1101
1102 /* REACTION-FIELD ELECTROSTATICS */
1103 felec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1104
1105 cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2);
1106
1107 fscal = felec;
1108
1109 fscal = _mm_and_ps(fscal,cutoff_mask);
1110
1111 /* Calculate temporary vectorial force */
1112 tx = _mm_mul_ps(fscal,dx30);
1113 ty = _mm_mul_ps(fscal,dy30);
1114 tz = _mm_mul_ps(fscal,dz30);
1115
1116 /* Update vectorial force */
1117 fix3 = _mm_add_ps(fix3,tx);
1118 fiy3 = _mm_add_ps(fiy3,ty);
1119 fiz3 = _mm_add_ps(fiz3,tz);
1120
1121 fjx0 = _mm_add_ps(fjx0,tx);
1122 fjy0 = _mm_add_ps(fjy0,ty);
1123 fjz0 = _mm_add_ps(fjz0,tz);
1124
1125 }
1126
1127 fjptrA = f+j_coord_offsetA;
1128 fjptrB = f+j_coord_offsetB;
1129 fjptrC = f+j_coord_offsetC;
1130 fjptrD = f+j_coord_offsetD;
1131
1132 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1133
1134 /* Inner loop uses 146 flops */
1135 }
1136
1137 if(jidx<j_index_end)
1138 {
1139
1140 /* Get j neighbor index, and coordinate index */
1141 jnrlistA = jjnr[jidx];
1142 jnrlistB = jjnr[jidx+1];
1143 jnrlistC = jjnr[jidx+2];
1144 jnrlistD = jjnr[jidx+3];
1145 /* Sign of each element will be negative for non-real atoms.
1146 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1147 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1148 */
1149 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1150 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1151 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1152 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1153 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1154 j_coord_offsetA = DIM*jnrA;
1155 j_coord_offsetB = DIM*jnrB;
1156 j_coord_offsetC = DIM*jnrC;
1157 j_coord_offsetD = DIM*jnrD;
1158
1159 /* load j atom coordinates */
1160 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1161 x+j_coord_offsetC,x+j_coord_offsetD,
1162 &jx0,&jy0,&jz0);
1163
1164 /* Calculate displacement vector */
1165 dx00 = _mm_sub_ps(ix0,jx0);
1166 dy00 = _mm_sub_ps(iy0,jy0);
1167 dz00 = _mm_sub_ps(iz0,jz0);
1168 dx10 = _mm_sub_ps(ix1,jx0);
1169 dy10 = _mm_sub_ps(iy1,jy0);
1170 dz10 = _mm_sub_ps(iz1,jz0);
1171 dx20 = _mm_sub_ps(ix2,jx0);
1172 dy20 = _mm_sub_ps(iy2,jy0);
1173 dz20 = _mm_sub_ps(iz2,jz0);
1174 dx30 = _mm_sub_ps(ix3,jx0);
1175 dy30 = _mm_sub_ps(iy3,jy0);
1176 dz30 = _mm_sub_ps(iz3,jz0);
1177
1178 /* Calculate squared distance and things based on it */
1179 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1180 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1181 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1182 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1183
1184 rinv00 = gmx_mm_invsqrt_ps(rsq00);
1185 rinv10 = gmx_mm_invsqrt_ps(rsq10);
1186 rinv20 = gmx_mm_invsqrt_ps(rsq20);
1187 rinv30 = gmx_mm_invsqrt_ps(rsq30);
1188
1189 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
1190 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
1191 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
1192 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
1193
1194 /* Load parameters for j particles */
1195 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1196 charge+jnrC+0,charge+jnrD+0);
1197 vdwjidx0A = 2*vdwtype[jnrA+0];
1198 vdwjidx0B = 2*vdwtype[jnrB+0];
1199 vdwjidx0C = 2*vdwtype[jnrC+0];
1200 vdwjidx0D = 2*vdwtype[jnrD+0];
1201
1202 fjx0 = _mm_setzero_ps();
1203 fjy0 = _mm_setzero_ps();
1204 fjz0 = _mm_setzero_ps();
1205
1206 /**************************
1207 * CALCULATE INTERACTIONS *
1208 **************************/
1209
1210 if (gmx_mm_any_lt(rsq00,rcutoff2))
1211 {
1212
1213 r00 = _mm_mul_ps(rsq00,rinv00);
1214 r00 = _mm_andnot_ps(dummy_mask,r00);
1215
1216 /* Compute parameters for interactions between i and j atoms */
1217 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1218 vdwparam+vdwioffset0+vdwjidx0B,
1219 vdwparam+vdwioffset0+vdwjidx0C,
1220 vdwparam+vdwioffset0+vdwjidx0D,
1221 &c6_00,&c12_00);
1222
1223 /* LENNARD-JONES DISPERSION/REPULSION */
1224
1225 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1226 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
1227 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
1228 vvdw = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
1229 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
1230
1231 d = _mm_sub_ps(r00,rswitch);
1232 d = _mm_max_ps(d,_mm_setzero_ps());
1233 d2 = _mm_mul_ps(d,d);
1234 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1235
1236 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1237
1238 /* Evaluate switch function */
1239 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1240 fvdw = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1241 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
1242
1243 fscal = fvdw;
1244
1245 fscal = _mm_and_ps(fscal,cutoff_mask);
1246
1247 fscal = _mm_andnot_ps(dummy_mask,fscal);
1248
1249 /* Calculate temporary vectorial force */
1250 tx = _mm_mul_ps(fscal,dx00);
1251 ty = _mm_mul_ps(fscal,dy00);
1252 tz = _mm_mul_ps(fscal,dz00);
1253
1254 /* Update vectorial force */
1255 fix0 = _mm_add_ps(fix0,tx);
1256 fiy0 = _mm_add_ps(fiy0,ty);
1257 fiz0 = _mm_add_ps(fiz0,tz);
1258
1259 fjx0 = _mm_add_ps(fjx0,tx);
1260 fjy0 = _mm_add_ps(fjy0,ty);
1261 fjz0 = _mm_add_ps(fjz0,tz);
1262
1263 }
1264
1265 /**************************
1266 * CALCULATE INTERACTIONS *
1267 **************************/
1268
1269 if (gmx_mm_any_lt(rsq10,rcutoff2))
1270 {
1271
1272 /* Compute parameters for interactions between i and j atoms */
1273 qq10 = _mm_mul_ps(iq1,jq0);
1274
1275 /* REACTION-FIELD ELECTROSTATICS */
1276 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1277
1278 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
1279
1280 fscal = felec;
1281
1282 fscal = _mm_and_ps(fscal,cutoff_mask);
1283
1284 fscal = _mm_andnot_ps(dummy_mask,fscal);
1285
1286 /* Calculate temporary vectorial force */
1287 tx = _mm_mul_ps(fscal,dx10);
1288 ty = _mm_mul_ps(fscal,dy10);
1289 tz = _mm_mul_ps(fscal,dz10);
1290
1291 /* Update vectorial force */
1292 fix1 = _mm_add_ps(fix1,tx);
1293 fiy1 = _mm_add_ps(fiy1,ty);
1294 fiz1 = _mm_add_ps(fiz1,tz);
1295
1296 fjx0 = _mm_add_ps(fjx0,tx);
1297 fjy0 = _mm_add_ps(fjy0,ty);
1298 fjz0 = _mm_add_ps(fjz0,tz);
1299
1300 }
1301
1302 /**************************
1303 * CALCULATE INTERACTIONS *
1304 **************************/
1305
1306 if (gmx_mm_any_lt(rsq20,rcutoff2))
1307 {
1308
1309 /* Compute parameters for interactions between i and j atoms */
1310 qq20 = _mm_mul_ps(iq2,jq0);
1311
1312 /* REACTION-FIELD ELECTROSTATICS */
1313 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1314
1315 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
1316
1317 fscal = felec;
1318
1319 fscal = _mm_and_ps(fscal,cutoff_mask);
1320
1321 fscal = _mm_andnot_ps(dummy_mask,fscal);
1322
1323 /* Calculate temporary vectorial force */
1324 tx = _mm_mul_ps(fscal,dx20);
1325 ty = _mm_mul_ps(fscal,dy20);
1326 tz = _mm_mul_ps(fscal,dz20);
1327
1328 /* Update vectorial force */
1329 fix2 = _mm_add_ps(fix2,tx);
1330 fiy2 = _mm_add_ps(fiy2,ty);
1331 fiz2 = _mm_add_ps(fiz2,tz);
1332
1333 fjx0 = _mm_add_ps(fjx0,tx);
1334 fjy0 = _mm_add_ps(fjy0,ty);
1335 fjz0 = _mm_add_ps(fjz0,tz);
1336
1337 }
1338
1339 /**************************
1340 * CALCULATE INTERACTIONS *
1341 **************************/
1342
1343 if (gmx_mm_any_lt(rsq30,rcutoff2))
1344 {
1345
1346 /* Compute parameters for interactions between i and j atoms */
1347 qq30 = _mm_mul_ps(iq3,jq0);
1348
1349 /* REACTION-FIELD ELECTROSTATICS */
1350 felec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1351
1352 cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2);
1353
1354 fscal = felec;
1355
1356 fscal = _mm_and_ps(fscal,cutoff_mask);
1357
1358 fscal = _mm_andnot_ps(dummy_mask,fscal);
1359
1360 /* Calculate temporary vectorial force */
1361 tx = _mm_mul_ps(fscal,dx30);
1362 ty = _mm_mul_ps(fscal,dy30);
1363 tz = _mm_mul_ps(fscal,dz30);
1364
1365 /* Update vectorial force */
1366 fix3 = _mm_add_ps(fix3,tx);
1367 fiy3 = _mm_add_ps(fiy3,ty);
1368 fiz3 = _mm_add_ps(fiz3,tz);
1369
1370 fjx0 = _mm_add_ps(fjx0,tx);
1371 fjy0 = _mm_add_ps(fjy0,ty);
1372 fjz0 = _mm_add_ps(fjz0,tz);
1373
1374 }
1375
1376 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1377 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1378 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1379 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1380
1381 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1382
1383 /* Inner loop uses 147 flops */
1384 }
1385
1386 /* End of innermost loop */
1387
1388 gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1389 f+i_coord_offset,fshift+i_shift_offset);
1390
1391 /* Increment number of inner iterations */
1392 inneriter += j_index_end - j_index_start;
1393
1394 /* Outer loop uses 24 flops */
1395 }
1396
1397 /* Increment number of outer iterations */
1398 outeriter += nri;
1399
1400 /* Update outer/inner flops */
1401
1402 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*147);
1403 }
The ultimate molecular dynamics simulation package