2 * Note: this file was generated by the Gromacs sse4_1_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_sse4_1_single.h"
34 #include "kernelutil_x86_sse4_1_single.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_VF_sse4_1_single
38 * Electrostatics interaction: ReactionField
39 * VdW interaction: LennardJones
40 * Geometry: Particle-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_VF_sse4_1_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 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
107 rcutoff_scalar = fr->rcoulomb;
108 rcutoff = _mm_set1_ps(rcutoff_scalar);
109 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
111 sh_vdw_invrcut6 = _mm_set1_ps(fr->ic->sh_invrc6);
112 rvdw = _mm_set1_ps(fr->rvdw);
114 /* Avoid stupid compiler warnings */
115 jnrA = jnrB = jnrC = jnrD = 0;
124 for(iidx=0;iidx<4*DIM;iidx++)
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_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
146 fix0 = _mm_setzero_ps();
147 fiy0 = _mm_setzero_ps();
148 fiz0 = _mm_setzero_ps();
150 /* Load parameters for i particles */
151 iq0 = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
152 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
154 /* Reset potential sums */
155 velecsum = _mm_setzero_ps();
156 vvdwsum = _mm_setzero_ps();
158 /* Start inner kernel loop */
159 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
162 /* Get j neighbor index, and coordinate index */
167 j_coord_offsetA = DIM*jnrA;
168 j_coord_offsetB = DIM*jnrB;
169 j_coord_offsetC = DIM*jnrC;
170 j_coord_offsetD = DIM*jnrD;
172 /* load j atom coordinates */
173 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
174 x+j_coord_offsetC,x+j_coord_offsetD,
177 /* Calculate displacement vector */
178 dx00 = _mm_sub_ps(ix0,jx0);
179 dy00 = _mm_sub_ps(iy0,jy0);
180 dz00 = _mm_sub_ps(iz0,jz0);
182 /* Calculate squared distance and things based on it */
183 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
185 rinv00 = gmx_mm_invsqrt_ps(rsq00);
187 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
189 /* Load parameters for j particles */
190 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
191 charge+jnrC+0,charge+jnrD+0);
192 vdwjidx0A = 2*vdwtype[jnrA+0];
193 vdwjidx0B = 2*vdwtype[jnrB+0];
194 vdwjidx0C = 2*vdwtype[jnrC+0];
195 vdwjidx0D = 2*vdwtype[jnrD+0];
197 /**************************
198 * CALCULATE INTERACTIONS *
199 **************************/
201 if (gmx_mm_any_lt(rsq00,rcutoff2))
204 /* Compute parameters for interactions between i and j atoms */
205 qq00 = _mm_mul_ps(iq0,jq0);
206 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
207 vdwparam+vdwioffset0+vdwjidx0B,
208 vdwparam+vdwioffset0+vdwjidx0C,
209 vdwparam+vdwioffset0+vdwjidx0D,
212 /* REACTION-FIELD ELECTROSTATICS */
213 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
214 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
216 /* LENNARD-JONES DISPERSION/REPULSION */
218 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
219 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
220 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
221 vvdw = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
222 _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
223 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
225 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
227 /* Update potential sum for this i atom from the interaction with this j atom. */
228 velec = _mm_and_ps(velec,cutoff_mask);
229 velecsum = _mm_add_ps(velecsum,velec);
230 vvdw = _mm_and_ps(vvdw,cutoff_mask);
231 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
233 fscal = _mm_add_ps(felec,fvdw);
235 fscal = _mm_and_ps(fscal,cutoff_mask);
237 /* Calculate temporary vectorial force */
238 tx = _mm_mul_ps(fscal,dx00);
239 ty = _mm_mul_ps(fscal,dy00);
240 tz = _mm_mul_ps(fscal,dz00);
242 /* Update vectorial force */
243 fix0 = _mm_add_ps(fix0,tx);
244 fiy0 = _mm_add_ps(fiy0,ty);
245 fiz0 = _mm_add_ps(fiz0,tz);
247 fjptrA = f+j_coord_offsetA;
248 fjptrB = f+j_coord_offsetB;
249 fjptrC = f+j_coord_offsetC;
250 fjptrD = f+j_coord_offsetD;
251 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
255 /* Inner loop uses 54 flops */
261 /* Get j neighbor index, and coordinate index */
262 jnrlistA = jjnr[jidx];
263 jnrlistB = jjnr[jidx+1];
264 jnrlistC = jjnr[jidx+2];
265 jnrlistD = jjnr[jidx+3];
266 /* Sign of each element will be negative for non-real atoms.
267 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
268 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
270 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
271 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
272 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
273 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
274 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
275 j_coord_offsetA = DIM*jnrA;
276 j_coord_offsetB = DIM*jnrB;
277 j_coord_offsetC = DIM*jnrC;
278 j_coord_offsetD = DIM*jnrD;
280 /* load j atom coordinates */
281 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
282 x+j_coord_offsetC,x+j_coord_offsetD,
285 /* Calculate displacement vector */
286 dx00 = _mm_sub_ps(ix0,jx0);
287 dy00 = _mm_sub_ps(iy0,jy0);
288 dz00 = _mm_sub_ps(iz0,jz0);
290 /* Calculate squared distance and things based on it */
291 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
293 rinv00 = gmx_mm_invsqrt_ps(rsq00);
295 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
297 /* Load parameters for j particles */
298 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
299 charge+jnrC+0,charge+jnrD+0);
300 vdwjidx0A = 2*vdwtype[jnrA+0];
301 vdwjidx0B = 2*vdwtype[jnrB+0];
302 vdwjidx0C = 2*vdwtype[jnrC+0];
303 vdwjidx0D = 2*vdwtype[jnrD+0];
305 /**************************
306 * CALCULATE INTERACTIONS *
307 **************************/
309 if (gmx_mm_any_lt(rsq00,rcutoff2))
312 /* Compute parameters for interactions between i and j atoms */
313 qq00 = _mm_mul_ps(iq0,jq0);
314 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
315 vdwparam+vdwioffset0+vdwjidx0B,
316 vdwparam+vdwioffset0+vdwjidx0C,
317 vdwparam+vdwioffset0+vdwjidx0D,
320 /* REACTION-FIELD ELECTROSTATICS */
321 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
322 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
324 /* LENNARD-JONES DISPERSION/REPULSION */
326 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
327 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
328 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
329 vvdw = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
330 _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
331 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
333 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
335 /* Update potential sum for this i atom from the interaction with this j atom. */
336 velec = _mm_and_ps(velec,cutoff_mask);
337 velec = _mm_andnot_ps(dummy_mask,velec);
338 velecsum = _mm_add_ps(velecsum,velec);
339 vvdw = _mm_and_ps(vvdw,cutoff_mask);
340 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
341 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
343 fscal = _mm_add_ps(felec,fvdw);
345 fscal = _mm_and_ps(fscal,cutoff_mask);
347 fscal = _mm_andnot_ps(dummy_mask,fscal);
349 /* Calculate temporary vectorial force */
350 tx = _mm_mul_ps(fscal,dx00);
351 ty = _mm_mul_ps(fscal,dy00);
352 tz = _mm_mul_ps(fscal,dz00);
354 /* Update vectorial force */
355 fix0 = _mm_add_ps(fix0,tx);
356 fiy0 = _mm_add_ps(fiy0,ty);
357 fiz0 = _mm_add_ps(fiz0,tz);
359 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
360 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
361 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
362 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
363 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
367 /* Inner loop uses 54 flops */
370 /* End of innermost loop */
372 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
373 f+i_coord_offset,fshift+i_shift_offset);
376 /* Update potential energies */
377 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
378 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
380 /* Increment number of inner iterations */
381 inneriter += j_index_end - j_index_start;
383 /* Outer loop uses 9 flops */
386 /* Increment number of outer iterations */
389 /* Update outer/inner flops */
391 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*54);
394 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_F_sse4_1_single
395 * Electrostatics interaction: ReactionField
396 * VdW interaction: LennardJones
397 * Geometry: Particle-Particle
398 * Calculate force/pot: Force
401 nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_F_sse4_1_single
402 (t_nblist * gmx_restrict nlist,
403 rvec * gmx_restrict xx,
404 rvec * gmx_restrict ff,
405 t_forcerec * gmx_restrict fr,
406 t_mdatoms * gmx_restrict mdatoms,
407 nb_kernel_data_t * gmx_restrict kernel_data,
408 t_nrnb * gmx_restrict nrnb)
410 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
411 * just 0 for non-waters.
412 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
413 * jnr indices corresponding to data put in the four positions in the SIMD register.
415 int i_shift_offset,i_coord_offset,outeriter,inneriter;
416 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
417 int jnrA,jnrB,jnrC,jnrD;
418 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
419 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
420 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
422 real *shiftvec,*fshift,*x,*f;
423 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
425 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
427 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
428 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
429 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
430 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
431 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
434 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
437 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
438 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
439 __m128 dummy_mask,cutoff_mask;
440 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
441 __m128 one = _mm_set1_ps(1.0);
442 __m128 two = _mm_set1_ps(2.0);
448 jindex = nlist->jindex;
450 shiftidx = nlist->shift;
452 shiftvec = fr->shift_vec[0];
453 fshift = fr->fshift[0];
454 facel = _mm_set1_ps(fr->epsfac);
455 charge = mdatoms->chargeA;
456 krf = _mm_set1_ps(fr->ic->k_rf);
457 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
458 crf = _mm_set1_ps(fr->ic->c_rf);
459 nvdwtype = fr->ntype;
461 vdwtype = mdatoms->typeA;
463 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
464 rcutoff_scalar = fr->rcoulomb;
465 rcutoff = _mm_set1_ps(rcutoff_scalar);
466 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
468 sh_vdw_invrcut6 = _mm_set1_ps(fr->ic->sh_invrc6);
469 rvdw = _mm_set1_ps(fr->rvdw);
471 /* Avoid stupid compiler warnings */
472 jnrA = jnrB = jnrC = jnrD = 0;
481 for(iidx=0;iidx<4*DIM;iidx++)
486 /* Start outer loop over neighborlists */
487 for(iidx=0; iidx<nri; iidx++)
489 /* Load shift vector for this list */
490 i_shift_offset = DIM*shiftidx[iidx];
492 /* Load limits for loop over neighbors */
493 j_index_start = jindex[iidx];
494 j_index_end = jindex[iidx+1];
496 /* Get outer coordinate index */
498 i_coord_offset = DIM*inr;
500 /* Load i particle coords and add shift vector */
501 gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
503 fix0 = _mm_setzero_ps();
504 fiy0 = _mm_setzero_ps();
505 fiz0 = _mm_setzero_ps();
507 /* Load parameters for i particles */
508 iq0 = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
509 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
511 /* Start inner kernel loop */
512 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
515 /* Get j neighbor index, and coordinate index */
520 j_coord_offsetA = DIM*jnrA;
521 j_coord_offsetB = DIM*jnrB;
522 j_coord_offsetC = DIM*jnrC;
523 j_coord_offsetD = DIM*jnrD;
525 /* load j atom coordinates */
526 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
527 x+j_coord_offsetC,x+j_coord_offsetD,
530 /* Calculate displacement vector */
531 dx00 = _mm_sub_ps(ix0,jx0);
532 dy00 = _mm_sub_ps(iy0,jy0);
533 dz00 = _mm_sub_ps(iz0,jz0);
535 /* Calculate squared distance and things based on it */
536 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
538 rinv00 = gmx_mm_invsqrt_ps(rsq00);
540 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
542 /* Load parameters for j particles */
543 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
544 charge+jnrC+0,charge+jnrD+0);
545 vdwjidx0A = 2*vdwtype[jnrA+0];
546 vdwjidx0B = 2*vdwtype[jnrB+0];
547 vdwjidx0C = 2*vdwtype[jnrC+0];
548 vdwjidx0D = 2*vdwtype[jnrD+0];
550 /**************************
551 * CALCULATE INTERACTIONS *
552 **************************/
554 if (gmx_mm_any_lt(rsq00,rcutoff2))
557 /* Compute parameters for interactions between i and j atoms */
558 qq00 = _mm_mul_ps(iq0,jq0);
559 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
560 vdwparam+vdwioffset0+vdwjidx0B,
561 vdwparam+vdwioffset0+vdwjidx0C,
562 vdwparam+vdwioffset0+vdwjidx0D,
565 /* REACTION-FIELD ELECTROSTATICS */
566 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
568 /* LENNARD-JONES DISPERSION/REPULSION */
570 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
571 fvdw = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
573 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
575 fscal = _mm_add_ps(felec,fvdw);
577 fscal = _mm_and_ps(fscal,cutoff_mask);
579 /* Calculate temporary vectorial force */
580 tx = _mm_mul_ps(fscal,dx00);
581 ty = _mm_mul_ps(fscal,dy00);
582 tz = _mm_mul_ps(fscal,dz00);
584 /* Update vectorial force */
585 fix0 = _mm_add_ps(fix0,tx);
586 fiy0 = _mm_add_ps(fiy0,ty);
587 fiz0 = _mm_add_ps(fiz0,tz);
589 fjptrA = f+j_coord_offsetA;
590 fjptrB = f+j_coord_offsetB;
591 fjptrC = f+j_coord_offsetC;
592 fjptrD = f+j_coord_offsetD;
593 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
597 /* Inner loop uses 37 flops */
603 /* Get j neighbor index, and coordinate index */
604 jnrlistA = jjnr[jidx];
605 jnrlistB = jjnr[jidx+1];
606 jnrlistC = jjnr[jidx+2];
607 jnrlistD = jjnr[jidx+3];
608 /* Sign of each element will be negative for non-real atoms.
609 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
610 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
612 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
613 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
614 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
615 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
616 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
617 j_coord_offsetA = DIM*jnrA;
618 j_coord_offsetB = DIM*jnrB;
619 j_coord_offsetC = DIM*jnrC;
620 j_coord_offsetD = DIM*jnrD;
622 /* load j atom coordinates */
623 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
624 x+j_coord_offsetC,x+j_coord_offsetD,
627 /* Calculate displacement vector */
628 dx00 = _mm_sub_ps(ix0,jx0);
629 dy00 = _mm_sub_ps(iy0,jy0);
630 dz00 = _mm_sub_ps(iz0,jz0);
632 /* Calculate squared distance and things based on it */
633 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
635 rinv00 = gmx_mm_invsqrt_ps(rsq00);
637 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
639 /* Load parameters for j particles */
640 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
641 charge+jnrC+0,charge+jnrD+0);
642 vdwjidx0A = 2*vdwtype[jnrA+0];
643 vdwjidx0B = 2*vdwtype[jnrB+0];
644 vdwjidx0C = 2*vdwtype[jnrC+0];
645 vdwjidx0D = 2*vdwtype[jnrD+0];
647 /**************************
648 * CALCULATE INTERACTIONS *
649 **************************/
651 if (gmx_mm_any_lt(rsq00,rcutoff2))
654 /* Compute parameters for interactions between i and j atoms */
655 qq00 = _mm_mul_ps(iq0,jq0);
656 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
657 vdwparam+vdwioffset0+vdwjidx0B,
658 vdwparam+vdwioffset0+vdwjidx0C,
659 vdwparam+vdwioffset0+vdwjidx0D,
662 /* REACTION-FIELD ELECTROSTATICS */
663 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
665 /* LENNARD-JONES DISPERSION/REPULSION */
667 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
668 fvdw = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
670 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
672 fscal = _mm_add_ps(felec,fvdw);
674 fscal = _mm_and_ps(fscal,cutoff_mask);
676 fscal = _mm_andnot_ps(dummy_mask,fscal);
678 /* Calculate temporary vectorial force */
679 tx = _mm_mul_ps(fscal,dx00);
680 ty = _mm_mul_ps(fscal,dy00);
681 tz = _mm_mul_ps(fscal,dz00);
683 /* Update vectorial force */
684 fix0 = _mm_add_ps(fix0,tx);
685 fiy0 = _mm_add_ps(fiy0,ty);
686 fiz0 = _mm_add_ps(fiz0,tz);
688 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
689 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
690 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
691 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
692 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
696 /* Inner loop uses 37 flops */
699 /* End of innermost loop */
701 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
702 f+i_coord_offset,fshift+i_shift_offset);
704 /* Increment number of inner iterations */
705 inneriter += j_index_end - j_index_start;
707 /* Outer loop uses 7 flops */
710 /* Increment number of outer iterations */
713 /* Update outer/inner flops */
715 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*37);