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.
42 #include "../nb_kernel.h"
43 #include "gromacs/legacyheaders/types/simple.h"
44 #include "gromacs/math/vec.h"
45 #include "gromacs/legacyheaders/nrnb.h"
47 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
48 #include "kernelutil_x86_avx_128_fma_double.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwLJ_GeomW3P1_VF_avx_128_fma_double
52 * Electrostatics interaction: Ewald
53 * VdW interaction: LennardJones
54 * Geometry: Water3-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecEw_VdwLJ_GeomW3P1_VF_avx_128_fma_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 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
89 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
90 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
91 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
94 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
97 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
98 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
100 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
102 __m128d dummy_mask,cutoff_mask;
103 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
104 __m128d one = _mm_set1_pd(1.0);
105 __m128d two = _mm_set1_pd(2.0);
111 jindex = nlist->jindex;
113 shiftidx = nlist->shift;
115 shiftvec = fr->shift_vec[0];
116 fshift = fr->fshift[0];
117 facel = _mm_set1_pd(fr->epsfac);
118 charge = mdatoms->chargeA;
119 nvdwtype = fr->ntype;
121 vdwtype = mdatoms->typeA;
123 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
124 ewtab = fr->ic->tabq_coul_FDV0;
125 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
126 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
128 /* Setup water-specific parameters */
129 inr = nlist->iinr[0];
130 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
131 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
132 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
133 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
135 /* Avoid stupid compiler warnings */
143 /* Start outer loop over neighborlists */
144 for(iidx=0; iidx<nri; iidx++)
146 /* Load shift vector for this list */
147 i_shift_offset = DIM*shiftidx[iidx];
149 /* Load limits for loop over neighbors */
150 j_index_start = jindex[iidx];
151 j_index_end = jindex[iidx+1];
153 /* Get outer coordinate index */
155 i_coord_offset = DIM*inr;
157 /* Load i particle coords and add shift vector */
158 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
159 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
161 fix0 = _mm_setzero_pd();
162 fiy0 = _mm_setzero_pd();
163 fiz0 = _mm_setzero_pd();
164 fix1 = _mm_setzero_pd();
165 fiy1 = _mm_setzero_pd();
166 fiz1 = _mm_setzero_pd();
167 fix2 = _mm_setzero_pd();
168 fiy2 = _mm_setzero_pd();
169 fiz2 = _mm_setzero_pd();
171 /* Reset potential sums */
172 velecsum = _mm_setzero_pd();
173 vvdwsum = _mm_setzero_pd();
175 /* Start inner kernel loop */
176 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
179 /* Get j neighbor index, and coordinate index */
182 j_coord_offsetA = DIM*jnrA;
183 j_coord_offsetB = DIM*jnrB;
185 /* load j atom coordinates */
186 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
189 /* Calculate displacement vector */
190 dx00 = _mm_sub_pd(ix0,jx0);
191 dy00 = _mm_sub_pd(iy0,jy0);
192 dz00 = _mm_sub_pd(iz0,jz0);
193 dx10 = _mm_sub_pd(ix1,jx0);
194 dy10 = _mm_sub_pd(iy1,jy0);
195 dz10 = _mm_sub_pd(iz1,jz0);
196 dx20 = _mm_sub_pd(ix2,jx0);
197 dy20 = _mm_sub_pd(iy2,jy0);
198 dz20 = _mm_sub_pd(iz2,jz0);
200 /* Calculate squared distance and things based on it */
201 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
202 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
203 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
205 rinv00 = gmx_mm_invsqrt_pd(rsq00);
206 rinv10 = gmx_mm_invsqrt_pd(rsq10);
207 rinv20 = gmx_mm_invsqrt_pd(rsq20);
209 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
210 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
211 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
213 /* Load parameters for j particles */
214 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
215 vdwjidx0A = 2*vdwtype[jnrA+0];
216 vdwjidx0B = 2*vdwtype[jnrB+0];
218 fjx0 = _mm_setzero_pd();
219 fjy0 = _mm_setzero_pd();
220 fjz0 = _mm_setzero_pd();
222 /**************************
223 * CALCULATE INTERACTIONS *
224 **************************/
226 r00 = _mm_mul_pd(rsq00,rinv00);
228 /* Compute parameters for interactions between i and j atoms */
229 qq00 = _mm_mul_pd(iq0,jq0);
230 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
231 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
233 /* EWALD ELECTROSTATICS */
235 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
236 ewrt = _mm_mul_pd(r00,ewtabscale);
237 ewitab = _mm_cvttpd_epi32(ewrt);
239 eweps = _mm_frcz_pd(ewrt);
241 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
243 twoeweps = _mm_add_pd(eweps,eweps);
244 ewitab = _mm_slli_epi32(ewitab,2);
245 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
246 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
247 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
248 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
249 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
250 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
251 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
252 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
253 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
254 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
256 /* LENNARD-JONES DISPERSION/REPULSION */
258 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
259 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
260 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
261 vvdw = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
262 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
264 /* Update potential sum for this i atom from the interaction with this j atom. */
265 velecsum = _mm_add_pd(velecsum,velec);
266 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
268 fscal = _mm_add_pd(felec,fvdw);
270 /* Update vectorial force */
271 fix0 = _mm_macc_pd(dx00,fscal,fix0);
272 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
273 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
275 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
276 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
277 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
279 /**************************
280 * CALCULATE INTERACTIONS *
281 **************************/
283 r10 = _mm_mul_pd(rsq10,rinv10);
285 /* Compute parameters for interactions between i and j atoms */
286 qq10 = _mm_mul_pd(iq1,jq0);
288 /* EWALD ELECTROSTATICS */
290 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
291 ewrt = _mm_mul_pd(r10,ewtabscale);
292 ewitab = _mm_cvttpd_epi32(ewrt);
294 eweps = _mm_frcz_pd(ewrt);
296 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
298 twoeweps = _mm_add_pd(eweps,eweps);
299 ewitab = _mm_slli_epi32(ewitab,2);
300 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
301 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
302 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
303 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
304 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
305 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
306 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
307 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
308 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
309 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
311 /* Update potential sum for this i atom from the interaction with this j atom. */
312 velecsum = _mm_add_pd(velecsum,velec);
316 /* Update vectorial force */
317 fix1 = _mm_macc_pd(dx10,fscal,fix1);
318 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
319 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
321 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
322 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
323 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
325 /**************************
326 * CALCULATE INTERACTIONS *
327 **************************/
329 r20 = _mm_mul_pd(rsq20,rinv20);
331 /* Compute parameters for interactions between i and j atoms */
332 qq20 = _mm_mul_pd(iq2,jq0);
334 /* EWALD ELECTROSTATICS */
336 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
337 ewrt = _mm_mul_pd(r20,ewtabscale);
338 ewitab = _mm_cvttpd_epi32(ewrt);
340 eweps = _mm_frcz_pd(ewrt);
342 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
344 twoeweps = _mm_add_pd(eweps,eweps);
345 ewitab = _mm_slli_epi32(ewitab,2);
346 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
347 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
348 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
349 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
350 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
351 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
352 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
353 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
354 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
355 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
357 /* Update potential sum for this i atom from the interaction with this j atom. */
358 velecsum = _mm_add_pd(velecsum,velec);
362 /* Update vectorial force */
363 fix2 = _mm_macc_pd(dx20,fscal,fix2);
364 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
365 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
367 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
368 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
369 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
371 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
373 /* Inner loop uses 147 flops */
380 j_coord_offsetA = DIM*jnrA;
382 /* load j atom coordinates */
383 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
386 /* Calculate displacement vector */
387 dx00 = _mm_sub_pd(ix0,jx0);
388 dy00 = _mm_sub_pd(iy0,jy0);
389 dz00 = _mm_sub_pd(iz0,jz0);
390 dx10 = _mm_sub_pd(ix1,jx0);
391 dy10 = _mm_sub_pd(iy1,jy0);
392 dz10 = _mm_sub_pd(iz1,jz0);
393 dx20 = _mm_sub_pd(ix2,jx0);
394 dy20 = _mm_sub_pd(iy2,jy0);
395 dz20 = _mm_sub_pd(iz2,jz0);
397 /* Calculate squared distance and things based on it */
398 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
399 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
400 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
402 rinv00 = gmx_mm_invsqrt_pd(rsq00);
403 rinv10 = gmx_mm_invsqrt_pd(rsq10);
404 rinv20 = gmx_mm_invsqrt_pd(rsq20);
406 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
407 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
408 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
410 /* Load parameters for j particles */
411 jq0 = _mm_load_sd(charge+jnrA+0);
412 vdwjidx0A = 2*vdwtype[jnrA+0];
414 fjx0 = _mm_setzero_pd();
415 fjy0 = _mm_setzero_pd();
416 fjz0 = _mm_setzero_pd();
418 /**************************
419 * CALCULATE INTERACTIONS *
420 **************************/
422 r00 = _mm_mul_pd(rsq00,rinv00);
424 /* Compute parameters for interactions between i and j atoms */
425 qq00 = _mm_mul_pd(iq0,jq0);
426 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
428 /* EWALD ELECTROSTATICS */
430 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
431 ewrt = _mm_mul_pd(r00,ewtabscale);
432 ewitab = _mm_cvttpd_epi32(ewrt);
434 eweps = _mm_frcz_pd(ewrt);
436 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
438 twoeweps = _mm_add_pd(eweps,eweps);
439 ewitab = _mm_slli_epi32(ewitab,2);
440 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
441 ewtabD = _mm_setzero_pd();
442 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
443 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
444 ewtabFn = _mm_setzero_pd();
445 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
446 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
447 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
448 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
449 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
451 /* LENNARD-JONES DISPERSION/REPULSION */
453 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
454 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
455 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
456 vvdw = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
457 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
459 /* Update potential sum for this i atom from the interaction with this j atom. */
460 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
461 velecsum = _mm_add_pd(velecsum,velec);
462 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
463 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
465 fscal = _mm_add_pd(felec,fvdw);
467 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
469 /* Update vectorial force */
470 fix0 = _mm_macc_pd(dx00,fscal,fix0);
471 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
472 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
474 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
475 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
476 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
478 /**************************
479 * CALCULATE INTERACTIONS *
480 **************************/
482 r10 = _mm_mul_pd(rsq10,rinv10);
484 /* Compute parameters for interactions between i and j atoms */
485 qq10 = _mm_mul_pd(iq1,jq0);
487 /* EWALD ELECTROSTATICS */
489 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
490 ewrt = _mm_mul_pd(r10,ewtabscale);
491 ewitab = _mm_cvttpd_epi32(ewrt);
493 eweps = _mm_frcz_pd(ewrt);
495 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
497 twoeweps = _mm_add_pd(eweps,eweps);
498 ewitab = _mm_slli_epi32(ewitab,2);
499 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
500 ewtabD = _mm_setzero_pd();
501 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
502 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
503 ewtabFn = _mm_setzero_pd();
504 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
505 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
506 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
507 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
508 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
510 /* Update potential sum for this i atom from the interaction with this j atom. */
511 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
512 velecsum = _mm_add_pd(velecsum,velec);
516 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
518 /* Update vectorial force */
519 fix1 = _mm_macc_pd(dx10,fscal,fix1);
520 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
521 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
523 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
524 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
525 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
527 /**************************
528 * CALCULATE INTERACTIONS *
529 **************************/
531 r20 = _mm_mul_pd(rsq20,rinv20);
533 /* Compute parameters for interactions between i and j atoms */
534 qq20 = _mm_mul_pd(iq2,jq0);
536 /* EWALD ELECTROSTATICS */
538 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
539 ewrt = _mm_mul_pd(r20,ewtabscale);
540 ewitab = _mm_cvttpd_epi32(ewrt);
542 eweps = _mm_frcz_pd(ewrt);
544 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
546 twoeweps = _mm_add_pd(eweps,eweps);
547 ewitab = _mm_slli_epi32(ewitab,2);
548 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
549 ewtabD = _mm_setzero_pd();
550 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
551 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
552 ewtabFn = _mm_setzero_pd();
553 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
554 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
555 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
556 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
557 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
559 /* Update potential sum for this i atom from the interaction with this j atom. */
560 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
561 velecsum = _mm_add_pd(velecsum,velec);
565 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
567 /* Update vectorial force */
568 fix2 = _mm_macc_pd(dx20,fscal,fix2);
569 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
570 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
572 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
573 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
574 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
576 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
578 /* Inner loop uses 147 flops */
581 /* End of innermost loop */
583 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
584 f+i_coord_offset,fshift+i_shift_offset);
587 /* Update potential energies */
588 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
589 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
591 /* Increment number of inner iterations */
592 inneriter += j_index_end - j_index_start;
594 /* Outer loop uses 20 flops */
597 /* Increment number of outer iterations */
600 /* Update outer/inner flops */
602 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*147);
605 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_avx_128_fma_double
606 * Electrostatics interaction: Ewald
607 * VdW interaction: LennardJones
608 * Geometry: Water3-Particle
609 * Calculate force/pot: Force
612 nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_avx_128_fma_double
613 (t_nblist * gmx_restrict nlist,
614 rvec * gmx_restrict xx,
615 rvec * gmx_restrict ff,
616 t_forcerec * gmx_restrict fr,
617 t_mdatoms * gmx_restrict mdatoms,
618 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
619 t_nrnb * gmx_restrict nrnb)
621 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
622 * just 0 for non-waters.
623 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
624 * jnr indices corresponding to data put in the four positions in the SIMD register.
626 int i_shift_offset,i_coord_offset,outeriter,inneriter;
627 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
629 int j_coord_offsetA,j_coord_offsetB;
630 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
632 real *shiftvec,*fshift,*x,*f;
633 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
635 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
637 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
639 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
640 int vdwjidx0A,vdwjidx0B;
641 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
642 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
643 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
644 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
645 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
648 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
651 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
652 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
654 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
656 __m128d dummy_mask,cutoff_mask;
657 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
658 __m128d one = _mm_set1_pd(1.0);
659 __m128d two = _mm_set1_pd(2.0);
665 jindex = nlist->jindex;
667 shiftidx = nlist->shift;
669 shiftvec = fr->shift_vec[0];
670 fshift = fr->fshift[0];
671 facel = _mm_set1_pd(fr->epsfac);
672 charge = mdatoms->chargeA;
673 nvdwtype = fr->ntype;
675 vdwtype = mdatoms->typeA;
677 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
678 ewtab = fr->ic->tabq_coul_F;
679 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
680 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
682 /* Setup water-specific parameters */
683 inr = nlist->iinr[0];
684 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
685 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
686 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
687 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
689 /* Avoid stupid compiler warnings */
697 /* Start outer loop over neighborlists */
698 for(iidx=0; iidx<nri; iidx++)
700 /* Load shift vector for this list */
701 i_shift_offset = DIM*shiftidx[iidx];
703 /* Load limits for loop over neighbors */
704 j_index_start = jindex[iidx];
705 j_index_end = jindex[iidx+1];
707 /* Get outer coordinate index */
709 i_coord_offset = DIM*inr;
711 /* Load i particle coords and add shift vector */
712 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
713 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
715 fix0 = _mm_setzero_pd();
716 fiy0 = _mm_setzero_pd();
717 fiz0 = _mm_setzero_pd();
718 fix1 = _mm_setzero_pd();
719 fiy1 = _mm_setzero_pd();
720 fiz1 = _mm_setzero_pd();
721 fix2 = _mm_setzero_pd();
722 fiy2 = _mm_setzero_pd();
723 fiz2 = _mm_setzero_pd();
725 /* Start inner kernel loop */
726 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
729 /* Get j neighbor index, and coordinate index */
732 j_coord_offsetA = DIM*jnrA;
733 j_coord_offsetB = DIM*jnrB;
735 /* load j atom coordinates */
736 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
739 /* Calculate displacement vector */
740 dx00 = _mm_sub_pd(ix0,jx0);
741 dy00 = _mm_sub_pd(iy0,jy0);
742 dz00 = _mm_sub_pd(iz0,jz0);
743 dx10 = _mm_sub_pd(ix1,jx0);
744 dy10 = _mm_sub_pd(iy1,jy0);
745 dz10 = _mm_sub_pd(iz1,jz0);
746 dx20 = _mm_sub_pd(ix2,jx0);
747 dy20 = _mm_sub_pd(iy2,jy0);
748 dz20 = _mm_sub_pd(iz2,jz0);
750 /* Calculate squared distance and things based on it */
751 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
752 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
753 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
755 rinv00 = gmx_mm_invsqrt_pd(rsq00);
756 rinv10 = gmx_mm_invsqrt_pd(rsq10);
757 rinv20 = gmx_mm_invsqrt_pd(rsq20);
759 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
760 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
761 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
763 /* Load parameters for j particles */
764 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
765 vdwjidx0A = 2*vdwtype[jnrA+0];
766 vdwjidx0B = 2*vdwtype[jnrB+0];
768 fjx0 = _mm_setzero_pd();
769 fjy0 = _mm_setzero_pd();
770 fjz0 = _mm_setzero_pd();
772 /**************************
773 * CALCULATE INTERACTIONS *
774 **************************/
776 r00 = _mm_mul_pd(rsq00,rinv00);
778 /* Compute parameters for interactions between i and j atoms */
779 qq00 = _mm_mul_pd(iq0,jq0);
780 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
781 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
783 /* EWALD ELECTROSTATICS */
785 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
786 ewrt = _mm_mul_pd(r00,ewtabscale);
787 ewitab = _mm_cvttpd_epi32(ewrt);
789 eweps = _mm_frcz_pd(ewrt);
791 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
793 twoeweps = _mm_add_pd(eweps,eweps);
794 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
796 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
797 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
799 /* LENNARD-JONES DISPERSION/REPULSION */
801 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
802 fvdw = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
804 fscal = _mm_add_pd(felec,fvdw);
806 /* Update vectorial force */
807 fix0 = _mm_macc_pd(dx00,fscal,fix0);
808 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
809 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
811 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
812 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
813 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
815 /**************************
816 * CALCULATE INTERACTIONS *
817 **************************/
819 r10 = _mm_mul_pd(rsq10,rinv10);
821 /* Compute parameters for interactions between i and j atoms */
822 qq10 = _mm_mul_pd(iq1,jq0);
824 /* EWALD ELECTROSTATICS */
826 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
827 ewrt = _mm_mul_pd(r10,ewtabscale);
828 ewitab = _mm_cvttpd_epi32(ewrt);
830 eweps = _mm_frcz_pd(ewrt);
832 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
834 twoeweps = _mm_add_pd(eweps,eweps);
835 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
837 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
838 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
842 /* Update vectorial force */
843 fix1 = _mm_macc_pd(dx10,fscal,fix1);
844 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
845 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
847 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
848 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
849 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
851 /**************************
852 * CALCULATE INTERACTIONS *
853 **************************/
855 r20 = _mm_mul_pd(rsq20,rinv20);
857 /* Compute parameters for interactions between i and j atoms */
858 qq20 = _mm_mul_pd(iq2,jq0);
860 /* EWALD ELECTROSTATICS */
862 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
863 ewrt = _mm_mul_pd(r20,ewtabscale);
864 ewitab = _mm_cvttpd_epi32(ewrt);
866 eweps = _mm_frcz_pd(ewrt);
868 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
870 twoeweps = _mm_add_pd(eweps,eweps);
871 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
873 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
874 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
878 /* Update vectorial force */
879 fix2 = _mm_macc_pd(dx20,fscal,fix2);
880 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
881 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
883 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
884 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
885 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
887 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
889 /* Inner loop uses 127 flops */
896 j_coord_offsetA = DIM*jnrA;
898 /* load j atom coordinates */
899 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
902 /* Calculate displacement vector */
903 dx00 = _mm_sub_pd(ix0,jx0);
904 dy00 = _mm_sub_pd(iy0,jy0);
905 dz00 = _mm_sub_pd(iz0,jz0);
906 dx10 = _mm_sub_pd(ix1,jx0);
907 dy10 = _mm_sub_pd(iy1,jy0);
908 dz10 = _mm_sub_pd(iz1,jz0);
909 dx20 = _mm_sub_pd(ix2,jx0);
910 dy20 = _mm_sub_pd(iy2,jy0);
911 dz20 = _mm_sub_pd(iz2,jz0);
913 /* Calculate squared distance and things based on it */
914 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
915 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
916 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
918 rinv00 = gmx_mm_invsqrt_pd(rsq00);
919 rinv10 = gmx_mm_invsqrt_pd(rsq10);
920 rinv20 = gmx_mm_invsqrt_pd(rsq20);
922 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
923 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
924 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
926 /* Load parameters for j particles */
927 jq0 = _mm_load_sd(charge+jnrA+0);
928 vdwjidx0A = 2*vdwtype[jnrA+0];
930 fjx0 = _mm_setzero_pd();
931 fjy0 = _mm_setzero_pd();
932 fjz0 = _mm_setzero_pd();
934 /**************************
935 * CALCULATE INTERACTIONS *
936 **************************/
938 r00 = _mm_mul_pd(rsq00,rinv00);
940 /* Compute parameters for interactions between i and j atoms */
941 qq00 = _mm_mul_pd(iq0,jq0);
942 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
944 /* EWALD ELECTROSTATICS */
946 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
947 ewrt = _mm_mul_pd(r00,ewtabscale);
948 ewitab = _mm_cvttpd_epi32(ewrt);
950 eweps = _mm_frcz_pd(ewrt);
952 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
954 twoeweps = _mm_add_pd(eweps,eweps);
955 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
956 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
957 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
959 /* LENNARD-JONES DISPERSION/REPULSION */
961 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
962 fvdw = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
964 fscal = _mm_add_pd(felec,fvdw);
966 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
968 /* Update vectorial force */
969 fix0 = _mm_macc_pd(dx00,fscal,fix0);
970 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
971 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
973 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
974 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
975 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
977 /**************************
978 * CALCULATE INTERACTIONS *
979 **************************/
981 r10 = _mm_mul_pd(rsq10,rinv10);
983 /* Compute parameters for interactions between i and j atoms */
984 qq10 = _mm_mul_pd(iq1,jq0);
986 /* EWALD ELECTROSTATICS */
988 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
989 ewrt = _mm_mul_pd(r10,ewtabscale);
990 ewitab = _mm_cvttpd_epi32(ewrt);
992 eweps = _mm_frcz_pd(ewrt);
994 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
996 twoeweps = _mm_add_pd(eweps,eweps);
997 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
998 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
999 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1003 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1005 /* Update vectorial force */
1006 fix1 = _mm_macc_pd(dx10,fscal,fix1);
1007 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
1008 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
1010 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
1011 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
1012 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
1014 /**************************
1015 * CALCULATE INTERACTIONS *
1016 **************************/
1018 r20 = _mm_mul_pd(rsq20,rinv20);
1020 /* Compute parameters for interactions between i and j atoms */
1021 qq20 = _mm_mul_pd(iq2,jq0);
1023 /* EWALD ELECTROSTATICS */
1025 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1026 ewrt = _mm_mul_pd(r20,ewtabscale);
1027 ewitab = _mm_cvttpd_epi32(ewrt);
1029 eweps = _mm_frcz_pd(ewrt);
1031 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1033 twoeweps = _mm_add_pd(eweps,eweps);
1034 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1035 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1036 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1040 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1042 /* Update vectorial force */
1043 fix2 = _mm_macc_pd(dx20,fscal,fix2);
1044 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
1045 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
1047 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
1048 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
1049 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
1051 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1053 /* Inner loop uses 127 flops */
1056 /* End of innermost loop */
1058 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1059 f+i_coord_offset,fshift+i_shift_offset);
1061 /* Increment number of inner iterations */
1062 inneriter += j_index_end - j_index_start;
1064 /* Outer loop uses 18 flops */
1067 /* Increment number of outer iterations */
1070 /* Update outer/inner flops */
1072 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*127);