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 sse2_single kernel generator.
44 #include "../nb_kernel.h"
45 #include "gromacs/legacyheaders/types/simple.h"
46 #include "gromacs/math/vec.h"
47 #include "gromacs/legacyheaders/nrnb.h"
49 #include "gromacs/simd/math_x86_sse2_single.h"
50 #include "kernelutil_x86_sse2_single.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwLJ_GeomW3P1_VF_sse2_single
54 * Electrostatics interaction: ReactionField
55 * VdW interaction: LennardJones
56 * Geometry: Water3-Particle
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecRF_VdwLJ_GeomW3P1_VF_sse2_single
61 (t_nblist * gmx_restrict nlist,
62 rvec * gmx_restrict xx,
63 rvec * gmx_restrict ff,
64 t_forcerec * gmx_restrict fr,
65 t_mdatoms * gmx_restrict mdatoms,
66 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
67 t_nrnb * gmx_restrict nrnb)
69 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
70 * just 0 for non-waters.
71 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
72 * jnr indices corresponding to data put in the four positions in the SIMD register.
74 int i_shift_offset,i_coord_offset,outeriter,inneriter;
75 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
76 int jnrA,jnrB,jnrC,jnrD;
77 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
78 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
79 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
81 real *shiftvec,*fshift,*x,*f;
82 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
84 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
86 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
88 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
90 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
91 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
92 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
93 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
94 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
95 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
96 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
99 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
102 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
103 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
104 __m128 dummy_mask,cutoff_mask;
105 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
106 __m128 one = _mm_set1_ps(1.0);
107 __m128 two = _mm_set1_ps(2.0);
113 jindex = nlist->jindex;
115 shiftidx = nlist->shift;
117 shiftvec = fr->shift_vec[0];
118 fshift = fr->fshift[0];
119 facel = _mm_set1_ps(fr->epsfac);
120 charge = mdatoms->chargeA;
121 krf = _mm_set1_ps(fr->ic->k_rf);
122 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
123 crf = _mm_set1_ps(fr->ic->c_rf);
124 nvdwtype = fr->ntype;
126 vdwtype = mdatoms->typeA;
128 /* Setup water-specific parameters */
129 inr = nlist->iinr[0];
130 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
131 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
132 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
133 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
135 /* Avoid stupid compiler warnings */
136 jnrA = jnrB = jnrC = jnrD = 0;
145 for(iidx=0;iidx<4*DIM;iidx++)
150 /* Start outer loop over neighborlists */
151 for(iidx=0; iidx<nri; iidx++)
153 /* Load shift vector for this list */
154 i_shift_offset = DIM*shiftidx[iidx];
156 /* Load limits for loop over neighbors */
157 j_index_start = jindex[iidx];
158 j_index_end = jindex[iidx+1];
160 /* Get outer coordinate index */
162 i_coord_offset = DIM*inr;
164 /* Load i particle coords and add shift vector */
165 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
166 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
168 fix0 = _mm_setzero_ps();
169 fiy0 = _mm_setzero_ps();
170 fiz0 = _mm_setzero_ps();
171 fix1 = _mm_setzero_ps();
172 fiy1 = _mm_setzero_ps();
173 fiz1 = _mm_setzero_ps();
174 fix2 = _mm_setzero_ps();
175 fiy2 = _mm_setzero_ps();
176 fiz2 = _mm_setzero_ps();
178 /* Reset potential sums */
179 velecsum = _mm_setzero_ps();
180 vvdwsum = _mm_setzero_ps();
182 /* Start inner kernel loop */
183 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
186 /* Get j neighbor index, and coordinate index */
191 j_coord_offsetA = DIM*jnrA;
192 j_coord_offsetB = DIM*jnrB;
193 j_coord_offsetC = DIM*jnrC;
194 j_coord_offsetD = DIM*jnrD;
196 /* load j atom coordinates */
197 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
198 x+j_coord_offsetC,x+j_coord_offsetD,
201 /* Calculate displacement vector */
202 dx00 = _mm_sub_ps(ix0,jx0);
203 dy00 = _mm_sub_ps(iy0,jy0);
204 dz00 = _mm_sub_ps(iz0,jz0);
205 dx10 = _mm_sub_ps(ix1,jx0);
206 dy10 = _mm_sub_ps(iy1,jy0);
207 dz10 = _mm_sub_ps(iz1,jz0);
208 dx20 = _mm_sub_ps(ix2,jx0);
209 dy20 = _mm_sub_ps(iy2,jy0);
210 dz20 = _mm_sub_ps(iz2,jz0);
212 /* Calculate squared distance and things based on it */
213 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
214 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
215 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
217 rinv00 = gmx_mm_invsqrt_ps(rsq00);
218 rinv10 = gmx_mm_invsqrt_ps(rsq10);
219 rinv20 = gmx_mm_invsqrt_ps(rsq20);
221 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
222 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
223 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
225 /* Load parameters for j particles */
226 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
227 charge+jnrC+0,charge+jnrD+0);
228 vdwjidx0A = 2*vdwtype[jnrA+0];
229 vdwjidx0B = 2*vdwtype[jnrB+0];
230 vdwjidx0C = 2*vdwtype[jnrC+0];
231 vdwjidx0D = 2*vdwtype[jnrD+0];
233 fjx0 = _mm_setzero_ps();
234 fjy0 = _mm_setzero_ps();
235 fjz0 = _mm_setzero_ps();
237 /**************************
238 * CALCULATE INTERACTIONS *
239 **************************/
241 /* Compute parameters for interactions between i and j atoms */
242 qq00 = _mm_mul_ps(iq0,jq0);
243 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
244 vdwparam+vdwioffset0+vdwjidx0B,
245 vdwparam+vdwioffset0+vdwjidx0C,
246 vdwparam+vdwioffset0+vdwjidx0D,
249 /* REACTION-FIELD ELECTROSTATICS */
250 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
251 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
253 /* LENNARD-JONES DISPERSION/REPULSION */
255 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
256 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
257 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
258 vvdw = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
259 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
261 /* Update potential sum for this i atom from the interaction with this j atom. */
262 velecsum = _mm_add_ps(velecsum,velec);
263 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
265 fscal = _mm_add_ps(felec,fvdw);
267 /* Calculate temporary vectorial force */
268 tx = _mm_mul_ps(fscal,dx00);
269 ty = _mm_mul_ps(fscal,dy00);
270 tz = _mm_mul_ps(fscal,dz00);
272 /* Update vectorial force */
273 fix0 = _mm_add_ps(fix0,tx);
274 fiy0 = _mm_add_ps(fiy0,ty);
275 fiz0 = _mm_add_ps(fiz0,tz);
277 fjx0 = _mm_add_ps(fjx0,tx);
278 fjy0 = _mm_add_ps(fjy0,ty);
279 fjz0 = _mm_add_ps(fjz0,tz);
281 /**************************
282 * CALCULATE INTERACTIONS *
283 **************************/
285 /* Compute parameters for interactions between i and j atoms */
286 qq10 = _mm_mul_ps(iq1,jq0);
288 /* REACTION-FIELD ELECTROSTATICS */
289 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
290 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
292 /* Update potential sum for this i atom from the interaction with this j atom. */
293 velecsum = _mm_add_ps(velecsum,velec);
297 /* Calculate temporary vectorial force */
298 tx = _mm_mul_ps(fscal,dx10);
299 ty = _mm_mul_ps(fscal,dy10);
300 tz = _mm_mul_ps(fscal,dz10);
302 /* Update vectorial force */
303 fix1 = _mm_add_ps(fix1,tx);
304 fiy1 = _mm_add_ps(fiy1,ty);
305 fiz1 = _mm_add_ps(fiz1,tz);
307 fjx0 = _mm_add_ps(fjx0,tx);
308 fjy0 = _mm_add_ps(fjy0,ty);
309 fjz0 = _mm_add_ps(fjz0,tz);
311 /**************************
312 * CALCULATE INTERACTIONS *
313 **************************/
315 /* Compute parameters for interactions between i and j atoms */
316 qq20 = _mm_mul_ps(iq2,jq0);
318 /* REACTION-FIELD ELECTROSTATICS */
319 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
320 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
322 /* Update potential sum for this i atom from the interaction with this j atom. */
323 velecsum = _mm_add_ps(velecsum,velec);
327 /* Calculate temporary vectorial force */
328 tx = _mm_mul_ps(fscal,dx20);
329 ty = _mm_mul_ps(fscal,dy20);
330 tz = _mm_mul_ps(fscal,dz20);
332 /* Update vectorial force */
333 fix2 = _mm_add_ps(fix2,tx);
334 fiy2 = _mm_add_ps(fiy2,ty);
335 fiz2 = _mm_add_ps(fiz2,tz);
337 fjx0 = _mm_add_ps(fjx0,tx);
338 fjy0 = _mm_add_ps(fjy0,ty);
339 fjz0 = _mm_add_ps(fjz0,tz);
341 fjptrA = f+j_coord_offsetA;
342 fjptrB = f+j_coord_offsetB;
343 fjptrC = f+j_coord_offsetC;
344 fjptrD = f+j_coord_offsetD;
346 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
348 /* Inner loop uses 108 flops */
354 /* Get j neighbor index, and coordinate index */
355 jnrlistA = jjnr[jidx];
356 jnrlistB = jjnr[jidx+1];
357 jnrlistC = jjnr[jidx+2];
358 jnrlistD = jjnr[jidx+3];
359 /* Sign of each element will be negative for non-real atoms.
360 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
361 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
363 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
364 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
365 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
366 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
367 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
368 j_coord_offsetA = DIM*jnrA;
369 j_coord_offsetB = DIM*jnrB;
370 j_coord_offsetC = DIM*jnrC;
371 j_coord_offsetD = DIM*jnrD;
373 /* load j atom coordinates */
374 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
375 x+j_coord_offsetC,x+j_coord_offsetD,
378 /* Calculate displacement vector */
379 dx00 = _mm_sub_ps(ix0,jx0);
380 dy00 = _mm_sub_ps(iy0,jy0);
381 dz00 = _mm_sub_ps(iz0,jz0);
382 dx10 = _mm_sub_ps(ix1,jx0);
383 dy10 = _mm_sub_ps(iy1,jy0);
384 dz10 = _mm_sub_ps(iz1,jz0);
385 dx20 = _mm_sub_ps(ix2,jx0);
386 dy20 = _mm_sub_ps(iy2,jy0);
387 dz20 = _mm_sub_ps(iz2,jz0);
389 /* Calculate squared distance and things based on it */
390 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
391 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
392 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
394 rinv00 = gmx_mm_invsqrt_ps(rsq00);
395 rinv10 = gmx_mm_invsqrt_ps(rsq10);
396 rinv20 = gmx_mm_invsqrt_ps(rsq20);
398 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
399 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
400 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
402 /* Load parameters for j particles */
403 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
404 charge+jnrC+0,charge+jnrD+0);
405 vdwjidx0A = 2*vdwtype[jnrA+0];
406 vdwjidx0B = 2*vdwtype[jnrB+0];
407 vdwjidx0C = 2*vdwtype[jnrC+0];
408 vdwjidx0D = 2*vdwtype[jnrD+0];
410 fjx0 = _mm_setzero_ps();
411 fjy0 = _mm_setzero_ps();
412 fjz0 = _mm_setzero_ps();
414 /**************************
415 * CALCULATE INTERACTIONS *
416 **************************/
418 /* Compute parameters for interactions between i and j atoms */
419 qq00 = _mm_mul_ps(iq0,jq0);
420 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
421 vdwparam+vdwioffset0+vdwjidx0B,
422 vdwparam+vdwioffset0+vdwjidx0C,
423 vdwparam+vdwioffset0+vdwjidx0D,
426 /* REACTION-FIELD ELECTROSTATICS */
427 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
428 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
430 /* LENNARD-JONES DISPERSION/REPULSION */
432 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
433 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
434 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
435 vvdw = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
436 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
438 /* Update potential sum for this i atom from the interaction with this j atom. */
439 velec = _mm_andnot_ps(dummy_mask,velec);
440 velecsum = _mm_add_ps(velecsum,velec);
441 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
442 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
444 fscal = _mm_add_ps(felec,fvdw);
446 fscal = _mm_andnot_ps(dummy_mask,fscal);
448 /* Calculate temporary vectorial force */
449 tx = _mm_mul_ps(fscal,dx00);
450 ty = _mm_mul_ps(fscal,dy00);
451 tz = _mm_mul_ps(fscal,dz00);
453 /* Update vectorial force */
454 fix0 = _mm_add_ps(fix0,tx);
455 fiy0 = _mm_add_ps(fiy0,ty);
456 fiz0 = _mm_add_ps(fiz0,tz);
458 fjx0 = _mm_add_ps(fjx0,tx);
459 fjy0 = _mm_add_ps(fjy0,ty);
460 fjz0 = _mm_add_ps(fjz0,tz);
462 /**************************
463 * CALCULATE INTERACTIONS *
464 **************************/
466 /* Compute parameters for interactions between i and j atoms */
467 qq10 = _mm_mul_ps(iq1,jq0);
469 /* REACTION-FIELD ELECTROSTATICS */
470 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
471 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
473 /* Update potential sum for this i atom from the interaction with this j atom. */
474 velec = _mm_andnot_ps(dummy_mask,velec);
475 velecsum = _mm_add_ps(velecsum,velec);
479 fscal = _mm_andnot_ps(dummy_mask,fscal);
481 /* Calculate temporary vectorial force */
482 tx = _mm_mul_ps(fscal,dx10);
483 ty = _mm_mul_ps(fscal,dy10);
484 tz = _mm_mul_ps(fscal,dz10);
486 /* Update vectorial force */
487 fix1 = _mm_add_ps(fix1,tx);
488 fiy1 = _mm_add_ps(fiy1,ty);
489 fiz1 = _mm_add_ps(fiz1,tz);
491 fjx0 = _mm_add_ps(fjx0,tx);
492 fjy0 = _mm_add_ps(fjy0,ty);
493 fjz0 = _mm_add_ps(fjz0,tz);
495 /**************************
496 * CALCULATE INTERACTIONS *
497 **************************/
499 /* Compute parameters for interactions between i and j atoms */
500 qq20 = _mm_mul_ps(iq2,jq0);
502 /* REACTION-FIELD ELECTROSTATICS */
503 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
504 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
506 /* Update potential sum for this i atom from the interaction with this j atom. */
507 velec = _mm_andnot_ps(dummy_mask,velec);
508 velecsum = _mm_add_ps(velecsum,velec);
512 fscal = _mm_andnot_ps(dummy_mask,fscal);
514 /* Calculate temporary vectorial force */
515 tx = _mm_mul_ps(fscal,dx20);
516 ty = _mm_mul_ps(fscal,dy20);
517 tz = _mm_mul_ps(fscal,dz20);
519 /* Update vectorial force */
520 fix2 = _mm_add_ps(fix2,tx);
521 fiy2 = _mm_add_ps(fiy2,ty);
522 fiz2 = _mm_add_ps(fiz2,tz);
524 fjx0 = _mm_add_ps(fjx0,tx);
525 fjy0 = _mm_add_ps(fjy0,ty);
526 fjz0 = _mm_add_ps(fjz0,tz);
528 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
529 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
530 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
531 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
533 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
535 /* Inner loop uses 108 flops */
538 /* End of innermost loop */
540 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
541 f+i_coord_offset,fshift+i_shift_offset);
544 /* Update potential energies */
545 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
546 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
548 /* Increment number of inner iterations */
549 inneriter += j_index_end - j_index_start;
551 /* Outer loop uses 20 flops */
554 /* Increment number of outer iterations */
557 /* Update outer/inner flops */
559 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*108);
562 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwLJ_GeomW3P1_F_sse2_single
563 * Electrostatics interaction: ReactionField
564 * VdW interaction: LennardJones
565 * Geometry: Water3-Particle
566 * Calculate force/pot: Force
569 nb_kernel_ElecRF_VdwLJ_GeomW3P1_F_sse2_single
570 (t_nblist * gmx_restrict nlist,
571 rvec * gmx_restrict xx,
572 rvec * gmx_restrict ff,
573 t_forcerec * gmx_restrict fr,
574 t_mdatoms * gmx_restrict mdatoms,
575 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
576 t_nrnb * gmx_restrict nrnb)
578 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
579 * just 0 for non-waters.
580 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
581 * jnr indices corresponding to data put in the four positions in the SIMD register.
583 int i_shift_offset,i_coord_offset,outeriter,inneriter;
584 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
585 int jnrA,jnrB,jnrC,jnrD;
586 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
587 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
588 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
590 real *shiftvec,*fshift,*x,*f;
591 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
593 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
595 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
597 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
599 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
600 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
601 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
602 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
603 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
604 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
605 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
608 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
611 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
612 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
613 __m128 dummy_mask,cutoff_mask;
614 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
615 __m128 one = _mm_set1_ps(1.0);
616 __m128 two = _mm_set1_ps(2.0);
622 jindex = nlist->jindex;
624 shiftidx = nlist->shift;
626 shiftvec = fr->shift_vec[0];
627 fshift = fr->fshift[0];
628 facel = _mm_set1_ps(fr->epsfac);
629 charge = mdatoms->chargeA;
630 krf = _mm_set1_ps(fr->ic->k_rf);
631 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
632 crf = _mm_set1_ps(fr->ic->c_rf);
633 nvdwtype = fr->ntype;
635 vdwtype = mdatoms->typeA;
637 /* Setup water-specific parameters */
638 inr = nlist->iinr[0];
639 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
640 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
641 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
642 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
644 /* Avoid stupid compiler warnings */
645 jnrA = jnrB = jnrC = jnrD = 0;
654 for(iidx=0;iidx<4*DIM;iidx++)
659 /* Start outer loop over neighborlists */
660 for(iidx=0; iidx<nri; iidx++)
662 /* Load shift vector for this list */
663 i_shift_offset = DIM*shiftidx[iidx];
665 /* Load limits for loop over neighbors */
666 j_index_start = jindex[iidx];
667 j_index_end = jindex[iidx+1];
669 /* Get outer coordinate index */
671 i_coord_offset = DIM*inr;
673 /* Load i particle coords and add shift vector */
674 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
675 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
677 fix0 = _mm_setzero_ps();
678 fiy0 = _mm_setzero_ps();
679 fiz0 = _mm_setzero_ps();
680 fix1 = _mm_setzero_ps();
681 fiy1 = _mm_setzero_ps();
682 fiz1 = _mm_setzero_ps();
683 fix2 = _mm_setzero_ps();
684 fiy2 = _mm_setzero_ps();
685 fiz2 = _mm_setzero_ps();
687 /* Start inner kernel loop */
688 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
691 /* Get j neighbor index, and coordinate index */
696 j_coord_offsetA = DIM*jnrA;
697 j_coord_offsetB = DIM*jnrB;
698 j_coord_offsetC = DIM*jnrC;
699 j_coord_offsetD = DIM*jnrD;
701 /* load j atom coordinates */
702 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
703 x+j_coord_offsetC,x+j_coord_offsetD,
706 /* Calculate displacement vector */
707 dx00 = _mm_sub_ps(ix0,jx0);
708 dy00 = _mm_sub_ps(iy0,jy0);
709 dz00 = _mm_sub_ps(iz0,jz0);
710 dx10 = _mm_sub_ps(ix1,jx0);
711 dy10 = _mm_sub_ps(iy1,jy0);
712 dz10 = _mm_sub_ps(iz1,jz0);
713 dx20 = _mm_sub_ps(ix2,jx0);
714 dy20 = _mm_sub_ps(iy2,jy0);
715 dz20 = _mm_sub_ps(iz2,jz0);
717 /* Calculate squared distance and things based on it */
718 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
719 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
720 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
722 rinv00 = gmx_mm_invsqrt_ps(rsq00);
723 rinv10 = gmx_mm_invsqrt_ps(rsq10);
724 rinv20 = gmx_mm_invsqrt_ps(rsq20);
726 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
727 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
728 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
730 /* Load parameters for j particles */
731 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
732 charge+jnrC+0,charge+jnrD+0);
733 vdwjidx0A = 2*vdwtype[jnrA+0];
734 vdwjidx0B = 2*vdwtype[jnrB+0];
735 vdwjidx0C = 2*vdwtype[jnrC+0];
736 vdwjidx0D = 2*vdwtype[jnrD+0];
738 fjx0 = _mm_setzero_ps();
739 fjy0 = _mm_setzero_ps();
740 fjz0 = _mm_setzero_ps();
742 /**************************
743 * CALCULATE INTERACTIONS *
744 **************************/
746 /* Compute parameters for interactions between i and j atoms */
747 qq00 = _mm_mul_ps(iq0,jq0);
748 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
749 vdwparam+vdwioffset0+vdwjidx0B,
750 vdwparam+vdwioffset0+vdwjidx0C,
751 vdwparam+vdwioffset0+vdwjidx0D,
754 /* REACTION-FIELD ELECTROSTATICS */
755 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
757 /* LENNARD-JONES DISPERSION/REPULSION */
759 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
760 fvdw = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
762 fscal = _mm_add_ps(felec,fvdw);
764 /* Calculate temporary vectorial force */
765 tx = _mm_mul_ps(fscal,dx00);
766 ty = _mm_mul_ps(fscal,dy00);
767 tz = _mm_mul_ps(fscal,dz00);
769 /* Update vectorial force */
770 fix0 = _mm_add_ps(fix0,tx);
771 fiy0 = _mm_add_ps(fiy0,ty);
772 fiz0 = _mm_add_ps(fiz0,tz);
774 fjx0 = _mm_add_ps(fjx0,tx);
775 fjy0 = _mm_add_ps(fjy0,ty);
776 fjz0 = _mm_add_ps(fjz0,tz);
778 /**************************
779 * CALCULATE INTERACTIONS *
780 **************************/
782 /* Compute parameters for interactions between i and j atoms */
783 qq10 = _mm_mul_ps(iq1,jq0);
785 /* REACTION-FIELD ELECTROSTATICS */
786 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
790 /* Calculate temporary vectorial force */
791 tx = _mm_mul_ps(fscal,dx10);
792 ty = _mm_mul_ps(fscal,dy10);
793 tz = _mm_mul_ps(fscal,dz10);
795 /* Update vectorial force */
796 fix1 = _mm_add_ps(fix1,tx);
797 fiy1 = _mm_add_ps(fiy1,ty);
798 fiz1 = _mm_add_ps(fiz1,tz);
800 fjx0 = _mm_add_ps(fjx0,tx);
801 fjy0 = _mm_add_ps(fjy0,ty);
802 fjz0 = _mm_add_ps(fjz0,tz);
804 /**************************
805 * CALCULATE INTERACTIONS *
806 **************************/
808 /* Compute parameters for interactions between i and j atoms */
809 qq20 = _mm_mul_ps(iq2,jq0);
811 /* REACTION-FIELD ELECTROSTATICS */
812 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
816 /* Calculate temporary vectorial force */
817 tx = _mm_mul_ps(fscal,dx20);
818 ty = _mm_mul_ps(fscal,dy20);
819 tz = _mm_mul_ps(fscal,dz20);
821 /* Update vectorial force */
822 fix2 = _mm_add_ps(fix2,tx);
823 fiy2 = _mm_add_ps(fiy2,ty);
824 fiz2 = _mm_add_ps(fiz2,tz);
826 fjx0 = _mm_add_ps(fjx0,tx);
827 fjy0 = _mm_add_ps(fjy0,ty);
828 fjz0 = _mm_add_ps(fjz0,tz);
830 fjptrA = f+j_coord_offsetA;
831 fjptrB = f+j_coord_offsetB;
832 fjptrC = f+j_coord_offsetC;
833 fjptrD = f+j_coord_offsetD;
835 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
837 /* Inner loop uses 88 flops */
843 /* Get j neighbor index, and coordinate index */
844 jnrlistA = jjnr[jidx];
845 jnrlistB = jjnr[jidx+1];
846 jnrlistC = jjnr[jidx+2];
847 jnrlistD = jjnr[jidx+3];
848 /* Sign of each element will be negative for non-real atoms.
849 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
850 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
852 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
853 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
854 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
855 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
856 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
857 j_coord_offsetA = DIM*jnrA;
858 j_coord_offsetB = DIM*jnrB;
859 j_coord_offsetC = DIM*jnrC;
860 j_coord_offsetD = DIM*jnrD;
862 /* load j atom coordinates */
863 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
864 x+j_coord_offsetC,x+j_coord_offsetD,
867 /* Calculate displacement vector */
868 dx00 = _mm_sub_ps(ix0,jx0);
869 dy00 = _mm_sub_ps(iy0,jy0);
870 dz00 = _mm_sub_ps(iz0,jz0);
871 dx10 = _mm_sub_ps(ix1,jx0);
872 dy10 = _mm_sub_ps(iy1,jy0);
873 dz10 = _mm_sub_ps(iz1,jz0);
874 dx20 = _mm_sub_ps(ix2,jx0);
875 dy20 = _mm_sub_ps(iy2,jy0);
876 dz20 = _mm_sub_ps(iz2,jz0);
878 /* Calculate squared distance and things based on it */
879 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
880 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
881 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
883 rinv00 = gmx_mm_invsqrt_ps(rsq00);
884 rinv10 = gmx_mm_invsqrt_ps(rsq10);
885 rinv20 = gmx_mm_invsqrt_ps(rsq20);
887 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
888 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
889 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
891 /* Load parameters for j particles */
892 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
893 charge+jnrC+0,charge+jnrD+0);
894 vdwjidx0A = 2*vdwtype[jnrA+0];
895 vdwjidx0B = 2*vdwtype[jnrB+0];
896 vdwjidx0C = 2*vdwtype[jnrC+0];
897 vdwjidx0D = 2*vdwtype[jnrD+0];
899 fjx0 = _mm_setzero_ps();
900 fjy0 = _mm_setzero_ps();
901 fjz0 = _mm_setzero_ps();
903 /**************************
904 * CALCULATE INTERACTIONS *
905 **************************/
907 /* Compute parameters for interactions between i and j atoms */
908 qq00 = _mm_mul_ps(iq0,jq0);
909 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
910 vdwparam+vdwioffset0+vdwjidx0B,
911 vdwparam+vdwioffset0+vdwjidx0C,
912 vdwparam+vdwioffset0+vdwjidx0D,
915 /* REACTION-FIELD ELECTROSTATICS */
916 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
918 /* LENNARD-JONES DISPERSION/REPULSION */
920 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
921 fvdw = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
923 fscal = _mm_add_ps(felec,fvdw);
925 fscal = _mm_andnot_ps(dummy_mask,fscal);
927 /* Calculate temporary vectorial force */
928 tx = _mm_mul_ps(fscal,dx00);
929 ty = _mm_mul_ps(fscal,dy00);
930 tz = _mm_mul_ps(fscal,dz00);
932 /* Update vectorial force */
933 fix0 = _mm_add_ps(fix0,tx);
934 fiy0 = _mm_add_ps(fiy0,ty);
935 fiz0 = _mm_add_ps(fiz0,tz);
937 fjx0 = _mm_add_ps(fjx0,tx);
938 fjy0 = _mm_add_ps(fjy0,ty);
939 fjz0 = _mm_add_ps(fjz0,tz);
941 /**************************
942 * CALCULATE INTERACTIONS *
943 **************************/
945 /* Compute parameters for interactions between i and j atoms */
946 qq10 = _mm_mul_ps(iq1,jq0);
948 /* REACTION-FIELD ELECTROSTATICS */
949 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
953 fscal = _mm_andnot_ps(dummy_mask,fscal);
955 /* Calculate temporary vectorial force */
956 tx = _mm_mul_ps(fscal,dx10);
957 ty = _mm_mul_ps(fscal,dy10);
958 tz = _mm_mul_ps(fscal,dz10);
960 /* Update vectorial force */
961 fix1 = _mm_add_ps(fix1,tx);
962 fiy1 = _mm_add_ps(fiy1,ty);
963 fiz1 = _mm_add_ps(fiz1,tz);
965 fjx0 = _mm_add_ps(fjx0,tx);
966 fjy0 = _mm_add_ps(fjy0,ty);
967 fjz0 = _mm_add_ps(fjz0,tz);
969 /**************************
970 * CALCULATE INTERACTIONS *
971 **************************/
973 /* Compute parameters for interactions between i and j atoms */
974 qq20 = _mm_mul_ps(iq2,jq0);
976 /* REACTION-FIELD ELECTROSTATICS */
977 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
981 fscal = _mm_andnot_ps(dummy_mask,fscal);
983 /* Calculate temporary vectorial force */
984 tx = _mm_mul_ps(fscal,dx20);
985 ty = _mm_mul_ps(fscal,dy20);
986 tz = _mm_mul_ps(fscal,dz20);
988 /* Update vectorial force */
989 fix2 = _mm_add_ps(fix2,tx);
990 fiy2 = _mm_add_ps(fiy2,ty);
991 fiz2 = _mm_add_ps(fiz2,tz);
993 fjx0 = _mm_add_ps(fjx0,tx);
994 fjy0 = _mm_add_ps(fjy0,ty);
995 fjz0 = _mm_add_ps(fjz0,tz);
997 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
998 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
999 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1000 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1002 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1004 /* Inner loop uses 88 flops */
1007 /* End of innermost loop */
1009 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1010 f+i_coord_offset,fshift+i_shift_offset);
1012 /* Increment number of inner iterations */
1013 inneriter += j_index_end - j_index_start;
1015 /* Outer loop uses 18 flops */
1018 /* Increment number of outer iterations */
1021 /* Update outer/inner flops */
1023 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*88);