src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_sse4_1_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse4_1_double kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  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
  13  *
  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
  17  * later version.
  18  *
  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.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_sse4_1_double.h"
  34 #include "kernelutil_x86_sse4_1_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_VF_sse4_1_double
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_VF_sse4_1_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)
  52 {
  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.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB;
  61     int              j_coord_offsetA,j_coord_offsetB;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwioffset1;
  69     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  70     int              vdwioffset2;
  71     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  72     int              vdwjidx0A,vdwjidx0B;
  73     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  74     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  75     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  76     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  77     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  78     real             *charge;
  79     int              nvdwtype;
  80     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  81     int              *vdwtype;
  82     real             *vdwparam;
  83     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  84     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  85     __m128i          vfitab;
  86     __m128i          ifour       = _mm_set1_epi32(4);
  87     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  88     real             *vftab;
  89     __m128d          dummy_mask,cutoff_mask;
  90     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  91     __m128d          one     = _mm_set1_pd(1.0);
  92     __m128d          two     = _mm_set1_pd(2.0);
  93     x                = xx[0];
  94     f                = ff[0];
  95
  96     nri              = nlist->nri;
  97     iinr             = nlist->iinr;
  98     jindex           = nlist->jindex;
  99     jjnr             = nlist->jjnr;
 100     shiftidx         = nlist->shift;
 101     gid              = nlist->gid;
 102     shiftvec         = fr->shift_vec[0];
 103     fshift           = fr->fshift[0];
 104     facel            = _mm_set1_pd(fr->epsfac);
 105     charge           = mdatoms->chargeA;
 106     krf              = _mm_set1_pd(fr->ic->k_rf);
 107     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 108     crf              = _mm_set1_pd(fr->ic->c_rf);
 109     nvdwtype         = fr->ntype;
 110     vdwparam         = fr->nbfp;
 111     vdwtype          = mdatoms->typeA;
 112
 113     vftab            = kernel_data->table_vdw->data;
 114     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 115
 116     /* Setup water-specific parameters */
 117     inr              = nlist->iinr[0];
 118     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 119     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 120     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 121     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 122
 123     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 124     rcutoff_scalar   = fr->rcoulomb;
 125     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 126     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 127
 128     /* Avoid stupid compiler warnings */
 129     jnrA = jnrB = 0;
 130     j_coord_offsetA = 0;
 131     j_coord_offsetB = 0;
 132
 133     outeriter        = 0;
 134     inneriter        = 0;
 135
 136     /* Start outer loop over neighborlists */
 137     for(iidx=0; iidx<nri; iidx++)
 138     {
 139         /* Load shift vector for this list */
 140         i_shift_offset   = DIM*shiftidx[iidx];
 141
 142         /* Load limits for loop over neighbors */
 143         j_index_start    = jindex[iidx];
 144         j_index_end      = jindex[iidx+1];
 145
 146         /* Get outer coordinate index */
 147         inr              = iinr[iidx];
 148         i_coord_offset   = DIM*inr;
 149
 150         /* Load i particle coords and add shift vector */
 151         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 152                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 153
 154         fix0             = _mm_setzero_pd();
 155         fiy0             = _mm_setzero_pd();
 156         fiz0             = _mm_setzero_pd();
 157         fix1             = _mm_setzero_pd();
 158         fiy1             = _mm_setzero_pd();
 159         fiz1             = _mm_setzero_pd();
 160         fix2             = _mm_setzero_pd();
 161         fiy2             = _mm_setzero_pd();
 162         fiz2             = _mm_setzero_pd();
 163
 164         /* Reset potential sums */
 165         velecsum         = _mm_setzero_pd();
 166         vvdwsum          = _mm_setzero_pd();
 167
 168         /* Start inner kernel loop */
 169         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 170         {
 171
 172             /* Get j neighbor index, and coordinate index */
 173             jnrA             = jjnr[jidx];
 174             jnrB             = jjnr[jidx+1];
 175             j_coord_offsetA  = DIM*jnrA;
 176             j_coord_offsetB  = DIM*jnrB;
 177
 178             /* load j atom coordinates */
 179             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 180                                               &jx0,&jy0,&jz0);
 181
 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);
 186             dx10             = _mm_sub_pd(ix1,jx0);
 187             dy10             = _mm_sub_pd(iy1,jy0);
 188             dz10             = _mm_sub_pd(iz1,jz0);
 189             dx20             = _mm_sub_pd(ix2,jx0);
 190             dy20             = _mm_sub_pd(iy2,jy0);
 191             dz20             = _mm_sub_pd(iz2,jz0);
 192
 193             /* Calculate squared distance and things based on it */
 194             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 195             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 196             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 197
 198             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 199             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 200             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 201
 202             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 203             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 204             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 205
 206             /* Load parameters for j particles */
 207             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 208             vdwjidx0A        = 2*vdwtype[jnrA+0];
 209             vdwjidx0B        = 2*vdwtype[jnrB+0];
 210
 211             fjx0             = _mm_setzero_pd();
 212             fjy0             = _mm_setzero_pd();
 213             fjz0             = _mm_setzero_pd();
 214
 215             /**************************
 216              * CALCULATE INTERACTIONS *
 217              **************************/
 218
 219             if (gmx_mm_any_lt(rsq00,rcutoff2))
 220             {
 221
 222             r00              = _mm_mul_pd(rsq00,rinv00);
 223
 224             /* Compute parameters for interactions between i and j atoms */
 225             qq00             = _mm_mul_pd(iq0,jq0);
 226             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 227                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 228
 229             /* Calculate table index by multiplying r with table scale and truncate to integer */
 230             rt               = _mm_mul_pd(r00,vftabscale);
 231             vfitab           = _mm_cvttpd_epi32(rt);
 232             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 233             vfitab           = _mm_slli_epi32(vfitab,3);
 234
 235             /* REACTION-FIELD ELECTROSTATICS */
 236             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 237             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 238
 239             /* CUBIC SPLINE TABLE DISPERSION */
 240             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 241             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 242             GMX_MM_TRANSPOSE2_PD(Y,F);
 243             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 244             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 245             GMX_MM_TRANSPOSE2_PD(G,H);
 246             Heps             = _mm_mul_pd(vfeps,H);
 247             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 248             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 249             vvdw6            = _mm_mul_pd(c6_00,VV);
 250             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 251             fvdw6            = _mm_mul_pd(c6_00,FF);
 252
 253             /* CUBIC SPLINE TABLE REPULSION */
 254             vfitab           = _mm_add_epi32(vfitab,ifour);
 255             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 256             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 257             GMX_MM_TRANSPOSE2_PD(Y,F);
 258             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 259             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 260             GMX_MM_TRANSPOSE2_PD(G,H);
 261             Heps             = _mm_mul_pd(vfeps,H);
 262             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 263             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 264             vvdw12           = _mm_mul_pd(c12_00,VV);
 265             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 266             fvdw12           = _mm_mul_pd(c12_00,FF);
 267             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 268             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 269
 270             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 271
 272             /* Update potential sum for this i atom from the interaction with this j atom. */
 273             velec            = _mm_and_pd(velec,cutoff_mask);
 274             velecsum         = _mm_add_pd(velecsum,velec);
 275             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 276             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 277
 278             fscal            = _mm_add_pd(felec,fvdw);
 279
 280             fscal            = _mm_and_pd(fscal,cutoff_mask);
 281
 282             /* Calculate temporary vectorial force */
 283             tx               = _mm_mul_pd(fscal,dx00);
 284             ty               = _mm_mul_pd(fscal,dy00);
 285             tz               = _mm_mul_pd(fscal,dz00);
 286
 287             /* Update vectorial force */
 288             fix0             = _mm_add_pd(fix0,tx);
 289             fiy0             = _mm_add_pd(fiy0,ty);
 290             fiz0             = _mm_add_pd(fiz0,tz);
 291
 292             fjx0             = _mm_add_pd(fjx0,tx);
 293             fjy0             = _mm_add_pd(fjy0,ty);
 294             fjz0             = _mm_add_pd(fjz0,tz);
 295
 296             }
 297
 298             /**************************
 299              * CALCULATE INTERACTIONS *
 300              **************************/
 301
 302             if (gmx_mm_any_lt(rsq10,rcutoff2))
 303             {
 304
 305             /* Compute parameters for interactions between i and j atoms */
 306             qq10             = _mm_mul_pd(iq1,jq0);
 307
 308             /* REACTION-FIELD ELECTROSTATICS */
 309             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 310             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 311
 312             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 313
 314             /* Update potential sum for this i atom from the interaction with this j atom. */
 315             velec            = _mm_and_pd(velec,cutoff_mask);
 316             velecsum         = _mm_add_pd(velecsum,velec);
 317
 318             fscal            = felec;
 319
 320             fscal            = _mm_and_pd(fscal,cutoff_mask);
 321
 322             /* Calculate temporary vectorial force */
 323             tx               = _mm_mul_pd(fscal,dx10);
 324             ty               = _mm_mul_pd(fscal,dy10);
 325             tz               = _mm_mul_pd(fscal,dz10);
 326
 327             /* Update vectorial force */
 328             fix1             = _mm_add_pd(fix1,tx);
 329             fiy1             = _mm_add_pd(fiy1,ty);
 330             fiz1             = _mm_add_pd(fiz1,tz);
 331
 332             fjx0             = _mm_add_pd(fjx0,tx);
 333             fjy0             = _mm_add_pd(fjy0,ty);
 334             fjz0             = _mm_add_pd(fjz0,tz);
 335
 336             }
 337
 338             /**************************
 339              * CALCULATE INTERACTIONS *
 340              **************************/
 341
 342             if (gmx_mm_any_lt(rsq20,rcutoff2))
 343             {
 344
 345             /* Compute parameters for interactions between i and j atoms */
 346             qq20             = _mm_mul_pd(iq2,jq0);
 347
 348             /* REACTION-FIELD ELECTROSTATICS */
 349             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 350             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 351
 352             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 353
 354             /* Update potential sum for this i atom from the interaction with this j atom. */
 355             velec            = _mm_and_pd(velec,cutoff_mask);
 356             velecsum         = _mm_add_pd(velecsum,velec);
 357
 358             fscal            = felec;
 359
 360             fscal            = _mm_and_pd(fscal,cutoff_mask);
 361
 362             /* Calculate temporary vectorial force */
 363             tx               = _mm_mul_pd(fscal,dx20);
 364             ty               = _mm_mul_pd(fscal,dy20);
 365             tz               = _mm_mul_pd(fscal,dz20);
 366
 367             /* Update vectorial force */
 368             fix2             = _mm_add_pd(fix2,tx);
 369             fiy2             = _mm_add_pd(fiy2,ty);
 370             fiz2             = _mm_add_pd(fiz2,tz);
 371
 372             fjx0             = _mm_add_pd(fjx0,tx);
 373             fjy0             = _mm_add_pd(fjy0,ty);
 374             fjz0             = _mm_add_pd(fjz0,tz);
 375
 376             }
 377
 378             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 379
 380             /* Inner loop uses 147 flops */
 381         }
 382
 383         if(jidx<j_index_end)
 384         {
 385
 386             jnrA             = jjnr[jidx];
 387             j_coord_offsetA  = DIM*jnrA;
 388
 389             /* load j atom coordinates */
 390             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 391                                               &jx0,&jy0,&jz0);
 392
 393             /* Calculate displacement vector */
 394             dx00             = _mm_sub_pd(ix0,jx0);
 395             dy00             = _mm_sub_pd(iy0,jy0);
 396             dz00             = _mm_sub_pd(iz0,jz0);
 397             dx10             = _mm_sub_pd(ix1,jx0);
 398             dy10             = _mm_sub_pd(iy1,jy0);
 399             dz10             = _mm_sub_pd(iz1,jz0);
 400             dx20             = _mm_sub_pd(ix2,jx0);
 401             dy20             = _mm_sub_pd(iy2,jy0);
 402             dz20             = _mm_sub_pd(iz2,jz0);
 403
 404             /* Calculate squared distance and things based on it */
 405             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 406             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 407             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 408
 409             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 410             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 411             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 412
 413             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 414             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 415             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 416
 417             /* Load parameters for j particles */
 418             jq0              = _mm_load_sd(charge+jnrA+0);
 419             vdwjidx0A        = 2*vdwtype[jnrA+0];
 420
 421             fjx0             = _mm_setzero_pd();
 422             fjy0             = _mm_setzero_pd();
 423             fjz0             = _mm_setzero_pd();
 424
 425             /**************************
 426              * CALCULATE INTERACTIONS *
 427              **************************/
 428
 429             if (gmx_mm_any_lt(rsq00,rcutoff2))
 430             {
 431
 432             r00              = _mm_mul_pd(rsq00,rinv00);
 433
 434             /* Compute parameters for interactions between i and j atoms */
 435             qq00             = _mm_mul_pd(iq0,jq0);
 436             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 437
 438             /* Calculate table index by multiplying r with table scale and truncate to integer */
 439             rt               = _mm_mul_pd(r00,vftabscale);
 440             vfitab           = _mm_cvttpd_epi32(rt);
 441             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 442             vfitab           = _mm_slli_epi32(vfitab,3);
 443
 444             /* REACTION-FIELD ELECTROSTATICS */
 445             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 446             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 447
 448             /* CUBIC SPLINE TABLE DISPERSION */
 449             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 450             F                = _mm_setzero_pd();
 451             GMX_MM_TRANSPOSE2_PD(Y,F);
 452             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 453             H                = _mm_setzero_pd();
 454             GMX_MM_TRANSPOSE2_PD(G,H);
 455             Heps             = _mm_mul_pd(vfeps,H);
 456             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 457             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 458             vvdw6            = _mm_mul_pd(c6_00,VV);
 459             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 460             fvdw6            = _mm_mul_pd(c6_00,FF);
 461
 462             /* CUBIC SPLINE TABLE REPULSION */
 463             vfitab           = _mm_add_epi32(vfitab,ifour);
 464             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 465             F                = _mm_setzero_pd();
 466             GMX_MM_TRANSPOSE2_PD(Y,F);
 467             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 468             H                = _mm_setzero_pd();
 469             GMX_MM_TRANSPOSE2_PD(G,H);
 470             Heps             = _mm_mul_pd(vfeps,H);
 471             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 472             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 473             vvdw12           = _mm_mul_pd(c12_00,VV);
 474             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 475             fvdw12           = _mm_mul_pd(c12_00,FF);
 476             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 477             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 478
 479             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 480
 481             /* Update potential sum for this i atom from the interaction with this j atom. */
 482             velec            = _mm_and_pd(velec,cutoff_mask);
 483             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 484             velecsum         = _mm_add_pd(velecsum,velec);
 485             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 486             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 487             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 488
 489             fscal            = _mm_add_pd(felec,fvdw);
 490
 491             fscal            = _mm_and_pd(fscal,cutoff_mask);
 492
 493             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 494
 495             /* Calculate temporary vectorial force */
 496             tx               = _mm_mul_pd(fscal,dx00);
 497             ty               = _mm_mul_pd(fscal,dy00);
 498             tz               = _mm_mul_pd(fscal,dz00);
 499
 500             /* Update vectorial force */
 501             fix0             = _mm_add_pd(fix0,tx);
 502             fiy0             = _mm_add_pd(fiy0,ty);
 503             fiz0             = _mm_add_pd(fiz0,tz);
 504
 505             fjx0             = _mm_add_pd(fjx0,tx);
 506             fjy0             = _mm_add_pd(fjy0,ty);
 507             fjz0             = _mm_add_pd(fjz0,tz);
 508
 509             }
 510
 511             /**************************
 512              * CALCULATE INTERACTIONS *
 513              **************************/
 514
 515             if (gmx_mm_any_lt(rsq10,rcutoff2))
 516             {
 517
 518             /* Compute parameters for interactions between i and j atoms */
 519             qq10             = _mm_mul_pd(iq1,jq0);
 520
 521             /* REACTION-FIELD ELECTROSTATICS */
 522             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 523             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 524
 525             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 526
 527             /* Update potential sum for this i atom from the interaction with this j atom. */
 528             velec            = _mm_and_pd(velec,cutoff_mask);
 529             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 530             velecsum         = _mm_add_pd(velecsum,velec);
 531
 532             fscal            = felec;
 533
 534             fscal            = _mm_and_pd(fscal,cutoff_mask);
 535
 536             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 537
 538             /* Calculate temporary vectorial force */
 539             tx               = _mm_mul_pd(fscal,dx10);
 540             ty               = _mm_mul_pd(fscal,dy10);
 541             tz               = _mm_mul_pd(fscal,dz10);
 542
 543             /* Update vectorial force */
 544             fix1             = _mm_add_pd(fix1,tx);
 545             fiy1             = _mm_add_pd(fiy1,ty);
 546             fiz1             = _mm_add_pd(fiz1,tz);
 547
 548             fjx0             = _mm_add_pd(fjx0,tx);
 549             fjy0             = _mm_add_pd(fjy0,ty);
 550             fjz0             = _mm_add_pd(fjz0,tz);
 551
 552             }
 553
 554             /**************************
 555              * CALCULATE INTERACTIONS *
 556              **************************/
 557
 558             if (gmx_mm_any_lt(rsq20,rcutoff2))
 559             {
 560
 561             /* Compute parameters for interactions between i and j atoms */
 562             qq20             = _mm_mul_pd(iq2,jq0);
 563
 564             /* REACTION-FIELD ELECTROSTATICS */
 565             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 566             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 567
 568             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 569
 570             /* Update potential sum for this i atom from the interaction with this j atom. */
 571             velec            = _mm_and_pd(velec,cutoff_mask);
 572             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 573             velecsum         = _mm_add_pd(velecsum,velec);
 574
 575             fscal            = felec;
 576
 577             fscal            = _mm_and_pd(fscal,cutoff_mask);
 578
 579             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 580
 581             /* Calculate temporary vectorial force */
 582             tx               = _mm_mul_pd(fscal,dx20);
 583             ty               = _mm_mul_pd(fscal,dy20);
 584             tz               = _mm_mul_pd(fscal,dz20);
 585
 586             /* Update vectorial force */
 587             fix2             = _mm_add_pd(fix2,tx);
 588             fiy2             = _mm_add_pd(fiy2,ty);
 589             fiz2             = _mm_add_pd(fiz2,tz);
 590
 591             fjx0             = _mm_add_pd(fjx0,tx);
 592             fjy0             = _mm_add_pd(fjy0,ty);
 593             fjz0             = _mm_add_pd(fjz0,tz);
 594
 595             }
 596
 597             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 598
 599             /* Inner loop uses 147 flops */
 600         }
 601
 602         /* End of innermost loop */
 603
 604         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 605                                               f+i_coord_offset,fshift+i_shift_offset);
 606
 607         ggid                        = gid[iidx];
 608         /* Update potential energies */
 609         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 610         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 611
 612         /* Increment number of inner iterations */
 613         inneriter                  += j_index_end - j_index_start;
 614
 615         /* Outer loop uses 20 flops */
 616     }
 617
 618     /* Increment number of outer iterations */
 619     outeriter        += nri;
 620
 621     /* Update outer/inner flops */
 622
 623     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*147);
 624 }
 625 /*
 626  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_sse4_1_double
 627  * Electrostatics interaction: ReactionField
 628  * VdW interaction:            CubicSplineTable
 629  * Geometry:                   Water3-Particle
 630  * Calculate force/pot:        Force
 631  */
 632 void
 633 nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_sse4_1_double
 634                     (t_nblist * gmx_restrict                nlist,
 635                      rvec * gmx_restrict                    xx,
 636                      rvec * gmx_restrict                    ff,
 637                      t_forcerec * gmx_restrict              fr,
 638                      t_mdatoms * gmx_restrict               mdatoms,
 639                      nb_kernel_data_t * gmx_restrict        kernel_data,
 640                      t_nrnb * gmx_restrict                  nrnb)
 641 {
 642     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 643      * just 0 for non-waters.
 644      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 645      * jnr indices corresponding to data put in the four positions in the SIMD register.
 646      */
 647     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 648     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 649     int              jnrA,jnrB;
 650     int              j_coord_offsetA,j_coord_offsetB;
 651     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 652     real             rcutoff_scalar;
 653     real             *shiftvec,*fshift,*x,*f;
 654     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 655     int              vdwioffset0;
 656     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 657     int              vdwioffset1;
 658     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 659     int              vdwioffset2;
 660     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 661     int              vdwjidx0A,vdwjidx0B;
 662     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 663     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 664     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 665     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 666     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 667     real             *charge;
 668     int              nvdwtype;
 669     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 670     int              *vdwtype;
 671     real             *vdwparam;
 672     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 673     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 674     __m128i          vfitab;
 675     __m128i          ifour       = _mm_set1_epi32(4);
 676     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 677     real             *vftab;
 678     __m128d          dummy_mask,cutoff_mask;
 679     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 680     __m128d          one     = _mm_set1_pd(1.0);
 681     __m128d          two     = _mm_set1_pd(2.0);
 682     x                = xx[0];
 683     f                = ff[0];
 684
 685     nri              = nlist->nri;
 686     iinr             = nlist->iinr;
 687     jindex           = nlist->jindex;
 688     jjnr             = nlist->jjnr;
 689     shiftidx         = nlist->shift;
 690     gid              = nlist->gid;
 691     shiftvec         = fr->shift_vec[0];
 692     fshift           = fr->fshift[0];
 693     facel            = _mm_set1_pd(fr->epsfac);
 694     charge           = mdatoms->chargeA;
 695     krf              = _mm_set1_pd(fr->ic->k_rf);
 696     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 697     crf              = _mm_set1_pd(fr->ic->c_rf);
 698     nvdwtype         = fr->ntype;
 699     vdwparam         = fr->nbfp;
 700     vdwtype          = mdatoms->typeA;
 701
 702     vftab            = kernel_data->table_vdw->data;
 703     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 704
 705     /* Setup water-specific parameters */
 706     inr              = nlist->iinr[0];
 707     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 708     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 709     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 710     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 711
 712     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 713     rcutoff_scalar   = fr->rcoulomb;
 714     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 715     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 716
 717     /* Avoid stupid compiler warnings */
 718     jnrA = jnrB = 0;
 719     j_coord_offsetA = 0;
 720     j_coord_offsetB = 0;
 721
 722     outeriter        = 0;
 723     inneriter        = 0;
 724
 725     /* Start outer loop over neighborlists */
 726     for(iidx=0; iidx<nri; iidx++)
 727     {
 728         /* Load shift vector for this list */
 729         i_shift_offset   = DIM*shiftidx[iidx];
 730
 731         /* Load limits for loop over neighbors */
 732         j_index_start    = jindex[iidx];
 733         j_index_end      = jindex[iidx+1];
 734
 735         /* Get outer coordinate index */
 736         inr              = iinr[iidx];
 737         i_coord_offset   = DIM*inr;
 738
 739         /* Load i particle coords and add shift vector */
 740         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 741                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 742
 743         fix0             = _mm_setzero_pd();
 744         fiy0             = _mm_setzero_pd();
 745         fiz0             = _mm_setzero_pd();
 746         fix1             = _mm_setzero_pd();
 747         fiy1             = _mm_setzero_pd();
 748         fiz1             = _mm_setzero_pd();
 749         fix2             = _mm_setzero_pd();
 750         fiy2             = _mm_setzero_pd();
 751         fiz2             = _mm_setzero_pd();
 752
 753         /* Start inner kernel loop */
 754         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 755         {
 756
 757             /* Get j neighbor index, and coordinate index */
 758             jnrA             = jjnr[jidx];
 759             jnrB             = jjnr[jidx+1];
 760             j_coord_offsetA  = DIM*jnrA;
 761             j_coord_offsetB  = DIM*jnrB;
 762
 763             /* load j atom coordinates */
 764             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 765                                               &jx0,&jy0,&jz0);
 766
 767             /* Calculate displacement vector */
 768             dx00             = _mm_sub_pd(ix0,jx0);
 769             dy00             = _mm_sub_pd(iy0,jy0);
 770             dz00             = _mm_sub_pd(iz0,jz0);
 771             dx10             = _mm_sub_pd(ix1,jx0);
 772             dy10             = _mm_sub_pd(iy1,jy0);
 773             dz10             = _mm_sub_pd(iz1,jz0);
 774             dx20             = _mm_sub_pd(ix2,jx0);
 775             dy20             = _mm_sub_pd(iy2,jy0);
 776             dz20             = _mm_sub_pd(iz2,jz0);
 777
 778             /* Calculate squared distance and things based on it */
 779             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 780             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 781             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 782
 783             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 784             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 785             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 786
 787             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 788             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 789             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 790
 791             /* Load parameters for j particles */
 792             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 793             vdwjidx0A        = 2*vdwtype[jnrA+0];
 794             vdwjidx0B        = 2*vdwtype[jnrB+0];
 795
 796             fjx0             = _mm_setzero_pd();
 797             fjy0             = _mm_setzero_pd();
 798             fjz0             = _mm_setzero_pd();
 799
 800             /**************************
 801              * CALCULATE INTERACTIONS *
 802              **************************/
 803
 804             if (gmx_mm_any_lt(rsq00,rcutoff2))
 805             {
 806
 807             r00              = _mm_mul_pd(rsq00,rinv00);
 808
 809             /* Compute parameters for interactions between i and j atoms */
 810             qq00             = _mm_mul_pd(iq0,jq0);
 811             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 812                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 813
 814             /* Calculate table index by multiplying r with table scale and truncate to integer */
 815             rt               = _mm_mul_pd(r00,vftabscale);
 816             vfitab           = _mm_cvttpd_epi32(rt);
 817             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 818             vfitab           = _mm_slli_epi32(vfitab,3);
 819
 820             /* REACTION-FIELD ELECTROSTATICS */
 821             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 822
 823             /* CUBIC SPLINE TABLE DISPERSION */
 824             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 825             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 826             GMX_MM_TRANSPOSE2_PD(Y,F);
 827             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 828             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 829             GMX_MM_TRANSPOSE2_PD(G,H);
 830             Heps             = _mm_mul_pd(vfeps,H);
 831             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 832             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 833             fvdw6            = _mm_mul_pd(c6_00,FF);
 834
 835             /* CUBIC SPLINE TABLE REPULSION */
 836             vfitab           = _mm_add_epi32(vfitab,ifour);
 837             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 838             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 839             GMX_MM_TRANSPOSE2_PD(Y,F);
 840             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 841             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 842             GMX_MM_TRANSPOSE2_PD(G,H);
 843             Heps             = _mm_mul_pd(vfeps,H);
 844             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 845             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 846             fvdw12           = _mm_mul_pd(c12_00,FF);
 847             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 848
 849             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 850
 851             fscal            = _mm_add_pd(felec,fvdw);
 852
 853             fscal            = _mm_and_pd(fscal,cutoff_mask);
 854
 855             /* Calculate temporary vectorial force */
 856             tx               = _mm_mul_pd(fscal,dx00);
 857             ty               = _mm_mul_pd(fscal,dy00);
 858             tz               = _mm_mul_pd(fscal,dz00);
 859
 860             /* Update vectorial force */
 861             fix0             = _mm_add_pd(fix0,tx);
 862             fiy0             = _mm_add_pd(fiy0,ty);
 863             fiz0             = _mm_add_pd(fiz0,tz);
 864
 865             fjx0             = _mm_add_pd(fjx0,tx);
 866             fjy0             = _mm_add_pd(fjy0,ty);
 867             fjz0             = _mm_add_pd(fjz0,tz);
 868
 869             }
 870
 871             /**************************
 872              * CALCULATE INTERACTIONS *
 873              **************************/
 874
 875             if (gmx_mm_any_lt(rsq10,rcutoff2))
 876             {
 877
 878             /* Compute parameters for interactions between i and j atoms */
 879             qq10             = _mm_mul_pd(iq1,jq0);
 880
 881             /* REACTION-FIELD ELECTROSTATICS */
 882             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 883
 884             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 885
 886             fscal            = felec;
 887
 888             fscal            = _mm_and_pd(fscal,cutoff_mask);
 889
 890             /* Calculate temporary vectorial force */
 891             tx               = _mm_mul_pd(fscal,dx10);
 892             ty               = _mm_mul_pd(fscal,dy10);
 893             tz               = _mm_mul_pd(fscal,dz10);
 894
 895             /* Update vectorial force */
 896             fix1             = _mm_add_pd(fix1,tx);
 897             fiy1             = _mm_add_pd(fiy1,ty);
 898             fiz1             = _mm_add_pd(fiz1,tz);
 899
 900             fjx0             = _mm_add_pd(fjx0,tx);
 901             fjy0             = _mm_add_pd(fjy0,ty);
 902             fjz0             = _mm_add_pd(fjz0,tz);
 903
 904             }
 905
 906             /**************************
 907              * CALCULATE INTERACTIONS *
 908              **************************/
 909
 910             if (gmx_mm_any_lt(rsq20,rcutoff2))
 911             {
 912
 913             /* Compute parameters for interactions between i and j atoms */
 914             qq20             = _mm_mul_pd(iq2,jq0);
 915
 916             /* REACTION-FIELD ELECTROSTATICS */
 917             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 918
 919             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 920
 921             fscal            = felec;
 922
 923             fscal            = _mm_and_pd(fscal,cutoff_mask);
 924
 925             /* Calculate temporary vectorial force */
 926             tx               = _mm_mul_pd(fscal,dx20);
 927             ty               = _mm_mul_pd(fscal,dy20);
 928             tz               = _mm_mul_pd(fscal,dz20);
 929
 930             /* Update vectorial force */
 931             fix2             = _mm_add_pd(fix2,tx);
 932             fiy2             = _mm_add_pd(fiy2,ty);
 933             fiz2             = _mm_add_pd(fiz2,tz);
 934
 935             fjx0             = _mm_add_pd(fjx0,tx);
 936             fjy0             = _mm_add_pd(fjy0,ty);
 937             fjz0             = _mm_add_pd(fjz0,tz);
 938
 939             }
 940
 941             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 942
 943             /* Inner loop uses 120 flops */
 944         }
 945
 946         if(jidx<j_index_end)
 947         {
 948
 949             jnrA             = jjnr[jidx];
 950             j_coord_offsetA  = DIM*jnrA;
 951
 952             /* load j atom coordinates */
 953             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 954                                               &jx0,&jy0,&jz0);
 955
 956             /* Calculate displacement vector */
 957             dx00             = _mm_sub_pd(ix0,jx0);
 958             dy00             = _mm_sub_pd(iy0,jy0);
 959             dz00             = _mm_sub_pd(iz0,jz0);
 960             dx10             = _mm_sub_pd(ix1,jx0);
 961             dy10             = _mm_sub_pd(iy1,jy0);
 962             dz10             = _mm_sub_pd(iz1,jz0);
 963             dx20             = _mm_sub_pd(ix2,jx0);
 964             dy20             = _mm_sub_pd(iy2,jy0);
 965             dz20             = _mm_sub_pd(iz2,jz0);
 966
 967             /* Calculate squared distance and things based on it */
 968             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 969             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 970             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 971
 972             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 973             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 974             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 975
 976             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 977             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 978             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 979
 980             /* Load parameters for j particles */
 981             jq0              = _mm_load_sd(charge+jnrA+0);
 982             vdwjidx0A        = 2*vdwtype[jnrA+0];
 983
 984             fjx0             = _mm_setzero_pd();
 985             fjy0             = _mm_setzero_pd();
 986             fjz0             = _mm_setzero_pd();
 987
 988             /**************************
 989              * CALCULATE INTERACTIONS *
 990              **************************/
 991
 992             if (gmx_mm_any_lt(rsq00,rcutoff2))
 993             {
 994
 995             r00              = _mm_mul_pd(rsq00,rinv00);
 996
 997             /* Compute parameters for interactions between i and j atoms */
 998             qq00             = _mm_mul_pd(iq0,jq0);
 999             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1000
1001             /* Calculate table index by multiplying r with table scale and truncate to integer */
1002             rt               = _mm_mul_pd(r00,vftabscale);
1003             vfitab           = _mm_cvttpd_epi32(rt);
1004             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1005             vfitab           = _mm_slli_epi32(vfitab,3);
1006
1007             /* REACTION-FIELD ELECTROSTATICS */
1008             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
1009
1010             /* CUBIC SPLINE TABLE DISPERSION */
1011             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1012             F                = _mm_setzero_pd();
1013             GMX_MM_TRANSPOSE2_PD(Y,F);
1014             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1015             H                = _mm_setzero_pd();
1016             GMX_MM_TRANSPOSE2_PD(G,H);
1017             Heps             = _mm_mul_pd(vfeps,H);
1018             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1019             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1020             fvdw6            = _mm_mul_pd(c6_00,FF);
1021
1022             /* CUBIC SPLINE TABLE REPULSION */
1023             vfitab           = _mm_add_epi32(vfitab,ifour);
1024             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1025             F                = _mm_setzero_pd();
1026             GMX_MM_TRANSPOSE2_PD(Y,F);
1027             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1028             H                = _mm_setzero_pd();
1029             GMX_MM_TRANSPOSE2_PD(G,H);
1030             Heps             = _mm_mul_pd(vfeps,H);
1031             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1032             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1033             fvdw12           = _mm_mul_pd(c12_00,FF);
1034             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1035
1036             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1037
1038             fscal            = _mm_add_pd(felec,fvdw);
1039
1040             fscal            = _mm_and_pd(fscal,cutoff_mask);
1041
1042             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1043
1044             /* Calculate temporary vectorial force */
1045             tx               = _mm_mul_pd(fscal,dx00);
1046             ty               = _mm_mul_pd(fscal,dy00);
1047             tz               = _mm_mul_pd(fscal,dz00);
1048
1049             /* Update vectorial force */
1050             fix0             = _mm_add_pd(fix0,tx);
1051             fiy0             = _mm_add_pd(fiy0,ty);
1052             fiz0             = _mm_add_pd(fiz0,tz);
1053
1054             fjx0             = _mm_add_pd(fjx0,tx);
1055             fjy0             = _mm_add_pd(fjy0,ty);
1056             fjz0             = _mm_add_pd(fjz0,tz);
1057
1058             }
1059
1060             /**************************
1061              * CALCULATE INTERACTIONS *
1062              **************************/
1063
1064             if (gmx_mm_any_lt(rsq10,rcutoff2))
1065             {
1066
1067             /* Compute parameters for interactions between i and j atoms */
1068             qq10             = _mm_mul_pd(iq1,jq0);
1069
1070             /* REACTION-FIELD ELECTROSTATICS */
1071             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
1072
1073             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1074
1075             fscal            = felec;
1076
1077             fscal            = _mm_and_pd(fscal,cutoff_mask);
1078
1079             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1080
1081             /* Calculate temporary vectorial force */
1082             tx               = _mm_mul_pd(fscal,dx10);
1083             ty               = _mm_mul_pd(fscal,dy10);
1084             tz               = _mm_mul_pd(fscal,dz10);
1085
1086             /* Update vectorial force */
1087             fix1             = _mm_add_pd(fix1,tx);
1088             fiy1             = _mm_add_pd(fiy1,ty);
1089             fiz1             = _mm_add_pd(fiz1,tz);
1090
1091             fjx0             = _mm_add_pd(fjx0,tx);
1092             fjy0             = _mm_add_pd(fjy0,ty);
1093             fjz0             = _mm_add_pd(fjz0,tz);
1094
1095             }
1096
1097             /**************************
1098              * CALCULATE INTERACTIONS *
1099              **************************/
1100
1101             if (gmx_mm_any_lt(rsq20,rcutoff2))
1102             {
1103
1104             /* Compute parameters for interactions between i and j atoms */
1105             qq20             = _mm_mul_pd(iq2,jq0);
1106
1107             /* REACTION-FIELD ELECTROSTATICS */
1108             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
1109
1110             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1111
1112             fscal            = felec;
1113
1114             fscal            = _mm_and_pd(fscal,cutoff_mask);
1115
1116             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1117
1118             /* Calculate temporary vectorial force */
1119             tx               = _mm_mul_pd(fscal,dx20);
1120             ty               = _mm_mul_pd(fscal,dy20);
1121             tz               = _mm_mul_pd(fscal,dz20);
1122
1123             /* Update vectorial force */
1124             fix2             = _mm_add_pd(fix2,tx);
1125             fiy2             = _mm_add_pd(fiy2,ty);
1126             fiz2             = _mm_add_pd(fiz2,tz);
1127
1128             fjx0             = _mm_add_pd(fjx0,tx);
1129             fjy0             = _mm_add_pd(fjy0,ty);
1130             fjz0             = _mm_add_pd(fjz0,tz);
1131
1132             }
1133
1134             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1135
1136             /* Inner loop uses 120 flops */
1137         }
1138
1139         /* End of innermost loop */
1140
1141         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1142                                               f+i_coord_offset,fshift+i_shift_offset);
1143
1144         /* Increment number of inner iterations */
1145         inneriter                  += j_index_end - j_index_start;
1146
1147         /* Outer loop uses 18 flops */
1148     }
1149
1150     /* Increment number of outer iterations */
1151     outeriter        += nri;
1152
1153     /* Update outer/inner flops */
1154
1155     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*120);
1156 }