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 avx_128_fma_double kernel generator.
42 #include "../nb_kernel.h"
43 #include "types/simple.h"
44 #include "gromacs/math/vec.h"
47 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
48 #include "kernelutil_x86_avx_128_fma_double.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwLJ_GeomW4P1_VF_avx_128_fma_double
52 * Electrostatics interaction: ReactionField
53 * VdW interaction: LennardJones
54 * Geometry: Water4-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecRF_VdwLJ_GeomW4P1_VF_avx_128_fma_double
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 refer to j loop unrolling done with SSE double precision, e.g. for the two 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;
75 int j_coord_offsetA,j_coord_offsetB;
76 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
78 real *shiftvec,*fshift,*x,*f;
79 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
81 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
83 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
85 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
87 __m128d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
88 int vdwjidx0A,vdwjidx0B;
89 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
90 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
91 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
92 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
93 __m128d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
94 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
97 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
100 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
101 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
102 __m128d dummy_mask,cutoff_mask;
103 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
104 __m128d one = _mm_set1_pd(1.0);
105 __m128d two = _mm_set1_pd(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_pd(fr->epsfac);
118 charge = mdatoms->chargeA;
119 krf = _mm_set1_pd(fr->ic->k_rf);
120 krf2 = _mm_set1_pd(fr->ic->k_rf*2.0);
121 crf = _mm_set1_pd(fr->ic->c_rf);
122 nvdwtype = fr->ntype;
124 vdwtype = mdatoms->typeA;
126 /* Setup water-specific parameters */
127 inr = nlist->iinr[0];
128 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
129 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
130 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
131 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
133 /* Avoid stupid compiler warnings */
141 /* Start outer loop over neighborlists */
142 for(iidx=0; iidx<nri; iidx++)
144 /* Load shift vector for this list */
145 i_shift_offset = DIM*shiftidx[iidx];
147 /* Load limits for loop over neighbors */
148 j_index_start = jindex[iidx];
149 j_index_end = jindex[iidx+1];
151 /* Get outer coordinate index */
153 i_coord_offset = DIM*inr;
155 /* Load i particle coords and add shift vector */
156 gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
157 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
159 fix0 = _mm_setzero_pd();
160 fiy0 = _mm_setzero_pd();
161 fiz0 = _mm_setzero_pd();
162 fix1 = _mm_setzero_pd();
163 fiy1 = _mm_setzero_pd();
164 fiz1 = _mm_setzero_pd();
165 fix2 = _mm_setzero_pd();
166 fiy2 = _mm_setzero_pd();
167 fiz2 = _mm_setzero_pd();
168 fix3 = _mm_setzero_pd();
169 fiy3 = _mm_setzero_pd();
170 fiz3 = _mm_setzero_pd();
172 /* Reset potential sums */
173 velecsum = _mm_setzero_pd();
174 vvdwsum = _mm_setzero_pd();
176 /* Start inner kernel loop */
177 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
180 /* Get j neighbor index, and coordinate index */
183 j_coord_offsetA = DIM*jnrA;
184 j_coord_offsetB = DIM*jnrB;
186 /* load j atom coordinates */
187 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
190 /* Calculate displacement vector */
191 dx00 = _mm_sub_pd(ix0,jx0);
192 dy00 = _mm_sub_pd(iy0,jy0);
193 dz00 = _mm_sub_pd(iz0,jz0);
194 dx10 = _mm_sub_pd(ix1,jx0);
195 dy10 = _mm_sub_pd(iy1,jy0);
196 dz10 = _mm_sub_pd(iz1,jz0);
197 dx20 = _mm_sub_pd(ix2,jx0);
198 dy20 = _mm_sub_pd(iy2,jy0);
199 dz20 = _mm_sub_pd(iz2,jz0);
200 dx30 = _mm_sub_pd(ix3,jx0);
201 dy30 = _mm_sub_pd(iy3,jy0);
202 dz30 = _mm_sub_pd(iz3,jz0);
204 /* Calculate squared distance and things based on it */
205 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
206 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
207 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
208 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
210 rinv10 = gmx_mm_invsqrt_pd(rsq10);
211 rinv20 = gmx_mm_invsqrt_pd(rsq20);
212 rinv30 = gmx_mm_invsqrt_pd(rsq30);
214 rinvsq00 = gmx_mm_inv_pd(rsq00);
215 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
216 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
217 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
219 /* Load parameters for j particles */
220 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
221 vdwjidx0A = 2*vdwtype[jnrA+0];
222 vdwjidx0B = 2*vdwtype[jnrB+0];
224 fjx0 = _mm_setzero_pd();
225 fjy0 = _mm_setzero_pd();
226 fjz0 = _mm_setzero_pd();
228 /**************************
229 * CALCULATE INTERACTIONS *
230 **************************/
232 /* Compute parameters for interactions between i and j atoms */
233 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
234 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
236 /* LENNARD-JONES DISPERSION/REPULSION */
238 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
239 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
240 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
241 vvdw = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
242 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
244 /* Update potential sum for this i atom from the interaction with this j atom. */
245 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
249 /* Update vectorial force */
250 fix0 = _mm_macc_pd(dx00,fscal,fix0);
251 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
252 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
254 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
255 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
256 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
258 /**************************
259 * CALCULATE INTERACTIONS *
260 **************************/
262 /* Compute parameters for interactions between i and j atoms */
263 qq10 = _mm_mul_pd(iq1,jq0);
265 /* REACTION-FIELD ELECTROSTATICS */
266 velec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
267 felec = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
269 /* Update potential sum for this i atom from the interaction with this j atom. */
270 velecsum = _mm_add_pd(velecsum,velec);
274 /* Update vectorial force */
275 fix1 = _mm_macc_pd(dx10,fscal,fix1);
276 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
277 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
279 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
280 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
281 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
283 /**************************
284 * CALCULATE INTERACTIONS *
285 **************************/
287 /* Compute parameters for interactions between i and j atoms */
288 qq20 = _mm_mul_pd(iq2,jq0);
290 /* REACTION-FIELD ELECTROSTATICS */
291 velec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
292 felec = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
294 /* Update potential sum for this i atom from the interaction with this j atom. */
295 velecsum = _mm_add_pd(velecsum,velec);
299 /* Update vectorial force */
300 fix2 = _mm_macc_pd(dx20,fscal,fix2);
301 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
302 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
304 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
305 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
306 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
308 /**************************
309 * CALCULATE INTERACTIONS *
310 **************************/
312 /* Compute parameters for interactions between i and j atoms */
313 qq30 = _mm_mul_pd(iq3,jq0);
315 /* REACTION-FIELD ELECTROSTATICS */
316 velec = _mm_mul_pd(qq30,_mm_sub_pd(_mm_macc_pd(krf,rsq30,rinv30),crf));
317 felec = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
319 /* Update potential sum for this i atom from the interaction with this j atom. */
320 velecsum = _mm_add_pd(velecsum,velec);
324 /* Update vectorial force */
325 fix3 = _mm_macc_pd(dx30,fscal,fix3);
326 fiy3 = _mm_macc_pd(dy30,fscal,fiy3);
327 fiz3 = _mm_macc_pd(dz30,fscal,fiz3);
329 fjx0 = _mm_macc_pd(dx30,fscal,fjx0);
330 fjy0 = _mm_macc_pd(dy30,fscal,fjy0);
331 fjz0 = _mm_macc_pd(dz30,fscal,fjz0);
333 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
335 /* Inner loop uses 143 flops */
342 j_coord_offsetA = DIM*jnrA;
344 /* load j atom coordinates */
345 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
348 /* Calculate displacement vector */
349 dx00 = _mm_sub_pd(ix0,jx0);
350 dy00 = _mm_sub_pd(iy0,jy0);
351 dz00 = _mm_sub_pd(iz0,jz0);
352 dx10 = _mm_sub_pd(ix1,jx0);
353 dy10 = _mm_sub_pd(iy1,jy0);
354 dz10 = _mm_sub_pd(iz1,jz0);
355 dx20 = _mm_sub_pd(ix2,jx0);
356 dy20 = _mm_sub_pd(iy2,jy0);
357 dz20 = _mm_sub_pd(iz2,jz0);
358 dx30 = _mm_sub_pd(ix3,jx0);
359 dy30 = _mm_sub_pd(iy3,jy0);
360 dz30 = _mm_sub_pd(iz3,jz0);
362 /* Calculate squared distance and things based on it */
363 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
364 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
365 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
366 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
368 rinv10 = gmx_mm_invsqrt_pd(rsq10);
369 rinv20 = gmx_mm_invsqrt_pd(rsq20);
370 rinv30 = gmx_mm_invsqrt_pd(rsq30);
372 rinvsq00 = gmx_mm_inv_pd(rsq00);
373 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
374 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
375 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
377 /* Load parameters for j particles */
378 jq0 = _mm_load_sd(charge+jnrA+0);
379 vdwjidx0A = 2*vdwtype[jnrA+0];
381 fjx0 = _mm_setzero_pd();
382 fjy0 = _mm_setzero_pd();
383 fjz0 = _mm_setzero_pd();
385 /**************************
386 * CALCULATE INTERACTIONS *
387 **************************/
389 /* Compute parameters for interactions between i and j atoms */
390 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
392 /* LENNARD-JONES DISPERSION/REPULSION */
394 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
395 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
396 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
397 vvdw = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
398 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
400 /* Update potential sum for this i atom from the interaction with this j atom. */
401 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
402 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
406 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
408 /* Update vectorial force */
409 fix0 = _mm_macc_pd(dx00,fscal,fix0);
410 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
411 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
413 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
414 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
415 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
417 /**************************
418 * CALCULATE INTERACTIONS *
419 **************************/
421 /* Compute parameters for interactions between i and j atoms */
422 qq10 = _mm_mul_pd(iq1,jq0);
424 /* REACTION-FIELD ELECTROSTATICS */
425 velec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
426 felec = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
428 /* Update potential sum for this i atom from the interaction with this j atom. */
429 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
430 velecsum = _mm_add_pd(velecsum,velec);
434 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
436 /* Update vectorial force */
437 fix1 = _mm_macc_pd(dx10,fscal,fix1);
438 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
439 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
441 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
442 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
443 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
445 /**************************
446 * CALCULATE INTERACTIONS *
447 **************************/
449 /* Compute parameters for interactions between i and j atoms */
450 qq20 = _mm_mul_pd(iq2,jq0);
452 /* REACTION-FIELD ELECTROSTATICS */
453 velec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
454 felec = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
456 /* Update potential sum for this i atom from the interaction with this j atom. */
457 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
458 velecsum = _mm_add_pd(velecsum,velec);
462 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
464 /* Update vectorial force */
465 fix2 = _mm_macc_pd(dx20,fscal,fix2);
466 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
467 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
469 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
470 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
471 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
473 /**************************
474 * CALCULATE INTERACTIONS *
475 **************************/
477 /* Compute parameters for interactions between i and j atoms */
478 qq30 = _mm_mul_pd(iq3,jq0);
480 /* REACTION-FIELD ELECTROSTATICS */
481 velec = _mm_mul_pd(qq30,_mm_sub_pd(_mm_macc_pd(krf,rsq30,rinv30),crf));
482 felec = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
484 /* Update potential sum for this i atom from the interaction with this j atom. */
485 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
486 velecsum = _mm_add_pd(velecsum,velec);
490 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
492 /* Update vectorial force */
493 fix3 = _mm_macc_pd(dx30,fscal,fix3);
494 fiy3 = _mm_macc_pd(dy30,fscal,fiy3);
495 fiz3 = _mm_macc_pd(dz30,fscal,fiz3);
497 fjx0 = _mm_macc_pd(dx30,fscal,fjx0);
498 fjy0 = _mm_macc_pd(dy30,fscal,fjy0);
499 fjz0 = _mm_macc_pd(dz30,fscal,fjz0);
501 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
503 /* Inner loop uses 143 flops */
506 /* End of innermost loop */
508 gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
509 f+i_coord_offset,fshift+i_shift_offset);
512 /* Update potential energies */
513 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
514 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
516 /* Increment number of inner iterations */
517 inneriter += j_index_end - j_index_start;
519 /* Outer loop uses 26 flops */
522 /* Increment number of outer iterations */
525 /* Update outer/inner flops */
527 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*143);
530 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwLJ_GeomW4P1_F_avx_128_fma_double
531 * Electrostatics interaction: ReactionField
532 * VdW interaction: LennardJones
533 * Geometry: Water4-Particle
534 * Calculate force/pot: Force
537 nb_kernel_ElecRF_VdwLJ_GeomW4P1_F_avx_128_fma_double
538 (t_nblist * gmx_restrict nlist,
539 rvec * gmx_restrict xx,
540 rvec * gmx_restrict ff,
541 t_forcerec * gmx_restrict fr,
542 t_mdatoms * gmx_restrict mdatoms,
543 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
544 t_nrnb * gmx_restrict nrnb)
546 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
547 * just 0 for non-waters.
548 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
549 * jnr indices corresponding to data put in the four positions in the SIMD register.
551 int i_shift_offset,i_coord_offset,outeriter,inneriter;
552 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
554 int j_coord_offsetA,j_coord_offsetB;
555 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
557 real *shiftvec,*fshift,*x,*f;
558 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
560 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
562 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
564 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
566 __m128d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
567 int vdwjidx0A,vdwjidx0B;
568 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
569 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
570 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
571 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
572 __m128d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
573 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
576 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
579 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
580 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
581 __m128d dummy_mask,cutoff_mask;
582 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
583 __m128d one = _mm_set1_pd(1.0);
584 __m128d two = _mm_set1_pd(2.0);
590 jindex = nlist->jindex;
592 shiftidx = nlist->shift;
594 shiftvec = fr->shift_vec[0];
595 fshift = fr->fshift[0];
596 facel = _mm_set1_pd(fr->epsfac);
597 charge = mdatoms->chargeA;
598 krf = _mm_set1_pd(fr->ic->k_rf);
599 krf2 = _mm_set1_pd(fr->ic->k_rf*2.0);
600 crf = _mm_set1_pd(fr->ic->c_rf);
601 nvdwtype = fr->ntype;
603 vdwtype = mdatoms->typeA;
605 /* Setup water-specific parameters */
606 inr = nlist->iinr[0];
607 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
608 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
609 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
610 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
612 /* Avoid stupid compiler warnings */
620 /* Start outer loop over neighborlists */
621 for(iidx=0; iidx<nri; iidx++)
623 /* Load shift vector for this list */
624 i_shift_offset = DIM*shiftidx[iidx];
626 /* Load limits for loop over neighbors */
627 j_index_start = jindex[iidx];
628 j_index_end = jindex[iidx+1];
630 /* Get outer coordinate index */
632 i_coord_offset = DIM*inr;
634 /* Load i particle coords and add shift vector */
635 gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
636 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
638 fix0 = _mm_setzero_pd();
639 fiy0 = _mm_setzero_pd();
640 fiz0 = _mm_setzero_pd();
641 fix1 = _mm_setzero_pd();
642 fiy1 = _mm_setzero_pd();
643 fiz1 = _mm_setzero_pd();
644 fix2 = _mm_setzero_pd();
645 fiy2 = _mm_setzero_pd();
646 fiz2 = _mm_setzero_pd();
647 fix3 = _mm_setzero_pd();
648 fiy3 = _mm_setzero_pd();
649 fiz3 = _mm_setzero_pd();
651 /* Start inner kernel loop */
652 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
655 /* Get j neighbor index, and coordinate index */
658 j_coord_offsetA = DIM*jnrA;
659 j_coord_offsetB = DIM*jnrB;
661 /* load j atom coordinates */
662 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
665 /* Calculate displacement vector */
666 dx00 = _mm_sub_pd(ix0,jx0);
667 dy00 = _mm_sub_pd(iy0,jy0);
668 dz00 = _mm_sub_pd(iz0,jz0);
669 dx10 = _mm_sub_pd(ix1,jx0);
670 dy10 = _mm_sub_pd(iy1,jy0);
671 dz10 = _mm_sub_pd(iz1,jz0);
672 dx20 = _mm_sub_pd(ix2,jx0);
673 dy20 = _mm_sub_pd(iy2,jy0);
674 dz20 = _mm_sub_pd(iz2,jz0);
675 dx30 = _mm_sub_pd(ix3,jx0);
676 dy30 = _mm_sub_pd(iy3,jy0);
677 dz30 = _mm_sub_pd(iz3,jz0);
679 /* Calculate squared distance and things based on it */
680 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
681 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
682 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
683 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
685 rinv10 = gmx_mm_invsqrt_pd(rsq10);
686 rinv20 = gmx_mm_invsqrt_pd(rsq20);
687 rinv30 = gmx_mm_invsqrt_pd(rsq30);
689 rinvsq00 = gmx_mm_inv_pd(rsq00);
690 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
691 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
692 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
694 /* Load parameters for j particles */
695 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
696 vdwjidx0A = 2*vdwtype[jnrA+0];
697 vdwjidx0B = 2*vdwtype[jnrB+0];
699 fjx0 = _mm_setzero_pd();
700 fjy0 = _mm_setzero_pd();
701 fjz0 = _mm_setzero_pd();
703 /**************************
704 * CALCULATE INTERACTIONS *
705 **************************/
707 /* Compute parameters for interactions between i and j atoms */
708 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
709 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
711 /* LENNARD-JONES DISPERSION/REPULSION */
713 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
714 fvdw = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
718 /* Update vectorial force */
719 fix0 = _mm_macc_pd(dx00,fscal,fix0);
720 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
721 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
723 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
724 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
725 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
727 /**************************
728 * CALCULATE INTERACTIONS *
729 **************************/
731 /* Compute parameters for interactions between i and j atoms */
732 qq10 = _mm_mul_pd(iq1,jq0);
734 /* REACTION-FIELD ELECTROSTATICS */
735 felec = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
739 /* Update vectorial force */
740 fix1 = _mm_macc_pd(dx10,fscal,fix1);
741 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
742 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
744 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
745 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
746 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
748 /**************************
749 * CALCULATE INTERACTIONS *
750 **************************/
752 /* Compute parameters for interactions between i and j atoms */
753 qq20 = _mm_mul_pd(iq2,jq0);
755 /* REACTION-FIELD ELECTROSTATICS */
756 felec = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
760 /* Update vectorial force */
761 fix2 = _mm_macc_pd(dx20,fscal,fix2);
762 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
763 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
765 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
766 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
767 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
769 /**************************
770 * CALCULATE INTERACTIONS *
771 **************************/
773 /* Compute parameters for interactions between i and j atoms */
774 qq30 = _mm_mul_pd(iq3,jq0);
776 /* REACTION-FIELD ELECTROSTATICS */
777 felec = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
781 /* Update vectorial force */
782 fix3 = _mm_macc_pd(dx30,fscal,fix3);
783 fiy3 = _mm_macc_pd(dy30,fscal,fiy3);
784 fiz3 = _mm_macc_pd(dz30,fscal,fiz3);
786 fjx0 = _mm_macc_pd(dx30,fscal,fjx0);
787 fjy0 = _mm_macc_pd(dy30,fscal,fjy0);
788 fjz0 = _mm_macc_pd(dz30,fscal,fjz0);
790 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
792 /* Inner loop uses 123 flops */
799 j_coord_offsetA = DIM*jnrA;
801 /* load j atom coordinates */
802 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
805 /* Calculate displacement vector */
806 dx00 = _mm_sub_pd(ix0,jx0);
807 dy00 = _mm_sub_pd(iy0,jy0);
808 dz00 = _mm_sub_pd(iz0,jz0);
809 dx10 = _mm_sub_pd(ix1,jx0);
810 dy10 = _mm_sub_pd(iy1,jy0);
811 dz10 = _mm_sub_pd(iz1,jz0);
812 dx20 = _mm_sub_pd(ix2,jx0);
813 dy20 = _mm_sub_pd(iy2,jy0);
814 dz20 = _mm_sub_pd(iz2,jz0);
815 dx30 = _mm_sub_pd(ix3,jx0);
816 dy30 = _mm_sub_pd(iy3,jy0);
817 dz30 = _mm_sub_pd(iz3,jz0);
819 /* Calculate squared distance and things based on it */
820 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
821 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
822 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
823 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
825 rinv10 = gmx_mm_invsqrt_pd(rsq10);
826 rinv20 = gmx_mm_invsqrt_pd(rsq20);
827 rinv30 = gmx_mm_invsqrt_pd(rsq30);
829 rinvsq00 = gmx_mm_inv_pd(rsq00);
830 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
831 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
832 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
834 /* Load parameters for j particles */
835 jq0 = _mm_load_sd(charge+jnrA+0);
836 vdwjidx0A = 2*vdwtype[jnrA+0];
838 fjx0 = _mm_setzero_pd();
839 fjy0 = _mm_setzero_pd();
840 fjz0 = _mm_setzero_pd();
842 /**************************
843 * CALCULATE INTERACTIONS *
844 **************************/
846 /* Compute parameters for interactions between i and j atoms */
847 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
849 /* LENNARD-JONES DISPERSION/REPULSION */
851 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
852 fvdw = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
856 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
858 /* Update vectorial force */
859 fix0 = _mm_macc_pd(dx00,fscal,fix0);
860 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
861 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
863 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
864 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
865 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
867 /**************************
868 * CALCULATE INTERACTIONS *
869 **************************/
871 /* Compute parameters for interactions between i and j atoms */
872 qq10 = _mm_mul_pd(iq1,jq0);
874 /* REACTION-FIELD ELECTROSTATICS */
875 felec = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
879 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
881 /* Update vectorial force */
882 fix1 = _mm_macc_pd(dx10,fscal,fix1);
883 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
884 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
886 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
887 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
888 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
890 /**************************
891 * CALCULATE INTERACTIONS *
892 **************************/
894 /* Compute parameters for interactions between i and j atoms */
895 qq20 = _mm_mul_pd(iq2,jq0);
897 /* REACTION-FIELD ELECTROSTATICS */
898 felec = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
902 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
904 /* Update vectorial force */
905 fix2 = _mm_macc_pd(dx20,fscal,fix2);
906 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
907 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
909 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
910 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
911 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
913 /**************************
914 * CALCULATE INTERACTIONS *
915 **************************/
917 /* Compute parameters for interactions between i and j atoms */
918 qq30 = _mm_mul_pd(iq3,jq0);
920 /* REACTION-FIELD ELECTROSTATICS */
921 felec = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
925 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
927 /* Update vectorial force */
928 fix3 = _mm_macc_pd(dx30,fscal,fix3);
929 fiy3 = _mm_macc_pd(dy30,fscal,fiy3);
930 fiz3 = _mm_macc_pd(dz30,fscal,fiz3);
932 fjx0 = _mm_macc_pd(dx30,fscal,fjx0);
933 fjy0 = _mm_macc_pd(dy30,fscal,fjy0);
934 fjz0 = _mm_macc_pd(dz30,fscal,fjz0);
936 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
938 /* Inner loop uses 123 flops */
941 /* End of innermost loop */
943 gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
944 f+i_coord_offset,fshift+i_shift_offset);
946 /* Increment number of inner iterations */
947 inneriter += j_index_end - j_index_start;
949 /* Outer loop uses 24 flops */
952 /* Increment number of outer iterations */
955 /* Update outer/inner flops */
957 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*123);