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_ElecRFCut_VdwNone_GeomW3P1_VF_sse2_single
38 * Electrostatics interaction: ReactionField
39 * VdW interaction: None
40 * Geometry: Water3-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecRFCut_VdwNone_GeomW3P1_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 j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
62 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
63 real shX,shY,shZ,rcutoff_scalar;
64 real *shiftvec,*fshift,*x,*f;
65 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
67 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
69 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
71 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
72 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
73 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
74 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
75 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
76 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
77 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
79 __m128 dummy_mask,cutoff_mask;
80 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
81 __m128 one = _mm_set1_ps(1.0);
82 __m128 two = _mm_set1_ps(2.0);
88 jindex = nlist->jindex;
90 shiftidx = nlist->shift;
92 shiftvec = fr->shift_vec[0];
93 fshift = fr->fshift[0];
94 facel = _mm_set1_ps(fr->epsfac);
95 charge = mdatoms->chargeA;
96 krf = _mm_set1_ps(fr->ic->k_rf);
97 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
98 crf = _mm_set1_ps(fr->ic->c_rf);
100 /* Setup water-specific parameters */
101 inr = nlist->iinr[0];
102 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
103 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
104 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
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 /* Avoid stupid compiler warnings */
112 jnrA = jnrB = jnrC = jnrD = 0;
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];
126 shX = shiftvec[i_shift_offset+XX];
127 shY = shiftvec[i_shift_offset+YY];
128 shZ = shiftvec[i_shift_offset+ZZ];
130 /* Load limits for loop over neighbors */
131 j_index_start = jindex[iidx];
132 j_index_end = jindex[iidx+1];
134 /* Get outer coordinate index */
136 i_coord_offset = DIM*inr;
138 /* Load i particle coords and add shift vector */
139 ix0 = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
140 iy0 = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
141 iz0 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
142 ix1 = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
143 iy1 = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
144 iz1 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
145 ix2 = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
146 iy2 = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
147 iz2 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
149 fix0 = _mm_setzero_ps();
150 fiy0 = _mm_setzero_ps();
151 fiz0 = _mm_setzero_ps();
152 fix1 = _mm_setzero_ps();
153 fiy1 = _mm_setzero_ps();
154 fiz1 = _mm_setzero_ps();
155 fix2 = _mm_setzero_ps();
156 fiy2 = _mm_setzero_ps();
157 fiz2 = _mm_setzero_ps();
159 /* Reset potential sums */
160 velecsum = _mm_setzero_ps();
162 /* Start inner kernel loop */
163 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
166 /* Get j neighbor index, and coordinate index */
172 j_coord_offsetA = DIM*jnrA;
173 j_coord_offsetB = DIM*jnrB;
174 j_coord_offsetC = DIM*jnrC;
175 j_coord_offsetD = DIM*jnrD;
177 /* load j atom coordinates */
178 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
179 x+j_coord_offsetC,x+j_coord_offsetD,
182 /* Calculate displacement vector */
183 dx00 = _mm_sub_ps(ix0,jx0);
184 dy00 = _mm_sub_ps(iy0,jy0);
185 dz00 = _mm_sub_ps(iz0,jz0);
186 dx10 = _mm_sub_ps(ix1,jx0);
187 dy10 = _mm_sub_ps(iy1,jy0);
188 dz10 = _mm_sub_ps(iz1,jz0);
189 dx20 = _mm_sub_ps(ix2,jx0);
190 dy20 = _mm_sub_ps(iy2,jy0);
191 dz20 = _mm_sub_ps(iz2,jz0);
193 /* Calculate squared distance and things based on it */
194 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
195 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
196 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
198 rinv00 = gmx_mm_invsqrt_ps(rsq00);
199 rinv10 = gmx_mm_invsqrt_ps(rsq10);
200 rinv20 = gmx_mm_invsqrt_ps(rsq20);
202 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
203 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
204 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
206 /* Load parameters for j particles */
207 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
208 charge+jnrC+0,charge+jnrD+0);
210 /**************************
211 * CALCULATE INTERACTIONS *
212 **************************/
214 if (gmx_mm_any_lt(rsq00,rcutoff2))
217 /* Compute parameters for interactions between i and j atoms */
218 qq00 = _mm_mul_ps(iq0,jq0);
220 /* REACTION-FIELD ELECTROSTATICS */
221 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
222 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
224 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
226 /* Update potential sum for this i atom from the interaction with this j atom. */
227 velec = _mm_and_ps(velec,cutoff_mask);
228 velecsum = _mm_add_ps(velecsum,velec);
232 fscal = _mm_and_ps(fscal,cutoff_mask);
234 /* Calculate temporary vectorial force */
235 tx = _mm_mul_ps(fscal,dx00);
236 ty = _mm_mul_ps(fscal,dy00);
237 tz = _mm_mul_ps(fscal,dz00);
239 /* Update vectorial force */
240 fix0 = _mm_add_ps(fix0,tx);
241 fiy0 = _mm_add_ps(fiy0,ty);
242 fiz0 = _mm_add_ps(fiz0,tz);
244 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
245 f+j_coord_offsetC,f+j_coord_offsetD,
250 /**************************
251 * CALCULATE INTERACTIONS *
252 **************************/
254 if (gmx_mm_any_lt(rsq10,rcutoff2))
257 /* Compute parameters for interactions between i and j atoms */
258 qq10 = _mm_mul_ps(iq1,jq0);
260 /* REACTION-FIELD ELECTROSTATICS */
261 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
262 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
264 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
266 /* Update potential sum for this i atom from the interaction with this j atom. */
267 velec = _mm_and_ps(velec,cutoff_mask);
268 velecsum = _mm_add_ps(velecsum,velec);
272 fscal = _mm_and_ps(fscal,cutoff_mask);
274 /* Calculate temporary vectorial force */
275 tx = _mm_mul_ps(fscal,dx10);
276 ty = _mm_mul_ps(fscal,dy10);
277 tz = _mm_mul_ps(fscal,dz10);
279 /* Update vectorial force */
280 fix1 = _mm_add_ps(fix1,tx);
281 fiy1 = _mm_add_ps(fiy1,ty);
282 fiz1 = _mm_add_ps(fiz1,tz);
284 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
285 f+j_coord_offsetC,f+j_coord_offsetD,
290 /**************************
291 * CALCULATE INTERACTIONS *
292 **************************/
294 if (gmx_mm_any_lt(rsq20,rcutoff2))
297 /* Compute parameters for interactions between i and j atoms */
298 qq20 = _mm_mul_ps(iq2,jq0);
300 /* REACTION-FIELD ELECTROSTATICS */
301 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
302 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
304 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
306 /* Update potential sum for this i atom from the interaction with this j atom. */
307 velec = _mm_and_ps(velec,cutoff_mask);
308 velecsum = _mm_add_ps(velecsum,velec);
312 fscal = _mm_and_ps(fscal,cutoff_mask);
314 /* Calculate temporary vectorial force */
315 tx = _mm_mul_ps(fscal,dx20);
316 ty = _mm_mul_ps(fscal,dy20);
317 tz = _mm_mul_ps(fscal,dz20);
319 /* Update vectorial force */
320 fix2 = _mm_add_ps(fix2,tx);
321 fiy2 = _mm_add_ps(fiy2,ty);
322 fiz2 = _mm_add_ps(fiz2,tz);
324 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
325 f+j_coord_offsetC,f+j_coord_offsetD,
330 /* Inner loop uses 108 flops */
336 /* Get j neighbor index, and coordinate index */
342 /* Sign of each element will be negative for non-real atoms.
343 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
344 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
346 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
347 jnrA = (jnrA>=0) ? jnrA : 0;
348 jnrB = (jnrB>=0) ? jnrB : 0;
349 jnrC = (jnrC>=0) ? jnrC : 0;
350 jnrD = (jnrD>=0) ? jnrD : 0;
352 j_coord_offsetA = DIM*jnrA;
353 j_coord_offsetB = DIM*jnrB;
354 j_coord_offsetC = DIM*jnrC;
355 j_coord_offsetD = DIM*jnrD;
357 /* load j atom coordinates */
358 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
359 x+j_coord_offsetC,x+j_coord_offsetD,
362 /* Calculate displacement vector */
363 dx00 = _mm_sub_ps(ix0,jx0);
364 dy00 = _mm_sub_ps(iy0,jy0);
365 dz00 = _mm_sub_ps(iz0,jz0);
366 dx10 = _mm_sub_ps(ix1,jx0);
367 dy10 = _mm_sub_ps(iy1,jy0);
368 dz10 = _mm_sub_ps(iz1,jz0);
369 dx20 = _mm_sub_ps(ix2,jx0);
370 dy20 = _mm_sub_ps(iy2,jy0);
371 dz20 = _mm_sub_ps(iz2,jz0);
373 /* Calculate squared distance and things based on it */
374 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
375 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
376 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
378 rinv00 = gmx_mm_invsqrt_ps(rsq00);
379 rinv10 = gmx_mm_invsqrt_ps(rsq10);
380 rinv20 = gmx_mm_invsqrt_ps(rsq20);
382 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
383 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
384 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
386 /* Load parameters for j particles */
387 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
388 charge+jnrC+0,charge+jnrD+0);
390 /**************************
391 * CALCULATE INTERACTIONS *
392 **************************/
394 if (gmx_mm_any_lt(rsq00,rcutoff2))
397 /* Compute parameters for interactions between i and j atoms */
398 qq00 = _mm_mul_ps(iq0,jq0);
400 /* REACTION-FIELD ELECTROSTATICS */
401 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
402 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
404 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
406 /* Update potential sum for this i atom from the interaction with this j atom. */
407 velec = _mm_and_ps(velec,cutoff_mask);
408 velec = _mm_andnot_ps(dummy_mask,velec);
409 velecsum = _mm_add_ps(velecsum,velec);
413 fscal = _mm_and_ps(fscal,cutoff_mask);
415 fscal = _mm_andnot_ps(dummy_mask,fscal);
417 /* Calculate temporary vectorial force */
418 tx = _mm_mul_ps(fscal,dx00);
419 ty = _mm_mul_ps(fscal,dy00);
420 tz = _mm_mul_ps(fscal,dz00);
422 /* Update vectorial force */
423 fix0 = _mm_add_ps(fix0,tx);
424 fiy0 = _mm_add_ps(fiy0,ty);
425 fiz0 = _mm_add_ps(fiz0,tz);
427 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
428 f+j_coord_offsetC,f+j_coord_offsetD,
433 /**************************
434 * CALCULATE INTERACTIONS *
435 **************************/
437 if (gmx_mm_any_lt(rsq10,rcutoff2))
440 /* Compute parameters for interactions between i and j atoms */
441 qq10 = _mm_mul_ps(iq1,jq0);
443 /* REACTION-FIELD ELECTROSTATICS */
444 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
445 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
447 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
449 /* Update potential sum for this i atom from the interaction with this j atom. */
450 velec = _mm_and_ps(velec,cutoff_mask);
451 velec = _mm_andnot_ps(dummy_mask,velec);
452 velecsum = _mm_add_ps(velecsum,velec);
456 fscal = _mm_and_ps(fscal,cutoff_mask);
458 fscal = _mm_andnot_ps(dummy_mask,fscal);
460 /* Calculate temporary vectorial force */
461 tx = _mm_mul_ps(fscal,dx10);
462 ty = _mm_mul_ps(fscal,dy10);
463 tz = _mm_mul_ps(fscal,dz10);
465 /* Update vectorial force */
466 fix1 = _mm_add_ps(fix1,tx);
467 fiy1 = _mm_add_ps(fiy1,ty);
468 fiz1 = _mm_add_ps(fiz1,tz);
470 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
471 f+j_coord_offsetC,f+j_coord_offsetD,
476 /**************************
477 * CALCULATE INTERACTIONS *
478 **************************/
480 if (gmx_mm_any_lt(rsq20,rcutoff2))
483 /* Compute parameters for interactions between i and j atoms */
484 qq20 = _mm_mul_ps(iq2,jq0);
486 /* REACTION-FIELD ELECTROSTATICS */
487 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
488 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
490 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
492 /* Update potential sum for this i atom from the interaction with this j atom. */
493 velec = _mm_and_ps(velec,cutoff_mask);
494 velec = _mm_andnot_ps(dummy_mask,velec);
495 velecsum = _mm_add_ps(velecsum,velec);
499 fscal = _mm_and_ps(fscal,cutoff_mask);
501 fscal = _mm_andnot_ps(dummy_mask,fscal);
503 /* Calculate temporary vectorial force */
504 tx = _mm_mul_ps(fscal,dx20);
505 ty = _mm_mul_ps(fscal,dy20);
506 tz = _mm_mul_ps(fscal,dz20);
508 /* Update vectorial force */
509 fix2 = _mm_add_ps(fix2,tx);
510 fiy2 = _mm_add_ps(fiy2,ty);
511 fiz2 = _mm_add_ps(fiz2,tz);
513 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
514 f+j_coord_offsetC,f+j_coord_offsetD,
519 /* Inner loop uses 108 flops */
522 /* End of innermost loop */
524 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
525 f+i_coord_offset,fshift+i_shift_offset);
528 /* Update potential energies */
529 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
531 /* Increment number of inner iterations */
532 inneriter += j_index_end - j_index_start;
534 /* Outer loop uses 28 flops */
537 /* Increment number of outer iterations */
540 /* Update outer/inner flops */
542 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*28 + inneriter*108);
545 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwNone_GeomW3P1_F_sse2_single
546 * Electrostatics interaction: ReactionField
547 * VdW interaction: None
548 * Geometry: Water3-Particle
549 * Calculate force/pot: Force
552 nb_kernel_ElecRFCut_VdwNone_GeomW3P1_F_sse2_single
553 (t_nblist * gmx_restrict nlist,
554 rvec * gmx_restrict xx,
555 rvec * gmx_restrict ff,
556 t_forcerec * gmx_restrict fr,
557 t_mdatoms * gmx_restrict mdatoms,
558 nb_kernel_data_t * gmx_restrict kernel_data,
559 t_nrnb * gmx_restrict nrnb)
561 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
562 * just 0 for non-waters.
563 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
564 * jnr indices corresponding to data put in the four positions in the SIMD register.
566 int i_shift_offset,i_coord_offset,outeriter,inneriter;
567 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
568 int jnrA,jnrB,jnrC,jnrD;
569 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
570 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
571 real shX,shY,shZ,rcutoff_scalar;
572 real *shiftvec,*fshift,*x,*f;
573 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
575 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
577 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
579 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
580 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
581 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
582 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
583 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
584 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
585 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
587 __m128 dummy_mask,cutoff_mask;
588 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
589 __m128 one = _mm_set1_ps(1.0);
590 __m128 two = _mm_set1_ps(2.0);
596 jindex = nlist->jindex;
598 shiftidx = nlist->shift;
600 shiftvec = fr->shift_vec[0];
601 fshift = fr->fshift[0];
602 facel = _mm_set1_ps(fr->epsfac);
603 charge = mdatoms->chargeA;
604 krf = _mm_set1_ps(fr->ic->k_rf);
605 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
606 crf = _mm_set1_ps(fr->ic->c_rf);
608 /* Setup water-specific parameters */
609 inr = nlist->iinr[0];
610 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
611 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
612 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
614 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
615 rcutoff_scalar = fr->rcoulomb;
616 rcutoff = _mm_set1_ps(rcutoff_scalar);
617 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
619 /* Avoid stupid compiler warnings */
620 jnrA = jnrB = jnrC = jnrD = 0;
629 /* Start outer loop over neighborlists */
630 for(iidx=0; iidx<nri; iidx++)
632 /* Load shift vector for this list */
633 i_shift_offset = DIM*shiftidx[iidx];
634 shX = shiftvec[i_shift_offset+XX];
635 shY = shiftvec[i_shift_offset+YY];
636 shZ = shiftvec[i_shift_offset+ZZ];
638 /* Load limits for loop over neighbors */
639 j_index_start = jindex[iidx];
640 j_index_end = jindex[iidx+1];
642 /* Get outer coordinate index */
644 i_coord_offset = DIM*inr;
646 /* Load i particle coords and add shift vector */
647 ix0 = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
648 iy0 = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
649 iz0 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
650 ix1 = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
651 iy1 = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
652 iz1 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
653 ix2 = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
654 iy2 = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
655 iz2 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
657 fix0 = _mm_setzero_ps();
658 fiy0 = _mm_setzero_ps();
659 fiz0 = _mm_setzero_ps();
660 fix1 = _mm_setzero_ps();
661 fiy1 = _mm_setzero_ps();
662 fiz1 = _mm_setzero_ps();
663 fix2 = _mm_setzero_ps();
664 fiy2 = _mm_setzero_ps();
665 fiz2 = _mm_setzero_ps();
667 /* Start inner kernel loop */
668 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
671 /* Get j neighbor index, and coordinate index */
677 j_coord_offsetA = DIM*jnrA;
678 j_coord_offsetB = DIM*jnrB;
679 j_coord_offsetC = DIM*jnrC;
680 j_coord_offsetD = DIM*jnrD;
682 /* load j atom coordinates */
683 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
684 x+j_coord_offsetC,x+j_coord_offsetD,
687 /* Calculate displacement vector */
688 dx00 = _mm_sub_ps(ix0,jx0);
689 dy00 = _mm_sub_ps(iy0,jy0);
690 dz00 = _mm_sub_ps(iz0,jz0);
691 dx10 = _mm_sub_ps(ix1,jx0);
692 dy10 = _mm_sub_ps(iy1,jy0);
693 dz10 = _mm_sub_ps(iz1,jz0);
694 dx20 = _mm_sub_ps(ix2,jx0);
695 dy20 = _mm_sub_ps(iy2,jy0);
696 dz20 = _mm_sub_ps(iz2,jz0);
698 /* Calculate squared distance and things based on it */
699 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
700 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
701 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
703 rinv00 = gmx_mm_invsqrt_ps(rsq00);
704 rinv10 = gmx_mm_invsqrt_ps(rsq10);
705 rinv20 = gmx_mm_invsqrt_ps(rsq20);
707 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
708 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
709 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
711 /* Load parameters for j particles */
712 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
713 charge+jnrC+0,charge+jnrD+0);
715 /**************************
716 * CALCULATE INTERACTIONS *
717 **************************/
719 if (gmx_mm_any_lt(rsq00,rcutoff2))
722 /* Compute parameters for interactions between i and j atoms */
723 qq00 = _mm_mul_ps(iq0,jq0);
725 /* REACTION-FIELD ELECTROSTATICS */
726 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
728 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
732 fscal = _mm_and_ps(fscal,cutoff_mask);
734 /* Calculate temporary vectorial force */
735 tx = _mm_mul_ps(fscal,dx00);
736 ty = _mm_mul_ps(fscal,dy00);
737 tz = _mm_mul_ps(fscal,dz00);
739 /* Update vectorial force */
740 fix0 = _mm_add_ps(fix0,tx);
741 fiy0 = _mm_add_ps(fiy0,ty);
742 fiz0 = _mm_add_ps(fiz0,tz);
744 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
745 f+j_coord_offsetC,f+j_coord_offsetD,
750 /**************************
751 * CALCULATE INTERACTIONS *
752 **************************/
754 if (gmx_mm_any_lt(rsq10,rcutoff2))
757 /* Compute parameters for interactions between i and j atoms */
758 qq10 = _mm_mul_ps(iq1,jq0);
760 /* REACTION-FIELD ELECTROSTATICS */
761 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
763 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
767 fscal = _mm_and_ps(fscal,cutoff_mask);
769 /* Calculate temporary vectorial force */
770 tx = _mm_mul_ps(fscal,dx10);
771 ty = _mm_mul_ps(fscal,dy10);
772 tz = _mm_mul_ps(fscal,dz10);
774 /* Update vectorial force */
775 fix1 = _mm_add_ps(fix1,tx);
776 fiy1 = _mm_add_ps(fiy1,ty);
777 fiz1 = _mm_add_ps(fiz1,tz);
779 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
780 f+j_coord_offsetC,f+j_coord_offsetD,
785 /**************************
786 * CALCULATE INTERACTIONS *
787 **************************/
789 if (gmx_mm_any_lt(rsq20,rcutoff2))
792 /* Compute parameters for interactions between i and j atoms */
793 qq20 = _mm_mul_ps(iq2,jq0);
795 /* REACTION-FIELD ELECTROSTATICS */
796 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
798 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
802 fscal = _mm_and_ps(fscal,cutoff_mask);
804 /* Calculate temporary vectorial force */
805 tx = _mm_mul_ps(fscal,dx20);
806 ty = _mm_mul_ps(fscal,dy20);
807 tz = _mm_mul_ps(fscal,dz20);
809 /* Update vectorial force */
810 fix2 = _mm_add_ps(fix2,tx);
811 fiy2 = _mm_add_ps(fiy2,ty);
812 fiz2 = _mm_add_ps(fiz2,tz);
814 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
815 f+j_coord_offsetC,f+j_coord_offsetD,
820 /* Inner loop uses 90 flops */
826 /* Get j neighbor index, and coordinate index */
832 /* Sign of each element will be negative for non-real atoms.
833 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
834 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
836 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
837 jnrA = (jnrA>=0) ? jnrA : 0;
838 jnrB = (jnrB>=0) ? jnrB : 0;
839 jnrC = (jnrC>=0) ? jnrC : 0;
840 jnrD = (jnrD>=0) ? jnrD : 0;
842 j_coord_offsetA = DIM*jnrA;
843 j_coord_offsetB = DIM*jnrB;
844 j_coord_offsetC = DIM*jnrC;
845 j_coord_offsetD = DIM*jnrD;
847 /* load j atom coordinates */
848 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
849 x+j_coord_offsetC,x+j_coord_offsetD,
852 /* Calculate displacement vector */
853 dx00 = _mm_sub_ps(ix0,jx0);
854 dy00 = _mm_sub_ps(iy0,jy0);
855 dz00 = _mm_sub_ps(iz0,jz0);
856 dx10 = _mm_sub_ps(ix1,jx0);
857 dy10 = _mm_sub_ps(iy1,jy0);
858 dz10 = _mm_sub_ps(iz1,jz0);
859 dx20 = _mm_sub_ps(ix2,jx0);
860 dy20 = _mm_sub_ps(iy2,jy0);
861 dz20 = _mm_sub_ps(iz2,jz0);
863 /* Calculate squared distance and things based on it */
864 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
865 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
866 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
868 rinv00 = gmx_mm_invsqrt_ps(rsq00);
869 rinv10 = gmx_mm_invsqrt_ps(rsq10);
870 rinv20 = gmx_mm_invsqrt_ps(rsq20);
872 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
873 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
874 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
876 /* Load parameters for j particles */
877 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
878 charge+jnrC+0,charge+jnrD+0);
880 /**************************
881 * CALCULATE INTERACTIONS *
882 **************************/
884 if (gmx_mm_any_lt(rsq00,rcutoff2))
887 /* Compute parameters for interactions between i and j atoms */
888 qq00 = _mm_mul_ps(iq0,jq0);
890 /* REACTION-FIELD ELECTROSTATICS */
891 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
893 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
897 fscal = _mm_and_ps(fscal,cutoff_mask);
899 fscal = _mm_andnot_ps(dummy_mask,fscal);
901 /* Calculate temporary vectorial force */
902 tx = _mm_mul_ps(fscal,dx00);
903 ty = _mm_mul_ps(fscal,dy00);
904 tz = _mm_mul_ps(fscal,dz00);
906 /* Update vectorial force */
907 fix0 = _mm_add_ps(fix0,tx);
908 fiy0 = _mm_add_ps(fiy0,ty);
909 fiz0 = _mm_add_ps(fiz0,tz);
911 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
912 f+j_coord_offsetC,f+j_coord_offsetD,
917 /**************************
918 * CALCULATE INTERACTIONS *
919 **************************/
921 if (gmx_mm_any_lt(rsq10,rcutoff2))
924 /* Compute parameters for interactions between i and j atoms */
925 qq10 = _mm_mul_ps(iq1,jq0);
927 /* REACTION-FIELD ELECTROSTATICS */
928 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
930 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
934 fscal = _mm_and_ps(fscal,cutoff_mask);
936 fscal = _mm_andnot_ps(dummy_mask,fscal);
938 /* Calculate temporary vectorial force */
939 tx = _mm_mul_ps(fscal,dx10);
940 ty = _mm_mul_ps(fscal,dy10);
941 tz = _mm_mul_ps(fscal,dz10);
943 /* Update vectorial force */
944 fix1 = _mm_add_ps(fix1,tx);
945 fiy1 = _mm_add_ps(fiy1,ty);
946 fiz1 = _mm_add_ps(fiz1,tz);
948 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
949 f+j_coord_offsetC,f+j_coord_offsetD,
954 /**************************
955 * CALCULATE INTERACTIONS *
956 **************************/
958 if (gmx_mm_any_lt(rsq20,rcutoff2))
961 /* Compute parameters for interactions between i and j atoms */
962 qq20 = _mm_mul_ps(iq2,jq0);
964 /* REACTION-FIELD ELECTROSTATICS */
965 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
967 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
971 fscal = _mm_and_ps(fscal,cutoff_mask);
973 fscal = _mm_andnot_ps(dummy_mask,fscal);
975 /* Calculate temporary vectorial force */
976 tx = _mm_mul_ps(fscal,dx20);
977 ty = _mm_mul_ps(fscal,dy20);
978 tz = _mm_mul_ps(fscal,dz20);
980 /* Update vectorial force */
981 fix2 = _mm_add_ps(fix2,tx);
982 fiy2 = _mm_add_ps(fiy2,ty);
983 fiz2 = _mm_add_ps(fiz2,tz);
985 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
986 f+j_coord_offsetC,f+j_coord_offsetD,
991 /* Inner loop uses 90 flops */
994 /* End of innermost loop */
996 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
997 f+i_coord_offset,fshift+i_shift_offset);
999 /* Increment number of inner iterations */
1000 inneriter += j_index_end - j_index_start;
1002 /* Outer loop uses 27 flops */
1005 /* Increment number of outer iterations */
1008 /* Update outer/inner flops */
1010 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*27 + inneriter*90);