2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
36 * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
42 #include "../nb_kernel.h"
43 #include "gromacs/legacyheaders/types/simple.h"
44 #include "gromacs/math/vec.h"
45 #include "gromacs/legacyheaders/nrnb.h"
47 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
48 #include "kernelutil_x86_avx_128_fma_double.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_VF_avx_128_fma_double
52 * Electrostatics interaction: CubicSplineTable
53 * VdW interaction: LennardJones
54 * Geometry: Water4-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_VF_avx_128_fma_double
59 (t_nblist * gmx_restrict nlist,
60 rvec * gmx_restrict xx,
61 rvec * gmx_restrict ff,
62 t_forcerec * gmx_restrict fr,
63 t_mdatoms * gmx_restrict mdatoms,
64 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
65 t_nrnb * gmx_restrict nrnb)
67 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
68 * just 0 for non-waters.
69 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
70 * jnr indices corresponding to data put in the four positions in the SIMD register.
72 int i_shift_offset,i_coord_offset,outeriter,inneriter;
73 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
75 int j_coord_offsetA,j_coord_offsetB;
76 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
78 real *shiftvec,*fshift,*x,*f;
79 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
81 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
83 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
85 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
87 __m128d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
88 int vdwjidx0A,vdwjidx0B;
89 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
90 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
91 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
92 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
93 __m128d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
94 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
97 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
100 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
101 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
103 __m128i ifour = _mm_set1_epi32(4);
104 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
106 __m128d dummy_mask,cutoff_mask;
107 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
108 __m128d one = _mm_set1_pd(1.0);
109 __m128d two = _mm_set1_pd(2.0);
115 jindex = nlist->jindex;
117 shiftidx = nlist->shift;
119 shiftvec = fr->shift_vec[0];
120 fshift = fr->fshift[0];
121 facel = _mm_set1_pd(fr->epsfac);
122 charge = mdatoms->chargeA;
123 nvdwtype = fr->ntype;
125 vdwtype = mdatoms->typeA;
127 vftab = kernel_data->table_elec->data;
128 vftabscale = _mm_set1_pd(kernel_data->table_elec->scale);
130 /* Setup water-specific parameters */
131 inr = nlist->iinr[0];
132 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
133 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
134 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
135 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
137 /* Avoid stupid compiler warnings */
145 /* Start outer loop over neighborlists */
146 for(iidx=0; iidx<nri; iidx++)
148 /* Load shift vector for this list */
149 i_shift_offset = DIM*shiftidx[iidx];
151 /* Load limits for loop over neighbors */
152 j_index_start = jindex[iidx];
153 j_index_end = jindex[iidx+1];
155 /* Get outer coordinate index */
157 i_coord_offset = DIM*inr;
159 /* Load i particle coords and add shift vector */
160 gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
161 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
163 fix0 = _mm_setzero_pd();
164 fiy0 = _mm_setzero_pd();
165 fiz0 = _mm_setzero_pd();
166 fix1 = _mm_setzero_pd();
167 fiy1 = _mm_setzero_pd();
168 fiz1 = _mm_setzero_pd();
169 fix2 = _mm_setzero_pd();
170 fiy2 = _mm_setzero_pd();
171 fiz2 = _mm_setzero_pd();
172 fix3 = _mm_setzero_pd();
173 fiy3 = _mm_setzero_pd();
174 fiz3 = _mm_setzero_pd();
176 /* Reset potential sums */
177 velecsum = _mm_setzero_pd();
178 vvdwsum = _mm_setzero_pd();
180 /* Start inner kernel loop */
181 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
184 /* Get j neighbor index, and coordinate index */
187 j_coord_offsetA = DIM*jnrA;
188 j_coord_offsetB = DIM*jnrB;
190 /* load j atom coordinates */
191 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
194 /* Calculate displacement vector */
195 dx00 = _mm_sub_pd(ix0,jx0);
196 dy00 = _mm_sub_pd(iy0,jy0);
197 dz00 = _mm_sub_pd(iz0,jz0);
198 dx10 = _mm_sub_pd(ix1,jx0);
199 dy10 = _mm_sub_pd(iy1,jy0);
200 dz10 = _mm_sub_pd(iz1,jz0);
201 dx20 = _mm_sub_pd(ix2,jx0);
202 dy20 = _mm_sub_pd(iy2,jy0);
203 dz20 = _mm_sub_pd(iz2,jz0);
204 dx30 = _mm_sub_pd(ix3,jx0);
205 dy30 = _mm_sub_pd(iy3,jy0);
206 dz30 = _mm_sub_pd(iz3,jz0);
208 /* Calculate squared distance and things based on it */
209 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
210 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
211 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
212 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
214 rinv10 = gmx_mm_invsqrt_pd(rsq10);
215 rinv20 = gmx_mm_invsqrt_pd(rsq20);
216 rinv30 = gmx_mm_invsqrt_pd(rsq30);
218 rinvsq00 = gmx_mm_inv_pd(rsq00);
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 /* Compute parameters for interactions between i and j atoms */
234 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
235 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
237 /* LENNARD-JONES DISPERSION/REPULSION */
239 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
240 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
241 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
242 vvdw = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
243 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
245 /* Update potential sum for this i atom from the interaction with this j atom. */
246 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
250 /* Update vectorial force */
251 fix0 = _mm_macc_pd(dx00,fscal,fix0);
252 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
253 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
255 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
256 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
257 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
259 /**************************
260 * CALCULATE INTERACTIONS *
261 **************************/
263 r10 = _mm_mul_pd(rsq10,rinv10);
265 /* Compute parameters for interactions between i and j atoms */
266 qq10 = _mm_mul_pd(iq1,jq0);
268 /* Calculate table index by multiplying r with table scale and truncate to integer */
269 rt = _mm_mul_pd(r10,vftabscale);
270 vfitab = _mm_cvttpd_epi32(rt);
272 vfeps = _mm_frcz_pd(rt);
274 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
276 twovfeps = _mm_add_pd(vfeps,vfeps);
277 vfitab = _mm_slli_epi32(vfitab,2);
279 /* CUBIC SPLINE TABLE ELECTROSTATICS */
280 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
281 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
282 GMX_MM_TRANSPOSE2_PD(Y,F);
283 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
284 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
285 GMX_MM_TRANSPOSE2_PD(G,H);
286 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
287 VV = _mm_macc_pd(vfeps,Fp,Y);
288 velec = _mm_mul_pd(qq10,VV);
289 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
290 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
292 /* Update potential sum for this i atom from the interaction with this j atom. */
293 velecsum = _mm_add_pd(velecsum,velec);
297 /* Update vectorial force */
298 fix1 = _mm_macc_pd(dx10,fscal,fix1);
299 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
300 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
302 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
303 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
304 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
306 /**************************
307 * CALCULATE INTERACTIONS *
308 **************************/
310 r20 = _mm_mul_pd(rsq20,rinv20);
312 /* Compute parameters for interactions between i and j atoms */
313 qq20 = _mm_mul_pd(iq2,jq0);
315 /* Calculate table index by multiplying r with table scale and truncate to integer */
316 rt = _mm_mul_pd(r20,vftabscale);
317 vfitab = _mm_cvttpd_epi32(rt);
319 vfeps = _mm_frcz_pd(rt);
321 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
323 twovfeps = _mm_add_pd(vfeps,vfeps);
324 vfitab = _mm_slli_epi32(vfitab,2);
326 /* CUBIC SPLINE TABLE ELECTROSTATICS */
327 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
328 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
329 GMX_MM_TRANSPOSE2_PD(Y,F);
330 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
331 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
332 GMX_MM_TRANSPOSE2_PD(G,H);
333 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
334 VV = _mm_macc_pd(vfeps,Fp,Y);
335 velec = _mm_mul_pd(qq20,VV);
336 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
337 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
339 /* Update potential sum for this i atom from the interaction with this j atom. */
340 velecsum = _mm_add_pd(velecsum,velec);
344 /* Update vectorial force */
345 fix2 = _mm_macc_pd(dx20,fscal,fix2);
346 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
347 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
349 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
350 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
351 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
353 /**************************
354 * CALCULATE INTERACTIONS *
355 **************************/
357 r30 = _mm_mul_pd(rsq30,rinv30);
359 /* Compute parameters for interactions between i and j atoms */
360 qq30 = _mm_mul_pd(iq3,jq0);
362 /* Calculate table index by multiplying r with table scale and truncate to integer */
363 rt = _mm_mul_pd(r30,vftabscale);
364 vfitab = _mm_cvttpd_epi32(rt);
366 vfeps = _mm_frcz_pd(rt);
368 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
370 twovfeps = _mm_add_pd(vfeps,vfeps);
371 vfitab = _mm_slli_epi32(vfitab,2);
373 /* CUBIC SPLINE TABLE ELECTROSTATICS */
374 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
375 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
376 GMX_MM_TRANSPOSE2_PD(Y,F);
377 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
378 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
379 GMX_MM_TRANSPOSE2_PD(G,H);
380 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
381 VV = _mm_macc_pd(vfeps,Fp,Y);
382 velec = _mm_mul_pd(qq30,VV);
383 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
384 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
386 /* Update potential sum for this i atom from the interaction with this j atom. */
387 velecsum = _mm_add_pd(velecsum,velec);
391 /* Update vectorial force */
392 fix3 = _mm_macc_pd(dx30,fscal,fix3);
393 fiy3 = _mm_macc_pd(dy30,fscal,fiy3);
394 fiz3 = _mm_macc_pd(dz30,fscal,fiz3);
396 fjx0 = _mm_macc_pd(dx30,fscal,fjx0);
397 fjy0 = _mm_macc_pd(dy30,fscal,fjy0);
398 fjz0 = _mm_macc_pd(dz30,fscal,fjz0);
400 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
402 /* Inner loop uses 176 flops */
409 j_coord_offsetA = DIM*jnrA;
411 /* load j atom coordinates */
412 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
415 /* Calculate displacement vector */
416 dx00 = _mm_sub_pd(ix0,jx0);
417 dy00 = _mm_sub_pd(iy0,jy0);
418 dz00 = _mm_sub_pd(iz0,jz0);
419 dx10 = _mm_sub_pd(ix1,jx0);
420 dy10 = _mm_sub_pd(iy1,jy0);
421 dz10 = _mm_sub_pd(iz1,jz0);
422 dx20 = _mm_sub_pd(ix2,jx0);
423 dy20 = _mm_sub_pd(iy2,jy0);
424 dz20 = _mm_sub_pd(iz2,jz0);
425 dx30 = _mm_sub_pd(ix3,jx0);
426 dy30 = _mm_sub_pd(iy3,jy0);
427 dz30 = _mm_sub_pd(iz3,jz0);
429 /* Calculate squared distance and things based on it */
430 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
431 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
432 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
433 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
435 rinv10 = gmx_mm_invsqrt_pd(rsq10);
436 rinv20 = gmx_mm_invsqrt_pd(rsq20);
437 rinv30 = gmx_mm_invsqrt_pd(rsq30);
439 rinvsq00 = gmx_mm_inv_pd(rsq00);
441 /* Load parameters for j particles */
442 jq0 = _mm_load_sd(charge+jnrA+0);
443 vdwjidx0A = 2*vdwtype[jnrA+0];
445 fjx0 = _mm_setzero_pd();
446 fjy0 = _mm_setzero_pd();
447 fjz0 = _mm_setzero_pd();
449 /**************************
450 * CALCULATE INTERACTIONS *
451 **************************/
453 /* Compute parameters for interactions between i and j atoms */
454 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
456 /* LENNARD-JONES DISPERSION/REPULSION */
458 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
459 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
460 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
461 vvdw = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
462 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
464 /* Update potential sum for this i atom from the interaction with this j atom. */
465 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
466 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
470 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
472 /* Update vectorial force */
473 fix0 = _mm_macc_pd(dx00,fscal,fix0);
474 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
475 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
477 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
478 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
479 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
481 /**************************
482 * CALCULATE INTERACTIONS *
483 **************************/
485 r10 = _mm_mul_pd(rsq10,rinv10);
487 /* Compute parameters for interactions between i and j atoms */
488 qq10 = _mm_mul_pd(iq1,jq0);
490 /* Calculate table index by multiplying r with table scale and truncate to integer */
491 rt = _mm_mul_pd(r10,vftabscale);
492 vfitab = _mm_cvttpd_epi32(rt);
494 vfeps = _mm_frcz_pd(rt);
496 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
498 twovfeps = _mm_add_pd(vfeps,vfeps);
499 vfitab = _mm_slli_epi32(vfitab,2);
501 /* CUBIC SPLINE TABLE ELECTROSTATICS */
502 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
503 F = _mm_setzero_pd();
504 GMX_MM_TRANSPOSE2_PD(Y,F);
505 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
506 H = _mm_setzero_pd();
507 GMX_MM_TRANSPOSE2_PD(G,H);
508 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
509 VV = _mm_macc_pd(vfeps,Fp,Y);
510 velec = _mm_mul_pd(qq10,VV);
511 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
512 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
514 /* Update potential sum for this i atom from the interaction with this j atom. */
515 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
516 velecsum = _mm_add_pd(velecsum,velec);
520 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
522 /* Update vectorial force */
523 fix1 = _mm_macc_pd(dx10,fscal,fix1);
524 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
525 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
527 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
528 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
529 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
531 /**************************
532 * CALCULATE INTERACTIONS *
533 **************************/
535 r20 = _mm_mul_pd(rsq20,rinv20);
537 /* Compute parameters for interactions between i and j atoms */
538 qq20 = _mm_mul_pd(iq2,jq0);
540 /* Calculate table index by multiplying r with table scale and truncate to integer */
541 rt = _mm_mul_pd(r20,vftabscale);
542 vfitab = _mm_cvttpd_epi32(rt);
544 vfeps = _mm_frcz_pd(rt);
546 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
548 twovfeps = _mm_add_pd(vfeps,vfeps);
549 vfitab = _mm_slli_epi32(vfitab,2);
551 /* CUBIC SPLINE TABLE ELECTROSTATICS */
552 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
553 F = _mm_setzero_pd();
554 GMX_MM_TRANSPOSE2_PD(Y,F);
555 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
556 H = _mm_setzero_pd();
557 GMX_MM_TRANSPOSE2_PD(G,H);
558 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
559 VV = _mm_macc_pd(vfeps,Fp,Y);
560 velec = _mm_mul_pd(qq20,VV);
561 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
562 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
564 /* Update potential sum for this i atom from the interaction with this j atom. */
565 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
566 velecsum = _mm_add_pd(velecsum,velec);
570 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
572 /* Update vectorial force */
573 fix2 = _mm_macc_pd(dx20,fscal,fix2);
574 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
575 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
577 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
578 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
579 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
581 /**************************
582 * CALCULATE INTERACTIONS *
583 **************************/
585 r30 = _mm_mul_pd(rsq30,rinv30);
587 /* Compute parameters for interactions between i and j atoms */
588 qq30 = _mm_mul_pd(iq3,jq0);
590 /* Calculate table index by multiplying r with table scale and truncate to integer */
591 rt = _mm_mul_pd(r30,vftabscale);
592 vfitab = _mm_cvttpd_epi32(rt);
594 vfeps = _mm_frcz_pd(rt);
596 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
598 twovfeps = _mm_add_pd(vfeps,vfeps);
599 vfitab = _mm_slli_epi32(vfitab,2);
601 /* CUBIC SPLINE TABLE ELECTROSTATICS */
602 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
603 F = _mm_setzero_pd();
604 GMX_MM_TRANSPOSE2_PD(Y,F);
605 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
606 H = _mm_setzero_pd();
607 GMX_MM_TRANSPOSE2_PD(G,H);
608 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
609 VV = _mm_macc_pd(vfeps,Fp,Y);
610 velec = _mm_mul_pd(qq30,VV);
611 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
612 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
614 /* Update potential sum for this i atom from the interaction with this j atom. */
615 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
616 velecsum = _mm_add_pd(velecsum,velec);
620 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
622 /* Update vectorial force */
623 fix3 = _mm_macc_pd(dx30,fscal,fix3);
624 fiy3 = _mm_macc_pd(dy30,fscal,fiy3);
625 fiz3 = _mm_macc_pd(dz30,fscal,fiz3);
627 fjx0 = _mm_macc_pd(dx30,fscal,fjx0);
628 fjy0 = _mm_macc_pd(dy30,fscal,fjy0);
629 fjz0 = _mm_macc_pd(dz30,fscal,fjz0);
631 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
633 /* Inner loop uses 176 flops */
636 /* End of innermost loop */
638 gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
639 f+i_coord_offset,fshift+i_shift_offset);
642 /* Update potential energies */
643 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
644 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
646 /* Increment number of inner iterations */
647 inneriter += j_index_end - j_index_start;
649 /* Outer loop uses 26 flops */
652 /* Increment number of outer iterations */
655 /* Update outer/inner flops */
657 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*176);
660 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_avx_128_fma_double
661 * Electrostatics interaction: CubicSplineTable
662 * VdW interaction: LennardJones
663 * Geometry: Water4-Particle
664 * Calculate force/pot: Force
667 nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_avx_128_fma_double
668 (t_nblist * gmx_restrict nlist,
669 rvec * gmx_restrict xx,
670 rvec * gmx_restrict ff,
671 t_forcerec * gmx_restrict fr,
672 t_mdatoms * gmx_restrict mdatoms,
673 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
674 t_nrnb * gmx_restrict nrnb)
676 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
677 * just 0 for non-waters.
678 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
679 * jnr indices corresponding to data put in the four positions in the SIMD register.
681 int i_shift_offset,i_coord_offset,outeriter,inneriter;
682 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
684 int j_coord_offsetA,j_coord_offsetB;
685 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
687 real *shiftvec,*fshift,*x,*f;
688 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
690 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
692 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
694 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
696 __m128d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
697 int vdwjidx0A,vdwjidx0B;
698 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
699 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
700 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
701 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
702 __m128d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
703 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
706 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
709 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
710 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
712 __m128i ifour = _mm_set1_epi32(4);
713 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
715 __m128d dummy_mask,cutoff_mask;
716 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
717 __m128d one = _mm_set1_pd(1.0);
718 __m128d two = _mm_set1_pd(2.0);
724 jindex = nlist->jindex;
726 shiftidx = nlist->shift;
728 shiftvec = fr->shift_vec[0];
729 fshift = fr->fshift[0];
730 facel = _mm_set1_pd(fr->epsfac);
731 charge = mdatoms->chargeA;
732 nvdwtype = fr->ntype;
734 vdwtype = mdatoms->typeA;
736 vftab = kernel_data->table_elec->data;
737 vftabscale = _mm_set1_pd(kernel_data->table_elec->scale);
739 /* Setup water-specific parameters */
740 inr = nlist->iinr[0];
741 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
742 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
743 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
744 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
746 /* Avoid stupid compiler warnings */
754 /* Start outer loop over neighborlists */
755 for(iidx=0; iidx<nri; iidx++)
757 /* Load shift vector for this list */
758 i_shift_offset = DIM*shiftidx[iidx];
760 /* Load limits for loop over neighbors */
761 j_index_start = jindex[iidx];
762 j_index_end = jindex[iidx+1];
764 /* Get outer coordinate index */
766 i_coord_offset = DIM*inr;
768 /* Load i particle coords and add shift vector */
769 gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
770 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
772 fix0 = _mm_setzero_pd();
773 fiy0 = _mm_setzero_pd();
774 fiz0 = _mm_setzero_pd();
775 fix1 = _mm_setzero_pd();
776 fiy1 = _mm_setzero_pd();
777 fiz1 = _mm_setzero_pd();
778 fix2 = _mm_setzero_pd();
779 fiy2 = _mm_setzero_pd();
780 fiz2 = _mm_setzero_pd();
781 fix3 = _mm_setzero_pd();
782 fiy3 = _mm_setzero_pd();
783 fiz3 = _mm_setzero_pd();
785 /* Start inner kernel loop */
786 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
789 /* Get j neighbor index, and coordinate index */
792 j_coord_offsetA = DIM*jnrA;
793 j_coord_offsetB = DIM*jnrB;
795 /* load j atom coordinates */
796 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
799 /* Calculate displacement vector */
800 dx00 = _mm_sub_pd(ix0,jx0);
801 dy00 = _mm_sub_pd(iy0,jy0);
802 dz00 = _mm_sub_pd(iz0,jz0);
803 dx10 = _mm_sub_pd(ix1,jx0);
804 dy10 = _mm_sub_pd(iy1,jy0);
805 dz10 = _mm_sub_pd(iz1,jz0);
806 dx20 = _mm_sub_pd(ix2,jx0);
807 dy20 = _mm_sub_pd(iy2,jy0);
808 dz20 = _mm_sub_pd(iz2,jz0);
809 dx30 = _mm_sub_pd(ix3,jx0);
810 dy30 = _mm_sub_pd(iy3,jy0);
811 dz30 = _mm_sub_pd(iz3,jz0);
813 /* Calculate squared distance and things based on it */
814 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
815 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
816 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
817 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
819 rinv10 = gmx_mm_invsqrt_pd(rsq10);
820 rinv20 = gmx_mm_invsqrt_pd(rsq20);
821 rinv30 = gmx_mm_invsqrt_pd(rsq30);
823 rinvsq00 = gmx_mm_inv_pd(rsq00);
825 /* Load parameters for j particles */
826 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
827 vdwjidx0A = 2*vdwtype[jnrA+0];
828 vdwjidx0B = 2*vdwtype[jnrB+0];
830 fjx0 = _mm_setzero_pd();
831 fjy0 = _mm_setzero_pd();
832 fjz0 = _mm_setzero_pd();
834 /**************************
835 * CALCULATE INTERACTIONS *
836 **************************/
838 /* Compute parameters for interactions between i and j atoms */
839 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
840 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
842 /* LENNARD-JONES DISPERSION/REPULSION */
844 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
845 fvdw = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
849 /* Update vectorial force */
850 fix0 = _mm_macc_pd(dx00,fscal,fix0);
851 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
852 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
854 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
855 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
856 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
858 /**************************
859 * CALCULATE INTERACTIONS *
860 **************************/
862 r10 = _mm_mul_pd(rsq10,rinv10);
864 /* Compute parameters for interactions between i and j atoms */
865 qq10 = _mm_mul_pd(iq1,jq0);
867 /* Calculate table index by multiplying r with table scale and truncate to integer */
868 rt = _mm_mul_pd(r10,vftabscale);
869 vfitab = _mm_cvttpd_epi32(rt);
871 vfeps = _mm_frcz_pd(rt);
873 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
875 twovfeps = _mm_add_pd(vfeps,vfeps);
876 vfitab = _mm_slli_epi32(vfitab,2);
878 /* CUBIC SPLINE TABLE ELECTROSTATICS */
879 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
880 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
881 GMX_MM_TRANSPOSE2_PD(Y,F);
882 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
883 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
884 GMX_MM_TRANSPOSE2_PD(G,H);
885 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
886 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
887 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
891 /* Update vectorial force */
892 fix1 = _mm_macc_pd(dx10,fscal,fix1);
893 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
894 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
896 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
897 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
898 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
900 /**************************
901 * CALCULATE INTERACTIONS *
902 **************************/
904 r20 = _mm_mul_pd(rsq20,rinv20);
906 /* Compute parameters for interactions between i and j atoms */
907 qq20 = _mm_mul_pd(iq2,jq0);
909 /* Calculate table index by multiplying r with table scale and truncate to integer */
910 rt = _mm_mul_pd(r20,vftabscale);
911 vfitab = _mm_cvttpd_epi32(rt);
913 vfeps = _mm_frcz_pd(rt);
915 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
917 twovfeps = _mm_add_pd(vfeps,vfeps);
918 vfitab = _mm_slli_epi32(vfitab,2);
920 /* CUBIC SPLINE TABLE ELECTROSTATICS */
921 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
922 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
923 GMX_MM_TRANSPOSE2_PD(Y,F);
924 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
925 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
926 GMX_MM_TRANSPOSE2_PD(G,H);
927 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
928 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
929 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
933 /* Update vectorial force */
934 fix2 = _mm_macc_pd(dx20,fscal,fix2);
935 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
936 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
938 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
939 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
940 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
942 /**************************
943 * CALCULATE INTERACTIONS *
944 **************************/
946 r30 = _mm_mul_pd(rsq30,rinv30);
948 /* Compute parameters for interactions between i and j atoms */
949 qq30 = _mm_mul_pd(iq3,jq0);
951 /* Calculate table index by multiplying r with table scale and truncate to integer */
952 rt = _mm_mul_pd(r30,vftabscale);
953 vfitab = _mm_cvttpd_epi32(rt);
955 vfeps = _mm_frcz_pd(rt);
957 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
959 twovfeps = _mm_add_pd(vfeps,vfeps);
960 vfitab = _mm_slli_epi32(vfitab,2);
962 /* CUBIC SPLINE TABLE ELECTROSTATICS */
963 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
964 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
965 GMX_MM_TRANSPOSE2_PD(Y,F);
966 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
967 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
968 GMX_MM_TRANSPOSE2_PD(G,H);
969 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
970 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
971 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
975 /* Update vectorial force */
976 fix3 = _mm_macc_pd(dx30,fscal,fix3);
977 fiy3 = _mm_macc_pd(dy30,fscal,fiy3);
978 fiz3 = _mm_macc_pd(dz30,fscal,fiz3);
980 fjx0 = _mm_macc_pd(dx30,fscal,fjx0);
981 fjy0 = _mm_macc_pd(dy30,fscal,fjy0);
982 fjz0 = _mm_macc_pd(dz30,fscal,fjz0);
984 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
986 /* Inner loop uses 159 flops */
993 j_coord_offsetA = DIM*jnrA;
995 /* load j atom coordinates */
996 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
999 /* Calculate displacement vector */
1000 dx00 = _mm_sub_pd(ix0,jx0);
1001 dy00 = _mm_sub_pd(iy0,jy0);
1002 dz00 = _mm_sub_pd(iz0,jz0);
1003 dx10 = _mm_sub_pd(ix1,jx0);
1004 dy10 = _mm_sub_pd(iy1,jy0);
1005 dz10 = _mm_sub_pd(iz1,jz0);
1006 dx20 = _mm_sub_pd(ix2,jx0);
1007 dy20 = _mm_sub_pd(iy2,jy0);
1008 dz20 = _mm_sub_pd(iz2,jz0);
1009 dx30 = _mm_sub_pd(ix3,jx0);
1010 dy30 = _mm_sub_pd(iy3,jy0);
1011 dz30 = _mm_sub_pd(iz3,jz0);
1013 /* Calculate squared distance and things based on it */
1014 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1015 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1016 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1017 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1019 rinv10 = gmx_mm_invsqrt_pd(rsq10);
1020 rinv20 = gmx_mm_invsqrt_pd(rsq20);
1021 rinv30 = gmx_mm_invsqrt_pd(rsq30);
1023 rinvsq00 = gmx_mm_inv_pd(rsq00);
1025 /* Load parameters for j particles */
1026 jq0 = _mm_load_sd(charge+jnrA+0);
1027 vdwjidx0A = 2*vdwtype[jnrA+0];
1029 fjx0 = _mm_setzero_pd();
1030 fjy0 = _mm_setzero_pd();
1031 fjz0 = _mm_setzero_pd();
1033 /**************************
1034 * CALCULATE INTERACTIONS *
1035 **************************/
1037 /* Compute parameters for interactions between i and j atoms */
1038 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1040 /* LENNARD-JONES DISPERSION/REPULSION */
1042 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1043 fvdw = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1047 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1049 /* Update vectorial force */
1050 fix0 = _mm_macc_pd(dx00,fscal,fix0);
1051 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
1052 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
1054 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
1055 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
1056 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
1058 /**************************
1059 * CALCULATE INTERACTIONS *
1060 **************************/
1062 r10 = _mm_mul_pd(rsq10,rinv10);
1064 /* Compute parameters for interactions between i and j atoms */
1065 qq10 = _mm_mul_pd(iq1,jq0);
1067 /* Calculate table index by multiplying r with table scale and truncate to integer */
1068 rt = _mm_mul_pd(r10,vftabscale);
1069 vfitab = _mm_cvttpd_epi32(rt);
1071 vfeps = _mm_frcz_pd(rt);
1073 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1075 twovfeps = _mm_add_pd(vfeps,vfeps);
1076 vfitab = _mm_slli_epi32(vfitab,2);
1078 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1079 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1080 F = _mm_setzero_pd();
1081 GMX_MM_TRANSPOSE2_PD(Y,F);
1082 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1083 H = _mm_setzero_pd();
1084 GMX_MM_TRANSPOSE2_PD(G,H);
1085 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1086 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1087 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
1091 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1093 /* Update vectorial force */
1094 fix1 = _mm_macc_pd(dx10,fscal,fix1);
1095 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
1096 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
1098 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
1099 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
1100 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
1102 /**************************
1103 * CALCULATE INTERACTIONS *
1104 **************************/
1106 r20 = _mm_mul_pd(rsq20,rinv20);
1108 /* Compute parameters for interactions between i and j atoms */
1109 qq20 = _mm_mul_pd(iq2,jq0);
1111 /* Calculate table index by multiplying r with table scale and truncate to integer */
1112 rt = _mm_mul_pd(r20,vftabscale);
1113 vfitab = _mm_cvttpd_epi32(rt);
1115 vfeps = _mm_frcz_pd(rt);
1117 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1119 twovfeps = _mm_add_pd(vfeps,vfeps);
1120 vfitab = _mm_slli_epi32(vfitab,2);
1122 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1123 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1124 F = _mm_setzero_pd();
1125 GMX_MM_TRANSPOSE2_PD(Y,F);
1126 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1127 H = _mm_setzero_pd();
1128 GMX_MM_TRANSPOSE2_PD(G,H);
1129 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1130 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1131 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
1135 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1137 /* Update vectorial force */
1138 fix2 = _mm_macc_pd(dx20,fscal,fix2);
1139 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
1140 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
1142 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
1143 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
1144 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
1146 /**************************
1147 * CALCULATE INTERACTIONS *
1148 **************************/
1150 r30 = _mm_mul_pd(rsq30,rinv30);
1152 /* Compute parameters for interactions between i and j atoms */
1153 qq30 = _mm_mul_pd(iq3,jq0);
1155 /* Calculate table index by multiplying r with table scale and truncate to integer */
1156 rt = _mm_mul_pd(r30,vftabscale);
1157 vfitab = _mm_cvttpd_epi32(rt);
1159 vfeps = _mm_frcz_pd(rt);
1161 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1163 twovfeps = _mm_add_pd(vfeps,vfeps);
1164 vfitab = _mm_slli_epi32(vfitab,2);
1166 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1167 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1168 F = _mm_setzero_pd();
1169 GMX_MM_TRANSPOSE2_PD(Y,F);
1170 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1171 H = _mm_setzero_pd();
1172 GMX_MM_TRANSPOSE2_PD(G,H);
1173 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1174 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1175 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
1179 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1181 /* Update vectorial force */
1182 fix3 = _mm_macc_pd(dx30,fscal,fix3);
1183 fiy3 = _mm_macc_pd(dy30,fscal,fiy3);
1184 fiz3 = _mm_macc_pd(dz30,fscal,fiz3);
1186 fjx0 = _mm_macc_pd(dx30,fscal,fjx0);
1187 fjy0 = _mm_macc_pd(dy30,fscal,fjy0);
1188 fjz0 = _mm_macc_pd(dz30,fscal,fjz0);
1190 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1192 /* Inner loop uses 159 flops */
1195 /* End of innermost loop */
1197 gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1198 f+i_coord_offset,fshift+i_shift_offset);
1200 /* Increment number of inner iterations */
1201 inneriter += j_index_end - j_index_start;
1203 /* Outer loop uses 24 flops */
1206 /* Increment number of outer iterations */
1209 /* Update outer/inner flops */
1211 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*159);