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 avx_128_fma_double kernel generator.
44 #include "../nb_kernel.h"
45 #include "gromacs/legacyheaders/types/simple.h"
46 #include "gromacs/math/vec.h"
47 #include "gromacs/legacyheaders/nrnb.h"
49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
50 #include "kernelutil_x86_avx_128_fma_double.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3W3_VF_avx_128_fma_double
54 * Electrostatics interaction: Ewald
55 * VdW interaction: LJEwald
56 * Geometry: Water3-Water3
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3W3_VF_avx_128_fma_double
61 (t_nblist * gmx_restrict nlist,
62 rvec * gmx_restrict xx,
63 rvec * gmx_restrict ff,
64 t_forcerec * gmx_restrict fr,
65 t_mdatoms * gmx_restrict mdatoms,
66 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
67 t_nrnb * gmx_restrict nrnb)
69 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
70 * just 0 for non-waters.
71 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
72 * jnr indices corresponding to data put in the four positions in the SIMD register.
74 int i_shift_offset,i_coord_offset,outeriter,inneriter;
75 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
77 int j_coord_offsetA,j_coord_offsetB;
78 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
80 real *shiftvec,*fshift,*x,*f;
81 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
83 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
85 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
87 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
88 int vdwjidx0A,vdwjidx0B;
89 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
90 int vdwjidx1A,vdwjidx1B;
91 __m128d jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
92 int vdwjidx2A,vdwjidx2B;
93 __m128d jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
94 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
95 __m128d dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
96 __m128d dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
97 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
98 __m128d dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
99 __m128d dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
100 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
101 __m128d dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
102 __m128d dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
103 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
106 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
109 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
110 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
121 __m128d ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
122 __m128d one_half = _mm_set1_pd(0.5);
123 __m128d minus_one = _mm_set1_pd(-1.0);
125 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
127 __m128d dummy_mask,cutoff_mask;
128 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
129 __m128d one = _mm_set1_pd(1.0);
130 __m128d two = _mm_set1_pd(2.0);
136 jindex = nlist->jindex;
138 shiftidx = nlist->shift;
140 shiftvec = fr->shift_vec[0];
141 fshift = fr->fshift[0];
142 facel = _mm_set1_pd(fr->epsfac);
143 charge = mdatoms->chargeA;
144 nvdwtype = fr->ntype;
146 vdwtype = mdatoms->typeA;
147 vdwgridparam = fr->ljpme_c6grid;
148 sh_lj_ewald = _mm_set1_pd(fr->ic->sh_lj_ewald);
149 ewclj = _mm_set1_pd(fr->ewaldcoeff_lj);
150 ewclj2 = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
152 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
153 ewtab = fr->ic->tabq_coul_FDV0;
154 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
155 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
157 /* Setup water-specific parameters */
158 inr = nlist->iinr[0];
159 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
160 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
161 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
162 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
164 jq0 = _mm_set1_pd(charge[inr+0]);
165 jq1 = _mm_set1_pd(charge[inr+1]);
166 jq2 = _mm_set1_pd(charge[inr+2]);
167 vdwjidx0A = 2*vdwtype[inr+0];
168 qq00 = _mm_mul_pd(iq0,jq0);
169 c6_00 = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
170 c12_00 = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
171 c6grid_00 = _mm_set1_pd(vdwgridparam[vdwioffset0+vdwjidx0A]);
172 qq01 = _mm_mul_pd(iq0,jq1);
173 qq02 = _mm_mul_pd(iq0,jq2);
174 qq10 = _mm_mul_pd(iq1,jq0);
175 qq11 = _mm_mul_pd(iq1,jq1);
176 qq12 = _mm_mul_pd(iq1,jq2);
177 qq20 = _mm_mul_pd(iq2,jq0);
178 qq21 = _mm_mul_pd(iq2,jq1);
179 qq22 = _mm_mul_pd(iq2,jq2);
181 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
182 rcutoff_scalar = fr->rcoulomb;
183 rcutoff = _mm_set1_pd(rcutoff_scalar);
184 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
186 sh_vdw_invrcut6 = _mm_set1_pd(fr->ic->sh_invrc6);
187 rvdw = _mm_set1_pd(fr->rvdw);
189 /* Avoid stupid compiler warnings */
197 /* Start outer loop over neighborlists */
198 for(iidx=0; iidx<nri; iidx++)
200 /* Load shift vector for this list */
201 i_shift_offset = DIM*shiftidx[iidx];
203 /* Load limits for loop over neighbors */
204 j_index_start = jindex[iidx];
205 j_index_end = jindex[iidx+1];
207 /* Get outer coordinate index */
209 i_coord_offset = DIM*inr;
211 /* Load i particle coords and add shift vector */
212 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
213 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
215 fix0 = _mm_setzero_pd();
216 fiy0 = _mm_setzero_pd();
217 fiz0 = _mm_setzero_pd();
218 fix1 = _mm_setzero_pd();
219 fiy1 = _mm_setzero_pd();
220 fiz1 = _mm_setzero_pd();
221 fix2 = _mm_setzero_pd();
222 fiy2 = _mm_setzero_pd();
223 fiz2 = _mm_setzero_pd();
225 /* Reset potential sums */
226 velecsum = _mm_setzero_pd();
227 vvdwsum = _mm_setzero_pd();
229 /* Start inner kernel loop */
230 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
233 /* Get j neighbor index, and coordinate index */
236 j_coord_offsetA = DIM*jnrA;
237 j_coord_offsetB = DIM*jnrB;
239 /* load j atom coordinates */
240 gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
241 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
243 /* Calculate displacement vector */
244 dx00 = _mm_sub_pd(ix0,jx0);
245 dy00 = _mm_sub_pd(iy0,jy0);
246 dz00 = _mm_sub_pd(iz0,jz0);
247 dx01 = _mm_sub_pd(ix0,jx1);
248 dy01 = _mm_sub_pd(iy0,jy1);
249 dz01 = _mm_sub_pd(iz0,jz1);
250 dx02 = _mm_sub_pd(ix0,jx2);
251 dy02 = _mm_sub_pd(iy0,jy2);
252 dz02 = _mm_sub_pd(iz0,jz2);
253 dx10 = _mm_sub_pd(ix1,jx0);
254 dy10 = _mm_sub_pd(iy1,jy0);
255 dz10 = _mm_sub_pd(iz1,jz0);
256 dx11 = _mm_sub_pd(ix1,jx1);
257 dy11 = _mm_sub_pd(iy1,jy1);
258 dz11 = _mm_sub_pd(iz1,jz1);
259 dx12 = _mm_sub_pd(ix1,jx2);
260 dy12 = _mm_sub_pd(iy1,jy2);
261 dz12 = _mm_sub_pd(iz1,jz2);
262 dx20 = _mm_sub_pd(ix2,jx0);
263 dy20 = _mm_sub_pd(iy2,jy0);
264 dz20 = _mm_sub_pd(iz2,jz0);
265 dx21 = _mm_sub_pd(ix2,jx1);
266 dy21 = _mm_sub_pd(iy2,jy1);
267 dz21 = _mm_sub_pd(iz2,jz1);
268 dx22 = _mm_sub_pd(ix2,jx2);
269 dy22 = _mm_sub_pd(iy2,jy2);
270 dz22 = _mm_sub_pd(iz2,jz2);
272 /* Calculate squared distance and things based on it */
273 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
274 rsq01 = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
275 rsq02 = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
276 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
277 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
278 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
279 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
280 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
281 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
283 rinv00 = gmx_mm_invsqrt_pd(rsq00);
284 rinv01 = gmx_mm_invsqrt_pd(rsq01);
285 rinv02 = gmx_mm_invsqrt_pd(rsq02);
286 rinv10 = gmx_mm_invsqrt_pd(rsq10);
287 rinv11 = gmx_mm_invsqrt_pd(rsq11);
288 rinv12 = gmx_mm_invsqrt_pd(rsq12);
289 rinv20 = gmx_mm_invsqrt_pd(rsq20);
290 rinv21 = gmx_mm_invsqrt_pd(rsq21);
291 rinv22 = gmx_mm_invsqrt_pd(rsq22);
293 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
294 rinvsq01 = _mm_mul_pd(rinv01,rinv01);
295 rinvsq02 = _mm_mul_pd(rinv02,rinv02);
296 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
297 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
298 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
299 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
300 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
301 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
303 fjx0 = _mm_setzero_pd();
304 fjy0 = _mm_setzero_pd();
305 fjz0 = _mm_setzero_pd();
306 fjx1 = _mm_setzero_pd();
307 fjy1 = _mm_setzero_pd();
308 fjz1 = _mm_setzero_pd();
309 fjx2 = _mm_setzero_pd();
310 fjy2 = _mm_setzero_pd();
311 fjz2 = _mm_setzero_pd();
313 /**************************
314 * CALCULATE INTERACTIONS *
315 **************************/
317 if (gmx_mm_any_lt(rsq00,rcutoff2))
320 r00 = _mm_mul_pd(rsq00,rinv00);
322 /* EWALD ELECTROSTATICS */
324 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
325 ewrt = _mm_mul_pd(r00,ewtabscale);
326 ewitab = _mm_cvttpd_epi32(ewrt);
328 eweps = _mm_frcz_pd(ewrt);
330 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
332 twoeweps = _mm_add_pd(eweps,eweps);
333 ewitab = _mm_slli_epi32(ewitab,2);
334 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
335 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
336 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
337 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
338 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
339 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
340 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
341 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
342 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
343 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
345 /* Analytical LJ-PME */
346 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
347 ewcljrsq = _mm_mul_pd(ewclj2,rsq00);
348 ewclj6 = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
349 exponent = gmx_simd_exp_d(ewcljrsq);
350 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
351 poly = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
352 /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
353 vvdw6 = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
354 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
355 vvdw = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
356 _mm_mul_pd(_mm_sub_pd(vvdw6,_mm_macc_pd(c6grid_00,sh_lj_ewald,_mm_mul_pd(c6_00,sh_vdw_invrcut6))),one_sixth));
357 /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
358 fvdw = _mm_mul_pd(_mm_add_pd(vvdw12,_mm_msub_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6),vvdw6)),rinvsq00);
360 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
362 /* Update potential sum for this i atom from the interaction with this j atom. */
363 velec = _mm_and_pd(velec,cutoff_mask);
364 velecsum = _mm_add_pd(velecsum,velec);
365 vvdw = _mm_and_pd(vvdw,cutoff_mask);
366 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
368 fscal = _mm_add_pd(felec,fvdw);
370 fscal = _mm_and_pd(fscal,cutoff_mask);
372 /* Update vectorial force */
373 fix0 = _mm_macc_pd(dx00,fscal,fix0);
374 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
375 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
377 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
378 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
379 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
383 /**************************
384 * CALCULATE INTERACTIONS *
385 **************************/
387 if (gmx_mm_any_lt(rsq01,rcutoff2))
390 r01 = _mm_mul_pd(rsq01,rinv01);
392 /* EWALD ELECTROSTATICS */
394 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
395 ewrt = _mm_mul_pd(r01,ewtabscale);
396 ewitab = _mm_cvttpd_epi32(ewrt);
398 eweps = _mm_frcz_pd(ewrt);
400 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
402 twoeweps = _mm_add_pd(eweps,eweps);
403 ewitab = _mm_slli_epi32(ewitab,2);
404 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
405 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
406 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
407 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
408 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
409 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
410 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
411 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
412 velec = _mm_mul_pd(qq01,_mm_sub_pd(_mm_sub_pd(rinv01,sh_ewald),velec));
413 felec = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
415 cutoff_mask = _mm_cmplt_pd(rsq01,rcutoff2);
417 /* Update potential sum for this i atom from the interaction with this j atom. */
418 velec = _mm_and_pd(velec,cutoff_mask);
419 velecsum = _mm_add_pd(velecsum,velec);
423 fscal = _mm_and_pd(fscal,cutoff_mask);
425 /* Update vectorial force */
426 fix0 = _mm_macc_pd(dx01,fscal,fix0);
427 fiy0 = _mm_macc_pd(dy01,fscal,fiy0);
428 fiz0 = _mm_macc_pd(dz01,fscal,fiz0);
430 fjx1 = _mm_macc_pd(dx01,fscal,fjx1);
431 fjy1 = _mm_macc_pd(dy01,fscal,fjy1);
432 fjz1 = _mm_macc_pd(dz01,fscal,fjz1);
436 /**************************
437 * CALCULATE INTERACTIONS *
438 **************************/
440 if (gmx_mm_any_lt(rsq02,rcutoff2))
443 r02 = _mm_mul_pd(rsq02,rinv02);
445 /* EWALD ELECTROSTATICS */
447 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
448 ewrt = _mm_mul_pd(r02,ewtabscale);
449 ewitab = _mm_cvttpd_epi32(ewrt);
451 eweps = _mm_frcz_pd(ewrt);
453 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
455 twoeweps = _mm_add_pd(eweps,eweps);
456 ewitab = _mm_slli_epi32(ewitab,2);
457 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
458 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
459 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
460 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
461 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
462 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
463 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
464 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
465 velec = _mm_mul_pd(qq02,_mm_sub_pd(_mm_sub_pd(rinv02,sh_ewald),velec));
466 felec = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
468 cutoff_mask = _mm_cmplt_pd(rsq02,rcutoff2);
470 /* Update potential sum for this i atom from the interaction with this j atom. */
471 velec = _mm_and_pd(velec,cutoff_mask);
472 velecsum = _mm_add_pd(velecsum,velec);
476 fscal = _mm_and_pd(fscal,cutoff_mask);
478 /* Update vectorial force */
479 fix0 = _mm_macc_pd(dx02,fscal,fix0);
480 fiy0 = _mm_macc_pd(dy02,fscal,fiy0);
481 fiz0 = _mm_macc_pd(dz02,fscal,fiz0);
483 fjx2 = _mm_macc_pd(dx02,fscal,fjx2);
484 fjy2 = _mm_macc_pd(dy02,fscal,fjy2);
485 fjz2 = _mm_macc_pd(dz02,fscal,fjz2);
489 /**************************
490 * CALCULATE INTERACTIONS *
491 **************************/
493 if (gmx_mm_any_lt(rsq10,rcutoff2))
496 r10 = _mm_mul_pd(rsq10,rinv10);
498 /* EWALD ELECTROSTATICS */
500 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
501 ewrt = _mm_mul_pd(r10,ewtabscale);
502 ewitab = _mm_cvttpd_epi32(ewrt);
504 eweps = _mm_frcz_pd(ewrt);
506 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
508 twoeweps = _mm_add_pd(eweps,eweps);
509 ewitab = _mm_slli_epi32(ewitab,2);
510 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
511 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
512 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
513 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
514 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
515 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
516 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
517 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
518 velec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
519 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
521 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
523 /* Update potential sum for this i atom from the interaction with this j atom. */
524 velec = _mm_and_pd(velec,cutoff_mask);
525 velecsum = _mm_add_pd(velecsum,velec);
529 fscal = _mm_and_pd(fscal,cutoff_mask);
531 /* Update vectorial force */
532 fix1 = _mm_macc_pd(dx10,fscal,fix1);
533 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
534 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
536 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
537 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
538 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
542 /**************************
543 * CALCULATE INTERACTIONS *
544 **************************/
546 if (gmx_mm_any_lt(rsq11,rcutoff2))
549 r11 = _mm_mul_pd(rsq11,rinv11);
551 /* EWALD ELECTROSTATICS */
553 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
554 ewrt = _mm_mul_pd(r11,ewtabscale);
555 ewitab = _mm_cvttpd_epi32(ewrt);
557 eweps = _mm_frcz_pd(ewrt);
559 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
561 twoeweps = _mm_add_pd(eweps,eweps);
562 ewitab = _mm_slli_epi32(ewitab,2);
563 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
564 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
565 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
566 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
567 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
568 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
569 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
570 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
571 velec = _mm_mul_pd(qq11,_mm_sub_pd(_mm_sub_pd(rinv11,sh_ewald),velec));
572 felec = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
574 cutoff_mask = _mm_cmplt_pd(rsq11,rcutoff2);
576 /* Update potential sum for this i atom from the interaction with this j atom. */
577 velec = _mm_and_pd(velec,cutoff_mask);
578 velecsum = _mm_add_pd(velecsum,velec);
582 fscal = _mm_and_pd(fscal,cutoff_mask);
584 /* Update vectorial force */
585 fix1 = _mm_macc_pd(dx11,fscal,fix1);
586 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
587 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
589 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
590 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
591 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
595 /**************************
596 * CALCULATE INTERACTIONS *
597 **************************/
599 if (gmx_mm_any_lt(rsq12,rcutoff2))
602 r12 = _mm_mul_pd(rsq12,rinv12);
604 /* EWALD ELECTROSTATICS */
606 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
607 ewrt = _mm_mul_pd(r12,ewtabscale);
608 ewitab = _mm_cvttpd_epi32(ewrt);
610 eweps = _mm_frcz_pd(ewrt);
612 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
614 twoeweps = _mm_add_pd(eweps,eweps);
615 ewitab = _mm_slli_epi32(ewitab,2);
616 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
617 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
618 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
619 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
620 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
621 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
622 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
623 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
624 velec = _mm_mul_pd(qq12,_mm_sub_pd(_mm_sub_pd(rinv12,sh_ewald),velec));
625 felec = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
627 cutoff_mask = _mm_cmplt_pd(rsq12,rcutoff2);
629 /* Update potential sum for this i atom from the interaction with this j atom. */
630 velec = _mm_and_pd(velec,cutoff_mask);
631 velecsum = _mm_add_pd(velecsum,velec);
635 fscal = _mm_and_pd(fscal,cutoff_mask);
637 /* Update vectorial force */
638 fix1 = _mm_macc_pd(dx12,fscal,fix1);
639 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
640 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
642 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
643 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
644 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
648 /**************************
649 * CALCULATE INTERACTIONS *
650 **************************/
652 if (gmx_mm_any_lt(rsq20,rcutoff2))
655 r20 = _mm_mul_pd(rsq20,rinv20);
657 /* EWALD ELECTROSTATICS */
659 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
660 ewrt = _mm_mul_pd(r20,ewtabscale);
661 ewitab = _mm_cvttpd_epi32(ewrt);
663 eweps = _mm_frcz_pd(ewrt);
665 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
667 twoeweps = _mm_add_pd(eweps,eweps);
668 ewitab = _mm_slli_epi32(ewitab,2);
669 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
670 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
671 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
672 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
673 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
674 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
675 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
676 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
677 velec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
678 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
680 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
682 /* Update potential sum for this i atom from the interaction with this j atom. */
683 velec = _mm_and_pd(velec,cutoff_mask);
684 velecsum = _mm_add_pd(velecsum,velec);
688 fscal = _mm_and_pd(fscal,cutoff_mask);
690 /* Update vectorial force */
691 fix2 = _mm_macc_pd(dx20,fscal,fix2);
692 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
693 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
695 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
696 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
697 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
701 /**************************
702 * CALCULATE INTERACTIONS *
703 **************************/
705 if (gmx_mm_any_lt(rsq21,rcutoff2))
708 r21 = _mm_mul_pd(rsq21,rinv21);
710 /* EWALD ELECTROSTATICS */
712 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
713 ewrt = _mm_mul_pd(r21,ewtabscale);
714 ewitab = _mm_cvttpd_epi32(ewrt);
716 eweps = _mm_frcz_pd(ewrt);
718 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
720 twoeweps = _mm_add_pd(eweps,eweps);
721 ewitab = _mm_slli_epi32(ewitab,2);
722 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
723 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
724 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
725 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
726 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
727 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
728 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
729 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
730 velec = _mm_mul_pd(qq21,_mm_sub_pd(_mm_sub_pd(rinv21,sh_ewald),velec));
731 felec = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
733 cutoff_mask = _mm_cmplt_pd(rsq21,rcutoff2);
735 /* Update potential sum for this i atom from the interaction with this j atom. */
736 velec = _mm_and_pd(velec,cutoff_mask);
737 velecsum = _mm_add_pd(velecsum,velec);
741 fscal = _mm_and_pd(fscal,cutoff_mask);
743 /* Update vectorial force */
744 fix2 = _mm_macc_pd(dx21,fscal,fix2);
745 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
746 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
748 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
749 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
750 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
754 /**************************
755 * CALCULATE INTERACTIONS *
756 **************************/
758 if (gmx_mm_any_lt(rsq22,rcutoff2))
761 r22 = _mm_mul_pd(rsq22,rinv22);
763 /* EWALD ELECTROSTATICS */
765 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
766 ewrt = _mm_mul_pd(r22,ewtabscale);
767 ewitab = _mm_cvttpd_epi32(ewrt);
769 eweps = _mm_frcz_pd(ewrt);
771 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
773 twoeweps = _mm_add_pd(eweps,eweps);
774 ewitab = _mm_slli_epi32(ewitab,2);
775 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
776 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
777 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
778 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
779 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
780 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
781 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
782 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
783 velec = _mm_mul_pd(qq22,_mm_sub_pd(_mm_sub_pd(rinv22,sh_ewald),velec));
784 felec = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
786 cutoff_mask = _mm_cmplt_pd(rsq22,rcutoff2);
788 /* Update potential sum for this i atom from the interaction with this j atom. */
789 velec = _mm_and_pd(velec,cutoff_mask);
790 velecsum = _mm_add_pd(velecsum,velec);
794 fscal = _mm_and_pd(fscal,cutoff_mask);
796 /* Update vectorial force */
797 fix2 = _mm_macc_pd(dx22,fscal,fix2);
798 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
799 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
801 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
802 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
803 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
807 gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
809 /* Inner loop uses 470 flops */
816 j_coord_offsetA = DIM*jnrA;
818 /* load j atom coordinates */
819 gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
820 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
822 /* Calculate displacement vector */
823 dx00 = _mm_sub_pd(ix0,jx0);
824 dy00 = _mm_sub_pd(iy0,jy0);
825 dz00 = _mm_sub_pd(iz0,jz0);
826 dx01 = _mm_sub_pd(ix0,jx1);
827 dy01 = _mm_sub_pd(iy0,jy1);
828 dz01 = _mm_sub_pd(iz0,jz1);
829 dx02 = _mm_sub_pd(ix0,jx2);
830 dy02 = _mm_sub_pd(iy0,jy2);
831 dz02 = _mm_sub_pd(iz0,jz2);
832 dx10 = _mm_sub_pd(ix1,jx0);
833 dy10 = _mm_sub_pd(iy1,jy0);
834 dz10 = _mm_sub_pd(iz1,jz0);
835 dx11 = _mm_sub_pd(ix1,jx1);
836 dy11 = _mm_sub_pd(iy1,jy1);
837 dz11 = _mm_sub_pd(iz1,jz1);
838 dx12 = _mm_sub_pd(ix1,jx2);
839 dy12 = _mm_sub_pd(iy1,jy2);
840 dz12 = _mm_sub_pd(iz1,jz2);
841 dx20 = _mm_sub_pd(ix2,jx0);
842 dy20 = _mm_sub_pd(iy2,jy0);
843 dz20 = _mm_sub_pd(iz2,jz0);
844 dx21 = _mm_sub_pd(ix2,jx1);
845 dy21 = _mm_sub_pd(iy2,jy1);
846 dz21 = _mm_sub_pd(iz2,jz1);
847 dx22 = _mm_sub_pd(ix2,jx2);
848 dy22 = _mm_sub_pd(iy2,jy2);
849 dz22 = _mm_sub_pd(iz2,jz2);
851 /* Calculate squared distance and things based on it */
852 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
853 rsq01 = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
854 rsq02 = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
855 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
856 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
857 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
858 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
859 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
860 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
862 rinv00 = gmx_mm_invsqrt_pd(rsq00);
863 rinv01 = gmx_mm_invsqrt_pd(rsq01);
864 rinv02 = gmx_mm_invsqrt_pd(rsq02);
865 rinv10 = gmx_mm_invsqrt_pd(rsq10);
866 rinv11 = gmx_mm_invsqrt_pd(rsq11);
867 rinv12 = gmx_mm_invsqrt_pd(rsq12);
868 rinv20 = gmx_mm_invsqrt_pd(rsq20);
869 rinv21 = gmx_mm_invsqrt_pd(rsq21);
870 rinv22 = gmx_mm_invsqrt_pd(rsq22);
872 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
873 rinvsq01 = _mm_mul_pd(rinv01,rinv01);
874 rinvsq02 = _mm_mul_pd(rinv02,rinv02);
875 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
876 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
877 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
878 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
879 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
880 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
882 fjx0 = _mm_setzero_pd();
883 fjy0 = _mm_setzero_pd();
884 fjz0 = _mm_setzero_pd();
885 fjx1 = _mm_setzero_pd();
886 fjy1 = _mm_setzero_pd();
887 fjz1 = _mm_setzero_pd();
888 fjx2 = _mm_setzero_pd();
889 fjy2 = _mm_setzero_pd();
890 fjz2 = _mm_setzero_pd();
892 /**************************
893 * CALCULATE INTERACTIONS *
894 **************************/
896 if (gmx_mm_any_lt(rsq00,rcutoff2))
899 r00 = _mm_mul_pd(rsq00,rinv00);
901 /* EWALD ELECTROSTATICS */
903 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
904 ewrt = _mm_mul_pd(r00,ewtabscale);
905 ewitab = _mm_cvttpd_epi32(ewrt);
907 eweps = _mm_frcz_pd(ewrt);
909 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
911 twoeweps = _mm_add_pd(eweps,eweps);
912 ewitab = _mm_slli_epi32(ewitab,2);
913 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
914 ewtabD = _mm_setzero_pd();
915 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
916 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
917 ewtabFn = _mm_setzero_pd();
918 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
919 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
920 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
921 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
922 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
924 /* Analytical LJ-PME */
925 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
926 ewcljrsq = _mm_mul_pd(ewclj2,rsq00);
927 ewclj6 = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
928 exponent = gmx_simd_exp_d(ewcljrsq);
929 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
930 poly = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
931 /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
932 vvdw6 = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
933 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
934 vvdw = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
935 _mm_mul_pd(_mm_sub_pd(vvdw6,_mm_macc_pd(c6grid_00,sh_lj_ewald,_mm_mul_pd(c6_00,sh_vdw_invrcut6))),one_sixth));
936 /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
937 fvdw = _mm_mul_pd(_mm_add_pd(vvdw12,_mm_msub_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6),vvdw6)),rinvsq00);
939 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
941 /* Update potential sum for this i atom from the interaction with this j atom. */
942 velec = _mm_and_pd(velec,cutoff_mask);
943 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
944 velecsum = _mm_add_pd(velecsum,velec);
945 vvdw = _mm_and_pd(vvdw,cutoff_mask);
946 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
947 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
949 fscal = _mm_add_pd(felec,fvdw);
951 fscal = _mm_and_pd(fscal,cutoff_mask);
953 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
955 /* Update vectorial force */
956 fix0 = _mm_macc_pd(dx00,fscal,fix0);
957 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
958 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
960 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
961 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
962 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
966 /**************************
967 * CALCULATE INTERACTIONS *
968 **************************/
970 if (gmx_mm_any_lt(rsq01,rcutoff2))
973 r01 = _mm_mul_pd(rsq01,rinv01);
975 /* EWALD ELECTROSTATICS */
977 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
978 ewrt = _mm_mul_pd(r01,ewtabscale);
979 ewitab = _mm_cvttpd_epi32(ewrt);
981 eweps = _mm_frcz_pd(ewrt);
983 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
985 twoeweps = _mm_add_pd(eweps,eweps);
986 ewitab = _mm_slli_epi32(ewitab,2);
987 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
988 ewtabD = _mm_setzero_pd();
989 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
990 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
991 ewtabFn = _mm_setzero_pd();
992 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
993 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
994 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
995 velec = _mm_mul_pd(qq01,_mm_sub_pd(_mm_sub_pd(rinv01,sh_ewald),velec));
996 felec = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
998 cutoff_mask = _mm_cmplt_pd(rsq01,rcutoff2);
1000 /* Update potential sum for this i atom from the interaction with this j atom. */
1001 velec = _mm_and_pd(velec,cutoff_mask);
1002 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1003 velecsum = _mm_add_pd(velecsum,velec);
1007 fscal = _mm_and_pd(fscal,cutoff_mask);
1009 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1011 /* Update vectorial force */
1012 fix0 = _mm_macc_pd(dx01,fscal,fix0);
1013 fiy0 = _mm_macc_pd(dy01,fscal,fiy0);
1014 fiz0 = _mm_macc_pd(dz01,fscal,fiz0);
1016 fjx1 = _mm_macc_pd(dx01,fscal,fjx1);
1017 fjy1 = _mm_macc_pd(dy01,fscal,fjy1);
1018 fjz1 = _mm_macc_pd(dz01,fscal,fjz1);
1022 /**************************
1023 * CALCULATE INTERACTIONS *
1024 **************************/
1026 if (gmx_mm_any_lt(rsq02,rcutoff2))
1029 r02 = _mm_mul_pd(rsq02,rinv02);
1031 /* EWALD ELECTROSTATICS */
1033 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1034 ewrt = _mm_mul_pd(r02,ewtabscale);
1035 ewitab = _mm_cvttpd_epi32(ewrt);
1037 eweps = _mm_frcz_pd(ewrt);
1039 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1041 twoeweps = _mm_add_pd(eweps,eweps);
1042 ewitab = _mm_slli_epi32(ewitab,2);
1043 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1044 ewtabD = _mm_setzero_pd();
1045 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1046 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1047 ewtabFn = _mm_setzero_pd();
1048 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1049 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1050 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1051 velec = _mm_mul_pd(qq02,_mm_sub_pd(_mm_sub_pd(rinv02,sh_ewald),velec));
1052 felec = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
1054 cutoff_mask = _mm_cmplt_pd(rsq02,rcutoff2);
1056 /* Update potential sum for this i atom from the interaction with this j atom. */
1057 velec = _mm_and_pd(velec,cutoff_mask);
1058 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1059 velecsum = _mm_add_pd(velecsum,velec);
1063 fscal = _mm_and_pd(fscal,cutoff_mask);
1065 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1067 /* Update vectorial force */
1068 fix0 = _mm_macc_pd(dx02,fscal,fix0);
1069 fiy0 = _mm_macc_pd(dy02,fscal,fiy0);
1070 fiz0 = _mm_macc_pd(dz02,fscal,fiz0);
1072 fjx2 = _mm_macc_pd(dx02,fscal,fjx2);
1073 fjy2 = _mm_macc_pd(dy02,fscal,fjy2);
1074 fjz2 = _mm_macc_pd(dz02,fscal,fjz2);
1078 /**************************
1079 * CALCULATE INTERACTIONS *
1080 **************************/
1082 if (gmx_mm_any_lt(rsq10,rcutoff2))
1085 r10 = _mm_mul_pd(rsq10,rinv10);
1087 /* EWALD ELECTROSTATICS */
1089 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1090 ewrt = _mm_mul_pd(r10,ewtabscale);
1091 ewitab = _mm_cvttpd_epi32(ewrt);
1093 eweps = _mm_frcz_pd(ewrt);
1095 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1097 twoeweps = _mm_add_pd(eweps,eweps);
1098 ewitab = _mm_slli_epi32(ewitab,2);
1099 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1100 ewtabD = _mm_setzero_pd();
1101 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1102 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1103 ewtabFn = _mm_setzero_pd();
1104 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1105 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1106 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1107 velec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
1108 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1110 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
1112 /* Update potential sum for this i atom from the interaction with this j atom. */
1113 velec = _mm_and_pd(velec,cutoff_mask);
1114 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1115 velecsum = _mm_add_pd(velecsum,velec);
1119 fscal = _mm_and_pd(fscal,cutoff_mask);
1121 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1123 /* Update vectorial force */
1124 fix1 = _mm_macc_pd(dx10,fscal,fix1);
1125 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
1126 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
1128 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
1129 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
1130 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
1134 /**************************
1135 * CALCULATE INTERACTIONS *
1136 **************************/
1138 if (gmx_mm_any_lt(rsq11,rcutoff2))
1141 r11 = _mm_mul_pd(rsq11,rinv11);
1143 /* EWALD ELECTROSTATICS */
1145 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1146 ewrt = _mm_mul_pd(r11,ewtabscale);
1147 ewitab = _mm_cvttpd_epi32(ewrt);
1149 eweps = _mm_frcz_pd(ewrt);
1151 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1153 twoeweps = _mm_add_pd(eweps,eweps);
1154 ewitab = _mm_slli_epi32(ewitab,2);
1155 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1156 ewtabD = _mm_setzero_pd();
1157 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1158 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1159 ewtabFn = _mm_setzero_pd();
1160 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1161 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1162 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1163 velec = _mm_mul_pd(qq11,_mm_sub_pd(_mm_sub_pd(rinv11,sh_ewald),velec));
1164 felec = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1166 cutoff_mask = _mm_cmplt_pd(rsq11,rcutoff2);
1168 /* Update potential sum for this i atom from the interaction with this j atom. */
1169 velec = _mm_and_pd(velec,cutoff_mask);
1170 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1171 velecsum = _mm_add_pd(velecsum,velec);
1175 fscal = _mm_and_pd(fscal,cutoff_mask);
1177 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1179 /* Update vectorial force */
1180 fix1 = _mm_macc_pd(dx11,fscal,fix1);
1181 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
1182 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
1184 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
1185 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
1186 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
1190 /**************************
1191 * CALCULATE INTERACTIONS *
1192 **************************/
1194 if (gmx_mm_any_lt(rsq12,rcutoff2))
1197 r12 = _mm_mul_pd(rsq12,rinv12);
1199 /* EWALD ELECTROSTATICS */
1201 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1202 ewrt = _mm_mul_pd(r12,ewtabscale);
1203 ewitab = _mm_cvttpd_epi32(ewrt);
1205 eweps = _mm_frcz_pd(ewrt);
1207 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1209 twoeweps = _mm_add_pd(eweps,eweps);
1210 ewitab = _mm_slli_epi32(ewitab,2);
1211 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1212 ewtabD = _mm_setzero_pd();
1213 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1214 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1215 ewtabFn = _mm_setzero_pd();
1216 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1217 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1218 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1219 velec = _mm_mul_pd(qq12,_mm_sub_pd(_mm_sub_pd(rinv12,sh_ewald),velec));
1220 felec = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1222 cutoff_mask = _mm_cmplt_pd(rsq12,rcutoff2);
1224 /* Update potential sum for this i atom from the interaction with this j atom. */
1225 velec = _mm_and_pd(velec,cutoff_mask);
1226 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1227 velecsum = _mm_add_pd(velecsum,velec);
1231 fscal = _mm_and_pd(fscal,cutoff_mask);
1233 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1235 /* Update vectorial force */
1236 fix1 = _mm_macc_pd(dx12,fscal,fix1);
1237 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
1238 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
1240 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
1241 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
1242 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
1246 /**************************
1247 * CALCULATE INTERACTIONS *
1248 **************************/
1250 if (gmx_mm_any_lt(rsq20,rcutoff2))
1253 r20 = _mm_mul_pd(rsq20,rinv20);
1255 /* EWALD ELECTROSTATICS */
1257 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1258 ewrt = _mm_mul_pd(r20,ewtabscale);
1259 ewitab = _mm_cvttpd_epi32(ewrt);
1261 eweps = _mm_frcz_pd(ewrt);
1263 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1265 twoeweps = _mm_add_pd(eweps,eweps);
1266 ewitab = _mm_slli_epi32(ewitab,2);
1267 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1268 ewtabD = _mm_setzero_pd();
1269 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1270 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1271 ewtabFn = _mm_setzero_pd();
1272 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1273 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1274 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1275 velec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
1276 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1278 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
1280 /* Update potential sum for this i atom from the interaction with this j atom. */
1281 velec = _mm_and_pd(velec,cutoff_mask);
1282 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1283 velecsum = _mm_add_pd(velecsum,velec);
1287 fscal = _mm_and_pd(fscal,cutoff_mask);
1289 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1291 /* Update vectorial force */
1292 fix2 = _mm_macc_pd(dx20,fscal,fix2);
1293 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
1294 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
1296 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
1297 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
1298 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
1302 /**************************
1303 * CALCULATE INTERACTIONS *
1304 **************************/
1306 if (gmx_mm_any_lt(rsq21,rcutoff2))
1309 r21 = _mm_mul_pd(rsq21,rinv21);
1311 /* EWALD ELECTROSTATICS */
1313 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1314 ewrt = _mm_mul_pd(r21,ewtabscale);
1315 ewitab = _mm_cvttpd_epi32(ewrt);
1317 eweps = _mm_frcz_pd(ewrt);
1319 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1321 twoeweps = _mm_add_pd(eweps,eweps);
1322 ewitab = _mm_slli_epi32(ewitab,2);
1323 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1324 ewtabD = _mm_setzero_pd();
1325 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1326 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1327 ewtabFn = _mm_setzero_pd();
1328 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1329 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1330 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1331 velec = _mm_mul_pd(qq21,_mm_sub_pd(_mm_sub_pd(rinv21,sh_ewald),velec));
1332 felec = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1334 cutoff_mask = _mm_cmplt_pd(rsq21,rcutoff2);
1336 /* Update potential sum for this i atom from the interaction with this j atom. */
1337 velec = _mm_and_pd(velec,cutoff_mask);
1338 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1339 velecsum = _mm_add_pd(velecsum,velec);
1343 fscal = _mm_and_pd(fscal,cutoff_mask);
1345 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1347 /* Update vectorial force */
1348 fix2 = _mm_macc_pd(dx21,fscal,fix2);
1349 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
1350 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
1352 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
1353 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
1354 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
1358 /**************************
1359 * CALCULATE INTERACTIONS *
1360 **************************/
1362 if (gmx_mm_any_lt(rsq22,rcutoff2))
1365 r22 = _mm_mul_pd(rsq22,rinv22);
1367 /* EWALD ELECTROSTATICS */
1369 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1370 ewrt = _mm_mul_pd(r22,ewtabscale);
1371 ewitab = _mm_cvttpd_epi32(ewrt);
1373 eweps = _mm_frcz_pd(ewrt);
1375 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1377 twoeweps = _mm_add_pd(eweps,eweps);
1378 ewitab = _mm_slli_epi32(ewitab,2);
1379 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1380 ewtabD = _mm_setzero_pd();
1381 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1382 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1383 ewtabFn = _mm_setzero_pd();
1384 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1385 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1386 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1387 velec = _mm_mul_pd(qq22,_mm_sub_pd(_mm_sub_pd(rinv22,sh_ewald),velec));
1388 felec = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1390 cutoff_mask = _mm_cmplt_pd(rsq22,rcutoff2);
1392 /* Update potential sum for this i atom from the interaction with this j atom. */
1393 velec = _mm_and_pd(velec,cutoff_mask);
1394 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1395 velecsum = _mm_add_pd(velecsum,velec);
1399 fscal = _mm_and_pd(fscal,cutoff_mask);
1401 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1403 /* Update vectorial force */
1404 fix2 = _mm_macc_pd(dx22,fscal,fix2);
1405 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
1406 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
1408 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
1409 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
1410 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
1414 gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1416 /* Inner loop uses 470 flops */
1419 /* End of innermost loop */
1421 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1422 f+i_coord_offset,fshift+i_shift_offset);
1425 /* Update potential energies */
1426 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1427 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1429 /* Increment number of inner iterations */
1430 inneriter += j_index_end - j_index_start;
1432 /* Outer loop uses 20 flops */
1435 /* Increment number of outer iterations */
1438 /* Update outer/inner flops */
1440 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*470);
1443 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3W3_F_avx_128_fma_double
1444 * Electrostatics interaction: Ewald
1445 * VdW interaction: LJEwald
1446 * Geometry: Water3-Water3
1447 * Calculate force/pot: Force
1450 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3W3_F_avx_128_fma_double
1451 (t_nblist * gmx_restrict nlist,
1452 rvec * gmx_restrict xx,
1453 rvec * gmx_restrict ff,
1454 t_forcerec * gmx_restrict fr,
1455 t_mdatoms * gmx_restrict mdatoms,
1456 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1457 t_nrnb * gmx_restrict nrnb)
1459 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1460 * just 0 for non-waters.
1461 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
1462 * jnr indices corresponding to data put in the four positions in the SIMD register.
1464 int i_shift_offset,i_coord_offset,outeriter,inneriter;
1465 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1467 int j_coord_offsetA,j_coord_offsetB;
1468 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
1469 real rcutoff_scalar;
1470 real *shiftvec,*fshift,*x,*f;
1471 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1473 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1475 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1477 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1478 int vdwjidx0A,vdwjidx0B;
1479 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1480 int vdwjidx1A,vdwjidx1B;
1481 __m128d jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1482 int vdwjidx2A,vdwjidx2B;
1483 __m128d jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1484 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1485 __m128d dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1486 __m128d dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1487 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1488 __m128d dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1489 __m128d dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1490 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1491 __m128d dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1492 __m128d dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1493 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
1496 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1499 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
1500 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
1511 __m128d ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
1512 __m128d one_half = _mm_set1_pd(0.5);
1513 __m128d minus_one = _mm_set1_pd(-1.0);
1515 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1517 __m128d dummy_mask,cutoff_mask;
1518 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
1519 __m128d one = _mm_set1_pd(1.0);
1520 __m128d two = _mm_set1_pd(2.0);
1526 jindex = nlist->jindex;
1528 shiftidx = nlist->shift;
1530 shiftvec = fr->shift_vec[0];
1531 fshift = fr->fshift[0];
1532 facel = _mm_set1_pd(fr->epsfac);
1533 charge = mdatoms->chargeA;
1534 nvdwtype = fr->ntype;
1535 vdwparam = fr->nbfp;
1536 vdwtype = mdatoms->typeA;
1537 vdwgridparam = fr->ljpme_c6grid;
1538 sh_lj_ewald = _mm_set1_pd(fr->ic->sh_lj_ewald);
1539 ewclj = _mm_set1_pd(fr->ewaldcoeff_lj);
1540 ewclj2 = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
1542 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
1543 ewtab = fr->ic->tabq_coul_F;
1544 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
1545 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
1547 /* Setup water-specific parameters */
1548 inr = nlist->iinr[0];
1549 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
1550 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
1551 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
1552 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
1554 jq0 = _mm_set1_pd(charge[inr+0]);
1555 jq1 = _mm_set1_pd(charge[inr+1]);
1556 jq2 = _mm_set1_pd(charge[inr+2]);
1557 vdwjidx0A = 2*vdwtype[inr+0];
1558 qq00 = _mm_mul_pd(iq0,jq0);
1559 c6_00 = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
1560 c12_00 = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
1561 c6grid_00 = _mm_set1_pd(vdwgridparam[vdwioffset0+vdwjidx0A]);
1562 qq01 = _mm_mul_pd(iq0,jq1);
1563 qq02 = _mm_mul_pd(iq0,jq2);
1564 qq10 = _mm_mul_pd(iq1,jq0);
1565 qq11 = _mm_mul_pd(iq1,jq1);
1566 qq12 = _mm_mul_pd(iq1,jq2);
1567 qq20 = _mm_mul_pd(iq2,jq0);
1568 qq21 = _mm_mul_pd(iq2,jq1);
1569 qq22 = _mm_mul_pd(iq2,jq2);
1571 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1572 rcutoff_scalar = fr->rcoulomb;
1573 rcutoff = _mm_set1_pd(rcutoff_scalar);
1574 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
1576 sh_vdw_invrcut6 = _mm_set1_pd(fr->ic->sh_invrc6);
1577 rvdw = _mm_set1_pd(fr->rvdw);
1579 /* Avoid stupid compiler warnings */
1581 j_coord_offsetA = 0;
1582 j_coord_offsetB = 0;
1587 /* Start outer loop over neighborlists */
1588 for(iidx=0; iidx<nri; iidx++)
1590 /* Load shift vector for this list */
1591 i_shift_offset = DIM*shiftidx[iidx];
1593 /* Load limits for loop over neighbors */
1594 j_index_start = jindex[iidx];
1595 j_index_end = jindex[iidx+1];
1597 /* Get outer coordinate index */
1599 i_coord_offset = DIM*inr;
1601 /* Load i particle coords and add shift vector */
1602 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1603 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1605 fix0 = _mm_setzero_pd();
1606 fiy0 = _mm_setzero_pd();
1607 fiz0 = _mm_setzero_pd();
1608 fix1 = _mm_setzero_pd();
1609 fiy1 = _mm_setzero_pd();
1610 fiz1 = _mm_setzero_pd();
1611 fix2 = _mm_setzero_pd();
1612 fiy2 = _mm_setzero_pd();
1613 fiz2 = _mm_setzero_pd();
1615 /* Start inner kernel loop */
1616 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
1619 /* Get j neighbor index, and coordinate index */
1621 jnrB = jjnr[jidx+1];
1622 j_coord_offsetA = DIM*jnrA;
1623 j_coord_offsetB = DIM*jnrB;
1625 /* load j atom coordinates */
1626 gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1627 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1629 /* Calculate displacement vector */
1630 dx00 = _mm_sub_pd(ix0,jx0);
1631 dy00 = _mm_sub_pd(iy0,jy0);
1632 dz00 = _mm_sub_pd(iz0,jz0);
1633 dx01 = _mm_sub_pd(ix0,jx1);
1634 dy01 = _mm_sub_pd(iy0,jy1);
1635 dz01 = _mm_sub_pd(iz0,jz1);
1636 dx02 = _mm_sub_pd(ix0,jx2);
1637 dy02 = _mm_sub_pd(iy0,jy2);
1638 dz02 = _mm_sub_pd(iz0,jz2);
1639 dx10 = _mm_sub_pd(ix1,jx0);
1640 dy10 = _mm_sub_pd(iy1,jy0);
1641 dz10 = _mm_sub_pd(iz1,jz0);
1642 dx11 = _mm_sub_pd(ix1,jx1);
1643 dy11 = _mm_sub_pd(iy1,jy1);
1644 dz11 = _mm_sub_pd(iz1,jz1);
1645 dx12 = _mm_sub_pd(ix1,jx2);
1646 dy12 = _mm_sub_pd(iy1,jy2);
1647 dz12 = _mm_sub_pd(iz1,jz2);
1648 dx20 = _mm_sub_pd(ix2,jx0);
1649 dy20 = _mm_sub_pd(iy2,jy0);
1650 dz20 = _mm_sub_pd(iz2,jz0);
1651 dx21 = _mm_sub_pd(ix2,jx1);
1652 dy21 = _mm_sub_pd(iy2,jy1);
1653 dz21 = _mm_sub_pd(iz2,jz1);
1654 dx22 = _mm_sub_pd(ix2,jx2);
1655 dy22 = _mm_sub_pd(iy2,jy2);
1656 dz22 = _mm_sub_pd(iz2,jz2);
1658 /* Calculate squared distance and things based on it */
1659 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1660 rsq01 = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
1661 rsq02 = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
1662 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1663 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1664 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1665 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1666 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1667 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1669 rinv00 = gmx_mm_invsqrt_pd(rsq00);
1670 rinv01 = gmx_mm_invsqrt_pd(rsq01);
1671 rinv02 = gmx_mm_invsqrt_pd(rsq02);
1672 rinv10 = gmx_mm_invsqrt_pd(rsq10);
1673 rinv11 = gmx_mm_invsqrt_pd(rsq11);
1674 rinv12 = gmx_mm_invsqrt_pd(rsq12);
1675 rinv20 = gmx_mm_invsqrt_pd(rsq20);
1676 rinv21 = gmx_mm_invsqrt_pd(rsq21);
1677 rinv22 = gmx_mm_invsqrt_pd(rsq22);
1679 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
1680 rinvsq01 = _mm_mul_pd(rinv01,rinv01);
1681 rinvsq02 = _mm_mul_pd(rinv02,rinv02);
1682 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
1683 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
1684 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
1685 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
1686 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
1687 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
1689 fjx0 = _mm_setzero_pd();
1690 fjy0 = _mm_setzero_pd();
1691 fjz0 = _mm_setzero_pd();
1692 fjx1 = _mm_setzero_pd();
1693 fjy1 = _mm_setzero_pd();
1694 fjz1 = _mm_setzero_pd();
1695 fjx2 = _mm_setzero_pd();
1696 fjy2 = _mm_setzero_pd();
1697 fjz2 = _mm_setzero_pd();
1699 /**************************
1700 * CALCULATE INTERACTIONS *
1701 **************************/
1703 if (gmx_mm_any_lt(rsq00,rcutoff2))
1706 r00 = _mm_mul_pd(rsq00,rinv00);
1708 /* EWALD ELECTROSTATICS */
1710 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1711 ewrt = _mm_mul_pd(r00,ewtabscale);
1712 ewitab = _mm_cvttpd_epi32(ewrt);
1714 eweps = _mm_frcz_pd(ewrt);
1716 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1718 twoeweps = _mm_add_pd(eweps,eweps);
1719 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1721 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1722 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1724 /* Analytical LJ-PME */
1725 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1726 ewcljrsq = _mm_mul_pd(ewclj2,rsq00);
1727 ewclj6 = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1728 exponent = gmx_simd_exp_d(ewcljrsq);
1729 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1730 poly = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
1731 /* f6A = 6 * C6grid * (1 - poly) */
1732 f6A = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1733 /* f6B = C6grid * exponent * beta^6 */
1734 f6B = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1735 /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1736 fvdw = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
1738 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
1740 fscal = _mm_add_pd(felec,fvdw);
1742 fscal = _mm_and_pd(fscal,cutoff_mask);
1744 /* Update vectorial force */
1745 fix0 = _mm_macc_pd(dx00,fscal,fix0);
1746 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
1747 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
1749 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
1750 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
1751 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
1755 /**************************
1756 * CALCULATE INTERACTIONS *
1757 **************************/
1759 if (gmx_mm_any_lt(rsq01,rcutoff2))
1762 r01 = _mm_mul_pd(rsq01,rinv01);
1764 /* EWALD ELECTROSTATICS */
1766 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1767 ewrt = _mm_mul_pd(r01,ewtabscale);
1768 ewitab = _mm_cvttpd_epi32(ewrt);
1770 eweps = _mm_frcz_pd(ewrt);
1772 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1774 twoeweps = _mm_add_pd(eweps,eweps);
1775 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1777 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1778 felec = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
1780 cutoff_mask = _mm_cmplt_pd(rsq01,rcutoff2);
1784 fscal = _mm_and_pd(fscal,cutoff_mask);
1786 /* Update vectorial force */
1787 fix0 = _mm_macc_pd(dx01,fscal,fix0);
1788 fiy0 = _mm_macc_pd(dy01,fscal,fiy0);
1789 fiz0 = _mm_macc_pd(dz01,fscal,fiz0);
1791 fjx1 = _mm_macc_pd(dx01,fscal,fjx1);
1792 fjy1 = _mm_macc_pd(dy01,fscal,fjy1);
1793 fjz1 = _mm_macc_pd(dz01,fscal,fjz1);
1797 /**************************
1798 * CALCULATE INTERACTIONS *
1799 **************************/
1801 if (gmx_mm_any_lt(rsq02,rcutoff2))
1804 r02 = _mm_mul_pd(rsq02,rinv02);
1806 /* EWALD ELECTROSTATICS */
1808 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1809 ewrt = _mm_mul_pd(r02,ewtabscale);
1810 ewitab = _mm_cvttpd_epi32(ewrt);
1812 eweps = _mm_frcz_pd(ewrt);
1814 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1816 twoeweps = _mm_add_pd(eweps,eweps);
1817 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1819 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1820 felec = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
1822 cutoff_mask = _mm_cmplt_pd(rsq02,rcutoff2);
1826 fscal = _mm_and_pd(fscal,cutoff_mask);
1828 /* Update vectorial force */
1829 fix0 = _mm_macc_pd(dx02,fscal,fix0);
1830 fiy0 = _mm_macc_pd(dy02,fscal,fiy0);
1831 fiz0 = _mm_macc_pd(dz02,fscal,fiz0);
1833 fjx2 = _mm_macc_pd(dx02,fscal,fjx2);
1834 fjy2 = _mm_macc_pd(dy02,fscal,fjy2);
1835 fjz2 = _mm_macc_pd(dz02,fscal,fjz2);
1839 /**************************
1840 * CALCULATE INTERACTIONS *
1841 **************************/
1843 if (gmx_mm_any_lt(rsq10,rcutoff2))
1846 r10 = _mm_mul_pd(rsq10,rinv10);
1848 /* EWALD ELECTROSTATICS */
1850 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1851 ewrt = _mm_mul_pd(r10,ewtabscale);
1852 ewitab = _mm_cvttpd_epi32(ewrt);
1854 eweps = _mm_frcz_pd(ewrt);
1856 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1858 twoeweps = _mm_add_pd(eweps,eweps);
1859 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1861 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1862 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1864 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
1868 fscal = _mm_and_pd(fscal,cutoff_mask);
1870 /* Update vectorial force */
1871 fix1 = _mm_macc_pd(dx10,fscal,fix1);
1872 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
1873 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
1875 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
1876 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
1877 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
1881 /**************************
1882 * CALCULATE INTERACTIONS *
1883 **************************/
1885 if (gmx_mm_any_lt(rsq11,rcutoff2))
1888 r11 = _mm_mul_pd(rsq11,rinv11);
1890 /* EWALD ELECTROSTATICS */
1892 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1893 ewrt = _mm_mul_pd(r11,ewtabscale);
1894 ewitab = _mm_cvttpd_epi32(ewrt);
1896 eweps = _mm_frcz_pd(ewrt);
1898 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1900 twoeweps = _mm_add_pd(eweps,eweps);
1901 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1903 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1904 felec = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1906 cutoff_mask = _mm_cmplt_pd(rsq11,rcutoff2);
1910 fscal = _mm_and_pd(fscal,cutoff_mask);
1912 /* Update vectorial force */
1913 fix1 = _mm_macc_pd(dx11,fscal,fix1);
1914 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
1915 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
1917 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
1918 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
1919 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
1923 /**************************
1924 * CALCULATE INTERACTIONS *
1925 **************************/
1927 if (gmx_mm_any_lt(rsq12,rcutoff2))
1930 r12 = _mm_mul_pd(rsq12,rinv12);
1932 /* EWALD ELECTROSTATICS */
1934 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1935 ewrt = _mm_mul_pd(r12,ewtabscale);
1936 ewitab = _mm_cvttpd_epi32(ewrt);
1938 eweps = _mm_frcz_pd(ewrt);
1940 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1942 twoeweps = _mm_add_pd(eweps,eweps);
1943 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1945 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1946 felec = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1948 cutoff_mask = _mm_cmplt_pd(rsq12,rcutoff2);
1952 fscal = _mm_and_pd(fscal,cutoff_mask);
1954 /* Update vectorial force */
1955 fix1 = _mm_macc_pd(dx12,fscal,fix1);
1956 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
1957 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
1959 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
1960 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
1961 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
1965 /**************************
1966 * CALCULATE INTERACTIONS *
1967 **************************/
1969 if (gmx_mm_any_lt(rsq20,rcutoff2))
1972 r20 = _mm_mul_pd(rsq20,rinv20);
1974 /* EWALD ELECTROSTATICS */
1976 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1977 ewrt = _mm_mul_pd(r20,ewtabscale);
1978 ewitab = _mm_cvttpd_epi32(ewrt);
1980 eweps = _mm_frcz_pd(ewrt);
1982 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1984 twoeweps = _mm_add_pd(eweps,eweps);
1985 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1987 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1988 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1990 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
1994 fscal = _mm_and_pd(fscal,cutoff_mask);
1996 /* Update vectorial force */
1997 fix2 = _mm_macc_pd(dx20,fscal,fix2);
1998 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
1999 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
2001 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
2002 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
2003 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
2007 /**************************
2008 * CALCULATE INTERACTIONS *
2009 **************************/
2011 if (gmx_mm_any_lt(rsq21,rcutoff2))
2014 r21 = _mm_mul_pd(rsq21,rinv21);
2016 /* EWALD ELECTROSTATICS */
2018 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2019 ewrt = _mm_mul_pd(r21,ewtabscale);
2020 ewitab = _mm_cvttpd_epi32(ewrt);
2022 eweps = _mm_frcz_pd(ewrt);
2024 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2026 twoeweps = _mm_add_pd(eweps,eweps);
2027 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
2029 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2030 felec = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
2032 cutoff_mask = _mm_cmplt_pd(rsq21,rcutoff2);
2036 fscal = _mm_and_pd(fscal,cutoff_mask);
2038 /* Update vectorial force */
2039 fix2 = _mm_macc_pd(dx21,fscal,fix2);
2040 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
2041 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
2043 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
2044 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
2045 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
2049 /**************************
2050 * CALCULATE INTERACTIONS *
2051 **************************/
2053 if (gmx_mm_any_lt(rsq22,rcutoff2))
2056 r22 = _mm_mul_pd(rsq22,rinv22);
2058 /* EWALD ELECTROSTATICS */
2060 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2061 ewrt = _mm_mul_pd(r22,ewtabscale);
2062 ewitab = _mm_cvttpd_epi32(ewrt);
2064 eweps = _mm_frcz_pd(ewrt);
2066 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2068 twoeweps = _mm_add_pd(eweps,eweps);
2069 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
2071 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2072 felec = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
2074 cutoff_mask = _mm_cmplt_pd(rsq22,rcutoff2);
2078 fscal = _mm_and_pd(fscal,cutoff_mask);
2080 /* Update vectorial force */
2081 fix2 = _mm_macc_pd(dx22,fscal,fix2);
2082 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
2083 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
2085 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
2086 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
2087 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
2091 gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2093 /* Inner loop uses 399 flops */
2096 if(jidx<j_index_end)
2100 j_coord_offsetA = DIM*jnrA;
2102 /* load j atom coordinates */
2103 gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
2104 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
2106 /* Calculate displacement vector */
2107 dx00 = _mm_sub_pd(ix0,jx0);
2108 dy00 = _mm_sub_pd(iy0,jy0);
2109 dz00 = _mm_sub_pd(iz0,jz0);
2110 dx01 = _mm_sub_pd(ix0,jx1);
2111 dy01 = _mm_sub_pd(iy0,jy1);
2112 dz01 = _mm_sub_pd(iz0,jz1);
2113 dx02 = _mm_sub_pd(ix0,jx2);
2114 dy02 = _mm_sub_pd(iy0,jy2);
2115 dz02 = _mm_sub_pd(iz0,jz2);
2116 dx10 = _mm_sub_pd(ix1,jx0);
2117 dy10 = _mm_sub_pd(iy1,jy0);
2118 dz10 = _mm_sub_pd(iz1,jz0);
2119 dx11 = _mm_sub_pd(ix1,jx1);
2120 dy11 = _mm_sub_pd(iy1,jy1);
2121 dz11 = _mm_sub_pd(iz1,jz1);
2122 dx12 = _mm_sub_pd(ix1,jx2);
2123 dy12 = _mm_sub_pd(iy1,jy2);
2124 dz12 = _mm_sub_pd(iz1,jz2);
2125 dx20 = _mm_sub_pd(ix2,jx0);
2126 dy20 = _mm_sub_pd(iy2,jy0);
2127 dz20 = _mm_sub_pd(iz2,jz0);
2128 dx21 = _mm_sub_pd(ix2,jx1);
2129 dy21 = _mm_sub_pd(iy2,jy1);
2130 dz21 = _mm_sub_pd(iz2,jz1);
2131 dx22 = _mm_sub_pd(ix2,jx2);
2132 dy22 = _mm_sub_pd(iy2,jy2);
2133 dz22 = _mm_sub_pd(iz2,jz2);
2135 /* Calculate squared distance and things based on it */
2136 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
2137 rsq01 = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
2138 rsq02 = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
2139 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
2140 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
2141 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
2142 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
2143 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
2144 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
2146 rinv00 = gmx_mm_invsqrt_pd(rsq00);
2147 rinv01 = gmx_mm_invsqrt_pd(rsq01);
2148 rinv02 = gmx_mm_invsqrt_pd(rsq02);
2149 rinv10 = gmx_mm_invsqrt_pd(rsq10);
2150 rinv11 = gmx_mm_invsqrt_pd(rsq11);
2151 rinv12 = gmx_mm_invsqrt_pd(rsq12);
2152 rinv20 = gmx_mm_invsqrt_pd(rsq20);
2153 rinv21 = gmx_mm_invsqrt_pd(rsq21);
2154 rinv22 = gmx_mm_invsqrt_pd(rsq22);
2156 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
2157 rinvsq01 = _mm_mul_pd(rinv01,rinv01);
2158 rinvsq02 = _mm_mul_pd(rinv02,rinv02);
2159 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
2160 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
2161 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
2162 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
2163 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
2164 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
2166 fjx0 = _mm_setzero_pd();
2167 fjy0 = _mm_setzero_pd();
2168 fjz0 = _mm_setzero_pd();
2169 fjx1 = _mm_setzero_pd();
2170 fjy1 = _mm_setzero_pd();
2171 fjz1 = _mm_setzero_pd();
2172 fjx2 = _mm_setzero_pd();
2173 fjy2 = _mm_setzero_pd();
2174 fjz2 = _mm_setzero_pd();
2176 /**************************
2177 * CALCULATE INTERACTIONS *
2178 **************************/
2180 if (gmx_mm_any_lt(rsq00,rcutoff2))
2183 r00 = _mm_mul_pd(rsq00,rinv00);
2185 /* EWALD ELECTROSTATICS */
2187 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2188 ewrt = _mm_mul_pd(r00,ewtabscale);
2189 ewitab = _mm_cvttpd_epi32(ewrt);
2191 eweps = _mm_frcz_pd(ewrt);
2193 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2195 twoeweps = _mm_add_pd(eweps,eweps);
2196 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2197 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2198 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
2200 /* Analytical LJ-PME */
2201 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
2202 ewcljrsq = _mm_mul_pd(ewclj2,rsq00);
2203 ewclj6 = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
2204 exponent = gmx_simd_exp_d(ewcljrsq);
2205 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
2206 poly = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
2207 /* f6A = 6 * C6grid * (1 - poly) */
2208 f6A = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
2209 /* f6B = C6grid * exponent * beta^6 */
2210 f6B = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
2211 /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
2212 fvdw = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
2214 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
2216 fscal = _mm_add_pd(felec,fvdw);
2218 fscal = _mm_and_pd(fscal,cutoff_mask);
2220 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2222 /* Update vectorial force */
2223 fix0 = _mm_macc_pd(dx00,fscal,fix0);
2224 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
2225 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
2227 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
2228 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
2229 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
2233 /**************************
2234 * CALCULATE INTERACTIONS *
2235 **************************/
2237 if (gmx_mm_any_lt(rsq01,rcutoff2))
2240 r01 = _mm_mul_pd(rsq01,rinv01);
2242 /* EWALD ELECTROSTATICS */
2244 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2245 ewrt = _mm_mul_pd(r01,ewtabscale);
2246 ewitab = _mm_cvttpd_epi32(ewrt);
2248 eweps = _mm_frcz_pd(ewrt);
2250 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2252 twoeweps = _mm_add_pd(eweps,eweps);
2253 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2254 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2255 felec = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
2257 cutoff_mask = _mm_cmplt_pd(rsq01,rcutoff2);
2261 fscal = _mm_and_pd(fscal,cutoff_mask);
2263 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2265 /* Update vectorial force */
2266 fix0 = _mm_macc_pd(dx01,fscal,fix0);
2267 fiy0 = _mm_macc_pd(dy01,fscal,fiy0);
2268 fiz0 = _mm_macc_pd(dz01,fscal,fiz0);
2270 fjx1 = _mm_macc_pd(dx01,fscal,fjx1);
2271 fjy1 = _mm_macc_pd(dy01,fscal,fjy1);
2272 fjz1 = _mm_macc_pd(dz01,fscal,fjz1);
2276 /**************************
2277 * CALCULATE INTERACTIONS *
2278 **************************/
2280 if (gmx_mm_any_lt(rsq02,rcutoff2))
2283 r02 = _mm_mul_pd(rsq02,rinv02);
2285 /* EWALD ELECTROSTATICS */
2287 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2288 ewrt = _mm_mul_pd(r02,ewtabscale);
2289 ewitab = _mm_cvttpd_epi32(ewrt);
2291 eweps = _mm_frcz_pd(ewrt);
2293 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2295 twoeweps = _mm_add_pd(eweps,eweps);
2296 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2297 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2298 felec = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
2300 cutoff_mask = _mm_cmplt_pd(rsq02,rcutoff2);
2304 fscal = _mm_and_pd(fscal,cutoff_mask);
2306 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2308 /* Update vectorial force */
2309 fix0 = _mm_macc_pd(dx02,fscal,fix0);
2310 fiy0 = _mm_macc_pd(dy02,fscal,fiy0);
2311 fiz0 = _mm_macc_pd(dz02,fscal,fiz0);
2313 fjx2 = _mm_macc_pd(dx02,fscal,fjx2);
2314 fjy2 = _mm_macc_pd(dy02,fscal,fjy2);
2315 fjz2 = _mm_macc_pd(dz02,fscal,fjz2);
2319 /**************************
2320 * CALCULATE INTERACTIONS *
2321 **************************/
2323 if (gmx_mm_any_lt(rsq10,rcutoff2))
2326 r10 = _mm_mul_pd(rsq10,rinv10);
2328 /* EWALD ELECTROSTATICS */
2330 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2331 ewrt = _mm_mul_pd(r10,ewtabscale);
2332 ewitab = _mm_cvttpd_epi32(ewrt);
2334 eweps = _mm_frcz_pd(ewrt);
2336 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2338 twoeweps = _mm_add_pd(eweps,eweps);
2339 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2340 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2341 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
2343 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
2347 fscal = _mm_and_pd(fscal,cutoff_mask);
2349 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2351 /* Update vectorial force */
2352 fix1 = _mm_macc_pd(dx10,fscal,fix1);
2353 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
2354 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
2356 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
2357 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
2358 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
2362 /**************************
2363 * CALCULATE INTERACTIONS *
2364 **************************/
2366 if (gmx_mm_any_lt(rsq11,rcutoff2))
2369 r11 = _mm_mul_pd(rsq11,rinv11);
2371 /* EWALD ELECTROSTATICS */
2373 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2374 ewrt = _mm_mul_pd(r11,ewtabscale);
2375 ewitab = _mm_cvttpd_epi32(ewrt);
2377 eweps = _mm_frcz_pd(ewrt);
2379 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2381 twoeweps = _mm_add_pd(eweps,eweps);
2382 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2383 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2384 felec = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
2386 cutoff_mask = _mm_cmplt_pd(rsq11,rcutoff2);
2390 fscal = _mm_and_pd(fscal,cutoff_mask);
2392 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2394 /* Update vectorial force */
2395 fix1 = _mm_macc_pd(dx11,fscal,fix1);
2396 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
2397 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
2399 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
2400 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
2401 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
2405 /**************************
2406 * CALCULATE INTERACTIONS *
2407 **************************/
2409 if (gmx_mm_any_lt(rsq12,rcutoff2))
2412 r12 = _mm_mul_pd(rsq12,rinv12);
2414 /* EWALD ELECTROSTATICS */
2416 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2417 ewrt = _mm_mul_pd(r12,ewtabscale);
2418 ewitab = _mm_cvttpd_epi32(ewrt);
2420 eweps = _mm_frcz_pd(ewrt);
2422 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2424 twoeweps = _mm_add_pd(eweps,eweps);
2425 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2426 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2427 felec = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
2429 cutoff_mask = _mm_cmplt_pd(rsq12,rcutoff2);
2433 fscal = _mm_and_pd(fscal,cutoff_mask);
2435 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2437 /* Update vectorial force */
2438 fix1 = _mm_macc_pd(dx12,fscal,fix1);
2439 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
2440 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
2442 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
2443 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
2444 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
2448 /**************************
2449 * CALCULATE INTERACTIONS *
2450 **************************/
2452 if (gmx_mm_any_lt(rsq20,rcutoff2))
2455 r20 = _mm_mul_pd(rsq20,rinv20);
2457 /* EWALD ELECTROSTATICS */
2459 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2460 ewrt = _mm_mul_pd(r20,ewtabscale);
2461 ewitab = _mm_cvttpd_epi32(ewrt);
2463 eweps = _mm_frcz_pd(ewrt);
2465 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2467 twoeweps = _mm_add_pd(eweps,eweps);
2468 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2469 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2470 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
2472 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
2476 fscal = _mm_and_pd(fscal,cutoff_mask);
2478 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2480 /* Update vectorial force */
2481 fix2 = _mm_macc_pd(dx20,fscal,fix2);
2482 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
2483 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
2485 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
2486 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
2487 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
2491 /**************************
2492 * CALCULATE INTERACTIONS *
2493 **************************/
2495 if (gmx_mm_any_lt(rsq21,rcutoff2))
2498 r21 = _mm_mul_pd(rsq21,rinv21);
2500 /* EWALD ELECTROSTATICS */
2502 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2503 ewrt = _mm_mul_pd(r21,ewtabscale);
2504 ewitab = _mm_cvttpd_epi32(ewrt);
2506 eweps = _mm_frcz_pd(ewrt);
2508 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2510 twoeweps = _mm_add_pd(eweps,eweps);
2511 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2512 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2513 felec = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
2515 cutoff_mask = _mm_cmplt_pd(rsq21,rcutoff2);
2519 fscal = _mm_and_pd(fscal,cutoff_mask);
2521 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2523 /* Update vectorial force */
2524 fix2 = _mm_macc_pd(dx21,fscal,fix2);
2525 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
2526 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
2528 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
2529 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
2530 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
2534 /**************************
2535 * CALCULATE INTERACTIONS *
2536 **************************/
2538 if (gmx_mm_any_lt(rsq22,rcutoff2))
2541 r22 = _mm_mul_pd(rsq22,rinv22);
2543 /* EWALD ELECTROSTATICS */
2545 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2546 ewrt = _mm_mul_pd(r22,ewtabscale);
2547 ewitab = _mm_cvttpd_epi32(ewrt);
2549 eweps = _mm_frcz_pd(ewrt);
2551 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2553 twoeweps = _mm_add_pd(eweps,eweps);
2554 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2555 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2556 felec = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
2558 cutoff_mask = _mm_cmplt_pd(rsq22,rcutoff2);
2562 fscal = _mm_and_pd(fscal,cutoff_mask);
2564 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2566 /* Update vectorial force */
2567 fix2 = _mm_macc_pd(dx22,fscal,fix2);
2568 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
2569 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
2571 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
2572 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
2573 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
2577 gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2579 /* Inner loop uses 399 flops */
2582 /* End of innermost loop */
2584 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2585 f+i_coord_offset,fshift+i_shift_offset);
2587 /* Increment number of inner iterations */
2588 inneriter += j_index_end - j_index_start;
2590 /* Outer loop uses 18 flops */
2593 /* Increment number of outer iterations */
2596 /* Update outer/inner flops */
2598 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*399);