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_ElecEw_VdwLJEw_GeomP1P1_VF_avx_128_fma_double
52 * Electrostatics interaction: Ewald
53 * VdW interaction: LJEwald
54 * Geometry: Particle-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecEw_VdwLJEw_GeomP1P1_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;
82 int vdwjidx0A,vdwjidx0B;
83 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
84 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
85 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
88 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
91 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
92 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
95 __m128d ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
96 __m128d one_half = _mm_set1_pd(0.5);
97 __m128d minus_one = _mm_set1_pd(-1.0);
99 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
101 __m128d dummy_mask,cutoff_mask;
102 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
103 __m128d one = _mm_set1_pd(1.0);
104 __m128d two = _mm_set1_pd(2.0);
110 jindex = nlist->jindex;
112 shiftidx = nlist->shift;
114 shiftvec = fr->shift_vec[0];
115 fshift = fr->fshift[0];
116 facel = _mm_set1_pd(fr->epsfac);
117 charge = mdatoms->chargeA;
118 nvdwtype = fr->ntype;
120 vdwtype = mdatoms->typeA;
121 vdwgridparam = fr->ljpme_c6grid;
122 sh_lj_ewald = _mm_set1_pd(fr->ic->sh_lj_ewald);
123 ewclj = _mm_set1_pd(fr->ewaldcoeff_lj);
124 ewclj2 = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
126 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
127 ewtab = fr->ic->tabq_coul_FDV0;
128 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
129 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
131 /* Avoid stupid compiler warnings */
139 /* Start outer loop over neighborlists */
140 for(iidx=0; iidx<nri; iidx++)
142 /* Load shift vector for this list */
143 i_shift_offset = DIM*shiftidx[iidx];
145 /* Load limits for loop over neighbors */
146 j_index_start = jindex[iidx];
147 j_index_end = jindex[iidx+1];
149 /* Get outer coordinate index */
151 i_coord_offset = DIM*inr;
153 /* Load i particle coords and add shift vector */
154 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
156 fix0 = _mm_setzero_pd();
157 fiy0 = _mm_setzero_pd();
158 fiz0 = _mm_setzero_pd();
160 /* Load parameters for i particles */
161 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
162 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
164 /* Reset potential sums */
165 velecsum = _mm_setzero_pd();
166 vvdwsum = _mm_setzero_pd();
168 /* Start inner kernel loop */
169 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
172 /* Get j neighbor index, and coordinate index */
175 j_coord_offsetA = DIM*jnrA;
176 j_coord_offsetB = DIM*jnrB;
178 /* load j atom coordinates */
179 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
182 /* Calculate displacement vector */
183 dx00 = _mm_sub_pd(ix0,jx0);
184 dy00 = _mm_sub_pd(iy0,jy0);
185 dz00 = _mm_sub_pd(iz0,jz0);
187 /* Calculate squared distance and things based on it */
188 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
190 rinv00 = gmx_mm_invsqrt_pd(rsq00);
192 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
194 /* Load parameters for j particles */
195 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
196 vdwjidx0A = 2*vdwtype[jnrA+0];
197 vdwjidx0B = 2*vdwtype[jnrB+0];
199 /**************************
200 * CALCULATE INTERACTIONS *
201 **************************/
203 r00 = _mm_mul_pd(rsq00,rinv00);
205 /* Compute parameters for interactions between i and j atoms */
206 qq00 = _mm_mul_pd(iq0,jq0);
207 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
208 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
209 c6grid_00 = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
210 vdwgridparam+vdwioffset0+vdwjidx0B);
212 /* EWALD ELECTROSTATICS */
214 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
215 ewrt = _mm_mul_pd(r00,ewtabscale);
216 ewitab = _mm_cvttpd_epi32(ewrt);
218 eweps = _mm_frcz_pd(ewrt);
220 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
222 twoeweps = _mm_add_pd(eweps,eweps);
223 ewitab = _mm_slli_epi32(ewitab,2);
224 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
225 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
226 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
227 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
228 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
229 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
230 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
231 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
232 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
233 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
235 /* Analytical LJ-PME */
236 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
237 ewcljrsq = _mm_mul_pd(ewclj2,rsq00);
238 ewclj6 = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
239 exponent = gmx_simd_exp_d(ewcljrsq);
240 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
241 poly = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
242 /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
243 vvdw6 = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
244 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
245 vvdw = _mm_msub_pd(vvdw12,one_twelfth,_mm_mul_pd(vvdw6,one_sixth));
246 /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
247 fvdw = _mm_mul_pd(_mm_add_pd(vvdw12,_mm_msub_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6),vvdw6)),rinvsq00);
249 /* Update potential sum for this i atom from the interaction with this j atom. */
250 velecsum = _mm_add_pd(velecsum,velec);
251 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
253 fscal = _mm_add_pd(felec,fvdw);
255 /* Update vectorial force */
256 fix0 = _mm_macc_pd(dx00,fscal,fix0);
257 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
258 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
260 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
261 _mm_mul_pd(dx00,fscal),
262 _mm_mul_pd(dy00,fscal),
263 _mm_mul_pd(dz00,fscal));
265 /* Inner loop uses 68 flops */
272 j_coord_offsetA = DIM*jnrA;
274 /* load j atom coordinates */
275 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
278 /* Calculate displacement vector */
279 dx00 = _mm_sub_pd(ix0,jx0);
280 dy00 = _mm_sub_pd(iy0,jy0);
281 dz00 = _mm_sub_pd(iz0,jz0);
283 /* Calculate squared distance and things based on it */
284 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
286 rinv00 = gmx_mm_invsqrt_pd(rsq00);
288 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
290 /* Load parameters for j particles */
291 jq0 = _mm_load_sd(charge+jnrA+0);
292 vdwjidx0A = 2*vdwtype[jnrA+0];
294 /**************************
295 * CALCULATE INTERACTIONS *
296 **************************/
298 r00 = _mm_mul_pd(rsq00,rinv00);
300 /* Compute parameters for interactions between i and j atoms */
301 qq00 = _mm_mul_pd(iq0,jq0);
302 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
303 c6grid_00 = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
305 /* EWALD ELECTROSTATICS */
307 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
308 ewrt = _mm_mul_pd(r00,ewtabscale);
309 ewitab = _mm_cvttpd_epi32(ewrt);
311 eweps = _mm_frcz_pd(ewrt);
313 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
315 twoeweps = _mm_add_pd(eweps,eweps);
316 ewitab = _mm_slli_epi32(ewitab,2);
317 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
318 ewtabD = _mm_setzero_pd();
319 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
320 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
321 ewtabFn = _mm_setzero_pd();
322 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
323 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
324 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
325 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
326 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
328 /* Analytical LJ-PME */
329 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
330 ewcljrsq = _mm_mul_pd(ewclj2,rsq00);
331 ewclj6 = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
332 exponent = gmx_simd_exp_d(ewcljrsq);
333 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
334 poly = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
335 /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
336 vvdw6 = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
337 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
338 vvdw = _mm_msub_pd(vvdw12,one_twelfth,_mm_mul_pd(vvdw6,one_sixth));
339 /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
340 fvdw = _mm_mul_pd(_mm_add_pd(vvdw12,_mm_msub_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6),vvdw6)),rinvsq00);
342 /* Update potential sum for this i atom from the interaction with this j atom. */
343 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
344 velecsum = _mm_add_pd(velecsum,velec);
345 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
346 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
348 fscal = _mm_add_pd(felec,fvdw);
350 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
352 /* Update vectorial force */
353 fix0 = _mm_macc_pd(dx00,fscal,fix0);
354 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
355 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
357 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
358 _mm_mul_pd(dx00,fscal),
359 _mm_mul_pd(dy00,fscal),
360 _mm_mul_pd(dz00,fscal));
362 /* Inner loop uses 68 flops */
365 /* End of innermost loop */
367 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
368 f+i_coord_offset,fshift+i_shift_offset);
371 /* Update potential energies */
372 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
373 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
375 /* Increment number of inner iterations */
376 inneriter += j_index_end - j_index_start;
378 /* Outer loop uses 9 flops */
381 /* Increment number of outer iterations */
384 /* Update outer/inner flops */
386 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*68);
389 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwLJEw_GeomP1P1_F_avx_128_fma_double
390 * Electrostatics interaction: Ewald
391 * VdW interaction: LJEwald
392 * Geometry: Particle-Particle
393 * Calculate force/pot: Force
396 nb_kernel_ElecEw_VdwLJEw_GeomP1P1_F_avx_128_fma_double
397 (t_nblist * gmx_restrict nlist,
398 rvec * gmx_restrict xx,
399 rvec * gmx_restrict ff,
400 t_forcerec * gmx_restrict fr,
401 t_mdatoms * gmx_restrict mdatoms,
402 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
403 t_nrnb * gmx_restrict nrnb)
405 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
406 * just 0 for non-waters.
407 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
408 * jnr indices corresponding to data put in the four positions in the SIMD register.
410 int i_shift_offset,i_coord_offset,outeriter,inneriter;
411 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
413 int j_coord_offsetA,j_coord_offsetB;
414 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
416 real *shiftvec,*fshift,*x,*f;
417 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
419 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
420 int vdwjidx0A,vdwjidx0B;
421 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
422 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
423 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
426 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
429 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
430 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
433 __m128d ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
434 __m128d one_half = _mm_set1_pd(0.5);
435 __m128d minus_one = _mm_set1_pd(-1.0);
437 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
439 __m128d dummy_mask,cutoff_mask;
440 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
441 __m128d one = _mm_set1_pd(1.0);
442 __m128d two = _mm_set1_pd(2.0);
448 jindex = nlist->jindex;
450 shiftidx = nlist->shift;
452 shiftvec = fr->shift_vec[0];
453 fshift = fr->fshift[0];
454 facel = _mm_set1_pd(fr->epsfac);
455 charge = mdatoms->chargeA;
456 nvdwtype = fr->ntype;
458 vdwtype = mdatoms->typeA;
459 vdwgridparam = fr->ljpme_c6grid;
460 sh_lj_ewald = _mm_set1_pd(fr->ic->sh_lj_ewald);
461 ewclj = _mm_set1_pd(fr->ewaldcoeff_lj);
462 ewclj2 = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
464 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
465 ewtab = fr->ic->tabq_coul_F;
466 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
467 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
469 /* Avoid stupid compiler warnings */
477 /* Start outer loop over neighborlists */
478 for(iidx=0; iidx<nri; iidx++)
480 /* Load shift vector for this list */
481 i_shift_offset = DIM*shiftidx[iidx];
483 /* Load limits for loop over neighbors */
484 j_index_start = jindex[iidx];
485 j_index_end = jindex[iidx+1];
487 /* Get outer coordinate index */
489 i_coord_offset = DIM*inr;
491 /* Load i particle coords and add shift vector */
492 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
494 fix0 = _mm_setzero_pd();
495 fiy0 = _mm_setzero_pd();
496 fiz0 = _mm_setzero_pd();
498 /* Load parameters for i particles */
499 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
500 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
502 /* Start inner kernel loop */
503 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
506 /* Get j neighbor index, and coordinate index */
509 j_coord_offsetA = DIM*jnrA;
510 j_coord_offsetB = DIM*jnrB;
512 /* load j atom coordinates */
513 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
516 /* Calculate displacement vector */
517 dx00 = _mm_sub_pd(ix0,jx0);
518 dy00 = _mm_sub_pd(iy0,jy0);
519 dz00 = _mm_sub_pd(iz0,jz0);
521 /* Calculate squared distance and things based on it */
522 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
524 rinv00 = gmx_mm_invsqrt_pd(rsq00);
526 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
528 /* Load parameters for j particles */
529 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
530 vdwjidx0A = 2*vdwtype[jnrA+0];
531 vdwjidx0B = 2*vdwtype[jnrB+0];
533 /**************************
534 * CALCULATE INTERACTIONS *
535 **************************/
537 r00 = _mm_mul_pd(rsq00,rinv00);
539 /* Compute parameters for interactions between i and j atoms */
540 qq00 = _mm_mul_pd(iq0,jq0);
541 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
542 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
543 c6grid_00 = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
544 vdwgridparam+vdwioffset0+vdwjidx0B);
546 /* EWALD ELECTROSTATICS */
548 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
549 ewrt = _mm_mul_pd(r00,ewtabscale);
550 ewitab = _mm_cvttpd_epi32(ewrt);
552 eweps = _mm_frcz_pd(ewrt);
554 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
556 twoeweps = _mm_add_pd(eweps,eweps);
557 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
559 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
560 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
562 /* Analytical LJ-PME */
563 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
564 ewcljrsq = _mm_mul_pd(ewclj2,rsq00);
565 ewclj6 = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
566 exponent = gmx_simd_exp_d(ewcljrsq);
567 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
568 poly = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
569 /* f6A = 6 * C6grid * (1 - poly) */
570 f6A = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
571 /* f6B = C6grid * exponent * beta^6 */
572 f6B = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
573 /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
574 fvdw = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
576 fscal = _mm_add_pd(felec,fvdw);
578 /* Update vectorial force */
579 fix0 = _mm_macc_pd(dx00,fscal,fix0);
580 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
581 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
583 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
584 _mm_mul_pd(dx00,fscal),
585 _mm_mul_pd(dy00,fscal),
586 _mm_mul_pd(dz00,fscal));
588 /* Inner loop uses 60 flops */
595 j_coord_offsetA = DIM*jnrA;
597 /* load j atom coordinates */
598 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
601 /* Calculate displacement vector */
602 dx00 = _mm_sub_pd(ix0,jx0);
603 dy00 = _mm_sub_pd(iy0,jy0);
604 dz00 = _mm_sub_pd(iz0,jz0);
606 /* Calculate squared distance and things based on it */
607 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
609 rinv00 = gmx_mm_invsqrt_pd(rsq00);
611 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
613 /* Load parameters for j particles */
614 jq0 = _mm_load_sd(charge+jnrA+0);
615 vdwjidx0A = 2*vdwtype[jnrA+0];
617 /**************************
618 * CALCULATE INTERACTIONS *
619 **************************/
621 r00 = _mm_mul_pd(rsq00,rinv00);
623 /* Compute parameters for interactions between i and j atoms */
624 qq00 = _mm_mul_pd(iq0,jq0);
625 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
626 c6grid_00 = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
628 /* EWALD ELECTROSTATICS */
630 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
631 ewrt = _mm_mul_pd(r00,ewtabscale);
632 ewitab = _mm_cvttpd_epi32(ewrt);
634 eweps = _mm_frcz_pd(ewrt);
636 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
638 twoeweps = _mm_add_pd(eweps,eweps);
639 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
640 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
641 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
643 /* Analytical LJ-PME */
644 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
645 ewcljrsq = _mm_mul_pd(ewclj2,rsq00);
646 ewclj6 = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
647 exponent = gmx_simd_exp_d(ewcljrsq);
648 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
649 poly = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
650 /* f6A = 6 * C6grid * (1 - poly) */
651 f6A = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
652 /* f6B = C6grid * exponent * beta^6 */
653 f6B = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
654 /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
655 fvdw = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
657 fscal = _mm_add_pd(felec,fvdw);
659 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
661 /* Update vectorial force */
662 fix0 = _mm_macc_pd(dx00,fscal,fix0);
663 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
664 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
666 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
667 _mm_mul_pd(dx00,fscal),
668 _mm_mul_pd(dy00,fscal),
669 _mm_mul_pd(dz00,fscal));
671 /* Inner loop uses 60 flops */
674 /* End of innermost loop */
676 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
677 f+i_coord_offset,fshift+i_shift_offset);
679 /* Increment number of inner iterations */
680 inneriter += j_index_end - j_index_start;
682 /* Outer loop uses 7 flops */
685 /* Increment number of outer iterations */
688 /* Update outer/inner flops */
690 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*60);