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_ElecRF_VdwNone_GeomW3P1_VF_sse4_1_single
38 * Electrostatics interaction: ReactionField
39 * VdW interaction: None
40 * Geometry: Water3-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecRF_VdwNone_GeomW3P1_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;
72 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
74 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
75 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
76 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
77 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
78 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
79 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
80 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
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);
103 /* Setup water-specific parameters */
104 inr = nlist->iinr[0];
105 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
106 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
107 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
109 /* Avoid stupid compiler warnings */
110 jnrA = jnrB = jnrC = jnrD = 0;
119 for(iidx=0;iidx<4*DIM;iidx++)
124 /* Start outer loop over neighborlists */
125 for(iidx=0; iidx<nri; iidx++)
127 /* Load shift vector for this list */
128 i_shift_offset = DIM*shiftidx[iidx];
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 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
140 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
142 fix0 = _mm_setzero_ps();
143 fiy0 = _mm_setzero_ps();
144 fiz0 = _mm_setzero_ps();
145 fix1 = _mm_setzero_ps();
146 fiy1 = _mm_setzero_ps();
147 fiz1 = _mm_setzero_ps();
148 fix2 = _mm_setzero_ps();
149 fiy2 = _mm_setzero_ps();
150 fiz2 = _mm_setzero_ps();
152 /* Reset potential sums */
153 velecsum = _mm_setzero_ps();
155 /* Start inner kernel loop */
156 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
159 /* Get j neighbor index, and coordinate index */
164 j_coord_offsetA = DIM*jnrA;
165 j_coord_offsetB = DIM*jnrB;
166 j_coord_offsetC = DIM*jnrC;
167 j_coord_offsetD = DIM*jnrD;
169 /* load j atom coordinates */
170 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
171 x+j_coord_offsetC,x+j_coord_offsetD,
174 /* Calculate displacement vector */
175 dx00 = _mm_sub_ps(ix0,jx0);
176 dy00 = _mm_sub_ps(iy0,jy0);
177 dz00 = _mm_sub_ps(iz0,jz0);
178 dx10 = _mm_sub_ps(ix1,jx0);
179 dy10 = _mm_sub_ps(iy1,jy0);
180 dz10 = _mm_sub_ps(iz1,jz0);
181 dx20 = _mm_sub_ps(ix2,jx0);
182 dy20 = _mm_sub_ps(iy2,jy0);
183 dz20 = _mm_sub_ps(iz2,jz0);
185 /* Calculate squared distance and things based on it */
186 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
187 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
188 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
190 rinv00 = gmx_mm_invsqrt_ps(rsq00);
191 rinv10 = gmx_mm_invsqrt_ps(rsq10);
192 rinv20 = gmx_mm_invsqrt_ps(rsq20);
194 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
195 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
196 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
198 /* Load parameters for j particles */
199 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
200 charge+jnrC+0,charge+jnrD+0);
202 /**************************
203 * CALCULATE INTERACTIONS *
204 **************************/
206 /* Compute parameters for interactions between i and j atoms */
207 qq00 = _mm_mul_ps(iq0,jq0);
209 /* REACTION-FIELD ELECTROSTATICS */
210 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
211 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
213 /* Update potential sum for this i atom from the interaction with this j atom. */
214 velecsum = _mm_add_ps(velecsum,velec);
218 /* Calculate temporary vectorial force */
219 tx = _mm_mul_ps(fscal,dx00);
220 ty = _mm_mul_ps(fscal,dy00);
221 tz = _mm_mul_ps(fscal,dz00);
223 /* Update vectorial force */
224 fix0 = _mm_add_ps(fix0,tx);
225 fiy0 = _mm_add_ps(fiy0,ty);
226 fiz0 = _mm_add_ps(fiz0,tz);
228 fjptrA = f+j_coord_offsetA;
229 fjptrB = f+j_coord_offsetB;
230 fjptrC = f+j_coord_offsetC;
231 fjptrD = f+j_coord_offsetD;
232 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
234 /**************************
235 * CALCULATE INTERACTIONS *
236 **************************/
238 /* Compute parameters for interactions between i and j atoms */
239 qq10 = _mm_mul_ps(iq1,jq0);
241 /* REACTION-FIELD ELECTROSTATICS */
242 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
243 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
245 /* Update potential sum for this i atom from the interaction with this j atom. */
246 velecsum = _mm_add_ps(velecsum,velec);
250 /* Calculate temporary vectorial force */
251 tx = _mm_mul_ps(fscal,dx10);
252 ty = _mm_mul_ps(fscal,dy10);
253 tz = _mm_mul_ps(fscal,dz10);
255 /* Update vectorial force */
256 fix1 = _mm_add_ps(fix1,tx);
257 fiy1 = _mm_add_ps(fiy1,ty);
258 fiz1 = _mm_add_ps(fiz1,tz);
260 fjptrA = f+j_coord_offsetA;
261 fjptrB = f+j_coord_offsetB;
262 fjptrC = f+j_coord_offsetC;
263 fjptrD = f+j_coord_offsetD;
264 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
266 /**************************
267 * CALCULATE INTERACTIONS *
268 **************************/
270 /* Compute parameters for interactions between i and j atoms */
271 qq20 = _mm_mul_ps(iq2,jq0);
273 /* REACTION-FIELD ELECTROSTATICS */
274 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
275 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
277 /* Update potential sum for this i atom from the interaction with this j atom. */
278 velecsum = _mm_add_ps(velecsum,velec);
282 /* Calculate temporary vectorial force */
283 tx = _mm_mul_ps(fscal,dx20);
284 ty = _mm_mul_ps(fscal,dy20);
285 tz = _mm_mul_ps(fscal,dz20);
287 /* Update vectorial force */
288 fix2 = _mm_add_ps(fix2,tx);
289 fiy2 = _mm_add_ps(fiy2,ty);
290 fiz2 = _mm_add_ps(fiz2,tz);
292 fjptrA = f+j_coord_offsetA;
293 fjptrB = f+j_coord_offsetB;
294 fjptrC = f+j_coord_offsetC;
295 fjptrD = f+j_coord_offsetD;
296 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
298 /* Inner loop uses 96 flops */
304 /* Get j neighbor index, and coordinate index */
305 jnrlistA = jjnr[jidx];
306 jnrlistB = jjnr[jidx+1];
307 jnrlistC = jjnr[jidx+2];
308 jnrlistD = jjnr[jidx+3];
309 /* Sign of each element will be negative for non-real atoms.
310 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
311 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
313 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
314 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
315 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
316 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
317 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
318 j_coord_offsetA = DIM*jnrA;
319 j_coord_offsetB = DIM*jnrB;
320 j_coord_offsetC = DIM*jnrC;
321 j_coord_offsetD = DIM*jnrD;
323 /* load j atom coordinates */
324 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
325 x+j_coord_offsetC,x+j_coord_offsetD,
328 /* Calculate displacement vector */
329 dx00 = _mm_sub_ps(ix0,jx0);
330 dy00 = _mm_sub_ps(iy0,jy0);
331 dz00 = _mm_sub_ps(iz0,jz0);
332 dx10 = _mm_sub_ps(ix1,jx0);
333 dy10 = _mm_sub_ps(iy1,jy0);
334 dz10 = _mm_sub_ps(iz1,jz0);
335 dx20 = _mm_sub_ps(ix2,jx0);
336 dy20 = _mm_sub_ps(iy2,jy0);
337 dz20 = _mm_sub_ps(iz2,jz0);
339 /* Calculate squared distance and things based on it */
340 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
341 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
342 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
344 rinv00 = gmx_mm_invsqrt_ps(rsq00);
345 rinv10 = gmx_mm_invsqrt_ps(rsq10);
346 rinv20 = gmx_mm_invsqrt_ps(rsq20);
348 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
349 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
350 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
352 /* Load parameters for j particles */
353 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
354 charge+jnrC+0,charge+jnrD+0);
356 /**************************
357 * CALCULATE INTERACTIONS *
358 **************************/
360 /* Compute parameters for interactions between i and j atoms */
361 qq00 = _mm_mul_ps(iq0,jq0);
363 /* REACTION-FIELD ELECTROSTATICS */
364 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
365 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
367 /* Update potential sum for this i atom from the interaction with this j atom. */
368 velec = _mm_andnot_ps(dummy_mask,velec);
369 velecsum = _mm_add_ps(velecsum,velec);
373 fscal = _mm_andnot_ps(dummy_mask,fscal);
375 /* Calculate temporary vectorial force */
376 tx = _mm_mul_ps(fscal,dx00);
377 ty = _mm_mul_ps(fscal,dy00);
378 tz = _mm_mul_ps(fscal,dz00);
380 /* Update vectorial force */
381 fix0 = _mm_add_ps(fix0,tx);
382 fiy0 = _mm_add_ps(fiy0,ty);
383 fiz0 = _mm_add_ps(fiz0,tz);
385 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
386 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
387 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
388 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
389 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
391 /**************************
392 * CALCULATE INTERACTIONS *
393 **************************/
395 /* Compute parameters for interactions between i and j atoms */
396 qq10 = _mm_mul_ps(iq1,jq0);
398 /* REACTION-FIELD ELECTROSTATICS */
399 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
400 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
402 /* Update potential sum for this i atom from the interaction with this j atom. */
403 velec = _mm_andnot_ps(dummy_mask,velec);
404 velecsum = _mm_add_ps(velecsum,velec);
408 fscal = _mm_andnot_ps(dummy_mask,fscal);
410 /* Calculate temporary vectorial force */
411 tx = _mm_mul_ps(fscal,dx10);
412 ty = _mm_mul_ps(fscal,dy10);
413 tz = _mm_mul_ps(fscal,dz10);
415 /* Update vectorial force */
416 fix1 = _mm_add_ps(fix1,tx);
417 fiy1 = _mm_add_ps(fiy1,ty);
418 fiz1 = _mm_add_ps(fiz1,tz);
420 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
421 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
422 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
423 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
424 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
426 /**************************
427 * CALCULATE INTERACTIONS *
428 **************************/
430 /* Compute parameters for interactions between i and j atoms */
431 qq20 = _mm_mul_ps(iq2,jq0);
433 /* REACTION-FIELD ELECTROSTATICS */
434 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
435 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
437 /* Update potential sum for this i atom from the interaction with this j atom. */
438 velec = _mm_andnot_ps(dummy_mask,velec);
439 velecsum = _mm_add_ps(velecsum,velec);
443 fscal = _mm_andnot_ps(dummy_mask,fscal);
445 /* Calculate temporary vectorial force */
446 tx = _mm_mul_ps(fscal,dx20);
447 ty = _mm_mul_ps(fscal,dy20);
448 tz = _mm_mul_ps(fscal,dz20);
450 /* Update vectorial force */
451 fix2 = _mm_add_ps(fix2,tx);
452 fiy2 = _mm_add_ps(fiy2,ty);
453 fiz2 = _mm_add_ps(fiz2,tz);
455 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
456 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
457 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
458 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
459 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
461 /* Inner loop uses 96 flops */
464 /* End of innermost loop */
466 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
467 f+i_coord_offset,fshift+i_shift_offset);
470 /* Update potential energies */
471 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
473 /* Increment number of inner iterations */
474 inneriter += j_index_end - j_index_start;
476 /* Outer loop uses 19 flops */
479 /* Increment number of outer iterations */
482 /* Update outer/inner flops */
484 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*96);
487 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwNone_GeomW3P1_F_sse4_1_single
488 * Electrostatics interaction: ReactionField
489 * VdW interaction: None
490 * Geometry: Water3-Particle
491 * Calculate force/pot: Force
494 nb_kernel_ElecRF_VdwNone_GeomW3P1_F_sse4_1_single
495 (t_nblist * gmx_restrict nlist,
496 rvec * gmx_restrict xx,
497 rvec * gmx_restrict ff,
498 t_forcerec * gmx_restrict fr,
499 t_mdatoms * gmx_restrict mdatoms,
500 nb_kernel_data_t * gmx_restrict kernel_data,
501 t_nrnb * gmx_restrict nrnb)
503 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
504 * just 0 for non-waters.
505 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
506 * jnr indices corresponding to data put in the four positions in the SIMD register.
508 int i_shift_offset,i_coord_offset,outeriter,inneriter;
509 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
510 int jnrA,jnrB,jnrC,jnrD;
511 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
512 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
513 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
515 real *shiftvec,*fshift,*x,*f;
516 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
518 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
520 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
522 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
524 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
525 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
526 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
527 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
528 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
529 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
530 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
532 __m128 dummy_mask,cutoff_mask;
533 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
534 __m128 one = _mm_set1_ps(1.0);
535 __m128 two = _mm_set1_ps(2.0);
541 jindex = nlist->jindex;
543 shiftidx = nlist->shift;
545 shiftvec = fr->shift_vec[0];
546 fshift = fr->fshift[0];
547 facel = _mm_set1_ps(fr->epsfac);
548 charge = mdatoms->chargeA;
549 krf = _mm_set1_ps(fr->ic->k_rf);
550 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
551 crf = _mm_set1_ps(fr->ic->c_rf);
553 /* Setup water-specific parameters */
554 inr = nlist->iinr[0];
555 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
556 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
557 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
559 /* Avoid stupid compiler warnings */
560 jnrA = jnrB = jnrC = jnrD = 0;
569 for(iidx=0;iidx<4*DIM;iidx++)
574 /* Start outer loop over neighborlists */
575 for(iidx=0; iidx<nri; iidx++)
577 /* Load shift vector for this list */
578 i_shift_offset = DIM*shiftidx[iidx];
580 /* Load limits for loop over neighbors */
581 j_index_start = jindex[iidx];
582 j_index_end = jindex[iidx+1];
584 /* Get outer coordinate index */
586 i_coord_offset = DIM*inr;
588 /* Load i particle coords and add shift vector */
589 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
590 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
592 fix0 = _mm_setzero_ps();
593 fiy0 = _mm_setzero_ps();
594 fiz0 = _mm_setzero_ps();
595 fix1 = _mm_setzero_ps();
596 fiy1 = _mm_setzero_ps();
597 fiz1 = _mm_setzero_ps();
598 fix2 = _mm_setzero_ps();
599 fiy2 = _mm_setzero_ps();
600 fiz2 = _mm_setzero_ps();
602 /* Start inner kernel loop */
603 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
606 /* Get j neighbor index, and coordinate index */
611 j_coord_offsetA = DIM*jnrA;
612 j_coord_offsetB = DIM*jnrB;
613 j_coord_offsetC = DIM*jnrC;
614 j_coord_offsetD = DIM*jnrD;
616 /* load j atom coordinates */
617 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
618 x+j_coord_offsetC,x+j_coord_offsetD,
621 /* Calculate displacement vector */
622 dx00 = _mm_sub_ps(ix0,jx0);
623 dy00 = _mm_sub_ps(iy0,jy0);
624 dz00 = _mm_sub_ps(iz0,jz0);
625 dx10 = _mm_sub_ps(ix1,jx0);
626 dy10 = _mm_sub_ps(iy1,jy0);
627 dz10 = _mm_sub_ps(iz1,jz0);
628 dx20 = _mm_sub_ps(ix2,jx0);
629 dy20 = _mm_sub_ps(iy2,jy0);
630 dz20 = _mm_sub_ps(iz2,jz0);
632 /* Calculate squared distance and things based on it */
633 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
634 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
635 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
637 rinv00 = gmx_mm_invsqrt_ps(rsq00);
638 rinv10 = gmx_mm_invsqrt_ps(rsq10);
639 rinv20 = gmx_mm_invsqrt_ps(rsq20);
641 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
642 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
643 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
645 /* Load parameters for j particles */
646 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
647 charge+jnrC+0,charge+jnrD+0);
649 /**************************
650 * CALCULATE INTERACTIONS *
651 **************************/
653 /* Compute parameters for interactions between i and j atoms */
654 qq00 = _mm_mul_ps(iq0,jq0);
656 /* REACTION-FIELD ELECTROSTATICS */
657 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
661 /* Calculate temporary vectorial force */
662 tx = _mm_mul_ps(fscal,dx00);
663 ty = _mm_mul_ps(fscal,dy00);
664 tz = _mm_mul_ps(fscal,dz00);
666 /* Update vectorial force */
667 fix0 = _mm_add_ps(fix0,tx);
668 fiy0 = _mm_add_ps(fiy0,ty);
669 fiz0 = _mm_add_ps(fiz0,tz);
671 fjptrA = f+j_coord_offsetA;
672 fjptrB = f+j_coord_offsetB;
673 fjptrC = f+j_coord_offsetC;
674 fjptrD = f+j_coord_offsetD;
675 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
677 /**************************
678 * CALCULATE INTERACTIONS *
679 **************************/
681 /* Compute parameters for interactions between i and j atoms */
682 qq10 = _mm_mul_ps(iq1,jq0);
684 /* REACTION-FIELD ELECTROSTATICS */
685 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
689 /* Calculate temporary vectorial force */
690 tx = _mm_mul_ps(fscal,dx10);
691 ty = _mm_mul_ps(fscal,dy10);
692 tz = _mm_mul_ps(fscal,dz10);
694 /* Update vectorial force */
695 fix1 = _mm_add_ps(fix1,tx);
696 fiy1 = _mm_add_ps(fiy1,ty);
697 fiz1 = _mm_add_ps(fiz1,tz);
699 fjptrA = f+j_coord_offsetA;
700 fjptrB = f+j_coord_offsetB;
701 fjptrC = f+j_coord_offsetC;
702 fjptrD = f+j_coord_offsetD;
703 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
705 /**************************
706 * CALCULATE INTERACTIONS *
707 **************************/
709 /* Compute parameters for interactions between i and j atoms */
710 qq20 = _mm_mul_ps(iq2,jq0);
712 /* REACTION-FIELD ELECTROSTATICS */
713 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
717 /* Calculate temporary vectorial force */
718 tx = _mm_mul_ps(fscal,dx20);
719 ty = _mm_mul_ps(fscal,dy20);
720 tz = _mm_mul_ps(fscal,dz20);
722 /* Update vectorial force */
723 fix2 = _mm_add_ps(fix2,tx);
724 fiy2 = _mm_add_ps(fiy2,ty);
725 fiz2 = _mm_add_ps(fiz2,tz);
727 fjptrA = f+j_coord_offsetA;
728 fjptrB = f+j_coord_offsetB;
729 fjptrC = f+j_coord_offsetC;
730 fjptrD = f+j_coord_offsetD;
731 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
733 /* Inner loop uses 81 flops */
739 /* Get j neighbor index, and coordinate index */
740 jnrlistA = jjnr[jidx];
741 jnrlistB = jjnr[jidx+1];
742 jnrlistC = jjnr[jidx+2];
743 jnrlistD = jjnr[jidx+3];
744 /* Sign of each element will be negative for non-real atoms.
745 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
746 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
748 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
749 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
750 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
751 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
752 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
753 j_coord_offsetA = DIM*jnrA;
754 j_coord_offsetB = DIM*jnrB;
755 j_coord_offsetC = DIM*jnrC;
756 j_coord_offsetD = DIM*jnrD;
758 /* load j atom coordinates */
759 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
760 x+j_coord_offsetC,x+j_coord_offsetD,
763 /* Calculate displacement vector */
764 dx00 = _mm_sub_ps(ix0,jx0);
765 dy00 = _mm_sub_ps(iy0,jy0);
766 dz00 = _mm_sub_ps(iz0,jz0);
767 dx10 = _mm_sub_ps(ix1,jx0);
768 dy10 = _mm_sub_ps(iy1,jy0);
769 dz10 = _mm_sub_ps(iz1,jz0);
770 dx20 = _mm_sub_ps(ix2,jx0);
771 dy20 = _mm_sub_ps(iy2,jy0);
772 dz20 = _mm_sub_ps(iz2,jz0);
774 /* Calculate squared distance and things based on it */
775 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
776 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
777 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
779 rinv00 = gmx_mm_invsqrt_ps(rsq00);
780 rinv10 = gmx_mm_invsqrt_ps(rsq10);
781 rinv20 = gmx_mm_invsqrt_ps(rsq20);
783 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
784 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
785 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
787 /* Load parameters for j particles */
788 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
789 charge+jnrC+0,charge+jnrD+0);
791 /**************************
792 * CALCULATE INTERACTIONS *
793 **************************/
795 /* Compute parameters for interactions between i and j atoms */
796 qq00 = _mm_mul_ps(iq0,jq0);
798 /* REACTION-FIELD ELECTROSTATICS */
799 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
803 fscal = _mm_andnot_ps(dummy_mask,fscal);
805 /* Calculate temporary vectorial force */
806 tx = _mm_mul_ps(fscal,dx00);
807 ty = _mm_mul_ps(fscal,dy00);
808 tz = _mm_mul_ps(fscal,dz00);
810 /* Update vectorial force */
811 fix0 = _mm_add_ps(fix0,tx);
812 fiy0 = _mm_add_ps(fiy0,ty);
813 fiz0 = _mm_add_ps(fiz0,tz);
815 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
816 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
817 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
818 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
819 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
821 /**************************
822 * CALCULATE INTERACTIONS *
823 **************************/
825 /* Compute parameters for interactions between i and j atoms */
826 qq10 = _mm_mul_ps(iq1,jq0);
828 /* REACTION-FIELD ELECTROSTATICS */
829 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
833 fscal = _mm_andnot_ps(dummy_mask,fscal);
835 /* Calculate temporary vectorial force */
836 tx = _mm_mul_ps(fscal,dx10);
837 ty = _mm_mul_ps(fscal,dy10);
838 tz = _mm_mul_ps(fscal,dz10);
840 /* Update vectorial force */
841 fix1 = _mm_add_ps(fix1,tx);
842 fiy1 = _mm_add_ps(fiy1,ty);
843 fiz1 = _mm_add_ps(fiz1,tz);
845 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
846 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
847 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
848 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
849 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
851 /**************************
852 * CALCULATE INTERACTIONS *
853 **************************/
855 /* Compute parameters for interactions between i and j atoms */
856 qq20 = _mm_mul_ps(iq2,jq0);
858 /* REACTION-FIELD ELECTROSTATICS */
859 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
863 fscal = _mm_andnot_ps(dummy_mask,fscal);
865 /* Calculate temporary vectorial force */
866 tx = _mm_mul_ps(fscal,dx20);
867 ty = _mm_mul_ps(fscal,dy20);
868 tz = _mm_mul_ps(fscal,dz20);
870 /* Update vectorial force */
871 fix2 = _mm_add_ps(fix2,tx);
872 fiy2 = _mm_add_ps(fiy2,ty);
873 fiz2 = _mm_add_ps(fiz2,tz);
875 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
876 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
877 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
878 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
879 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
881 /* Inner loop uses 81 flops */
884 /* End of innermost loop */
886 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
887 f+i_coord_offset,fshift+i_shift_offset);
889 /* Increment number of inner iterations */
890 inneriter += j_index_end - j_index_start;
892 /* Outer loop uses 18 flops */
895 /* Increment number of outer iterations */
898 /* Update outer/inner flops */
900 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*81);