2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014,2015,2017,2018, 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/gmxlib/nrnb.h"
47 #include "kernelutil_x86_avx_128_fma_double.h"
50 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_VF_avx_128_fma_double
51 * Electrostatics interaction: Ewald
52 * VdW interaction: LennardJones
53 * Geometry: Water3-Particle
54 * Calculate force/pot: PotentialAndForce
57 nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_VF_avx_128_fma_double
58 (t_nblist * gmx_restrict nlist,
59 rvec * gmx_restrict xx,
60 rvec * gmx_restrict ff,
61 struct t_forcerec * gmx_restrict fr,
62 t_mdatoms * gmx_restrict mdatoms,
63 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
64 t_nrnb * gmx_restrict nrnb)
66 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
67 * just 0 for non-waters.
68 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
69 * jnr indices corresponding to data put in the four positions in the SIMD register.
71 int i_shift_offset,i_coord_offset,outeriter,inneriter;
72 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
74 int j_coord_offsetA,j_coord_offsetB;
75 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
77 real *shiftvec,*fshift,*x,*f;
78 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
80 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
82 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
84 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
85 int vdwjidx0A,vdwjidx0B;
86 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
87 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
88 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
89 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
90 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
93 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
96 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
97 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
99 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
101 __m128d dummy_mask,cutoff_mask;
102 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
103 __m128d one = _mm_set1_pd(1.0);
104 __m128d two = _mm_set1_pd(2.0);
110 jindex = nlist->jindex;
112 shiftidx = nlist->shift;
114 shiftvec = fr->shift_vec[0];
115 fshift = fr->fshift[0];
116 facel = _mm_set1_pd(fr->ic->epsfac);
117 charge = mdatoms->chargeA;
118 nvdwtype = fr->ntype;
120 vdwtype = mdatoms->typeA;
122 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
123 ewtab = fr->ic->tabq_coul_FDV0;
124 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
125 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
127 /* Setup water-specific parameters */
128 inr = nlist->iinr[0];
129 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
130 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
131 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
132 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
134 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
135 rcutoff_scalar = fr->ic->rcoulomb;
136 rcutoff = _mm_set1_pd(rcutoff_scalar);
137 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
139 sh_vdw_invrcut6 = _mm_set1_pd(fr->ic->sh_invrc6);
140 rvdw = _mm_set1_pd(fr->ic->rvdw);
142 /* Avoid stupid compiler warnings */
150 /* Start outer loop over neighborlists */
151 for(iidx=0; iidx<nri; iidx++)
153 /* Load shift vector for this list */
154 i_shift_offset = DIM*shiftidx[iidx];
156 /* Load limits for loop over neighbors */
157 j_index_start = jindex[iidx];
158 j_index_end = jindex[iidx+1];
160 /* Get outer coordinate index */
162 i_coord_offset = DIM*inr;
164 /* Load i particle coords and add shift vector */
165 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
166 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
168 fix0 = _mm_setzero_pd();
169 fiy0 = _mm_setzero_pd();
170 fiz0 = _mm_setzero_pd();
171 fix1 = _mm_setzero_pd();
172 fiy1 = _mm_setzero_pd();
173 fiz1 = _mm_setzero_pd();
174 fix2 = _mm_setzero_pd();
175 fiy2 = _mm_setzero_pd();
176 fiz2 = _mm_setzero_pd();
178 /* Reset potential sums */
179 velecsum = _mm_setzero_pd();
180 vvdwsum = _mm_setzero_pd();
182 /* Start inner kernel loop */
183 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
186 /* Get j neighbor index, and coordinate index */
189 j_coord_offsetA = DIM*jnrA;
190 j_coord_offsetB = DIM*jnrB;
192 /* load j atom coordinates */
193 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
196 /* Calculate displacement vector */
197 dx00 = _mm_sub_pd(ix0,jx0);
198 dy00 = _mm_sub_pd(iy0,jy0);
199 dz00 = _mm_sub_pd(iz0,jz0);
200 dx10 = _mm_sub_pd(ix1,jx0);
201 dy10 = _mm_sub_pd(iy1,jy0);
202 dz10 = _mm_sub_pd(iz1,jz0);
203 dx20 = _mm_sub_pd(ix2,jx0);
204 dy20 = _mm_sub_pd(iy2,jy0);
205 dz20 = _mm_sub_pd(iz2,jz0);
207 /* Calculate squared distance and things based on it */
208 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
209 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
210 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
212 rinv00 = avx128fma_invsqrt_d(rsq00);
213 rinv10 = avx128fma_invsqrt_d(rsq10);
214 rinv20 = avx128fma_invsqrt_d(rsq20);
216 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
217 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
218 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
220 /* Load parameters for j particles */
221 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
222 vdwjidx0A = 2*vdwtype[jnrA+0];
223 vdwjidx0B = 2*vdwtype[jnrB+0];
225 fjx0 = _mm_setzero_pd();
226 fjy0 = _mm_setzero_pd();
227 fjz0 = _mm_setzero_pd();
229 /**************************
230 * CALCULATE INTERACTIONS *
231 **************************/
233 if (gmx_mm_any_lt(rsq00,rcutoff2))
236 r00 = _mm_mul_pd(rsq00,rinv00);
238 /* Compute parameters for interactions between i and j atoms */
239 qq00 = _mm_mul_pd(iq0,jq0);
240 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
241 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
243 /* EWALD ELECTROSTATICS */
245 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
246 ewrt = _mm_mul_pd(r00,ewtabscale);
247 ewitab = _mm_cvttpd_epi32(ewrt);
249 eweps = _mm_frcz_pd(ewrt);
251 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
253 twoeweps = _mm_add_pd(eweps,eweps);
254 ewitab = _mm_slli_epi32(ewitab,2);
255 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
256 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
257 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
258 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
259 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
260 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
261 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
262 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
263 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
264 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
266 /* LENNARD-JONES DISPERSION/REPULSION */
268 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
269 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
270 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
271 vvdw = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
272 _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
273 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
275 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
277 /* Update potential sum for this i atom from the interaction with this j atom. */
278 velec = _mm_and_pd(velec,cutoff_mask);
279 velecsum = _mm_add_pd(velecsum,velec);
280 vvdw = _mm_and_pd(vvdw,cutoff_mask);
281 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
283 fscal = _mm_add_pd(felec,fvdw);
285 fscal = _mm_and_pd(fscal,cutoff_mask);
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);
298 /**************************
299 * CALCULATE INTERACTIONS *
300 **************************/
302 if (gmx_mm_any_lt(rsq10,rcutoff2))
305 r10 = _mm_mul_pd(rsq10,rinv10);
307 /* Compute parameters for interactions between i and j atoms */
308 qq10 = _mm_mul_pd(iq1,jq0);
310 /* EWALD ELECTROSTATICS */
312 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
313 ewrt = _mm_mul_pd(r10,ewtabscale);
314 ewitab = _mm_cvttpd_epi32(ewrt);
316 eweps = _mm_frcz_pd(ewrt);
318 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
320 twoeweps = _mm_add_pd(eweps,eweps);
321 ewitab = _mm_slli_epi32(ewitab,2);
322 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
323 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
324 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
325 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
326 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
327 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
328 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
329 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
330 velec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
331 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
333 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
335 /* Update potential sum for this i atom from the interaction with this j atom. */
336 velec = _mm_and_pd(velec,cutoff_mask);
337 velecsum = _mm_add_pd(velecsum,velec);
341 fscal = _mm_and_pd(fscal,cutoff_mask);
343 /* Update vectorial force */
344 fix1 = _mm_macc_pd(dx10,fscal,fix1);
345 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
346 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
348 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
349 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
350 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
354 /**************************
355 * CALCULATE INTERACTIONS *
356 **************************/
358 if (gmx_mm_any_lt(rsq20,rcutoff2))
361 r20 = _mm_mul_pd(rsq20,rinv20);
363 /* Compute parameters for interactions between i and j atoms */
364 qq20 = _mm_mul_pd(iq2,jq0);
366 /* EWALD ELECTROSTATICS */
368 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
369 ewrt = _mm_mul_pd(r20,ewtabscale);
370 ewitab = _mm_cvttpd_epi32(ewrt);
372 eweps = _mm_frcz_pd(ewrt);
374 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
376 twoeweps = _mm_add_pd(eweps,eweps);
377 ewitab = _mm_slli_epi32(ewitab,2);
378 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
379 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
380 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
381 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
382 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
383 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
384 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
385 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
386 velec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
387 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
389 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
391 /* Update potential sum for this i atom from the interaction with this j atom. */
392 velec = _mm_and_pd(velec,cutoff_mask);
393 velecsum = _mm_add_pd(velecsum,velec);
397 fscal = _mm_and_pd(fscal,cutoff_mask);
399 /* Update vectorial force */
400 fix2 = _mm_macc_pd(dx20,fscal,fix2);
401 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
402 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
404 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
405 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
406 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
410 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
412 /* Inner loop uses 168 flops */
419 j_coord_offsetA = DIM*jnrA;
421 /* load j atom coordinates */
422 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
425 /* Calculate displacement vector */
426 dx00 = _mm_sub_pd(ix0,jx0);
427 dy00 = _mm_sub_pd(iy0,jy0);
428 dz00 = _mm_sub_pd(iz0,jz0);
429 dx10 = _mm_sub_pd(ix1,jx0);
430 dy10 = _mm_sub_pd(iy1,jy0);
431 dz10 = _mm_sub_pd(iz1,jz0);
432 dx20 = _mm_sub_pd(ix2,jx0);
433 dy20 = _mm_sub_pd(iy2,jy0);
434 dz20 = _mm_sub_pd(iz2,jz0);
436 /* Calculate squared distance and things based on it */
437 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
438 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
439 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
441 rinv00 = avx128fma_invsqrt_d(rsq00);
442 rinv10 = avx128fma_invsqrt_d(rsq10);
443 rinv20 = avx128fma_invsqrt_d(rsq20);
445 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
446 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
447 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
449 /* Load parameters for j particles */
450 jq0 = _mm_load_sd(charge+jnrA+0);
451 vdwjidx0A = 2*vdwtype[jnrA+0];
453 fjx0 = _mm_setzero_pd();
454 fjy0 = _mm_setzero_pd();
455 fjz0 = _mm_setzero_pd();
457 /**************************
458 * CALCULATE INTERACTIONS *
459 **************************/
461 if (gmx_mm_any_lt(rsq00,rcutoff2))
464 r00 = _mm_mul_pd(rsq00,rinv00);
466 /* Compute parameters for interactions between i and j atoms */
467 qq00 = _mm_mul_pd(iq0,jq0);
468 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
470 /* EWALD ELECTROSTATICS */
472 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
473 ewrt = _mm_mul_pd(r00,ewtabscale);
474 ewitab = _mm_cvttpd_epi32(ewrt);
476 eweps = _mm_frcz_pd(ewrt);
478 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
480 twoeweps = _mm_add_pd(eweps,eweps);
481 ewitab = _mm_slli_epi32(ewitab,2);
482 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
483 ewtabD = _mm_setzero_pd();
484 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
485 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
486 ewtabFn = _mm_setzero_pd();
487 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
488 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
489 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
490 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
491 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
493 /* LENNARD-JONES DISPERSION/REPULSION */
495 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
496 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
497 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
498 vvdw = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
499 _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
500 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
502 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
504 /* Update potential sum for this i atom from the interaction with this j atom. */
505 velec = _mm_and_pd(velec,cutoff_mask);
506 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
507 velecsum = _mm_add_pd(velecsum,velec);
508 vvdw = _mm_and_pd(vvdw,cutoff_mask);
509 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
510 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
512 fscal = _mm_add_pd(felec,fvdw);
514 fscal = _mm_and_pd(fscal,cutoff_mask);
516 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
518 /* Update vectorial force */
519 fix0 = _mm_macc_pd(dx00,fscal,fix0);
520 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
521 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
523 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
524 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
525 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
529 /**************************
530 * CALCULATE INTERACTIONS *
531 **************************/
533 if (gmx_mm_any_lt(rsq10,rcutoff2))
536 r10 = _mm_mul_pd(rsq10,rinv10);
538 /* Compute parameters for interactions between i and j atoms */
539 qq10 = _mm_mul_pd(iq1,jq0);
541 /* EWALD ELECTROSTATICS */
543 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
544 ewrt = _mm_mul_pd(r10,ewtabscale);
545 ewitab = _mm_cvttpd_epi32(ewrt);
547 eweps = _mm_frcz_pd(ewrt);
549 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
551 twoeweps = _mm_add_pd(eweps,eweps);
552 ewitab = _mm_slli_epi32(ewitab,2);
553 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
554 ewtabD = _mm_setzero_pd();
555 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
556 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
557 ewtabFn = _mm_setzero_pd();
558 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
559 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
560 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
561 velec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
562 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
564 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
566 /* Update potential sum for this i atom from the interaction with this j atom. */
567 velec = _mm_and_pd(velec,cutoff_mask);
568 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
569 velecsum = _mm_add_pd(velecsum,velec);
573 fscal = _mm_and_pd(fscal,cutoff_mask);
575 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
577 /* Update vectorial force */
578 fix1 = _mm_macc_pd(dx10,fscal,fix1);
579 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
580 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
582 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
583 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
584 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
588 /**************************
589 * CALCULATE INTERACTIONS *
590 **************************/
592 if (gmx_mm_any_lt(rsq20,rcutoff2))
595 r20 = _mm_mul_pd(rsq20,rinv20);
597 /* Compute parameters for interactions between i and j atoms */
598 qq20 = _mm_mul_pd(iq2,jq0);
600 /* EWALD ELECTROSTATICS */
602 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
603 ewrt = _mm_mul_pd(r20,ewtabscale);
604 ewitab = _mm_cvttpd_epi32(ewrt);
606 eweps = _mm_frcz_pd(ewrt);
608 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
610 twoeweps = _mm_add_pd(eweps,eweps);
611 ewitab = _mm_slli_epi32(ewitab,2);
612 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
613 ewtabD = _mm_setzero_pd();
614 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
615 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
616 ewtabFn = _mm_setzero_pd();
617 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
618 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
619 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
620 velec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
621 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
623 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
625 /* Update potential sum for this i atom from the interaction with this j atom. */
626 velec = _mm_and_pd(velec,cutoff_mask);
627 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
628 velecsum = _mm_add_pd(velecsum,velec);
632 fscal = _mm_and_pd(fscal,cutoff_mask);
634 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
636 /* Update vectorial force */
637 fix2 = _mm_macc_pd(dx20,fscal,fix2);
638 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
639 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
641 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
642 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
643 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
647 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
649 /* Inner loop uses 168 flops */
652 /* End of innermost loop */
654 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
655 f+i_coord_offset,fshift+i_shift_offset);
658 /* Update potential energies */
659 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
660 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
662 /* Increment number of inner iterations */
663 inneriter += j_index_end - j_index_start;
665 /* Outer loop uses 20 flops */
668 /* Increment number of outer iterations */
671 /* Update outer/inner flops */
673 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*168);
676 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_avx_128_fma_double
677 * Electrostatics interaction: Ewald
678 * VdW interaction: LennardJones
679 * Geometry: Water3-Particle
680 * Calculate force/pot: Force
683 nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_avx_128_fma_double
684 (t_nblist * gmx_restrict nlist,
685 rvec * gmx_restrict xx,
686 rvec * gmx_restrict ff,
687 struct t_forcerec * gmx_restrict fr,
688 t_mdatoms * gmx_restrict mdatoms,
689 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
690 t_nrnb * gmx_restrict nrnb)
692 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
693 * just 0 for non-waters.
694 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
695 * jnr indices corresponding to data put in the four positions in the SIMD register.
697 int i_shift_offset,i_coord_offset,outeriter,inneriter;
698 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
700 int j_coord_offsetA,j_coord_offsetB;
701 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
703 real *shiftvec,*fshift,*x,*f;
704 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
706 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
708 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
710 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
711 int vdwjidx0A,vdwjidx0B;
712 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
713 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
714 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
715 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
716 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
719 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
722 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
723 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
725 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
727 __m128d dummy_mask,cutoff_mask;
728 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
729 __m128d one = _mm_set1_pd(1.0);
730 __m128d two = _mm_set1_pd(2.0);
736 jindex = nlist->jindex;
738 shiftidx = nlist->shift;
740 shiftvec = fr->shift_vec[0];
741 fshift = fr->fshift[0];
742 facel = _mm_set1_pd(fr->ic->epsfac);
743 charge = mdatoms->chargeA;
744 nvdwtype = fr->ntype;
746 vdwtype = mdatoms->typeA;
748 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
749 ewtab = fr->ic->tabq_coul_F;
750 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
751 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
753 /* Setup water-specific parameters */
754 inr = nlist->iinr[0];
755 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
756 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
757 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
758 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
760 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
761 rcutoff_scalar = fr->ic->rcoulomb;
762 rcutoff = _mm_set1_pd(rcutoff_scalar);
763 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
765 sh_vdw_invrcut6 = _mm_set1_pd(fr->ic->sh_invrc6);
766 rvdw = _mm_set1_pd(fr->ic->rvdw);
768 /* Avoid stupid compiler warnings */
776 /* Start outer loop over neighborlists */
777 for(iidx=0; iidx<nri; iidx++)
779 /* Load shift vector for this list */
780 i_shift_offset = DIM*shiftidx[iidx];
782 /* Load limits for loop over neighbors */
783 j_index_start = jindex[iidx];
784 j_index_end = jindex[iidx+1];
786 /* Get outer coordinate index */
788 i_coord_offset = DIM*inr;
790 /* Load i particle coords and add shift vector */
791 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
792 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
794 fix0 = _mm_setzero_pd();
795 fiy0 = _mm_setzero_pd();
796 fiz0 = _mm_setzero_pd();
797 fix1 = _mm_setzero_pd();
798 fiy1 = _mm_setzero_pd();
799 fiz1 = _mm_setzero_pd();
800 fix2 = _mm_setzero_pd();
801 fiy2 = _mm_setzero_pd();
802 fiz2 = _mm_setzero_pd();
804 /* Start inner kernel loop */
805 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
808 /* Get j neighbor index, and coordinate index */
811 j_coord_offsetA = DIM*jnrA;
812 j_coord_offsetB = DIM*jnrB;
814 /* load j atom coordinates */
815 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
818 /* Calculate displacement vector */
819 dx00 = _mm_sub_pd(ix0,jx0);
820 dy00 = _mm_sub_pd(iy0,jy0);
821 dz00 = _mm_sub_pd(iz0,jz0);
822 dx10 = _mm_sub_pd(ix1,jx0);
823 dy10 = _mm_sub_pd(iy1,jy0);
824 dz10 = _mm_sub_pd(iz1,jz0);
825 dx20 = _mm_sub_pd(ix2,jx0);
826 dy20 = _mm_sub_pd(iy2,jy0);
827 dz20 = _mm_sub_pd(iz2,jz0);
829 /* Calculate squared distance and things based on it */
830 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
831 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
832 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
834 rinv00 = avx128fma_invsqrt_d(rsq00);
835 rinv10 = avx128fma_invsqrt_d(rsq10);
836 rinv20 = avx128fma_invsqrt_d(rsq20);
838 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
839 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
840 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
842 /* Load parameters for j particles */
843 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
844 vdwjidx0A = 2*vdwtype[jnrA+0];
845 vdwjidx0B = 2*vdwtype[jnrB+0];
847 fjx0 = _mm_setzero_pd();
848 fjy0 = _mm_setzero_pd();
849 fjz0 = _mm_setzero_pd();
851 /**************************
852 * CALCULATE INTERACTIONS *
853 **************************/
855 if (gmx_mm_any_lt(rsq00,rcutoff2))
858 r00 = _mm_mul_pd(rsq00,rinv00);
860 /* Compute parameters for interactions between i and j atoms */
861 qq00 = _mm_mul_pd(iq0,jq0);
862 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
863 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
865 /* EWALD ELECTROSTATICS */
867 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
868 ewrt = _mm_mul_pd(r00,ewtabscale);
869 ewitab = _mm_cvttpd_epi32(ewrt);
871 eweps = _mm_frcz_pd(ewrt);
873 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
875 twoeweps = _mm_add_pd(eweps,eweps);
876 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
878 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
879 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
881 /* LENNARD-JONES DISPERSION/REPULSION */
883 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
884 fvdw = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
886 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
888 fscal = _mm_add_pd(felec,fvdw);
890 fscal = _mm_and_pd(fscal,cutoff_mask);
892 /* Update vectorial force */
893 fix0 = _mm_macc_pd(dx00,fscal,fix0);
894 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
895 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
897 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
898 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
899 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
903 /**************************
904 * CALCULATE INTERACTIONS *
905 **************************/
907 if (gmx_mm_any_lt(rsq10,rcutoff2))
910 r10 = _mm_mul_pd(rsq10,rinv10);
912 /* Compute parameters for interactions between i and j atoms */
913 qq10 = _mm_mul_pd(iq1,jq0);
915 /* EWALD ELECTROSTATICS */
917 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
918 ewrt = _mm_mul_pd(r10,ewtabscale);
919 ewitab = _mm_cvttpd_epi32(ewrt);
921 eweps = _mm_frcz_pd(ewrt);
923 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
925 twoeweps = _mm_add_pd(eweps,eweps);
926 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
928 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
929 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
931 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
935 fscal = _mm_and_pd(fscal,cutoff_mask);
937 /* Update vectorial force */
938 fix1 = _mm_macc_pd(dx10,fscal,fix1);
939 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
940 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
942 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
943 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
944 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
948 /**************************
949 * CALCULATE INTERACTIONS *
950 **************************/
952 if (gmx_mm_any_lt(rsq20,rcutoff2))
955 r20 = _mm_mul_pd(rsq20,rinv20);
957 /* Compute parameters for interactions between i and j atoms */
958 qq20 = _mm_mul_pd(iq2,jq0);
960 /* EWALD ELECTROSTATICS */
962 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
963 ewrt = _mm_mul_pd(r20,ewtabscale);
964 ewitab = _mm_cvttpd_epi32(ewrt);
966 eweps = _mm_frcz_pd(ewrt);
968 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
970 twoeweps = _mm_add_pd(eweps,eweps);
971 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
973 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
974 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
976 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
980 fscal = _mm_and_pd(fscal,cutoff_mask);
982 /* Update vectorial force */
983 fix2 = _mm_macc_pd(dx20,fscal,fix2);
984 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
985 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
987 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
988 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
989 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
993 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
995 /* Inner loop uses 136 flops */
1002 j_coord_offsetA = DIM*jnrA;
1004 /* load j atom coordinates */
1005 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1008 /* Calculate displacement vector */
1009 dx00 = _mm_sub_pd(ix0,jx0);
1010 dy00 = _mm_sub_pd(iy0,jy0);
1011 dz00 = _mm_sub_pd(iz0,jz0);
1012 dx10 = _mm_sub_pd(ix1,jx0);
1013 dy10 = _mm_sub_pd(iy1,jy0);
1014 dz10 = _mm_sub_pd(iz1,jz0);
1015 dx20 = _mm_sub_pd(ix2,jx0);
1016 dy20 = _mm_sub_pd(iy2,jy0);
1017 dz20 = _mm_sub_pd(iz2,jz0);
1019 /* Calculate squared distance and things based on it */
1020 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1021 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1022 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1024 rinv00 = avx128fma_invsqrt_d(rsq00);
1025 rinv10 = avx128fma_invsqrt_d(rsq10);
1026 rinv20 = avx128fma_invsqrt_d(rsq20);
1028 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
1029 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
1030 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
1032 /* Load parameters for j particles */
1033 jq0 = _mm_load_sd(charge+jnrA+0);
1034 vdwjidx0A = 2*vdwtype[jnrA+0];
1036 fjx0 = _mm_setzero_pd();
1037 fjy0 = _mm_setzero_pd();
1038 fjz0 = _mm_setzero_pd();
1040 /**************************
1041 * CALCULATE INTERACTIONS *
1042 **************************/
1044 if (gmx_mm_any_lt(rsq00,rcutoff2))
1047 r00 = _mm_mul_pd(rsq00,rinv00);
1049 /* Compute parameters for interactions between i and j atoms */
1050 qq00 = _mm_mul_pd(iq0,jq0);
1051 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1053 /* EWALD ELECTROSTATICS */
1055 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1056 ewrt = _mm_mul_pd(r00,ewtabscale);
1057 ewitab = _mm_cvttpd_epi32(ewrt);
1059 eweps = _mm_frcz_pd(ewrt);
1061 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1063 twoeweps = _mm_add_pd(eweps,eweps);
1064 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1065 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1066 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1068 /* LENNARD-JONES DISPERSION/REPULSION */
1070 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1071 fvdw = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1073 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
1075 fscal = _mm_add_pd(felec,fvdw);
1077 fscal = _mm_and_pd(fscal,cutoff_mask);
1079 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1081 /* Update vectorial force */
1082 fix0 = _mm_macc_pd(dx00,fscal,fix0);
1083 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
1084 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
1086 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
1087 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
1088 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
1092 /**************************
1093 * CALCULATE INTERACTIONS *
1094 **************************/
1096 if (gmx_mm_any_lt(rsq10,rcutoff2))
1099 r10 = _mm_mul_pd(rsq10,rinv10);
1101 /* Compute parameters for interactions between i and j atoms */
1102 qq10 = _mm_mul_pd(iq1,jq0);
1104 /* EWALD ELECTROSTATICS */
1106 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1107 ewrt = _mm_mul_pd(r10,ewtabscale);
1108 ewitab = _mm_cvttpd_epi32(ewrt);
1110 eweps = _mm_frcz_pd(ewrt);
1112 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1114 twoeweps = _mm_add_pd(eweps,eweps);
1115 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1116 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1117 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1119 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
1123 fscal = _mm_and_pd(fscal,cutoff_mask);
1125 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1127 /* Update vectorial force */
1128 fix1 = _mm_macc_pd(dx10,fscal,fix1);
1129 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
1130 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
1132 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
1133 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
1134 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
1138 /**************************
1139 * CALCULATE INTERACTIONS *
1140 **************************/
1142 if (gmx_mm_any_lt(rsq20,rcutoff2))
1145 r20 = _mm_mul_pd(rsq20,rinv20);
1147 /* Compute parameters for interactions between i and j atoms */
1148 qq20 = _mm_mul_pd(iq2,jq0);
1150 /* EWALD ELECTROSTATICS */
1152 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1153 ewrt = _mm_mul_pd(r20,ewtabscale);
1154 ewitab = _mm_cvttpd_epi32(ewrt);
1156 eweps = _mm_frcz_pd(ewrt);
1158 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1160 twoeweps = _mm_add_pd(eweps,eweps);
1161 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1162 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1163 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1165 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
1169 fscal = _mm_and_pd(fscal,cutoff_mask);
1171 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1173 /* Update vectorial force */
1174 fix2 = _mm_macc_pd(dx20,fscal,fix2);
1175 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
1176 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
1178 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
1179 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
1180 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
1184 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1186 /* Inner loop uses 136 flops */
1189 /* End of innermost loop */
1191 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1192 f+i_coord_offset,fshift+i_shift_offset);
1194 /* Increment number of inner iterations */
1195 inneriter += j_index_end - j_index_start;
1197 /* Outer loop uses 18 flops */
1200 /* Increment number of outer iterations */
1203 /* Update outer/inner flops */
1205 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*136);