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_ElecEwSh_VdwLJSh_GeomW3P1_VF_avx_128_fma_double
54 * Electrostatics interaction: Ewald
55 * VdW interaction: LennardJones
56 * Geometry: Water3-Particle
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecEwSh_VdwLJSh_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 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
104 __m128d dummy_mask,cutoff_mask;
105 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
106 __m128d one = _mm_set1_pd(1.0);
107 __m128d two = _mm_set1_pd(2.0);
113 jindex = nlist->jindex;
115 shiftidx = nlist->shift;
117 shiftvec = fr->shift_vec[0];
118 fshift = fr->fshift[0];
119 facel = _mm_set1_pd(fr->epsfac);
120 charge = mdatoms->chargeA;
121 nvdwtype = fr->ntype;
123 vdwtype = mdatoms->typeA;
125 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
126 ewtab = fr->ic->tabq_coul_FDV0;
127 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
128 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
130 /* Setup water-specific parameters */
131 inr = nlist->iinr[0];
132 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
133 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
134 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
135 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
137 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
138 rcutoff_scalar = fr->rcoulomb;
139 rcutoff = _mm_set1_pd(rcutoff_scalar);
140 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
142 sh_vdw_invrcut6 = _mm_set1_pd(fr->ic->sh_invrc6);
143 rvdw = _mm_set1_pd(fr->rvdw);
145 /* Avoid stupid compiler warnings */
153 /* Start outer loop over neighborlists */
154 for(iidx=0; iidx<nri; iidx++)
156 /* Load shift vector for this list */
157 i_shift_offset = DIM*shiftidx[iidx];
159 /* Load limits for loop over neighbors */
160 j_index_start = jindex[iidx];
161 j_index_end = jindex[iidx+1];
163 /* Get outer coordinate index */
165 i_coord_offset = DIM*inr;
167 /* Load i particle coords and add shift vector */
168 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
169 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
171 fix0 = _mm_setzero_pd();
172 fiy0 = _mm_setzero_pd();
173 fiz0 = _mm_setzero_pd();
174 fix1 = _mm_setzero_pd();
175 fiy1 = _mm_setzero_pd();
176 fiz1 = _mm_setzero_pd();
177 fix2 = _mm_setzero_pd();
178 fiy2 = _mm_setzero_pd();
179 fiz2 = _mm_setzero_pd();
181 /* Reset potential sums */
182 velecsum = _mm_setzero_pd();
183 vvdwsum = _mm_setzero_pd();
185 /* Start inner kernel loop */
186 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
189 /* Get j neighbor index, and coordinate index */
192 j_coord_offsetA = DIM*jnrA;
193 j_coord_offsetB = DIM*jnrB;
195 /* load j atom coordinates */
196 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
199 /* Calculate displacement vector */
200 dx00 = _mm_sub_pd(ix0,jx0);
201 dy00 = _mm_sub_pd(iy0,jy0);
202 dz00 = _mm_sub_pd(iz0,jz0);
203 dx10 = _mm_sub_pd(ix1,jx0);
204 dy10 = _mm_sub_pd(iy1,jy0);
205 dz10 = _mm_sub_pd(iz1,jz0);
206 dx20 = _mm_sub_pd(ix2,jx0);
207 dy20 = _mm_sub_pd(iy2,jy0);
208 dz20 = _mm_sub_pd(iz2,jz0);
210 /* Calculate squared distance and things based on it */
211 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
212 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
213 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
215 rinv00 = gmx_mm_invsqrt_pd(rsq00);
216 rinv10 = gmx_mm_invsqrt_pd(rsq10);
217 rinv20 = gmx_mm_invsqrt_pd(rsq20);
219 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
220 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
221 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
223 /* Load parameters for j particles */
224 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
225 vdwjidx0A = 2*vdwtype[jnrA+0];
226 vdwjidx0B = 2*vdwtype[jnrB+0];
228 fjx0 = _mm_setzero_pd();
229 fjy0 = _mm_setzero_pd();
230 fjz0 = _mm_setzero_pd();
232 /**************************
233 * CALCULATE INTERACTIONS *
234 **************************/
236 if (gmx_mm_any_lt(rsq00,rcutoff2))
239 r00 = _mm_mul_pd(rsq00,rinv00);
241 /* Compute parameters for interactions between i and j atoms */
242 qq00 = _mm_mul_pd(iq0,jq0);
243 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
244 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
246 /* EWALD ELECTROSTATICS */
248 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
249 ewrt = _mm_mul_pd(r00,ewtabscale);
250 ewitab = _mm_cvttpd_epi32(ewrt);
252 eweps = _mm_frcz_pd(ewrt);
254 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
256 twoeweps = _mm_add_pd(eweps,eweps);
257 ewitab = _mm_slli_epi32(ewitab,2);
258 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
259 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
260 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
261 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
262 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
263 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
264 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
265 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
266 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
267 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
269 /* LENNARD-JONES DISPERSION/REPULSION */
271 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
272 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
273 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
274 vvdw = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
275 _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
276 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
278 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
280 /* Update potential sum for this i atom from the interaction with this j atom. */
281 velec = _mm_and_pd(velec,cutoff_mask);
282 velecsum = _mm_add_pd(velecsum,velec);
283 vvdw = _mm_and_pd(vvdw,cutoff_mask);
284 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
286 fscal = _mm_add_pd(felec,fvdw);
288 fscal = _mm_and_pd(fscal,cutoff_mask);
290 /* Update vectorial force */
291 fix0 = _mm_macc_pd(dx00,fscal,fix0);
292 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
293 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
295 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
296 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
297 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
301 /**************************
302 * CALCULATE INTERACTIONS *
303 **************************/
305 if (gmx_mm_any_lt(rsq10,rcutoff2))
308 r10 = _mm_mul_pd(rsq10,rinv10);
310 /* Compute parameters for interactions between i and j atoms */
311 qq10 = _mm_mul_pd(iq1,jq0);
313 /* EWALD ELECTROSTATICS */
315 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
316 ewrt = _mm_mul_pd(r10,ewtabscale);
317 ewitab = _mm_cvttpd_epi32(ewrt);
319 eweps = _mm_frcz_pd(ewrt);
321 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
323 twoeweps = _mm_add_pd(eweps,eweps);
324 ewitab = _mm_slli_epi32(ewitab,2);
325 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
326 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
327 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
328 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
329 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
330 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
331 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
332 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
333 velec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
334 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
336 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
338 /* Update potential sum for this i atom from the interaction with this j atom. */
339 velec = _mm_and_pd(velec,cutoff_mask);
340 velecsum = _mm_add_pd(velecsum,velec);
344 fscal = _mm_and_pd(fscal,cutoff_mask);
346 /* Update vectorial force */
347 fix1 = _mm_macc_pd(dx10,fscal,fix1);
348 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
349 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
351 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
352 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
353 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
357 /**************************
358 * CALCULATE INTERACTIONS *
359 **************************/
361 if (gmx_mm_any_lt(rsq20,rcutoff2))
364 r20 = _mm_mul_pd(rsq20,rinv20);
366 /* Compute parameters for interactions between i and j atoms */
367 qq20 = _mm_mul_pd(iq2,jq0);
369 /* EWALD ELECTROSTATICS */
371 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
372 ewrt = _mm_mul_pd(r20,ewtabscale);
373 ewitab = _mm_cvttpd_epi32(ewrt);
375 eweps = _mm_frcz_pd(ewrt);
377 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
379 twoeweps = _mm_add_pd(eweps,eweps);
380 ewitab = _mm_slli_epi32(ewitab,2);
381 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
382 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
383 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
384 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
385 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
386 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
387 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
388 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
389 velec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
390 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
392 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
394 /* Update potential sum for this i atom from the interaction with this j atom. */
395 velec = _mm_and_pd(velec,cutoff_mask);
396 velecsum = _mm_add_pd(velecsum,velec);
400 fscal = _mm_and_pd(fscal,cutoff_mask);
402 /* Update vectorial force */
403 fix2 = _mm_macc_pd(dx20,fscal,fix2);
404 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
405 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
407 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
408 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
409 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
413 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
415 /* Inner loop uses 168 flops */
422 j_coord_offsetA = DIM*jnrA;
424 /* load j atom coordinates */
425 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
428 /* Calculate displacement vector */
429 dx00 = _mm_sub_pd(ix0,jx0);
430 dy00 = _mm_sub_pd(iy0,jy0);
431 dz00 = _mm_sub_pd(iz0,jz0);
432 dx10 = _mm_sub_pd(ix1,jx0);
433 dy10 = _mm_sub_pd(iy1,jy0);
434 dz10 = _mm_sub_pd(iz1,jz0);
435 dx20 = _mm_sub_pd(ix2,jx0);
436 dy20 = _mm_sub_pd(iy2,jy0);
437 dz20 = _mm_sub_pd(iz2,jz0);
439 /* Calculate squared distance and things based on it */
440 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
441 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
442 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
444 rinv00 = gmx_mm_invsqrt_pd(rsq00);
445 rinv10 = gmx_mm_invsqrt_pd(rsq10);
446 rinv20 = gmx_mm_invsqrt_pd(rsq20);
448 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
449 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
450 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
452 /* Load parameters for j particles */
453 jq0 = _mm_load_sd(charge+jnrA+0);
454 vdwjidx0A = 2*vdwtype[jnrA+0];
456 fjx0 = _mm_setzero_pd();
457 fjy0 = _mm_setzero_pd();
458 fjz0 = _mm_setzero_pd();
460 /**************************
461 * CALCULATE INTERACTIONS *
462 **************************/
464 if (gmx_mm_any_lt(rsq00,rcutoff2))
467 r00 = _mm_mul_pd(rsq00,rinv00);
469 /* Compute parameters for interactions between i and j atoms */
470 qq00 = _mm_mul_pd(iq0,jq0);
471 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
473 /* EWALD ELECTROSTATICS */
475 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
476 ewrt = _mm_mul_pd(r00,ewtabscale);
477 ewitab = _mm_cvttpd_epi32(ewrt);
479 eweps = _mm_frcz_pd(ewrt);
481 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
483 twoeweps = _mm_add_pd(eweps,eweps);
484 ewitab = _mm_slli_epi32(ewitab,2);
485 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
486 ewtabD = _mm_setzero_pd();
487 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
488 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
489 ewtabFn = _mm_setzero_pd();
490 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
491 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
492 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
493 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
494 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
496 /* LENNARD-JONES DISPERSION/REPULSION */
498 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
499 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
500 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
501 vvdw = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
502 _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
503 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
505 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
507 /* Update potential sum for this i atom from the interaction with this j atom. */
508 velec = _mm_and_pd(velec,cutoff_mask);
509 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
510 velecsum = _mm_add_pd(velecsum,velec);
511 vvdw = _mm_and_pd(vvdw,cutoff_mask);
512 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
513 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
515 fscal = _mm_add_pd(felec,fvdw);
517 fscal = _mm_and_pd(fscal,cutoff_mask);
519 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
521 /* Update vectorial force */
522 fix0 = _mm_macc_pd(dx00,fscal,fix0);
523 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
524 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
526 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
527 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
528 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
532 /**************************
533 * CALCULATE INTERACTIONS *
534 **************************/
536 if (gmx_mm_any_lt(rsq10,rcutoff2))
539 r10 = _mm_mul_pd(rsq10,rinv10);
541 /* Compute parameters for interactions between i and j atoms */
542 qq10 = _mm_mul_pd(iq1,jq0);
544 /* EWALD ELECTROSTATICS */
546 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
547 ewrt = _mm_mul_pd(r10,ewtabscale);
548 ewitab = _mm_cvttpd_epi32(ewrt);
550 eweps = _mm_frcz_pd(ewrt);
552 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
554 twoeweps = _mm_add_pd(eweps,eweps);
555 ewitab = _mm_slli_epi32(ewitab,2);
556 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
557 ewtabD = _mm_setzero_pd();
558 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
559 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
560 ewtabFn = _mm_setzero_pd();
561 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
562 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
563 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
564 velec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
565 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
567 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
569 /* Update potential sum for this i atom from the interaction with this j atom. */
570 velec = _mm_and_pd(velec,cutoff_mask);
571 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
572 velecsum = _mm_add_pd(velecsum,velec);
576 fscal = _mm_and_pd(fscal,cutoff_mask);
578 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
580 /* Update vectorial force */
581 fix1 = _mm_macc_pd(dx10,fscal,fix1);
582 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
583 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
585 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
586 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
587 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
591 /**************************
592 * CALCULATE INTERACTIONS *
593 **************************/
595 if (gmx_mm_any_lt(rsq20,rcutoff2))
598 r20 = _mm_mul_pd(rsq20,rinv20);
600 /* Compute parameters for interactions between i and j atoms */
601 qq20 = _mm_mul_pd(iq2,jq0);
603 /* EWALD ELECTROSTATICS */
605 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
606 ewrt = _mm_mul_pd(r20,ewtabscale);
607 ewitab = _mm_cvttpd_epi32(ewrt);
609 eweps = _mm_frcz_pd(ewrt);
611 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
613 twoeweps = _mm_add_pd(eweps,eweps);
614 ewitab = _mm_slli_epi32(ewitab,2);
615 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
616 ewtabD = _mm_setzero_pd();
617 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
618 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
619 ewtabFn = _mm_setzero_pd();
620 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
621 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
622 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
623 velec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
624 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
626 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
628 /* Update potential sum for this i atom from the interaction with this j atom. */
629 velec = _mm_and_pd(velec,cutoff_mask);
630 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
631 velecsum = _mm_add_pd(velecsum,velec);
635 fscal = _mm_and_pd(fscal,cutoff_mask);
637 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
639 /* Update vectorial force */
640 fix2 = _mm_macc_pd(dx20,fscal,fix2);
641 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
642 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
644 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
645 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
646 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
650 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
652 /* Inner loop uses 168 flops */
655 /* End of innermost loop */
657 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
658 f+i_coord_offset,fshift+i_shift_offset);
661 /* Update potential energies */
662 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
663 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
665 /* Increment number of inner iterations */
666 inneriter += j_index_end - j_index_start;
668 /* Outer loop uses 20 flops */
671 /* Increment number of outer iterations */
674 /* Update outer/inner flops */
676 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*168);
679 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_avx_128_fma_double
680 * Electrostatics interaction: Ewald
681 * VdW interaction: LennardJones
682 * Geometry: Water3-Particle
683 * Calculate force/pot: Force
686 nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_avx_128_fma_double
687 (t_nblist * gmx_restrict nlist,
688 rvec * gmx_restrict xx,
689 rvec * gmx_restrict ff,
690 t_forcerec * gmx_restrict fr,
691 t_mdatoms * gmx_restrict mdatoms,
692 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
693 t_nrnb * gmx_restrict nrnb)
695 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
696 * just 0 for non-waters.
697 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
698 * jnr indices corresponding to data put in the four positions in the SIMD register.
700 int i_shift_offset,i_coord_offset,outeriter,inneriter;
701 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
703 int j_coord_offsetA,j_coord_offsetB;
704 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
706 real *shiftvec,*fshift,*x,*f;
707 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
709 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
711 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
713 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
714 int vdwjidx0A,vdwjidx0B;
715 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
716 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
717 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
718 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
719 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
722 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
725 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
726 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
728 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
730 __m128d dummy_mask,cutoff_mask;
731 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
732 __m128d one = _mm_set1_pd(1.0);
733 __m128d two = _mm_set1_pd(2.0);
739 jindex = nlist->jindex;
741 shiftidx = nlist->shift;
743 shiftvec = fr->shift_vec[0];
744 fshift = fr->fshift[0];
745 facel = _mm_set1_pd(fr->epsfac);
746 charge = mdatoms->chargeA;
747 nvdwtype = fr->ntype;
749 vdwtype = mdatoms->typeA;
751 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
752 ewtab = fr->ic->tabq_coul_F;
753 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
754 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
756 /* Setup water-specific parameters */
757 inr = nlist->iinr[0];
758 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
759 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
760 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
761 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
763 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
764 rcutoff_scalar = fr->rcoulomb;
765 rcutoff = _mm_set1_pd(rcutoff_scalar);
766 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
768 sh_vdw_invrcut6 = _mm_set1_pd(fr->ic->sh_invrc6);
769 rvdw = _mm_set1_pd(fr->rvdw);
771 /* Avoid stupid compiler warnings */
779 /* Start outer loop over neighborlists */
780 for(iidx=0; iidx<nri; iidx++)
782 /* Load shift vector for this list */
783 i_shift_offset = DIM*shiftidx[iidx];
785 /* Load limits for loop over neighbors */
786 j_index_start = jindex[iidx];
787 j_index_end = jindex[iidx+1];
789 /* Get outer coordinate index */
791 i_coord_offset = DIM*inr;
793 /* Load i particle coords and add shift vector */
794 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
795 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
797 fix0 = _mm_setzero_pd();
798 fiy0 = _mm_setzero_pd();
799 fiz0 = _mm_setzero_pd();
800 fix1 = _mm_setzero_pd();
801 fiy1 = _mm_setzero_pd();
802 fiz1 = _mm_setzero_pd();
803 fix2 = _mm_setzero_pd();
804 fiy2 = _mm_setzero_pd();
805 fiz2 = _mm_setzero_pd();
807 /* Start inner kernel loop */
808 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
811 /* Get j neighbor index, and coordinate index */
814 j_coord_offsetA = DIM*jnrA;
815 j_coord_offsetB = DIM*jnrB;
817 /* load j atom coordinates */
818 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
821 /* Calculate displacement vector */
822 dx00 = _mm_sub_pd(ix0,jx0);
823 dy00 = _mm_sub_pd(iy0,jy0);
824 dz00 = _mm_sub_pd(iz0,jz0);
825 dx10 = _mm_sub_pd(ix1,jx0);
826 dy10 = _mm_sub_pd(iy1,jy0);
827 dz10 = _mm_sub_pd(iz1,jz0);
828 dx20 = _mm_sub_pd(ix2,jx0);
829 dy20 = _mm_sub_pd(iy2,jy0);
830 dz20 = _mm_sub_pd(iz2,jz0);
832 /* Calculate squared distance and things based on it */
833 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
834 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
835 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
837 rinv00 = gmx_mm_invsqrt_pd(rsq00);
838 rinv10 = gmx_mm_invsqrt_pd(rsq10);
839 rinv20 = gmx_mm_invsqrt_pd(rsq20);
841 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
842 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
843 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
845 /* Load parameters for j particles */
846 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
847 vdwjidx0A = 2*vdwtype[jnrA+0];
848 vdwjidx0B = 2*vdwtype[jnrB+0];
850 fjx0 = _mm_setzero_pd();
851 fjy0 = _mm_setzero_pd();
852 fjz0 = _mm_setzero_pd();
854 /**************************
855 * CALCULATE INTERACTIONS *
856 **************************/
858 if (gmx_mm_any_lt(rsq00,rcutoff2))
861 r00 = _mm_mul_pd(rsq00,rinv00);
863 /* Compute parameters for interactions between i and j atoms */
864 qq00 = _mm_mul_pd(iq0,jq0);
865 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
866 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
868 /* EWALD ELECTROSTATICS */
870 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
871 ewrt = _mm_mul_pd(r00,ewtabscale);
872 ewitab = _mm_cvttpd_epi32(ewrt);
874 eweps = _mm_frcz_pd(ewrt);
876 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
878 twoeweps = _mm_add_pd(eweps,eweps);
879 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
881 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
882 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
884 /* LENNARD-JONES DISPERSION/REPULSION */
886 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
887 fvdw = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
889 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
891 fscal = _mm_add_pd(felec,fvdw);
893 fscal = _mm_and_pd(fscal,cutoff_mask);
895 /* Update vectorial force */
896 fix0 = _mm_macc_pd(dx00,fscal,fix0);
897 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
898 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
900 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
901 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
902 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
906 /**************************
907 * CALCULATE INTERACTIONS *
908 **************************/
910 if (gmx_mm_any_lt(rsq10,rcutoff2))
913 r10 = _mm_mul_pd(rsq10,rinv10);
915 /* Compute parameters for interactions between i and j atoms */
916 qq10 = _mm_mul_pd(iq1,jq0);
918 /* EWALD ELECTROSTATICS */
920 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
921 ewrt = _mm_mul_pd(r10,ewtabscale);
922 ewitab = _mm_cvttpd_epi32(ewrt);
924 eweps = _mm_frcz_pd(ewrt);
926 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
928 twoeweps = _mm_add_pd(eweps,eweps);
929 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
931 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
932 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
934 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
938 fscal = _mm_and_pd(fscal,cutoff_mask);
940 /* Update vectorial force */
941 fix1 = _mm_macc_pd(dx10,fscal,fix1);
942 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
943 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
945 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
946 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
947 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
951 /**************************
952 * CALCULATE INTERACTIONS *
953 **************************/
955 if (gmx_mm_any_lt(rsq20,rcutoff2))
958 r20 = _mm_mul_pd(rsq20,rinv20);
960 /* Compute parameters for interactions between i and j atoms */
961 qq20 = _mm_mul_pd(iq2,jq0);
963 /* EWALD ELECTROSTATICS */
965 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
966 ewrt = _mm_mul_pd(r20,ewtabscale);
967 ewitab = _mm_cvttpd_epi32(ewrt);
969 eweps = _mm_frcz_pd(ewrt);
971 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
973 twoeweps = _mm_add_pd(eweps,eweps);
974 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
976 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
977 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
979 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
983 fscal = _mm_and_pd(fscal,cutoff_mask);
985 /* Update vectorial force */
986 fix2 = _mm_macc_pd(dx20,fscal,fix2);
987 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
988 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
990 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
991 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
992 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
996 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
998 /* Inner loop uses 136 flops */
1001 if(jidx<j_index_end)
1005 j_coord_offsetA = DIM*jnrA;
1007 /* load j atom coordinates */
1008 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1011 /* Calculate displacement vector */
1012 dx00 = _mm_sub_pd(ix0,jx0);
1013 dy00 = _mm_sub_pd(iy0,jy0);
1014 dz00 = _mm_sub_pd(iz0,jz0);
1015 dx10 = _mm_sub_pd(ix1,jx0);
1016 dy10 = _mm_sub_pd(iy1,jy0);
1017 dz10 = _mm_sub_pd(iz1,jz0);
1018 dx20 = _mm_sub_pd(ix2,jx0);
1019 dy20 = _mm_sub_pd(iy2,jy0);
1020 dz20 = _mm_sub_pd(iz2,jz0);
1022 /* Calculate squared distance and things based on it */
1023 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1024 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1025 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1027 rinv00 = gmx_mm_invsqrt_pd(rsq00);
1028 rinv10 = gmx_mm_invsqrt_pd(rsq10);
1029 rinv20 = gmx_mm_invsqrt_pd(rsq20);
1031 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
1032 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
1033 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
1035 /* Load parameters for j particles */
1036 jq0 = _mm_load_sd(charge+jnrA+0);
1037 vdwjidx0A = 2*vdwtype[jnrA+0];
1039 fjx0 = _mm_setzero_pd();
1040 fjy0 = _mm_setzero_pd();
1041 fjz0 = _mm_setzero_pd();
1043 /**************************
1044 * CALCULATE INTERACTIONS *
1045 **************************/
1047 if (gmx_mm_any_lt(rsq00,rcutoff2))
1050 r00 = _mm_mul_pd(rsq00,rinv00);
1052 /* Compute parameters for interactions between i and j atoms */
1053 qq00 = _mm_mul_pd(iq0,jq0);
1054 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1056 /* EWALD ELECTROSTATICS */
1058 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1059 ewrt = _mm_mul_pd(r00,ewtabscale);
1060 ewitab = _mm_cvttpd_epi32(ewrt);
1062 eweps = _mm_frcz_pd(ewrt);
1064 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1066 twoeweps = _mm_add_pd(eweps,eweps);
1067 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1068 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1069 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1071 /* LENNARD-JONES DISPERSION/REPULSION */
1073 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1074 fvdw = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1076 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
1078 fscal = _mm_add_pd(felec,fvdw);
1080 fscal = _mm_and_pd(fscal,cutoff_mask);
1082 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1084 /* Update vectorial force */
1085 fix0 = _mm_macc_pd(dx00,fscal,fix0);
1086 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
1087 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
1089 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
1090 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
1091 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
1095 /**************************
1096 * CALCULATE INTERACTIONS *
1097 **************************/
1099 if (gmx_mm_any_lt(rsq10,rcutoff2))
1102 r10 = _mm_mul_pd(rsq10,rinv10);
1104 /* Compute parameters for interactions between i and j atoms */
1105 qq10 = _mm_mul_pd(iq1,jq0);
1107 /* EWALD ELECTROSTATICS */
1109 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1110 ewrt = _mm_mul_pd(r10,ewtabscale);
1111 ewitab = _mm_cvttpd_epi32(ewrt);
1113 eweps = _mm_frcz_pd(ewrt);
1115 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1117 twoeweps = _mm_add_pd(eweps,eweps);
1118 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1119 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1120 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1122 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
1126 fscal = _mm_and_pd(fscal,cutoff_mask);
1128 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1130 /* Update vectorial force */
1131 fix1 = _mm_macc_pd(dx10,fscal,fix1);
1132 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
1133 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
1135 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
1136 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
1137 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
1141 /**************************
1142 * CALCULATE INTERACTIONS *
1143 **************************/
1145 if (gmx_mm_any_lt(rsq20,rcutoff2))
1148 r20 = _mm_mul_pd(rsq20,rinv20);
1150 /* Compute parameters for interactions between i and j atoms */
1151 qq20 = _mm_mul_pd(iq2,jq0);
1153 /* EWALD ELECTROSTATICS */
1155 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1156 ewrt = _mm_mul_pd(r20,ewtabscale);
1157 ewitab = _mm_cvttpd_epi32(ewrt);
1159 eweps = _mm_frcz_pd(ewrt);
1161 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1163 twoeweps = _mm_add_pd(eweps,eweps);
1164 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1165 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1166 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1168 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
1172 fscal = _mm_and_pd(fscal,cutoff_mask);
1174 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1176 /* Update vectorial force */
1177 fix2 = _mm_macc_pd(dx20,fscal,fix2);
1178 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
1179 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
1181 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
1182 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
1183 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
1187 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1189 /* Inner loop uses 136 flops */
1192 /* End of innermost loop */
1194 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1195 f+i_coord_offset,fshift+i_shift_offset);
1197 /* Increment number of inner iterations */
1198 inneriter += j_index_end - j_index_start;
1200 /* Outer loop uses 18 flops */
1203 /* Increment number of outer iterations */
1206 /* Update outer/inner flops */
1208 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*136);