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.
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_avx_128_fma_double.h"
50 #include "kernelutil_x86_avx_128_fma_double.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwCSTab_GeomW3P1_VF_avx_128_fma_double
54 * Electrostatics interaction: ReactionField
55 * VdW interaction: CubicSplineTable
56 * Geometry: Water3-Particle
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_VF_avx_128_fma_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,twovfeps;
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 krf = _mm_set1_pd(fr->ic->k_rf);
123 krf2 = _mm_set1_pd(fr->ic->k_rf*2.0);
124 crf = _mm_set1_pd(fr->ic->c_rf);
125 nvdwtype = fr->ntype;
127 vdwtype = mdatoms->typeA;
129 vftab = kernel_data->table_vdw->data;
130 vftabscale = _mm_set1_pd(kernel_data->table_vdw->scale);
132 /* Setup water-specific parameters */
133 inr = nlist->iinr[0];
134 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
135 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
136 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
137 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
139 /* Avoid stupid compiler warnings */
147 /* Start outer loop over neighborlists */
148 for(iidx=0; iidx<nri; iidx++)
150 /* Load shift vector for this list */
151 i_shift_offset = DIM*shiftidx[iidx];
153 /* Load limits for loop over neighbors */
154 j_index_start = jindex[iidx];
155 j_index_end = jindex[iidx+1];
157 /* Get outer coordinate index */
159 i_coord_offset = DIM*inr;
161 /* Load i particle coords and add shift vector */
162 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
163 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
165 fix0 = _mm_setzero_pd();
166 fiy0 = _mm_setzero_pd();
167 fiz0 = _mm_setzero_pd();
168 fix1 = _mm_setzero_pd();
169 fiy1 = _mm_setzero_pd();
170 fiz1 = _mm_setzero_pd();
171 fix2 = _mm_setzero_pd();
172 fiy2 = _mm_setzero_pd();
173 fiz2 = _mm_setzero_pd();
175 /* Reset potential sums */
176 velecsum = _mm_setzero_pd();
177 vvdwsum = _mm_setzero_pd();
179 /* Start inner kernel loop */
180 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
183 /* Get j neighbor index, and coordinate index */
186 j_coord_offsetA = DIM*jnrA;
187 j_coord_offsetB = DIM*jnrB;
189 /* load j atom coordinates */
190 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
193 /* Calculate displacement vector */
194 dx00 = _mm_sub_pd(ix0,jx0);
195 dy00 = _mm_sub_pd(iy0,jy0);
196 dz00 = _mm_sub_pd(iz0,jz0);
197 dx10 = _mm_sub_pd(ix1,jx0);
198 dy10 = _mm_sub_pd(iy1,jy0);
199 dz10 = _mm_sub_pd(iz1,jz0);
200 dx20 = _mm_sub_pd(ix2,jx0);
201 dy20 = _mm_sub_pd(iy2,jy0);
202 dz20 = _mm_sub_pd(iz2,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);
209 rinv00 = gmx_mm_invsqrt_pd(rsq00);
210 rinv10 = gmx_mm_invsqrt_pd(rsq10);
211 rinv20 = gmx_mm_invsqrt_pd(rsq20);
213 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
214 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
215 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
217 /* Load parameters for j particles */
218 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
219 vdwjidx0A = 2*vdwtype[jnrA+0];
220 vdwjidx0B = 2*vdwtype[jnrB+0];
222 fjx0 = _mm_setzero_pd();
223 fjy0 = _mm_setzero_pd();
224 fjz0 = _mm_setzero_pd();
226 /**************************
227 * CALCULATE INTERACTIONS *
228 **************************/
230 r00 = _mm_mul_pd(rsq00,rinv00);
232 /* Compute parameters for interactions between i and j atoms */
233 qq00 = _mm_mul_pd(iq0,jq0);
234 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
235 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
237 /* Calculate table index by multiplying r with table scale and truncate to integer */
238 rt = _mm_mul_pd(r00,vftabscale);
239 vfitab = _mm_cvttpd_epi32(rt);
241 vfeps = _mm_frcz_pd(rt);
243 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
245 twovfeps = _mm_add_pd(vfeps,vfeps);
246 vfitab = _mm_slli_epi32(vfitab,3);
248 /* REACTION-FIELD ELECTROSTATICS */
249 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
250 felec = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
252 /* CUBIC SPLINE TABLE DISPERSION */
253 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
254 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
255 GMX_MM_TRANSPOSE2_PD(Y,F);
256 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
257 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
258 GMX_MM_TRANSPOSE2_PD(G,H);
259 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
260 VV = _mm_macc_pd(vfeps,Fp,Y);
261 vvdw6 = _mm_mul_pd(c6_00,VV);
262 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
263 fvdw6 = _mm_mul_pd(c6_00,FF);
265 /* CUBIC SPLINE TABLE REPULSION */
266 vfitab = _mm_add_epi32(vfitab,ifour);
267 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
268 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
269 GMX_MM_TRANSPOSE2_PD(Y,F);
270 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
271 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
272 GMX_MM_TRANSPOSE2_PD(G,H);
273 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
274 VV = _mm_macc_pd(vfeps,Fp,Y);
275 vvdw12 = _mm_mul_pd(c12_00,VV);
276 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
277 fvdw12 = _mm_mul_pd(c12_00,FF);
278 vvdw = _mm_add_pd(vvdw12,vvdw6);
279 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
281 /* Update potential sum for this i atom from the interaction with this j atom. */
282 velecsum = _mm_add_pd(velecsum,velec);
283 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
285 fscal = _mm_add_pd(felec,fvdw);
287 /* Update vectorial force */
288 fix0 = _mm_macc_pd(dx00,fscal,fix0);
289 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
290 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
292 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
293 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
294 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
296 /**************************
297 * CALCULATE INTERACTIONS *
298 **************************/
300 /* Compute parameters for interactions between i and j atoms */
301 qq10 = _mm_mul_pd(iq1,jq0);
303 /* REACTION-FIELD ELECTROSTATICS */
304 velec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
305 felec = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
307 /* Update potential sum for this i atom from the interaction with this j atom. */
308 velecsum = _mm_add_pd(velecsum,velec);
312 /* Update vectorial force */
313 fix1 = _mm_macc_pd(dx10,fscal,fix1);
314 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
315 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
317 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
318 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
319 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
321 /**************************
322 * CALCULATE INTERACTIONS *
323 **************************/
325 /* Compute parameters for interactions between i and j atoms */
326 qq20 = _mm_mul_pd(iq2,jq0);
328 /* REACTION-FIELD ELECTROSTATICS */
329 velec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
330 felec = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
332 /* Update potential sum for this i atom from the interaction with this j atom. */
333 velecsum = _mm_add_pd(velecsum,velec);
337 /* Update vectorial force */
338 fix2 = _mm_macc_pd(dx20,fscal,fix2);
339 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
340 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
342 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
343 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
344 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
346 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
348 /* Inner loop uses 143 flops */
355 j_coord_offsetA = DIM*jnrA;
357 /* load j atom coordinates */
358 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
361 /* Calculate displacement vector */
362 dx00 = _mm_sub_pd(ix0,jx0);
363 dy00 = _mm_sub_pd(iy0,jy0);
364 dz00 = _mm_sub_pd(iz0,jz0);
365 dx10 = _mm_sub_pd(ix1,jx0);
366 dy10 = _mm_sub_pd(iy1,jy0);
367 dz10 = _mm_sub_pd(iz1,jz0);
368 dx20 = _mm_sub_pd(ix2,jx0);
369 dy20 = _mm_sub_pd(iy2,jy0);
370 dz20 = _mm_sub_pd(iz2,jz0);
372 /* Calculate squared distance and things based on it */
373 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
374 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
375 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
377 rinv00 = gmx_mm_invsqrt_pd(rsq00);
378 rinv10 = gmx_mm_invsqrt_pd(rsq10);
379 rinv20 = gmx_mm_invsqrt_pd(rsq20);
381 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
382 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
383 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
385 /* Load parameters for j particles */
386 jq0 = _mm_load_sd(charge+jnrA+0);
387 vdwjidx0A = 2*vdwtype[jnrA+0];
389 fjx0 = _mm_setzero_pd();
390 fjy0 = _mm_setzero_pd();
391 fjz0 = _mm_setzero_pd();
393 /**************************
394 * CALCULATE INTERACTIONS *
395 **************************/
397 r00 = _mm_mul_pd(rsq00,rinv00);
399 /* Compute parameters for interactions between i and j atoms */
400 qq00 = _mm_mul_pd(iq0,jq0);
401 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
403 /* Calculate table index by multiplying r with table scale and truncate to integer */
404 rt = _mm_mul_pd(r00,vftabscale);
405 vfitab = _mm_cvttpd_epi32(rt);
407 vfeps = _mm_frcz_pd(rt);
409 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
411 twovfeps = _mm_add_pd(vfeps,vfeps);
412 vfitab = _mm_slli_epi32(vfitab,3);
414 /* REACTION-FIELD ELECTROSTATICS */
415 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
416 felec = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
418 /* CUBIC SPLINE TABLE DISPERSION */
419 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
420 F = _mm_setzero_pd();
421 GMX_MM_TRANSPOSE2_PD(Y,F);
422 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
423 H = _mm_setzero_pd();
424 GMX_MM_TRANSPOSE2_PD(G,H);
425 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
426 VV = _mm_macc_pd(vfeps,Fp,Y);
427 vvdw6 = _mm_mul_pd(c6_00,VV);
428 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
429 fvdw6 = _mm_mul_pd(c6_00,FF);
431 /* CUBIC SPLINE TABLE REPULSION */
432 vfitab = _mm_add_epi32(vfitab,ifour);
433 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
434 F = _mm_setzero_pd();
435 GMX_MM_TRANSPOSE2_PD(Y,F);
436 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
437 H = _mm_setzero_pd();
438 GMX_MM_TRANSPOSE2_PD(G,H);
439 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
440 VV = _mm_macc_pd(vfeps,Fp,Y);
441 vvdw12 = _mm_mul_pd(c12_00,VV);
442 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
443 fvdw12 = _mm_mul_pd(c12_00,FF);
444 vvdw = _mm_add_pd(vvdw12,vvdw6);
445 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
447 /* Update potential sum for this i atom from the interaction with this j atom. */
448 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
449 velecsum = _mm_add_pd(velecsum,velec);
450 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
451 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
453 fscal = _mm_add_pd(felec,fvdw);
455 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
457 /* Update vectorial force */
458 fix0 = _mm_macc_pd(dx00,fscal,fix0);
459 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
460 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
462 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
463 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
464 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
466 /**************************
467 * CALCULATE INTERACTIONS *
468 **************************/
470 /* Compute parameters for interactions between i and j atoms */
471 qq10 = _mm_mul_pd(iq1,jq0);
473 /* REACTION-FIELD ELECTROSTATICS */
474 velec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
475 felec = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
477 /* Update potential sum for this i atom from the interaction with this j atom. */
478 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
479 velecsum = _mm_add_pd(velecsum,velec);
483 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
485 /* Update vectorial force */
486 fix1 = _mm_macc_pd(dx10,fscal,fix1);
487 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
488 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
490 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
491 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
492 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
494 /**************************
495 * CALCULATE INTERACTIONS *
496 **************************/
498 /* Compute parameters for interactions between i and j atoms */
499 qq20 = _mm_mul_pd(iq2,jq0);
501 /* REACTION-FIELD ELECTROSTATICS */
502 velec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
503 felec = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
505 /* Update potential sum for this i atom from the interaction with this j atom. */
506 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
507 velecsum = _mm_add_pd(velecsum,velec);
511 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
513 /* Update vectorial force */
514 fix2 = _mm_macc_pd(dx20,fscal,fix2);
515 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
516 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
518 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
519 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
520 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
522 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
524 /* Inner loop uses 143 flops */
527 /* End of innermost loop */
529 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
530 f+i_coord_offset,fshift+i_shift_offset);
533 /* Update potential energies */
534 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
535 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
537 /* Increment number of inner iterations */
538 inneriter += j_index_end - j_index_start;
540 /* Outer loop uses 20 flops */
543 /* Increment number of outer iterations */
546 /* Update outer/inner flops */
548 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*143);
551 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_avx_128_fma_double
552 * Electrostatics interaction: ReactionField
553 * VdW interaction: CubicSplineTable
554 * Geometry: Water3-Particle
555 * Calculate force/pot: Force
558 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_avx_128_fma_double
559 (t_nblist * gmx_restrict nlist,
560 rvec * gmx_restrict xx,
561 rvec * gmx_restrict ff,
562 t_forcerec * gmx_restrict fr,
563 t_mdatoms * gmx_restrict mdatoms,
564 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
565 t_nrnb * gmx_restrict nrnb)
567 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
568 * just 0 for non-waters.
569 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
570 * jnr indices corresponding to data put in the four positions in the SIMD register.
572 int i_shift_offset,i_coord_offset,outeriter,inneriter;
573 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
575 int j_coord_offsetA,j_coord_offsetB;
576 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
578 real *shiftvec,*fshift,*x,*f;
579 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
581 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
583 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
585 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
586 int vdwjidx0A,vdwjidx0B;
587 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
588 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
589 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
590 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
591 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
594 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
597 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
598 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
600 __m128i ifour = _mm_set1_epi32(4);
601 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
603 __m128d dummy_mask,cutoff_mask;
604 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
605 __m128d one = _mm_set1_pd(1.0);
606 __m128d two = _mm_set1_pd(2.0);
612 jindex = nlist->jindex;
614 shiftidx = nlist->shift;
616 shiftvec = fr->shift_vec[0];
617 fshift = fr->fshift[0];
618 facel = _mm_set1_pd(fr->epsfac);
619 charge = mdatoms->chargeA;
620 krf = _mm_set1_pd(fr->ic->k_rf);
621 krf2 = _mm_set1_pd(fr->ic->k_rf*2.0);
622 crf = _mm_set1_pd(fr->ic->c_rf);
623 nvdwtype = fr->ntype;
625 vdwtype = mdatoms->typeA;
627 vftab = kernel_data->table_vdw->data;
628 vftabscale = _mm_set1_pd(kernel_data->table_vdw->scale);
630 /* Setup water-specific parameters */
631 inr = nlist->iinr[0];
632 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
633 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
634 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
635 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
637 /* Avoid stupid compiler warnings */
645 /* Start outer loop over neighborlists */
646 for(iidx=0; iidx<nri; iidx++)
648 /* Load shift vector for this list */
649 i_shift_offset = DIM*shiftidx[iidx];
651 /* Load limits for loop over neighbors */
652 j_index_start = jindex[iidx];
653 j_index_end = jindex[iidx+1];
655 /* Get outer coordinate index */
657 i_coord_offset = DIM*inr;
659 /* Load i particle coords and add shift vector */
660 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
661 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
663 fix0 = _mm_setzero_pd();
664 fiy0 = _mm_setzero_pd();
665 fiz0 = _mm_setzero_pd();
666 fix1 = _mm_setzero_pd();
667 fiy1 = _mm_setzero_pd();
668 fiz1 = _mm_setzero_pd();
669 fix2 = _mm_setzero_pd();
670 fiy2 = _mm_setzero_pd();
671 fiz2 = _mm_setzero_pd();
673 /* Start inner kernel loop */
674 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
677 /* Get j neighbor index, and coordinate index */
680 j_coord_offsetA = DIM*jnrA;
681 j_coord_offsetB = DIM*jnrB;
683 /* load j atom coordinates */
684 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
687 /* Calculate displacement vector */
688 dx00 = _mm_sub_pd(ix0,jx0);
689 dy00 = _mm_sub_pd(iy0,jy0);
690 dz00 = _mm_sub_pd(iz0,jz0);
691 dx10 = _mm_sub_pd(ix1,jx0);
692 dy10 = _mm_sub_pd(iy1,jy0);
693 dz10 = _mm_sub_pd(iz1,jz0);
694 dx20 = _mm_sub_pd(ix2,jx0);
695 dy20 = _mm_sub_pd(iy2,jy0);
696 dz20 = _mm_sub_pd(iz2,jz0);
698 /* Calculate squared distance and things based on it */
699 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
700 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
701 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
703 rinv00 = gmx_mm_invsqrt_pd(rsq00);
704 rinv10 = gmx_mm_invsqrt_pd(rsq10);
705 rinv20 = gmx_mm_invsqrt_pd(rsq20);
707 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
708 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
709 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
711 /* Load parameters for j particles */
712 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
713 vdwjidx0A = 2*vdwtype[jnrA+0];
714 vdwjidx0B = 2*vdwtype[jnrB+0];
716 fjx0 = _mm_setzero_pd();
717 fjy0 = _mm_setzero_pd();
718 fjz0 = _mm_setzero_pd();
720 /**************************
721 * CALCULATE INTERACTIONS *
722 **************************/
724 r00 = _mm_mul_pd(rsq00,rinv00);
726 /* Compute parameters for interactions between i and j atoms */
727 qq00 = _mm_mul_pd(iq0,jq0);
728 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
729 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
731 /* Calculate table index by multiplying r with table scale and truncate to integer */
732 rt = _mm_mul_pd(r00,vftabscale);
733 vfitab = _mm_cvttpd_epi32(rt);
735 vfeps = _mm_frcz_pd(rt);
737 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
739 twovfeps = _mm_add_pd(vfeps,vfeps);
740 vfitab = _mm_slli_epi32(vfitab,3);
742 /* REACTION-FIELD ELECTROSTATICS */
743 felec = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
745 /* CUBIC SPLINE TABLE DISPERSION */
746 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
747 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
748 GMX_MM_TRANSPOSE2_PD(Y,F);
749 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
750 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
751 GMX_MM_TRANSPOSE2_PD(G,H);
752 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
753 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
754 fvdw6 = _mm_mul_pd(c6_00,FF);
756 /* CUBIC SPLINE TABLE REPULSION */
757 vfitab = _mm_add_epi32(vfitab,ifour);
758 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
759 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
760 GMX_MM_TRANSPOSE2_PD(Y,F);
761 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
762 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
763 GMX_MM_TRANSPOSE2_PD(G,H);
764 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
765 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
766 fvdw12 = _mm_mul_pd(c12_00,FF);
767 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
769 fscal = _mm_add_pd(felec,fvdw);
771 /* Update vectorial force */
772 fix0 = _mm_macc_pd(dx00,fscal,fix0);
773 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
774 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
776 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
777 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
778 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
780 /**************************
781 * CALCULATE INTERACTIONS *
782 **************************/
784 /* Compute parameters for interactions between i and j atoms */
785 qq10 = _mm_mul_pd(iq1,jq0);
787 /* REACTION-FIELD ELECTROSTATICS */
788 felec = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
792 /* Update vectorial force */
793 fix1 = _mm_macc_pd(dx10,fscal,fix1);
794 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
795 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
797 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
798 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
799 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
801 /**************************
802 * CALCULATE INTERACTIONS *
803 **************************/
805 /* Compute parameters for interactions between i and j atoms */
806 qq20 = _mm_mul_pd(iq2,jq0);
808 /* REACTION-FIELD ELECTROSTATICS */
809 felec = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
813 /* Update vectorial force */
814 fix2 = _mm_macc_pd(dx20,fscal,fix2);
815 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
816 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
818 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
819 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
820 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
822 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
824 /* Inner loop uses 120 flops */
831 j_coord_offsetA = DIM*jnrA;
833 /* load j atom coordinates */
834 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
837 /* Calculate displacement vector */
838 dx00 = _mm_sub_pd(ix0,jx0);
839 dy00 = _mm_sub_pd(iy0,jy0);
840 dz00 = _mm_sub_pd(iz0,jz0);
841 dx10 = _mm_sub_pd(ix1,jx0);
842 dy10 = _mm_sub_pd(iy1,jy0);
843 dz10 = _mm_sub_pd(iz1,jz0);
844 dx20 = _mm_sub_pd(ix2,jx0);
845 dy20 = _mm_sub_pd(iy2,jy0);
846 dz20 = _mm_sub_pd(iz2,jz0);
848 /* Calculate squared distance and things based on it */
849 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
850 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
851 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
853 rinv00 = gmx_mm_invsqrt_pd(rsq00);
854 rinv10 = gmx_mm_invsqrt_pd(rsq10);
855 rinv20 = gmx_mm_invsqrt_pd(rsq20);
857 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
858 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
859 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
861 /* Load parameters for j particles */
862 jq0 = _mm_load_sd(charge+jnrA+0);
863 vdwjidx0A = 2*vdwtype[jnrA+0];
865 fjx0 = _mm_setzero_pd();
866 fjy0 = _mm_setzero_pd();
867 fjz0 = _mm_setzero_pd();
869 /**************************
870 * CALCULATE INTERACTIONS *
871 **************************/
873 r00 = _mm_mul_pd(rsq00,rinv00);
875 /* Compute parameters for interactions between i and j atoms */
876 qq00 = _mm_mul_pd(iq0,jq0);
877 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
879 /* Calculate table index by multiplying r with table scale and truncate to integer */
880 rt = _mm_mul_pd(r00,vftabscale);
881 vfitab = _mm_cvttpd_epi32(rt);
883 vfeps = _mm_frcz_pd(rt);
885 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
887 twovfeps = _mm_add_pd(vfeps,vfeps);
888 vfitab = _mm_slli_epi32(vfitab,3);
890 /* REACTION-FIELD ELECTROSTATICS */
891 felec = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
893 /* CUBIC SPLINE TABLE DISPERSION */
894 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
895 F = _mm_setzero_pd();
896 GMX_MM_TRANSPOSE2_PD(Y,F);
897 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
898 H = _mm_setzero_pd();
899 GMX_MM_TRANSPOSE2_PD(G,H);
900 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
901 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
902 fvdw6 = _mm_mul_pd(c6_00,FF);
904 /* CUBIC SPLINE TABLE REPULSION */
905 vfitab = _mm_add_epi32(vfitab,ifour);
906 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
907 F = _mm_setzero_pd();
908 GMX_MM_TRANSPOSE2_PD(Y,F);
909 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
910 H = _mm_setzero_pd();
911 GMX_MM_TRANSPOSE2_PD(G,H);
912 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
913 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
914 fvdw12 = _mm_mul_pd(c12_00,FF);
915 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
917 fscal = _mm_add_pd(felec,fvdw);
919 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
921 /* Update vectorial force */
922 fix0 = _mm_macc_pd(dx00,fscal,fix0);
923 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
924 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
926 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
927 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
928 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
930 /**************************
931 * CALCULATE INTERACTIONS *
932 **************************/
934 /* Compute parameters for interactions between i and j atoms */
935 qq10 = _mm_mul_pd(iq1,jq0);
937 /* REACTION-FIELD ELECTROSTATICS */
938 felec = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
942 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
944 /* Update vectorial force */
945 fix1 = _mm_macc_pd(dx10,fscal,fix1);
946 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
947 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
949 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
950 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
951 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
953 /**************************
954 * CALCULATE INTERACTIONS *
955 **************************/
957 /* Compute parameters for interactions between i and j atoms */
958 qq20 = _mm_mul_pd(iq2,jq0);
960 /* REACTION-FIELD ELECTROSTATICS */
961 felec = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
965 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
967 /* Update vectorial force */
968 fix2 = _mm_macc_pd(dx20,fscal,fix2);
969 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
970 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
972 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
973 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
974 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
976 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
978 /* Inner loop uses 120 flops */
981 /* End of innermost loop */
983 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
984 f+i_coord_offset,fshift+i_shift_offset);
986 /* Increment number of inner iterations */
987 inneriter += j_index_end - j_index_start;
989 /* Outer loop uses 18 flops */
992 /* Increment number of outer iterations */
995 /* Update outer/inner flops */
997 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*120);