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_ElecCSTab_VdwLJ_GeomW3P1_VF_avx_128_fma_double
54 * Electrostatics interaction: CubicSplineTable
55 * VdW interaction: LennardJones
56 * Geometry: Water3-Particle
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_VF_avx_128_fma_double
61 (t_nblist * gmx_restrict nlist,
62 rvec * gmx_restrict xx,
63 rvec * gmx_restrict ff,
64 t_forcerec * gmx_restrict fr,
65 t_mdatoms * gmx_restrict mdatoms,
66 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
67 t_nrnb * gmx_restrict nrnb)
69 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
70 * just 0 for non-waters.
71 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
72 * jnr indices corresponding to data put in the four positions in the SIMD register.
74 int i_shift_offset,i_coord_offset,outeriter,inneriter;
75 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
77 int j_coord_offsetA,j_coord_offsetB;
78 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
80 real *shiftvec,*fshift,*x,*f;
81 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
83 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
85 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
87 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
88 int vdwjidx0A,vdwjidx0B;
89 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
90 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
91 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
92 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
93 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
96 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
99 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
100 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
102 __m128i ifour = _mm_set1_epi32(4);
103 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
105 __m128d dummy_mask,cutoff_mask;
106 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
107 __m128d one = _mm_set1_pd(1.0);
108 __m128d two = _mm_set1_pd(2.0);
114 jindex = nlist->jindex;
116 shiftidx = nlist->shift;
118 shiftvec = fr->shift_vec[0];
119 fshift = fr->fshift[0];
120 facel = _mm_set1_pd(fr->epsfac);
121 charge = mdatoms->chargeA;
122 nvdwtype = fr->ntype;
124 vdwtype = mdatoms->typeA;
126 vftab = kernel_data->table_elec->data;
127 vftabscale = _mm_set1_pd(kernel_data->table_elec->scale);
129 /* Setup water-specific parameters */
130 inr = nlist->iinr[0];
131 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
132 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
133 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
134 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
136 /* Avoid stupid compiler warnings */
144 /* Start outer loop over neighborlists */
145 for(iidx=0; iidx<nri; iidx++)
147 /* Load shift vector for this list */
148 i_shift_offset = DIM*shiftidx[iidx];
150 /* Load limits for loop over neighbors */
151 j_index_start = jindex[iidx];
152 j_index_end = jindex[iidx+1];
154 /* Get outer coordinate index */
156 i_coord_offset = DIM*inr;
158 /* Load i particle coords and add shift vector */
159 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
160 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
162 fix0 = _mm_setzero_pd();
163 fiy0 = _mm_setzero_pd();
164 fiz0 = _mm_setzero_pd();
165 fix1 = _mm_setzero_pd();
166 fiy1 = _mm_setzero_pd();
167 fiz1 = _mm_setzero_pd();
168 fix2 = _mm_setzero_pd();
169 fiy2 = _mm_setzero_pd();
170 fiz2 = _mm_setzero_pd();
172 /* Reset potential sums */
173 velecsum = _mm_setzero_pd();
174 vvdwsum = _mm_setzero_pd();
176 /* Start inner kernel loop */
177 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
180 /* Get j neighbor index, and coordinate index */
183 j_coord_offsetA = DIM*jnrA;
184 j_coord_offsetB = DIM*jnrB;
186 /* load j atom coordinates */
187 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
190 /* Calculate displacement vector */
191 dx00 = _mm_sub_pd(ix0,jx0);
192 dy00 = _mm_sub_pd(iy0,jy0);
193 dz00 = _mm_sub_pd(iz0,jz0);
194 dx10 = _mm_sub_pd(ix1,jx0);
195 dy10 = _mm_sub_pd(iy1,jy0);
196 dz10 = _mm_sub_pd(iz1,jz0);
197 dx20 = _mm_sub_pd(ix2,jx0);
198 dy20 = _mm_sub_pd(iy2,jy0);
199 dz20 = _mm_sub_pd(iz2,jz0);
201 /* Calculate squared distance and things based on it */
202 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
203 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
204 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
206 rinv00 = gmx_mm_invsqrt_pd(rsq00);
207 rinv10 = gmx_mm_invsqrt_pd(rsq10);
208 rinv20 = gmx_mm_invsqrt_pd(rsq20);
210 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
212 /* Load parameters for j particles */
213 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
214 vdwjidx0A = 2*vdwtype[jnrA+0];
215 vdwjidx0B = 2*vdwtype[jnrB+0];
217 fjx0 = _mm_setzero_pd();
218 fjy0 = _mm_setzero_pd();
219 fjz0 = _mm_setzero_pd();
221 /**************************
222 * CALCULATE INTERACTIONS *
223 **************************/
225 r00 = _mm_mul_pd(rsq00,rinv00);
227 /* Compute parameters for interactions between i and j atoms */
228 qq00 = _mm_mul_pd(iq0,jq0);
229 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
230 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
232 /* Calculate table index by multiplying r with table scale and truncate to integer */
233 rt = _mm_mul_pd(r00,vftabscale);
234 vfitab = _mm_cvttpd_epi32(rt);
236 vfeps = _mm_frcz_pd(rt);
238 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
240 twovfeps = _mm_add_pd(vfeps,vfeps);
241 vfitab = _mm_slli_epi32(vfitab,2);
243 /* CUBIC SPLINE TABLE ELECTROSTATICS */
244 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
245 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
246 GMX_MM_TRANSPOSE2_PD(Y,F);
247 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
248 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
249 GMX_MM_TRANSPOSE2_PD(G,H);
250 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
251 VV = _mm_macc_pd(vfeps,Fp,Y);
252 velec = _mm_mul_pd(qq00,VV);
253 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
254 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
256 /* LENNARD-JONES DISPERSION/REPULSION */
258 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
259 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
260 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
261 vvdw = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
262 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
264 /* Update potential sum for this i atom from the interaction with this j atom. */
265 velecsum = _mm_add_pd(velecsum,velec);
266 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
268 fscal = _mm_add_pd(felec,fvdw);
270 /* Update vectorial force */
271 fix0 = _mm_macc_pd(dx00,fscal,fix0);
272 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
273 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
275 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
276 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
277 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
279 /**************************
280 * CALCULATE INTERACTIONS *
281 **************************/
283 r10 = _mm_mul_pd(rsq10,rinv10);
285 /* Compute parameters for interactions between i and j atoms */
286 qq10 = _mm_mul_pd(iq1,jq0);
288 /* Calculate table index by multiplying r with table scale and truncate to integer */
289 rt = _mm_mul_pd(r10,vftabscale);
290 vfitab = _mm_cvttpd_epi32(rt);
292 vfeps = _mm_frcz_pd(rt);
294 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
296 twovfeps = _mm_add_pd(vfeps,vfeps);
297 vfitab = _mm_slli_epi32(vfitab,2);
299 /* CUBIC SPLINE TABLE ELECTROSTATICS */
300 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
301 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
302 GMX_MM_TRANSPOSE2_PD(Y,F);
303 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
304 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
305 GMX_MM_TRANSPOSE2_PD(G,H);
306 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
307 VV = _mm_macc_pd(vfeps,Fp,Y);
308 velec = _mm_mul_pd(qq10,VV);
309 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
310 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
312 /* Update potential sum for this i atom from the interaction with this j atom. */
313 velecsum = _mm_add_pd(velecsum,velec);
317 /* Update vectorial force */
318 fix1 = _mm_macc_pd(dx10,fscal,fix1);
319 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
320 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
322 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
323 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
324 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
326 /**************************
327 * CALCULATE INTERACTIONS *
328 **************************/
330 r20 = _mm_mul_pd(rsq20,rinv20);
332 /* Compute parameters for interactions between i and j atoms */
333 qq20 = _mm_mul_pd(iq2,jq0);
335 /* Calculate table index by multiplying r with table scale and truncate to integer */
336 rt = _mm_mul_pd(r20,vftabscale);
337 vfitab = _mm_cvttpd_epi32(rt);
339 vfeps = _mm_frcz_pd(rt);
341 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
343 twovfeps = _mm_add_pd(vfeps,vfeps);
344 vfitab = _mm_slli_epi32(vfitab,2);
346 /* CUBIC SPLINE TABLE ELECTROSTATICS */
347 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
348 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
349 GMX_MM_TRANSPOSE2_PD(Y,F);
350 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
351 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
352 GMX_MM_TRANSPOSE2_PD(G,H);
353 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
354 VV = _mm_macc_pd(vfeps,Fp,Y);
355 velec = _mm_mul_pd(qq20,VV);
356 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
357 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
359 /* Update potential sum for this i atom from the interaction with this j atom. */
360 velecsum = _mm_add_pd(velecsum,velec);
364 /* Update vectorial force */
365 fix2 = _mm_macc_pd(dx20,fscal,fix2);
366 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
367 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
369 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
370 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
371 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
373 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
375 /* Inner loop uses 154 flops */
382 j_coord_offsetA = DIM*jnrA;
384 /* load j atom coordinates */
385 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
388 /* Calculate displacement vector */
389 dx00 = _mm_sub_pd(ix0,jx0);
390 dy00 = _mm_sub_pd(iy0,jy0);
391 dz00 = _mm_sub_pd(iz0,jz0);
392 dx10 = _mm_sub_pd(ix1,jx0);
393 dy10 = _mm_sub_pd(iy1,jy0);
394 dz10 = _mm_sub_pd(iz1,jz0);
395 dx20 = _mm_sub_pd(ix2,jx0);
396 dy20 = _mm_sub_pd(iy2,jy0);
397 dz20 = _mm_sub_pd(iz2,jz0);
399 /* Calculate squared distance and things based on it */
400 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
401 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
402 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
404 rinv00 = gmx_mm_invsqrt_pd(rsq00);
405 rinv10 = gmx_mm_invsqrt_pd(rsq10);
406 rinv20 = gmx_mm_invsqrt_pd(rsq20);
408 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
410 /* Load parameters for j particles */
411 jq0 = _mm_load_sd(charge+jnrA+0);
412 vdwjidx0A = 2*vdwtype[jnrA+0];
414 fjx0 = _mm_setzero_pd();
415 fjy0 = _mm_setzero_pd();
416 fjz0 = _mm_setzero_pd();
418 /**************************
419 * CALCULATE INTERACTIONS *
420 **************************/
422 r00 = _mm_mul_pd(rsq00,rinv00);
424 /* Compute parameters for interactions between i and j atoms */
425 qq00 = _mm_mul_pd(iq0,jq0);
426 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
428 /* Calculate table index by multiplying r with table scale and truncate to integer */
429 rt = _mm_mul_pd(r00,vftabscale);
430 vfitab = _mm_cvttpd_epi32(rt);
432 vfeps = _mm_frcz_pd(rt);
434 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
436 twovfeps = _mm_add_pd(vfeps,vfeps);
437 vfitab = _mm_slli_epi32(vfitab,2);
439 /* CUBIC SPLINE TABLE ELECTROSTATICS */
440 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
441 F = _mm_setzero_pd();
442 GMX_MM_TRANSPOSE2_PD(Y,F);
443 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
444 H = _mm_setzero_pd();
445 GMX_MM_TRANSPOSE2_PD(G,H);
446 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
447 VV = _mm_macc_pd(vfeps,Fp,Y);
448 velec = _mm_mul_pd(qq00,VV);
449 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
450 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
452 /* LENNARD-JONES DISPERSION/REPULSION */
454 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
455 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
456 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
457 vvdw = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
458 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
460 /* Update potential sum for this i atom from the interaction with this j atom. */
461 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
462 velecsum = _mm_add_pd(velecsum,velec);
463 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
464 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
466 fscal = _mm_add_pd(felec,fvdw);
468 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
470 /* Update vectorial force */
471 fix0 = _mm_macc_pd(dx00,fscal,fix0);
472 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
473 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
475 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
476 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
477 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
479 /**************************
480 * CALCULATE INTERACTIONS *
481 **************************/
483 r10 = _mm_mul_pd(rsq10,rinv10);
485 /* Compute parameters for interactions between i and j atoms */
486 qq10 = _mm_mul_pd(iq1,jq0);
488 /* Calculate table index by multiplying r with table scale and truncate to integer */
489 rt = _mm_mul_pd(r10,vftabscale);
490 vfitab = _mm_cvttpd_epi32(rt);
492 vfeps = _mm_frcz_pd(rt);
494 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
496 twovfeps = _mm_add_pd(vfeps,vfeps);
497 vfitab = _mm_slli_epi32(vfitab,2);
499 /* CUBIC SPLINE TABLE ELECTROSTATICS */
500 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
501 F = _mm_setzero_pd();
502 GMX_MM_TRANSPOSE2_PD(Y,F);
503 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
504 H = _mm_setzero_pd();
505 GMX_MM_TRANSPOSE2_PD(G,H);
506 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
507 VV = _mm_macc_pd(vfeps,Fp,Y);
508 velec = _mm_mul_pd(qq10,VV);
509 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
510 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
512 /* Update potential sum for this i atom from the interaction with this j atom. */
513 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
514 velecsum = _mm_add_pd(velecsum,velec);
518 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
520 /* Update vectorial force */
521 fix1 = _mm_macc_pd(dx10,fscal,fix1);
522 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
523 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
525 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
526 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
527 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
529 /**************************
530 * CALCULATE INTERACTIONS *
531 **************************/
533 r20 = _mm_mul_pd(rsq20,rinv20);
535 /* Compute parameters for interactions between i and j atoms */
536 qq20 = _mm_mul_pd(iq2,jq0);
538 /* Calculate table index by multiplying r with table scale and truncate to integer */
539 rt = _mm_mul_pd(r20,vftabscale);
540 vfitab = _mm_cvttpd_epi32(rt);
542 vfeps = _mm_frcz_pd(rt);
544 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
546 twovfeps = _mm_add_pd(vfeps,vfeps);
547 vfitab = _mm_slli_epi32(vfitab,2);
549 /* CUBIC SPLINE TABLE ELECTROSTATICS */
550 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
551 F = _mm_setzero_pd();
552 GMX_MM_TRANSPOSE2_PD(Y,F);
553 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
554 H = _mm_setzero_pd();
555 GMX_MM_TRANSPOSE2_PD(G,H);
556 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
557 VV = _mm_macc_pd(vfeps,Fp,Y);
558 velec = _mm_mul_pd(qq20,VV);
559 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
560 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
562 /* Update potential sum for this i atom from the interaction with this j atom. */
563 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
564 velecsum = _mm_add_pd(velecsum,velec);
568 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
570 /* Update vectorial force */
571 fix2 = _mm_macc_pd(dx20,fscal,fix2);
572 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
573 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
575 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
576 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
577 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
579 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
581 /* Inner loop uses 154 flops */
584 /* End of innermost loop */
586 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
587 f+i_coord_offset,fshift+i_shift_offset);
590 /* Update potential energies */
591 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
592 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
594 /* Increment number of inner iterations */
595 inneriter += j_index_end - j_index_start;
597 /* Outer loop uses 20 flops */
600 /* Increment number of outer iterations */
603 /* Update outer/inner flops */
605 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*154);
608 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_avx_128_fma_double
609 * Electrostatics interaction: CubicSplineTable
610 * VdW interaction: LennardJones
611 * Geometry: Water3-Particle
612 * Calculate force/pot: Force
615 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_avx_128_fma_double
616 (t_nblist * gmx_restrict nlist,
617 rvec * gmx_restrict xx,
618 rvec * gmx_restrict ff,
619 t_forcerec * gmx_restrict fr,
620 t_mdatoms * gmx_restrict mdatoms,
621 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
622 t_nrnb * gmx_restrict nrnb)
624 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
625 * just 0 for non-waters.
626 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
627 * jnr indices corresponding to data put in the four positions in the SIMD register.
629 int i_shift_offset,i_coord_offset,outeriter,inneriter;
630 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
632 int j_coord_offsetA,j_coord_offsetB;
633 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
635 real *shiftvec,*fshift,*x,*f;
636 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
638 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
640 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
642 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
643 int vdwjidx0A,vdwjidx0B;
644 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
645 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
646 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
647 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
648 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
651 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
654 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
655 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
657 __m128i ifour = _mm_set1_epi32(4);
658 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
660 __m128d dummy_mask,cutoff_mask;
661 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
662 __m128d one = _mm_set1_pd(1.0);
663 __m128d two = _mm_set1_pd(2.0);
669 jindex = nlist->jindex;
671 shiftidx = nlist->shift;
673 shiftvec = fr->shift_vec[0];
674 fshift = fr->fshift[0];
675 facel = _mm_set1_pd(fr->epsfac);
676 charge = mdatoms->chargeA;
677 nvdwtype = fr->ntype;
679 vdwtype = mdatoms->typeA;
681 vftab = kernel_data->table_elec->data;
682 vftabscale = _mm_set1_pd(kernel_data->table_elec->scale);
684 /* Setup water-specific parameters */
685 inr = nlist->iinr[0];
686 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
687 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
688 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
689 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
691 /* Avoid stupid compiler warnings */
699 /* Start outer loop over neighborlists */
700 for(iidx=0; iidx<nri; iidx++)
702 /* Load shift vector for this list */
703 i_shift_offset = DIM*shiftidx[iidx];
705 /* Load limits for loop over neighbors */
706 j_index_start = jindex[iidx];
707 j_index_end = jindex[iidx+1];
709 /* Get outer coordinate index */
711 i_coord_offset = DIM*inr;
713 /* Load i particle coords and add shift vector */
714 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
715 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
717 fix0 = _mm_setzero_pd();
718 fiy0 = _mm_setzero_pd();
719 fiz0 = _mm_setzero_pd();
720 fix1 = _mm_setzero_pd();
721 fiy1 = _mm_setzero_pd();
722 fiz1 = _mm_setzero_pd();
723 fix2 = _mm_setzero_pd();
724 fiy2 = _mm_setzero_pd();
725 fiz2 = _mm_setzero_pd();
727 /* Start inner kernel loop */
728 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
731 /* Get j neighbor index, and coordinate index */
734 j_coord_offsetA = DIM*jnrA;
735 j_coord_offsetB = DIM*jnrB;
737 /* load j atom coordinates */
738 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
741 /* Calculate displacement vector */
742 dx00 = _mm_sub_pd(ix0,jx0);
743 dy00 = _mm_sub_pd(iy0,jy0);
744 dz00 = _mm_sub_pd(iz0,jz0);
745 dx10 = _mm_sub_pd(ix1,jx0);
746 dy10 = _mm_sub_pd(iy1,jy0);
747 dz10 = _mm_sub_pd(iz1,jz0);
748 dx20 = _mm_sub_pd(ix2,jx0);
749 dy20 = _mm_sub_pd(iy2,jy0);
750 dz20 = _mm_sub_pd(iz2,jz0);
752 /* Calculate squared distance and things based on it */
753 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
754 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
755 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
757 rinv00 = gmx_mm_invsqrt_pd(rsq00);
758 rinv10 = gmx_mm_invsqrt_pd(rsq10);
759 rinv20 = gmx_mm_invsqrt_pd(rsq20);
761 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
763 /* Load parameters for j particles */
764 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
765 vdwjidx0A = 2*vdwtype[jnrA+0];
766 vdwjidx0B = 2*vdwtype[jnrB+0];
768 fjx0 = _mm_setzero_pd();
769 fjy0 = _mm_setzero_pd();
770 fjz0 = _mm_setzero_pd();
772 /**************************
773 * CALCULATE INTERACTIONS *
774 **************************/
776 r00 = _mm_mul_pd(rsq00,rinv00);
778 /* Compute parameters for interactions between i and j atoms */
779 qq00 = _mm_mul_pd(iq0,jq0);
780 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
781 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
783 /* Calculate table index by multiplying r with table scale and truncate to integer */
784 rt = _mm_mul_pd(r00,vftabscale);
785 vfitab = _mm_cvttpd_epi32(rt);
787 vfeps = _mm_frcz_pd(rt);
789 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
791 twovfeps = _mm_add_pd(vfeps,vfeps);
792 vfitab = _mm_slli_epi32(vfitab,2);
794 /* CUBIC SPLINE TABLE ELECTROSTATICS */
795 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
796 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
797 GMX_MM_TRANSPOSE2_PD(Y,F);
798 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
799 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
800 GMX_MM_TRANSPOSE2_PD(G,H);
801 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
802 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
803 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
805 /* LENNARD-JONES DISPERSION/REPULSION */
807 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
808 fvdw = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
810 fscal = _mm_add_pd(felec,fvdw);
812 /* Update vectorial force */
813 fix0 = _mm_macc_pd(dx00,fscal,fix0);
814 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
815 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
817 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
818 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
819 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
821 /**************************
822 * CALCULATE INTERACTIONS *
823 **************************/
825 r10 = _mm_mul_pd(rsq10,rinv10);
827 /* Compute parameters for interactions between i and j atoms */
828 qq10 = _mm_mul_pd(iq1,jq0);
830 /* Calculate table index by multiplying r with table scale and truncate to integer */
831 rt = _mm_mul_pd(r10,vftabscale);
832 vfitab = _mm_cvttpd_epi32(rt);
834 vfeps = _mm_frcz_pd(rt);
836 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
838 twovfeps = _mm_add_pd(vfeps,vfeps);
839 vfitab = _mm_slli_epi32(vfitab,2);
841 /* CUBIC SPLINE TABLE ELECTROSTATICS */
842 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
843 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
844 GMX_MM_TRANSPOSE2_PD(Y,F);
845 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
846 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
847 GMX_MM_TRANSPOSE2_PD(G,H);
848 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
849 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
850 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
854 /* Update vectorial force */
855 fix1 = _mm_macc_pd(dx10,fscal,fix1);
856 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
857 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
859 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
860 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
861 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
863 /**************************
864 * CALCULATE INTERACTIONS *
865 **************************/
867 r20 = _mm_mul_pd(rsq20,rinv20);
869 /* Compute parameters for interactions between i and j atoms */
870 qq20 = _mm_mul_pd(iq2,jq0);
872 /* Calculate table index by multiplying r with table scale and truncate to integer */
873 rt = _mm_mul_pd(r20,vftabscale);
874 vfitab = _mm_cvttpd_epi32(rt);
876 vfeps = _mm_frcz_pd(rt);
878 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
880 twovfeps = _mm_add_pd(vfeps,vfeps);
881 vfitab = _mm_slli_epi32(vfitab,2);
883 /* CUBIC SPLINE TABLE ELECTROSTATICS */
884 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
885 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
886 GMX_MM_TRANSPOSE2_PD(Y,F);
887 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
888 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
889 GMX_MM_TRANSPOSE2_PD(G,H);
890 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
891 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
892 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
896 /* Update vectorial force */
897 fix2 = _mm_macc_pd(dx20,fscal,fix2);
898 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
899 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
901 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
902 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
903 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
905 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
907 /* Inner loop uses 137 flops */
914 j_coord_offsetA = DIM*jnrA;
916 /* load j atom coordinates */
917 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
920 /* Calculate displacement vector */
921 dx00 = _mm_sub_pd(ix0,jx0);
922 dy00 = _mm_sub_pd(iy0,jy0);
923 dz00 = _mm_sub_pd(iz0,jz0);
924 dx10 = _mm_sub_pd(ix1,jx0);
925 dy10 = _mm_sub_pd(iy1,jy0);
926 dz10 = _mm_sub_pd(iz1,jz0);
927 dx20 = _mm_sub_pd(ix2,jx0);
928 dy20 = _mm_sub_pd(iy2,jy0);
929 dz20 = _mm_sub_pd(iz2,jz0);
931 /* Calculate squared distance and things based on it */
932 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
933 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
934 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
936 rinv00 = gmx_mm_invsqrt_pd(rsq00);
937 rinv10 = gmx_mm_invsqrt_pd(rsq10);
938 rinv20 = gmx_mm_invsqrt_pd(rsq20);
940 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
942 /* Load parameters for j particles */
943 jq0 = _mm_load_sd(charge+jnrA+0);
944 vdwjidx0A = 2*vdwtype[jnrA+0];
946 fjx0 = _mm_setzero_pd();
947 fjy0 = _mm_setzero_pd();
948 fjz0 = _mm_setzero_pd();
950 /**************************
951 * CALCULATE INTERACTIONS *
952 **************************/
954 r00 = _mm_mul_pd(rsq00,rinv00);
956 /* Compute parameters for interactions between i and j atoms */
957 qq00 = _mm_mul_pd(iq0,jq0);
958 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
960 /* Calculate table index by multiplying r with table scale and truncate to integer */
961 rt = _mm_mul_pd(r00,vftabscale);
962 vfitab = _mm_cvttpd_epi32(rt);
964 vfeps = _mm_frcz_pd(rt);
966 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
968 twovfeps = _mm_add_pd(vfeps,vfeps);
969 vfitab = _mm_slli_epi32(vfitab,2);
971 /* CUBIC SPLINE TABLE ELECTROSTATICS */
972 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
973 F = _mm_setzero_pd();
974 GMX_MM_TRANSPOSE2_PD(Y,F);
975 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
976 H = _mm_setzero_pd();
977 GMX_MM_TRANSPOSE2_PD(G,H);
978 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
979 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
980 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
982 /* LENNARD-JONES DISPERSION/REPULSION */
984 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
985 fvdw = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
987 fscal = _mm_add_pd(felec,fvdw);
989 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
991 /* Update vectorial force */
992 fix0 = _mm_macc_pd(dx00,fscal,fix0);
993 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
994 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
996 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
997 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
998 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
1000 /**************************
1001 * CALCULATE INTERACTIONS *
1002 **************************/
1004 r10 = _mm_mul_pd(rsq10,rinv10);
1006 /* Compute parameters for interactions between i and j atoms */
1007 qq10 = _mm_mul_pd(iq1,jq0);
1009 /* Calculate table index by multiplying r with table scale and truncate to integer */
1010 rt = _mm_mul_pd(r10,vftabscale);
1011 vfitab = _mm_cvttpd_epi32(rt);
1013 vfeps = _mm_frcz_pd(rt);
1015 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1017 twovfeps = _mm_add_pd(vfeps,vfeps);
1018 vfitab = _mm_slli_epi32(vfitab,2);
1020 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1021 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1022 F = _mm_setzero_pd();
1023 GMX_MM_TRANSPOSE2_PD(Y,F);
1024 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1025 H = _mm_setzero_pd();
1026 GMX_MM_TRANSPOSE2_PD(G,H);
1027 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1028 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1029 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
1033 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1035 /* Update vectorial force */
1036 fix1 = _mm_macc_pd(dx10,fscal,fix1);
1037 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
1038 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
1040 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
1041 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
1042 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
1044 /**************************
1045 * CALCULATE INTERACTIONS *
1046 **************************/
1048 r20 = _mm_mul_pd(rsq20,rinv20);
1050 /* Compute parameters for interactions between i and j atoms */
1051 qq20 = _mm_mul_pd(iq2,jq0);
1053 /* Calculate table index by multiplying r with table scale and truncate to integer */
1054 rt = _mm_mul_pd(r20,vftabscale);
1055 vfitab = _mm_cvttpd_epi32(rt);
1057 vfeps = _mm_frcz_pd(rt);
1059 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1061 twovfeps = _mm_add_pd(vfeps,vfeps);
1062 vfitab = _mm_slli_epi32(vfitab,2);
1064 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1065 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1066 F = _mm_setzero_pd();
1067 GMX_MM_TRANSPOSE2_PD(Y,F);
1068 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1069 H = _mm_setzero_pd();
1070 GMX_MM_TRANSPOSE2_PD(G,H);
1071 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1072 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1073 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
1077 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1079 /* Update vectorial force */
1080 fix2 = _mm_macc_pd(dx20,fscal,fix2);
1081 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
1082 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
1084 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
1085 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
1086 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
1088 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1090 /* Inner loop uses 137 flops */
1093 /* End of innermost loop */
1095 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1096 f+i_coord_offset,fshift+i_shift_offset);
1098 /* Increment number of inner iterations */
1099 inneriter += j_index_end - j_index_start;
1101 /* Outer loop uses 18 flops */
1104 /* Increment number of outer iterations */
1107 /* Update outer/inner flops */
1109 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*137);