2 * Note: this file was generated by the Gromacs avx_128_fma_double kernel generator.
4 * This source code is part of
8 * Copyright (c) 2001-2012, The GROMACS Development Team
10 * Gromacs is a library for molecular simulation and trajectory analysis,
11 * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
12 * a full list of developers and information, check out http://www.gromacs.org
14 * This program is free software; you can redistribute it and/or modify it under
15 * the terms of the GNU Lesser General Public License as published by the Free
16 * Software Foundation; either version 2 of the License, or (at your option) any
19 * To help fund GROMACS development, we humbly ask that you cite
20 * the papers people have written on it - you can find them on the website.
28 #include "../nb_kernel.h"
29 #include "types/simple.h"
33 #include "gmx_math_x86_avx_128_fma_double.h"
34 #include "kernelutil_x86_avx_128_fma_double.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwLJ_GeomW3P1_VF_avx_128_fma_double
38 * Electrostatics interaction: Ewald
39 * VdW interaction: LennardJones
40 * Geometry: Water3-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecEw_VdwLJ_GeomW3P1_VF_avx_128_fma_double
45 (t_nblist * gmx_restrict nlist,
46 rvec * gmx_restrict xx,
47 rvec * gmx_restrict ff,
48 t_forcerec * gmx_restrict fr,
49 t_mdatoms * gmx_restrict mdatoms,
50 nb_kernel_data_t * gmx_restrict kernel_data,
51 t_nrnb * gmx_restrict nrnb)
53 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
54 * just 0 for non-waters.
55 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
56 * jnr indices corresponding to data put in the four positions in the SIMD register.
58 int i_shift_offset,i_coord_offset,outeriter,inneriter;
59 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
61 int j_coord_offsetA,j_coord_offsetB;
62 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
64 real *shiftvec,*fshift,*x,*f;
65 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
67 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
69 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
71 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
72 int vdwjidx0A,vdwjidx0B;
73 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
74 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
75 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
76 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
77 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
80 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
83 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
84 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
86 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
88 __m128d dummy_mask,cutoff_mask;
89 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
90 __m128d one = _mm_set1_pd(1.0);
91 __m128d two = _mm_set1_pd(2.0);
97 jindex = nlist->jindex;
99 shiftidx = nlist->shift;
101 shiftvec = fr->shift_vec[0];
102 fshift = fr->fshift[0];
103 facel = _mm_set1_pd(fr->epsfac);
104 charge = mdatoms->chargeA;
105 nvdwtype = fr->ntype;
107 vdwtype = mdatoms->typeA;
109 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
110 ewtab = fr->ic->tabq_coul_FDV0;
111 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
112 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
114 /* Setup water-specific parameters */
115 inr = nlist->iinr[0];
116 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
117 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
118 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
119 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
121 /* Avoid stupid compiler warnings */
129 /* Start outer loop over neighborlists */
130 for(iidx=0; iidx<nri; iidx++)
132 /* Load shift vector for this list */
133 i_shift_offset = DIM*shiftidx[iidx];
135 /* Load limits for loop over neighbors */
136 j_index_start = jindex[iidx];
137 j_index_end = jindex[iidx+1];
139 /* Get outer coordinate index */
141 i_coord_offset = DIM*inr;
143 /* Load i particle coords and add shift vector */
144 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
145 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
147 fix0 = _mm_setzero_pd();
148 fiy0 = _mm_setzero_pd();
149 fiz0 = _mm_setzero_pd();
150 fix1 = _mm_setzero_pd();
151 fiy1 = _mm_setzero_pd();
152 fiz1 = _mm_setzero_pd();
153 fix2 = _mm_setzero_pd();
154 fiy2 = _mm_setzero_pd();
155 fiz2 = _mm_setzero_pd();
157 /* Reset potential sums */
158 velecsum = _mm_setzero_pd();
159 vvdwsum = _mm_setzero_pd();
161 /* Start inner kernel loop */
162 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
165 /* Get j neighbor index, and coordinate index */
168 j_coord_offsetA = DIM*jnrA;
169 j_coord_offsetB = DIM*jnrB;
171 /* load j atom coordinates */
172 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
175 /* Calculate displacement vector */
176 dx00 = _mm_sub_pd(ix0,jx0);
177 dy00 = _mm_sub_pd(iy0,jy0);
178 dz00 = _mm_sub_pd(iz0,jz0);
179 dx10 = _mm_sub_pd(ix1,jx0);
180 dy10 = _mm_sub_pd(iy1,jy0);
181 dz10 = _mm_sub_pd(iz1,jz0);
182 dx20 = _mm_sub_pd(ix2,jx0);
183 dy20 = _mm_sub_pd(iy2,jy0);
184 dz20 = _mm_sub_pd(iz2,jz0);
186 /* Calculate squared distance and things based on it */
187 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
188 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
189 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
191 rinv00 = gmx_mm_invsqrt_pd(rsq00);
192 rinv10 = gmx_mm_invsqrt_pd(rsq10);
193 rinv20 = gmx_mm_invsqrt_pd(rsq20);
195 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
196 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
197 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
199 /* Load parameters for j particles */
200 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
201 vdwjidx0A = 2*vdwtype[jnrA+0];
202 vdwjidx0B = 2*vdwtype[jnrB+0];
204 fjx0 = _mm_setzero_pd();
205 fjy0 = _mm_setzero_pd();
206 fjz0 = _mm_setzero_pd();
208 /**************************
209 * CALCULATE INTERACTIONS *
210 **************************/
212 r00 = _mm_mul_pd(rsq00,rinv00);
214 /* Compute parameters for interactions between i and j atoms */
215 qq00 = _mm_mul_pd(iq0,jq0);
216 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
217 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
219 /* EWALD ELECTROSTATICS */
221 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
222 ewrt = _mm_mul_pd(r00,ewtabscale);
223 ewitab = _mm_cvttpd_epi32(ewrt);
225 eweps = _mm_frcz_pd(ewrt);
227 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
229 twoeweps = _mm_add_pd(eweps,eweps);
230 ewitab = _mm_slli_epi32(ewitab,2);
231 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
232 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
233 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
234 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
235 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
236 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
237 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
238 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
239 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
240 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
242 /* LENNARD-JONES DISPERSION/REPULSION */
244 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
245 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
246 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
247 vvdw = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
248 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
250 /* Update potential sum for this i atom from the interaction with this j atom. */
251 velecsum = _mm_add_pd(velecsum,velec);
252 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
254 fscal = _mm_add_pd(felec,fvdw);
256 /* Update vectorial force */
257 fix0 = _mm_macc_pd(dx00,fscal,fix0);
258 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
259 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
261 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
262 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
263 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
265 /**************************
266 * CALCULATE INTERACTIONS *
267 **************************/
269 r10 = _mm_mul_pd(rsq10,rinv10);
271 /* Compute parameters for interactions between i and j atoms */
272 qq10 = _mm_mul_pd(iq1,jq0);
274 /* EWALD ELECTROSTATICS */
276 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
277 ewrt = _mm_mul_pd(r10,ewtabscale);
278 ewitab = _mm_cvttpd_epi32(ewrt);
280 eweps = _mm_frcz_pd(ewrt);
282 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
284 twoeweps = _mm_add_pd(eweps,eweps);
285 ewitab = _mm_slli_epi32(ewitab,2);
286 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
287 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
288 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
289 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
290 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
291 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
292 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
293 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
294 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
295 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
297 /* Update potential sum for this i atom from the interaction with this j atom. */
298 velecsum = _mm_add_pd(velecsum,velec);
302 /* Update vectorial force */
303 fix1 = _mm_macc_pd(dx10,fscal,fix1);
304 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
305 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
307 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
308 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
309 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
311 /**************************
312 * CALCULATE INTERACTIONS *
313 **************************/
315 r20 = _mm_mul_pd(rsq20,rinv20);
317 /* Compute parameters for interactions between i and j atoms */
318 qq20 = _mm_mul_pd(iq2,jq0);
320 /* EWALD ELECTROSTATICS */
322 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
323 ewrt = _mm_mul_pd(r20,ewtabscale);
324 ewitab = _mm_cvttpd_epi32(ewrt);
326 eweps = _mm_frcz_pd(ewrt);
328 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
330 twoeweps = _mm_add_pd(eweps,eweps);
331 ewitab = _mm_slli_epi32(ewitab,2);
332 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
333 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
334 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
335 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
336 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
337 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
338 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
339 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
340 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
341 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
343 /* Update potential sum for this i atom from the interaction with this j atom. */
344 velecsum = _mm_add_pd(velecsum,velec);
348 /* Update vectorial force */
349 fix2 = _mm_macc_pd(dx20,fscal,fix2);
350 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
351 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
353 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
354 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
355 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
357 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
359 /* Inner loop uses 147 flops */
366 j_coord_offsetA = DIM*jnrA;
368 /* load j atom coordinates */
369 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
372 /* Calculate displacement vector */
373 dx00 = _mm_sub_pd(ix0,jx0);
374 dy00 = _mm_sub_pd(iy0,jy0);
375 dz00 = _mm_sub_pd(iz0,jz0);
376 dx10 = _mm_sub_pd(ix1,jx0);
377 dy10 = _mm_sub_pd(iy1,jy0);
378 dz10 = _mm_sub_pd(iz1,jz0);
379 dx20 = _mm_sub_pd(ix2,jx0);
380 dy20 = _mm_sub_pd(iy2,jy0);
381 dz20 = _mm_sub_pd(iz2,jz0);
383 /* Calculate squared distance and things based on it */
384 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
385 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
386 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
388 rinv00 = gmx_mm_invsqrt_pd(rsq00);
389 rinv10 = gmx_mm_invsqrt_pd(rsq10);
390 rinv20 = gmx_mm_invsqrt_pd(rsq20);
392 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
393 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
394 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
396 /* Load parameters for j particles */
397 jq0 = _mm_load_sd(charge+jnrA+0);
398 vdwjidx0A = 2*vdwtype[jnrA+0];
400 fjx0 = _mm_setzero_pd();
401 fjy0 = _mm_setzero_pd();
402 fjz0 = _mm_setzero_pd();
404 /**************************
405 * CALCULATE INTERACTIONS *
406 **************************/
408 r00 = _mm_mul_pd(rsq00,rinv00);
410 /* Compute parameters for interactions between i and j atoms */
411 qq00 = _mm_mul_pd(iq0,jq0);
412 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
414 /* EWALD ELECTROSTATICS */
416 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
417 ewrt = _mm_mul_pd(r00,ewtabscale);
418 ewitab = _mm_cvttpd_epi32(ewrt);
420 eweps = _mm_frcz_pd(ewrt);
422 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
424 twoeweps = _mm_add_pd(eweps,eweps);
425 ewitab = _mm_slli_epi32(ewitab,2);
426 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
427 ewtabD = _mm_setzero_pd();
428 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
429 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
430 ewtabFn = _mm_setzero_pd();
431 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
432 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
433 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
434 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
435 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
437 /* LENNARD-JONES DISPERSION/REPULSION */
439 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
440 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
441 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
442 vvdw = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
443 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
445 /* Update potential sum for this i atom from the interaction with this j atom. */
446 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
447 velecsum = _mm_add_pd(velecsum,velec);
448 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
449 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
451 fscal = _mm_add_pd(felec,fvdw);
453 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
455 /* Update vectorial force */
456 fix0 = _mm_macc_pd(dx00,fscal,fix0);
457 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
458 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
460 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
461 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
462 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
464 /**************************
465 * CALCULATE INTERACTIONS *
466 **************************/
468 r10 = _mm_mul_pd(rsq10,rinv10);
470 /* Compute parameters for interactions between i and j atoms */
471 qq10 = _mm_mul_pd(iq1,jq0);
473 /* EWALD ELECTROSTATICS */
475 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
476 ewrt = _mm_mul_pd(r10,ewtabscale);
477 ewitab = _mm_cvttpd_epi32(ewrt);
479 eweps = _mm_frcz_pd(ewrt);
481 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
483 twoeweps = _mm_add_pd(eweps,eweps);
484 ewitab = _mm_slli_epi32(ewitab,2);
485 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
486 ewtabD = _mm_setzero_pd();
487 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
488 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
489 ewtabFn = _mm_setzero_pd();
490 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
491 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
492 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
493 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
494 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
496 /* Update potential sum for this i atom from the interaction with this j atom. */
497 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
498 velecsum = _mm_add_pd(velecsum,velec);
502 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
504 /* Update vectorial force */
505 fix1 = _mm_macc_pd(dx10,fscal,fix1);
506 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
507 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
509 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
510 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
511 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
513 /**************************
514 * CALCULATE INTERACTIONS *
515 **************************/
517 r20 = _mm_mul_pd(rsq20,rinv20);
519 /* Compute parameters for interactions between i and j atoms */
520 qq20 = _mm_mul_pd(iq2,jq0);
522 /* EWALD ELECTROSTATICS */
524 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
525 ewrt = _mm_mul_pd(r20,ewtabscale);
526 ewitab = _mm_cvttpd_epi32(ewrt);
528 eweps = _mm_frcz_pd(ewrt);
530 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
532 twoeweps = _mm_add_pd(eweps,eweps);
533 ewitab = _mm_slli_epi32(ewitab,2);
534 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
535 ewtabD = _mm_setzero_pd();
536 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
537 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
538 ewtabFn = _mm_setzero_pd();
539 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
540 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
541 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
542 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
543 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
545 /* Update potential sum for this i atom from the interaction with this j atom. */
546 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
547 velecsum = _mm_add_pd(velecsum,velec);
551 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
553 /* Update vectorial force */
554 fix2 = _mm_macc_pd(dx20,fscal,fix2);
555 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
556 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
558 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
559 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
560 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
562 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
564 /* Inner loop uses 147 flops */
567 /* End of innermost loop */
569 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
570 f+i_coord_offset,fshift+i_shift_offset);
573 /* Update potential energies */
574 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
575 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
577 /* Increment number of inner iterations */
578 inneriter += j_index_end - j_index_start;
580 /* Outer loop uses 20 flops */
583 /* Increment number of outer iterations */
586 /* Update outer/inner flops */
588 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*147);
591 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_avx_128_fma_double
592 * Electrostatics interaction: Ewald
593 * VdW interaction: LennardJones
594 * Geometry: Water3-Particle
595 * Calculate force/pot: Force
598 nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_avx_128_fma_double
599 (t_nblist * gmx_restrict nlist,
600 rvec * gmx_restrict xx,
601 rvec * gmx_restrict ff,
602 t_forcerec * gmx_restrict fr,
603 t_mdatoms * gmx_restrict mdatoms,
604 nb_kernel_data_t * gmx_restrict kernel_data,
605 t_nrnb * gmx_restrict nrnb)
607 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
608 * just 0 for non-waters.
609 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
610 * jnr indices corresponding to data put in the four positions in the SIMD register.
612 int i_shift_offset,i_coord_offset,outeriter,inneriter;
613 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
615 int j_coord_offsetA,j_coord_offsetB;
616 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
618 real *shiftvec,*fshift,*x,*f;
619 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
621 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
623 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
625 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
626 int vdwjidx0A,vdwjidx0B;
627 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
628 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
629 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
630 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
631 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
634 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
637 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
638 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
640 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
642 __m128d dummy_mask,cutoff_mask;
643 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
644 __m128d one = _mm_set1_pd(1.0);
645 __m128d two = _mm_set1_pd(2.0);
651 jindex = nlist->jindex;
653 shiftidx = nlist->shift;
655 shiftvec = fr->shift_vec[0];
656 fshift = fr->fshift[0];
657 facel = _mm_set1_pd(fr->epsfac);
658 charge = mdatoms->chargeA;
659 nvdwtype = fr->ntype;
661 vdwtype = mdatoms->typeA;
663 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
664 ewtab = fr->ic->tabq_coul_F;
665 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
666 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
668 /* Setup water-specific parameters */
669 inr = nlist->iinr[0];
670 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
671 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
672 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
673 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
675 /* Avoid stupid compiler warnings */
683 /* Start outer loop over neighborlists */
684 for(iidx=0; iidx<nri; iidx++)
686 /* Load shift vector for this list */
687 i_shift_offset = DIM*shiftidx[iidx];
689 /* Load limits for loop over neighbors */
690 j_index_start = jindex[iidx];
691 j_index_end = jindex[iidx+1];
693 /* Get outer coordinate index */
695 i_coord_offset = DIM*inr;
697 /* Load i particle coords and add shift vector */
698 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
699 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
701 fix0 = _mm_setzero_pd();
702 fiy0 = _mm_setzero_pd();
703 fiz0 = _mm_setzero_pd();
704 fix1 = _mm_setzero_pd();
705 fiy1 = _mm_setzero_pd();
706 fiz1 = _mm_setzero_pd();
707 fix2 = _mm_setzero_pd();
708 fiy2 = _mm_setzero_pd();
709 fiz2 = _mm_setzero_pd();
711 /* Start inner kernel loop */
712 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
715 /* Get j neighbor index, and coordinate index */
718 j_coord_offsetA = DIM*jnrA;
719 j_coord_offsetB = DIM*jnrB;
721 /* load j atom coordinates */
722 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
725 /* Calculate displacement vector */
726 dx00 = _mm_sub_pd(ix0,jx0);
727 dy00 = _mm_sub_pd(iy0,jy0);
728 dz00 = _mm_sub_pd(iz0,jz0);
729 dx10 = _mm_sub_pd(ix1,jx0);
730 dy10 = _mm_sub_pd(iy1,jy0);
731 dz10 = _mm_sub_pd(iz1,jz0);
732 dx20 = _mm_sub_pd(ix2,jx0);
733 dy20 = _mm_sub_pd(iy2,jy0);
734 dz20 = _mm_sub_pd(iz2,jz0);
736 /* Calculate squared distance and things based on it */
737 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
738 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
739 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
741 rinv00 = gmx_mm_invsqrt_pd(rsq00);
742 rinv10 = gmx_mm_invsqrt_pd(rsq10);
743 rinv20 = gmx_mm_invsqrt_pd(rsq20);
745 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
746 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
747 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
749 /* Load parameters for j particles */
750 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
751 vdwjidx0A = 2*vdwtype[jnrA+0];
752 vdwjidx0B = 2*vdwtype[jnrB+0];
754 fjx0 = _mm_setzero_pd();
755 fjy0 = _mm_setzero_pd();
756 fjz0 = _mm_setzero_pd();
758 /**************************
759 * CALCULATE INTERACTIONS *
760 **************************/
762 r00 = _mm_mul_pd(rsq00,rinv00);
764 /* Compute parameters for interactions between i and j atoms */
765 qq00 = _mm_mul_pd(iq0,jq0);
766 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
767 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
769 /* EWALD ELECTROSTATICS */
771 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
772 ewrt = _mm_mul_pd(r00,ewtabscale);
773 ewitab = _mm_cvttpd_epi32(ewrt);
775 eweps = _mm_frcz_pd(ewrt);
777 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
779 twoeweps = _mm_add_pd(eweps,eweps);
780 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
782 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
783 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
785 /* LENNARD-JONES DISPERSION/REPULSION */
787 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
788 fvdw = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
790 fscal = _mm_add_pd(felec,fvdw);
792 /* Update vectorial force */
793 fix0 = _mm_macc_pd(dx00,fscal,fix0);
794 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
795 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
797 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
798 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
799 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
801 /**************************
802 * CALCULATE INTERACTIONS *
803 **************************/
805 r10 = _mm_mul_pd(rsq10,rinv10);
807 /* Compute parameters for interactions between i and j atoms */
808 qq10 = _mm_mul_pd(iq1,jq0);
810 /* EWALD ELECTROSTATICS */
812 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
813 ewrt = _mm_mul_pd(r10,ewtabscale);
814 ewitab = _mm_cvttpd_epi32(ewrt);
816 eweps = _mm_frcz_pd(ewrt);
818 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
820 twoeweps = _mm_add_pd(eweps,eweps);
821 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
823 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
824 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
828 /* Update vectorial force */
829 fix1 = _mm_macc_pd(dx10,fscal,fix1);
830 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
831 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
833 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
834 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
835 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
837 /**************************
838 * CALCULATE INTERACTIONS *
839 **************************/
841 r20 = _mm_mul_pd(rsq20,rinv20);
843 /* Compute parameters for interactions between i and j atoms */
844 qq20 = _mm_mul_pd(iq2,jq0);
846 /* EWALD ELECTROSTATICS */
848 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
849 ewrt = _mm_mul_pd(r20,ewtabscale);
850 ewitab = _mm_cvttpd_epi32(ewrt);
852 eweps = _mm_frcz_pd(ewrt);
854 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
856 twoeweps = _mm_add_pd(eweps,eweps);
857 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
859 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
860 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
864 /* Update vectorial force */
865 fix2 = _mm_macc_pd(dx20,fscal,fix2);
866 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
867 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
869 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
870 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
871 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
873 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
875 /* Inner loop uses 127 flops */
882 j_coord_offsetA = DIM*jnrA;
884 /* load j atom coordinates */
885 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
888 /* Calculate displacement vector */
889 dx00 = _mm_sub_pd(ix0,jx0);
890 dy00 = _mm_sub_pd(iy0,jy0);
891 dz00 = _mm_sub_pd(iz0,jz0);
892 dx10 = _mm_sub_pd(ix1,jx0);
893 dy10 = _mm_sub_pd(iy1,jy0);
894 dz10 = _mm_sub_pd(iz1,jz0);
895 dx20 = _mm_sub_pd(ix2,jx0);
896 dy20 = _mm_sub_pd(iy2,jy0);
897 dz20 = _mm_sub_pd(iz2,jz0);
899 /* Calculate squared distance and things based on it */
900 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
901 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
902 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
904 rinv00 = gmx_mm_invsqrt_pd(rsq00);
905 rinv10 = gmx_mm_invsqrt_pd(rsq10);
906 rinv20 = gmx_mm_invsqrt_pd(rsq20);
908 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
909 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
910 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
912 /* Load parameters for j particles */
913 jq0 = _mm_load_sd(charge+jnrA+0);
914 vdwjidx0A = 2*vdwtype[jnrA+0];
916 fjx0 = _mm_setzero_pd();
917 fjy0 = _mm_setzero_pd();
918 fjz0 = _mm_setzero_pd();
920 /**************************
921 * CALCULATE INTERACTIONS *
922 **************************/
924 r00 = _mm_mul_pd(rsq00,rinv00);
926 /* Compute parameters for interactions between i and j atoms */
927 qq00 = _mm_mul_pd(iq0,jq0);
928 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
930 /* EWALD ELECTROSTATICS */
932 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
933 ewrt = _mm_mul_pd(r00,ewtabscale);
934 ewitab = _mm_cvttpd_epi32(ewrt);
936 eweps = _mm_frcz_pd(ewrt);
938 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
940 twoeweps = _mm_add_pd(eweps,eweps);
941 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
942 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
943 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
945 /* LENNARD-JONES DISPERSION/REPULSION */
947 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
948 fvdw = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
950 fscal = _mm_add_pd(felec,fvdw);
952 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
954 /* Update vectorial force */
955 fix0 = _mm_macc_pd(dx00,fscal,fix0);
956 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
957 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
959 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
960 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
961 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
963 /**************************
964 * CALCULATE INTERACTIONS *
965 **************************/
967 r10 = _mm_mul_pd(rsq10,rinv10);
969 /* Compute parameters for interactions between i and j atoms */
970 qq10 = _mm_mul_pd(iq1,jq0);
972 /* EWALD ELECTROSTATICS */
974 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
975 ewrt = _mm_mul_pd(r10,ewtabscale);
976 ewitab = _mm_cvttpd_epi32(ewrt);
978 eweps = _mm_frcz_pd(ewrt);
980 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
982 twoeweps = _mm_add_pd(eweps,eweps);
983 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
984 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
985 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
989 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
991 /* Update vectorial force */
992 fix1 = _mm_macc_pd(dx10,fscal,fix1);
993 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
994 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
996 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
997 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
998 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
1000 /**************************
1001 * CALCULATE INTERACTIONS *
1002 **************************/
1004 r20 = _mm_mul_pd(rsq20,rinv20);
1006 /* Compute parameters for interactions between i and j atoms */
1007 qq20 = _mm_mul_pd(iq2,jq0);
1009 /* EWALD ELECTROSTATICS */
1011 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1012 ewrt = _mm_mul_pd(r20,ewtabscale);
1013 ewitab = _mm_cvttpd_epi32(ewrt);
1015 eweps = _mm_frcz_pd(ewrt);
1017 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1019 twoeweps = _mm_add_pd(eweps,eweps);
1020 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1021 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1022 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1026 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1028 /* Update vectorial force */
1029 fix2 = _mm_macc_pd(dx20,fscal,fix2);
1030 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
1031 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
1033 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
1034 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
1035 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
1037 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1039 /* Inner loop uses 127 flops */
1042 /* End of innermost loop */
1044 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1045 f+i_coord_offset,fshift+i_shift_offset);
1047 /* Increment number of inner iterations */
1048 inneriter += j_index_end - j_index_start;
1050 /* Outer loop uses 18 flops */
1053 /* Increment number of outer iterations */
1056 /* Update outer/inner flops */
1058 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*127);