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.
42 #include "../nb_kernel.h"
43 #include "types/simple.h"
44 #include "gromacs/math/vec.h"
47 #include "gromacs/simd/math_x86_sse2_single.h"
48 #include "kernelutil_x86_sse2_single.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwLJ_GeomW3P1_VF_sse2_single
52 * Electrostatics interaction: ReactionField
53 * VdW interaction: LennardJones
54 * Geometry: Water3-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecRF_VdwLJ_GeomW3P1_VF_sse2_single
59 (t_nblist * gmx_restrict nlist,
60 rvec * gmx_restrict xx,
61 rvec * gmx_restrict ff,
62 t_forcerec * gmx_restrict fr,
63 t_mdatoms * gmx_restrict mdatoms,
64 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
65 t_nrnb * gmx_restrict nrnb)
67 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
68 * just 0 for non-waters.
69 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
70 * jnr indices corresponding to data put in the four positions in the SIMD register.
72 int i_shift_offset,i_coord_offset,outeriter,inneriter;
73 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
74 int jnrA,jnrB,jnrC,jnrD;
75 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
76 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
77 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
79 real *shiftvec,*fshift,*x,*f;
80 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
82 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
84 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
86 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
88 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
89 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
90 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
91 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
92 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
93 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
94 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
97 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
100 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
101 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
102 __m128 dummy_mask,cutoff_mask;
103 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
104 __m128 one = _mm_set1_ps(1.0);
105 __m128 two = _mm_set1_ps(2.0);
111 jindex = nlist->jindex;
113 shiftidx = nlist->shift;
115 shiftvec = fr->shift_vec[0];
116 fshift = fr->fshift[0];
117 facel = _mm_set1_ps(fr->epsfac);
118 charge = mdatoms->chargeA;
119 krf = _mm_set1_ps(fr->ic->k_rf);
120 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
121 crf = _mm_set1_ps(fr->ic->c_rf);
122 nvdwtype = fr->ntype;
124 vdwtype = mdatoms->typeA;
126 /* Setup water-specific parameters */
127 inr = nlist->iinr[0];
128 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
129 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
130 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
131 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
133 /* Avoid stupid compiler warnings */
134 jnrA = jnrB = jnrC = jnrD = 0;
143 for(iidx=0;iidx<4*DIM;iidx++)
148 /* Start outer loop over neighborlists */
149 for(iidx=0; iidx<nri; iidx++)
151 /* Load shift vector for this list */
152 i_shift_offset = DIM*shiftidx[iidx];
154 /* Load limits for loop over neighbors */
155 j_index_start = jindex[iidx];
156 j_index_end = jindex[iidx+1];
158 /* Get outer coordinate index */
160 i_coord_offset = DIM*inr;
162 /* Load i particle coords and add shift vector */
163 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
164 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
166 fix0 = _mm_setzero_ps();
167 fiy0 = _mm_setzero_ps();
168 fiz0 = _mm_setzero_ps();
169 fix1 = _mm_setzero_ps();
170 fiy1 = _mm_setzero_ps();
171 fiz1 = _mm_setzero_ps();
172 fix2 = _mm_setzero_ps();
173 fiy2 = _mm_setzero_ps();
174 fiz2 = _mm_setzero_ps();
176 /* Reset potential sums */
177 velecsum = _mm_setzero_ps();
178 vvdwsum = _mm_setzero_ps();
180 /* Start inner kernel loop */
181 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
184 /* Get j neighbor index, and coordinate index */
189 j_coord_offsetA = DIM*jnrA;
190 j_coord_offsetB = DIM*jnrB;
191 j_coord_offsetC = DIM*jnrC;
192 j_coord_offsetD = DIM*jnrD;
194 /* load j atom coordinates */
195 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
196 x+j_coord_offsetC,x+j_coord_offsetD,
199 /* Calculate displacement vector */
200 dx00 = _mm_sub_ps(ix0,jx0);
201 dy00 = _mm_sub_ps(iy0,jy0);
202 dz00 = _mm_sub_ps(iz0,jz0);
203 dx10 = _mm_sub_ps(ix1,jx0);
204 dy10 = _mm_sub_ps(iy1,jy0);
205 dz10 = _mm_sub_ps(iz1,jz0);
206 dx20 = _mm_sub_ps(ix2,jx0);
207 dy20 = _mm_sub_ps(iy2,jy0);
208 dz20 = _mm_sub_ps(iz2,jz0);
210 /* Calculate squared distance and things based on it */
211 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
212 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
213 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
215 rinv00 = gmx_mm_invsqrt_ps(rsq00);
216 rinv10 = gmx_mm_invsqrt_ps(rsq10);
217 rinv20 = gmx_mm_invsqrt_ps(rsq20);
219 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
220 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
221 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
223 /* Load parameters for j particles */
224 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
225 charge+jnrC+0,charge+jnrD+0);
226 vdwjidx0A = 2*vdwtype[jnrA+0];
227 vdwjidx0B = 2*vdwtype[jnrB+0];
228 vdwjidx0C = 2*vdwtype[jnrC+0];
229 vdwjidx0D = 2*vdwtype[jnrD+0];
231 fjx0 = _mm_setzero_ps();
232 fjy0 = _mm_setzero_ps();
233 fjz0 = _mm_setzero_ps();
235 /**************************
236 * CALCULATE INTERACTIONS *
237 **************************/
239 /* Compute parameters for interactions between i and j atoms */
240 qq00 = _mm_mul_ps(iq0,jq0);
241 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
242 vdwparam+vdwioffset0+vdwjidx0B,
243 vdwparam+vdwioffset0+vdwjidx0C,
244 vdwparam+vdwioffset0+vdwjidx0D,
247 /* REACTION-FIELD ELECTROSTATICS */
248 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
249 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
251 /* LENNARD-JONES DISPERSION/REPULSION */
253 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
254 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
255 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
256 vvdw = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
257 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
259 /* Update potential sum for this i atom from the interaction with this j atom. */
260 velecsum = _mm_add_ps(velecsum,velec);
261 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
263 fscal = _mm_add_ps(felec,fvdw);
265 /* Calculate temporary vectorial force */
266 tx = _mm_mul_ps(fscal,dx00);
267 ty = _mm_mul_ps(fscal,dy00);
268 tz = _mm_mul_ps(fscal,dz00);
270 /* Update vectorial force */
271 fix0 = _mm_add_ps(fix0,tx);
272 fiy0 = _mm_add_ps(fiy0,ty);
273 fiz0 = _mm_add_ps(fiz0,tz);
275 fjx0 = _mm_add_ps(fjx0,tx);
276 fjy0 = _mm_add_ps(fjy0,ty);
277 fjz0 = _mm_add_ps(fjz0,tz);
279 /**************************
280 * CALCULATE INTERACTIONS *
281 **************************/
283 /* Compute parameters for interactions between i and j atoms */
284 qq10 = _mm_mul_ps(iq1,jq0);
286 /* REACTION-FIELD ELECTROSTATICS */
287 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
288 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
290 /* Update potential sum for this i atom from the interaction with this j atom. */
291 velecsum = _mm_add_ps(velecsum,velec);
295 /* Calculate temporary vectorial force */
296 tx = _mm_mul_ps(fscal,dx10);
297 ty = _mm_mul_ps(fscal,dy10);
298 tz = _mm_mul_ps(fscal,dz10);
300 /* Update vectorial force */
301 fix1 = _mm_add_ps(fix1,tx);
302 fiy1 = _mm_add_ps(fiy1,ty);
303 fiz1 = _mm_add_ps(fiz1,tz);
305 fjx0 = _mm_add_ps(fjx0,tx);
306 fjy0 = _mm_add_ps(fjy0,ty);
307 fjz0 = _mm_add_ps(fjz0,tz);
309 /**************************
310 * CALCULATE INTERACTIONS *
311 **************************/
313 /* Compute parameters for interactions between i and j atoms */
314 qq20 = _mm_mul_ps(iq2,jq0);
316 /* REACTION-FIELD ELECTROSTATICS */
317 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
318 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
320 /* Update potential sum for this i atom from the interaction with this j atom. */
321 velecsum = _mm_add_ps(velecsum,velec);
325 /* Calculate temporary vectorial force */
326 tx = _mm_mul_ps(fscal,dx20);
327 ty = _mm_mul_ps(fscal,dy20);
328 tz = _mm_mul_ps(fscal,dz20);
330 /* Update vectorial force */
331 fix2 = _mm_add_ps(fix2,tx);
332 fiy2 = _mm_add_ps(fiy2,ty);
333 fiz2 = _mm_add_ps(fiz2,tz);
335 fjx0 = _mm_add_ps(fjx0,tx);
336 fjy0 = _mm_add_ps(fjy0,ty);
337 fjz0 = _mm_add_ps(fjz0,tz);
339 fjptrA = f+j_coord_offsetA;
340 fjptrB = f+j_coord_offsetB;
341 fjptrC = f+j_coord_offsetC;
342 fjptrD = f+j_coord_offsetD;
344 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
346 /* Inner loop uses 108 flops */
352 /* Get j neighbor index, and coordinate index */
353 jnrlistA = jjnr[jidx];
354 jnrlistB = jjnr[jidx+1];
355 jnrlistC = jjnr[jidx+2];
356 jnrlistD = jjnr[jidx+3];
357 /* Sign of each element will be negative for non-real atoms.
358 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
359 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
361 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
362 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
363 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
364 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
365 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
366 j_coord_offsetA = DIM*jnrA;
367 j_coord_offsetB = DIM*jnrB;
368 j_coord_offsetC = DIM*jnrC;
369 j_coord_offsetD = DIM*jnrD;
371 /* load j atom coordinates */
372 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
373 x+j_coord_offsetC,x+j_coord_offsetD,
376 /* Calculate displacement vector */
377 dx00 = _mm_sub_ps(ix0,jx0);
378 dy00 = _mm_sub_ps(iy0,jy0);
379 dz00 = _mm_sub_ps(iz0,jz0);
380 dx10 = _mm_sub_ps(ix1,jx0);
381 dy10 = _mm_sub_ps(iy1,jy0);
382 dz10 = _mm_sub_ps(iz1,jz0);
383 dx20 = _mm_sub_ps(ix2,jx0);
384 dy20 = _mm_sub_ps(iy2,jy0);
385 dz20 = _mm_sub_ps(iz2,jz0);
387 /* Calculate squared distance and things based on it */
388 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
389 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
390 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
392 rinv00 = gmx_mm_invsqrt_ps(rsq00);
393 rinv10 = gmx_mm_invsqrt_ps(rsq10);
394 rinv20 = gmx_mm_invsqrt_ps(rsq20);
396 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
397 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
398 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
400 /* Load parameters for j particles */
401 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
402 charge+jnrC+0,charge+jnrD+0);
403 vdwjidx0A = 2*vdwtype[jnrA+0];
404 vdwjidx0B = 2*vdwtype[jnrB+0];
405 vdwjidx0C = 2*vdwtype[jnrC+0];
406 vdwjidx0D = 2*vdwtype[jnrD+0];
408 fjx0 = _mm_setzero_ps();
409 fjy0 = _mm_setzero_ps();
410 fjz0 = _mm_setzero_ps();
412 /**************************
413 * CALCULATE INTERACTIONS *
414 **************************/
416 /* Compute parameters for interactions between i and j atoms */
417 qq00 = _mm_mul_ps(iq0,jq0);
418 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
419 vdwparam+vdwioffset0+vdwjidx0B,
420 vdwparam+vdwioffset0+vdwjidx0C,
421 vdwparam+vdwioffset0+vdwjidx0D,
424 /* REACTION-FIELD ELECTROSTATICS */
425 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
426 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
428 /* LENNARD-JONES DISPERSION/REPULSION */
430 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
431 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
432 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
433 vvdw = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
434 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
436 /* Update potential sum for this i atom from the interaction with this j atom. */
437 velec = _mm_andnot_ps(dummy_mask,velec);
438 velecsum = _mm_add_ps(velecsum,velec);
439 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
440 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
442 fscal = _mm_add_ps(felec,fvdw);
444 fscal = _mm_andnot_ps(dummy_mask,fscal);
446 /* Calculate temporary vectorial force */
447 tx = _mm_mul_ps(fscal,dx00);
448 ty = _mm_mul_ps(fscal,dy00);
449 tz = _mm_mul_ps(fscal,dz00);
451 /* Update vectorial force */
452 fix0 = _mm_add_ps(fix0,tx);
453 fiy0 = _mm_add_ps(fiy0,ty);
454 fiz0 = _mm_add_ps(fiz0,tz);
456 fjx0 = _mm_add_ps(fjx0,tx);
457 fjy0 = _mm_add_ps(fjy0,ty);
458 fjz0 = _mm_add_ps(fjz0,tz);
460 /**************************
461 * CALCULATE INTERACTIONS *
462 **************************/
464 /* Compute parameters for interactions between i and j atoms */
465 qq10 = _mm_mul_ps(iq1,jq0);
467 /* REACTION-FIELD ELECTROSTATICS */
468 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
469 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
471 /* Update potential sum for this i atom from the interaction with this j atom. */
472 velec = _mm_andnot_ps(dummy_mask,velec);
473 velecsum = _mm_add_ps(velecsum,velec);
477 fscal = _mm_andnot_ps(dummy_mask,fscal);
479 /* Calculate temporary vectorial force */
480 tx = _mm_mul_ps(fscal,dx10);
481 ty = _mm_mul_ps(fscal,dy10);
482 tz = _mm_mul_ps(fscal,dz10);
484 /* Update vectorial force */
485 fix1 = _mm_add_ps(fix1,tx);
486 fiy1 = _mm_add_ps(fiy1,ty);
487 fiz1 = _mm_add_ps(fiz1,tz);
489 fjx0 = _mm_add_ps(fjx0,tx);
490 fjy0 = _mm_add_ps(fjy0,ty);
491 fjz0 = _mm_add_ps(fjz0,tz);
493 /**************************
494 * CALCULATE INTERACTIONS *
495 **************************/
497 /* Compute parameters for interactions between i and j atoms */
498 qq20 = _mm_mul_ps(iq2,jq0);
500 /* REACTION-FIELD ELECTROSTATICS */
501 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
502 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
504 /* Update potential sum for this i atom from the interaction with this j atom. */
505 velec = _mm_andnot_ps(dummy_mask,velec);
506 velecsum = _mm_add_ps(velecsum,velec);
510 fscal = _mm_andnot_ps(dummy_mask,fscal);
512 /* Calculate temporary vectorial force */
513 tx = _mm_mul_ps(fscal,dx20);
514 ty = _mm_mul_ps(fscal,dy20);
515 tz = _mm_mul_ps(fscal,dz20);
517 /* Update vectorial force */
518 fix2 = _mm_add_ps(fix2,tx);
519 fiy2 = _mm_add_ps(fiy2,ty);
520 fiz2 = _mm_add_ps(fiz2,tz);
522 fjx0 = _mm_add_ps(fjx0,tx);
523 fjy0 = _mm_add_ps(fjy0,ty);
524 fjz0 = _mm_add_ps(fjz0,tz);
526 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
527 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
528 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
529 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
531 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
533 /* Inner loop uses 108 flops */
536 /* End of innermost loop */
538 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
539 f+i_coord_offset,fshift+i_shift_offset);
542 /* Update potential energies */
543 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
544 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
546 /* Increment number of inner iterations */
547 inneriter += j_index_end - j_index_start;
549 /* Outer loop uses 20 flops */
552 /* Increment number of outer iterations */
555 /* Update outer/inner flops */
557 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*108);
560 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwLJ_GeomW3P1_F_sse2_single
561 * Electrostatics interaction: ReactionField
562 * VdW interaction: LennardJones
563 * Geometry: Water3-Particle
564 * Calculate force/pot: Force
567 nb_kernel_ElecRF_VdwLJ_GeomW3P1_F_sse2_single
568 (t_nblist * gmx_restrict nlist,
569 rvec * gmx_restrict xx,
570 rvec * gmx_restrict ff,
571 t_forcerec * gmx_restrict fr,
572 t_mdatoms * gmx_restrict mdatoms,
573 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
574 t_nrnb * gmx_restrict nrnb)
576 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
577 * just 0 for non-waters.
578 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
579 * jnr indices corresponding to data put in the four positions in the SIMD register.
581 int i_shift_offset,i_coord_offset,outeriter,inneriter;
582 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
583 int jnrA,jnrB,jnrC,jnrD;
584 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
585 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
586 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
588 real *shiftvec,*fshift,*x,*f;
589 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
591 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
593 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
595 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
597 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
598 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
599 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
600 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
601 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
602 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
603 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
606 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
609 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
610 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
611 __m128 dummy_mask,cutoff_mask;
612 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
613 __m128 one = _mm_set1_ps(1.0);
614 __m128 two = _mm_set1_ps(2.0);
620 jindex = nlist->jindex;
622 shiftidx = nlist->shift;
624 shiftvec = fr->shift_vec[0];
625 fshift = fr->fshift[0];
626 facel = _mm_set1_ps(fr->epsfac);
627 charge = mdatoms->chargeA;
628 krf = _mm_set1_ps(fr->ic->k_rf);
629 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
630 crf = _mm_set1_ps(fr->ic->c_rf);
631 nvdwtype = fr->ntype;
633 vdwtype = mdatoms->typeA;
635 /* Setup water-specific parameters */
636 inr = nlist->iinr[0];
637 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
638 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
639 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
640 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
642 /* Avoid stupid compiler warnings */
643 jnrA = jnrB = jnrC = jnrD = 0;
652 for(iidx=0;iidx<4*DIM;iidx++)
657 /* Start outer loop over neighborlists */
658 for(iidx=0; iidx<nri; iidx++)
660 /* Load shift vector for this list */
661 i_shift_offset = DIM*shiftidx[iidx];
663 /* Load limits for loop over neighbors */
664 j_index_start = jindex[iidx];
665 j_index_end = jindex[iidx+1];
667 /* Get outer coordinate index */
669 i_coord_offset = DIM*inr;
671 /* Load i particle coords and add shift vector */
672 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
673 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
675 fix0 = _mm_setzero_ps();
676 fiy0 = _mm_setzero_ps();
677 fiz0 = _mm_setzero_ps();
678 fix1 = _mm_setzero_ps();
679 fiy1 = _mm_setzero_ps();
680 fiz1 = _mm_setzero_ps();
681 fix2 = _mm_setzero_ps();
682 fiy2 = _mm_setzero_ps();
683 fiz2 = _mm_setzero_ps();
685 /* Start inner kernel loop */
686 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
689 /* Get j neighbor index, and coordinate index */
694 j_coord_offsetA = DIM*jnrA;
695 j_coord_offsetB = DIM*jnrB;
696 j_coord_offsetC = DIM*jnrC;
697 j_coord_offsetD = DIM*jnrD;
699 /* load j atom coordinates */
700 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
701 x+j_coord_offsetC,x+j_coord_offsetD,
704 /* Calculate displacement vector */
705 dx00 = _mm_sub_ps(ix0,jx0);
706 dy00 = _mm_sub_ps(iy0,jy0);
707 dz00 = _mm_sub_ps(iz0,jz0);
708 dx10 = _mm_sub_ps(ix1,jx0);
709 dy10 = _mm_sub_ps(iy1,jy0);
710 dz10 = _mm_sub_ps(iz1,jz0);
711 dx20 = _mm_sub_ps(ix2,jx0);
712 dy20 = _mm_sub_ps(iy2,jy0);
713 dz20 = _mm_sub_ps(iz2,jz0);
715 /* Calculate squared distance and things based on it */
716 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
717 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
718 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
720 rinv00 = gmx_mm_invsqrt_ps(rsq00);
721 rinv10 = gmx_mm_invsqrt_ps(rsq10);
722 rinv20 = gmx_mm_invsqrt_ps(rsq20);
724 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
725 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
726 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
728 /* Load parameters for j particles */
729 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
730 charge+jnrC+0,charge+jnrD+0);
731 vdwjidx0A = 2*vdwtype[jnrA+0];
732 vdwjidx0B = 2*vdwtype[jnrB+0];
733 vdwjidx0C = 2*vdwtype[jnrC+0];
734 vdwjidx0D = 2*vdwtype[jnrD+0];
736 fjx0 = _mm_setzero_ps();
737 fjy0 = _mm_setzero_ps();
738 fjz0 = _mm_setzero_ps();
740 /**************************
741 * CALCULATE INTERACTIONS *
742 **************************/
744 /* Compute parameters for interactions between i and j atoms */
745 qq00 = _mm_mul_ps(iq0,jq0);
746 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
747 vdwparam+vdwioffset0+vdwjidx0B,
748 vdwparam+vdwioffset0+vdwjidx0C,
749 vdwparam+vdwioffset0+vdwjidx0D,
752 /* REACTION-FIELD ELECTROSTATICS */
753 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
755 /* LENNARD-JONES DISPERSION/REPULSION */
757 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
758 fvdw = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
760 fscal = _mm_add_ps(felec,fvdw);
762 /* Calculate temporary vectorial force */
763 tx = _mm_mul_ps(fscal,dx00);
764 ty = _mm_mul_ps(fscal,dy00);
765 tz = _mm_mul_ps(fscal,dz00);
767 /* Update vectorial force */
768 fix0 = _mm_add_ps(fix0,tx);
769 fiy0 = _mm_add_ps(fiy0,ty);
770 fiz0 = _mm_add_ps(fiz0,tz);
772 fjx0 = _mm_add_ps(fjx0,tx);
773 fjy0 = _mm_add_ps(fjy0,ty);
774 fjz0 = _mm_add_ps(fjz0,tz);
776 /**************************
777 * CALCULATE INTERACTIONS *
778 **************************/
780 /* Compute parameters for interactions between i and j atoms */
781 qq10 = _mm_mul_ps(iq1,jq0);
783 /* REACTION-FIELD ELECTROSTATICS */
784 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
788 /* Calculate temporary vectorial force */
789 tx = _mm_mul_ps(fscal,dx10);
790 ty = _mm_mul_ps(fscal,dy10);
791 tz = _mm_mul_ps(fscal,dz10);
793 /* Update vectorial force */
794 fix1 = _mm_add_ps(fix1,tx);
795 fiy1 = _mm_add_ps(fiy1,ty);
796 fiz1 = _mm_add_ps(fiz1,tz);
798 fjx0 = _mm_add_ps(fjx0,tx);
799 fjy0 = _mm_add_ps(fjy0,ty);
800 fjz0 = _mm_add_ps(fjz0,tz);
802 /**************************
803 * CALCULATE INTERACTIONS *
804 **************************/
806 /* Compute parameters for interactions between i and j atoms */
807 qq20 = _mm_mul_ps(iq2,jq0);
809 /* REACTION-FIELD ELECTROSTATICS */
810 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
814 /* Calculate temporary vectorial force */
815 tx = _mm_mul_ps(fscal,dx20);
816 ty = _mm_mul_ps(fscal,dy20);
817 tz = _mm_mul_ps(fscal,dz20);
819 /* Update vectorial force */
820 fix2 = _mm_add_ps(fix2,tx);
821 fiy2 = _mm_add_ps(fiy2,ty);
822 fiz2 = _mm_add_ps(fiz2,tz);
824 fjx0 = _mm_add_ps(fjx0,tx);
825 fjy0 = _mm_add_ps(fjy0,ty);
826 fjz0 = _mm_add_ps(fjz0,tz);
828 fjptrA = f+j_coord_offsetA;
829 fjptrB = f+j_coord_offsetB;
830 fjptrC = f+j_coord_offsetC;
831 fjptrD = f+j_coord_offsetD;
833 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
835 /* Inner loop uses 88 flops */
841 /* Get j neighbor index, and coordinate index */
842 jnrlistA = jjnr[jidx];
843 jnrlistB = jjnr[jidx+1];
844 jnrlistC = jjnr[jidx+2];
845 jnrlistD = jjnr[jidx+3];
846 /* Sign of each element will be negative for non-real atoms.
847 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
848 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
850 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
851 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
852 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
853 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
854 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
855 j_coord_offsetA = DIM*jnrA;
856 j_coord_offsetB = DIM*jnrB;
857 j_coord_offsetC = DIM*jnrC;
858 j_coord_offsetD = DIM*jnrD;
860 /* load j atom coordinates */
861 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
862 x+j_coord_offsetC,x+j_coord_offsetD,
865 /* Calculate displacement vector */
866 dx00 = _mm_sub_ps(ix0,jx0);
867 dy00 = _mm_sub_ps(iy0,jy0);
868 dz00 = _mm_sub_ps(iz0,jz0);
869 dx10 = _mm_sub_ps(ix1,jx0);
870 dy10 = _mm_sub_ps(iy1,jy0);
871 dz10 = _mm_sub_ps(iz1,jz0);
872 dx20 = _mm_sub_ps(ix2,jx0);
873 dy20 = _mm_sub_ps(iy2,jy0);
874 dz20 = _mm_sub_ps(iz2,jz0);
876 /* Calculate squared distance and things based on it */
877 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
878 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
879 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
881 rinv00 = gmx_mm_invsqrt_ps(rsq00);
882 rinv10 = gmx_mm_invsqrt_ps(rsq10);
883 rinv20 = gmx_mm_invsqrt_ps(rsq20);
885 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
886 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
887 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
889 /* Load parameters for j particles */
890 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
891 charge+jnrC+0,charge+jnrD+0);
892 vdwjidx0A = 2*vdwtype[jnrA+0];
893 vdwjidx0B = 2*vdwtype[jnrB+0];
894 vdwjidx0C = 2*vdwtype[jnrC+0];
895 vdwjidx0D = 2*vdwtype[jnrD+0];
897 fjx0 = _mm_setzero_ps();
898 fjy0 = _mm_setzero_ps();
899 fjz0 = _mm_setzero_ps();
901 /**************************
902 * CALCULATE INTERACTIONS *
903 **************************/
905 /* Compute parameters for interactions between i and j atoms */
906 qq00 = _mm_mul_ps(iq0,jq0);
907 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
908 vdwparam+vdwioffset0+vdwjidx0B,
909 vdwparam+vdwioffset0+vdwjidx0C,
910 vdwparam+vdwioffset0+vdwjidx0D,
913 /* REACTION-FIELD ELECTROSTATICS */
914 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
916 /* LENNARD-JONES DISPERSION/REPULSION */
918 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
919 fvdw = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
921 fscal = _mm_add_ps(felec,fvdw);
923 fscal = _mm_andnot_ps(dummy_mask,fscal);
925 /* Calculate temporary vectorial force */
926 tx = _mm_mul_ps(fscal,dx00);
927 ty = _mm_mul_ps(fscal,dy00);
928 tz = _mm_mul_ps(fscal,dz00);
930 /* Update vectorial force */
931 fix0 = _mm_add_ps(fix0,tx);
932 fiy0 = _mm_add_ps(fiy0,ty);
933 fiz0 = _mm_add_ps(fiz0,tz);
935 fjx0 = _mm_add_ps(fjx0,tx);
936 fjy0 = _mm_add_ps(fjy0,ty);
937 fjz0 = _mm_add_ps(fjz0,tz);
939 /**************************
940 * CALCULATE INTERACTIONS *
941 **************************/
943 /* Compute parameters for interactions between i and j atoms */
944 qq10 = _mm_mul_ps(iq1,jq0);
946 /* REACTION-FIELD ELECTROSTATICS */
947 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
951 fscal = _mm_andnot_ps(dummy_mask,fscal);
953 /* Calculate temporary vectorial force */
954 tx = _mm_mul_ps(fscal,dx10);
955 ty = _mm_mul_ps(fscal,dy10);
956 tz = _mm_mul_ps(fscal,dz10);
958 /* Update vectorial force */
959 fix1 = _mm_add_ps(fix1,tx);
960 fiy1 = _mm_add_ps(fiy1,ty);
961 fiz1 = _mm_add_ps(fiz1,tz);
963 fjx0 = _mm_add_ps(fjx0,tx);
964 fjy0 = _mm_add_ps(fjy0,ty);
965 fjz0 = _mm_add_ps(fjz0,tz);
967 /**************************
968 * CALCULATE INTERACTIONS *
969 **************************/
971 /* Compute parameters for interactions between i and j atoms */
972 qq20 = _mm_mul_ps(iq2,jq0);
974 /* REACTION-FIELD ELECTROSTATICS */
975 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
979 fscal = _mm_andnot_ps(dummy_mask,fscal);
981 /* Calculate temporary vectorial force */
982 tx = _mm_mul_ps(fscal,dx20);
983 ty = _mm_mul_ps(fscal,dy20);
984 tz = _mm_mul_ps(fscal,dz20);
986 /* Update vectorial force */
987 fix2 = _mm_add_ps(fix2,tx);
988 fiy2 = _mm_add_ps(fiy2,ty);
989 fiz2 = _mm_add_ps(fiz2,tz);
991 fjx0 = _mm_add_ps(fjx0,tx);
992 fjy0 = _mm_add_ps(fjy0,ty);
993 fjz0 = _mm_add_ps(fjz0,tz);
995 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
996 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
997 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
998 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1000 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1002 /* Inner loop uses 88 flops */
1005 /* End of innermost loop */
1007 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1008 f+i_coord_offset,fshift+i_shift_offset);
1010 /* Increment number of inner iterations */
1011 inneriter += j_index_end - j_index_start;
1013 /* Outer loop uses 18 flops */
1016 /* Increment number of outer iterations */
1019 /* Update outer/inner flops */
1021 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*88);