2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
36 * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
42 #include "../nb_kernel.h"
43 #include "gromacs/legacyheaders/types/simple.h"
44 #include "gromacs/math/vec.h"
45 #include "gromacs/legacyheaders/nrnb.h"
47 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
48 #include "kernelutil_x86_avx_128_fma_double.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_VF_avx_128_fma_double
52 * Electrostatics interaction: CubicSplineTable
53 * VdW interaction: LennardJones
54 * Geometry: Particle-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecCSTab_VdwLJ_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);
94 __m128i ifour = _mm_set1_epi32(4);
95 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
97 __m128d dummy_mask,cutoff_mask;
98 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
99 __m128d one = _mm_set1_pd(1.0);
100 __m128d two = _mm_set1_pd(2.0);
106 jindex = nlist->jindex;
108 shiftidx = nlist->shift;
110 shiftvec = fr->shift_vec[0];
111 fshift = fr->fshift[0];
112 facel = _mm_set1_pd(fr->epsfac);
113 charge = mdatoms->chargeA;
114 nvdwtype = fr->ntype;
116 vdwtype = mdatoms->typeA;
118 vftab = kernel_data->table_elec->data;
119 vftabscale = _mm_set1_pd(kernel_data->table_elec->scale);
121 /* Avoid stupid compiler warnings */
129 /* Start outer loop over neighborlists */
130 for(iidx=0; iidx<nri; iidx++)
132 /* Load shift vector for this list */
133 i_shift_offset = DIM*shiftidx[iidx];
135 /* Load limits for loop over neighbors */
136 j_index_start = jindex[iidx];
137 j_index_end = jindex[iidx+1];
139 /* Get outer coordinate index */
141 i_coord_offset = DIM*inr;
143 /* Load i particle coords and add shift vector */
144 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
146 fix0 = _mm_setzero_pd();
147 fiy0 = _mm_setzero_pd();
148 fiz0 = _mm_setzero_pd();
150 /* Load parameters for i particles */
151 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
152 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
154 /* Reset potential sums */
155 velecsum = _mm_setzero_pd();
156 vvdwsum = _mm_setzero_pd();
158 /* Start inner kernel loop */
159 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
162 /* Get j neighbor index, and coordinate index */
165 j_coord_offsetA = DIM*jnrA;
166 j_coord_offsetB = DIM*jnrB;
168 /* load j atom coordinates */
169 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
172 /* Calculate displacement vector */
173 dx00 = _mm_sub_pd(ix0,jx0);
174 dy00 = _mm_sub_pd(iy0,jy0);
175 dz00 = _mm_sub_pd(iz0,jz0);
177 /* Calculate squared distance and things based on it */
178 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
180 rinv00 = gmx_mm_invsqrt_pd(rsq00);
182 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
184 /* Load parameters for j particles */
185 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
186 vdwjidx0A = 2*vdwtype[jnrA+0];
187 vdwjidx0B = 2*vdwtype[jnrB+0];
189 /**************************
190 * CALCULATE INTERACTIONS *
191 **************************/
193 r00 = _mm_mul_pd(rsq00,rinv00);
195 /* Compute parameters for interactions between i and j atoms */
196 qq00 = _mm_mul_pd(iq0,jq0);
197 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
198 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
200 /* Calculate table index by multiplying r with table scale and truncate to integer */
201 rt = _mm_mul_pd(r00,vftabscale);
202 vfitab = _mm_cvttpd_epi32(rt);
204 vfeps = _mm_frcz_pd(rt);
206 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
208 twovfeps = _mm_add_pd(vfeps,vfeps);
209 vfitab = _mm_slli_epi32(vfitab,2);
211 /* CUBIC SPLINE TABLE ELECTROSTATICS */
212 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
213 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
214 GMX_MM_TRANSPOSE2_PD(Y,F);
215 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
216 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
217 GMX_MM_TRANSPOSE2_PD(G,H);
218 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
219 VV = _mm_macc_pd(vfeps,Fp,Y);
220 velec = _mm_mul_pd(qq00,VV);
221 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
222 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
224 /* LENNARD-JONES DISPERSION/REPULSION */
226 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
227 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
228 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
229 vvdw = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
230 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
232 /* Update potential sum for this i atom from the interaction with this j atom. */
233 velecsum = _mm_add_pd(velecsum,velec);
234 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
236 fscal = _mm_add_pd(felec,fvdw);
238 /* Update vectorial force */
239 fix0 = _mm_macc_pd(dx00,fscal,fix0);
240 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
241 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
243 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
244 _mm_mul_pd(dx00,fscal),
245 _mm_mul_pd(dy00,fscal),
246 _mm_mul_pd(dz00,fscal));
248 /* Inner loop uses 59 flops */
255 j_coord_offsetA = DIM*jnrA;
257 /* load j atom coordinates */
258 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
261 /* Calculate displacement vector */
262 dx00 = _mm_sub_pd(ix0,jx0);
263 dy00 = _mm_sub_pd(iy0,jy0);
264 dz00 = _mm_sub_pd(iz0,jz0);
266 /* Calculate squared distance and things based on it */
267 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
269 rinv00 = gmx_mm_invsqrt_pd(rsq00);
271 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
273 /* Load parameters for j particles */
274 jq0 = _mm_load_sd(charge+jnrA+0);
275 vdwjidx0A = 2*vdwtype[jnrA+0];
277 /**************************
278 * CALCULATE INTERACTIONS *
279 **************************/
281 r00 = _mm_mul_pd(rsq00,rinv00);
283 /* Compute parameters for interactions between i and j atoms */
284 qq00 = _mm_mul_pd(iq0,jq0);
285 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
287 /* Calculate table index by multiplying r with table scale and truncate to integer */
288 rt = _mm_mul_pd(r00,vftabscale);
289 vfitab = _mm_cvttpd_epi32(rt);
291 vfeps = _mm_frcz_pd(rt);
293 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
295 twovfeps = _mm_add_pd(vfeps,vfeps);
296 vfitab = _mm_slli_epi32(vfitab,2);
298 /* CUBIC SPLINE TABLE ELECTROSTATICS */
299 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
300 F = _mm_setzero_pd();
301 GMX_MM_TRANSPOSE2_PD(Y,F);
302 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
303 H = _mm_setzero_pd();
304 GMX_MM_TRANSPOSE2_PD(G,H);
305 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
306 VV = _mm_macc_pd(vfeps,Fp,Y);
307 velec = _mm_mul_pd(qq00,VV);
308 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
309 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
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_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
317 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
319 /* Update potential sum for this i atom from the interaction with this j atom. */
320 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
321 velecsum = _mm_add_pd(velecsum,velec);
322 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
323 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
325 fscal = _mm_add_pd(felec,fvdw);
327 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
329 /* Update vectorial force */
330 fix0 = _mm_macc_pd(dx00,fscal,fix0);
331 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
332 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
334 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
335 _mm_mul_pd(dx00,fscal),
336 _mm_mul_pd(dy00,fscal),
337 _mm_mul_pd(dz00,fscal));
339 /* Inner loop uses 59 flops */
342 /* End of innermost loop */
344 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
345 f+i_coord_offset,fshift+i_shift_offset);
348 /* Update potential energies */
349 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
350 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
352 /* Increment number of inner iterations */
353 inneriter += j_index_end - j_index_start;
355 /* Outer loop uses 9 flops */
358 /* Increment number of outer iterations */
361 /* Update outer/inner flops */
363 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*59);
366 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_F_avx_128_fma_double
367 * Electrostatics interaction: CubicSplineTable
368 * VdW interaction: LennardJones
369 * Geometry: Particle-Particle
370 * Calculate force/pot: Force
373 nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_F_avx_128_fma_double
374 (t_nblist * gmx_restrict nlist,
375 rvec * gmx_restrict xx,
376 rvec * gmx_restrict ff,
377 t_forcerec * gmx_restrict fr,
378 t_mdatoms * gmx_restrict mdatoms,
379 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
380 t_nrnb * gmx_restrict nrnb)
382 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
383 * just 0 for non-waters.
384 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
385 * jnr indices corresponding to data put in the four positions in the SIMD register.
387 int i_shift_offset,i_coord_offset,outeriter,inneriter;
388 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
390 int j_coord_offsetA,j_coord_offsetB;
391 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
393 real *shiftvec,*fshift,*x,*f;
394 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
396 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
397 int vdwjidx0A,vdwjidx0B;
398 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
399 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
400 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
403 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
406 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
407 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
409 __m128i ifour = _mm_set1_epi32(4);
410 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
412 __m128d dummy_mask,cutoff_mask;
413 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
414 __m128d one = _mm_set1_pd(1.0);
415 __m128d two = _mm_set1_pd(2.0);
421 jindex = nlist->jindex;
423 shiftidx = nlist->shift;
425 shiftvec = fr->shift_vec[0];
426 fshift = fr->fshift[0];
427 facel = _mm_set1_pd(fr->epsfac);
428 charge = mdatoms->chargeA;
429 nvdwtype = fr->ntype;
431 vdwtype = mdatoms->typeA;
433 vftab = kernel_data->table_elec->data;
434 vftabscale = _mm_set1_pd(kernel_data->table_elec->scale);
436 /* Avoid stupid compiler warnings */
444 /* Start outer loop over neighborlists */
445 for(iidx=0; iidx<nri; iidx++)
447 /* Load shift vector for this list */
448 i_shift_offset = DIM*shiftidx[iidx];
450 /* Load limits for loop over neighbors */
451 j_index_start = jindex[iidx];
452 j_index_end = jindex[iidx+1];
454 /* Get outer coordinate index */
456 i_coord_offset = DIM*inr;
458 /* Load i particle coords and add shift vector */
459 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
461 fix0 = _mm_setzero_pd();
462 fiy0 = _mm_setzero_pd();
463 fiz0 = _mm_setzero_pd();
465 /* Load parameters for i particles */
466 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
467 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
469 /* Start inner kernel loop */
470 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
473 /* Get j neighbor index, and coordinate index */
476 j_coord_offsetA = DIM*jnrA;
477 j_coord_offsetB = DIM*jnrB;
479 /* load j atom coordinates */
480 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
483 /* Calculate displacement vector */
484 dx00 = _mm_sub_pd(ix0,jx0);
485 dy00 = _mm_sub_pd(iy0,jy0);
486 dz00 = _mm_sub_pd(iz0,jz0);
488 /* Calculate squared distance and things based on it */
489 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
491 rinv00 = gmx_mm_invsqrt_pd(rsq00);
493 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
495 /* Load parameters for j particles */
496 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
497 vdwjidx0A = 2*vdwtype[jnrA+0];
498 vdwjidx0B = 2*vdwtype[jnrB+0];
500 /**************************
501 * CALCULATE INTERACTIONS *
502 **************************/
504 r00 = _mm_mul_pd(rsq00,rinv00);
506 /* Compute parameters for interactions between i and j atoms */
507 qq00 = _mm_mul_pd(iq0,jq0);
508 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
509 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
511 /* Calculate table index by multiplying r with table scale and truncate to integer */
512 rt = _mm_mul_pd(r00,vftabscale);
513 vfitab = _mm_cvttpd_epi32(rt);
515 vfeps = _mm_frcz_pd(rt);
517 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
519 twovfeps = _mm_add_pd(vfeps,vfeps);
520 vfitab = _mm_slli_epi32(vfitab,2);
522 /* CUBIC SPLINE TABLE ELECTROSTATICS */
523 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
524 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
525 GMX_MM_TRANSPOSE2_PD(Y,F);
526 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
527 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
528 GMX_MM_TRANSPOSE2_PD(G,H);
529 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
530 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
531 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
533 /* LENNARD-JONES DISPERSION/REPULSION */
535 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
536 fvdw = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
538 fscal = _mm_add_pd(felec,fvdw);
540 /* Update vectorial force */
541 fix0 = _mm_macc_pd(dx00,fscal,fix0);
542 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
543 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
545 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
546 _mm_mul_pd(dx00,fscal),
547 _mm_mul_pd(dy00,fscal),
548 _mm_mul_pd(dz00,fscal));
550 /* Inner loop uses 50 flops */
557 j_coord_offsetA = DIM*jnrA;
559 /* load j atom coordinates */
560 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
563 /* Calculate displacement vector */
564 dx00 = _mm_sub_pd(ix0,jx0);
565 dy00 = _mm_sub_pd(iy0,jy0);
566 dz00 = _mm_sub_pd(iz0,jz0);
568 /* Calculate squared distance and things based on it */
569 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
571 rinv00 = gmx_mm_invsqrt_pd(rsq00);
573 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
575 /* Load parameters for j particles */
576 jq0 = _mm_load_sd(charge+jnrA+0);
577 vdwjidx0A = 2*vdwtype[jnrA+0];
579 /**************************
580 * CALCULATE INTERACTIONS *
581 **************************/
583 r00 = _mm_mul_pd(rsq00,rinv00);
585 /* Compute parameters for interactions between i and j atoms */
586 qq00 = _mm_mul_pd(iq0,jq0);
587 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
589 /* Calculate table index by multiplying r with table scale and truncate to integer */
590 rt = _mm_mul_pd(r00,vftabscale);
591 vfitab = _mm_cvttpd_epi32(rt);
593 vfeps = _mm_frcz_pd(rt);
595 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
597 twovfeps = _mm_add_pd(vfeps,vfeps);
598 vfitab = _mm_slli_epi32(vfitab,2);
600 /* CUBIC SPLINE TABLE ELECTROSTATICS */
601 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
602 F = _mm_setzero_pd();
603 GMX_MM_TRANSPOSE2_PD(Y,F);
604 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
605 H = _mm_setzero_pd();
606 GMX_MM_TRANSPOSE2_PD(G,H);
607 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
608 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
609 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
611 /* LENNARD-JONES DISPERSION/REPULSION */
613 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
614 fvdw = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
616 fscal = _mm_add_pd(felec,fvdw);
618 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
620 /* Update vectorial force */
621 fix0 = _mm_macc_pd(dx00,fscal,fix0);
622 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
623 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
625 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
626 _mm_mul_pd(dx00,fscal),
627 _mm_mul_pd(dy00,fscal),
628 _mm_mul_pd(dz00,fscal));
630 /* Inner loop uses 50 flops */
633 /* End of innermost loop */
635 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
636 f+i_coord_offset,fshift+i_shift_offset);
638 /* Increment number of inner iterations */
639 inneriter += j_index_end - j_index_start;
641 /* Outer loop uses 7 flops */
644 /* Increment number of outer iterations */
647 /* Update outer/inner flops */
649 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*50);