2 * Note: this file was generated by the Gromacs avx_128_fma_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_avx_128_fma_single.h"
34 #include "kernelutil_x86_avx_128_fma_single.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwLJ_GeomP1P1_VF_avx_128_fma_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_avx_128_fma_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 AVX_128, 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 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_macc_ps(krf,rsq00,rinv00),crf));
203 felec = _mm_mul_ps(qq00,_mm_msub_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_msub_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 /* Update vectorial force */
220 fix0 = _mm_macc_ps(dx00,fscal,fix0);
221 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
222 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
224 fjptrA = f+j_coord_offsetA;
225 fjptrB = f+j_coord_offsetB;
226 fjptrC = f+j_coord_offsetC;
227 fjptrD = f+j_coord_offsetD;
228 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
229 _mm_mul_ps(dx00,fscal),
230 _mm_mul_ps(dy00,fscal),
231 _mm_mul_ps(dz00,fscal));
233 /* Inner loop uses 47 flops */
239 /* Get j neighbor index, and coordinate index */
240 jnrlistA = jjnr[jidx];
241 jnrlistB = jjnr[jidx+1];
242 jnrlistC = jjnr[jidx+2];
243 jnrlistD = jjnr[jidx+3];
244 /* Sign of each element will be negative for non-real atoms.
245 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
246 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
248 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
249 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
250 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
251 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
252 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
253 j_coord_offsetA = DIM*jnrA;
254 j_coord_offsetB = DIM*jnrB;
255 j_coord_offsetC = DIM*jnrC;
256 j_coord_offsetD = DIM*jnrD;
258 /* load j atom coordinates */
259 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
260 x+j_coord_offsetC,x+j_coord_offsetD,
263 /* Calculate displacement vector */
264 dx00 = _mm_sub_ps(ix0,jx0);
265 dy00 = _mm_sub_ps(iy0,jy0);
266 dz00 = _mm_sub_ps(iz0,jz0);
268 /* Calculate squared distance and things based on it */
269 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
271 rinv00 = gmx_mm_invsqrt_ps(rsq00);
273 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
275 /* Load parameters for j particles */
276 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
277 charge+jnrC+0,charge+jnrD+0);
278 vdwjidx0A = 2*vdwtype[jnrA+0];
279 vdwjidx0B = 2*vdwtype[jnrB+0];
280 vdwjidx0C = 2*vdwtype[jnrC+0];
281 vdwjidx0D = 2*vdwtype[jnrD+0];
283 /**************************
284 * CALCULATE INTERACTIONS *
285 **************************/
287 /* Compute parameters for interactions between i and j atoms */
288 qq00 = _mm_mul_ps(iq0,jq0);
289 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
290 vdwparam+vdwioffset0+vdwjidx0B,
291 vdwparam+vdwioffset0+vdwjidx0C,
292 vdwparam+vdwioffset0+vdwjidx0D,
295 /* REACTION-FIELD ELECTROSTATICS */
296 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
297 felec = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
299 /* LENNARD-JONES DISPERSION/REPULSION */
301 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
302 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
303 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
304 vvdw = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
305 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
307 /* Update potential sum for this i atom from the interaction with this j atom. */
308 velec = _mm_andnot_ps(dummy_mask,velec);
309 velecsum = _mm_add_ps(velecsum,velec);
310 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
311 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
313 fscal = _mm_add_ps(felec,fvdw);
315 fscal = _mm_andnot_ps(dummy_mask,fscal);
317 /* Update vectorial force */
318 fix0 = _mm_macc_ps(dx00,fscal,fix0);
319 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
320 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
322 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
323 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
324 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
325 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
326 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
327 _mm_mul_ps(dx00,fscal),
328 _mm_mul_ps(dy00,fscal),
329 _mm_mul_ps(dz00,fscal));
331 /* Inner loop uses 47 flops */
334 /* End of innermost loop */
336 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
337 f+i_coord_offset,fshift+i_shift_offset);
340 /* Update potential energies */
341 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
342 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
344 /* Increment number of inner iterations */
345 inneriter += j_index_end - j_index_start;
347 /* Outer loop uses 9 flops */
350 /* Increment number of outer iterations */
353 /* Update outer/inner flops */
355 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*47);
358 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_avx_128_fma_single
359 * Electrostatics interaction: ReactionField
360 * VdW interaction: LennardJones
361 * Geometry: Particle-Particle
362 * Calculate force/pot: Force
365 nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_avx_128_fma_single
366 (t_nblist * gmx_restrict nlist,
367 rvec * gmx_restrict xx,
368 rvec * gmx_restrict ff,
369 t_forcerec * gmx_restrict fr,
370 t_mdatoms * gmx_restrict mdatoms,
371 nb_kernel_data_t * gmx_restrict kernel_data,
372 t_nrnb * gmx_restrict nrnb)
374 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
375 * just 0 for non-waters.
376 * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
377 * jnr indices corresponding to data put in the four positions in the SIMD register.
379 int i_shift_offset,i_coord_offset,outeriter,inneriter;
380 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
381 int jnrA,jnrB,jnrC,jnrD;
382 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
383 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
384 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
386 real *shiftvec,*fshift,*x,*f;
387 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
389 __m128 fscal,rcutoff,rcutoff2,jidxall;
391 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
392 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
393 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
394 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
395 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
398 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
401 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
402 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
403 __m128 dummy_mask,cutoff_mask;
404 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
405 __m128 one = _mm_set1_ps(1.0);
406 __m128 two = _mm_set1_ps(2.0);
412 jindex = nlist->jindex;
414 shiftidx = nlist->shift;
416 shiftvec = fr->shift_vec[0];
417 fshift = fr->fshift[0];
418 facel = _mm_set1_ps(fr->epsfac);
419 charge = mdatoms->chargeA;
420 krf = _mm_set1_ps(fr->ic->k_rf);
421 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
422 crf = _mm_set1_ps(fr->ic->c_rf);
423 nvdwtype = fr->ntype;
425 vdwtype = mdatoms->typeA;
427 /* Avoid stupid compiler warnings */
428 jnrA = jnrB = jnrC = jnrD = 0;
437 for(iidx=0;iidx<4*DIM;iidx++)
442 /* Start outer loop over neighborlists */
443 for(iidx=0; iidx<nri; iidx++)
445 /* Load shift vector for this list */
446 i_shift_offset = DIM*shiftidx[iidx];
448 /* Load limits for loop over neighbors */
449 j_index_start = jindex[iidx];
450 j_index_end = jindex[iidx+1];
452 /* Get outer coordinate index */
454 i_coord_offset = DIM*inr;
456 /* Load i particle coords and add shift vector */
457 gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
459 fix0 = _mm_setzero_ps();
460 fiy0 = _mm_setzero_ps();
461 fiz0 = _mm_setzero_ps();
463 /* Load parameters for i particles */
464 iq0 = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
465 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
467 /* Start inner kernel loop */
468 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
471 /* Get j neighbor index, and coordinate index */
476 j_coord_offsetA = DIM*jnrA;
477 j_coord_offsetB = DIM*jnrB;
478 j_coord_offsetC = DIM*jnrC;
479 j_coord_offsetD = DIM*jnrD;
481 /* load j atom coordinates */
482 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
483 x+j_coord_offsetC,x+j_coord_offsetD,
486 /* Calculate displacement vector */
487 dx00 = _mm_sub_ps(ix0,jx0);
488 dy00 = _mm_sub_ps(iy0,jy0);
489 dz00 = _mm_sub_ps(iz0,jz0);
491 /* Calculate squared distance and things based on it */
492 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
494 rinv00 = gmx_mm_invsqrt_ps(rsq00);
496 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
498 /* Load parameters for j particles */
499 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
500 charge+jnrC+0,charge+jnrD+0);
501 vdwjidx0A = 2*vdwtype[jnrA+0];
502 vdwjidx0B = 2*vdwtype[jnrB+0];
503 vdwjidx0C = 2*vdwtype[jnrC+0];
504 vdwjidx0D = 2*vdwtype[jnrD+0];
506 /**************************
507 * CALCULATE INTERACTIONS *
508 **************************/
510 /* Compute parameters for interactions between i and j atoms */
511 qq00 = _mm_mul_ps(iq0,jq0);
512 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
513 vdwparam+vdwioffset0+vdwjidx0B,
514 vdwparam+vdwioffset0+vdwjidx0C,
515 vdwparam+vdwioffset0+vdwjidx0D,
518 /* REACTION-FIELD ELECTROSTATICS */
519 felec = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
521 /* LENNARD-JONES DISPERSION/REPULSION */
523 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
524 fvdw = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
526 fscal = _mm_add_ps(felec,fvdw);
528 /* Update vectorial force */
529 fix0 = _mm_macc_ps(dx00,fscal,fix0);
530 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
531 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
533 fjptrA = f+j_coord_offsetA;
534 fjptrB = f+j_coord_offsetB;
535 fjptrC = f+j_coord_offsetC;
536 fjptrD = f+j_coord_offsetD;
537 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
538 _mm_mul_ps(dx00,fscal),
539 _mm_mul_ps(dy00,fscal),
540 _mm_mul_ps(dz00,fscal));
542 /* Inner loop uses 37 flops */
548 /* Get j neighbor index, and coordinate index */
549 jnrlistA = jjnr[jidx];
550 jnrlistB = jjnr[jidx+1];
551 jnrlistC = jjnr[jidx+2];
552 jnrlistD = jjnr[jidx+3];
553 /* Sign of each element will be negative for non-real atoms.
554 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
555 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
557 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
558 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
559 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
560 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
561 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
562 j_coord_offsetA = DIM*jnrA;
563 j_coord_offsetB = DIM*jnrB;
564 j_coord_offsetC = DIM*jnrC;
565 j_coord_offsetD = DIM*jnrD;
567 /* load j atom coordinates */
568 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
569 x+j_coord_offsetC,x+j_coord_offsetD,
572 /* Calculate displacement vector */
573 dx00 = _mm_sub_ps(ix0,jx0);
574 dy00 = _mm_sub_ps(iy0,jy0);
575 dz00 = _mm_sub_ps(iz0,jz0);
577 /* Calculate squared distance and things based on it */
578 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
580 rinv00 = gmx_mm_invsqrt_ps(rsq00);
582 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
584 /* Load parameters for j particles */
585 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
586 charge+jnrC+0,charge+jnrD+0);
587 vdwjidx0A = 2*vdwtype[jnrA+0];
588 vdwjidx0B = 2*vdwtype[jnrB+0];
589 vdwjidx0C = 2*vdwtype[jnrC+0];
590 vdwjidx0D = 2*vdwtype[jnrD+0];
592 /**************************
593 * CALCULATE INTERACTIONS *
594 **************************/
596 /* Compute parameters for interactions between i and j atoms */
597 qq00 = _mm_mul_ps(iq0,jq0);
598 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
599 vdwparam+vdwioffset0+vdwjidx0B,
600 vdwparam+vdwioffset0+vdwjidx0C,
601 vdwparam+vdwioffset0+vdwjidx0D,
604 /* REACTION-FIELD ELECTROSTATICS */
605 felec = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
607 /* LENNARD-JONES DISPERSION/REPULSION */
609 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
610 fvdw = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
612 fscal = _mm_add_ps(felec,fvdw);
614 fscal = _mm_andnot_ps(dummy_mask,fscal);
616 /* Update vectorial force */
617 fix0 = _mm_macc_ps(dx00,fscal,fix0);
618 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
619 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
621 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
622 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
623 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
624 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
625 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
626 _mm_mul_ps(dx00,fscal),
627 _mm_mul_ps(dy00,fscal),
628 _mm_mul_ps(dz00,fscal));
630 /* Inner loop uses 37 flops */
633 /* End of innermost loop */
635 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
636 f+i_coord_offset,fshift+i_shift_offset);
638 /* Increment number of inner iterations */
639 inneriter += j_index_end - j_index_start;
641 /* Outer loop uses 7 flops */
644 /* Increment number of outer iterations */
647 /* Update outer/inner flops */
649 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*37);