src/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecRF_VdwCSTab_GeomW4P1_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_ElecRF_VdwCSTab_GeomW4P1_VF_sse4_1_double
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRF_VdwCSTab_GeomW4P1_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              vdwioffset3;
  73     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  74     int              vdwjidx0A,vdwjidx0B;
  75     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  76     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  77     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  78     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  79     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  80     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     int              nvdwtype;
  83     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  84     int              *vdwtype;
  85     real             *vdwparam;
  86     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  87     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  88     __m128i          vfitab;
  89     __m128i          ifour       = _mm_set1_epi32(4);
  90     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  91     real             *vftab;
  92     __m128d          dummy_mask,cutoff_mask;
  93     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  94     __m128d          one     = _mm_set1_pd(1.0);
  95     __m128d          two     = _mm_set1_pd(2.0);
  96     x                = xx[0];
  97     f                = ff[0];
  98
  99     nri              = nlist->nri;
 100     iinr             = nlist->iinr;
 101     jindex           = nlist->jindex;
 102     jjnr             = nlist->jjnr;
 103     shiftidx         = nlist->shift;
 104     gid              = nlist->gid;
 105     shiftvec         = fr->shift_vec[0];
 106     fshift           = fr->fshift[0];
 107     facel            = _mm_set1_pd(fr->epsfac);
 108     charge           = mdatoms->chargeA;
 109     krf              = _mm_set1_pd(fr->ic->k_rf);
 110     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 111     crf              = _mm_set1_pd(fr->ic->c_rf);
 112     nvdwtype         = fr->ntype;
 113     vdwparam         = fr->nbfp;
 114     vdwtype          = mdatoms->typeA;
 115
 116     vftab            = kernel_data->table_vdw->data;
 117     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 118
 119     /* Setup water-specific parameters */
 120     inr              = nlist->iinr[0];
 121     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 122     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 123     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 124     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 125
 126     /* Avoid stupid compiler warnings */
 127     jnrA = jnrB = 0;
 128     j_coord_offsetA = 0;
 129     j_coord_offsetB = 0;
 130
 131     outeriter        = 0;
 132     inneriter        = 0;
 133
 134     /* Start outer loop over neighborlists */
 135     for(iidx=0; iidx<nri; iidx++)
 136     {
 137         /* Load shift vector for this list */
 138         i_shift_offset   = DIM*shiftidx[iidx];
 139
 140         /* Load limits for loop over neighbors */
 141         j_index_start    = jindex[iidx];
 142         j_index_end      = jindex[iidx+1];
 143
 144         /* Get outer coordinate index */
 145         inr              = iinr[iidx];
 146         i_coord_offset   = DIM*inr;
 147
 148         /* Load i particle coords and add shift vector */
 149         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 150                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 151
 152         fix0             = _mm_setzero_pd();
 153         fiy0             = _mm_setzero_pd();
 154         fiz0             = _mm_setzero_pd();
 155         fix1             = _mm_setzero_pd();
 156         fiy1             = _mm_setzero_pd();
 157         fiz1             = _mm_setzero_pd();
 158         fix2             = _mm_setzero_pd();
 159         fiy2             = _mm_setzero_pd();
 160         fiz2             = _mm_setzero_pd();
 161         fix3             = _mm_setzero_pd();
 162         fiy3             = _mm_setzero_pd();
 163         fiz3             = _mm_setzero_pd();
 164
 165         /* Reset potential sums */
 166         velecsum         = _mm_setzero_pd();
 167         vvdwsum          = _mm_setzero_pd();
 168
 169         /* Start inner kernel loop */
 170         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 171         {
 172
 173             /* Get j neighbor index, and coordinate index */
 174             jnrA             = jjnr[jidx];
 175             jnrB             = jjnr[jidx+1];
 176             j_coord_offsetA  = DIM*jnrA;
 177             j_coord_offsetB  = DIM*jnrB;
 178
 179             /* load j atom coordinates */
 180             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 181                                               &jx0,&jy0,&jz0);
 182
 183             /* Calculate displacement vector */
 184             dx00             = _mm_sub_pd(ix0,jx0);
 185             dy00             = _mm_sub_pd(iy0,jy0);
 186             dz00             = _mm_sub_pd(iz0,jz0);
 187             dx10             = _mm_sub_pd(ix1,jx0);
 188             dy10             = _mm_sub_pd(iy1,jy0);
 189             dz10             = _mm_sub_pd(iz1,jz0);
 190             dx20             = _mm_sub_pd(ix2,jx0);
 191             dy20             = _mm_sub_pd(iy2,jy0);
 192             dz20             = _mm_sub_pd(iz2,jz0);
 193             dx30             = _mm_sub_pd(ix3,jx0);
 194             dy30             = _mm_sub_pd(iy3,jy0);
 195             dz30             = _mm_sub_pd(iz3,jz0);
 196
 197             /* Calculate squared distance and things based on it */
 198             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 199             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 200             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 201             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 202
 203             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 204             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 205             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 206             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 207
 208             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 209             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 210             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 211
 212             /* Load parameters for j particles */
 213             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 214             vdwjidx0A        = 2*vdwtype[jnrA+0];
 215             vdwjidx0B        = 2*vdwtype[jnrB+0];
 216
 217             fjx0             = _mm_setzero_pd();
 218             fjy0             = _mm_setzero_pd();
 219             fjz0             = _mm_setzero_pd();
 220
 221             /**************************
 222              * CALCULATE INTERACTIONS *
 223              **************************/
 224
 225             r00              = _mm_mul_pd(rsq00,rinv00);
 226
 227             /* Compute parameters for interactions between i and j atoms */
 228             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 229                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 230
 231             /* Calculate table index by multiplying r with table scale and truncate to integer */
 232             rt               = _mm_mul_pd(r00,vftabscale);
 233             vfitab           = _mm_cvttpd_epi32(rt);
 234             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 235             vfitab           = _mm_slli_epi32(vfitab,3);
 236
 237             /* CUBIC SPLINE TABLE DISPERSION */
 238             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 239             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 240             GMX_MM_TRANSPOSE2_PD(Y,F);
 241             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 242             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 243             GMX_MM_TRANSPOSE2_PD(G,H);
 244             Heps             = _mm_mul_pd(vfeps,H);
 245             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 246             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 247             vvdw6            = _mm_mul_pd(c6_00,VV);
 248             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 249             fvdw6            = _mm_mul_pd(c6_00,FF);
 250
 251             /* CUBIC SPLINE TABLE REPULSION */
 252             vfitab           = _mm_add_epi32(vfitab,ifour);
 253             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 254             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 255             GMX_MM_TRANSPOSE2_PD(Y,F);
 256             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 257             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 258             GMX_MM_TRANSPOSE2_PD(G,H);
 259             Heps             = _mm_mul_pd(vfeps,H);
 260             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 261             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 262             vvdw12           = _mm_mul_pd(c12_00,VV);
 263             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 264             fvdw12           = _mm_mul_pd(c12_00,FF);
 265             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 266             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 267
 268             /* Update potential sum for this i atom from the interaction with this j atom. */
 269             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 270
 271             fscal            = fvdw;
 272
 273             /* Calculate temporary vectorial force */
 274             tx               = _mm_mul_pd(fscal,dx00);
 275             ty               = _mm_mul_pd(fscal,dy00);
 276             tz               = _mm_mul_pd(fscal,dz00);
 277
 278             /* Update vectorial force */
 279             fix0             = _mm_add_pd(fix0,tx);
 280             fiy0             = _mm_add_pd(fiy0,ty);
 281             fiz0             = _mm_add_pd(fiz0,tz);
 282
 283             fjx0             = _mm_add_pd(fjx0,tx);
 284             fjy0             = _mm_add_pd(fjy0,ty);
 285             fjz0             = _mm_add_pd(fjz0,tz);
 286
 287             /**************************
 288              * CALCULATE INTERACTIONS *
 289              **************************/
 290
 291             /* Compute parameters for interactions between i and j atoms */
 292             qq10             = _mm_mul_pd(iq1,jq0);
 293
 294             /* REACTION-FIELD ELECTROSTATICS */
 295             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 296             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 297
 298             /* Update potential sum for this i atom from the interaction with this j atom. */
 299             velecsum         = _mm_add_pd(velecsum,velec);
 300
 301             fscal            = felec;
 302
 303             /* Calculate temporary vectorial force */
 304             tx               = _mm_mul_pd(fscal,dx10);
 305             ty               = _mm_mul_pd(fscal,dy10);
 306             tz               = _mm_mul_pd(fscal,dz10);
 307
 308             /* Update vectorial force */
 309             fix1             = _mm_add_pd(fix1,tx);
 310             fiy1             = _mm_add_pd(fiy1,ty);
 311             fiz1             = _mm_add_pd(fiz1,tz);
 312
 313             fjx0             = _mm_add_pd(fjx0,tx);
 314             fjy0             = _mm_add_pd(fjy0,ty);
 315             fjz0             = _mm_add_pd(fjz0,tz);
 316
 317             /**************************
 318              * CALCULATE INTERACTIONS *
 319              **************************/
 320
 321             /* Compute parameters for interactions between i and j atoms */
 322             qq20             = _mm_mul_pd(iq2,jq0);
 323
 324             /* REACTION-FIELD ELECTROSTATICS */
 325             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 326             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 327
 328             /* Update potential sum for this i atom from the interaction with this j atom. */
 329             velecsum         = _mm_add_pd(velecsum,velec);
 330
 331             fscal            = felec;
 332
 333             /* Calculate temporary vectorial force */
 334             tx               = _mm_mul_pd(fscal,dx20);
 335             ty               = _mm_mul_pd(fscal,dy20);
 336             tz               = _mm_mul_pd(fscal,dz20);
 337
 338             /* Update vectorial force */
 339             fix2             = _mm_add_pd(fix2,tx);
 340             fiy2             = _mm_add_pd(fiy2,ty);
 341             fiz2             = _mm_add_pd(fiz2,tz);
 342
 343             fjx0             = _mm_add_pd(fjx0,tx);
 344             fjy0             = _mm_add_pd(fjy0,ty);
 345             fjz0             = _mm_add_pd(fjz0,tz);
 346
 347             /**************************
 348              * CALCULATE INTERACTIONS *
 349              **************************/
 350
 351             /* Compute parameters for interactions between i and j atoms */
 352             qq30             = _mm_mul_pd(iq3,jq0);
 353
 354             /* REACTION-FIELD ELECTROSTATICS */
 355             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_add_pd(rinv30,_mm_mul_pd(krf,rsq30)),crf));
 356             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
 357
 358             /* Update potential sum for this i atom from the interaction with this j atom. */
 359             velecsum         = _mm_add_pd(velecsum,velec);
 360
 361             fscal            = felec;
 362
 363             /* Calculate temporary vectorial force */
 364             tx               = _mm_mul_pd(fscal,dx30);
 365             ty               = _mm_mul_pd(fscal,dy30);
 366             tz               = _mm_mul_pd(fscal,dz30);
 367
 368             /* Update vectorial force */
 369             fix3             = _mm_add_pd(fix3,tx);
 370             fiy3             = _mm_add_pd(fiy3,ty);
 371             fiz3             = _mm_add_pd(fiz3,tz);
 372
 373             fjx0             = _mm_add_pd(fjx0,tx);
 374             fjy0             = _mm_add_pd(fjy0,ty);
 375             fjz0             = _mm_add_pd(fjz0,tz);
 376
 377             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 378
 379             /* Inner loop uses 155 flops */
 380         }
 381
 382         if(jidx<j_index_end)
 383         {
 384
 385             jnrA             = jjnr[jidx];
 386             j_coord_offsetA  = DIM*jnrA;
 387
 388             /* load j atom coordinates */
 389             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 390                                               &jx0,&jy0,&jz0);
 391
 392             /* Calculate displacement vector */
 393             dx00             = _mm_sub_pd(ix0,jx0);
 394             dy00             = _mm_sub_pd(iy0,jy0);
 395             dz00             = _mm_sub_pd(iz0,jz0);
 396             dx10             = _mm_sub_pd(ix1,jx0);
 397             dy10             = _mm_sub_pd(iy1,jy0);
 398             dz10             = _mm_sub_pd(iz1,jz0);
 399             dx20             = _mm_sub_pd(ix2,jx0);
 400             dy20             = _mm_sub_pd(iy2,jy0);
 401             dz20             = _mm_sub_pd(iz2,jz0);
 402             dx30             = _mm_sub_pd(ix3,jx0);
 403             dy30             = _mm_sub_pd(iy3,jy0);
 404             dz30             = _mm_sub_pd(iz3,jz0);
 405
 406             /* Calculate squared distance and things based on it */
 407             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 408             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 409             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 410             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 411
 412             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 413             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 414             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 415             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 416
 417             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 418             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 419             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 420
 421             /* Load parameters for j particles */
 422             jq0              = _mm_load_sd(charge+jnrA+0);
 423             vdwjidx0A        = 2*vdwtype[jnrA+0];
 424
 425             fjx0             = _mm_setzero_pd();
 426             fjy0             = _mm_setzero_pd();
 427             fjz0             = _mm_setzero_pd();
 428
 429             /**************************
 430              * CALCULATE INTERACTIONS *
 431              **************************/
 432
 433             r00              = _mm_mul_pd(rsq00,rinv00);
 434
 435             /* Compute parameters for interactions between i and j atoms */
 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             /* CUBIC SPLINE TABLE DISPERSION */
 445             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 446             F                = _mm_setzero_pd();
 447             GMX_MM_TRANSPOSE2_PD(Y,F);
 448             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 449             H                = _mm_setzero_pd();
 450             GMX_MM_TRANSPOSE2_PD(G,H);
 451             Heps             = _mm_mul_pd(vfeps,H);
 452             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 453             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 454             vvdw6            = _mm_mul_pd(c6_00,VV);
 455             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 456             fvdw6            = _mm_mul_pd(c6_00,FF);
 457
 458             /* CUBIC SPLINE TABLE REPULSION */
 459             vfitab           = _mm_add_epi32(vfitab,ifour);
 460             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 461             F                = _mm_setzero_pd();
 462             GMX_MM_TRANSPOSE2_PD(Y,F);
 463             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 464             H                = _mm_setzero_pd();
 465             GMX_MM_TRANSPOSE2_PD(G,H);
 466             Heps             = _mm_mul_pd(vfeps,H);
 467             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 468             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 469             vvdw12           = _mm_mul_pd(c12_00,VV);
 470             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 471             fvdw12           = _mm_mul_pd(c12_00,FF);
 472             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 473             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 474
 475             /* Update potential sum for this i atom from the interaction with this j atom. */
 476             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 477             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 478
 479             fscal            = fvdw;
 480
 481             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 482
 483             /* Calculate temporary vectorial force */
 484             tx               = _mm_mul_pd(fscal,dx00);
 485             ty               = _mm_mul_pd(fscal,dy00);
 486             tz               = _mm_mul_pd(fscal,dz00);
 487
 488             /* Update vectorial force */
 489             fix0             = _mm_add_pd(fix0,tx);
 490             fiy0             = _mm_add_pd(fiy0,ty);
 491             fiz0             = _mm_add_pd(fiz0,tz);
 492
 493             fjx0             = _mm_add_pd(fjx0,tx);
 494             fjy0             = _mm_add_pd(fjy0,ty);
 495             fjz0             = _mm_add_pd(fjz0,tz);
 496
 497             /**************************
 498              * CALCULATE INTERACTIONS *
 499              **************************/
 500
 501             /* Compute parameters for interactions between i and j atoms */
 502             qq10             = _mm_mul_pd(iq1,jq0);
 503
 504             /* REACTION-FIELD ELECTROSTATICS */
 505             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 506             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 507
 508             /* Update potential sum for this i atom from the interaction with this j atom. */
 509             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 510             velecsum         = _mm_add_pd(velecsum,velec);
 511
 512             fscal            = felec;
 513
 514             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 515
 516             /* Calculate temporary vectorial force */
 517             tx               = _mm_mul_pd(fscal,dx10);
 518             ty               = _mm_mul_pd(fscal,dy10);
 519             tz               = _mm_mul_pd(fscal,dz10);
 520
 521             /* Update vectorial force */
 522             fix1             = _mm_add_pd(fix1,tx);
 523             fiy1             = _mm_add_pd(fiy1,ty);
 524             fiz1             = _mm_add_pd(fiz1,tz);
 525
 526             fjx0             = _mm_add_pd(fjx0,tx);
 527             fjy0             = _mm_add_pd(fjy0,ty);
 528             fjz0             = _mm_add_pd(fjz0,tz);
 529
 530             /**************************
 531              * CALCULATE INTERACTIONS *
 532              **************************/
 533
 534             /* Compute parameters for interactions between i and j atoms */
 535             qq20             = _mm_mul_pd(iq2,jq0);
 536
 537             /* REACTION-FIELD ELECTROSTATICS */
 538             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 539             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 540
 541             /* Update potential sum for this i atom from the interaction with this j atom. */
 542             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 543             velecsum         = _mm_add_pd(velecsum,velec);
 544
 545             fscal            = felec;
 546
 547             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 548
 549             /* Calculate temporary vectorial force */
 550             tx               = _mm_mul_pd(fscal,dx20);
 551             ty               = _mm_mul_pd(fscal,dy20);
 552             tz               = _mm_mul_pd(fscal,dz20);
 553
 554             /* Update vectorial force */
 555             fix2             = _mm_add_pd(fix2,tx);
 556             fiy2             = _mm_add_pd(fiy2,ty);
 557             fiz2             = _mm_add_pd(fiz2,tz);
 558
 559             fjx0             = _mm_add_pd(fjx0,tx);
 560             fjy0             = _mm_add_pd(fjy0,ty);
 561             fjz0             = _mm_add_pd(fjz0,tz);
 562
 563             /**************************
 564              * CALCULATE INTERACTIONS *
 565              **************************/
 566
 567             /* Compute parameters for interactions between i and j atoms */
 568             qq30             = _mm_mul_pd(iq3,jq0);
 569
 570             /* REACTION-FIELD ELECTROSTATICS */
 571             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_add_pd(rinv30,_mm_mul_pd(krf,rsq30)),crf));
 572             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
 573
 574             /* Update potential sum for this i atom from the interaction with this j atom. */
 575             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 576             velecsum         = _mm_add_pd(velecsum,velec);
 577
 578             fscal            = felec;
 579
 580             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 581
 582             /* Calculate temporary vectorial force */
 583             tx               = _mm_mul_pd(fscal,dx30);
 584             ty               = _mm_mul_pd(fscal,dy30);
 585             tz               = _mm_mul_pd(fscal,dz30);
 586
 587             /* Update vectorial force */
 588             fix3             = _mm_add_pd(fix3,tx);
 589             fiy3             = _mm_add_pd(fiy3,ty);
 590             fiz3             = _mm_add_pd(fiz3,tz);
 591
 592             fjx0             = _mm_add_pd(fjx0,tx);
 593             fjy0             = _mm_add_pd(fjy0,ty);
 594             fjz0             = _mm_add_pd(fjz0,tz);
 595
 596             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 597
 598             /* Inner loop uses 155 flops */
 599         }
 600
 601         /* End of innermost loop */
 602
 603         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 604                                               f+i_coord_offset,fshift+i_shift_offset);
 605
 606         ggid                        = gid[iidx];
 607         /* Update potential energies */
 608         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 609         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 610
 611         /* Increment number of inner iterations */
 612         inneriter                  += j_index_end - j_index_start;
 613
 614         /* Outer loop uses 26 flops */
 615     }
 616
 617     /* Increment number of outer iterations */
 618     outeriter        += nri;
 619
 620     /* Update outer/inner flops */
 621
 622     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*155);
 623 }
 624 /*
 625  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_sse4_1_double
 626  * Electrostatics interaction: ReactionField
 627  * VdW interaction:            CubicSplineTable
 628  * Geometry:                   Water4-Particle
 629  * Calculate force/pot:        Force
 630  */
 631 void
 632 nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_sse4_1_double
 633                     (t_nblist * gmx_restrict                nlist,
 634                      rvec * gmx_restrict                    xx,
 635                      rvec * gmx_restrict                    ff,
 636                      t_forcerec * gmx_restrict              fr,
 637                      t_mdatoms * gmx_restrict               mdatoms,
 638                      nb_kernel_data_t * gmx_restrict        kernel_data,
 639                      t_nrnb * gmx_restrict                  nrnb)
 640 {
 641     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 642      * just 0 for non-waters.
 643      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 644      * jnr indices corresponding to data put in the four positions in the SIMD register.
 645      */
 646     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 647     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 648     int              jnrA,jnrB;
 649     int              j_coord_offsetA,j_coord_offsetB;
 650     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 651     real             rcutoff_scalar;
 652     real             *shiftvec,*fshift,*x,*f;
 653     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 654     int              vdwioffset0;
 655     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 656     int              vdwioffset1;
 657     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 658     int              vdwioffset2;
 659     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 660     int              vdwioffset3;
 661     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 662     int              vdwjidx0A,vdwjidx0B;
 663     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 664     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 665     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 666     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 667     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 668     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 669     real             *charge;
 670     int              nvdwtype;
 671     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 672     int              *vdwtype;
 673     real             *vdwparam;
 674     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 675     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 676     __m128i          vfitab;
 677     __m128i          ifour       = _mm_set1_epi32(4);
 678     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 679     real             *vftab;
 680     __m128d          dummy_mask,cutoff_mask;
 681     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 682     __m128d          one     = _mm_set1_pd(1.0);
 683     __m128d          two     = _mm_set1_pd(2.0);
 684     x                = xx[0];
 685     f                = ff[0];
 686
 687     nri              = nlist->nri;
 688     iinr             = nlist->iinr;
 689     jindex           = nlist->jindex;
 690     jjnr             = nlist->jjnr;
 691     shiftidx         = nlist->shift;
 692     gid              = nlist->gid;
 693     shiftvec         = fr->shift_vec[0];
 694     fshift           = fr->fshift[0];
 695     facel            = _mm_set1_pd(fr->epsfac);
 696     charge           = mdatoms->chargeA;
 697     krf              = _mm_set1_pd(fr->ic->k_rf);
 698     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 699     crf              = _mm_set1_pd(fr->ic->c_rf);
 700     nvdwtype         = fr->ntype;
 701     vdwparam         = fr->nbfp;
 702     vdwtype          = mdatoms->typeA;
 703
 704     vftab            = kernel_data->table_vdw->data;
 705     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 706
 707     /* Setup water-specific parameters */
 708     inr              = nlist->iinr[0];
 709     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 710     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 711     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 712     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 713
 714     /* Avoid stupid compiler warnings */
 715     jnrA = jnrB = 0;
 716     j_coord_offsetA = 0;
 717     j_coord_offsetB = 0;
 718
 719     outeriter        = 0;
 720     inneriter        = 0;
 721
 722     /* Start outer loop over neighborlists */
 723     for(iidx=0; iidx<nri; iidx++)
 724     {
 725         /* Load shift vector for this list */
 726         i_shift_offset   = DIM*shiftidx[iidx];
 727
 728         /* Load limits for loop over neighbors */
 729         j_index_start    = jindex[iidx];
 730         j_index_end      = jindex[iidx+1];
 731
 732         /* Get outer coordinate index */
 733         inr              = iinr[iidx];
 734         i_coord_offset   = DIM*inr;
 735
 736         /* Load i particle coords and add shift vector */
 737         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 738                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 739
 740         fix0             = _mm_setzero_pd();
 741         fiy0             = _mm_setzero_pd();
 742         fiz0             = _mm_setzero_pd();
 743         fix1             = _mm_setzero_pd();
 744         fiy1             = _mm_setzero_pd();
 745         fiz1             = _mm_setzero_pd();
 746         fix2             = _mm_setzero_pd();
 747         fiy2             = _mm_setzero_pd();
 748         fiz2             = _mm_setzero_pd();
 749         fix3             = _mm_setzero_pd();
 750         fiy3             = _mm_setzero_pd();
 751         fiz3             = _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             dx30             = _mm_sub_pd(ix3,jx0);
 778             dy30             = _mm_sub_pd(iy3,jy0);
 779             dz30             = _mm_sub_pd(iz3,jz0);
 780
 781             /* Calculate squared distance and things based on it */
 782             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 783             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 784             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 785             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 786
 787             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 788             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 789             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 790             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 791
 792             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 793             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 794             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 795
 796             /* Load parameters for j particles */
 797             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 798             vdwjidx0A        = 2*vdwtype[jnrA+0];
 799             vdwjidx0B        = 2*vdwtype[jnrB+0];
 800
 801             fjx0             = _mm_setzero_pd();
 802             fjy0             = _mm_setzero_pd();
 803             fjz0             = _mm_setzero_pd();
 804
 805             /**************************
 806              * CALCULATE INTERACTIONS *
 807              **************************/
 808
 809             r00              = _mm_mul_pd(rsq00,rinv00);
 810
 811             /* Compute parameters for interactions between i and j atoms */
 812             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 813                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 814
 815             /* Calculate table index by multiplying r with table scale and truncate to integer */
 816             rt               = _mm_mul_pd(r00,vftabscale);
 817             vfitab           = _mm_cvttpd_epi32(rt);
 818             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 819             vfitab           = _mm_slli_epi32(vfitab,3);
 820
 821             /* CUBIC SPLINE TABLE DISPERSION */
 822             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 823             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 824             GMX_MM_TRANSPOSE2_PD(Y,F);
 825             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 826             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 827             GMX_MM_TRANSPOSE2_PD(G,H);
 828             Heps             = _mm_mul_pd(vfeps,H);
 829             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 830             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 831             fvdw6            = _mm_mul_pd(c6_00,FF);
 832
 833             /* CUBIC SPLINE TABLE REPULSION */
 834             vfitab           = _mm_add_epi32(vfitab,ifour);
 835             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 836             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 837             GMX_MM_TRANSPOSE2_PD(Y,F);
 838             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 839             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 840             GMX_MM_TRANSPOSE2_PD(G,H);
 841             Heps             = _mm_mul_pd(vfeps,H);
 842             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 843             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 844             fvdw12           = _mm_mul_pd(c12_00,FF);
 845             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 846
 847             fscal            = fvdw;
 848
 849             /* Calculate temporary vectorial force */
 850             tx               = _mm_mul_pd(fscal,dx00);
 851             ty               = _mm_mul_pd(fscal,dy00);
 852             tz               = _mm_mul_pd(fscal,dz00);
 853
 854             /* Update vectorial force */
 855             fix0             = _mm_add_pd(fix0,tx);
 856             fiy0             = _mm_add_pd(fiy0,ty);
 857             fiz0             = _mm_add_pd(fiz0,tz);
 858
 859             fjx0             = _mm_add_pd(fjx0,tx);
 860             fjy0             = _mm_add_pd(fjy0,ty);
 861             fjz0             = _mm_add_pd(fjz0,tz);
 862
 863             /**************************
 864              * CALCULATE INTERACTIONS *
 865              **************************/
 866
 867             /* Compute parameters for interactions between i and j atoms */
 868             qq10             = _mm_mul_pd(iq1,jq0);
 869
 870             /* REACTION-FIELD ELECTROSTATICS */
 871             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 872
 873             fscal            = felec;
 874
 875             /* Calculate temporary vectorial force */
 876             tx               = _mm_mul_pd(fscal,dx10);
 877             ty               = _mm_mul_pd(fscal,dy10);
 878             tz               = _mm_mul_pd(fscal,dz10);
 879
 880             /* Update vectorial force */
 881             fix1             = _mm_add_pd(fix1,tx);
 882             fiy1             = _mm_add_pd(fiy1,ty);
 883             fiz1             = _mm_add_pd(fiz1,tz);
 884
 885             fjx0             = _mm_add_pd(fjx0,tx);
 886             fjy0             = _mm_add_pd(fjy0,ty);
 887             fjz0             = _mm_add_pd(fjz0,tz);
 888
 889             /**************************
 890              * CALCULATE INTERACTIONS *
 891              **************************/
 892
 893             /* Compute parameters for interactions between i and j atoms */
 894             qq20             = _mm_mul_pd(iq2,jq0);
 895
 896             /* REACTION-FIELD ELECTROSTATICS */
 897             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 898
 899             fscal            = felec;
 900
 901             /* Calculate temporary vectorial force */
 902             tx               = _mm_mul_pd(fscal,dx20);
 903             ty               = _mm_mul_pd(fscal,dy20);
 904             tz               = _mm_mul_pd(fscal,dz20);
 905
 906             /* Update vectorial force */
 907             fix2             = _mm_add_pd(fix2,tx);
 908             fiy2             = _mm_add_pd(fiy2,ty);
 909             fiz2             = _mm_add_pd(fiz2,tz);
 910
 911             fjx0             = _mm_add_pd(fjx0,tx);
 912             fjy0             = _mm_add_pd(fjy0,ty);
 913             fjz0             = _mm_add_pd(fjz0,tz);
 914
 915             /**************************
 916              * CALCULATE INTERACTIONS *
 917              **************************/
 918
 919             /* Compute parameters for interactions between i and j atoms */
 920             qq30             = _mm_mul_pd(iq3,jq0);
 921
 922             /* REACTION-FIELD ELECTROSTATICS */
 923             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
 924
 925             fscal            = felec;
 926
 927             /* Calculate temporary vectorial force */
 928             tx               = _mm_mul_pd(fscal,dx30);
 929             ty               = _mm_mul_pd(fscal,dy30);
 930             tz               = _mm_mul_pd(fscal,dz30);
 931
 932             /* Update vectorial force */
 933             fix3             = _mm_add_pd(fix3,tx);
 934             fiy3             = _mm_add_pd(fiy3,ty);
 935             fiz3             = _mm_add_pd(fiz3,tz);
 936
 937             fjx0             = _mm_add_pd(fjx0,tx);
 938             fjy0             = _mm_add_pd(fjy0,ty);
 939             fjz0             = _mm_add_pd(fjz0,tz);
 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 132 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             dx30             = _mm_sub_pd(ix3,jx0);
 967             dy30             = _mm_sub_pd(iy3,jy0);
 968             dz30             = _mm_sub_pd(iz3,jz0);
 969
 970             /* Calculate squared distance and things based on it */
 971             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 972             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 973             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 974             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 975
 976             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 977             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 978             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 979             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 980
 981             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 982             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 983             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 984
 985             /* Load parameters for j particles */
 986             jq0              = _mm_load_sd(charge+jnrA+0);
 987             vdwjidx0A        = 2*vdwtype[jnrA+0];
 988
 989             fjx0             = _mm_setzero_pd();
 990             fjy0             = _mm_setzero_pd();
 991             fjz0             = _mm_setzero_pd();
 992
 993             /**************************
 994              * CALCULATE INTERACTIONS *
 995              **************************/
 996
 997             r00              = _mm_mul_pd(rsq00,rinv00);
 998
 999             /* Compute parameters for interactions between i and j atoms */
1000             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1001
1002             /* Calculate table index by multiplying r with table scale and truncate to integer */
1003             rt               = _mm_mul_pd(r00,vftabscale);
1004             vfitab           = _mm_cvttpd_epi32(rt);
1005             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1006             vfitab           = _mm_slli_epi32(vfitab,3);
1007
1008             /* CUBIC SPLINE TABLE DISPERSION */
1009             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1010             F                = _mm_setzero_pd();
1011             GMX_MM_TRANSPOSE2_PD(Y,F);
1012             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1013             H                = _mm_setzero_pd();
1014             GMX_MM_TRANSPOSE2_PD(G,H);
1015             Heps             = _mm_mul_pd(vfeps,H);
1016             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1017             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1018             fvdw6            = _mm_mul_pd(c6_00,FF);
1019
1020             /* CUBIC SPLINE TABLE REPULSION */
1021             vfitab           = _mm_add_epi32(vfitab,ifour);
1022             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1023             F                = _mm_setzero_pd();
1024             GMX_MM_TRANSPOSE2_PD(Y,F);
1025             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1026             H                = _mm_setzero_pd();
1027             GMX_MM_TRANSPOSE2_PD(G,H);
1028             Heps             = _mm_mul_pd(vfeps,H);
1029             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1030             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1031             fvdw12           = _mm_mul_pd(c12_00,FF);
1032             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1033
1034             fscal            = fvdw;
1035
1036             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1037
1038             /* Calculate temporary vectorial force */
1039             tx               = _mm_mul_pd(fscal,dx00);
1040             ty               = _mm_mul_pd(fscal,dy00);
1041             tz               = _mm_mul_pd(fscal,dz00);
1042
1043             /* Update vectorial force */
1044             fix0             = _mm_add_pd(fix0,tx);
1045             fiy0             = _mm_add_pd(fiy0,ty);
1046             fiz0             = _mm_add_pd(fiz0,tz);
1047
1048             fjx0             = _mm_add_pd(fjx0,tx);
1049             fjy0             = _mm_add_pd(fjy0,ty);
1050             fjz0             = _mm_add_pd(fjz0,tz);
1051
1052             /**************************
1053              * CALCULATE INTERACTIONS *
1054              **************************/
1055
1056             /* Compute parameters for interactions between i and j atoms */
1057             qq10             = _mm_mul_pd(iq1,jq0);
1058
1059             /* REACTION-FIELD ELECTROSTATICS */
1060             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
1061
1062             fscal            = felec;
1063
1064             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1065
1066             /* Calculate temporary vectorial force */
1067             tx               = _mm_mul_pd(fscal,dx10);
1068             ty               = _mm_mul_pd(fscal,dy10);
1069             tz               = _mm_mul_pd(fscal,dz10);
1070
1071             /* Update vectorial force */
1072             fix1             = _mm_add_pd(fix1,tx);
1073             fiy1             = _mm_add_pd(fiy1,ty);
1074             fiz1             = _mm_add_pd(fiz1,tz);
1075
1076             fjx0             = _mm_add_pd(fjx0,tx);
1077             fjy0             = _mm_add_pd(fjy0,ty);
1078             fjz0             = _mm_add_pd(fjz0,tz);
1079
1080             /**************************
1081              * CALCULATE INTERACTIONS *
1082              **************************/
1083
1084             /* Compute parameters for interactions between i and j atoms */
1085             qq20             = _mm_mul_pd(iq2,jq0);
1086
1087             /* REACTION-FIELD ELECTROSTATICS */
1088             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
1089
1090             fscal            = felec;
1091
1092             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1093
1094             /* Calculate temporary vectorial force */
1095             tx               = _mm_mul_pd(fscal,dx20);
1096             ty               = _mm_mul_pd(fscal,dy20);
1097             tz               = _mm_mul_pd(fscal,dz20);
1098
1099             /* Update vectorial force */
1100             fix2             = _mm_add_pd(fix2,tx);
1101             fiy2             = _mm_add_pd(fiy2,ty);
1102             fiz2             = _mm_add_pd(fiz2,tz);
1103
1104             fjx0             = _mm_add_pd(fjx0,tx);
1105             fjy0             = _mm_add_pd(fjy0,ty);
1106             fjz0             = _mm_add_pd(fjz0,tz);
1107
1108             /**************************
1109              * CALCULATE INTERACTIONS *
1110              **************************/
1111
1112             /* Compute parameters for interactions between i and j atoms */
1113             qq30             = _mm_mul_pd(iq3,jq0);
1114
1115             /* REACTION-FIELD ELECTROSTATICS */
1116             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
1117
1118             fscal            = felec;
1119
1120             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1121
1122             /* Calculate temporary vectorial force */
1123             tx               = _mm_mul_pd(fscal,dx30);
1124             ty               = _mm_mul_pd(fscal,dy30);
1125             tz               = _mm_mul_pd(fscal,dz30);
1126
1127             /* Update vectorial force */
1128             fix3             = _mm_add_pd(fix3,tx);
1129             fiy3             = _mm_add_pd(fiy3,ty);
1130             fiz3             = _mm_add_pd(fiz3,tz);
1131
1132             fjx0             = _mm_add_pd(fjx0,tx);
1133             fjy0             = _mm_add_pd(fjy0,ty);
1134             fjz0             = _mm_add_pd(fjz0,tz);
1135
1136             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1137
1138             /* Inner loop uses 132 flops */
1139         }
1140
1141         /* End of innermost loop */
1142
1143         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1144                                               f+i_coord_offset,fshift+i_shift_offset);
1145
1146         /* Increment number of inner iterations */
1147         inneriter                  += j_index_end - j_index_start;
1148
1149         /* Outer loop uses 24 flops */
1150     }
1151
1152     /* Increment number of outer iterations */
1153     outeriter        += nri;
1154
1155     /* Update outer/inner flops */
1156
1157     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*132);
1158 }