2 * Note: this file was generated by the Gromacs avx_128_fma_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_avx_128_fma_double.h"
34 #include "kernelutil_x86_avx_128_fma_double.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_VF_avx_128_fma_double
38 * Electrostatics interaction: CubicSplineTable
39 * VdW interaction: LennardJones
40 * Geometry: Particle-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_VF_avx_128_fma_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 __m128i ifour = _mm_set1_epi32(4);
81 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
83 __m128d dummy_mask,cutoff_mask;
84 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
85 __m128d one = _mm_set1_pd(1.0);
86 __m128d two = _mm_set1_pd(2.0);
92 jindex = nlist->jindex;
94 shiftidx = nlist->shift;
96 shiftvec = fr->shift_vec[0];
97 fshift = fr->fshift[0];
98 facel = _mm_set1_pd(fr->epsfac);
99 charge = mdatoms->chargeA;
100 nvdwtype = fr->ntype;
102 vdwtype = mdatoms->typeA;
104 vftab = kernel_data->table_elec->data;
105 vftabscale = _mm_set1_pd(kernel_data->table_elec->scale);
107 /* Avoid stupid compiler warnings */
115 /* Start outer loop over neighborlists */
116 for(iidx=0; iidx<nri; iidx++)
118 /* Load shift vector for this list */
119 i_shift_offset = DIM*shiftidx[iidx];
121 /* Load limits for loop over neighbors */
122 j_index_start = jindex[iidx];
123 j_index_end = jindex[iidx+1];
125 /* Get outer coordinate index */
127 i_coord_offset = DIM*inr;
129 /* Load i particle coords and add shift vector */
130 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
132 fix0 = _mm_setzero_pd();
133 fiy0 = _mm_setzero_pd();
134 fiz0 = _mm_setzero_pd();
136 /* Load parameters for i particles */
137 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
138 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
140 /* Reset potential sums */
141 velecsum = _mm_setzero_pd();
142 vvdwsum = _mm_setzero_pd();
144 /* Start inner kernel loop */
145 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
148 /* Get j neighbor index, and coordinate index */
151 j_coord_offsetA = DIM*jnrA;
152 j_coord_offsetB = DIM*jnrB;
154 /* load j atom coordinates */
155 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
158 /* Calculate displacement vector */
159 dx00 = _mm_sub_pd(ix0,jx0);
160 dy00 = _mm_sub_pd(iy0,jy0);
161 dz00 = _mm_sub_pd(iz0,jz0);
163 /* Calculate squared distance and things based on it */
164 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
166 rinv00 = gmx_mm_invsqrt_pd(rsq00);
168 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
170 /* Load parameters for j particles */
171 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
172 vdwjidx0A = 2*vdwtype[jnrA+0];
173 vdwjidx0B = 2*vdwtype[jnrB+0];
175 /**************************
176 * CALCULATE INTERACTIONS *
177 **************************/
179 r00 = _mm_mul_pd(rsq00,rinv00);
181 /* Compute parameters for interactions between i and j atoms */
182 qq00 = _mm_mul_pd(iq0,jq0);
183 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
184 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
186 /* Calculate table index by multiplying r with table scale and truncate to integer */
187 rt = _mm_mul_pd(r00,vftabscale);
188 vfitab = _mm_cvttpd_epi32(rt);
190 vfeps = _mm_frcz_pd(rt);
192 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
194 twovfeps = _mm_add_pd(vfeps,vfeps);
195 vfitab = _mm_slli_epi32(vfitab,2);
197 /* CUBIC SPLINE TABLE ELECTROSTATICS */
198 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
199 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
200 GMX_MM_TRANSPOSE2_PD(Y,F);
201 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
202 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
203 GMX_MM_TRANSPOSE2_PD(G,H);
204 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
205 VV = _mm_macc_pd(vfeps,Fp,Y);
206 velec = _mm_mul_pd(qq00,VV);
207 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
208 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
210 /* LENNARD-JONES DISPERSION/REPULSION */
212 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
213 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
214 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
215 vvdw = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
216 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
218 /* Update potential sum for this i atom from the interaction with this j atom. */
219 velecsum = _mm_add_pd(velecsum,velec);
220 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
222 fscal = _mm_add_pd(felec,fvdw);
224 /* Update vectorial force */
225 fix0 = _mm_macc_pd(dx00,fscal,fix0);
226 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
227 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
229 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
230 _mm_mul_pd(dx00,fscal),
231 _mm_mul_pd(dy00,fscal),
232 _mm_mul_pd(dz00,fscal));
234 /* Inner loop uses 59 flops */
241 j_coord_offsetA = DIM*jnrA;
243 /* load j atom coordinates */
244 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
247 /* Calculate displacement vector */
248 dx00 = _mm_sub_pd(ix0,jx0);
249 dy00 = _mm_sub_pd(iy0,jy0);
250 dz00 = _mm_sub_pd(iz0,jz0);
252 /* Calculate squared distance and things based on it */
253 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
255 rinv00 = gmx_mm_invsqrt_pd(rsq00);
257 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
259 /* Load parameters for j particles */
260 jq0 = _mm_load_sd(charge+jnrA+0);
261 vdwjidx0A = 2*vdwtype[jnrA+0];
263 /**************************
264 * CALCULATE INTERACTIONS *
265 **************************/
267 r00 = _mm_mul_pd(rsq00,rinv00);
269 /* Compute parameters for interactions between i and j atoms */
270 qq00 = _mm_mul_pd(iq0,jq0);
271 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
273 /* Calculate table index by multiplying r with table scale and truncate to integer */
274 rt = _mm_mul_pd(r00,vftabscale);
275 vfitab = _mm_cvttpd_epi32(rt);
277 vfeps = _mm_frcz_pd(rt);
279 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
281 twovfeps = _mm_add_pd(vfeps,vfeps);
282 vfitab = _mm_slli_epi32(vfitab,2);
284 /* CUBIC SPLINE TABLE ELECTROSTATICS */
285 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
286 F = _mm_setzero_pd();
287 GMX_MM_TRANSPOSE2_PD(Y,F);
288 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
289 H = _mm_setzero_pd();
290 GMX_MM_TRANSPOSE2_PD(G,H);
291 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
292 VV = _mm_macc_pd(vfeps,Fp,Y);
293 velec = _mm_mul_pd(qq00,VV);
294 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
295 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
297 /* LENNARD-JONES DISPERSION/REPULSION */
299 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
300 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
301 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
302 vvdw = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
303 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
305 /* Update potential sum for this i atom from the interaction with this j atom. */
306 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
307 velecsum = _mm_add_pd(velecsum,velec);
308 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
309 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
311 fscal = _mm_add_pd(felec,fvdw);
313 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
315 /* Update vectorial force */
316 fix0 = _mm_macc_pd(dx00,fscal,fix0);
317 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
318 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
320 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
321 _mm_mul_pd(dx00,fscal),
322 _mm_mul_pd(dy00,fscal),
323 _mm_mul_pd(dz00,fscal));
325 /* Inner loop uses 59 flops */
328 /* End of innermost loop */
330 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
331 f+i_coord_offset,fshift+i_shift_offset);
334 /* Update potential energies */
335 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
336 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
338 /* Increment number of inner iterations */
339 inneriter += j_index_end - j_index_start;
341 /* Outer loop uses 9 flops */
344 /* Increment number of outer iterations */
347 /* Update outer/inner flops */
349 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*59);
352 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_F_avx_128_fma_double
353 * Electrostatics interaction: CubicSplineTable
354 * VdW interaction: LennardJones
355 * Geometry: Particle-Particle
356 * Calculate force/pot: Force
359 nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_F_avx_128_fma_double
360 (t_nblist * gmx_restrict nlist,
361 rvec * gmx_restrict xx,
362 rvec * gmx_restrict ff,
363 t_forcerec * gmx_restrict fr,
364 t_mdatoms * gmx_restrict mdatoms,
365 nb_kernel_data_t * gmx_restrict kernel_data,
366 t_nrnb * gmx_restrict nrnb)
368 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
369 * just 0 for non-waters.
370 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
371 * jnr indices corresponding to data put in the four positions in the SIMD register.
373 int i_shift_offset,i_coord_offset,outeriter,inneriter;
374 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
376 int j_coord_offsetA,j_coord_offsetB;
377 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
379 real *shiftvec,*fshift,*x,*f;
380 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
382 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
383 int vdwjidx0A,vdwjidx0B;
384 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
385 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
386 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
389 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
392 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
393 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
395 __m128i ifour = _mm_set1_epi32(4);
396 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
398 __m128d dummy_mask,cutoff_mask;
399 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
400 __m128d one = _mm_set1_pd(1.0);
401 __m128d two = _mm_set1_pd(2.0);
407 jindex = nlist->jindex;
409 shiftidx = nlist->shift;
411 shiftvec = fr->shift_vec[0];
412 fshift = fr->fshift[0];
413 facel = _mm_set1_pd(fr->epsfac);
414 charge = mdatoms->chargeA;
415 nvdwtype = fr->ntype;
417 vdwtype = mdatoms->typeA;
419 vftab = kernel_data->table_elec->data;
420 vftabscale = _mm_set1_pd(kernel_data->table_elec->scale);
422 /* Avoid stupid compiler warnings */
430 /* Start outer loop over neighborlists */
431 for(iidx=0; iidx<nri; iidx++)
433 /* Load shift vector for this list */
434 i_shift_offset = DIM*shiftidx[iidx];
436 /* Load limits for loop over neighbors */
437 j_index_start = jindex[iidx];
438 j_index_end = jindex[iidx+1];
440 /* Get outer coordinate index */
442 i_coord_offset = DIM*inr;
444 /* Load i particle coords and add shift vector */
445 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
447 fix0 = _mm_setzero_pd();
448 fiy0 = _mm_setzero_pd();
449 fiz0 = _mm_setzero_pd();
451 /* Load parameters for i particles */
452 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
453 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
455 /* Start inner kernel loop */
456 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
459 /* Get j neighbor index, and coordinate index */
462 j_coord_offsetA = DIM*jnrA;
463 j_coord_offsetB = DIM*jnrB;
465 /* load j atom coordinates */
466 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
469 /* Calculate displacement vector */
470 dx00 = _mm_sub_pd(ix0,jx0);
471 dy00 = _mm_sub_pd(iy0,jy0);
472 dz00 = _mm_sub_pd(iz0,jz0);
474 /* Calculate squared distance and things based on it */
475 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
477 rinv00 = gmx_mm_invsqrt_pd(rsq00);
479 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
481 /* Load parameters for j particles */
482 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
483 vdwjidx0A = 2*vdwtype[jnrA+0];
484 vdwjidx0B = 2*vdwtype[jnrB+0];
486 /**************************
487 * CALCULATE INTERACTIONS *
488 **************************/
490 r00 = _mm_mul_pd(rsq00,rinv00);
492 /* Compute parameters for interactions between i and j atoms */
493 qq00 = _mm_mul_pd(iq0,jq0);
494 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
495 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
497 /* Calculate table index by multiplying r with table scale and truncate to integer */
498 rt = _mm_mul_pd(r00,vftabscale);
499 vfitab = _mm_cvttpd_epi32(rt);
501 vfeps = _mm_frcz_pd(rt);
503 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
505 twovfeps = _mm_add_pd(vfeps,vfeps);
506 vfitab = _mm_slli_epi32(vfitab,2);
508 /* CUBIC SPLINE TABLE ELECTROSTATICS */
509 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
510 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
511 GMX_MM_TRANSPOSE2_PD(Y,F);
512 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
513 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
514 GMX_MM_TRANSPOSE2_PD(G,H);
515 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
516 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
517 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
519 /* LENNARD-JONES DISPERSION/REPULSION */
521 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
522 fvdw = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
524 fscal = _mm_add_pd(felec,fvdw);
526 /* Update vectorial force */
527 fix0 = _mm_macc_pd(dx00,fscal,fix0);
528 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
529 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
531 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
532 _mm_mul_pd(dx00,fscal),
533 _mm_mul_pd(dy00,fscal),
534 _mm_mul_pd(dz00,fscal));
536 /* Inner loop uses 50 flops */
543 j_coord_offsetA = DIM*jnrA;
545 /* load j atom coordinates */
546 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
549 /* Calculate displacement vector */
550 dx00 = _mm_sub_pd(ix0,jx0);
551 dy00 = _mm_sub_pd(iy0,jy0);
552 dz00 = _mm_sub_pd(iz0,jz0);
554 /* Calculate squared distance and things based on it */
555 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
557 rinv00 = gmx_mm_invsqrt_pd(rsq00);
559 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
561 /* Load parameters for j particles */
562 jq0 = _mm_load_sd(charge+jnrA+0);
563 vdwjidx0A = 2*vdwtype[jnrA+0];
565 /**************************
566 * CALCULATE INTERACTIONS *
567 **************************/
569 r00 = _mm_mul_pd(rsq00,rinv00);
571 /* Compute parameters for interactions between i and j atoms */
572 qq00 = _mm_mul_pd(iq0,jq0);
573 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
575 /* Calculate table index by multiplying r with table scale and truncate to integer */
576 rt = _mm_mul_pd(r00,vftabscale);
577 vfitab = _mm_cvttpd_epi32(rt);
579 vfeps = _mm_frcz_pd(rt);
581 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
583 twovfeps = _mm_add_pd(vfeps,vfeps);
584 vfitab = _mm_slli_epi32(vfitab,2);
586 /* CUBIC SPLINE TABLE ELECTROSTATICS */
587 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
588 F = _mm_setzero_pd();
589 GMX_MM_TRANSPOSE2_PD(Y,F);
590 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
591 H = _mm_setzero_pd();
592 GMX_MM_TRANSPOSE2_PD(G,H);
593 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
594 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
595 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
597 /* LENNARD-JONES DISPERSION/REPULSION */
599 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
600 fvdw = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
602 fscal = _mm_add_pd(felec,fvdw);
604 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
606 /* Update vectorial force */
607 fix0 = _mm_macc_pd(dx00,fscal,fix0);
608 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
609 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
611 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
612 _mm_mul_pd(dx00,fscal),
613 _mm_mul_pd(dy00,fscal),
614 _mm_mul_pd(dz00,fscal));
616 /* Inner loop uses 50 flops */
619 /* End of innermost loop */
621 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
622 f+i_coord_offset,fshift+i_shift_offset);
624 /* Increment number of inner iterations */
625 inneriter += j_index_end - j_index_start;
627 /* Outer loop uses 7 flops */
630 /* Increment number of outer iterations */
633 /* Update outer/inner flops */
635 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*50);