2 * Note: this file was generated by the Gromacs sse4_1_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_sse4_1_double.h"
34 #include "kernelutil_x86_sse4_1_double.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecGB_VdwNone_GeomP1P1_VF_sse4_1_double
38 * Electrostatics interaction: GeneralizedBorn
39 * VdW interaction: None
40 * Geometry: Particle-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecGB_VdwNone_GeomP1P1_VF_sse4_1_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;
68 int vdwjidx0A,vdwjidx0B;
69 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
70 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
71 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
74 __m128d vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,dvdatmp;
75 __m128d minushalf = _mm_set1_pd(-0.5);
76 real *invsqrta,*dvda,*gbtab;
78 __m128i ifour = _mm_set1_epi32(4);
79 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
81 __m128d dummy_mask,cutoff_mask;
82 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
83 __m128d one = _mm_set1_pd(1.0);
84 __m128d two = _mm_set1_pd(2.0);
90 jindex = nlist->jindex;
92 shiftidx = nlist->shift;
94 shiftvec = fr->shift_vec[0];
95 fshift = fr->fshift[0];
96 facel = _mm_set1_pd(fr->epsfac);
97 charge = mdatoms->chargeA;
99 invsqrta = fr->invsqrta;
101 gbtabscale = _mm_set1_pd(fr->gbtab.scale);
102 gbtab = fr->gbtab.data;
103 gbinvepsdiff = _mm_set1_pd((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
105 /* Avoid stupid compiler warnings */
113 /* Start outer loop over neighborlists */
114 for(iidx=0; iidx<nri; iidx++)
116 /* Load shift vector for this list */
117 i_shift_offset = DIM*shiftidx[iidx];
119 /* Load limits for loop over neighbors */
120 j_index_start = jindex[iidx];
121 j_index_end = jindex[iidx+1];
123 /* Get outer coordinate index */
125 i_coord_offset = DIM*inr;
127 /* Load i particle coords and add shift vector */
128 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
130 fix0 = _mm_setzero_pd();
131 fiy0 = _mm_setzero_pd();
132 fiz0 = _mm_setzero_pd();
134 /* Load parameters for i particles */
135 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
136 isai0 = _mm_load1_pd(invsqrta+inr+0);
138 /* Reset potential sums */
139 velecsum = _mm_setzero_pd();
140 vgbsum = _mm_setzero_pd();
141 dvdasum = _mm_setzero_pd();
143 /* Start inner kernel loop */
144 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
147 /* Get j neighbor index, and coordinate index */
150 j_coord_offsetA = DIM*jnrA;
151 j_coord_offsetB = DIM*jnrB;
153 /* load j atom coordinates */
154 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
157 /* Calculate displacement vector */
158 dx00 = _mm_sub_pd(ix0,jx0);
159 dy00 = _mm_sub_pd(iy0,jy0);
160 dz00 = _mm_sub_pd(iz0,jz0);
162 /* Calculate squared distance and things based on it */
163 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
165 rinv00 = gmx_mm_invsqrt_pd(rsq00);
167 /* Load parameters for j particles */
168 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
169 isaj0 = gmx_mm_load_2real_swizzle_pd(invsqrta+jnrA+0,invsqrta+jnrB+0);
171 /**************************
172 * CALCULATE INTERACTIONS *
173 **************************/
175 r00 = _mm_mul_pd(rsq00,rinv00);
177 /* Compute parameters for interactions between i and j atoms */
178 qq00 = _mm_mul_pd(iq0,jq0);
180 /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
181 isaprod = _mm_mul_pd(isai0,isaj0);
182 gbqqfactor = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
183 gbscale = _mm_mul_pd(isaprod,gbtabscale);
185 /* Calculate generalized born table index - this is a separate table from the normal one,
186 * but we use the same procedure by multiplying r with scale and truncating to integer.
188 rt = _mm_mul_pd(r00,gbscale);
189 gbitab = _mm_cvttpd_epi32(rt);
190 gbeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
191 gbitab = _mm_slli_epi32(gbitab,2);
193 Y = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
194 F = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) );
195 GMX_MM_TRANSPOSE2_PD(Y,F);
196 G = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
197 H = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) +2);
198 GMX_MM_TRANSPOSE2_PD(G,H);
199 Heps = _mm_mul_pd(gbeps,H);
200 Fp = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
201 VV = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
202 vgb = _mm_mul_pd(gbqqfactor,VV);
204 FF = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
205 fgb = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
206 dvdatmp = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
207 dvdasum = _mm_add_pd(dvdasum,dvdatmp);
208 gmx_mm_increment_2real_swizzle_pd(dvda+jnrA,dvda+jnrB,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
209 velec = _mm_mul_pd(qq00,rinv00);
210 felec = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
212 /* Update potential sum for this i atom from the interaction with this j atom. */
213 velecsum = _mm_add_pd(velecsum,velec);
214 vgbsum = _mm_add_pd(vgbsum,vgb);
218 /* Calculate temporary vectorial force */
219 tx = _mm_mul_pd(fscal,dx00);
220 ty = _mm_mul_pd(fscal,dy00);
221 tz = _mm_mul_pd(fscal,dz00);
223 /* Update vectorial force */
224 fix0 = _mm_add_pd(fix0,tx);
225 fiy0 = _mm_add_pd(fiy0,ty);
226 fiz0 = _mm_add_pd(fiz0,tz);
228 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
230 /* Inner loop uses 58 flops */
237 j_coord_offsetA = DIM*jnrA;
239 /* load j atom coordinates */
240 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
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);
248 /* Calculate squared distance and things based on it */
249 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
251 rinv00 = gmx_mm_invsqrt_pd(rsq00);
253 /* Load parameters for j particles */
254 jq0 = _mm_load_sd(charge+jnrA+0);
255 isaj0 = _mm_load_sd(invsqrta+jnrA+0);
257 /**************************
258 * CALCULATE INTERACTIONS *
259 **************************/
261 r00 = _mm_mul_pd(rsq00,rinv00);
263 /* Compute parameters for interactions between i and j atoms */
264 qq00 = _mm_mul_pd(iq0,jq0);
266 /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
267 isaprod = _mm_mul_pd(isai0,isaj0);
268 gbqqfactor = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
269 gbscale = _mm_mul_pd(isaprod,gbtabscale);
271 /* Calculate generalized born table index - this is a separate table from the normal one,
272 * but we use the same procedure by multiplying r with scale and truncating to integer.
274 rt = _mm_mul_pd(r00,gbscale);
275 gbitab = _mm_cvttpd_epi32(rt);
276 gbeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
277 gbitab = _mm_slli_epi32(gbitab,2);
279 Y = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
280 F = _mm_setzero_pd();
281 GMX_MM_TRANSPOSE2_PD(Y,F);
282 G = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
283 H = _mm_setzero_pd();
284 GMX_MM_TRANSPOSE2_PD(G,H);
285 Heps = _mm_mul_pd(gbeps,H);
286 Fp = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
287 VV = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
288 vgb = _mm_mul_pd(gbqqfactor,VV);
290 FF = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
291 fgb = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
292 dvdatmp = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
293 dvdatmp = _mm_unpacklo_pd(dvdatmp,_mm_setzero_pd());
294 dvdasum = _mm_add_pd(dvdasum,dvdatmp);
295 gmx_mm_increment_1real_pd(dvda+jnrA,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
296 velec = _mm_mul_pd(qq00,rinv00);
297 felec = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
299 /* Update potential sum for this i atom from the interaction with this j atom. */
300 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
301 velecsum = _mm_add_pd(velecsum,velec);
302 vgb = _mm_unpacklo_pd(vgb,_mm_setzero_pd());
303 vgbsum = _mm_add_pd(vgbsum,vgb);
307 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
309 /* Calculate temporary vectorial force */
310 tx = _mm_mul_pd(fscal,dx00);
311 ty = _mm_mul_pd(fscal,dy00);
312 tz = _mm_mul_pd(fscal,dz00);
314 /* Update vectorial force */
315 fix0 = _mm_add_pd(fix0,tx);
316 fiy0 = _mm_add_pd(fiy0,ty);
317 fiz0 = _mm_add_pd(fiz0,tz);
319 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
321 /* Inner loop uses 58 flops */
324 /* End of innermost loop */
326 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
327 f+i_coord_offset,fshift+i_shift_offset);
330 /* Update potential energies */
331 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
332 gmx_mm_update_1pot_pd(vgbsum,kernel_data->energygrp_polarization+ggid);
333 dvdasum = _mm_mul_pd(dvdasum, _mm_mul_pd(isai0,isai0));
334 gmx_mm_update_1pot_pd(dvdasum,dvda+inr);
336 /* Increment number of inner iterations */
337 inneriter += j_index_end - j_index_start;
339 /* Outer loop uses 9 flops */
342 /* Increment number of outer iterations */
345 /* Update outer/inner flops */
347 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*9 + inneriter*58);
350 * Gromacs nonbonded kernel: nb_kernel_ElecGB_VdwNone_GeomP1P1_F_sse4_1_double
351 * Electrostatics interaction: GeneralizedBorn
352 * VdW interaction: None
353 * Geometry: Particle-Particle
354 * Calculate force/pot: Force
357 nb_kernel_ElecGB_VdwNone_GeomP1P1_F_sse4_1_double
358 (t_nblist * gmx_restrict nlist,
359 rvec * gmx_restrict xx,
360 rvec * gmx_restrict ff,
361 t_forcerec * gmx_restrict fr,
362 t_mdatoms * gmx_restrict mdatoms,
363 nb_kernel_data_t * gmx_restrict kernel_data,
364 t_nrnb * gmx_restrict nrnb)
366 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
367 * just 0 for non-waters.
368 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
369 * jnr indices corresponding to data put in the four positions in the SIMD register.
371 int i_shift_offset,i_coord_offset,outeriter,inneriter;
372 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
374 int j_coord_offsetA,j_coord_offsetB;
375 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
377 real *shiftvec,*fshift,*x,*f;
378 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
380 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
381 int vdwjidx0A,vdwjidx0B;
382 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
383 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
384 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
387 __m128d vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,dvdatmp;
388 __m128d minushalf = _mm_set1_pd(-0.5);
389 real *invsqrta,*dvda,*gbtab;
391 __m128i ifour = _mm_set1_epi32(4);
392 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
394 __m128d dummy_mask,cutoff_mask;
395 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
396 __m128d one = _mm_set1_pd(1.0);
397 __m128d two = _mm_set1_pd(2.0);
403 jindex = nlist->jindex;
405 shiftidx = nlist->shift;
407 shiftvec = fr->shift_vec[0];
408 fshift = fr->fshift[0];
409 facel = _mm_set1_pd(fr->epsfac);
410 charge = mdatoms->chargeA;
412 invsqrta = fr->invsqrta;
414 gbtabscale = _mm_set1_pd(fr->gbtab.scale);
415 gbtab = fr->gbtab.data;
416 gbinvepsdiff = _mm_set1_pd((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
418 /* Avoid stupid compiler warnings */
426 /* Start outer loop over neighborlists */
427 for(iidx=0; iidx<nri; iidx++)
429 /* Load shift vector for this list */
430 i_shift_offset = DIM*shiftidx[iidx];
432 /* Load limits for loop over neighbors */
433 j_index_start = jindex[iidx];
434 j_index_end = jindex[iidx+1];
436 /* Get outer coordinate index */
438 i_coord_offset = DIM*inr;
440 /* Load i particle coords and add shift vector */
441 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
443 fix0 = _mm_setzero_pd();
444 fiy0 = _mm_setzero_pd();
445 fiz0 = _mm_setzero_pd();
447 /* Load parameters for i particles */
448 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
449 isai0 = _mm_load1_pd(invsqrta+inr+0);
451 dvdasum = _mm_setzero_pd();
453 /* Start inner kernel loop */
454 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
457 /* Get j neighbor index, and coordinate index */
460 j_coord_offsetA = DIM*jnrA;
461 j_coord_offsetB = DIM*jnrB;
463 /* load j atom coordinates */
464 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
467 /* Calculate displacement vector */
468 dx00 = _mm_sub_pd(ix0,jx0);
469 dy00 = _mm_sub_pd(iy0,jy0);
470 dz00 = _mm_sub_pd(iz0,jz0);
472 /* Calculate squared distance and things based on it */
473 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
475 rinv00 = gmx_mm_invsqrt_pd(rsq00);
477 /* Load parameters for j particles */
478 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
479 isaj0 = gmx_mm_load_2real_swizzle_pd(invsqrta+jnrA+0,invsqrta+jnrB+0);
481 /**************************
482 * CALCULATE INTERACTIONS *
483 **************************/
485 r00 = _mm_mul_pd(rsq00,rinv00);
487 /* Compute parameters for interactions between i and j atoms */
488 qq00 = _mm_mul_pd(iq0,jq0);
490 /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
491 isaprod = _mm_mul_pd(isai0,isaj0);
492 gbqqfactor = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
493 gbscale = _mm_mul_pd(isaprod,gbtabscale);
495 /* Calculate generalized born table index - this is a separate table from the normal one,
496 * but we use the same procedure by multiplying r with scale and truncating to integer.
498 rt = _mm_mul_pd(r00,gbscale);
499 gbitab = _mm_cvttpd_epi32(rt);
500 gbeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
501 gbitab = _mm_slli_epi32(gbitab,2);
503 Y = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
504 F = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) );
505 GMX_MM_TRANSPOSE2_PD(Y,F);
506 G = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
507 H = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) +2);
508 GMX_MM_TRANSPOSE2_PD(G,H);
509 Heps = _mm_mul_pd(gbeps,H);
510 Fp = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
511 VV = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
512 vgb = _mm_mul_pd(gbqqfactor,VV);
514 FF = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
515 fgb = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
516 dvdatmp = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
517 dvdasum = _mm_add_pd(dvdasum,dvdatmp);
518 gmx_mm_increment_2real_swizzle_pd(dvda+jnrA,dvda+jnrB,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
519 velec = _mm_mul_pd(qq00,rinv00);
520 felec = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
524 /* Calculate temporary vectorial force */
525 tx = _mm_mul_pd(fscal,dx00);
526 ty = _mm_mul_pd(fscal,dy00);
527 tz = _mm_mul_pd(fscal,dz00);
529 /* Update vectorial force */
530 fix0 = _mm_add_pd(fix0,tx);
531 fiy0 = _mm_add_pd(fiy0,ty);
532 fiz0 = _mm_add_pd(fiz0,tz);
534 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
536 /* Inner loop uses 56 flops */
543 j_coord_offsetA = DIM*jnrA;
545 /* load j atom coordinates */
546 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
549 /* Calculate displacement vector */
550 dx00 = _mm_sub_pd(ix0,jx0);
551 dy00 = _mm_sub_pd(iy0,jy0);
552 dz00 = _mm_sub_pd(iz0,jz0);
554 /* Calculate squared distance and things based on it */
555 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
557 rinv00 = gmx_mm_invsqrt_pd(rsq00);
559 /* Load parameters for j particles */
560 jq0 = _mm_load_sd(charge+jnrA+0);
561 isaj0 = _mm_load_sd(invsqrta+jnrA+0);
563 /**************************
564 * CALCULATE INTERACTIONS *
565 **************************/
567 r00 = _mm_mul_pd(rsq00,rinv00);
569 /* Compute parameters for interactions between i and j atoms */
570 qq00 = _mm_mul_pd(iq0,jq0);
572 /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
573 isaprod = _mm_mul_pd(isai0,isaj0);
574 gbqqfactor = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
575 gbscale = _mm_mul_pd(isaprod,gbtabscale);
577 /* Calculate generalized born table index - this is a separate table from the normal one,
578 * but we use the same procedure by multiplying r with scale and truncating to integer.
580 rt = _mm_mul_pd(r00,gbscale);
581 gbitab = _mm_cvttpd_epi32(rt);
582 gbeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
583 gbitab = _mm_slli_epi32(gbitab,2);
585 Y = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
586 F = _mm_setzero_pd();
587 GMX_MM_TRANSPOSE2_PD(Y,F);
588 G = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
589 H = _mm_setzero_pd();
590 GMX_MM_TRANSPOSE2_PD(G,H);
591 Heps = _mm_mul_pd(gbeps,H);
592 Fp = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
593 VV = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
594 vgb = _mm_mul_pd(gbqqfactor,VV);
596 FF = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
597 fgb = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
598 dvdatmp = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
599 dvdatmp = _mm_unpacklo_pd(dvdatmp,_mm_setzero_pd());
600 dvdasum = _mm_add_pd(dvdasum,dvdatmp);
601 gmx_mm_increment_1real_pd(dvda+jnrA,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
602 velec = _mm_mul_pd(qq00,rinv00);
603 felec = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
607 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
609 /* Calculate temporary vectorial force */
610 tx = _mm_mul_pd(fscal,dx00);
611 ty = _mm_mul_pd(fscal,dy00);
612 tz = _mm_mul_pd(fscal,dz00);
614 /* Update vectorial force */
615 fix0 = _mm_add_pd(fix0,tx);
616 fiy0 = _mm_add_pd(fiy0,ty);
617 fiz0 = _mm_add_pd(fiz0,tz);
619 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
621 /* Inner loop uses 56 flops */
624 /* End of innermost loop */
626 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
627 f+i_coord_offset,fshift+i_shift_offset);
629 dvdasum = _mm_mul_pd(dvdasum, _mm_mul_pd(isai0,isai0));
630 gmx_mm_update_1pot_pd(dvdasum,dvda+inr);
632 /* Increment number of inner iterations */
633 inneriter += j_index_end - j_index_start;
635 /* Outer loop uses 7 flops */
638 /* Increment number of outer iterations */
641 /* Update outer/inner flops */
643 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*56);