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_ElecNone_VdwLJSw_GeomP1P1_VF_sse2_double
38 * Electrostatics interaction: None
39 * VdW interaction: LennardJones
40 * Geometry: Particle-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecNone_VdwLJSw_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;
72 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
75 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
76 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
77 __m128d rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
78 real rswitch_scalar,d_scalar;
79 __m128d dummy_mask,cutoff_mask;
80 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
81 __m128d one = _mm_set1_pd(1.0);
82 __m128d two = _mm_set1_pd(2.0);
88 jindex = nlist->jindex;
90 shiftidx = nlist->shift;
92 shiftvec = fr->shift_vec[0];
93 fshift = fr->fshift[0];
96 vdwtype = mdatoms->typeA;
98 rcutoff_scalar = fr->rvdw;
99 rcutoff = _mm_set1_pd(rcutoff_scalar);
100 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
102 rswitch_scalar = fr->rvdw_switch;
103 rswitch = _mm_set1_pd(rswitch_scalar);
104 /* Setup switch parameters */
105 d_scalar = rcutoff_scalar-rswitch_scalar;
106 d = _mm_set1_pd(d_scalar);
107 swV3 = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
108 swV4 = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
109 swV5 = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
110 swF2 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
111 swF3 = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
112 swF4 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
114 /* Avoid stupid compiler warnings */
122 /* Start outer loop over neighborlists */
123 for(iidx=0; iidx<nri; iidx++)
125 /* Load shift vector for this list */
126 i_shift_offset = DIM*shiftidx[iidx];
128 /* Load limits for loop over neighbors */
129 j_index_start = jindex[iidx];
130 j_index_end = jindex[iidx+1];
132 /* Get outer coordinate index */
134 i_coord_offset = DIM*inr;
136 /* Load i particle coords and add shift vector */
137 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
139 fix0 = _mm_setzero_pd();
140 fiy0 = _mm_setzero_pd();
141 fiz0 = _mm_setzero_pd();
143 /* Load parameters for i particles */
144 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
146 /* Reset potential sums */
147 vvdwsum = _mm_setzero_pd();
149 /* Start inner kernel loop */
150 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
153 /* Get j neighbor index, and coordinate index */
156 j_coord_offsetA = DIM*jnrA;
157 j_coord_offsetB = DIM*jnrB;
159 /* load j atom coordinates */
160 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
163 /* Calculate displacement vector */
164 dx00 = _mm_sub_pd(ix0,jx0);
165 dy00 = _mm_sub_pd(iy0,jy0);
166 dz00 = _mm_sub_pd(iz0,jz0);
168 /* Calculate squared distance and things based on it */
169 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
171 rinv00 = gmx_mm_invsqrt_pd(rsq00);
173 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
175 /* Load parameters for j particles */
176 vdwjidx0A = 2*vdwtype[jnrA+0];
177 vdwjidx0B = 2*vdwtype[jnrB+0];
179 /**************************
180 * CALCULATE INTERACTIONS *
181 **************************/
183 if (gmx_mm_any_lt(rsq00,rcutoff2))
186 r00 = _mm_mul_pd(rsq00,rinv00);
188 /* Compute parameters for interactions between i and j atoms */
189 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
190 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
192 /* LENNARD-JONES DISPERSION/REPULSION */
194 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
195 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
196 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
197 vvdw = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
198 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
200 d = _mm_sub_pd(r00,rswitch);
201 d = _mm_max_pd(d,_mm_setzero_pd());
202 d2 = _mm_mul_pd(d,d);
203 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
205 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
207 /* Evaluate switch function */
208 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
209 fvdw = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
210 vvdw = _mm_mul_pd(vvdw,sw);
211 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
213 /* Update potential sum for this i atom from the interaction with this j atom. */
214 vvdw = _mm_and_pd(vvdw,cutoff_mask);
215 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
219 fscal = _mm_and_pd(fscal,cutoff_mask);
221 /* Calculate temporary vectorial force */
222 tx = _mm_mul_pd(fscal,dx00);
223 ty = _mm_mul_pd(fscal,dy00);
224 tz = _mm_mul_pd(fscal,dz00);
226 /* Update vectorial force */
227 fix0 = _mm_add_pd(fix0,tx);
228 fiy0 = _mm_add_pd(fiy0,ty);
229 fiz0 = _mm_add_pd(fiz0,tz);
231 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
235 /* Inner loop uses 59 flops */
242 j_coord_offsetA = DIM*jnrA;
244 /* load j atom coordinates */
245 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
248 /* Calculate displacement vector */
249 dx00 = _mm_sub_pd(ix0,jx0);
250 dy00 = _mm_sub_pd(iy0,jy0);
251 dz00 = _mm_sub_pd(iz0,jz0);
253 /* Calculate squared distance and things based on it */
254 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
256 rinv00 = gmx_mm_invsqrt_pd(rsq00);
258 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
260 /* Load parameters for j particles */
261 vdwjidx0A = 2*vdwtype[jnrA+0];
263 /**************************
264 * CALCULATE INTERACTIONS *
265 **************************/
267 if (gmx_mm_any_lt(rsq00,rcutoff2))
270 r00 = _mm_mul_pd(rsq00,rinv00);
272 /* Compute parameters for interactions between i and j atoms */
273 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
275 /* LENNARD-JONES DISPERSION/REPULSION */
277 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
278 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
279 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
280 vvdw = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
281 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
283 d = _mm_sub_pd(r00,rswitch);
284 d = _mm_max_pd(d,_mm_setzero_pd());
285 d2 = _mm_mul_pd(d,d);
286 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
288 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
290 /* Evaluate switch function */
291 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
292 fvdw = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
293 vvdw = _mm_mul_pd(vvdw,sw);
294 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
296 /* Update potential sum for this i atom from the interaction with this j atom. */
297 vvdw = _mm_and_pd(vvdw,cutoff_mask);
298 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
299 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
303 fscal = _mm_and_pd(fscal,cutoff_mask);
305 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
307 /* Calculate temporary vectorial force */
308 tx = _mm_mul_pd(fscal,dx00);
309 ty = _mm_mul_pd(fscal,dy00);
310 tz = _mm_mul_pd(fscal,dz00);
312 /* Update vectorial force */
313 fix0 = _mm_add_pd(fix0,tx);
314 fiy0 = _mm_add_pd(fiy0,ty);
315 fiz0 = _mm_add_pd(fiz0,tz);
317 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
321 /* Inner loop uses 59 flops */
324 /* End of innermost loop */
326 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
327 f+i_coord_offset,fshift+i_shift_offset);
330 /* Update potential energies */
331 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
333 /* Increment number of inner iterations */
334 inneriter += j_index_end - j_index_start;
336 /* Outer loop uses 7 flops */
339 /* Increment number of outer iterations */
342 /* Update outer/inner flops */
344 inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*59);
347 * Gromacs nonbonded kernel: nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_sse2_double
348 * Electrostatics interaction: None
349 * VdW interaction: LennardJones
350 * Geometry: Particle-Particle
351 * Calculate force/pot: Force
354 nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_sse2_double
355 (t_nblist * gmx_restrict nlist,
356 rvec * gmx_restrict xx,
357 rvec * gmx_restrict ff,
358 t_forcerec * gmx_restrict fr,
359 t_mdatoms * gmx_restrict mdatoms,
360 nb_kernel_data_t * gmx_restrict kernel_data,
361 t_nrnb * gmx_restrict nrnb)
363 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
364 * just 0 for non-waters.
365 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
366 * jnr indices corresponding to data put in the four positions in the SIMD register.
368 int i_shift_offset,i_coord_offset,outeriter,inneriter;
369 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
371 int j_coord_offsetA,j_coord_offsetB;
372 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
374 real *shiftvec,*fshift,*x,*f;
375 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
377 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
378 int vdwjidx0A,vdwjidx0B;
379 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
380 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
382 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
385 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
386 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
387 __m128d rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
388 real rswitch_scalar,d_scalar;
389 __m128d dummy_mask,cutoff_mask;
390 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
391 __m128d one = _mm_set1_pd(1.0);
392 __m128d two = _mm_set1_pd(2.0);
398 jindex = nlist->jindex;
400 shiftidx = nlist->shift;
402 shiftvec = fr->shift_vec[0];
403 fshift = fr->fshift[0];
404 nvdwtype = fr->ntype;
406 vdwtype = mdatoms->typeA;
408 rcutoff_scalar = fr->rvdw;
409 rcutoff = _mm_set1_pd(rcutoff_scalar);
410 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
412 rswitch_scalar = fr->rvdw_switch;
413 rswitch = _mm_set1_pd(rswitch_scalar);
414 /* Setup switch parameters */
415 d_scalar = rcutoff_scalar-rswitch_scalar;
416 d = _mm_set1_pd(d_scalar);
417 swV3 = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
418 swV4 = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
419 swV5 = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
420 swF2 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
421 swF3 = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
422 swF4 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
424 /* Avoid stupid compiler warnings */
432 /* Start outer loop over neighborlists */
433 for(iidx=0; iidx<nri; iidx++)
435 /* Load shift vector for this list */
436 i_shift_offset = DIM*shiftidx[iidx];
438 /* Load limits for loop over neighbors */
439 j_index_start = jindex[iidx];
440 j_index_end = jindex[iidx+1];
442 /* Get outer coordinate index */
444 i_coord_offset = DIM*inr;
446 /* Load i particle coords and add shift vector */
447 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
449 fix0 = _mm_setzero_pd();
450 fiy0 = _mm_setzero_pd();
451 fiz0 = _mm_setzero_pd();
453 /* Load parameters for i particles */
454 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
456 /* Start inner kernel loop */
457 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
460 /* Get j neighbor index, and coordinate index */
463 j_coord_offsetA = DIM*jnrA;
464 j_coord_offsetB = DIM*jnrB;
466 /* load j atom coordinates */
467 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
470 /* Calculate displacement vector */
471 dx00 = _mm_sub_pd(ix0,jx0);
472 dy00 = _mm_sub_pd(iy0,jy0);
473 dz00 = _mm_sub_pd(iz0,jz0);
475 /* Calculate squared distance and things based on it */
476 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
478 rinv00 = gmx_mm_invsqrt_pd(rsq00);
480 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
482 /* Load parameters for j particles */
483 vdwjidx0A = 2*vdwtype[jnrA+0];
484 vdwjidx0B = 2*vdwtype[jnrB+0];
486 /**************************
487 * CALCULATE INTERACTIONS *
488 **************************/
490 if (gmx_mm_any_lt(rsq00,rcutoff2))
493 r00 = _mm_mul_pd(rsq00,rinv00);
495 /* Compute parameters for interactions between i and j atoms */
496 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
497 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
499 /* LENNARD-JONES DISPERSION/REPULSION */
501 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
502 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
503 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
504 vvdw = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
505 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
507 d = _mm_sub_pd(r00,rswitch);
508 d = _mm_max_pd(d,_mm_setzero_pd());
509 d2 = _mm_mul_pd(d,d);
510 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
512 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
514 /* Evaluate switch function */
515 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
516 fvdw = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
517 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
521 fscal = _mm_and_pd(fscal,cutoff_mask);
523 /* Calculate temporary vectorial force */
524 tx = _mm_mul_pd(fscal,dx00);
525 ty = _mm_mul_pd(fscal,dy00);
526 tz = _mm_mul_pd(fscal,dz00);
528 /* Update vectorial force */
529 fix0 = _mm_add_pd(fix0,tx);
530 fiy0 = _mm_add_pd(fiy0,ty);
531 fiz0 = _mm_add_pd(fiz0,tz);
533 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
537 /* Inner loop uses 56 flops */
544 j_coord_offsetA = DIM*jnrA;
546 /* load j atom coordinates */
547 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
550 /* Calculate displacement vector */
551 dx00 = _mm_sub_pd(ix0,jx0);
552 dy00 = _mm_sub_pd(iy0,jy0);
553 dz00 = _mm_sub_pd(iz0,jz0);
555 /* Calculate squared distance and things based on it */
556 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
558 rinv00 = gmx_mm_invsqrt_pd(rsq00);
560 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
562 /* Load parameters for j particles */
563 vdwjidx0A = 2*vdwtype[jnrA+0];
565 /**************************
566 * CALCULATE INTERACTIONS *
567 **************************/
569 if (gmx_mm_any_lt(rsq00,rcutoff2))
572 r00 = _mm_mul_pd(rsq00,rinv00);
574 /* Compute parameters for interactions between i and j atoms */
575 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
577 /* LENNARD-JONES DISPERSION/REPULSION */
579 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
580 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
581 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
582 vvdw = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
583 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
585 d = _mm_sub_pd(r00,rswitch);
586 d = _mm_max_pd(d,_mm_setzero_pd());
587 d2 = _mm_mul_pd(d,d);
588 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
590 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
592 /* Evaluate switch function */
593 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
594 fvdw = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
595 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
599 fscal = _mm_and_pd(fscal,cutoff_mask);
601 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
603 /* Calculate temporary vectorial force */
604 tx = _mm_mul_pd(fscal,dx00);
605 ty = _mm_mul_pd(fscal,dy00);
606 tz = _mm_mul_pd(fscal,dz00);
608 /* Update vectorial force */
609 fix0 = _mm_add_pd(fix0,tx);
610 fiy0 = _mm_add_pd(fiy0,ty);
611 fiz0 = _mm_add_pd(fiz0,tz);
613 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
617 /* Inner loop uses 56 flops */
620 /* End of innermost loop */
622 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
623 f+i_coord_offset,fshift+i_shift_offset);
625 /* Increment number of inner iterations */
626 inneriter += j_index_end - j_index_start;
628 /* Outer loop uses 6 flops */
631 /* Increment number of outer iterations */
634 /* Update outer/inner flops */
636 inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*56);