2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
36 * Note: this file was generated by the GROMACS sse4_1_single kernel generator.
42 #include "../nb_kernel.h"
43 #include "types/simple.h"
44 #include "gromacs/math/vec.h"
47 #include "gromacs/simd/math_x86_sse4_1_single.h"
48 #include "kernelutil_x86_sse4_1_single.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwLJ_GeomP1P1_VF_sse4_1_single
52 * Electrostatics interaction: ReactionField
53 * VdW interaction: LennardJones
54 * Geometry: Particle-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecRF_VdwLJ_GeomP1P1_VF_sse4_1_single
59 (t_nblist * gmx_restrict nlist,
60 rvec * gmx_restrict xx,
61 rvec * gmx_restrict ff,
62 t_forcerec * gmx_restrict fr,
63 t_mdatoms * gmx_restrict mdatoms,
64 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
65 t_nrnb * gmx_restrict nrnb)
67 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
68 * just 0 for non-waters.
69 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
70 * jnr indices corresponding to data put in the four positions in the SIMD register.
72 int i_shift_offset,i_coord_offset,outeriter,inneriter;
73 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
74 int jnrA,jnrB,jnrC,jnrD;
75 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
76 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
77 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
79 real *shiftvec,*fshift,*x,*f;
80 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
82 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
84 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
85 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
86 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
87 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
88 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
91 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
94 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
95 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
96 __m128 dummy_mask,cutoff_mask;
97 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
98 __m128 one = _mm_set1_ps(1.0);
99 __m128 two = _mm_set1_ps(2.0);
105 jindex = nlist->jindex;
107 shiftidx = nlist->shift;
109 shiftvec = fr->shift_vec[0];
110 fshift = fr->fshift[0];
111 facel = _mm_set1_ps(fr->epsfac);
112 charge = mdatoms->chargeA;
113 krf = _mm_set1_ps(fr->ic->k_rf);
114 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
115 crf = _mm_set1_ps(fr->ic->c_rf);
116 nvdwtype = fr->ntype;
118 vdwtype = mdatoms->typeA;
120 /* Avoid stupid compiler warnings */
121 jnrA = jnrB = jnrC = jnrD = 0;
130 for(iidx=0;iidx<4*DIM;iidx++)
135 /* Start outer loop over neighborlists */
136 for(iidx=0; iidx<nri; iidx++)
138 /* Load shift vector for this list */
139 i_shift_offset = DIM*shiftidx[iidx];
141 /* Load limits for loop over neighbors */
142 j_index_start = jindex[iidx];
143 j_index_end = jindex[iidx+1];
145 /* Get outer coordinate index */
147 i_coord_offset = DIM*inr;
149 /* Load i particle coords and add shift vector */
150 gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
152 fix0 = _mm_setzero_ps();
153 fiy0 = _mm_setzero_ps();
154 fiz0 = _mm_setzero_ps();
156 /* Load parameters for i particles */
157 iq0 = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
158 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
160 /* Reset potential sums */
161 velecsum = _mm_setzero_ps();
162 vvdwsum = _mm_setzero_ps();
164 /* Start inner kernel loop */
165 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
168 /* Get j neighbor index, and coordinate index */
173 j_coord_offsetA = DIM*jnrA;
174 j_coord_offsetB = DIM*jnrB;
175 j_coord_offsetC = DIM*jnrC;
176 j_coord_offsetD = DIM*jnrD;
178 /* load j atom coordinates */
179 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
180 x+j_coord_offsetC,x+j_coord_offsetD,
183 /* Calculate displacement vector */
184 dx00 = _mm_sub_ps(ix0,jx0);
185 dy00 = _mm_sub_ps(iy0,jy0);
186 dz00 = _mm_sub_ps(iz0,jz0);
188 /* Calculate squared distance and things based on it */
189 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
191 rinv00 = gmx_mm_invsqrt_ps(rsq00);
193 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
195 /* Load parameters for j particles */
196 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
197 charge+jnrC+0,charge+jnrD+0);
198 vdwjidx0A = 2*vdwtype[jnrA+0];
199 vdwjidx0B = 2*vdwtype[jnrB+0];
200 vdwjidx0C = 2*vdwtype[jnrC+0];
201 vdwjidx0D = 2*vdwtype[jnrD+0];
203 /**************************
204 * CALCULATE INTERACTIONS *
205 **************************/
207 /* Compute parameters for interactions between i and j atoms */
208 qq00 = _mm_mul_ps(iq0,jq0);
209 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
210 vdwparam+vdwioffset0+vdwjidx0B,
211 vdwparam+vdwioffset0+vdwjidx0C,
212 vdwparam+vdwioffset0+vdwjidx0D,
215 /* REACTION-FIELD ELECTROSTATICS */
216 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
217 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
219 /* LENNARD-JONES DISPERSION/REPULSION */
221 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
222 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
223 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
224 vvdw = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
225 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
227 /* Update potential sum for this i atom from the interaction with this j atom. */
228 velecsum = _mm_add_ps(velecsum,velec);
229 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
231 fscal = _mm_add_ps(felec,fvdw);
233 /* Calculate temporary vectorial force */
234 tx = _mm_mul_ps(fscal,dx00);
235 ty = _mm_mul_ps(fscal,dy00);
236 tz = _mm_mul_ps(fscal,dz00);
238 /* Update vectorial force */
239 fix0 = _mm_add_ps(fix0,tx);
240 fiy0 = _mm_add_ps(fiy0,ty);
241 fiz0 = _mm_add_ps(fiz0,tz);
243 fjptrA = f+j_coord_offsetA;
244 fjptrB = f+j_coord_offsetB;
245 fjptrC = f+j_coord_offsetC;
246 fjptrD = f+j_coord_offsetD;
247 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
249 /* Inner loop uses 44 flops */
255 /* Get j neighbor index, and coordinate index */
256 jnrlistA = jjnr[jidx];
257 jnrlistB = jjnr[jidx+1];
258 jnrlistC = jjnr[jidx+2];
259 jnrlistD = jjnr[jidx+3];
260 /* Sign of each element will be negative for non-real atoms.
261 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
262 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
264 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
265 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
266 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
267 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
268 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
269 j_coord_offsetA = DIM*jnrA;
270 j_coord_offsetB = DIM*jnrB;
271 j_coord_offsetC = DIM*jnrC;
272 j_coord_offsetD = DIM*jnrD;
274 /* load j atom coordinates */
275 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
276 x+j_coord_offsetC,x+j_coord_offsetD,
279 /* Calculate displacement vector */
280 dx00 = _mm_sub_ps(ix0,jx0);
281 dy00 = _mm_sub_ps(iy0,jy0);
282 dz00 = _mm_sub_ps(iz0,jz0);
284 /* Calculate squared distance and things based on it */
285 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
287 rinv00 = gmx_mm_invsqrt_ps(rsq00);
289 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
291 /* Load parameters for j particles */
292 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
293 charge+jnrC+0,charge+jnrD+0);
294 vdwjidx0A = 2*vdwtype[jnrA+0];
295 vdwjidx0B = 2*vdwtype[jnrB+0];
296 vdwjidx0C = 2*vdwtype[jnrC+0];
297 vdwjidx0D = 2*vdwtype[jnrD+0];
299 /**************************
300 * CALCULATE INTERACTIONS *
301 **************************/
303 /* Compute parameters for interactions between i and j atoms */
304 qq00 = _mm_mul_ps(iq0,jq0);
305 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
306 vdwparam+vdwioffset0+vdwjidx0B,
307 vdwparam+vdwioffset0+vdwjidx0C,
308 vdwparam+vdwioffset0+vdwjidx0D,
311 /* REACTION-FIELD ELECTROSTATICS */
312 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
313 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
315 /* LENNARD-JONES DISPERSION/REPULSION */
317 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
318 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
319 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
320 vvdw = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
321 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
323 /* Update potential sum for this i atom from the interaction with this j atom. */
324 velec = _mm_andnot_ps(dummy_mask,velec);
325 velecsum = _mm_add_ps(velecsum,velec);
326 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
327 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
329 fscal = _mm_add_ps(felec,fvdw);
331 fscal = _mm_andnot_ps(dummy_mask,fscal);
333 /* Calculate temporary vectorial force */
334 tx = _mm_mul_ps(fscal,dx00);
335 ty = _mm_mul_ps(fscal,dy00);
336 tz = _mm_mul_ps(fscal,dz00);
338 /* Update vectorial force */
339 fix0 = _mm_add_ps(fix0,tx);
340 fiy0 = _mm_add_ps(fiy0,ty);
341 fiz0 = _mm_add_ps(fiz0,tz);
343 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
344 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
345 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
346 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
347 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
349 /* Inner loop uses 44 flops */
352 /* End of innermost loop */
354 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
355 f+i_coord_offset,fshift+i_shift_offset);
358 /* Update potential energies */
359 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
360 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
362 /* Increment number of inner iterations */
363 inneriter += j_index_end - j_index_start;
365 /* Outer loop uses 9 flops */
368 /* Increment number of outer iterations */
371 /* Update outer/inner flops */
373 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*44);
376 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_sse4_1_single
377 * Electrostatics interaction: ReactionField
378 * VdW interaction: LennardJones
379 * Geometry: Particle-Particle
380 * Calculate force/pot: Force
383 nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_sse4_1_single
384 (t_nblist * gmx_restrict nlist,
385 rvec * gmx_restrict xx,
386 rvec * gmx_restrict ff,
387 t_forcerec * gmx_restrict fr,
388 t_mdatoms * gmx_restrict mdatoms,
389 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
390 t_nrnb * gmx_restrict nrnb)
392 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
393 * just 0 for non-waters.
394 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
395 * jnr indices corresponding to data put in the four positions in the SIMD register.
397 int i_shift_offset,i_coord_offset,outeriter,inneriter;
398 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
399 int jnrA,jnrB,jnrC,jnrD;
400 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
401 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
402 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
404 real *shiftvec,*fshift,*x,*f;
405 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
407 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
409 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
410 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
411 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
412 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
413 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
416 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
419 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
420 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
421 __m128 dummy_mask,cutoff_mask;
422 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
423 __m128 one = _mm_set1_ps(1.0);
424 __m128 two = _mm_set1_ps(2.0);
430 jindex = nlist->jindex;
432 shiftidx = nlist->shift;
434 shiftvec = fr->shift_vec[0];
435 fshift = fr->fshift[0];
436 facel = _mm_set1_ps(fr->epsfac);
437 charge = mdatoms->chargeA;
438 krf = _mm_set1_ps(fr->ic->k_rf);
439 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
440 crf = _mm_set1_ps(fr->ic->c_rf);
441 nvdwtype = fr->ntype;
443 vdwtype = mdatoms->typeA;
445 /* Avoid stupid compiler warnings */
446 jnrA = jnrB = jnrC = jnrD = 0;
455 for(iidx=0;iidx<4*DIM;iidx++)
460 /* Start outer loop over neighborlists */
461 for(iidx=0; iidx<nri; iidx++)
463 /* Load shift vector for this list */
464 i_shift_offset = DIM*shiftidx[iidx];
466 /* Load limits for loop over neighbors */
467 j_index_start = jindex[iidx];
468 j_index_end = jindex[iidx+1];
470 /* Get outer coordinate index */
472 i_coord_offset = DIM*inr;
474 /* Load i particle coords and add shift vector */
475 gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
477 fix0 = _mm_setzero_ps();
478 fiy0 = _mm_setzero_ps();
479 fiz0 = _mm_setzero_ps();
481 /* Load parameters for i particles */
482 iq0 = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
483 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
485 /* Start inner kernel loop */
486 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
489 /* Get j neighbor index, and coordinate index */
494 j_coord_offsetA = DIM*jnrA;
495 j_coord_offsetB = DIM*jnrB;
496 j_coord_offsetC = DIM*jnrC;
497 j_coord_offsetD = DIM*jnrD;
499 /* load j atom coordinates */
500 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
501 x+j_coord_offsetC,x+j_coord_offsetD,
504 /* Calculate displacement vector */
505 dx00 = _mm_sub_ps(ix0,jx0);
506 dy00 = _mm_sub_ps(iy0,jy0);
507 dz00 = _mm_sub_ps(iz0,jz0);
509 /* Calculate squared distance and things based on it */
510 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
512 rinv00 = gmx_mm_invsqrt_ps(rsq00);
514 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
516 /* Load parameters for j particles */
517 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
518 charge+jnrC+0,charge+jnrD+0);
519 vdwjidx0A = 2*vdwtype[jnrA+0];
520 vdwjidx0B = 2*vdwtype[jnrB+0];
521 vdwjidx0C = 2*vdwtype[jnrC+0];
522 vdwjidx0D = 2*vdwtype[jnrD+0];
524 /**************************
525 * CALCULATE INTERACTIONS *
526 **************************/
528 /* Compute parameters for interactions between i and j atoms */
529 qq00 = _mm_mul_ps(iq0,jq0);
530 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
531 vdwparam+vdwioffset0+vdwjidx0B,
532 vdwparam+vdwioffset0+vdwjidx0C,
533 vdwparam+vdwioffset0+vdwjidx0D,
536 /* REACTION-FIELD ELECTROSTATICS */
537 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
539 /* LENNARD-JONES DISPERSION/REPULSION */
541 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
542 fvdw = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
544 fscal = _mm_add_ps(felec,fvdw);
546 /* Calculate temporary vectorial force */
547 tx = _mm_mul_ps(fscal,dx00);
548 ty = _mm_mul_ps(fscal,dy00);
549 tz = _mm_mul_ps(fscal,dz00);
551 /* Update vectorial force */
552 fix0 = _mm_add_ps(fix0,tx);
553 fiy0 = _mm_add_ps(fiy0,ty);
554 fiz0 = _mm_add_ps(fiz0,tz);
556 fjptrA = f+j_coord_offsetA;
557 fjptrB = f+j_coord_offsetB;
558 fjptrC = f+j_coord_offsetC;
559 fjptrD = f+j_coord_offsetD;
560 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
562 /* Inner loop uses 34 flops */
568 /* Get j neighbor index, and coordinate index */
569 jnrlistA = jjnr[jidx];
570 jnrlistB = jjnr[jidx+1];
571 jnrlistC = jjnr[jidx+2];
572 jnrlistD = jjnr[jidx+3];
573 /* Sign of each element will be negative for non-real atoms.
574 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
575 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
577 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
578 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
579 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
580 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
581 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
582 j_coord_offsetA = DIM*jnrA;
583 j_coord_offsetB = DIM*jnrB;
584 j_coord_offsetC = DIM*jnrC;
585 j_coord_offsetD = DIM*jnrD;
587 /* load j atom coordinates */
588 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
589 x+j_coord_offsetC,x+j_coord_offsetD,
592 /* Calculate displacement vector */
593 dx00 = _mm_sub_ps(ix0,jx0);
594 dy00 = _mm_sub_ps(iy0,jy0);
595 dz00 = _mm_sub_ps(iz0,jz0);
597 /* Calculate squared distance and things based on it */
598 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
600 rinv00 = gmx_mm_invsqrt_ps(rsq00);
602 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
604 /* Load parameters for j particles */
605 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
606 charge+jnrC+0,charge+jnrD+0);
607 vdwjidx0A = 2*vdwtype[jnrA+0];
608 vdwjidx0B = 2*vdwtype[jnrB+0];
609 vdwjidx0C = 2*vdwtype[jnrC+0];
610 vdwjidx0D = 2*vdwtype[jnrD+0];
612 /**************************
613 * CALCULATE INTERACTIONS *
614 **************************/
616 /* Compute parameters for interactions between i and j atoms */
617 qq00 = _mm_mul_ps(iq0,jq0);
618 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
619 vdwparam+vdwioffset0+vdwjidx0B,
620 vdwparam+vdwioffset0+vdwjidx0C,
621 vdwparam+vdwioffset0+vdwjidx0D,
624 /* REACTION-FIELD ELECTROSTATICS */
625 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
627 /* LENNARD-JONES DISPERSION/REPULSION */
629 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
630 fvdw = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
632 fscal = _mm_add_ps(felec,fvdw);
634 fscal = _mm_andnot_ps(dummy_mask,fscal);
636 /* Calculate temporary vectorial force */
637 tx = _mm_mul_ps(fscal,dx00);
638 ty = _mm_mul_ps(fscal,dy00);
639 tz = _mm_mul_ps(fscal,dz00);
641 /* Update vectorial force */
642 fix0 = _mm_add_ps(fix0,tx);
643 fiy0 = _mm_add_ps(fiy0,ty);
644 fiz0 = _mm_add_ps(fiz0,tz);
646 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
647 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
648 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
649 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
650 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
652 /* Inner loop uses 34 flops */
655 /* End of innermost loop */
657 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
658 f+i_coord_offset,fshift+i_shift_offset);
660 /* Increment number of inner iterations */
661 inneriter += j_index_end - j_index_start;
663 /* Outer loop uses 7 flops */
666 /* Increment number of outer iterations */
669 /* Update outer/inner flops */
671 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*34);