?  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_ElecEwSw_VdwLJSw_GeomW4P1_sse2_double.c
blob5d48d3a58a4591600bc8d3ca494a5c6f5364a1be
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_ElecEwSw_VdwLJSw_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_ElecEwSw_VdwLJSw_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 rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
107 real rswitch_scalar,d_scalar;
108 __m128d dummy_mask,cutoff_mask;
109 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
110 __m128d one = _mm_set1_pd(1.0);
111 __m128d two = _mm_set1_pd(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_pd(fr->epsfac);
124 charge = mdatoms->chargeA;
125 nvdwtype = fr->ntype;
126 vdwparam = fr->nbfp;
127 vdwtype = mdatoms->typeA;
128
129 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
130 ewtab = fr->ic->tabq_coul_FDV0;
131 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
132 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
133
134 /* Setup water-specific parameters */
135 inr = nlist->iinr[0];
136 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
137 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
138 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
139 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
140
141 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
142 rcutoff_scalar = fr->rcoulomb;
143 rcutoff = _mm_set1_pd(rcutoff_scalar);
144 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
145
146 rswitch_scalar = fr->rcoulomb_switch;
147 rswitch = _mm_set1_pd(rswitch_scalar);
148 /* Setup switch parameters */
149 d_scalar = rcutoff_scalar-rswitch_scalar;
150 d = _mm_set1_pd(d_scalar);
151 swV3 = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
152 swV4 = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
153 swV5 = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
154 swF2 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
155 swF3 = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
156 swF4 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
157
158 /* Avoid stupid compiler warnings */
159 jnrA = jnrB = 0;
160 j_coord_offsetA = 0;
161 j_coord_offsetB = 0;
162
163 outeriter = 0;
164 inneriter = 0;
165
166 /* Start outer loop over neighborlists */
167 for(iidx=0; iidx<nri; iidx++)
168 {
169 /* Load shift vector for this list */
170 i_shift_offset = DIM*shiftidx[iidx];
171
172 /* Load limits for loop over neighbors */
173 j_index_start = jindex[iidx];
174 j_index_end = jindex[iidx+1];
175
176 /* Get outer coordinate index */
177 inr = iinr[iidx];
178 i_coord_offset = DIM*inr;
179
180 /* Load i particle coords and add shift vector */
181 gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
182 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
183
184 fix0 = _mm_setzero_pd();
185 fiy0 = _mm_setzero_pd();
186 fiz0 = _mm_setzero_pd();
187 fix1 = _mm_setzero_pd();
188 fiy1 = _mm_setzero_pd();
189 fiz1 = _mm_setzero_pd();
190 fix2 = _mm_setzero_pd();
191 fiy2 = _mm_setzero_pd();
192 fiz2 = _mm_setzero_pd();
193 fix3 = _mm_setzero_pd();
194 fiy3 = _mm_setzero_pd();
195 fiz3 = _mm_setzero_pd();
196
197 /* Reset potential sums */
198 velecsum = _mm_setzero_pd();
199 vvdwsum = _mm_setzero_pd();
200
201 /* Start inner kernel loop */
202 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
203 {
204
205 /* Get j neighbor index, and coordinate index */
206 jnrA = jjnr[jidx];
207 jnrB = jjnr[jidx+1];
208 j_coord_offsetA = DIM*jnrA;
209 j_coord_offsetB = DIM*jnrB;
210
211 /* load j atom coordinates */
212 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
213 &jx0,&jy0,&jz0);
214
215 /* Calculate displacement vector */
216 dx00 = _mm_sub_pd(ix0,jx0);
217 dy00 = _mm_sub_pd(iy0,jy0);
218 dz00 = _mm_sub_pd(iz0,jz0);
219 dx10 = _mm_sub_pd(ix1,jx0);
220 dy10 = _mm_sub_pd(iy1,jy0);
221 dz10 = _mm_sub_pd(iz1,jz0);
222 dx20 = _mm_sub_pd(ix2,jx0);
223 dy20 = _mm_sub_pd(iy2,jy0);
224 dz20 = _mm_sub_pd(iz2,jz0);
225 dx30 = _mm_sub_pd(ix3,jx0);
226 dy30 = _mm_sub_pd(iy3,jy0);
227 dz30 = _mm_sub_pd(iz3,jz0);
228
229 /* Calculate squared distance and things based on it */
230 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
231 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
232 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
233 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
234
235 rinv00 = gmx_mm_invsqrt_pd(rsq00);
236 rinv10 = gmx_mm_invsqrt_pd(rsq10);
237 rinv20 = gmx_mm_invsqrt_pd(rsq20);
238 rinv30 = gmx_mm_invsqrt_pd(rsq30);
239
240 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
241 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
242 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
243 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
244
245 /* Load parameters for j particles */
246 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
247 vdwjidx0A = 2*vdwtype[jnrA+0];
248 vdwjidx0B = 2*vdwtype[jnrB+0];
249
250 fjx0 = _mm_setzero_pd();
251 fjy0 = _mm_setzero_pd();
252 fjz0 = _mm_setzero_pd();
253
254 /**************************
255 * CALCULATE INTERACTIONS *
256 **************************/
257
258 if (gmx_mm_any_lt(rsq00,rcutoff2))
259 {
260
261 r00 = _mm_mul_pd(rsq00,rinv00);
262
263 /* Compute parameters for interactions between i and j atoms */
264 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
265 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
266
267 /* LENNARD-JONES DISPERSION/REPULSION */
268
269 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
270 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
271 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
272 vvdw = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
273 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
274
275 d = _mm_sub_pd(r00,rswitch);
276 d = _mm_max_pd(d,_mm_setzero_pd());
277 d2 = _mm_mul_pd(d,d);
278 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
279
280 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
281
282 /* Evaluate switch function */
283 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
284 fvdw = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
285 vvdw = _mm_mul_pd(vvdw,sw);
286 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
287
288 /* Update potential sum for this i atom from the interaction with this j atom. */
289 vvdw = _mm_and_pd(vvdw,cutoff_mask);
290 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
291
292 fscal = fvdw;
293
294 fscal = _mm_and_pd(fscal,cutoff_mask);
295
296 /* Calculate temporary vectorial force */
297 tx = _mm_mul_pd(fscal,dx00);
298 ty = _mm_mul_pd(fscal,dy00);
299 tz = _mm_mul_pd(fscal,dz00);
300
301 /* Update vectorial force */
302 fix0 = _mm_add_pd(fix0,tx);
303 fiy0 = _mm_add_pd(fiy0,ty);
304 fiz0 = _mm_add_pd(fiz0,tz);
305
306 fjx0 = _mm_add_pd(fjx0,tx);
307 fjy0 = _mm_add_pd(fjy0,ty);
308 fjz0 = _mm_add_pd(fjz0,tz);
309
310 }
311
312 /**************************
313 * CALCULATE INTERACTIONS *
314 **************************/
315
316 if (gmx_mm_any_lt(rsq10,rcutoff2))
317 {
318
319 r10 = _mm_mul_pd(rsq10,rinv10);
320
321 /* Compute parameters for interactions between i and j atoms */
322 qq10 = _mm_mul_pd(iq1,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(r10,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(qq10,_mm_sub_pd(rinv10,velec));
340 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
341
342 d = _mm_sub_pd(r10,rswitch);
343 d = _mm_max_pd(d,_mm_setzero_pd());
344 d2 = _mm_mul_pd(d,d);
345 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
346
347 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
348
349 /* Evaluate switch function */
350 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
351 felec = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
352 velec = _mm_mul_pd(velec,sw);
353 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
354
355 /* Update potential sum for this i atom from the interaction with this j atom. */
356 velec = _mm_and_pd(velec,cutoff_mask);
357 velecsum = _mm_add_pd(velecsum,velec);
358
359 fscal = felec;
360
361 fscal = _mm_and_pd(fscal,cutoff_mask);
362
363 /* Calculate temporary vectorial force */
364 tx = _mm_mul_pd(fscal,dx10);
365 ty = _mm_mul_pd(fscal,dy10);
366 tz = _mm_mul_pd(fscal,dz10);
367
368 /* Update vectorial force */
369 fix1 = _mm_add_pd(fix1,tx);
370 fiy1 = _mm_add_pd(fiy1,ty);
371 fiz1 = _mm_add_pd(fiz1,tz);
372
373 fjx0 = _mm_add_pd(fjx0,tx);
374 fjy0 = _mm_add_pd(fjy0,ty);
375 fjz0 = _mm_add_pd(fjz0,tz);
376
377 }
378
379 /**************************
380 * CALCULATE INTERACTIONS *
381 **************************/
382
383 if (gmx_mm_any_lt(rsq20,rcutoff2))
384 {
385
386 r20 = _mm_mul_pd(rsq20,rinv20);
387
388 /* Compute parameters for interactions between i and j atoms */
389 qq20 = _mm_mul_pd(iq2,jq0);
390
391 /* EWALD ELECTROSTATICS */
392
393 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
394 ewrt = _mm_mul_pd(r20,ewtabscale);
395 ewitab = _mm_cvttpd_epi32(ewrt);
396 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
397 ewitab = _mm_slli_epi32(ewitab,2);
398 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
399 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
400 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
401 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
402 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
403 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
404 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
405 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
406 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
407 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
408
409 d = _mm_sub_pd(r20,rswitch);
410 d = _mm_max_pd(d,_mm_setzero_pd());
411 d2 = _mm_mul_pd(d,d);
412 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
413
414 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
415
416 /* Evaluate switch function */
417 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
418 felec = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
419 velec = _mm_mul_pd(velec,sw);
420 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
421
422 /* Update potential sum for this i atom from the interaction with this j atom. */
423 velec = _mm_and_pd(velec,cutoff_mask);
424 velecsum = _mm_add_pd(velecsum,velec);
425
426 fscal = felec;
427
428 fscal = _mm_and_pd(fscal,cutoff_mask);
429
430 /* Calculate temporary vectorial force */
431 tx = _mm_mul_pd(fscal,dx20);
432 ty = _mm_mul_pd(fscal,dy20);
433 tz = _mm_mul_pd(fscal,dz20);
434
435 /* Update vectorial force */
436 fix2 = _mm_add_pd(fix2,tx);
437 fiy2 = _mm_add_pd(fiy2,ty);
438 fiz2 = _mm_add_pd(fiz2,tz);
439
440 fjx0 = _mm_add_pd(fjx0,tx);
441 fjy0 = _mm_add_pd(fjy0,ty);
442 fjz0 = _mm_add_pd(fjz0,tz);
443
444 }
445
446 /**************************
447 * CALCULATE INTERACTIONS *
448 **************************/
449
450 if (gmx_mm_any_lt(rsq30,rcutoff2))
451 {
452
453 r30 = _mm_mul_pd(rsq30,rinv30);
454
455 /* Compute parameters for interactions between i and j atoms */
456 qq30 = _mm_mul_pd(iq3,jq0);
457
458 /* EWALD ELECTROSTATICS */
459
460 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
461 ewrt = _mm_mul_pd(r30,ewtabscale);
462 ewitab = _mm_cvttpd_epi32(ewrt);
463 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
464 ewitab = _mm_slli_epi32(ewitab,2);
465 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
466 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
467 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
468 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
469 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
470 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
471 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
472 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
473 velec = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
474 felec = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
475
476 d = _mm_sub_pd(r30,rswitch);
477 d = _mm_max_pd(d,_mm_setzero_pd());
478 d2 = _mm_mul_pd(d,d);
479 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
480
481 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
482
483 /* Evaluate switch function */
484 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
485 felec = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv30,_mm_mul_pd(velec,dsw)) );
486 velec = _mm_mul_pd(velec,sw);
487 cutoff_mask = _mm_cmplt_pd(rsq30,rcutoff2);
488
489 /* Update potential sum for this i atom from the interaction with this j atom. */
490 velec = _mm_and_pd(velec,cutoff_mask);
491 velecsum = _mm_add_pd(velecsum,velec);
492
493 fscal = felec;
494
495 fscal = _mm_and_pd(fscal,cutoff_mask);
496
497 /* Calculate temporary vectorial force */
498 tx = _mm_mul_pd(fscal,dx30);
499 ty = _mm_mul_pd(fscal,dy30);
500 tz = _mm_mul_pd(fscal,dz30);
501
502 /* Update vectorial force */
503 fix3 = _mm_add_pd(fix3,tx);
504 fiy3 = _mm_add_pd(fiy3,ty);
505 fiz3 = _mm_add_pd(fiz3,tz);
506
507 fjx0 = _mm_add_pd(fjx0,tx);
508 fjy0 = _mm_add_pd(fjy0,ty);
509 fjz0 = _mm_add_pd(fjz0,tz);
510
511 }
512
513 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
514
515 /* Inner loop uses 257 flops */
516 }
517
518 if(jidx<j_index_end)
519 {
520
521 jnrA = jjnr[jidx];
522 j_coord_offsetA = DIM*jnrA;
523
524 /* load j atom coordinates */
525 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
526 &jx0,&jy0,&jz0);
527
528 /* Calculate displacement vector */
529 dx00 = _mm_sub_pd(ix0,jx0);
530 dy00 = _mm_sub_pd(iy0,jy0);
531 dz00 = _mm_sub_pd(iz0,jz0);
532 dx10 = _mm_sub_pd(ix1,jx0);
533 dy10 = _mm_sub_pd(iy1,jy0);
534 dz10 = _mm_sub_pd(iz1,jz0);
535 dx20 = _mm_sub_pd(ix2,jx0);
536 dy20 = _mm_sub_pd(iy2,jy0);
537 dz20 = _mm_sub_pd(iz2,jz0);
538 dx30 = _mm_sub_pd(ix3,jx0);
539 dy30 = _mm_sub_pd(iy3,jy0);
540 dz30 = _mm_sub_pd(iz3,jz0);
541
542 /* Calculate squared distance and things based on it */
543 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
544 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
545 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
546 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
547
548 rinv00 = gmx_mm_invsqrt_pd(rsq00);
549 rinv10 = gmx_mm_invsqrt_pd(rsq10);
550 rinv20 = gmx_mm_invsqrt_pd(rsq20);
551 rinv30 = gmx_mm_invsqrt_pd(rsq30);
552
553 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
554 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
555 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
556 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
557
558 /* Load parameters for j particles */
559 jq0 = _mm_load_sd(charge+jnrA+0);
560 vdwjidx0A = 2*vdwtype[jnrA+0];
561
562 fjx0 = _mm_setzero_pd();
563 fjy0 = _mm_setzero_pd();
564 fjz0 = _mm_setzero_pd();
565
566 /**************************
567 * CALCULATE INTERACTIONS *
568 **************************/
569
570 if (gmx_mm_any_lt(rsq00,rcutoff2))
571 {
572
573 r00 = _mm_mul_pd(rsq00,rinv00);
574
575 /* Compute parameters for interactions between i and j atoms */
576 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
577
578 /* LENNARD-JONES DISPERSION/REPULSION */
579
580 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
581 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
582 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
583 vvdw = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
584 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
585
586 d = _mm_sub_pd(r00,rswitch);
587 d = _mm_max_pd(d,_mm_setzero_pd());
588 d2 = _mm_mul_pd(d,d);
589 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
590
591 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
592
593 /* Evaluate switch function */
594 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
595 fvdw = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
596 vvdw = _mm_mul_pd(vvdw,sw);
597 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
598
599 /* Update potential sum for this i atom from the interaction with this j atom. */
600 vvdw = _mm_and_pd(vvdw,cutoff_mask);
601 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
602 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
603
604 fscal = fvdw;
605
606 fscal = _mm_and_pd(fscal,cutoff_mask);
607
608 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
609
610 /* Calculate temporary vectorial force */
611 tx = _mm_mul_pd(fscal,dx00);
612 ty = _mm_mul_pd(fscal,dy00);
613 tz = _mm_mul_pd(fscal,dz00);
614
615 /* Update vectorial force */
616 fix0 = _mm_add_pd(fix0,tx);
617 fiy0 = _mm_add_pd(fiy0,ty);
618 fiz0 = _mm_add_pd(fiz0,tz);
619
620 fjx0 = _mm_add_pd(fjx0,tx);
621 fjy0 = _mm_add_pd(fjy0,ty);
622 fjz0 = _mm_add_pd(fjz0,tz);
623
624 }
625
626 /**************************
627 * CALCULATE INTERACTIONS *
628 **************************/
629
630 if (gmx_mm_any_lt(rsq10,rcutoff2))
631 {
632
633 r10 = _mm_mul_pd(rsq10,rinv10);
634
635 /* Compute parameters for interactions between i and j atoms */
636 qq10 = _mm_mul_pd(iq1,jq0);
637
638 /* EWALD ELECTROSTATICS */
639
640 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
641 ewrt = _mm_mul_pd(r10,ewtabscale);
642 ewitab = _mm_cvttpd_epi32(ewrt);
643 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
644 ewitab = _mm_slli_epi32(ewitab,2);
645 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
646 ewtabD = _mm_setzero_pd();
647 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
648 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
649 ewtabFn = _mm_setzero_pd();
650 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
651 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
652 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
653 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
654 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
655
656 d = _mm_sub_pd(r10,rswitch);
657 d = _mm_max_pd(d,_mm_setzero_pd());
658 d2 = _mm_mul_pd(d,d);
659 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
660
661 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
662
663 /* Evaluate switch function */
664 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
665 felec = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
666 velec = _mm_mul_pd(velec,sw);
667 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
668
669 /* Update potential sum for this i atom from the interaction with this j atom. */
670 velec = _mm_and_pd(velec,cutoff_mask);
671 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
672 velecsum = _mm_add_pd(velecsum,velec);
673
674 fscal = felec;
675
676 fscal = _mm_and_pd(fscal,cutoff_mask);
677
678 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
679
680 /* Calculate temporary vectorial force */
681 tx = _mm_mul_pd(fscal,dx10);
682 ty = _mm_mul_pd(fscal,dy10);
683 tz = _mm_mul_pd(fscal,dz10);
684
685 /* Update vectorial force */
686 fix1 = _mm_add_pd(fix1,tx);
687 fiy1 = _mm_add_pd(fiy1,ty);
688 fiz1 = _mm_add_pd(fiz1,tz);
689
690 fjx0 = _mm_add_pd(fjx0,tx);
691 fjy0 = _mm_add_pd(fjy0,ty);
692 fjz0 = _mm_add_pd(fjz0,tz);
693
694 }
695
696 /**************************
697 * CALCULATE INTERACTIONS *
698 **************************/
699
700 if (gmx_mm_any_lt(rsq20,rcutoff2))
701 {
702
703 r20 = _mm_mul_pd(rsq20,rinv20);
704
705 /* Compute parameters for interactions between i and j atoms */
706 qq20 = _mm_mul_pd(iq2,jq0);
707
708 /* EWALD ELECTROSTATICS */
709
710 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
711 ewrt = _mm_mul_pd(r20,ewtabscale);
712 ewitab = _mm_cvttpd_epi32(ewrt);
713 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
714 ewitab = _mm_slli_epi32(ewitab,2);
715 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
716 ewtabD = _mm_setzero_pd();
717 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
718 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
719 ewtabFn = _mm_setzero_pd();
720 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
721 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
722 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
723 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
724 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
725
726 d = _mm_sub_pd(r20,rswitch);
727 d = _mm_max_pd(d,_mm_setzero_pd());
728 d2 = _mm_mul_pd(d,d);
729 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
730
731 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
732
733 /* Evaluate switch function */
734 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
735 felec = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
736 velec = _mm_mul_pd(velec,sw);
737 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
738
739 /* Update potential sum for this i atom from the interaction with this j atom. */
740 velec = _mm_and_pd(velec,cutoff_mask);
741 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
742 velecsum = _mm_add_pd(velecsum,velec);
743
744 fscal = felec;
745
746 fscal = _mm_and_pd(fscal,cutoff_mask);
747
748 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
749
750 /* Calculate temporary vectorial force */
751 tx = _mm_mul_pd(fscal,dx20);
752 ty = _mm_mul_pd(fscal,dy20);
753 tz = _mm_mul_pd(fscal,dz20);
754
755 /* Update vectorial force */
756 fix2 = _mm_add_pd(fix2,tx);
757 fiy2 = _mm_add_pd(fiy2,ty);
758 fiz2 = _mm_add_pd(fiz2,tz);
759
760 fjx0 = _mm_add_pd(fjx0,tx);
761 fjy0 = _mm_add_pd(fjy0,ty);
762 fjz0 = _mm_add_pd(fjz0,tz);
763
764 }
765
766 /**************************
767 * CALCULATE INTERACTIONS *
768 **************************/
769
770 if (gmx_mm_any_lt(rsq30,rcutoff2))
771 {
772
773 r30 = _mm_mul_pd(rsq30,rinv30);
774
775 /* Compute parameters for interactions between i and j atoms */
776 qq30 = _mm_mul_pd(iq3,jq0);
777
778 /* EWALD ELECTROSTATICS */
779
780 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
781 ewrt = _mm_mul_pd(r30,ewtabscale);
782 ewitab = _mm_cvttpd_epi32(ewrt);
783 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
784 ewitab = _mm_slli_epi32(ewitab,2);
785 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
786 ewtabD = _mm_setzero_pd();
787 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
788 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
789 ewtabFn = _mm_setzero_pd();
790 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
791 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
792 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
793 velec = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
794 felec = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
795
796 d = _mm_sub_pd(r30,rswitch);
797 d = _mm_max_pd(d,_mm_setzero_pd());
798 d2 = _mm_mul_pd(d,d);
799 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
800
801 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
802
803 /* Evaluate switch function */
804 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
805 felec = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv30,_mm_mul_pd(velec,dsw)) );
806 velec = _mm_mul_pd(velec,sw);
807 cutoff_mask = _mm_cmplt_pd(rsq30,rcutoff2);
808
809 /* Update potential sum for this i atom from the interaction with this j atom. */
810 velec = _mm_and_pd(velec,cutoff_mask);
811 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
812 velecsum = _mm_add_pd(velecsum,velec);
813
814 fscal = felec;
815
816 fscal = _mm_and_pd(fscal,cutoff_mask);
817
818 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
819
820 /* Calculate temporary vectorial force */
821 tx = _mm_mul_pd(fscal,dx30);
822 ty = _mm_mul_pd(fscal,dy30);
823 tz = _mm_mul_pd(fscal,dz30);
824
825 /* Update vectorial force */
826 fix3 = _mm_add_pd(fix3,tx);
827 fiy3 = _mm_add_pd(fiy3,ty);
828 fiz3 = _mm_add_pd(fiz3,tz);
829
830 fjx0 = _mm_add_pd(fjx0,tx);
831 fjy0 = _mm_add_pd(fjy0,ty);
832 fjz0 = _mm_add_pd(fjz0,tz);
833
834 }
835
836 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
837
838 /* Inner loop uses 257 flops */
839 }
840
841 /* End of innermost loop */
842
843 gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
844 f+i_coord_offset,fshift+i_shift_offset);
845
846 ggid = gid[iidx];
847 /* Update potential energies */
848 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
849 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
850
851 /* Increment number of inner iterations */
852 inneriter += j_index_end - j_index_start;
853
854 /* Outer loop uses 26 flops */
855 }
856
857 /* Increment number of outer iterations */
858 outeriter += nri;
859
860 /* Update outer/inner flops */
861
862 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*257);
863 }
864 /*
865 * Gromacs nonbonded kernel: nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_F_sse2_double
866 * Electrostatics interaction: Ewald
867 * VdW interaction: LennardJones
868 * Geometry: Water4-Particle
869 * Calculate force/pot: Force
870 */
871 void
872 nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_F_sse2_double
873 (t_nblist * gmx_restrict nlist,
874 rvec * gmx_restrict xx,
875 rvec * gmx_restrict ff,
876 t_forcerec * gmx_restrict fr,
877 t_mdatoms * gmx_restrict mdatoms,
878 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
879 t_nrnb * gmx_restrict nrnb)
880 {
881 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
882 * just 0 for non-waters.
883 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
884 * jnr indices corresponding to data put in the four positions in the SIMD register.
885 */
886 int i_shift_offset,i_coord_offset,outeriter,inneriter;
887 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
888 int jnrA,jnrB;
889 int j_coord_offsetA,j_coord_offsetB;
890 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
891 real rcutoff_scalar;
892 real *shiftvec,*fshift,*x,*f;
893 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
894 int vdwioffset0;
895 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
896 int vdwioffset1;
897 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
898 int vdwioffset2;
899 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
900 int vdwioffset3;
901 __m128d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
902 int vdwjidx0A,vdwjidx0B;
903 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
904 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
905 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
906 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
907 __m128d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
908 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
909 real *charge;
910 int nvdwtype;
911 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
912 int *vdwtype;
913 real *vdwparam;
914 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
915 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
916 __m128i ewitab;
917 __m128d ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
918 real *ewtab;
919 __m128d rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
920 real rswitch_scalar,d_scalar;
921 __m128d dummy_mask,cutoff_mask;
922 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
923 __m128d one = _mm_set1_pd(1.0);
924 __m128d two = _mm_set1_pd(2.0);
925 x = xx[0];
926 f = ff[0];
927
928 nri = nlist->nri;
929 iinr = nlist->iinr;
930 jindex = nlist->jindex;
931 jjnr = nlist->jjnr;
932 shiftidx = nlist->shift;
933 gid = nlist->gid;
934 shiftvec = fr->shift_vec[0];
935 fshift = fr->fshift[0];
936 facel = _mm_set1_pd(fr->epsfac);
937 charge = mdatoms->chargeA;
938 nvdwtype = fr->ntype;
939 vdwparam = fr->nbfp;
940 vdwtype = mdatoms->typeA;
941
942 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
943 ewtab = fr->ic->tabq_coul_FDV0;
944 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
945 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
946
947 /* Setup water-specific parameters */
948 inr = nlist->iinr[0];
949 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
950 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
951 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
952 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
953
954 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
955 rcutoff_scalar = fr->rcoulomb;
956 rcutoff = _mm_set1_pd(rcutoff_scalar);
957 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
958
959 rswitch_scalar = fr->rcoulomb_switch;
960 rswitch = _mm_set1_pd(rswitch_scalar);
961 /* Setup switch parameters */
962 d_scalar = rcutoff_scalar-rswitch_scalar;
963 d = _mm_set1_pd(d_scalar);
964 swV3 = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
965 swV4 = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
966 swV5 = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
967 swF2 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
968 swF3 = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
969 swF4 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
970
971 /* Avoid stupid compiler warnings */
972 jnrA = jnrB = 0;
973 j_coord_offsetA = 0;
974 j_coord_offsetB = 0;
975
976 outeriter = 0;
977 inneriter = 0;
978
979 /* Start outer loop over neighborlists */
980 for(iidx=0; iidx<nri; iidx++)
981 {
982 /* Load shift vector for this list */
983 i_shift_offset = DIM*shiftidx[iidx];
984
985 /* Load limits for loop over neighbors */
986 j_index_start = jindex[iidx];
987 j_index_end = jindex[iidx+1];
988
989 /* Get outer coordinate index */
990 inr = iinr[iidx];
991 i_coord_offset = DIM*inr;
992
993 /* Load i particle coords and add shift vector */
994 gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
995 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
996
997 fix0 = _mm_setzero_pd();
998 fiy0 = _mm_setzero_pd();
999 fiz0 = _mm_setzero_pd();
1000 fix1 = _mm_setzero_pd();
1001 fiy1 = _mm_setzero_pd();
1002 fiz1 = _mm_setzero_pd();
1003 fix2 = _mm_setzero_pd();
1004 fiy2 = _mm_setzero_pd();
1005 fiz2 = _mm_setzero_pd();
1006 fix3 = _mm_setzero_pd();
1007 fiy3 = _mm_setzero_pd();
1008 fiz3 = _mm_setzero_pd();
1009
1010 /* Start inner kernel loop */
1011 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
1012 {
1013
1014 /* Get j neighbor index, and coordinate index */
1015 jnrA = jjnr[jidx];
1016 jnrB = jjnr[jidx+1];
1017 j_coord_offsetA = DIM*jnrA;
1018 j_coord_offsetB = DIM*jnrB;
1019
1020 /* load j atom coordinates */
1021 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1022 &jx0,&jy0,&jz0);
1023
1024 /* Calculate displacement vector */
1025 dx00 = _mm_sub_pd(ix0,jx0);
1026 dy00 = _mm_sub_pd(iy0,jy0);
1027 dz00 = _mm_sub_pd(iz0,jz0);
1028 dx10 = _mm_sub_pd(ix1,jx0);
1029 dy10 = _mm_sub_pd(iy1,jy0);
1030 dz10 = _mm_sub_pd(iz1,jz0);
1031 dx20 = _mm_sub_pd(ix2,jx0);
1032 dy20 = _mm_sub_pd(iy2,jy0);
1033 dz20 = _mm_sub_pd(iz2,jz0);
1034 dx30 = _mm_sub_pd(ix3,jx0);
1035 dy30 = _mm_sub_pd(iy3,jy0);
1036 dz30 = _mm_sub_pd(iz3,jz0);
1037
1038 /* Calculate squared distance and things based on it */
1039 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1040 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1041 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1042 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1043
1044 rinv00 = gmx_mm_invsqrt_pd(rsq00);
1045 rinv10 = gmx_mm_invsqrt_pd(rsq10);
1046 rinv20 = gmx_mm_invsqrt_pd(rsq20);
1047 rinv30 = gmx_mm_invsqrt_pd(rsq30);
1048
1049 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
1050 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
1051 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
1052 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
1053
1054 /* Load parameters for j particles */
1055 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
1056 vdwjidx0A = 2*vdwtype[jnrA+0];
1057 vdwjidx0B = 2*vdwtype[jnrB+0];
1058
1059 fjx0 = _mm_setzero_pd();
1060 fjy0 = _mm_setzero_pd();
1061 fjz0 = _mm_setzero_pd();
1062
1063 /**************************
1064 * CALCULATE INTERACTIONS *
1065 **************************/
1066
1067 if (gmx_mm_any_lt(rsq00,rcutoff2))
1068 {
1069
1070 r00 = _mm_mul_pd(rsq00,rinv00);
1071
1072 /* Compute parameters for interactions between i and j atoms */
1073 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
1074 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
1075
1076 /* LENNARD-JONES DISPERSION/REPULSION */
1077
1078 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1079 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
1080 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
1081 vvdw = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
1082 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
1083
1084 d = _mm_sub_pd(r00,rswitch);
1085 d = _mm_max_pd(d,_mm_setzero_pd());
1086 d2 = _mm_mul_pd(d,d);
1087 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
1088
1089 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1090
1091 /* Evaluate switch function */
1092 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1093 fvdw = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
1094 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
1095
1096 fscal = fvdw;
1097
1098 fscal = _mm_and_pd(fscal,cutoff_mask);
1099
1100 /* Calculate temporary vectorial force */
1101 tx = _mm_mul_pd(fscal,dx00);
1102 ty = _mm_mul_pd(fscal,dy00);
1103 tz = _mm_mul_pd(fscal,dz00);
1104
1105 /* Update vectorial force */
1106 fix0 = _mm_add_pd(fix0,tx);
1107 fiy0 = _mm_add_pd(fiy0,ty);
1108 fiz0 = _mm_add_pd(fiz0,tz);
1109
1110 fjx0 = _mm_add_pd(fjx0,tx);
1111 fjy0 = _mm_add_pd(fjy0,ty);
1112 fjz0 = _mm_add_pd(fjz0,tz);
1113
1114 }
1115
1116 /**************************
1117 * CALCULATE INTERACTIONS *
1118 **************************/
1119
1120 if (gmx_mm_any_lt(rsq10,rcutoff2))
1121 {
1122
1123 r10 = _mm_mul_pd(rsq10,rinv10);
1124
1125 /* Compute parameters for interactions between i and j atoms */
1126 qq10 = _mm_mul_pd(iq1,jq0);
1127
1128 /* EWALD ELECTROSTATICS */
1129
1130 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1131 ewrt = _mm_mul_pd(r10,ewtabscale);
1132 ewitab = _mm_cvttpd_epi32(ewrt);
1133 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1134 ewitab = _mm_slli_epi32(ewitab,2);
1135 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1136 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1137 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1138 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1139 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
1140 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1141 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1142 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1143 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
1144 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1145
1146 d = _mm_sub_pd(r10,rswitch);
1147 d = _mm_max_pd(d,_mm_setzero_pd());
1148 d2 = _mm_mul_pd(d,d);
1149 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
1150
1151 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1152
1153 /* Evaluate switch function */
1154 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1155 felec = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
1156 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
1157
1158 fscal = felec;
1159
1160 fscal = _mm_and_pd(fscal,cutoff_mask);
1161
1162 /* Calculate temporary vectorial force */
1163 tx = _mm_mul_pd(fscal,dx10);
1164 ty = _mm_mul_pd(fscal,dy10);
1165 tz = _mm_mul_pd(fscal,dz10);
1166
1167 /* Update vectorial force */
1168 fix1 = _mm_add_pd(fix1,tx);
1169 fiy1 = _mm_add_pd(fiy1,ty);
1170 fiz1 = _mm_add_pd(fiz1,tz);
1171
1172 fjx0 = _mm_add_pd(fjx0,tx);
1173 fjy0 = _mm_add_pd(fjy0,ty);
1174 fjz0 = _mm_add_pd(fjz0,tz);
1175
1176 }
1177
1178 /**************************
1179 * CALCULATE INTERACTIONS *
1180 **************************/
1181
1182 if (gmx_mm_any_lt(rsq20,rcutoff2))
1183 {
1184
1185 r20 = _mm_mul_pd(rsq20,rinv20);
1186
1187 /* Compute parameters for interactions between i and j atoms */
1188 qq20 = _mm_mul_pd(iq2,jq0);
1189
1190 /* EWALD ELECTROSTATICS */
1191
1192 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1193 ewrt = _mm_mul_pd(r20,ewtabscale);
1194 ewitab = _mm_cvttpd_epi32(ewrt);
1195 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1196 ewitab = _mm_slli_epi32(ewitab,2);
1197 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1198 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1199 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1200 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1201 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
1202 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1203 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1204 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1205 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
1206 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1207
1208 d = _mm_sub_pd(r20,rswitch);
1209 d = _mm_max_pd(d,_mm_setzero_pd());
1210 d2 = _mm_mul_pd(d,d);
1211 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
1212
1213 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1214
1215 /* Evaluate switch function */
1216 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1217 felec = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
1218 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
1219
1220 fscal = felec;
1221
1222 fscal = _mm_and_pd(fscal,cutoff_mask);
1223
1224 /* Calculate temporary vectorial force */
1225 tx = _mm_mul_pd(fscal,dx20);
1226 ty = _mm_mul_pd(fscal,dy20);
1227 tz = _mm_mul_pd(fscal,dz20);
1228
1229 /* Update vectorial force */
1230 fix2 = _mm_add_pd(fix2,tx);
1231 fiy2 = _mm_add_pd(fiy2,ty);
1232 fiz2 = _mm_add_pd(fiz2,tz);
1233
1234 fjx0 = _mm_add_pd(fjx0,tx);
1235 fjy0 = _mm_add_pd(fjy0,ty);
1236 fjz0 = _mm_add_pd(fjz0,tz);
1237
1238 }
1239
1240 /**************************
1241 * CALCULATE INTERACTIONS *
1242 **************************/
1243
1244 if (gmx_mm_any_lt(rsq30,rcutoff2))
1245 {
1246
1247 r30 = _mm_mul_pd(rsq30,rinv30);
1248
1249 /* Compute parameters for interactions between i and j atoms */
1250 qq30 = _mm_mul_pd(iq3,jq0);
1251
1252 /* EWALD ELECTROSTATICS */
1253
1254 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1255 ewrt = _mm_mul_pd(r30,ewtabscale);
1256 ewitab = _mm_cvttpd_epi32(ewrt);
1257 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1258 ewitab = _mm_slli_epi32(ewitab,2);
1259 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1260 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1261 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1262 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1263 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
1264 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1265 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1266 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1267 velec = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
1268 felec = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1269
1270 d = _mm_sub_pd(r30,rswitch);
1271 d = _mm_max_pd(d,_mm_setzero_pd());
1272 d2 = _mm_mul_pd(d,d);
1273 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
1274
1275 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1276
1277 /* Evaluate switch function */
1278 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1279 felec = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv30,_mm_mul_pd(velec,dsw)) );
1280 cutoff_mask = _mm_cmplt_pd(rsq30,rcutoff2);
1281
1282 fscal = felec;
1283
1284 fscal = _mm_and_pd(fscal,cutoff_mask);
1285
1286 /* Calculate temporary vectorial force */
1287 tx = _mm_mul_pd(fscal,dx30);
1288 ty = _mm_mul_pd(fscal,dy30);
1289 tz = _mm_mul_pd(fscal,dz30);
1290
1291 /* Update vectorial force */
1292 fix3 = _mm_add_pd(fix3,tx);
1293 fiy3 = _mm_add_pd(fiy3,ty);
1294 fiz3 = _mm_add_pd(fiz3,tz);
1295
1296 fjx0 = _mm_add_pd(fjx0,tx);
1297 fjy0 = _mm_add_pd(fjy0,ty);
1298 fjz0 = _mm_add_pd(fjz0,tz);
1299
1300 }
1301
1302 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1303
1304 /* Inner loop uses 245 flops */
1305 }
1306
1307 if(jidx<j_index_end)
1308 {
1309
1310 jnrA = jjnr[jidx];
1311 j_coord_offsetA = DIM*jnrA;
1312
1313 /* load j atom coordinates */
1314 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1315 &jx0,&jy0,&jz0);
1316
1317 /* Calculate displacement vector */
1318 dx00 = _mm_sub_pd(ix0,jx0);
1319 dy00 = _mm_sub_pd(iy0,jy0);
1320 dz00 = _mm_sub_pd(iz0,jz0);
1321 dx10 = _mm_sub_pd(ix1,jx0);
1322 dy10 = _mm_sub_pd(iy1,jy0);
1323 dz10 = _mm_sub_pd(iz1,jz0);
1324 dx20 = _mm_sub_pd(ix2,jx0);
1325 dy20 = _mm_sub_pd(iy2,jy0);
1326 dz20 = _mm_sub_pd(iz2,jz0);
1327 dx30 = _mm_sub_pd(ix3,jx0);
1328 dy30 = _mm_sub_pd(iy3,jy0);
1329 dz30 = _mm_sub_pd(iz3,jz0);
1330
1331 /* Calculate squared distance and things based on it */
1332 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1333 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1334 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1335 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1336
1337 rinv00 = gmx_mm_invsqrt_pd(rsq00);
1338 rinv10 = gmx_mm_invsqrt_pd(rsq10);
1339 rinv20 = gmx_mm_invsqrt_pd(rsq20);
1340 rinv30 = gmx_mm_invsqrt_pd(rsq30);
1341
1342 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
1343 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
1344 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
1345 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
1346
1347 /* Load parameters for j particles */
1348 jq0 = _mm_load_sd(charge+jnrA+0);
1349 vdwjidx0A = 2*vdwtype[jnrA+0];
1350
1351 fjx0 = _mm_setzero_pd();
1352 fjy0 = _mm_setzero_pd();
1353 fjz0 = _mm_setzero_pd();
1354
1355 /**************************
1356 * CALCULATE INTERACTIONS *
1357 **************************/
1358
1359 if (gmx_mm_any_lt(rsq00,rcutoff2))
1360 {
1361
1362 r00 = _mm_mul_pd(rsq00,rinv00);
1363
1364 /* Compute parameters for interactions between i and j atoms */
1365 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1366
1367 /* LENNARD-JONES DISPERSION/REPULSION */
1368
1369 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1370 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
1371 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
1372 vvdw = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
1373 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
1374
1375 d = _mm_sub_pd(r00,rswitch);
1376 d = _mm_max_pd(d,_mm_setzero_pd());
1377 d2 = _mm_mul_pd(d,d);
1378 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
1379
1380 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1381
1382 /* Evaluate switch function */
1383 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1384 fvdw = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
1385 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
1386
1387 fscal = fvdw;
1388
1389 fscal = _mm_and_pd(fscal,cutoff_mask);
1390
1391 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1392
1393 /* Calculate temporary vectorial force */
1394 tx = _mm_mul_pd(fscal,dx00);
1395 ty = _mm_mul_pd(fscal,dy00);
1396 tz = _mm_mul_pd(fscal,dz00);
1397
1398 /* Update vectorial force */
1399 fix0 = _mm_add_pd(fix0,tx);
1400 fiy0 = _mm_add_pd(fiy0,ty);
1401 fiz0 = _mm_add_pd(fiz0,tz);
1402
1403 fjx0 = _mm_add_pd(fjx0,tx);
1404 fjy0 = _mm_add_pd(fjy0,ty);
1405 fjz0 = _mm_add_pd(fjz0,tz);
1406
1407 }
1408
1409 /**************************
1410 * CALCULATE INTERACTIONS *
1411 **************************/
1412
1413 if (gmx_mm_any_lt(rsq10,rcutoff2))
1414 {
1415
1416 r10 = _mm_mul_pd(rsq10,rinv10);
1417
1418 /* Compute parameters for interactions between i and j atoms */
1419 qq10 = _mm_mul_pd(iq1,jq0);
1420
1421 /* EWALD ELECTROSTATICS */
1422
1423 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1424 ewrt = _mm_mul_pd(r10,ewtabscale);
1425 ewitab = _mm_cvttpd_epi32(ewrt);
1426 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1427 ewitab = _mm_slli_epi32(ewitab,2);
1428 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1429 ewtabD = _mm_setzero_pd();
1430 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1431 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1432 ewtabFn = _mm_setzero_pd();
1433 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1434 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1435 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1436 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
1437 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1438
1439 d = _mm_sub_pd(r10,rswitch);
1440 d = _mm_max_pd(d,_mm_setzero_pd());
1441 d2 = _mm_mul_pd(d,d);
1442 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
1443
1444 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1445
1446 /* Evaluate switch function */
1447 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1448 felec = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
1449 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
1450
1451 fscal = felec;
1452
1453 fscal = _mm_and_pd(fscal,cutoff_mask);
1454
1455 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1456
1457 /* Calculate temporary vectorial force */
1458 tx = _mm_mul_pd(fscal,dx10);
1459 ty = _mm_mul_pd(fscal,dy10);
1460 tz = _mm_mul_pd(fscal,dz10);
1461
1462 /* Update vectorial force */
1463 fix1 = _mm_add_pd(fix1,tx);
1464 fiy1 = _mm_add_pd(fiy1,ty);
1465 fiz1 = _mm_add_pd(fiz1,tz);
1466
1467 fjx0 = _mm_add_pd(fjx0,tx);
1468 fjy0 = _mm_add_pd(fjy0,ty);
1469 fjz0 = _mm_add_pd(fjz0,tz);
1470
1471 }
1472
1473 /**************************
1474 * CALCULATE INTERACTIONS *
1475 **************************/
1476
1477 if (gmx_mm_any_lt(rsq20,rcutoff2))
1478 {
1479
1480 r20 = _mm_mul_pd(rsq20,rinv20);
1481
1482 /* Compute parameters for interactions between i and j atoms */
1483 qq20 = _mm_mul_pd(iq2,jq0);
1484
1485 /* EWALD ELECTROSTATICS */
1486
1487 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1488 ewrt = _mm_mul_pd(r20,ewtabscale);
1489 ewitab = _mm_cvttpd_epi32(ewrt);
1490 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1491 ewitab = _mm_slli_epi32(ewitab,2);
1492 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1493 ewtabD = _mm_setzero_pd();
1494 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1495 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1496 ewtabFn = _mm_setzero_pd();
1497 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1498 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1499 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1500 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
1501 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1502
1503 d = _mm_sub_pd(r20,rswitch);
1504 d = _mm_max_pd(d,_mm_setzero_pd());
1505 d2 = _mm_mul_pd(d,d);
1506 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
1507
1508 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1509
1510 /* Evaluate switch function */
1511 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1512 felec = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
1513 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
1514
1515 fscal = felec;
1516
1517 fscal = _mm_and_pd(fscal,cutoff_mask);
1518
1519 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1520
1521 /* Calculate temporary vectorial force */
1522 tx = _mm_mul_pd(fscal,dx20);
1523 ty = _mm_mul_pd(fscal,dy20);
1524 tz = _mm_mul_pd(fscal,dz20);
1525
1526 /* Update vectorial force */
1527 fix2 = _mm_add_pd(fix2,tx);
1528 fiy2 = _mm_add_pd(fiy2,ty);
1529 fiz2 = _mm_add_pd(fiz2,tz);
1530
1531 fjx0 = _mm_add_pd(fjx0,tx);
1532 fjy0 = _mm_add_pd(fjy0,ty);
1533 fjz0 = _mm_add_pd(fjz0,tz);
1534
1535 }
1536
1537 /**************************
1538 * CALCULATE INTERACTIONS *
1539 **************************/
1540
1541 if (gmx_mm_any_lt(rsq30,rcutoff2))
1542 {
1543
1544 r30 = _mm_mul_pd(rsq30,rinv30);
1545
1546 /* Compute parameters for interactions between i and j atoms */
1547 qq30 = _mm_mul_pd(iq3,jq0);
1548
1549 /* EWALD ELECTROSTATICS */
1550
1551 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1552 ewrt = _mm_mul_pd(r30,ewtabscale);
1553 ewitab = _mm_cvttpd_epi32(ewrt);
1554 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1555 ewitab = _mm_slli_epi32(ewitab,2);
1556 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1557 ewtabD = _mm_setzero_pd();
1558 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1559 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1560 ewtabFn = _mm_setzero_pd();
1561 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1562 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1563 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1564 velec = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
1565 felec = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1566
1567 d = _mm_sub_pd(r30,rswitch);
1568 d = _mm_max_pd(d,_mm_setzero_pd());
1569 d2 = _mm_mul_pd(d,d);
1570 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
1571
1572 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1573
1574 /* Evaluate switch function */
1575 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1576 felec = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv30,_mm_mul_pd(velec,dsw)) );
1577 cutoff_mask = _mm_cmplt_pd(rsq30,rcutoff2);
1578
1579 fscal = felec;
1580
1581 fscal = _mm_and_pd(fscal,cutoff_mask);
1582
1583 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1584
1585 /* Calculate temporary vectorial force */
1586 tx = _mm_mul_pd(fscal,dx30);
1587 ty = _mm_mul_pd(fscal,dy30);
1588 tz = _mm_mul_pd(fscal,dz30);
1589
1590 /* Update vectorial force */
1591 fix3 = _mm_add_pd(fix3,tx);
1592 fiy3 = _mm_add_pd(fiy3,ty);
1593 fiz3 = _mm_add_pd(fiz3,tz);
1594
1595 fjx0 = _mm_add_pd(fjx0,tx);
1596 fjy0 = _mm_add_pd(fjy0,ty);
1597 fjz0 = _mm_add_pd(fjz0,tz);
1598
1599 }
1600
1601 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1602
1603 /* Inner loop uses 245 flops */
1604 }
1605
1606 /* End of innermost loop */
1607
1608 gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1609 f+i_coord_offset,fshift+i_shift_offset);
1610
1611 /* Increment number of inner iterations */
1612 inneriter += j_index_end - j_index_start;
1613
1614 /* Outer loop uses 24 flops */
1615 }
1616
1617 /* Increment number of outer iterations */
1618 outeriter += nri;
1619
1620 /* Update outer/inner flops */
1621
1622 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*245);
1623 }
The ultimate molecular dynamics simulation package