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 sse4_1_double 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_sse4_1_double.h"
50 #include "kernelutil_x86_sse4_1_double.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_VF_sse4_1_double
54 * Electrostatics interaction: Coulomb
55 * VdW interaction: CubicSplineTable
56 * Geometry: Water3-Particle
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_VF_sse4_1_double
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 refer to j loop unrolling done with SSE double precision, e.g. for the two 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;
77 int j_coord_offsetA,j_coord_offsetB;
78 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
80 real *shiftvec,*fshift,*x,*f;
81 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
83 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
85 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
87 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
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 velec,felec,velecsum,facel,crf,krf,krf2;
96 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
99 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
100 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
102 __m128i ifour = _mm_set1_epi32(4);
103 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
105 __m128d dummy_mask,cutoff_mask;
106 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
107 __m128d one = _mm_set1_pd(1.0);
108 __m128d two = _mm_set1_pd(2.0);
114 jindex = nlist->jindex;
116 shiftidx = nlist->shift;
118 shiftvec = fr->shift_vec[0];
119 fshift = fr->fshift[0];
120 facel = _mm_set1_pd(fr->epsfac);
121 charge = mdatoms->chargeA;
122 nvdwtype = fr->ntype;
124 vdwtype = mdatoms->typeA;
126 vftab = kernel_data->table_vdw->data;
127 vftabscale = _mm_set1_pd(kernel_data->table_vdw->scale);
129 /* Setup water-specific parameters */
130 inr = nlist->iinr[0];
131 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
132 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
133 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
134 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
136 /* Avoid stupid compiler warnings */
144 /* Start outer loop over neighborlists */
145 for(iidx=0; iidx<nri; iidx++)
147 /* Load shift vector for this list */
148 i_shift_offset = DIM*shiftidx[iidx];
150 /* Load limits for loop over neighbors */
151 j_index_start = jindex[iidx];
152 j_index_end = jindex[iidx+1];
154 /* Get outer coordinate index */
156 i_coord_offset = DIM*inr;
158 /* Load i particle coords and add shift vector */
159 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
160 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
162 fix0 = _mm_setzero_pd();
163 fiy0 = _mm_setzero_pd();
164 fiz0 = _mm_setzero_pd();
165 fix1 = _mm_setzero_pd();
166 fiy1 = _mm_setzero_pd();
167 fiz1 = _mm_setzero_pd();
168 fix2 = _mm_setzero_pd();
169 fiy2 = _mm_setzero_pd();
170 fiz2 = _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);
201 /* Calculate squared distance and things based on it */
202 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
203 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
204 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
206 rinv00 = gmx_mm_invsqrt_pd(rsq00);
207 rinv10 = gmx_mm_invsqrt_pd(rsq10);
208 rinv20 = gmx_mm_invsqrt_pd(rsq20);
210 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
211 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
212 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
214 /* Load parameters for j particles */
215 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
216 vdwjidx0A = 2*vdwtype[jnrA+0];
217 vdwjidx0B = 2*vdwtype[jnrB+0];
219 fjx0 = _mm_setzero_pd();
220 fjy0 = _mm_setzero_pd();
221 fjz0 = _mm_setzero_pd();
223 /**************************
224 * CALCULATE INTERACTIONS *
225 **************************/
227 r00 = _mm_mul_pd(rsq00,rinv00);
229 /* Compute parameters for interactions between i and j atoms */
230 qq00 = _mm_mul_pd(iq0,jq0);
231 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
232 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
234 /* Calculate table index by multiplying r with table scale and truncate to integer */
235 rt = _mm_mul_pd(r00,vftabscale);
236 vfitab = _mm_cvttpd_epi32(rt);
237 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
238 vfitab = _mm_slli_epi32(vfitab,3);
240 /* COULOMB ELECTROSTATICS */
241 velec = _mm_mul_pd(qq00,rinv00);
242 felec = _mm_mul_pd(velec,rinvsq00);
244 /* CUBIC SPLINE TABLE DISPERSION */
245 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
246 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
247 GMX_MM_TRANSPOSE2_PD(Y,F);
248 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
249 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
250 GMX_MM_TRANSPOSE2_PD(G,H);
251 Heps = _mm_mul_pd(vfeps,H);
252 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
253 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
254 vvdw6 = _mm_mul_pd(c6_00,VV);
255 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
256 fvdw6 = _mm_mul_pd(c6_00,FF);
258 /* CUBIC SPLINE TABLE REPULSION */
259 vfitab = _mm_add_epi32(vfitab,ifour);
260 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
261 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
262 GMX_MM_TRANSPOSE2_PD(Y,F);
263 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
264 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
265 GMX_MM_TRANSPOSE2_PD(G,H);
266 Heps = _mm_mul_pd(vfeps,H);
267 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
268 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
269 vvdw12 = _mm_mul_pd(c12_00,VV);
270 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
271 fvdw12 = _mm_mul_pd(c12_00,FF);
272 vvdw = _mm_add_pd(vvdw12,vvdw6);
273 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
275 /* Update potential sum for this i atom from the interaction with this j atom. */
276 velecsum = _mm_add_pd(velecsum,velec);
277 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
279 fscal = _mm_add_pd(felec,fvdw);
281 /* Calculate temporary vectorial force */
282 tx = _mm_mul_pd(fscal,dx00);
283 ty = _mm_mul_pd(fscal,dy00);
284 tz = _mm_mul_pd(fscal,dz00);
286 /* Update vectorial force */
287 fix0 = _mm_add_pd(fix0,tx);
288 fiy0 = _mm_add_pd(fiy0,ty);
289 fiz0 = _mm_add_pd(fiz0,tz);
291 fjx0 = _mm_add_pd(fjx0,tx);
292 fjy0 = _mm_add_pd(fjy0,ty);
293 fjz0 = _mm_add_pd(fjz0,tz);
295 /**************************
296 * CALCULATE INTERACTIONS *
297 **************************/
299 /* Compute parameters for interactions between i and j atoms */
300 qq10 = _mm_mul_pd(iq1,jq0);
302 /* COULOMB ELECTROSTATICS */
303 velec = _mm_mul_pd(qq10,rinv10);
304 felec = _mm_mul_pd(velec,rinvsq10);
306 /* Update potential sum for this i atom from the interaction with this j atom. */
307 velecsum = _mm_add_pd(velecsum,velec);
311 /* Calculate temporary vectorial force */
312 tx = _mm_mul_pd(fscal,dx10);
313 ty = _mm_mul_pd(fscal,dy10);
314 tz = _mm_mul_pd(fscal,dz10);
316 /* Update vectorial force */
317 fix1 = _mm_add_pd(fix1,tx);
318 fiy1 = _mm_add_pd(fiy1,ty);
319 fiz1 = _mm_add_pd(fiz1,tz);
321 fjx0 = _mm_add_pd(fjx0,tx);
322 fjy0 = _mm_add_pd(fjy0,ty);
323 fjz0 = _mm_add_pd(fjz0,tz);
325 /**************************
326 * CALCULATE INTERACTIONS *
327 **************************/
329 /* Compute parameters for interactions between i and j atoms */
330 qq20 = _mm_mul_pd(iq2,jq0);
332 /* COULOMB ELECTROSTATICS */
333 velec = _mm_mul_pd(qq20,rinv20);
334 felec = _mm_mul_pd(velec,rinvsq20);
336 /* Update potential sum for this i atom from the interaction with this j atom. */
337 velecsum = _mm_add_pd(velecsum,velec);
341 /* Calculate temporary vectorial force */
342 tx = _mm_mul_pd(fscal,dx20);
343 ty = _mm_mul_pd(fscal,dy20);
344 tz = _mm_mul_pd(fscal,dz20);
346 /* Update vectorial force */
347 fix2 = _mm_add_pd(fix2,tx);
348 fiy2 = _mm_add_pd(fiy2,ty);
349 fiz2 = _mm_add_pd(fiz2,tz);
351 fjx0 = _mm_add_pd(fjx0,tx);
352 fjy0 = _mm_add_pd(fjy0,ty);
353 fjz0 = _mm_add_pd(fjz0,tz);
355 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
357 /* Inner loop uses 122 flops */
364 j_coord_offsetA = DIM*jnrA;
366 /* load j atom coordinates */
367 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
370 /* Calculate displacement vector */
371 dx00 = _mm_sub_pd(ix0,jx0);
372 dy00 = _mm_sub_pd(iy0,jy0);
373 dz00 = _mm_sub_pd(iz0,jz0);
374 dx10 = _mm_sub_pd(ix1,jx0);
375 dy10 = _mm_sub_pd(iy1,jy0);
376 dz10 = _mm_sub_pd(iz1,jz0);
377 dx20 = _mm_sub_pd(ix2,jx0);
378 dy20 = _mm_sub_pd(iy2,jy0);
379 dz20 = _mm_sub_pd(iz2,jz0);
381 /* Calculate squared distance and things based on it */
382 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
383 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
384 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
386 rinv00 = gmx_mm_invsqrt_pd(rsq00);
387 rinv10 = gmx_mm_invsqrt_pd(rsq10);
388 rinv20 = gmx_mm_invsqrt_pd(rsq20);
390 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
391 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
392 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
394 /* Load parameters for j particles */
395 jq0 = _mm_load_sd(charge+jnrA+0);
396 vdwjidx0A = 2*vdwtype[jnrA+0];
398 fjx0 = _mm_setzero_pd();
399 fjy0 = _mm_setzero_pd();
400 fjz0 = _mm_setzero_pd();
402 /**************************
403 * CALCULATE INTERACTIONS *
404 **************************/
406 r00 = _mm_mul_pd(rsq00,rinv00);
408 /* Compute parameters for interactions between i and j atoms */
409 qq00 = _mm_mul_pd(iq0,jq0);
410 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
412 /* Calculate table index by multiplying r with table scale and truncate to integer */
413 rt = _mm_mul_pd(r00,vftabscale);
414 vfitab = _mm_cvttpd_epi32(rt);
415 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
416 vfitab = _mm_slli_epi32(vfitab,3);
418 /* COULOMB ELECTROSTATICS */
419 velec = _mm_mul_pd(qq00,rinv00);
420 felec = _mm_mul_pd(velec,rinvsq00);
422 /* CUBIC SPLINE TABLE DISPERSION */
423 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
424 F = _mm_setzero_pd();
425 GMX_MM_TRANSPOSE2_PD(Y,F);
426 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
427 H = _mm_setzero_pd();
428 GMX_MM_TRANSPOSE2_PD(G,H);
429 Heps = _mm_mul_pd(vfeps,H);
430 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
431 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
432 vvdw6 = _mm_mul_pd(c6_00,VV);
433 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
434 fvdw6 = _mm_mul_pd(c6_00,FF);
436 /* CUBIC SPLINE TABLE REPULSION */
437 vfitab = _mm_add_epi32(vfitab,ifour);
438 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
439 F = _mm_setzero_pd();
440 GMX_MM_TRANSPOSE2_PD(Y,F);
441 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
442 H = _mm_setzero_pd();
443 GMX_MM_TRANSPOSE2_PD(G,H);
444 Heps = _mm_mul_pd(vfeps,H);
445 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
446 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
447 vvdw12 = _mm_mul_pd(c12_00,VV);
448 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
449 fvdw12 = _mm_mul_pd(c12_00,FF);
450 vvdw = _mm_add_pd(vvdw12,vvdw6);
451 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
453 /* Update potential sum for this i atom from the interaction with this j atom. */
454 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
455 velecsum = _mm_add_pd(velecsum,velec);
456 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
457 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
459 fscal = _mm_add_pd(felec,fvdw);
461 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
463 /* Calculate temporary vectorial force */
464 tx = _mm_mul_pd(fscal,dx00);
465 ty = _mm_mul_pd(fscal,dy00);
466 tz = _mm_mul_pd(fscal,dz00);
468 /* Update vectorial force */
469 fix0 = _mm_add_pd(fix0,tx);
470 fiy0 = _mm_add_pd(fiy0,ty);
471 fiz0 = _mm_add_pd(fiz0,tz);
473 fjx0 = _mm_add_pd(fjx0,tx);
474 fjy0 = _mm_add_pd(fjy0,ty);
475 fjz0 = _mm_add_pd(fjz0,tz);
477 /**************************
478 * CALCULATE INTERACTIONS *
479 **************************/
481 /* Compute parameters for interactions between i and j atoms */
482 qq10 = _mm_mul_pd(iq1,jq0);
484 /* COULOMB ELECTROSTATICS */
485 velec = _mm_mul_pd(qq10,rinv10);
486 felec = _mm_mul_pd(velec,rinvsq10);
488 /* Update potential sum for this i atom from the interaction with this j atom. */
489 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
490 velecsum = _mm_add_pd(velecsum,velec);
494 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
496 /* Calculate temporary vectorial force */
497 tx = _mm_mul_pd(fscal,dx10);
498 ty = _mm_mul_pd(fscal,dy10);
499 tz = _mm_mul_pd(fscal,dz10);
501 /* Update vectorial force */
502 fix1 = _mm_add_pd(fix1,tx);
503 fiy1 = _mm_add_pd(fiy1,ty);
504 fiz1 = _mm_add_pd(fiz1,tz);
506 fjx0 = _mm_add_pd(fjx0,tx);
507 fjy0 = _mm_add_pd(fjy0,ty);
508 fjz0 = _mm_add_pd(fjz0,tz);
510 /**************************
511 * CALCULATE INTERACTIONS *
512 **************************/
514 /* Compute parameters for interactions between i and j atoms */
515 qq20 = _mm_mul_pd(iq2,jq0);
517 /* COULOMB ELECTROSTATICS */
518 velec = _mm_mul_pd(qq20,rinv20);
519 felec = _mm_mul_pd(velec,rinvsq20);
521 /* Update potential sum for this i atom from the interaction with this j atom. */
522 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
523 velecsum = _mm_add_pd(velecsum,velec);
527 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
529 /* Calculate temporary vectorial force */
530 tx = _mm_mul_pd(fscal,dx20);
531 ty = _mm_mul_pd(fscal,dy20);
532 tz = _mm_mul_pd(fscal,dz20);
534 /* Update vectorial force */
535 fix2 = _mm_add_pd(fix2,tx);
536 fiy2 = _mm_add_pd(fiy2,ty);
537 fiz2 = _mm_add_pd(fiz2,tz);
539 fjx0 = _mm_add_pd(fjx0,tx);
540 fjy0 = _mm_add_pd(fjy0,ty);
541 fjz0 = _mm_add_pd(fjz0,tz);
543 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
545 /* Inner loop uses 122 flops */
548 /* End of innermost loop */
550 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
551 f+i_coord_offset,fshift+i_shift_offset);
554 /* Update potential energies */
555 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
556 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
558 /* Increment number of inner iterations */
559 inneriter += j_index_end - j_index_start;
561 /* Outer loop uses 20 flops */
564 /* Increment number of outer iterations */
567 /* Update outer/inner flops */
569 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*122);
572 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_sse4_1_double
573 * Electrostatics interaction: Coulomb
574 * VdW interaction: CubicSplineTable
575 * Geometry: Water3-Particle
576 * Calculate force/pot: Force
579 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_sse4_1_double
580 (t_nblist * gmx_restrict nlist,
581 rvec * gmx_restrict xx,
582 rvec * gmx_restrict ff,
583 t_forcerec * gmx_restrict fr,
584 t_mdatoms * gmx_restrict mdatoms,
585 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
586 t_nrnb * gmx_restrict nrnb)
588 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
589 * just 0 for non-waters.
590 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
591 * jnr indices corresponding to data put in the four positions in the SIMD register.
593 int i_shift_offset,i_coord_offset,outeriter,inneriter;
594 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
596 int j_coord_offsetA,j_coord_offsetB;
597 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
599 real *shiftvec,*fshift,*x,*f;
600 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
602 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
604 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
606 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
607 int vdwjidx0A,vdwjidx0B;
608 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
609 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
610 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
611 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
612 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
615 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
618 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
619 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
621 __m128i ifour = _mm_set1_epi32(4);
622 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
624 __m128d dummy_mask,cutoff_mask;
625 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
626 __m128d one = _mm_set1_pd(1.0);
627 __m128d two = _mm_set1_pd(2.0);
633 jindex = nlist->jindex;
635 shiftidx = nlist->shift;
637 shiftvec = fr->shift_vec[0];
638 fshift = fr->fshift[0];
639 facel = _mm_set1_pd(fr->epsfac);
640 charge = mdatoms->chargeA;
641 nvdwtype = fr->ntype;
643 vdwtype = mdatoms->typeA;
645 vftab = kernel_data->table_vdw->data;
646 vftabscale = _mm_set1_pd(kernel_data->table_vdw->scale);
648 /* Setup water-specific parameters */
649 inr = nlist->iinr[0];
650 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
651 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
652 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
653 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
655 /* Avoid stupid compiler warnings */
663 /* Start outer loop over neighborlists */
664 for(iidx=0; iidx<nri; iidx++)
666 /* Load shift vector for this list */
667 i_shift_offset = DIM*shiftidx[iidx];
669 /* Load limits for loop over neighbors */
670 j_index_start = jindex[iidx];
671 j_index_end = jindex[iidx+1];
673 /* Get outer coordinate index */
675 i_coord_offset = DIM*inr;
677 /* Load i particle coords and add shift vector */
678 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
679 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
681 fix0 = _mm_setzero_pd();
682 fiy0 = _mm_setzero_pd();
683 fiz0 = _mm_setzero_pd();
684 fix1 = _mm_setzero_pd();
685 fiy1 = _mm_setzero_pd();
686 fiz1 = _mm_setzero_pd();
687 fix2 = _mm_setzero_pd();
688 fiy2 = _mm_setzero_pd();
689 fiz2 = _mm_setzero_pd();
691 /* Start inner kernel loop */
692 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
695 /* Get j neighbor index, and coordinate index */
698 j_coord_offsetA = DIM*jnrA;
699 j_coord_offsetB = DIM*jnrB;
701 /* load j atom coordinates */
702 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
705 /* Calculate displacement vector */
706 dx00 = _mm_sub_pd(ix0,jx0);
707 dy00 = _mm_sub_pd(iy0,jy0);
708 dz00 = _mm_sub_pd(iz0,jz0);
709 dx10 = _mm_sub_pd(ix1,jx0);
710 dy10 = _mm_sub_pd(iy1,jy0);
711 dz10 = _mm_sub_pd(iz1,jz0);
712 dx20 = _mm_sub_pd(ix2,jx0);
713 dy20 = _mm_sub_pd(iy2,jy0);
714 dz20 = _mm_sub_pd(iz2,jz0);
716 /* Calculate squared distance and things based on it */
717 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
718 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
719 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
721 rinv00 = gmx_mm_invsqrt_pd(rsq00);
722 rinv10 = gmx_mm_invsqrt_pd(rsq10);
723 rinv20 = gmx_mm_invsqrt_pd(rsq20);
725 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
726 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
727 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
729 /* Load parameters for j particles */
730 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
731 vdwjidx0A = 2*vdwtype[jnrA+0];
732 vdwjidx0B = 2*vdwtype[jnrB+0];
734 fjx0 = _mm_setzero_pd();
735 fjy0 = _mm_setzero_pd();
736 fjz0 = _mm_setzero_pd();
738 /**************************
739 * CALCULATE INTERACTIONS *
740 **************************/
742 r00 = _mm_mul_pd(rsq00,rinv00);
744 /* Compute parameters for interactions between i and j atoms */
745 qq00 = _mm_mul_pd(iq0,jq0);
746 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
747 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
749 /* Calculate table index by multiplying r with table scale and truncate to integer */
750 rt = _mm_mul_pd(r00,vftabscale);
751 vfitab = _mm_cvttpd_epi32(rt);
752 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
753 vfitab = _mm_slli_epi32(vfitab,3);
755 /* COULOMB ELECTROSTATICS */
756 velec = _mm_mul_pd(qq00,rinv00);
757 felec = _mm_mul_pd(velec,rinvsq00);
759 /* CUBIC SPLINE TABLE DISPERSION */
760 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
761 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
762 GMX_MM_TRANSPOSE2_PD(Y,F);
763 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
764 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
765 GMX_MM_TRANSPOSE2_PD(G,H);
766 Heps = _mm_mul_pd(vfeps,H);
767 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
768 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
769 fvdw6 = _mm_mul_pd(c6_00,FF);
771 /* CUBIC SPLINE TABLE REPULSION */
772 vfitab = _mm_add_epi32(vfitab,ifour);
773 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
774 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
775 GMX_MM_TRANSPOSE2_PD(Y,F);
776 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
777 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
778 GMX_MM_TRANSPOSE2_PD(G,H);
779 Heps = _mm_mul_pd(vfeps,H);
780 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
781 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
782 fvdw12 = _mm_mul_pd(c12_00,FF);
783 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
785 fscal = _mm_add_pd(felec,fvdw);
787 /* Calculate temporary vectorial force */
788 tx = _mm_mul_pd(fscal,dx00);
789 ty = _mm_mul_pd(fscal,dy00);
790 tz = _mm_mul_pd(fscal,dz00);
792 /* Update vectorial force */
793 fix0 = _mm_add_pd(fix0,tx);
794 fiy0 = _mm_add_pd(fiy0,ty);
795 fiz0 = _mm_add_pd(fiz0,tz);
797 fjx0 = _mm_add_pd(fjx0,tx);
798 fjy0 = _mm_add_pd(fjy0,ty);
799 fjz0 = _mm_add_pd(fjz0,tz);
801 /**************************
802 * CALCULATE INTERACTIONS *
803 **************************/
805 /* Compute parameters for interactions between i and j atoms */
806 qq10 = _mm_mul_pd(iq1,jq0);
808 /* COULOMB ELECTROSTATICS */
809 velec = _mm_mul_pd(qq10,rinv10);
810 felec = _mm_mul_pd(velec,rinvsq10);
814 /* Calculate temporary vectorial force */
815 tx = _mm_mul_pd(fscal,dx10);
816 ty = _mm_mul_pd(fscal,dy10);
817 tz = _mm_mul_pd(fscal,dz10);
819 /* Update vectorial force */
820 fix1 = _mm_add_pd(fix1,tx);
821 fiy1 = _mm_add_pd(fiy1,ty);
822 fiz1 = _mm_add_pd(fiz1,tz);
824 fjx0 = _mm_add_pd(fjx0,tx);
825 fjy0 = _mm_add_pd(fjy0,ty);
826 fjz0 = _mm_add_pd(fjz0,tz);
828 /**************************
829 * CALCULATE INTERACTIONS *
830 **************************/
832 /* Compute parameters for interactions between i and j atoms */
833 qq20 = _mm_mul_pd(iq2,jq0);
835 /* COULOMB ELECTROSTATICS */
836 velec = _mm_mul_pd(qq20,rinv20);
837 felec = _mm_mul_pd(velec,rinvsq20);
841 /* Calculate temporary vectorial force */
842 tx = _mm_mul_pd(fscal,dx20);
843 ty = _mm_mul_pd(fscal,dy20);
844 tz = _mm_mul_pd(fscal,dz20);
846 /* Update vectorial force */
847 fix2 = _mm_add_pd(fix2,tx);
848 fiy2 = _mm_add_pd(fiy2,ty);
849 fiz2 = _mm_add_pd(fiz2,tz);
851 fjx0 = _mm_add_pd(fjx0,tx);
852 fjy0 = _mm_add_pd(fjy0,ty);
853 fjz0 = _mm_add_pd(fjz0,tz);
855 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
857 /* Inner loop uses 111 flops */
864 j_coord_offsetA = DIM*jnrA;
866 /* load j atom coordinates */
867 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
870 /* Calculate displacement vector */
871 dx00 = _mm_sub_pd(ix0,jx0);
872 dy00 = _mm_sub_pd(iy0,jy0);
873 dz00 = _mm_sub_pd(iz0,jz0);
874 dx10 = _mm_sub_pd(ix1,jx0);
875 dy10 = _mm_sub_pd(iy1,jy0);
876 dz10 = _mm_sub_pd(iz1,jz0);
877 dx20 = _mm_sub_pd(ix2,jx0);
878 dy20 = _mm_sub_pd(iy2,jy0);
879 dz20 = _mm_sub_pd(iz2,jz0);
881 /* Calculate squared distance and things based on it */
882 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
883 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
884 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
886 rinv00 = gmx_mm_invsqrt_pd(rsq00);
887 rinv10 = gmx_mm_invsqrt_pd(rsq10);
888 rinv20 = gmx_mm_invsqrt_pd(rsq20);
890 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
891 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
892 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
894 /* Load parameters for j particles */
895 jq0 = _mm_load_sd(charge+jnrA+0);
896 vdwjidx0A = 2*vdwtype[jnrA+0];
898 fjx0 = _mm_setzero_pd();
899 fjy0 = _mm_setzero_pd();
900 fjz0 = _mm_setzero_pd();
902 /**************************
903 * CALCULATE INTERACTIONS *
904 **************************/
906 r00 = _mm_mul_pd(rsq00,rinv00);
908 /* Compute parameters for interactions between i and j atoms */
909 qq00 = _mm_mul_pd(iq0,jq0);
910 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
912 /* Calculate table index by multiplying r with table scale and truncate to integer */
913 rt = _mm_mul_pd(r00,vftabscale);
914 vfitab = _mm_cvttpd_epi32(rt);
915 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
916 vfitab = _mm_slli_epi32(vfitab,3);
918 /* COULOMB ELECTROSTATICS */
919 velec = _mm_mul_pd(qq00,rinv00);
920 felec = _mm_mul_pd(velec,rinvsq00);
922 /* CUBIC SPLINE TABLE DISPERSION */
923 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
924 F = _mm_setzero_pd();
925 GMX_MM_TRANSPOSE2_PD(Y,F);
926 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
927 H = _mm_setzero_pd();
928 GMX_MM_TRANSPOSE2_PD(G,H);
929 Heps = _mm_mul_pd(vfeps,H);
930 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
931 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
932 fvdw6 = _mm_mul_pd(c6_00,FF);
934 /* CUBIC SPLINE TABLE REPULSION */
935 vfitab = _mm_add_epi32(vfitab,ifour);
936 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
937 F = _mm_setzero_pd();
938 GMX_MM_TRANSPOSE2_PD(Y,F);
939 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
940 H = _mm_setzero_pd();
941 GMX_MM_TRANSPOSE2_PD(G,H);
942 Heps = _mm_mul_pd(vfeps,H);
943 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
944 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
945 fvdw12 = _mm_mul_pd(c12_00,FF);
946 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
948 fscal = _mm_add_pd(felec,fvdw);
950 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
952 /* Calculate temporary vectorial force */
953 tx = _mm_mul_pd(fscal,dx00);
954 ty = _mm_mul_pd(fscal,dy00);
955 tz = _mm_mul_pd(fscal,dz00);
957 /* Update vectorial force */
958 fix0 = _mm_add_pd(fix0,tx);
959 fiy0 = _mm_add_pd(fiy0,ty);
960 fiz0 = _mm_add_pd(fiz0,tz);
962 fjx0 = _mm_add_pd(fjx0,tx);
963 fjy0 = _mm_add_pd(fjy0,ty);
964 fjz0 = _mm_add_pd(fjz0,tz);
966 /**************************
967 * CALCULATE INTERACTIONS *
968 **************************/
970 /* Compute parameters for interactions between i and j atoms */
971 qq10 = _mm_mul_pd(iq1,jq0);
973 /* COULOMB ELECTROSTATICS */
974 velec = _mm_mul_pd(qq10,rinv10);
975 felec = _mm_mul_pd(velec,rinvsq10);
979 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
981 /* Calculate temporary vectorial force */
982 tx = _mm_mul_pd(fscal,dx10);
983 ty = _mm_mul_pd(fscal,dy10);
984 tz = _mm_mul_pd(fscal,dz10);
986 /* Update vectorial force */
987 fix1 = _mm_add_pd(fix1,tx);
988 fiy1 = _mm_add_pd(fiy1,ty);
989 fiz1 = _mm_add_pd(fiz1,tz);
991 fjx0 = _mm_add_pd(fjx0,tx);
992 fjy0 = _mm_add_pd(fjy0,ty);
993 fjz0 = _mm_add_pd(fjz0,tz);
995 /**************************
996 * CALCULATE INTERACTIONS *
997 **************************/
999 /* Compute parameters for interactions between i and j atoms */
1000 qq20 = _mm_mul_pd(iq2,jq0);
1002 /* COULOMB ELECTROSTATICS */
1003 velec = _mm_mul_pd(qq20,rinv20);
1004 felec = _mm_mul_pd(velec,rinvsq20);
1008 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1010 /* Calculate temporary vectorial force */
1011 tx = _mm_mul_pd(fscal,dx20);
1012 ty = _mm_mul_pd(fscal,dy20);
1013 tz = _mm_mul_pd(fscal,dz20);
1015 /* Update vectorial force */
1016 fix2 = _mm_add_pd(fix2,tx);
1017 fiy2 = _mm_add_pd(fiy2,ty);
1018 fiz2 = _mm_add_pd(fiz2,tz);
1020 fjx0 = _mm_add_pd(fjx0,tx);
1021 fjy0 = _mm_add_pd(fjy0,ty);
1022 fjz0 = _mm_add_pd(fjz0,tz);
1024 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1026 /* Inner loop uses 111 flops */
1029 /* End of innermost loop */
1031 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1032 f+i_coord_offset,fshift+i_shift_offset);
1034 /* Increment number of inner iterations */
1035 inneriter += j_index_end - j_index_start;
1037 /* Outer loop uses 18 flops */
1040 /* Increment number of outer iterations */
1043 /* Update outer/inner flops */
1045 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*111);