2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014, 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.
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.
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.
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.
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.
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.
36 * Note: this file was generated by the GROMACS sse4_1_double kernel generator.
42 #include "../nb_kernel.h"
43 #include "types/simple.h"
44 #include "gromacs/math/vec.h"
47 #include "gromacs/simd/math_x86_sse4_1_double.h"
48 #include "kernelutil_x86_sse4_1_double.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3W3_VF_sse4_1_double
52 * Electrostatics interaction: Ewald
53 * VdW interaction: LJEwald
54 * Geometry: Water3-Water3
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3W3_VF_sse4_1_double
59 (t_nblist * gmx_restrict nlist,
60 rvec * gmx_restrict xx,
61 rvec * gmx_restrict ff,
62 t_forcerec * gmx_restrict fr,
63 t_mdatoms * gmx_restrict mdatoms,
64 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
65 t_nrnb * gmx_restrict nrnb)
67 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
68 * just 0 for non-waters.
69 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
70 * jnr indices corresponding to data put in the four positions in the SIMD register.
72 int i_shift_offset,i_coord_offset,outeriter,inneriter;
73 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
75 int j_coord_offsetA,j_coord_offsetB;
76 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
78 real *shiftvec,*fshift,*x,*f;
79 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
81 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
83 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
85 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
86 int vdwjidx0A,vdwjidx0B;
87 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
88 int vdwjidx1A,vdwjidx1B;
89 __m128d jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
90 int vdwjidx2A,vdwjidx2B;
91 __m128d jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
92 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
93 __m128d dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
94 __m128d dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
95 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
96 __m128d dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
97 __m128d dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
98 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
99 __m128d dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
100 __m128d dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
101 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
104 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
107 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
108 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
118 __m128d ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
120 __m128d one_half = _mm_set1_pd(0.5);
121 __m128d minus_one = _mm_set1_pd(-1.0);
123 __m128d ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
125 __m128d dummy_mask,cutoff_mask;
126 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
127 __m128d one = _mm_set1_pd(1.0);
128 __m128d two = _mm_set1_pd(2.0);
134 jindex = nlist->jindex;
136 shiftidx = nlist->shift;
138 shiftvec = fr->shift_vec[0];
139 fshift = fr->fshift[0];
140 facel = _mm_set1_pd(fr->epsfac);
141 charge = mdatoms->chargeA;
142 nvdwtype = fr->ntype;
144 vdwtype = mdatoms->typeA;
145 vdwgridparam = fr->ljpme_c6grid;
146 sh_lj_ewald = _mm_set1_pd(fr->ic->sh_lj_ewald);
147 ewclj = _mm_set1_pd(fr->ewaldcoeff_lj);
148 ewclj2 = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
150 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
151 ewtab = fr->ic->tabq_coul_FDV0;
152 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
153 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
155 /* Setup water-specific parameters */
156 inr = nlist->iinr[0];
157 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
158 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
159 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
160 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
162 jq0 = _mm_set1_pd(charge[inr+0]);
163 jq1 = _mm_set1_pd(charge[inr+1]);
164 jq2 = _mm_set1_pd(charge[inr+2]);
165 vdwjidx0A = 2*vdwtype[inr+0];
166 qq00 = _mm_mul_pd(iq0,jq0);
167 c6_00 = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
168 c12_00 = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
169 c6grid_00 = _mm_set1_pd(vdwgridparam[vdwioffset0+vdwjidx0A]);
170 qq01 = _mm_mul_pd(iq0,jq1);
171 qq02 = _mm_mul_pd(iq0,jq2);
172 qq10 = _mm_mul_pd(iq1,jq0);
173 qq11 = _mm_mul_pd(iq1,jq1);
174 qq12 = _mm_mul_pd(iq1,jq2);
175 qq20 = _mm_mul_pd(iq2,jq0);
176 qq21 = _mm_mul_pd(iq2,jq1);
177 qq22 = _mm_mul_pd(iq2,jq2);
179 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
180 rcutoff_scalar = fr->rcoulomb;
181 rcutoff = _mm_set1_pd(rcutoff_scalar);
182 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
184 sh_vdw_invrcut6 = _mm_set1_pd(fr->ic->sh_invrc6);
185 rvdw = _mm_set1_pd(fr->rvdw);
187 /* Avoid stupid compiler warnings */
195 /* Start outer loop over neighborlists */
196 for(iidx=0; iidx<nri; iidx++)
198 /* Load shift vector for this list */
199 i_shift_offset = DIM*shiftidx[iidx];
201 /* Load limits for loop over neighbors */
202 j_index_start = jindex[iidx];
203 j_index_end = jindex[iidx+1];
205 /* Get outer coordinate index */
207 i_coord_offset = DIM*inr;
209 /* Load i particle coords and add shift vector */
210 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
211 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
213 fix0 = _mm_setzero_pd();
214 fiy0 = _mm_setzero_pd();
215 fiz0 = _mm_setzero_pd();
216 fix1 = _mm_setzero_pd();
217 fiy1 = _mm_setzero_pd();
218 fiz1 = _mm_setzero_pd();
219 fix2 = _mm_setzero_pd();
220 fiy2 = _mm_setzero_pd();
221 fiz2 = _mm_setzero_pd();
223 /* Reset potential sums */
224 velecsum = _mm_setzero_pd();
225 vvdwsum = _mm_setzero_pd();
227 /* Start inner kernel loop */
228 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
231 /* Get j neighbor index, and coordinate index */
234 j_coord_offsetA = DIM*jnrA;
235 j_coord_offsetB = DIM*jnrB;
237 /* load j atom coordinates */
238 gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
239 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
241 /* Calculate displacement vector */
242 dx00 = _mm_sub_pd(ix0,jx0);
243 dy00 = _mm_sub_pd(iy0,jy0);
244 dz00 = _mm_sub_pd(iz0,jz0);
245 dx01 = _mm_sub_pd(ix0,jx1);
246 dy01 = _mm_sub_pd(iy0,jy1);
247 dz01 = _mm_sub_pd(iz0,jz1);
248 dx02 = _mm_sub_pd(ix0,jx2);
249 dy02 = _mm_sub_pd(iy0,jy2);
250 dz02 = _mm_sub_pd(iz0,jz2);
251 dx10 = _mm_sub_pd(ix1,jx0);
252 dy10 = _mm_sub_pd(iy1,jy0);
253 dz10 = _mm_sub_pd(iz1,jz0);
254 dx11 = _mm_sub_pd(ix1,jx1);
255 dy11 = _mm_sub_pd(iy1,jy1);
256 dz11 = _mm_sub_pd(iz1,jz1);
257 dx12 = _mm_sub_pd(ix1,jx2);
258 dy12 = _mm_sub_pd(iy1,jy2);
259 dz12 = _mm_sub_pd(iz1,jz2);
260 dx20 = _mm_sub_pd(ix2,jx0);
261 dy20 = _mm_sub_pd(iy2,jy0);
262 dz20 = _mm_sub_pd(iz2,jz0);
263 dx21 = _mm_sub_pd(ix2,jx1);
264 dy21 = _mm_sub_pd(iy2,jy1);
265 dz21 = _mm_sub_pd(iz2,jz1);
266 dx22 = _mm_sub_pd(ix2,jx2);
267 dy22 = _mm_sub_pd(iy2,jy2);
268 dz22 = _mm_sub_pd(iz2,jz2);
270 /* Calculate squared distance and things based on it */
271 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
272 rsq01 = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
273 rsq02 = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
274 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
275 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
276 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
277 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
278 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
279 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
281 rinv00 = gmx_mm_invsqrt_pd(rsq00);
282 rinv01 = gmx_mm_invsqrt_pd(rsq01);
283 rinv02 = gmx_mm_invsqrt_pd(rsq02);
284 rinv10 = gmx_mm_invsqrt_pd(rsq10);
285 rinv11 = gmx_mm_invsqrt_pd(rsq11);
286 rinv12 = gmx_mm_invsqrt_pd(rsq12);
287 rinv20 = gmx_mm_invsqrt_pd(rsq20);
288 rinv21 = gmx_mm_invsqrt_pd(rsq21);
289 rinv22 = gmx_mm_invsqrt_pd(rsq22);
291 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
292 rinvsq01 = _mm_mul_pd(rinv01,rinv01);
293 rinvsq02 = _mm_mul_pd(rinv02,rinv02);
294 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
295 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
296 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
297 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
298 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
299 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
301 fjx0 = _mm_setzero_pd();
302 fjy0 = _mm_setzero_pd();
303 fjz0 = _mm_setzero_pd();
304 fjx1 = _mm_setzero_pd();
305 fjy1 = _mm_setzero_pd();
306 fjz1 = _mm_setzero_pd();
307 fjx2 = _mm_setzero_pd();
308 fjy2 = _mm_setzero_pd();
309 fjz2 = _mm_setzero_pd();
311 /**************************
312 * CALCULATE INTERACTIONS *
313 **************************/
315 if (gmx_mm_any_lt(rsq00,rcutoff2))
318 r00 = _mm_mul_pd(rsq00,rinv00);
320 /* EWALD ELECTROSTATICS */
322 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
323 ewrt = _mm_mul_pd(r00,ewtabscale);
324 ewitab = _mm_cvttpd_epi32(ewrt);
325 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
326 ewitab = _mm_slli_epi32(ewitab,2);
327 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
328 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
329 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
330 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
331 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
332 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
333 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
334 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
335 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
336 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
338 /* Analytical LJ-PME */
339 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
340 ewcljrsq = _mm_mul_pd(ewclj2,rsq00);
341 ewclj6 = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
342 exponent = gmx_simd_exp_d(ewcljrsq);
343 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
344 poly = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
345 /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
346 vvdw6 = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
347 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
348 vvdw = _mm_sub_pd(_mm_mul_pd( _mm_sub_pd(vvdw12 , _mm_mul_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))),one_twelfth),
349 _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_add_pd(_mm_mul_pd(c6_00,sh_vdw_invrcut6),_mm_mul_pd(c6grid_00,sh_lj_ewald))),one_sixth));
350 /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
351 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,_mm_sub_pd(vvdw6,_mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6)))),rinvsq00);
353 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
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 vvdw = _mm_and_pd(vvdw,cutoff_mask);
359 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
361 fscal = _mm_add_pd(felec,fvdw);
363 fscal = _mm_and_pd(fscal,cutoff_mask);
365 /* Calculate temporary vectorial force */
366 tx = _mm_mul_pd(fscal,dx00);
367 ty = _mm_mul_pd(fscal,dy00);
368 tz = _mm_mul_pd(fscal,dz00);
370 /* Update vectorial force */
371 fix0 = _mm_add_pd(fix0,tx);
372 fiy0 = _mm_add_pd(fiy0,ty);
373 fiz0 = _mm_add_pd(fiz0,tz);
375 fjx0 = _mm_add_pd(fjx0,tx);
376 fjy0 = _mm_add_pd(fjy0,ty);
377 fjz0 = _mm_add_pd(fjz0,tz);
381 /**************************
382 * CALCULATE INTERACTIONS *
383 **************************/
385 if (gmx_mm_any_lt(rsq01,rcutoff2))
388 r01 = _mm_mul_pd(rsq01,rinv01);
390 /* EWALD ELECTROSTATICS */
392 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
393 ewrt = _mm_mul_pd(r01,ewtabscale);
394 ewitab = _mm_cvttpd_epi32(ewrt);
395 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
396 ewitab = _mm_slli_epi32(ewitab,2);
397 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
398 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
399 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
400 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
401 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
402 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
403 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
404 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
405 velec = _mm_mul_pd(qq01,_mm_sub_pd(_mm_sub_pd(rinv01,sh_ewald),velec));
406 felec = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
408 cutoff_mask = _mm_cmplt_pd(rsq01,rcutoff2);
410 /* Update potential sum for this i atom from the interaction with this j atom. */
411 velec = _mm_and_pd(velec,cutoff_mask);
412 velecsum = _mm_add_pd(velecsum,velec);
416 fscal = _mm_and_pd(fscal,cutoff_mask);
418 /* Calculate temporary vectorial force */
419 tx = _mm_mul_pd(fscal,dx01);
420 ty = _mm_mul_pd(fscal,dy01);
421 tz = _mm_mul_pd(fscal,dz01);
423 /* Update vectorial force */
424 fix0 = _mm_add_pd(fix0,tx);
425 fiy0 = _mm_add_pd(fiy0,ty);
426 fiz0 = _mm_add_pd(fiz0,tz);
428 fjx1 = _mm_add_pd(fjx1,tx);
429 fjy1 = _mm_add_pd(fjy1,ty);
430 fjz1 = _mm_add_pd(fjz1,tz);
434 /**************************
435 * CALCULATE INTERACTIONS *
436 **************************/
438 if (gmx_mm_any_lt(rsq02,rcutoff2))
441 r02 = _mm_mul_pd(rsq02,rinv02);
443 /* EWALD ELECTROSTATICS */
445 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
446 ewrt = _mm_mul_pd(r02,ewtabscale);
447 ewitab = _mm_cvttpd_epi32(ewrt);
448 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
449 ewitab = _mm_slli_epi32(ewitab,2);
450 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
451 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
452 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
453 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
454 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
455 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
456 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
457 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
458 velec = _mm_mul_pd(qq02,_mm_sub_pd(_mm_sub_pd(rinv02,sh_ewald),velec));
459 felec = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
461 cutoff_mask = _mm_cmplt_pd(rsq02,rcutoff2);
463 /* Update potential sum for this i atom from the interaction with this j atom. */
464 velec = _mm_and_pd(velec,cutoff_mask);
465 velecsum = _mm_add_pd(velecsum,velec);
469 fscal = _mm_and_pd(fscal,cutoff_mask);
471 /* Calculate temporary vectorial force */
472 tx = _mm_mul_pd(fscal,dx02);
473 ty = _mm_mul_pd(fscal,dy02);
474 tz = _mm_mul_pd(fscal,dz02);
476 /* Update vectorial force */
477 fix0 = _mm_add_pd(fix0,tx);
478 fiy0 = _mm_add_pd(fiy0,ty);
479 fiz0 = _mm_add_pd(fiz0,tz);
481 fjx2 = _mm_add_pd(fjx2,tx);
482 fjy2 = _mm_add_pd(fjy2,ty);
483 fjz2 = _mm_add_pd(fjz2,tz);
487 /**************************
488 * CALCULATE INTERACTIONS *
489 **************************/
491 if (gmx_mm_any_lt(rsq10,rcutoff2))
494 r10 = _mm_mul_pd(rsq10,rinv10);
496 /* EWALD ELECTROSTATICS */
498 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
499 ewrt = _mm_mul_pd(r10,ewtabscale);
500 ewitab = _mm_cvttpd_epi32(ewrt);
501 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
502 ewitab = _mm_slli_epi32(ewitab,2);
503 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
504 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
505 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
506 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
507 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
508 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
509 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
510 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
511 velec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
512 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
514 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
516 /* Update potential sum for this i atom from the interaction with this j atom. */
517 velec = _mm_and_pd(velec,cutoff_mask);
518 velecsum = _mm_add_pd(velecsum,velec);
522 fscal = _mm_and_pd(fscal,cutoff_mask);
524 /* Calculate temporary vectorial force */
525 tx = _mm_mul_pd(fscal,dx10);
526 ty = _mm_mul_pd(fscal,dy10);
527 tz = _mm_mul_pd(fscal,dz10);
529 /* Update vectorial force */
530 fix1 = _mm_add_pd(fix1,tx);
531 fiy1 = _mm_add_pd(fiy1,ty);
532 fiz1 = _mm_add_pd(fiz1,tz);
534 fjx0 = _mm_add_pd(fjx0,tx);
535 fjy0 = _mm_add_pd(fjy0,ty);
536 fjz0 = _mm_add_pd(fjz0,tz);
540 /**************************
541 * CALCULATE INTERACTIONS *
542 **************************/
544 if (gmx_mm_any_lt(rsq11,rcutoff2))
547 r11 = _mm_mul_pd(rsq11,rinv11);
549 /* EWALD ELECTROSTATICS */
551 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
552 ewrt = _mm_mul_pd(r11,ewtabscale);
553 ewitab = _mm_cvttpd_epi32(ewrt);
554 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
555 ewitab = _mm_slli_epi32(ewitab,2);
556 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
557 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
558 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
559 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
560 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
561 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
562 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
563 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
564 velec = _mm_mul_pd(qq11,_mm_sub_pd(_mm_sub_pd(rinv11,sh_ewald),velec));
565 felec = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
567 cutoff_mask = _mm_cmplt_pd(rsq11,rcutoff2);
569 /* Update potential sum for this i atom from the interaction with this j atom. */
570 velec = _mm_and_pd(velec,cutoff_mask);
571 velecsum = _mm_add_pd(velecsum,velec);
575 fscal = _mm_and_pd(fscal,cutoff_mask);
577 /* Calculate temporary vectorial force */
578 tx = _mm_mul_pd(fscal,dx11);
579 ty = _mm_mul_pd(fscal,dy11);
580 tz = _mm_mul_pd(fscal,dz11);
582 /* Update vectorial force */
583 fix1 = _mm_add_pd(fix1,tx);
584 fiy1 = _mm_add_pd(fiy1,ty);
585 fiz1 = _mm_add_pd(fiz1,tz);
587 fjx1 = _mm_add_pd(fjx1,tx);
588 fjy1 = _mm_add_pd(fjy1,ty);
589 fjz1 = _mm_add_pd(fjz1,tz);
593 /**************************
594 * CALCULATE INTERACTIONS *
595 **************************/
597 if (gmx_mm_any_lt(rsq12,rcutoff2))
600 r12 = _mm_mul_pd(rsq12,rinv12);
602 /* EWALD ELECTROSTATICS */
604 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
605 ewrt = _mm_mul_pd(r12,ewtabscale);
606 ewitab = _mm_cvttpd_epi32(ewrt);
607 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
608 ewitab = _mm_slli_epi32(ewitab,2);
609 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
610 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
611 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
612 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
613 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
614 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
615 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
616 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
617 velec = _mm_mul_pd(qq12,_mm_sub_pd(_mm_sub_pd(rinv12,sh_ewald),velec));
618 felec = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
620 cutoff_mask = _mm_cmplt_pd(rsq12,rcutoff2);
622 /* Update potential sum for this i atom from the interaction with this j atom. */
623 velec = _mm_and_pd(velec,cutoff_mask);
624 velecsum = _mm_add_pd(velecsum,velec);
628 fscal = _mm_and_pd(fscal,cutoff_mask);
630 /* Calculate temporary vectorial force */
631 tx = _mm_mul_pd(fscal,dx12);
632 ty = _mm_mul_pd(fscal,dy12);
633 tz = _mm_mul_pd(fscal,dz12);
635 /* Update vectorial force */
636 fix1 = _mm_add_pd(fix1,tx);
637 fiy1 = _mm_add_pd(fiy1,ty);
638 fiz1 = _mm_add_pd(fiz1,tz);
640 fjx2 = _mm_add_pd(fjx2,tx);
641 fjy2 = _mm_add_pd(fjy2,ty);
642 fjz2 = _mm_add_pd(fjz2,tz);
646 /**************************
647 * CALCULATE INTERACTIONS *
648 **************************/
650 if (gmx_mm_any_lt(rsq20,rcutoff2))
653 r20 = _mm_mul_pd(rsq20,rinv20);
655 /* EWALD ELECTROSTATICS */
657 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
658 ewrt = _mm_mul_pd(r20,ewtabscale);
659 ewitab = _mm_cvttpd_epi32(ewrt);
660 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
661 ewitab = _mm_slli_epi32(ewitab,2);
662 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
663 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
664 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
665 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
666 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
667 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
668 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
669 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
670 velec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
671 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
673 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
675 /* Update potential sum for this i atom from the interaction with this j atom. */
676 velec = _mm_and_pd(velec,cutoff_mask);
677 velecsum = _mm_add_pd(velecsum,velec);
681 fscal = _mm_and_pd(fscal,cutoff_mask);
683 /* Calculate temporary vectorial force */
684 tx = _mm_mul_pd(fscal,dx20);
685 ty = _mm_mul_pd(fscal,dy20);
686 tz = _mm_mul_pd(fscal,dz20);
688 /* Update vectorial force */
689 fix2 = _mm_add_pd(fix2,tx);
690 fiy2 = _mm_add_pd(fiy2,ty);
691 fiz2 = _mm_add_pd(fiz2,tz);
693 fjx0 = _mm_add_pd(fjx0,tx);
694 fjy0 = _mm_add_pd(fjy0,ty);
695 fjz0 = _mm_add_pd(fjz0,tz);
699 /**************************
700 * CALCULATE INTERACTIONS *
701 **************************/
703 if (gmx_mm_any_lt(rsq21,rcutoff2))
706 r21 = _mm_mul_pd(rsq21,rinv21);
708 /* EWALD ELECTROSTATICS */
710 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
711 ewrt = _mm_mul_pd(r21,ewtabscale);
712 ewitab = _mm_cvttpd_epi32(ewrt);
713 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
714 ewitab = _mm_slli_epi32(ewitab,2);
715 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
716 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
717 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
718 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
719 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
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(qq21,_mm_sub_pd(_mm_sub_pd(rinv21,sh_ewald),velec));
724 felec = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
726 cutoff_mask = _mm_cmplt_pd(rsq21,rcutoff2);
728 /* Update potential sum for this i atom from the interaction with this j atom. */
729 velec = _mm_and_pd(velec,cutoff_mask);
730 velecsum = _mm_add_pd(velecsum,velec);
734 fscal = _mm_and_pd(fscal,cutoff_mask);
736 /* Calculate temporary vectorial force */
737 tx = _mm_mul_pd(fscal,dx21);
738 ty = _mm_mul_pd(fscal,dy21);
739 tz = _mm_mul_pd(fscal,dz21);
741 /* Update vectorial force */
742 fix2 = _mm_add_pd(fix2,tx);
743 fiy2 = _mm_add_pd(fiy2,ty);
744 fiz2 = _mm_add_pd(fiz2,tz);
746 fjx1 = _mm_add_pd(fjx1,tx);
747 fjy1 = _mm_add_pd(fjy1,ty);
748 fjz1 = _mm_add_pd(fjz1,tz);
752 /**************************
753 * CALCULATE INTERACTIONS *
754 **************************/
756 if (gmx_mm_any_lt(rsq22,rcutoff2))
759 r22 = _mm_mul_pd(rsq22,rinv22);
761 /* EWALD ELECTROSTATICS */
763 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
764 ewrt = _mm_mul_pd(r22,ewtabscale);
765 ewitab = _mm_cvttpd_epi32(ewrt);
766 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
767 ewitab = _mm_slli_epi32(ewitab,2);
768 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
769 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
770 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
771 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
772 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
773 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
774 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
775 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
776 velec = _mm_mul_pd(qq22,_mm_sub_pd(_mm_sub_pd(rinv22,sh_ewald),velec));
777 felec = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
779 cutoff_mask = _mm_cmplt_pd(rsq22,rcutoff2);
781 /* Update potential sum for this i atom from the interaction with this j atom. */
782 velec = _mm_and_pd(velec,cutoff_mask);
783 velecsum = _mm_add_pd(velecsum,velec);
787 fscal = _mm_and_pd(fscal,cutoff_mask);
789 /* Calculate temporary vectorial force */
790 tx = _mm_mul_pd(fscal,dx22);
791 ty = _mm_mul_pd(fscal,dy22);
792 tz = _mm_mul_pd(fscal,dz22);
794 /* Update vectorial force */
795 fix2 = _mm_add_pd(fix2,tx);
796 fiy2 = _mm_add_pd(fiy2,ty);
797 fiz2 = _mm_add_pd(fiz2,tz);
799 fjx2 = _mm_add_pd(fjx2,tx);
800 fjy2 = _mm_add_pd(fjy2,ty);
801 fjz2 = _mm_add_pd(fjz2,tz);
805 gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
807 /* Inner loop uses 449 flops */
814 j_coord_offsetA = DIM*jnrA;
816 /* load j atom coordinates */
817 gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
818 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
820 /* Calculate displacement vector */
821 dx00 = _mm_sub_pd(ix0,jx0);
822 dy00 = _mm_sub_pd(iy0,jy0);
823 dz00 = _mm_sub_pd(iz0,jz0);
824 dx01 = _mm_sub_pd(ix0,jx1);
825 dy01 = _mm_sub_pd(iy0,jy1);
826 dz01 = _mm_sub_pd(iz0,jz1);
827 dx02 = _mm_sub_pd(ix0,jx2);
828 dy02 = _mm_sub_pd(iy0,jy2);
829 dz02 = _mm_sub_pd(iz0,jz2);
830 dx10 = _mm_sub_pd(ix1,jx0);
831 dy10 = _mm_sub_pd(iy1,jy0);
832 dz10 = _mm_sub_pd(iz1,jz0);
833 dx11 = _mm_sub_pd(ix1,jx1);
834 dy11 = _mm_sub_pd(iy1,jy1);
835 dz11 = _mm_sub_pd(iz1,jz1);
836 dx12 = _mm_sub_pd(ix1,jx2);
837 dy12 = _mm_sub_pd(iy1,jy2);
838 dz12 = _mm_sub_pd(iz1,jz2);
839 dx20 = _mm_sub_pd(ix2,jx0);
840 dy20 = _mm_sub_pd(iy2,jy0);
841 dz20 = _mm_sub_pd(iz2,jz0);
842 dx21 = _mm_sub_pd(ix2,jx1);
843 dy21 = _mm_sub_pd(iy2,jy1);
844 dz21 = _mm_sub_pd(iz2,jz1);
845 dx22 = _mm_sub_pd(ix2,jx2);
846 dy22 = _mm_sub_pd(iy2,jy2);
847 dz22 = _mm_sub_pd(iz2,jz2);
849 /* Calculate squared distance and things based on it */
850 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
851 rsq01 = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
852 rsq02 = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
853 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
854 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
855 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
856 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
857 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
858 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
860 rinv00 = gmx_mm_invsqrt_pd(rsq00);
861 rinv01 = gmx_mm_invsqrt_pd(rsq01);
862 rinv02 = gmx_mm_invsqrt_pd(rsq02);
863 rinv10 = gmx_mm_invsqrt_pd(rsq10);
864 rinv11 = gmx_mm_invsqrt_pd(rsq11);
865 rinv12 = gmx_mm_invsqrt_pd(rsq12);
866 rinv20 = gmx_mm_invsqrt_pd(rsq20);
867 rinv21 = gmx_mm_invsqrt_pd(rsq21);
868 rinv22 = gmx_mm_invsqrt_pd(rsq22);
870 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
871 rinvsq01 = _mm_mul_pd(rinv01,rinv01);
872 rinvsq02 = _mm_mul_pd(rinv02,rinv02);
873 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
874 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
875 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
876 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
877 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
878 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
880 fjx0 = _mm_setzero_pd();
881 fjy0 = _mm_setzero_pd();
882 fjz0 = _mm_setzero_pd();
883 fjx1 = _mm_setzero_pd();
884 fjy1 = _mm_setzero_pd();
885 fjz1 = _mm_setzero_pd();
886 fjx2 = _mm_setzero_pd();
887 fjy2 = _mm_setzero_pd();
888 fjz2 = _mm_setzero_pd();
890 /**************************
891 * CALCULATE INTERACTIONS *
892 **************************/
894 if (gmx_mm_any_lt(rsq00,rcutoff2))
897 r00 = _mm_mul_pd(rsq00,rinv00);
899 /* EWALD ELECTROSTATICS */
901 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
902 ewrt = _mm_mul_pd(r00,ewtabscale);
903 ewitab = _mm_cvttpd_epi32(ewrt);
904 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
905 ewitab = _mm_slli_epi32(ewitab,2);
906 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
907 ewtabD = _mm_setzero_pd();
908 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
909 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
910 ewtabFn = _mm_setzero_pd();
911 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
912 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
913 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
914 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
915 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
917 /* Analytical LJ-PME */
918 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
919 ewcljrsq = _mm_mul_pd(ewclj2,rsq00);
920 ewclj6 = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
921 exponent = gmx_simd_exp_d(ewcljrsq);
922 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
923 poly = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
924 /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
925 vvdw6 = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
926 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
927 vvdw = _mm_sub_pd(_mm_mul_pd( _mm_sub_pd(vvdw12 , _mm_mul_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))),one_twelfth),
928 _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_add_pd(_mm_mul_pd(c6_00,sh_vdw_invrcut6),_mm_mul_pd(c6grid_00,sh_lj_ewald))),one_sixth));
929 /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
930 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,_mm_sub_pd(vvdw6,_mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6)))),rinvsq00);
932 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
934 /* Update potential sum for this i atom from the interaction with this j atom. */
935 velec = _mm_and_pd(velec,cutoff_mask);
936 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
937 velecsum = _mm_add_pd(velecsum,velec);
938 vvdw = _mm_and_pd(vvdw,cutoff_mask);
939 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
940 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
942 fscal = _mm_add_pd(felec,fvdw);
944 fscal = _mm_and_pd(fscal,cutoff_mask);
946 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
948 /* Calculate temporary vectorial force */
949 tx = _mm_mul_pd(fscal,dx00);
950 ty = _mm_mul_pd(fscal,dy00);
951 tz = _mm_mul_pd(fscal,dz00);
953 /* Update vectorial force */
954 fix0 = _mm_add_pd(fix0,tx);
955 fiy0 = _mm_add_pd(fiy0,ty);
956 fiz0 = _mm_add_pd(fiz0,tz);
958 fjx0 = _mm_add_pd(fjx0,tx);
959 fjy0 = _mm_add_pd(fjy0,ty);
960 fjz0 = _mm_add_pd(fjz0,tz);
964 /**************************
965 * CALCULATE INTERACTIONS *
966 **************************/
968 if (gmx_mm_any_lt(rsq01,rcutoff2))
971 r01 = _mm_mul_pd(rsq01,rinv01);
973 /* EWALD ELECTROSTATICS */
975 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
976 ewrt = _mm_mul_pd(r01,ewtabscale);
977 ewitab = _mm_cvttpd_epi32(ewrt);
978 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
979 ewitab = _mm_slli_epi32(ewitab,2);
980 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
981 ewtabD = _mm_setzero_pd();
982 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
983 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
984 ewtabFn = _mm_setzero_pd();
985 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
986 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
987 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
988 velec = _mm_mul_pd(qq01,_mm_sub_pd(_mm_sub_pd(rinv01,sh_ewald),velec));
989 felec = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
991 cutoff_mask = _mm_cmplt_pd(rsq01,rcutoff2);
993 /* Update potential sum for this i atom from the interaction with this j atom. */
994 velec = _mm_and_pd(velec,cutoff_mask);
995 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
996 velecsum = _mm_add_pd(velecsum,velec);
1000 fscal = _mm_and_pd(fscal,cutoff_mask);
1002 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1004 /* Calculate temporary vectorial force */
1005 tx = _mm_mul_pd(fscal,dx01);
1006 ty = _mm_mul_pd(fscal,dy01);
1007 tz = _mm_mul_pd(fscal,dz01);
1009 /* Update vectorial force */
1010 fix0 = _mm_add_pd(fix0,tx);
1011 fiy0 = _mm_add_pd(fiy0,ty);
1012 fiz0 = _mm_add_pd(fiz0,tz);
1014 fjx1 = _mm_add_pd(fjx1,tx);
1015 fjy1 = _mm_add_pd(fjy1,ty);
1016 fjz1 = _mm_add_pd(fjz1,tz);
1020 /**************************
1021 * CALCULATE INTERACTIONS *
1022 **************************/
1024 if (gmx_mm_any_lt(rsq02,rcutoff2))
1027 r02 = _mm_mul_pd(rsq02,rinv02);
1029 /* EWALD ELECTROSTATICS */
1031 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1032 ewrt = _mm_mul_pd(r02,ewtabscale);
1033 ewitab = _mm_cvttpd_epi32(ewrt);
1034 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1035 ewitab = _mm_slli_epi32(ewitab,2);
1036 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1037 ewtabD = _mm_setzero_pd();
1038 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1039 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1040 ewtabFn = _mm_setzero_pd();
1041 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1042 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1043 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1044 velec = _mm_mul_pd(qq02,_mm_sub_pd(_mm_sub_pd(rinv02,sh_ewald),velec));
1045 felec = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
1047 cutoff_mask = _mm_cmplt_pd(rsq02,rcutoff2);
1049 /* Update potential sum for this i atom from the interaction with this j atom. */
1050 velec = _mm_and_pd(velec,cutoff_mask);
1051 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1052 velecsum = _mm_add_pd(velecsum,velec);
1056 fscal = _mm_and_pd(fscal,cutoff_mask);
1058 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1060 /* Calculate temporary vectorial force */
1061 tx = _mm_mul_pd(fscal,dx02);
1062 ty = _mm_mul_pd(fscal,dy02);
1063 tz = _mm_mul_pd(fscal,dz02);
1065 /* Update vectorial force */
1066 fix0 = _mm_add_pd(fix0,tx);
1067 fiy0 = _mm_add_pd(fiy0,ty);
1068 fiz0 = _mm_add_pd(fiz0,tz);
1070 fjx2 = _mm_add_pd(fjx2,tx);
1071 fjy2 = _mm_add_pd(fjy2,ty);
1072 fjz2 = _mm_add_pd(fjz2,tz);
1076 /**************************
1077 * CALCULATE INTERACTIONS *
1078 **************************/
1080 if (gmx_mm_any_lt(rsq10,rcutoff2))
1083 r10 = _mm_mul_pd(rsq10,rinv10);
1085 /* EWALD ELECTROSTATICS */
1087 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1088 ewrt = _mm_mul_pd(r10,ewtabscale);
1089 ewitab = _mm_cvttpd_epi32(ewrt);
1090 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1091 ewitab = _mm_slli_epi32(ewitab,2);
1092 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1093 ewtabD = _mm_setzero_pd();
1094 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1095 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1096 ewtabFn = _mm_setzero_pd();
1097 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1098 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1099 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1100 velec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
1101 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1103 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
1105 /* Update potential sum for this i atom from the interaction with this j atom. */
1106 velec = _mm_and_pd(velec,cutoff_mask);
1107 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1108 velecsum = _mm_add_pd(velecsum,velec);
1112 fscal = _mm_and_pd(fscal,cutoff_mask);
1114 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1116 /* Calculate temporary vectorial force */
1117 tx = _mm_mul_pd(fscal,dx10);
1118 ty = _mm_mul_pd(fscal,dy10);
1119 tz = _mm_mul_pd(fscal,dz10);
1121 /* Update vectorial force */
1122 fix1 = _mm_add_pd(fix1,tx);
1123 fiy1 = _mm_add_pd(fiy1,ty);
1124 fiz1 = _mm_add_pd(fiz1,tz);
1126 fjx0 = _mm_add_pd(fjx0,tx);
1127 fjy0 = _mm_add_pd(fjy0,ty);
1128 fjz0 = _mm_add_pd(fjz0,tz);
1132 /**************************
1133 * CALCULATE INTERACTIONS *
1134 **************************/
1136 if (gmx_mm_any_lt(rsq11,rcutoff2))
1139 r11 = _mm_mul_pd(rsq11,rinv11);
1141 /* EWALD ELECTROSTATICS */
1143 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1144 ewrt = _mm_mul_pd(r11,ewtabscale);
1145 ewitab = _mm_cvttpd_epi32(ewrt);
1146 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1147 ewitab = _mm_slli_epi32(ewitab,2);
1148 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1149 ewtabD = _mm_setzero_pd();
1150 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1151 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1152 ewtabFn = _mm_setzero_pd();
1153 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1154 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1155 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1156 velec = _mm_mul_pd(qq11,_mm_sub_pd(_mm_sub_pd(rinv11,sh_ewald),velec));
1157 felec = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1159 cutoff_mask = _mm_cmplt_pd(rsq11,rcutoff2);
1161 /* Update potential sum for this i atom from the interaction with this j atom. */
1162 velec = _mm_and_pd(velec,cutoff_mask);
1163 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1164 velecsum = _mm_add_pd(velecsum,velec);
1168 fscal = _mm_and_pd(fscal,cutoff_mask);
1170 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1172 /* Calculate temporary vectorial force */
1173 tx = _mm_mul_pd(fscal,dx11);
1174 ty = _mm_mul_pd(fscal,dy11);
1175 tz = _mm_mul_pd(fscal,dz11);
1177 /* Update vectorial force */
1178 fix1 = _mm_add_pd(fix1,tx);
1179 fiy1 = _mm_add_pd(fiy1,ty);
1180 fiz1 = _mm_add_pd(fiz1,tz);
1182 fjx1 = _mm_add_pd(fjx1,tx);
1183 fjy1 = _mm_add_pd(fjy1,ty);
1184 fjz1 = _mm_add_pd(fjz1,tz);
1188 /**************************
1189 * CALCULATE INTERACTIONS *
1190 **************************/
1192 if (gmx_mm_any_lt(rsq12,rcutoff2))
1195 r12 = _mm_mul_pd(rsq12,rinv12);
1197 /* EWALD ELECTROSTATICS */
1199 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1200 ewrt = _mm_mul_pd(r12,ewtabscale);
1201 ewitab = _mm_cvttpd_epi32(ewrt);
1202 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1203 ewitab = _mm_slli_epi32(ewitab,2);
1204 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1205 ewtabD = _mm_setzero_pd();
1206 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1207 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1208 ewtabFn = _mm_setzero_pd();
1209 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1210 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1211 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1212 velec = _mm_mul_pd(qq12,_mm_sub_pd(_mm_sub_pd(rinv12,sh_ewald),velec));
1213 felec = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1215 cutoff_mask = _mm_cmplt_pd(rsq12,rcutoff2);
1217 /* Update potential sum for this i atom from the interaction with this j atom. */
1218 velec = _mm_and_pd(velec,cutoff_mask);
1219 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1220 velecsum = _mm_add_pd(velecsum,velec);
1224 fscal = _mm_and_pd(fscal,cutoff_mask);
1226 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1228 /* Calculate temporary vectorial force */
1229 tx = _mm_mul_pd(fscal,dx12);
1230 ty = _mm_mul_pd(fscal,dy12);
1231 tz = _mm_mul_pd(fscal,dz12);
1233 /* Update vectorial force */
1234 fix1 = _mm_add_pd(fix1,tx);
1235 fiy1 = _mm_add_pd(fiy1,ty);
1236 fiz1 = _mm_add_pd(fiz1,tz);
1238 fjx2 = _mm_add_pd(fjx2,tx);
1239 fjy2 = _mm_add_pd(fjy2,ty);
1240 fjz2 = _mm_add_pd(fjz2,tz);
1244 /**************************
1245 * CALCULATE INTERACTIONS *
1246 **************************/
1248 if (gmx_mm_any_lt(rsq20,rcutoff2))
1251 r20 = _mm_mul_pd(rsq20,rinv20);
1253 /* EWALD ELECTROSTATICS */
1255 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1256 ewrt = _mm_mul_pd(r20,ewtabscale);
1257 ewitab = _mm_cvttpd_epi32(ewrt);
1258 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1259 ewitab = _mm_slli_epi32(ewitab,2);
1260 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1261 ewtabD = _mm_setzero_pd();
1262 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1263 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1264 ewtabFn = _mm_setzero_pd();
1265 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1266 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1267 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1268 velec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
1269 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1271 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
1273 /* Update potential sum for this i atom from the interaction with this j atom. */
1274 velec = _mm_and_pd(velec,cutoff_mask);
1275 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1276 velecsum = _mm_add_pd(velecsum,velec);
1280 fscal = _mm_and_pd(fscal,cutoff_mask);
1282 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1284 /* Calculate temporary vectorial force */
1285 tx = _mm_mul_pd(fscal,dx20);
1286 ty = _mm_mul_pd(fscal,dy20);
1287 tz = _mm_mul_pd(fscal,dz20);
1289 /* Update vectorial force */
1290 fix2 = _mm_add_pd(fix2,tx);
1291 fiy2 = _mm_add_pd(fiy2,ty);
1292 fiz2 = _mm_add_pd(fiz2,tz);
1294 fjx0 = _mm_add_pd(fjx0,tx);
1295 fjy0 = _mm_add_pd(fjy0,ty);
1296 fjz0 = _mm_add_pd(fjz0,tz);
1300 /**************************
1301 * CALCULATE INTERACTIONS *
1302 **************************/
1304 if (gmx_mm_any_lt(rsq21,rcutoff2))
1307 r21 = _mm_mul_pd(rsq21,rinv21);
1309 /* EWALD ELECTROSTATICS */
1311 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1312 ewrt = _mm_mul_pd(r21,ewtabscale);
1313 ewitab = _mm_cvttpd_epi32(ewrt);
1314 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1315 ewitab = _mm_slli_epi32(ewitab,2);
1316 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1317 ewtabD = _mm_setzero_pd();
1318 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1319 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1320 ewtabFn = _mm_setzero_pd();
1321 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1322 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1323 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1324 velec = _mm_mul_pd(qq21,_mm_sub_pd(_mm_sub_pd(rinv21,sh_ewald),velec));
1325 felec = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1327 cutoff_mask = _mm_cmplt_pd(rsq21,rcutoff2);
1329 /* Update potential sum for this i atom from the interaction with this j atom. */
1330 velec = _mm_and_pd(velec,cutoff_mask);
1331 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1332 velecsum = _mm_add_pd(velecsum,velec);
1336 fscal = _mm_and_pd(fscal,cutoff_mask);
1338 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1340 /* Calculate temporary vectorial force */
1341 tx = _mm_mul_pd(fscal,dx21);
1342 ty = _mm_mul_pd(fscal,dy21);
1343 tz = _mm_mul_pd(fscal,dz21);
1345 /* Update vectorial force */
1346 fix2 = _mm_add_pd(fix2,tx);
1347 fiy2 = _mm_add_pd(fiy2,ty);
1348 fiz2 = _mm_add_pd(fiz2,tz);
1350 fjx1 = _mm_add_pd(fjx1,tx);
1351 fjy1 = _mm_add_pd(fjy1,ty);
1352 fjz1 = _mm_add_pd(fjz1,tz);
1356 /**************************
1357 * CALCULATE INTERACTIONS *
1358 **************************/
1360 if (gmx_mm_any_lt(rsq22,rcutoff2))
1363 r22 = _mm_mul_pd(rsq22,rinv22);
1365 /* EWALD ELECTROSTATICS */
1367 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1368 ewrt = _mm_mul_pd(r22,ewtabscale);
1369 ewitab = _mm_cvttpd_epi32(ewrt);
1370 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1371 ewitab = _mm_slli_epi32(ewitab,2);
1372 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1373 ewtabD = _mm_setzero_pd();
1374 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1375 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1376 ewtabFn = _mm_setzero_pd();
1377 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1378 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1379 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1380 velec = _mm_mul_pd(qq22,_mm_sub_pd(_mm_sub_pd(rinv22,sh_ewald),velec));
1381 felec = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1383 cutoff_mask = _mm_cmplt_pd(rsq22,rcutoff2);
1385 /* Update potential sum for this i atom from the interaction with this j atom. */
1386 velec = _mm_and_pd(velec,cutoff_mask);
1387 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1388 velecsum = _mm_add_pd(velecsum,velec);
1392 fscal = _mm_and_pd(fscal,cutoff_mask);
1394 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1396 /* Calculate temporary vectorial force */
1397 tx = _mm_mul_pd(fscal,dx22);
1398 ty = _mm_mul_pd(fscal,dy22);
1399 tz = _mm_mul_pd(fscal,dz22);
1401 /* Update vectorial force */
1402 fix2 = _mm_add_pd(fix2,tx);
1403 fiy2 = _mm_add_pd(fiy2,ty);
1404 fiz2 = _mm_add_pd(fiz2,tz);
1406 fjx2 = _mm_add_pd(fjx2,tx);
1407 fjy2 = _mm_add_pd(fjy2,ty);
1408 fjz2 = _mm_add_pd(fjz2,tz);
1412 gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1414 /* Inner loop uses 449 flops */
1417 /* End of innermost loop */
1419 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1420 f+i_coord_offset,fshift+i_shift_offset);
1423 /* Update potential energies */
1424 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1425 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1427 /* Increment number of inner iterations */
1428 inneriter += j_index_end - j_index_start;
1430 /* Outer loop uses 20 flops */
1433 /* Increment number of outer iterations */
1436 /* Update outer/inner flops */
1438 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*449);
1441 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3W3_F_sse4_1_double
1442 * Electrostatics interaction: Ewald
1443 * VdW interaction: LJEwald
1444 * Geometry: Water3-Water3
1445 * Calculate force/pot: Force
1448 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3W3_F_sse4_1_double
1449 (t_nblist * gmx_restrict nlist,
1450 rvec * gmx_restrict xx,
1451 rvec * gmx_restrict ff,
1452 t_forcerec * gmx_restrict fr,
1453 t_mdatoms * gmx_restrict mdatoms,
1454 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1455 t_nrnb * gmx_restrict nrnb)
1457 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1458 * just 0 for non-waters.
1459 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
1460 * jnr indices corresponding to data put in the four positions in the SIMD register.
1462 int i_shift_offset,i_coord_offset,outeriter,inneriter;
1463 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1465 int j_coord_offsetA,j_coord_offsetB;
1466 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
1467 real rcutoff_scalar;
1468 real *shiftvec,*fshift,*x,*f;
1469 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1471 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1473 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1475 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1476 int vdwjidx0A,vdwjidx0B;
1477 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1478 int vdwjidx1A,vdwjidx1B;
1479 __m128d jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1480 int vdwjidx2A,vdwjidx2B;
1481 __m128d jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1482 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1483 __m128d dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1484 __m128d dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1485 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1486 __m128d dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1487 __m128d dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1488 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1489 __m128d dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1490 __m128d dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1491 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
1494 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1497 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
1498 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
1508 __m128d ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
1510 __m128d one_half = _mm_set1_pd(0.5);
1511 __m128d minus_one = _mm_set1_pd(-1.0);
1513 __m128d ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1515 __m128d dummy_mask,cutoff_mask;
1516 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
1517 __m128d one = _mm_set1_pd(1.0);
1518 __m128d two = _mm_set1_pd(2.0);
1524 jindex = nlist->jindex;
1526 shiftidx = nlist->shift;
1528 shiftvec = fr->shift_vec[0];
1529 fshift = fr->fshift[0];
1530 facel = _mm_set1_pd(fr->epsfac);
1531 charge = mdatoms->chargeA;
1532 nvdwtype = fr->ntype;
1533 vdwparam = fr->nbfp;
1534 vdwtype = mdatoms->typeA;
1535 vdwgridparam = fr->ljpme_c6grid;
1536 sh_lj_ewald = _mm_set1_pd(fr->ic->sh_lj_ewald);
1537 ewclj = _mm_set1_pd(fr->ewaldcoeff_lj);
1538 ewclj2 = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
1540 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
1541 ewtab = fr->ic->tabq_coul_F;
1542 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
1543 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
1545 /* Setup water-specific parameters */
1546 inr = nlist->iinr[0];
1547 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
1548 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
1549 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
1550 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
1552 jq0 = _mm_set1_pd(charge[inr+0]);
1553 jq1 = _mm_set1_pd(charge[inr+1]);
1554 jq2 = _mm_set1_pd(charge[inr+2]);
1555 vdwjidx0A = 2*vdwtype[inr+0];
1556 qq00 = _mm_mul_pd(iq0,jq0);
1557 c6_00 = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
1558 c12_00 = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
1559 c6grid_00 = _mm_set1_pd(vdwgridparam[vdwioffset0+vdwjidx0A]);
1560 qq01 = _mm_mul_pd(iq0,jq1);
1561 qq02 = _mm_mul_pd(iq0,jq2);
1562 qq10 = _mm_mul_pd(iq1,jq0);
1563 qq11 = _mm_mul_pd(iq1,jq1);
1564 qq12 = _mm_mul_pd(iq1,jq2);
1565 qq20 = _mm_mul_pd(iq2,jq0);
1566 qq21 = _mm_mul_pd(iq2,jq1);
1567 qq22 = _mm_mul_pd(iq2,jq2);
1569 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1570 rcutoff_scalar = fr->rcoulomb;
1571 rcutoff = _mm_set1_pd(rcutoff_scalar);
1572 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
1574 sh_vdw_invrcut6 = _mm_set1_pd(fr->ic->sh_invrc6);
1575 rvdw = _mm_set1_pd(fr->rvdw);
1577 /* Avoid stupid compiler warnings */
1579 j_coord_offsetA = 0;
1580 j_coord_offsetB = 0;
1585 /* Start outer loop over neighborlists */
1586 for(iidx=0; iidx<nri; iidx++)
1588 /* Load shift vector for this list */
1589 i_shift_offset = DIM*shiftidx[iidx];
1591 /* Load limits for loop over neighbors */
1592 j_index_start = jindex[iidx];
1593 j_index_end = jindex[iidx+1];
1595 /* Get outer coordinate index */
1597 i_coord_offset = DIM*inr;
1599 /* Load i particle coords and add shift vector */
1600 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1601 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1603 fix0 = _mm_setzero_pd();
1604 fiy0 = _mm_setzero_pd();
1605 fiz0 = _mm_setzero_pd();
1606 fix1 = _mm_setzero_pd();
1607 fiy1 = _mm_setzero_pd();
1608 fiz1 = _mm_setzero_pd();
1609 fix2 = _mm_setzero_pd();
1610 fiy2 = _mm_setzero_pd();
1611 fiz2 = _mm_setzero_pd();
1613 /* Start inner kernel loop */
1614 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
1617 /* Get j neighbor index, and coordinate index */
1619 jnrB = jjnr[jidx+1];
1620 j_coord_offsetA = DIM*jnrA;
1621 j_coord_offsetB = DIM*jnrB;
1623 /* load j atom coordinates */
1624 gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1625 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1627 /* Calculate displacement vector */
1628 dx00 = _mm_sub_pd(ix0,jx0);
1629 dy00 = _mm_sub_pd(iy0,jy0);
1630 dz00 = _mm_sub_pd(iz0,jz0);
1631 dx01 = _mm_sub_pd(ix0,jx1);
1632 dy01 = _mm_sub_pd(iy0,jy1);
1633 dz01 = _mm_sub_pd(iz0,jz1);
1634 dx02 = _mm_sub_pd(ix0,jx2);
1635 dy02 = _mm_sub_pd(iy0,jy2);
1636 dz02 = _mm_sub_pd(iz0,jz2);
1637 dx10 = _mm_sub_pd(ix1,jx0);
1638 dy10 = _mm_sub_pd(iy1,jy0);
1639 dz10 = _mm_sub_pd(iz1,jz0);
1640 dx11 = _mm_sub_pd(ix1,jx1);
1641 dy11 = _mm_sub_pd(iy1,jy1);
1642 dz11 = _mm_sub_pd(iz1,jz1);
1643 dx12 = _mm_sub_pd(ix1,jx2);
1644 dy12 = _mm_sub_pd(iy1,jy2);
1645 dz12 = _mm_sub_pd(iz1,jz2);
1646 dx20 = _mm_sub_pd(ix2,jx0);
1647 dy20 = _mm_sub_pd(iy2,jy0);
1648 dz20 = _mm_sub_pd(iz2,jz0);
1649 dx21 = _mm_sub_pd(ix2,jx1);
1650 dy21 = _mm_sub_pd(iy2,jy1);
1651 dz21 = _mm_sub_pd(iz2,jz1);
1652 dx22 = _mm_sub_pd(ix2,jx2);
1653 dy22 = _mm_sub_pd(iy2,jy2);
1654 dz22 = _mm_sub_pd(iz2,jz2);
1656 /* Calculate squared distance and things based on it */
1657 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1658 rsq01 = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
1659 rsq02 = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
1660 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1661 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1662 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1663 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1664 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1665 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1667 rinv00 = gmx_mm_invsqrt_pd(rsq00);
1668 rinv01 = gmx_mm_invsqrt_pd(rsq01);
1669 rinv02 = gmx_mm_invsqrt_pd(rsq02);
1670 rinv10 = gmx_mm_invsqrt_pd(rsq10);
1671 rinv11 = gmx_mm_invsqrt_pd(rsq11);
1672 rinv12 = gmx_mm_invsqrt_pd(rsq12);
1673 rinv20 = gmx_mm_invsqrt_pd(rsq20);
1674 rinv21 = gmx_mm_invsqrt_pd(rsq21);
1675 rinv22 = gmx_mm_invsqrt_pd(rsq22);
1677 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
1678 rinvsq01 = _mm_mul_pd(rinv01,rinv01);
1679 rinvsq02 = _mm_mul_pd(rinv02,rinv02);
1680 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
1681 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
1682 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
1683 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
1684 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
1685 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
1687 fjx0 = _mm_setzero_pd();
1688 fjy0 = _mm_setzero_pd();
1689 fjz0 = _mm_setzero_pd();
1690 fjx1 = _mm_setzero_pd();
1691 fjy1 = _mm_setzero_pd();
1692 fjz1 = _mm_setzero_pd();
1693 fjx2 = _mm_setzero_pd();
1694 fjy2 = _mm_setzero_pd();
1695 fjz2 = _mm_setzero_pd();
1697 /**************************
1698 * CALCULATE INTERACTIONS *
1699 **************************/
1701 if (gmx_mm_any_lt(rsq00,rcutoff2))
1704 r00 = _mm_mul_pd(rsq00,rinv00);
1706 /* EWALD ELECTROSTATICS */
1708 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1709 ewrt = _mm_mul_pd(r00,ewtabscale);
1710 ewitab = _mm_cvttpd_epi32(ewrt);
1711 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1712 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1714 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1715 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1717 /* Analytical LJ-PME */
1718 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1719 ewcljrsq = _mm_mul_pd(ewclj2,rsq00);
1720 ewclj6 = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1721 exponent = gmx_simd_exp_d(ewcljrsq);
1722 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1723 poly = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1724 /* f6A = 6 * C6grid * (1 - poly) */
1725 f6A = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1726 /* f6B = C6grid * exponent * beta^6 */
1727 f6B = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1728 /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1729 fvdw = _mm_mul_pd(_mm_add_pd(_mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),_mm_sub_pd(c6_00,f6A)),rinvsix),f6B),rinvsq00);
1731 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
1733 fscal = _mm_add_pd(felec,fvdw);
1735 fscal = _mm_and_pd(fscal,cutoff_mask);
1737 /* Calculate temporary vectorial force */
1738 tx = _mm_mul_pd(fscal,dx00);
1739 ty = _mm_mul_pd(fscal,dy00);
1740 tz = _mm_mul_pd(fscal,dz00);
1742 /* Update vectorial force */
1743 fix0 = _mm_add_pd(fix0,tx);
1744 fiy0 = _mm_add_pd(fiy0,ty);
1745 fiz0 = _mm_add_pd(fiz0,tz);
1747 fjx0 = _mm_add_pd(fjx0,tx);
1748 fjy0 = _mm_add_pd(fjy0,ty);
1749 fjz0 = _mm_add_pd(fjz0,tz);
1753 /**************************
1754 * CALCULATE INTERACTIONS *
1755 **************************/
1757 if (gmx_mm_any_lt(rsq01,rcutoff2))
1760 r01 = _mm_mul_pd(rsq01,rinv01);
1762 /* EWALD ELECTROSTATICS */
1764 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1765 ewrt = _mm_mul_pd(r01,ewtabscale);
1766 ewitab = _mm_cvttpd_epi32(ewrt);
1767 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1768 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1770 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1771 felec = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
1773 cutoff_mask = _mm_cmplt_pd(rsq01,rcutoff2);
1777 fscal = _mm_and_pd(fscal,cutoff_mask);
1779 /* Calculate temporary vectorial force */
1780 tx = _mm_mul_pd(fscal,dx01);
1781 ty = _mm_mul_pd(fscal,dy01);
1782 tz = _mm_mul_pd(fscal,dz01);
1784 /* Update vectorial force */
1785 fix0 = _mm_add_pd(fix0,tx);
1786 fiy0 = _mm_add_pd(fiy0,ty);
1787 fiz0 = _mm_add_pd(fiz0,tz);
1789 fjx1 = _mm_add_pd(fjx1,tx);
1790 fjy1 = _mm_add_pd(fjy1,ty);
1791 fjz1 = _mm_add_pd(fjz1,tz);
1795 /**************************
1796 * CALCULATE INTERACTIONS *
1797 **************************/
1799 if (gmx_mm_any_lt(rsq02,rcutoff2))
1802 r02 = _mm_mul_pd(rsq02,rinv02);
1804 /* EWALD ELECTROSTATICS */
1806 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1807 ewrt = _mm_mul_pd(r02,ewtabscale);
1808 ewitab = _mm_cvttpd_epi32(ewrt);
1809 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1810 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1812 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1813 felec = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
1815 cutoff_mask = _mm_cmplt_pd(rsq02,rcutoff2);
1819 fscal = _mm_and_pd(fscal,cutoff_mask);
1821 /* Calculate temporary vectorial force */
1822 tx = _mm_mul_pd(fscal,dx02);
1823 ty = _mm_mul_pd(fscal,dy02);
1824 tz = _mm_mul_pd(fscal,dz02);
1826 /* Update vectorial force */
1827 fix0 = _mm_add_pd(fix0,tx);
1828 fiy0 = _mm_add_pd(fiy0,ty);
1829 fiz0 = _mm_add_pd(fiz0,tz);
1831 fjx2 = _mm_add_pd(fjx2,tx);
1832 fjy2 = _mm_add_pd(fjy2,ty);
1833 fjz2 = _mm_add_pd(fjz2,tz);
1837 /**************************
1838 * CALCULATE INTERACTIONS *
1839 **************************/
1841 if (gmx_mm_any_lt(rsq10,rcutoff2))
1844 r10 = _mm_mul_pd(rsq10,rinv10);
1846 /* EWALD ELECTROSTATICS */
1848 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1849 ewrt = _mm_mul_pd(r10,ewtabscale);
1850 ewitab = _mm_cvttpd_epi32(ewrt);
1851 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1852 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1854 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1855 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1857 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
1861 fscal = _mm_and_pd(fscal,cutoff_mask);
1863 /* Calculate temporary vectorial force */
1864 tx = _mm_mul_pd(fscal,dx10);
1865 ty = _mm_mul_pd(fscal,dy10);
1866 tz = _mm_mul_pd(fscal,dz10);
1868 /* Update vectorial force */
1869 fix1 = _mm_add_pd(fix1,tx);
1870 fiy1 = _mm_add_pd(fiy1,ty);
1871 fiz1 = _mm_add_pd(fiz1,tz);
1873 fjx0 = _mm_add_pd(fjx0,tx);
1874 fjy0 = _mm_add_pd(fjy0,ty);
1875 fjz0 = _mm_add_pd(fjz0,tz);
1879 /**************************
1880 * CALCULATE INTERACTIONS *
1881 **************************/
1883 if (gmx_mm_any_lt(rsq11,rcutoff2))
1886 r11 = _mm_mul_pd(rsq11,rinv11);
1888 /* EWALD ELECTROSTATICS */
1890 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1891 ewrt = _mm_mul_pd(r11,ewtabscale);
1892 ewitab = _mm_cvttpd_epi32(ewrt);
1893 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1894 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1896 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1897 felec = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1899 cutoff_mask = _mm_cmplt_pd(rsq11,rcutoff2);
1903 fscal = _mm_and_pd(fscal,cutoff_mask);
1905 /* Calculate temporary vectorial force */
1906 tx = _mm_mul_pd(fscal,dx11);
1907 ty = _mm_mul_pd(fscal,dy11);
1908 tz = _mm_mul_pd(fscal,dz11);
1910 /* Update vectorial force */
1911 fix1 = _mm_add_pd(fix1,tx);
1912 fiy1 = _mm_add_pd(fiy1,ty);
1913 fiz1 = _mm_add_pd(fiz1,tz);
1915 fjx1 = _mm_add_pd(fjx1,tx);
1916 fjy1 = _mm_add_pd(fjy1,ty);
1917 fjz1 = _mm_add_pd(fjz1,tz);
1921 /**************************
1922 * CALCULATE INTERACTIONS *
1923 **************************/
1925 if (gmx_mm_any_lt(rsq12,rcutoff2))
1928 r12 = _mm_mul_pd(rsq12,rinv12);
1930 /* EWALD ELECTROSTATICS */
1932 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1933 ewrt = _mm_mul_pd(r12,ewtabscale);
1934 ewitab = _mm_cvttpd_epi32(ewrt);
1935 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1936 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1938 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1939 felec = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1941 cutoff_mask = _mm_cmplt_pd(rsq12,rcutoff2);
1945 fscal = _mm_and_pd(fscal,cutoff_mask);
1947 /* Calculate temporary vectorial force */
1948 tx = _mm_mul_pd(fscal,dx12);
1949 ty = _mm_mul_pd(fscal,dy12);
1950 tz = _mm_mul_pd(fscal,dz12);
1952 /* Update vectorial force */
1953 fix1 = _mm_add_pd(fix1,tx);
1954 fiy1 = _mm_add_pd(fiy1,ty);
1955 fiz1 = _mm_add_pd(fiz1,tz);
1957 fjx2 = _mm_add_pd(fjx2,tx);
1958 fjy2 = _mm_add_pd(fjy2,ty);
1959 fjz2 = _mm_add_pd(fjz2,tz);
1963 /**************************
1964 * CALCULATE INTERACTIONS *
1965 **************************/
1967 if (gmx_mm_any_lt(rsq20,rcutoff2))
1970 r20 = _mm_mul_pd(rsq20,rinv20);
1972 /* EWALD ELECTROSTATICS */
1974 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1975 ewrt = _mm_mul_pd(r20,ewtabscale);
1976 ewitab = _mm_cvttpd_epi32(ewrt);
1977 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1978 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1980 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1981 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1983 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
1987 fscal = _mm_and_pd(fscal,cutoff_mask);
1989 /* Calculate temporary vectorial force */
1990 tx = _mm_mul_pd(fscal,dx20);
1991 ty = _mm_mul_pd(fscal,dy20);
1992 tz = _mm_mul_pd(fscal,dz20);
1994 /* Update vectorial force */
1995 fix2 = _mm_add_pd(fix2,tx);
1996 fiy2 = _mm_add_pd(fiy2,ty);
1997 fiz2 = _mm_add_pd(fiz2,tz);
1999 fjx0 = _mm_add_pd(fjx0,tx);
2000 fjy0 = _mm_add_pd(fjy0,ty);
2001 fjz0 = _mm_add_pd(fjz0,tz);
2005 /**************************
2006 * CALCULATE INTERACTIONS *
2007 **************************/
2009 if (gmx_mm_any_lt(rsq21,rcutoff2))
2012 r21 = _mm_mul_pd(rsq21,rinv21);
2014 /* EWALD ELECTROSTATICS */
2016 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2017 ewrt = _mm_mul_pd(r21,ewtabscale);
2018 ewitab = _mm_cvttpd_epi32(ewrt);
2019 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2020 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
2022 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
2023 felec = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
2025 cutoff_mask = _mm_cmplt_pd(rsq21,rcutoff2);
2029 fscal = _mm_and_pd(fscal,cutoff_mask);
2031 /* Calculate temporary vectorial force */
2032 tx = _mm_mul_pd(fscal,dx21);
2033 ty = _mm_mul_pd(fscal,dy21);
2034 tz = _mm_mul_pd(fscal,dz21);
2036 /* Update vectorial force */
2037 fix2 = _mm_add_pd(fix2,tx);
2038 fiy2 = _mm_add_pd(fiy2,ty);
2039 fiz2 = _mm_add_pd(fiz2,tz);
2041 fjx1 = _mm_add_pd(fjx1,tx);
2042 fjy1 = _mm_add_pd(fjy1,ty);
2043 fjz1 = _mm_add_pd(fjz1,tz);
2047 /**************************
2048 * CALCULATE INTERACTIONS *
2049 **************************/
2051 if (gmx_mm_any_lt(rsq22,rcutoff2))
2054 r22 = _mm_mul_pd(rsq22,rinv22);
2056 /* EWALD ELECTROSTATICS */
2058 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2059 ewrt = _mm_mul_pd(r22,ewtabscale);
2060 ewitab = _mm_cvttpd_epi32(ewrt);
2061 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2062 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
2064 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
2065 felec = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
2067 cutoff_mask = _mm_cmplt_pd(rsq22,rcutoff2);
2071 fscal = _mm_and_pd(fscal,cutoff_mask);
2073 /* Calculate temporary vectorial force */
2074 tx = _mm_mul_pd(fscal,dx22);
2075 ty = _mm_mul_pd(fscal,dy22);
2076 tz = _mm_mul_pd(fscal,dz22);
2078 /* Update vectorial force */
2079 fix2 = _mm_add_pd(fix2,tx);
2080 fiy2 = _mm_add_pd(fiy2,ty);
2081 fiz2 = _mm_add_pd(fiz2,tz);
2083 fjx2 = _mm_add_pd(fjx2,tx);
2084 fjy2 = _mm_add_pd(fjy2,ty);
2085 fjz2 = _mm_add_pd(fjz2,tz);
2089 gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2091 /* Inner loop uses 374 flops */
2094 if(jidx<j_index_end)
2098 j_coord_offsetA = DIM*jnrA;
2100 /* load j atom coordinates */
2101 gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
2102 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
2104 /* Calculate displacement vector */
2105 dx00 = _mm_sub_pd(ix0,jx0);
2106 dy00 = _mm_sub_pd(iy0,jy0);
2107 dz00 = _mm_sub_pd(iz0,jz0);
2108 dx01 = _mm_sub_pd(ix0,jx1);
2109 dy01 = _mm_sub_pd(iy0,jy1);
2110 dz01 = _mm_sub_pd(iz0,jz1);
2111 dx02 = _mm_sub_pd(ix0,jx2);
2112 dy02 = _mm_sub_pd(iy0,jy2);
2113 dz02 = _mm_sub_pd(iz0,jz2);
2114 dx10 = _mm_sub_pd(ix1,jx0);
2115 dy10 = _mm_sub_pd(iy1,jy0);
2116 dz10 = _mm_sub_pd(iz1,jz0);
2117 dx11 = _mm_sub_pd(ix1,jx1);
2118 dy11 = _mm_sub_pd(iy1,jy1);
2119 dz11 = _mm_sub_pd(iz1,jz1);
2120 dx12 = _mm_sub_pd(ix1,jx2);
2121 dy12 = _mm_sub_pd(iy1,jy2);
2122 dz12 = _mm_sub_pd(iz1,jz2);
2123 dx20 = _mm_sub_pd(ix2,jx0);
2124 dy20 = _mm_sub_pd(iy2,jy0);
2125 dz20 = _mm_sub_pd(iz2,jz0);
2126 dx21 = _mm_sub_pd(ix2,jx1);
2127 dy21 = _mm_sub_pd(iy2,jy1);
2128 dz21 = _mm_sub_pd(iz2,jz1);
2129 dx22 = _mm_sub_pd(ix2,jx2);
2130 dy22 = _mm_sub_pd(iy2,jy2);
2131 dz22 = _mm_sub_pd(iz2,jz2);
2133 /* Calculate squared distance and things based on it */
2134 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
2135 rsq01 = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
2136 rsq02 = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
2137 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
2138 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
2139 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
2140 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
2141 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
2142 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
2144 rinv00 = gmx_mm_invsqrt_pd(rsq00);
2145 rinv01 = gmx_mm_invsqrt_pd(rsq01);
2146 rinv02 = gmx_mm_invsqrt_pd(rsq02);
2147 rinv10 = gmx_mm_invsqrt_pd(rsq10);
2148 rinv11 = gmx_mm_invsqrt_pd(rsq11);
2149 rinv12 = gmx_mm_invsqrt_pd(rsq12);
2150 rinv20 = gmx_mm_invsqrt_pd(rsq20);
2151 rinv21 = gmx_mm_invsqrt_pd(rsq21);
2152 rinv22 = gmx_mm_invsqrt_pd(rsq22);
2154 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
2155 rinvsq01 = _mm_mul_pd(rinv01,rinv01);
2156 rinvsq02 = _mm_mul_pd(rinv02,rinv02);
2157 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
2158 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
2159 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
2160 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
2161 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
2162 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
2164 fjx0 = _mm_setzero_pd();
2165 fjy0 = _mm_setzero_pd();
2166 fjz0 = _mm_setzero_pd();
2167 fjx1 = _mm_setzero_pd();
2168 fjy1 = _mm_setzero_pd();
2169 fjz1 = _mm_setzero_pd();
2170 fjx2 = _mm_setzero_pd();
2171 fjy2 = _mm_setzero_pd();
2172 fjz2 = _mm_setzero_pd();
2174 /**************************
2175 * CALCULATE INTERACTIONS *
2176 **************************/
2178 if (gmx_mm_any_lt(rsq00,rcutoff2))
2181 r00 = _mm_mul_pd(rsq00,rinv00);
2183 /* EWALD ELECTROSTATICS */
2185 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2186 ewrt = _mm_mul_pd(r00,ewtabscale);
2187 ewitab = _mm_cvttpd_epi32(ewrt);
2188 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2189 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2190 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
2191 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
2193 /* Analytical LJ-PME */
2194 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
2195 ewcljrsq = _mm_mul_pd(ewclj2,rsq00);
2196 ewclj6 = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
2197 exponent = gmx_simd_exp_d(ewcljrsq);
2198 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
2199 poly = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
2200 /* f6A = 6 * C6grid * (1 - poly) */
2201 f6A = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
2202 /* f6B = C6grid * exponent * beta^6 */
2203 f6B = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
2204 /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
2205 fvdw = _mm_mul_pd(_mm_add_pd(_mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),_mm_sub_pd(c6_00,f6A)),rinvsix),f6B),rinvsq00);
2207 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
2209 fscal = _mm_add_pd(felec,fvdw);
2211 fscal = _mm_and_pd(fscal,cutoff_mask);
2213 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2215 /* Calculate temporary vectorial force */
2216 tx = _mm_mul_pd(fscal,dx00);
2217 ty = _mm_mul_pd(fscal,dy00);
2218 tz = _mm_mul_pd(fscal,dz00);
2220 /* Update vectorial force */
2221 fix0 = _mm_add_pd(fix0,tx);
2222 fiy0 = _mm_add_pd(fiy0,ty);
2223 fiz0 = _mm_add_pd(fiz0,tz);
2225 fjx0 = _mm_add_pd(fjx0,tx);
2226 fjy0 = _mm_add_pd(fjy0,ty);
2227 fjz0 = _mm_add_pd(fjz0,tz);
2231 /**************************
2232 * CALCULATE INTERACTIONS *
2233 **************************/
2235 if (gmx_mm_any_lt(rsq01,rcutoff2))
2238 r01 = _mm_mul_pd(rsq01,rinv01);
2240 /* EWALD ELECTROSTATICS */
2242 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2243 ewrt = _mm_mul_pd(r01,ewtabscale);
2244 ewitab = _mm_cvttpd_epi32(ewrt);
2245 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2246 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2247 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
2248 felec = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
2250 cutoff_mask = _mm_cmplt_pd(rsq01,rcutoff2);
2254 fscal = _mm_and_pd(fscal,cutoff_mask);
2256 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2258 /* Calculate temporary vectorial force */
2259 tx = _mm_mul_pd(fscal,dx01);
2260 ty = _mm_mul_pd(fscal,dy01);
2261 tz = _mm_mul_pd(fscal,dz01);
2263 /* Update vectorial force */
2264 fix0 = _mm_add_pd(fix0,tx);
2265 fiy0 = _mm_add_pd(fiy0,ty);
2266 fiz0 = _mm_add_pd(fiz0,tz);
2268 fjx1 = _mm_add_pd(fjx1,tx);
2269 fjy1 = _mm_add_pd(fjy1,ty);
2270 fjz1 = _mm_add_pd(fjz1,tz);
2274 /**************************
2275 * CALCULATE INTERACTIONS *
2276 **************************/
2278 if (gmx_mm_any_lt(rsq02,rcutoff2))
2281 r02 = _mm_mul_pd(rsq02,rinv02);
2283 /* EWALD ELECTROSTATICS */
2285 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2286 ewrt = _mm_mul_pd(r02,ewtabscale);
2287 ewitab = _mm_cvttpd_epi32(ewrt);
2288 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2289 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2290 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
2291 felec = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
2293 cutoff_mask = _mm_cmplt_pd(rsq02,rcutoff2);
2297 fscal = _mm_and_pd(fscal,cutoff_mask);
2299 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2301 /* Calculate temporary vectorial force */
2302 tx = _mm_mul_pd(fscal,dx02);
2303 ty = _mm_mul_pd(fscal,dy02);
2304 tz = _mm_mul_pd(fscal,dz02);
2306 /* Update vectorial force */
2307 fix0 = _mm_add_pd(fix0,tx);
2308 fiy0 = _mm_add_pd(fiy0,ty);
2309 fiz0 = _mm_add_pd(fiz0,tz);
2311 fjx2 = _mm_add_pd(fjx2,tx);
2312 fjy2 = _mm_add_pd(fjy2,ty);
2313 fjz2 = _mm_add_pd(fjz2,tz);
2317 /**************************
2318 * CALCULATE INTERACTIONS *
2319 **************************/
2321 if (gmx_mm_any_lt(rsq10,rcutoff2))
2324 r10 = _mm_mul_pd(rsq10,rinv10);
2326 /* EWALD ELECTROSTATICS */
2328 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2329 ewrt = _mm_mul_pd(r10,ewtabscale);
2330 ewitab = _mm_cvttpd_epi32(ewrt);
2331 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2332 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2333 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
2334 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
2336 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
2340 fscal = _mm_and_pd(fscal,cutoff_mask);
2342 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2344 /* Calculate temporary vectorial force */
2345 tx = _mm_mul_pd(fscal,dx10);
2346 ty = _mm_mul_pd(fscal,dy10);
2347 tz = _mm_mul_pd(fscal,dz10);
2349 /* Update vectorial force */
2350 fix1 = _mm_add_pd(fix1,tx);
2351 fiy1 = _mm_add_pd(fiy1,ty);
2352 fiz1 = _mm_add_pd(fiz1,tz);
2354 fjx0 = _mm_add_pd(fjx0,tx);
2355 fjy0 = _mm_add_pd(fjy0,ty);
2356 fjz0 = _mm_add_pd(fjz0,tz);
2360 /**************************
2361 * CALCULATE INTERACTIONS *
2362 **************************/
2364 if (gmx_mm_any_lt(rsq11,rcutoff2))
2367 r11 = _mm_mul_pd(rsq11,rinv11);
2369 /* EWALD ELECTROSTATICS */
2371 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2372 ewrt = _mm_mul_pd(r11,ewtabscale);
2373 ewitab = _mm_cvttpd_epi32(ewrt);
2374 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2375 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2376 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
2377 felec = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
2379 cutoff_mask = _mm_cmplt_pd(rsq11,rcutoff2);
2383 fscal = _mm_and_pd(fscal,cutoff_mask);
2385 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2387 /* Calculate temporary vectorial force */
2388 tx = _mm_mul_pd(fscal,dx11);
2389 ty = _mm_mul_pd(fscal,dy11);
2390 tz = _mm_mul_pd(fscal,dz11);
2392 /* Update vectorial force */
2393 fix1 = _mm_add_pd(fix1,tx);
2394 fiy1 = _mm_add_pd(fiy1,ty);
2395 fiz1 = _mm_add_pd(fiz1,tz);
2397 fjx1 = _mm_add_pd(fjx1,tx);
2398 fjy1 = _mm_add_pd(fjy1,ty);
2399 fjz1 = _mm_add_pd(fjz1,tz);
2403 /**************************
2404 * CALCULATE INTERACTIONS *
2405 **************************/
2407 if (gmx_mm_any_lt(rsq12,rcutoff2))
2410 r12 = _mm_mul_pd(rsq12,rinv12);
2412 /* EWALD ELECTROSTATICS */
2414 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2415 ewrt = _mm_mul_pd(r12,ewtabscale);
2416 ewitab = _mm_cvttpd_epi32(ewrt);
2417 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2418 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2419 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
2420 felec = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
2422 cutoff_mask = _mm_cmplt_pd(rsq12,rcutoff2);
2426 fscal = _mm_and_pd(fscal,cutoff_mask);
2428 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2430 /* Calculate temporary vectorial force */
2431 tx = _mm_mul_pd(fscal,dx12);
2432 ty = _mm_mul_pd(fscal,dy12);
2433 tz = _mm_mul_pd(fscal,dz12);
2435 /* Update vectorial force */
2436 fix1 = _mm_add_pd(fix1,tx);
2437 fiy1 = _mm_add_pd(fiy1,ty);
2438 fiz1 = _mm_add_pd(fiz1,tz);
2440 fjx2 = _mm_add_pd(fjx2,tx);
2441 fjy2 = _mm_add_pd(fjy2,ty);
2442 fjz2 = _mm_add_pd(fjz2,tz);
2446 /**************************
2447 * CALCULATE INTERACTIONS *
2448 **************************/
2450 if (gmx_mm_any_lt(rsq20,rcutoff2))
2453 r20 = _mm_mul_pd(rsq20,rinv20);
2455 /* EWALD ELECTROSTATICS */
2457 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2458 ewrt = _mm_mul_pd(r20,ewtabscale);
2459 ewitab = _mm_cvttpd_epi32(ewrt);
2460 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2461 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2462 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
2463 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
2465 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
2469 fscal = _mm_and_pd(fscal,cutoff_mask);
2471 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2473 /* Calculate temporary vectorial force */
2474 tx = _mm_mul_pd(fscal,dx20);
2475 ty = _mm_mul_pd(fscal,dy20);
2476 tz = _mm_mul_pd(fscal,dz20);
2478 /* Update vectorial force */
2479 fix2 = _mm_add_pd(fix2,tx);
2480 fiy2 = _mm_add_pd(fiy2,ty);
2481 fiz2 = _mm_add_pd(fiz2,tz);
2483 fjx0 = _mm_add_pd(fjx0,tx);
2484 fjy0 = _mm_add_pd(fjy0,ty);
2485 fjz0 = _mm_add_pd(fjz0,tz);
2489 /**************************
2490 * CALCULATE INTERACTIONS *
2491 **************************/
2493 if (gmx_mm_any_lt(rsq21,rcutoff2))
2496 r21 = _mm_mul_pd(rsq21,rinv21);
2498 /* EWALD ELECTROSTATICS */
2500 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2501 ewrt = _mm_mul_pd(r21,ewtabscale);
2502 ewitab = _mm_cvttpd_epi32(ewrt);
2503 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2504 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2505 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
2506 felec = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
2508 cutoff_mask = _mm_cmplt_pd(rsq21,rcutoff2);
2512 fscal = _mm_and_pd(fscal,cutoff_mask);
2514 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2516 /* Calculate temporary vectorial force */
2517 tx = _mm_mul_pd(fscal,dx21);
2518 ty = _mm_mul_pd(fscal,dy21);
2519 tz = _mm_mul_pd(fscal,dz21);
2521 /* Update vectorial force */
2522 fix2 = _mm_add_pd(fix2,tx);
2523 fiy2 = _mm_add_pd(fiy2,ty);
2524 fiz2 = _mm_add_pd(fiz2,tz);
2526 fjx1 = _mm_add_pd(fjx1,tx);
2527 fjy1 = _mm_add_pd(fjy1,ty);
2528 fjz1 = _mm_add_pd(fjz1,tz);
2532 /**************************
2533 * CALCULATE INTERACTIONS *
2534 **************************/
2536 if (gmx_mm_any_lt(rsq22,rcutoff2))
2539 r22 = _mm_mul_pd(rsq22,rinv22);
2541 /* EWALD ELECTROSTATICS */
2543 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2544 ewrt = _mm_mul_pd(r22,ewtabscale);
2545 ewitab = _mm_cvttpd_epi32(ewrt);
2546 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2547 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2548 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
2549 felec = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
2551 cutoff_mask = _mm_cmplt_pd(rsq22,rcutoff2);
2555 fscal = _mm_and_pd(fscal,cutoff_mask);
2557 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2559 /* Calculate temporary vectorial force */
2560 tx = _mm_mul_pd(fscal,dx22);
2561 ty = _mm_mul_pd(fscal,dy22);
2562 tz = _mm_mul_pd(fscal,dz22);
2564 /* Update vectorial force */
2565 fix2 = _mm_add_pd(fix2,tx);
2566 fiy2 = _mm_add_pd(fiy2,ty);
2567 fiz2 = _mm_add_pd(fiz2,tz);
2569 fjx2 = _mm_add_pd(fjx2,tx);
2570 fjy2 = _mm_add_pd(fjy2,ty);
2571 fjz2 = _mm_add_pd(fjz2,tz);
2575 gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2577 /* Inner loop uses 374 flops */
2580 /* End of innermost loop */
2582 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2583 f+i_coord_offset,fshift+i_shift_offset);
2585 /* Increment number of inner iterations */
2586 inneriter += j_index_end - j_index_start;
2588 /* Outer loop uses 18 flops */
2591 /* Increment number of outer iterations */
2594 /* Update outer/inner flops */
2596 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*374);