2 * Note: this file was generated by the Gromacs sse2_single 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_sse2_single.h"
34 #include "kernelutil_x86_sse2_single.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwLJ_GeomP1P1_VF_sse2_single
38 * Electrostatics interaction: ReactionField
39 * VdW interaction: LennardJones
40 * Geometry: Particle-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecRF_VdwLJ_GeomP1P1_VF_sse2_single
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,C,D refer to j loop unrolling done with SSE, e.g. for the four 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;
60 int jnrA,jnrB,jnrC,jnrD;
61 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
62 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
63 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
65 real *shiftvec,*fshift,*x,*f;
66 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
68 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
70 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
71 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
72 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
73 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
74 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
77 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
80 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
81 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
82 __m128 dummy_mask,cutoff_mask;
83 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
84 __m128 one = _mm_set1_ps(1.0);
85 __m128 two = _mm_set1_ps(2.0);
91 jindex = nlist->jindex;
93 shiftidx = nlist->shift;
95 shiftvec = fr->shift_vec[0];
96 fshift = fr->fshift[0];
97 facel = _mm_set1_ps(fr->epsfac);
98 charge = mdatoms->chargeA;
99 krf = _mm_set1_ps(fr->ic->k_rf);
100 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
101 crf = _mm_set1_ps(fr->ic->c_rf);
102 nvdwtype = fr->ntype;
104 vdwtype = mdatoms->typeA;
106 /* Avoid stupid compiler warnings */
107 jnrA = jnrB = jnrC = jnrD = 0;
116 for(iidx=0;iidx<4*DIM;iidx++)
121 /* Start outer loop over neighborlists */
122 for(iidx=0; iidx<nri; iidx++)
124 /* Load shift vector for this list */
125 i_shift_offset = DIM*shiftidx[iidx];
127 /* Load limits for loop over neighbors */
128 j_index_start = jindex[iidx];
129 j_index_end = jindex[iidx+1];
131 /* Get outer coordinate index */
133 i_coord_offset = DIM*inr;
135 /* Load i particle coords and add shift vector */
136 gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
138 fix0 = _mm_setzero_ps();
139 fiy0 = _mm_setzero_ps();
140 fiz0 = _mm_setzero_ps();
142 /* Load parameters for i particles */
143 iq0 = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
144 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
146 /* Reset potential sums */
147 velecsum = _mm_setzero_ps();
148 vvdwsum = _mm_setzero_ps();
150 /* Start inner kernel loop */
151 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
154 /* Get j neighbor index, and coordinate index */
159 j_coord_offsetA = DIM*jnrA;
160 j_coord_offsetB = DIM*jnrB;
161 j_coord_offsetC = DIM*jnrC;
162 j_coord_offsetD = DIM*jnrD;
164 /* load j atom coordinates */
165 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
166 x+j_coord_offsetC,x+j_coord_offsetD,
169 /* Calculate displacement vector */
170 dx00 = _mm_sub_ps(ix0,jx0);
171 dy00 = _mm_sub_ps(iy0,jy0);
172 dz00 = _mm_sub_ps(iz0,jz0);
174 /* Calculate squared distance and things based on it */
175 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
177 rinv00 = gmx_mm_invsqrt_ps(rsq00);
179 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
181 /* Load parameters for j particles */
182 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
183 charge+jnrC+0,charge+jnrD+0);
184 vdwjidx0A = 2*vdwtype[jnrA+0];
185 vdwjidx0B = 2*vdwtype[jnrB+0];
186 vdwjidx0C = 2*vdwtype[jnrC+0];
187 vdwjidx0D = 2*vdwtype[jnrD+0];
189 /**************************
190 * CALCULATE INTERACTIONS *
191 **************************/
193 /* Compute parameters for interactions between i and j atoms */
194 qq00 = _mm_mul_ps(iq0,jq0);
195 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
196 vdwparam+vdwioffset0+vdwjidx0B,
197 vdwparam+vdwioffset0+vdwjidx0C,
198 vdwparam+vdwioffset0+vdwjidx0D,
201 /* REACTION-FIELD ELECTROSTATICS */
202 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
203 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
205 /* LENNARD-JONES DISPERSION/REPULSION */
207 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
208 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
209 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
210 vvdw = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
211 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
213 /* Update potential sum for this i atom from the interaction with this j atom. */
214 velecsum = _mm_add_ps(velecsum,velec);
215 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
217 fscal = _mm_add_ps(felec,fvdw);
219 /* Calculate temporary vectorial force */
220 tx = _mm_mul_ps(fscal,dx00);
221 ty = _mm_mul_ps(fscal,dy00);
222 tz = _mm_mul_ps(fscal,dz00);
224 /* Update vectorial force */
225 fix0 = _mm_add_ps(fix0,tx);
226 fiy0 = _mm_add_ps(fiy0,ty);
227 fiz0 = _mm_add_ps(fiz0,tz);
229 fjptrA = f+j_coord_offsetA;
230 fjptrB = f+j_coord_offsetB;
231 fjptrC = f+j_coord_offsetC;
232 fjptrD = f+j_coord_offsetD;
233 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
235 /* Inner loop uses 44 flops */
241 /* Get j neighbor index, and coordinate index */
242 jnrlistA = jjnr[jidx];
243 jnrlistB = jjnr[jidx+1];
244 jnrlistC = jjnr[jidx+2];
245 jnrlistD = jjnr[jidx+3];
246 /* Sign of each element will be negative for non-real atoms.
247 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
248 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
250 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
251 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
252 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
253 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
254 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
255 j_coord_offsetA = DIM*jnrA;
256 j_coord_offsetB = DIM*jnrB;
257 j_coord_offsetC = DIM*jnrC;
258 j_coord_offsetD = DIM*jnrD;
260 /* load j atom coordinates */
261 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
262 x+j_coord_offsetC,x+j_coord_offsetD,
265 /* Calculate displacement vector */
266 dx00 = _mm_sub_ps(ix0,jx0);
267 dy00 = _mm_sub_ps(iy0,jy0);
268 dz00 = _mm_sub_ps(iz0,jz0);
270 /* Calculate squared distance and things based on it */
271 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
273 rinv00 = gmx_mm_invsqrt_ps(rsq00);
275 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
277 /* Load parameters for j particles */
278 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
279 charge+jnrC+0,charge+jnrD+0);
280 vdwjidx0A = 2*vdwtype[jnrA+0];
281 vdwjidx0B = 2*vdwtype[jnrB+0];
282 vdwjidx0C = 2*vdwtype[jnrC+0];
283 vdwjidx0D = 2*vdwtype[jnrD+0];
285 /**************************
286 * CALCULATE INTERACTIONS *
287 **************************/
289 /* Compute parameters for interactions between i and j atoms */
290 qq00 = _mm_mul_ps(iq0,jq0);
291 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
292 vdwparam+vdwioffset0+vdwjidx0B,
293 vdwparam+vdwioffset0+vdwjidx0C,
294 vdwparam+vdwioffset0+vdwjidx0D,
297 /* REACTION-FIELD ELECTROSTATICS */
298 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
299 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
301 /* LENNARD-JONES DISPERSION/REPULSION */
303 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
304 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
305 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
306 vvdw = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
307 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
309 /* Update potential sum for this i atom from the interaction with this j atom. */
310 velec = _mm_andnot_ps(dummy_mask,velec);
311 velecsum = _mm_add_ps(velecsum,velec);
312 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
313 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
315 fscal = _mm_add_ps(felec,fvdw);
317 fscal = _mm_andnot_ps(dummy_mask,fscal);
319 /* Calculate temporary vectorial force */
320 tx = _mm_mul_ps(fscal,dx00);
321 ty = _mm_mul_ps(fscal,dy00);
322 tz = _mm_mul_ps(fscal,dz00);
324 /* Update vectorial force */
325 fix0 = _mm_add_ps(fix0,tx);
326 fiy0 = _mm_add_ps(fiy0,ty);
327 fiz0 = _mm_add_ps(fiz0,tz);
329 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
330 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
331 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
332 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
333 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
335 /* Inner loop uses 44 flops */
338 /* End of innermost loop */
340 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
341 f+i_coord_offset,fshift+i_shift_offset);
344 /* Update potential energies */
345 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
346 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
348 /* Increment number of inner iterations */
349 inneriter += j_index_end - j_index_start;
351 /* Outer loop uses 9 flops */
354 /* Increment number of outer iterations */
357 /* Update outer/inner flops */
359 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*44);
362 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_sse2_single
363 * Electrostatics interaction: ReactionField
364 * VdW interaction: LennardJones
365 * Geometry: Particle-Particle
366 * Calculate force/pot: Force
369 nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_sse2_single
370 (t_nblist * gmx_restrict nlist,
371 rvec * gmx_restrict xx,
372 rvec * gmx_restrict ff,
373 t_forcerec * gmx_restrict fr,
374 t_mdatoms * gmx_restrict mdatoms,
375 nb_kernel_data_t * gmx_restrict kernel_data,
376 t_nrnb * gmx_restrict nrnb)
378 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
379 * just 0 for non-waters.
380 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
381 * jnr indices corresponding to data put in the four positions in the SIMD register.
383 int i_shift_offset,i_coord_offset,outeriter,inneriter;
384 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
385 int jnrA,jnrB,jnrC,jnrD;
386 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
387 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
388 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
390 real *shiftvec,*fshift,*x,*f;
391 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
393 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
395 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
396 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
397 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
398 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
399 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
402 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
405 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
406 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
407 __m128 dummy_mask,cutoff_mask;
408 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
409 __m128 one = _mm_set1_ps(1.0);
410 __m128 two = _mm_set1_ps(2.0);
416 jindex = nlist->jindex;
418 shiftidx = nlist->shift;
420 shiftvec = fr->shift_vec[0];
421 fshift = fr->fshift[0];
422 facel = _mm_set1_ps(fr->epsfac);
423 charge = mdatoms->chargeA;
424 krf = _mm_set1_ps(fr->ic->k_rf);
425 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
426 crf = _mm_set1_ps(fr->ic->c_rf);
427 nvdwtype = fr->ntype;
429 vdwtype = mdatoms->typeA;
431 /* Avoid stupid compiler warnings */
432 jnrA = jnrB = jnrC = jnrD = 0;
441 for(iidx=0;iidx<4*DIM;iidx++)
446 /* Start outer loop over neighborlists */
447 for(iidx=0; iidx<nri; iidx++)
449 /* Load shift vector for this list */
450 i_shift_offset = DIM*shiftidx[iidx];
452 /* Load limits for loop over neighbors */
453 j_index_start = jindex[iidx];
454 j_index_end = jindex[iidx+1];
456 /* Get outer coordinate index */
458 i_coord_offset = DIM*inr;
460 /* Load i particle coords and add shift vector */
461 gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
463 fix0 = _mm_setzero_ps();
464 fiy0 = _mm_setzero_ps();
465 fiz0 = _mm_setzero_ps();
467 /* Load parameters for i particles */
468 iq0 = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
469 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
471 /* Start inner kernel loop */
472 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
475 /* Get j neighbor index, and coordinate index */
480 j_coord_offsetA = DIM*jnrA;
481 j_coord_offsetB = DIM*jnrB;
482 j_coord_offsetC = DIM*jnrC;
483 j_coord_offsetD = DIM*jnrD;
485 /* load j atom coordinates */
486 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
487 x+j_coord_offsetC,x+j_coord_offsetD,
490 /* Calculate displacement vector */
491 dx00 = _mm_sub_ps(ix0,jx0);
492 dy00 = _mm_sub_ps(iy0,jy0);
493 dz00 = _mm_sub_ps(iz0,jz0);
495 /* Calculate squared distance and things based on it */
496 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
498 rinv00 = gmx_mm_invsqrt_ps(rsq00);
500 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
502 /* Load parameters for j particles */
503 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
504 charge+jnrC+0,charge+jnrD+0);
505 vdwjidx0A = 2*vdwtype[jnrA+0];
506 vdwjidx0B = 2*vdwtype[jnrB+0];
507 vdwjidx0C = 2*vdwtype[jnrC+0];
508 vdwjidx0D = 2*vdwtype[jnrD+0];
510 /**************************
511 * CALCULATE INTERACTIONS *
512 **************************/
514 /* Compute parameters for interactions between i and j atoms */
515 qq00 = _mm_mul_ps(iq0,jq0);
516 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
517 vdwparam+vdwioffset0+vdwjidx0B,
518 vdwparam+vdwioffset0+vdwjidx0C,
519 vdwparam+vdwioffset0+vdwjidx0D,
522 /* REACTION-FIELD ELECTROSTATICS */
523 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
525 /* LENNARD-JONES DISPERSION/REPULSION */
527 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
528 fvdw = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
530 fscal = _mm_add_ps(felec,fvdw);
532 /* Calculate temporary vectorial force */
533 tx = _mm_mul_ps(fscal,dx00);
534 ty = _mm_mul_ps(fscal,dy00);
535 tz = _mm_mul_ps(fscal,dz00);
537 /* Update vectorial force */
538 fix0 = _mm_add_ps(fix0,tx);
539 fiy0 = _mm_add_ps(fiy0,ty);
540 fiz0 = _mm_add_ps(fiz0,tz);
542 fjptrA = f+j_coord_offsetA;
543 fjptrB = f+j_coord_offsetB;
544 fjptrC = f+j_coord_offsetC;
545 fjptrD = f+j_coord_offsetD;
546 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
548 /* Inner loop uses 34 flops */
554 /* Get j neighbor index, and coordinate index */
555 jnrlistA = jjnr[jidx];
556 jnrlistB = jjnr[jidx+1];
557 jnrlistC = jjnr[jidx+2];
558 jnrlistD = jjnr[jidx+3];
559 /* Sign of each element will be negative for non-real atoms.
560 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
561 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
563 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
564 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
565 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
566 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
567 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
568 j_coord_offsetA = DIM*jnrA;
569 j_coord_offsetB = DIM*jnrB;
570 j_coord_offsetC = DIM*jnrC;
571 j_coord_offsetD = DIM*jnrD;
573 /* load j atom coordinates */
574 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
575 x+j_coord_offsetC,x+j_coord_offsetD,
578 /* Calculate displacement vector */
579 dx00 = _mm_sub_ps(ix0,jx0);
580 dy00 = _mm_sub_ps(iy0,jy0);
581 dz00 = _mm_sub_ps(iz0,jz0);
583 /* Calculate squared distance and things based on it */
584 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
586 rinv00 = gmx_mm_invsqrt_ps(rsq00);
588 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
590 /* Load parameters for j particles */
591 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
592 charge+jnrC+0,charge+jnrD+0);
593 vdwjidx0A = 2*vdwtype[jnrA+0];
594 vdwjidx0B = 2*vdwtype[jnrB+0];
595 vdwjidx0C = 2*vdwtype[jnrC+0];
596 vdwjidx0D = 2*vdwtype[jnrD+0];
598 /**************************
599 * CALCULATE INTERACTIONS *
600 **************************/
602 /* Compute parameters for interactions between i and j atoms */
603 qq00 = _mm_mul_ps(iq0,jq0);
604 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
605 vdwparam+vdwioffset0+vdwjidx0B,
606 vdwparam+vdwioffset0+vdwjidx0C,
607 vdwparam+vdwioffset0+vdwjidx0D,
610 /* REACTION-FIELD ELECTROSTATICS */
611 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
613 /* LENNARD-JONES DISPERSION/REPULSION */
615 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
616 fvdw = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
618 fscal = _mm_add_ps(felec,fvdw);
620 fscal = _mm_andnot_ps(dummy_mask,fscal);
622 /* Calculate temporary vectorial force */
623 tx = _mm_mul_ps(fscal,dx00);
624 ty = _mm_mul_ps(fscal,dy00);
625 tz = _mm_mul_ps(fscal,dz00);
627 /* Update vectorial force */
628 fix0 = _mm_add_ps(fix0,tx);
629 fiy0 = _mm_add_ps(fiy0,ty);
630 fiz0 = _mm_add_ps(fiz0,tz);
632 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
633 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
634 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
635 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
636 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
638 /* Inner loop uses 34 flops */
641 /* End of innermost loop */
643 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
644 f+i_coord_offset,fshift+i_shift_offset);
646 /* Increment number of inner iterations */
647 inneriter += j_index_end - j_index_start;
649 /* Outer loop uses 7 flops */
652 /* Increment number of outer iterations */
655 /* Update outer/inner flops */
657 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*34);