2 * Note: this file was generated by the Gromacs sse2_double kernel generator.
4 * This source code is part of
8 * Copyright (c) 2001-2012, The GROMACS Development Team
10 * Gromacs is a library for molecular simulation and trajectory analysis,
11 * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
12 * a full list of developers and information, check out http://www.gromacs.org
14 * This program is free software; you can redistribute it and/or modify it under
15 * the terms of the GNU Lesser General Public License as published by the Free
16 * Software Foundation; either version 2 of the License, or (at your option) any
19 * To help fund GROMACS development, we humbly ask that you cite
20 * the papers people have written on it - you can find them on the website.
28 #include "../nb_kernel.h"
29 #include "types/simple.h"
33 #include "gmx_math_x86_sse2_double.h"
34 #include "kernelutil_x86_sse2_double.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_VF_sse2_double
38 * Electrostatics interaction: Ewald
39 * VdW interaction: LennardJones
40 * Geometry: Particle-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_VF_sse2_double
45 (t_nblist * gmx_restrict nlist,
46 rvec * gmx_restrict xx,
47 rvec * gmx_restrict ff,
48 t_forcerec * gmx_restrict fr,
49 t_mdatoms * gmx_restrict mdatoms,
50 nb_kernel_data_t * gmx_restrict kernel_data,
51 t_nrnb * gmx_restrict nrnb)
53 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
54 * just 0 for non-waters.
55 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
56 * jnr indices corresponding to data put in the four positions in the SIMD register.
58 int i_shift_offset,i_coord_offset,outeriter,inneriter;
59 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
61 int j_coord_offsetA,j_coord_offsetB;
62 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
64 real *shiftvec,*fshift,*x,*f;
65 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
67 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
68 int vdwjidx0A,vdwjidx0B;
69 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
70 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
71 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
74 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
77 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
78 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
80 __m128d ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
82 __m128d dummy_mask,cutoff_mask;
83 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
84 __m128d one = _mm_set1_pd(1.0);
85 __m128d two = _mm_set1_pd(2.0);
91 jindex = nlist->jindex;
93 shiftidx = nlist->shift;
95 shiftvec = fr->shift_vec[0];
96 fshift = fr->fshift[0];
97 facel = _mm_set1_pd(fr->epsfac);
98 charge = mdatoms->chargeA;
101 vdwtype = mdatoms->typeA;
103 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
104 ewtab = fr->ic->tabq_coul_FDV0;
105 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
106 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
108 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
109 rcutoff_scalar = fr->rcoulomb;
110 rcutoff = _mm_set1_pd(rcutoff_scalar);
111 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
113 sh_vdw_invrcut6 = _mm_set1_pd(fr->ic->sh_invrc6);
114 rvdw = _mm_set1_pd(fr->rvdw);
116 /* Avoid stupid compiler warnings */
124 /* Start outer loop over neighborlists */
125 for(iidx=0; iidx<nri; iidx++)
127 /* Load shift vector for this list */
128 i_shift_offset = DIM*shiftidx[iidx];
130 /* Load limits for loop over neighbors */
131 j_index_start = jindex[iidx];
132 j_index_end = jindex[iidx+1];
134 /* Get outer coordinate index */
136 i_coord_offset = DIM*inr;
138 /* Load i particle coords and add shift vector */
139 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
141 fix0 = _mm_setzero_pd();
142 fiy0 = _mm_setzero_pd();
143 fiz0 = _mm_setzero_pd();
145 /* Load parameters for i particles */
146 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
147 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
149 /* Reset potential sums */
150 velecsum = _mm_setzero_pd();
151 vvdwsum = _mm_setzero_pd();
153 /* Start inner kernel loop */
154 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
157 /* Get j neighbor index, and coordinate index */
160 j_coord_offsetA = DIM*jnrA;
161 j_coord_offsetB = DIM*jnrB;
163 /* load j atom coordinates */
164 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
167 /* Calculate displacement vector */
168 dx00 = _mm_sub_pd(ix0,jx0);
169 dy00 = _mm_sub_pd(iy0,jy0);
170 dz00 = _mm_sub_pd(iz0,jz0);
172 /* Calculate squared distance and things based on it */
173 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
175 rinv00 = gmx_mm_invsqrt_pd(rsq00);
177 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
179 /* Load parameters for j particles */
180 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
181 vdwjidx0A = 2*vdwtype[jnrA+0];
182 vdwjidx0B = 2*vdwtype[jnrB+0];
184 /**************************
185 * CALCULATE INTERACTIONS *
186 **************************/
188 if (gmx_mm_any_lt(rsq00,rcutoff2))
191 r00 = _mm_mul_pd(rsq00,rinv00);
193 /* Compute parameters for interactions between i and j atoms */
194 qq00 = _mm_mul_pd(iq0,jq0);
195 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
196 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
198 /* EWALD ELECTROSTATICS */
200 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
201 ewrt = _mm_mul_pd(r00,ewtabscale);
202 ewitab = _mm_cvttpd_epi32(ewrt);
203 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
204 ewitab = _mm_slli_epi32(ewitab,2);
205 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
206 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
207 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
208 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
209 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
210 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
211 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
212 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
213 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
214 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
216 /* LENNARD-JONES DISPERSION/REPULSION */
218 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
219 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
220 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
221 vvdw = _mm_sub_pd(_mm_mul_pd( _mm_sub_pd(vvdw12 , _mm_mul_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
222 _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
223 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
225 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
227 /* Update potential sum for this i atom from the interaction with this j atom. */
228 velec = _mm_and_pd(velec,cutoff_mask);
229 velecsum = _mm_add_pd(velecsum,velec);
230 vvdw = _mm_and_pd(vvdw,cutoff_mask);
231 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
233 fscal = _mm_add_pd(felec,fvdw);
235 fscal = _mm_and_pd(fscal,cutoff_mask);
237 /* Calculate temporary vectorial force */
238 tx = _mm_mul_pd(fscal,dx00);
239 ty = _mm_mul_pd(fscal,dy00);
240 tz = _mm_mul_pd(fscal,dz00);
242 /* Update vectorial force */
243 fix0 = _mm_add_pd(fix0,tx);
244 fiy0 = _mm_add_pd(fiy0,ty);
245 fiz0 = _mm_add_pd(fiz0,tz);
247 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
251 /* Inner loop uses 64 flops */
258 j_coord_offsetA = DIM*jnrA;
260 /* load j atom coordinates */
261 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
264 /* Calculate displacement vector */
265 dx00 = _mm_sub_pd(ix0,jx0);
266 dy00 = _mm_sub_pd(iy0,jy0);
267 dz00 = _mm_sub_pd(iz0,jz0);
269 /* Calculate squared distance and things based on it */
270 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
272 rinv00 = gmx_mm_invsqrt_pd(rsq00);
274 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
276 /* Load parameters for j particles */
277 jq0 = _mm_load_sd(charge+jnrA+0);
278 vdwjidx0A = 2*vdwtype[jnrA+0];
280 /**************************
281 * CALCULATE INTERACTIONS *
282 **************************/
284 if (gmx_mm_any_lt(rsq00,rcutoff2))
287 r00 = _mm_mul_pd(rsq00,rinv00);
289 /* Compute parameters for interactions between i and j atoms */
290 qq00 = _mm_mul_pd(iq0,jq0);
291 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
293 /* EWALD ELECTROSTATICS */
295 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
296 ewrt = _mm_mul_pd(r00,ewtabscale);
297 ewitab = _mm_cvttpd_epi32(ewrt);
298 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
299 ewitab = _mm_slli_epi32(ewitab,2);
300 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
301 ewtabD = _mm_setzero_pd();
302 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
303 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
304 ewtabFn = _mm_setzero_pd();
305 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
306 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
307 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
308 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
309 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
311 /* LENNARD-JONES DISPERSION/REPULSION */
313 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
314 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
315 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
316 vvdw = _mm_sub_pd(_mm_mul_pd( _mm_sub_pd(vvdw12 , _mm_mul_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
317 _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
318 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
320 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
322 /* Update potential sum for this i atom from the interaction with this j atom. */
323 velec = _mm_and_pd(velec,cutoff_mask);
324 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
325 velecsum = _mm_add_pd(velecsum,velec);
326 vvdw = _mm_and_pd(vvdw,cutoff_mask);
327 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
328 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
330 fscal = _mm_add_pd(felec,fvdw);
332 fscal = _mm_and_pd(fscal,cutoff_mask);
334 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
336 /* Calculate temporary vectorial force */
337 tx = _mm_mul_pd(fscal,dx00);
338 ty = _mm_mul_pd(fscal,dy00);
339 tz = _mm_mul_pd(fscal,dz00);
341 /* Update vectorial force */
342 fix0 = _mm_add_pd(fix0,tx);
343 fiy0 = _mm_add_pd(fiy0,ty);
344 fiz0 = _mm_add_pd(fiz0,tz);
346 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
350 /* Inner loop uses 64 flops */
353 /* End of innermost loop */
355 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
356 f+i_coord_offset,fshift+i_shift_offset);
359 /* Update potential energies */
360 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
361 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
363 /* Increment number of inner iterations */
364 inneriter += j_index_end - j_index_start;
366 /* Outer loop uses 9 flops */
369 /* Increment number of outer iterations */
372 /* Update outer/inner flops */
374 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*64);
377 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_F_sse2_double
378 * Electrostatics interaction: Ewald
379 * VdW interaction: LennardJones
380 * Geometry: Particle-Particle
381 * Calculate force/pot: Force
384 nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_F_sse2_double
385 (t_nblist * gmx_restrict nlist,
386 rvec * gmx_restrict xx,
387 rvec * gmx_restrict ff,
388 t_forcerec * gmx_restrict fr,
389 t_mdatoms * gmx_restrict mdatoms,
390 nb_kernel_data_t * gmx_restrict kernel_data,
391 t_nrnb * gmx_restrict nrnb)
393 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
394 * just 0 for non-waters.
395 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
396 * jnr indices corresponding to data put in the four positions in the SIMD register.
398 int i_shift_offset,i_coord_offset,outeriter,inneriter;
399 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
401 int j_coord_offsetA,j_coord_offsetB;
402 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
404 real *shiftvec,*fshift,*x,*f;
405 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
407 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
408 int vdwjidx0A,vdwjidx0B;
409 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
410 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
411 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
414 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
417 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
418 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
420 __m128d ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
422 __m128d dummy_mask,cutoff_mask;
423 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
424 __m128d one = _mm_set1_pd(1.0);
425 __m128d two = _mm_set1_pd(2.0);
431 jindex = nlist->jindex;
433 shiftidx = nlist->shift;
435 shiftvec = fr->shift_vec[0];
436 fshift = fr->fshift[0];
437 facel = _mm_set1_pd(fr->epsfac);
438 charge = mdatoms->chargeA;
439 nvdwtype = fr->ntype;
441 vdwtype = mdatoms->typeA;
443 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
444 ewtab = fr->ic->tabq_coul_F;
445 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
446 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
448 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
449 rcutoff_scalar = fr->rcoulomb;
450 rcutoff = _mm_set1_pd(rcutoff_scalar);
451 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
453 sh_vdw_invrcut6 = _mm_set1_pd(fr->ic->sh_invrc6);
454 rvdw = _mm_set1_pd(fr->rvdw);
456 /* Avoid stupid compiler warnings */
464 /* Start outer loop over neighborlists */
465 for(iidx=0; iidx<nri; iidx++)
467 /* Load shift vector for this list */
468 i_shift_offset = DIM*shiftidx[iidx];
470 /* Load limits for loop over neighbors */
471 j_index_start = jindex[iidx];
472 j_index_end = jindex[iidx+1];
474 /* Get outer coordinate index */
476 i_coord_offset = DIM*inr;
478 /* Load i particle coords and add shift vector */
479 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
481 fix0 = _mm_setzero_pd();
482 fiy0 = _mm_setzero_pd();
483 fiz0 = _mm_setzero_pd();
485 /* Load parameters for i particles */
486 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
487 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
489 /* Start inner kernel loop */
490 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
493 /* Get j neighbor index, and coordinate index */
496 j_coord_offsetA = DIM*jnrA;
497 j_coord_offsetB = DIM*jnrB;
499 /* load j atom coordinates */
500 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
503 /* Calculate displacement vector */
504 dx00 = _mm_sub_pd(ix0,jx0);
505 dy00 = _mm_sub_pd(iy0,jy0);
506 dz00 = _mm_sub_pd(iz0,jz0);
508 /* Calculate squared distance and things based on it */
509 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
511 rinv00 = gmx_mm_invsqrt_pd(rsq00);
513 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
515 /* Load parameters for j particles */
516 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
517 vdwjidx0A = 2*vdwtype[jnrA+0];
518 vdwjidx0B = 2*vdwtype[jnrB+0];
520 /**************************
521 * CALCULATE INTERACTIONS *
522 **************************/
524 if (gmx_mm_any_lt(rsq00,rcutoff2))
527 r00 = _mm_mul_pd(rsq00,rinv00);
529 /* Compute parameters for interactions between i and j atoms */
530 qq00 = _mm_mul_pd(iq0,jq0);
531 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
532 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
534 /* EWALD ELECTROSTATICS */
536 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
537 ewrt = _mm_mul_pd(r00,ewtabscale);
538 ewitab = _mm_cvttpd_epi32(ewrt);
539 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
540 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
542 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
543 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
545 /* LENNARD-JONES DISPERSION/REPULSION */
547 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
548 fvdw = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
550 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
552 fscal = _mm_add_pd(felec,fvdw);
554 fscal = _mm_and_pd(fscal,cutoff_mask);
556 /* Calculate temporary vectorial force */
557 tx = _mm_mul_pd(fscal,dx00);
558 ty = _mm_mul_pd(fscal,dy00);
559 tz = _mm_mul_pd(fscal,dz00);
561 /* Update vectorial force */
562 fix0 = _mm_add_pd(fix0,tx);
563 fiy0 = _mm_add_pd(fiy0,ty);
564 fiz0 = _mm_add_pd(fiz0,tz);
566 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
570 /* Inner loop uses 46 flops */
577 j_coord_offsetA = DIM*jnrA;
579 /* load j atom coordinates */
580 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
583 /* Calculate displacement vector */
584 dx00 = _mm_sub_pd(ix0,jx0);
585 dy00 = _mm_sub_pd(iy0,jy0);
586 dz00 = _mm_sub_pd(iz0,jz0);
588 /* Calculate squared distance and things based on it */
589 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
591 rinv00 = gmx_mm_invsqrt_pd(rsq00);
593 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
595 /* Load parameters for j particles */
596 jq0 = _mm_load_sd(charge+jnrA+0);
597 vdwjidx0A = 2*vdwtype[jnrA+0];
599 /**************************
600 * CALCULATE INTERACTIONS *
601 **************************/
603 if (gmx_mm_any_lt(rsq00,rcutoff2))
606 r00 = _mm_mul_pd(rsq00,rinv00);
608 /* Compute parameters for interactions between i and j atoms */
609 qq00 = _mm_mul_pd(iq0,jq0);
610 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
612 /* EWALD ELECTROSTATICS */
614 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
615 ewrt = _mm_mul_pd(r00,ewtabscale);
616 ewitab = _mm_cvttpd_epi32(ewrt);
617 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
618 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
619 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
620 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
622 /* LENNARD-JONES DISPERSION/REPULSION */
624 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
625 fvdw = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
627 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
629 fscal = _mm_add_pd(felec,fvdw);
631 fscal = _mm_and_pd(fscal,cutoff_mask);
633 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
635 /* Calculate temporary vectorial force */
636 tx = _mm_mul_pd(fscal,dx00);
637 ty = _mm_mul_pd(fscal,dy00);
638 tz = _mm_mul_pd(fscal,dz00);
640 /* Update vectorial force */
641 fix0 = _mm_add_pd(fix0,tx);
642 fiy0 = _mm_add_pd(fiy0,ty);
643 fiz0 = _mm_add_pd(fiz0,tz);
645 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
649 /* Inner loop uses 46 flops */
652 /* End of innermost loop */
654 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
655 f+i_coord_offset,fshift+i_shift_offset);
657 /* Increment number of inner iterations */
658 inneriter += j_index_end - j_index_start;
660 /* Outer loop uses 7 flops */
663 /* Increment number of outer iterations */
666 /* Update outer/inner flops */
668 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*46);