src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_sse4_1_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse4_1_single 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_single.h"
  34 #include "kernelutil_x86_sse4_1_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_VF_sse4_1_single
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_VF_sse4_1_single
  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,C,D refer to j loop unrolling done with SSE, e.g. for the four 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,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwioffset1;
  72     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  73     int              vdwioffset2;
  74     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  75     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  76     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  77     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  78     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  79     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  80     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     int              nvdwtype;
  83     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  84     int              *vdwtype;
  85     real             *vdwparam;
  86     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  87     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  88     __m128i          vfitab;
  89     __m128i          ifour       = _mm_set1_epi32(4);
  90     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  91     real             *vftab;
  92     __m128           dummy_mask,cutoff_mask;
  93     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  94     __m128           one     = _mm_set1_ps(1.0);
  95     __m128           two     = _mm_set1_ps(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_ps(fr->epsfac);
 108     charge           = mdatoms->chargeA;
 109     nvdwtype         = fr->ntype;
 110     vdwparam         = fr->nbfp;
 111     vdwtype          = mdatoms->typeA;
 112
 113     vftab            = kernel_data->table_elec_vdw->data;
 114     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 115
 116     /* Setup water-specific parameters */
 117     inr              = nlist->iinr[0];
 118     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 119     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 120     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 121     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 122
 123     /* Avoid stupid compiler warnings */
 124     jnrA = jnrB = jnrC = jnrD = 0;
 125     j_coord_offsetA = 0;
 126     j_coord_offsetB = 0;
 127     j_coord_offsetC = 0;
 128     j_coord_offsetD = 0;
 129
 130     outeriter        = 0;
 131     inneriter        = 0;
 132
 133     for(iidx=0;iidx<4*DIM;iidx++)
 134     {
 135         scratch[iidx] = 0.0;
 136     }
 137
 138     /* Start outer loop over neighborlists */
 139     for(iidx=0; iidx<nri; iidx++)
 140     {
 141         /* Load shift vector for this list */
 142         i_shift_offset   = DIM*shiftidx[iidx];
 143
 144         /* Load limits for loop over neighbors */
 145         j_index_start    = jindex[iidx];
 146         j_index_end      = jindex[iidx+1];
 147
 148         /* Get outer coordinate index */
 149         inr              = iinr[iidx];
 150         i_coord_offset   = DIM*inr;
 151
 152         /* Load i particle coords and add shift vector */
 153         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 154                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 155
 156         fix0             = _mm_setzero_ps();
 157         fiy0             = _mm_setzero_ps();
 158         fiz0             = _mm_setzero_ps();
 159         fix1             = _mm_setzero_ps();
 160         fiy1             = _mm_setzero_ps();
 161         fiz1             = _mm_setzero_ps();
 162         fix2             = _mm_setzero_ps();
 163         fiy2             = _mm_setzero_ps();
 164         fiz2             = _mm_setzero_ps();
 165
 166         /* Reset potential sums */
 167         velecsum         = _mm_setzero_ps();
 168         vvdwsum          = _mm_setzero_ps();
 169
 170         /* Start inner kernel loop */
 171         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 172         {
 173
 174             /* Get j neighbor index, and coordinate index */
 175             jnrA             = jjnr[jidx];
 176             jnrB             = jjnr[jidx+1];
 177             jnrC             = jjnr[jidx+2];
 178             jnrD             = jjnr[jidx+3];
 179             j_coord_offsetA  = DIM*jnrA;
 180             j_coord_offsetB  = DIM*jnrB;
 181             j_coord_offsetC  = DIM*jnrC;
 182             j_coord_offsetD  = DIM*jnrD;
 183
 184             /* load j atom coordinates */
 185             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 186                                               x+j_coord_offsetC,x+j_coord_offsetD,
 187                                               &jx0,&jy0,&jz0);
 188
 189             /* Calculate displacement vector */
 190             dx00             = _mm_sub_ps(ix0,jx0);
 191             dy00             = _mm_sub_ps(iy0,jy0);
 192             dz00             = _mm_sub_ps(iz0,jz0);
 193             dx10             = _mm_sub_ps(ix1,jx0);
 194             dy10             = _mm_sub_ps(iy1,jy0);
 195             dz10             = _mm_sub_ps(iz1,jz0);
 196             dx20             = _mm_sub_ps(ix2,jx0);
 197             dy20             = _mm_sub_ps(iy2,jy0);
 198             dz20             = _mm_sub_ps(iz2,jz0);
 199
 200             /* Calculate squared distance and things based on it */
 201             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 202             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 203             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 204
 205             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 206             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 207             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 208
 209             /* Load parameters for j particles */
 210             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 211                                                               charge+jnrC+0,charge+jnrD+0);
 212             vdwjidx0A        = 2*vdwtype[jnrA+0];
 213             vdwjidx0B        = 2*vdwtype[jnrB+0];
 214             vdwjidx0C        = 2*vdwtype[jnrC+0];
 215             vdwjidx0D        = 2*vdwtype[jnrD+0];
 216
 217             fjx0             = _mm_setzero_ps();
 218             fjy0             = _mm_setzero_ps();
 219             fjz0             = _mm_setzero_ps();
 220
 221             /**************************
 222              * CALCULATE INTERACTIONS *
 223              **************************/
 224
 225             r00              = _mm_mul_ps(rsq00,rinv00);
 226
 227             /* Compute parameters for interactions between i and j atoms */
 228             qq00             = _mm_mul_ps(iq0,jq0);
 229             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 230                                          vdwparam+vdwioffset0+vdwjidx0B,
 231                                          vdwparam+vdwioffset0+vdwjidx0C,
 232                                          vdwparam+vdwioffset0+vdwjidx0D,
 233                                          &c6_00,&c12_00);
 234
 235             /* Calculate table index by multiplying r with table scale and truncate to integer */
 236             rt               = _mm_mul_ps(r00,vftabscale);
 237             vfitab           = _mm_cvttps_epi32(rt);
 238             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 239             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 240
 241             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 242             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 243             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 244             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 245             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 246             _MM_TRANSPOSE4_PS(Y,F,G,H);
 247             Heps             = _mm_mul_ps(vfeps,H);
 248             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 249             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 250             velec            = _mm_mul_ps(qq00,VV);
 251             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 252             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 253
 254             /* CUBIC SPLINE TABLE DISPERSION */
 255             vfitab           = _mm_add_epi32(vfitab,ifour);
 256             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 257             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 258             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 259             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 260             _MM_TRANSPOSE4_PS(Y,F,G,H);
 261             Heps             = _mm_mul_ps(vfeps,H);
 262             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 263             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 264             vvdw6            = _mm_mul_ps(c6_00,VV);
 265             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 266             fvdw6            = _mm_mul_ps(c6_00,FF);
 267
 268             /* CUBIC SPLINE TABLE REPULSION */
 269             vfitab           = _mm_add_epi32(vfitab,ifour);
 270             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 271             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 272             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 273             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 274             _MM_TRANSPOSE4_PS(Y,F,G,H);
 275             Heps             = _mm_mul_ps(vfeps,H);
 276             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 277             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 278             vvdw12           = _mm_mul_ps(c12_00,VV);
 279             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 280             fvdw12           = _mm_mul_ps(c12_00,FF);
 281             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 282             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 283
 284             /* Update potential sum for this i atom from the interaction with this j atom. */
 285             velecsum         = _mm_add_ps(velecsum,velec);
 286             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 287
 288             fscal            = _mm_add_ps(felec,fvdw);
 289
 290             /* Calculate temporary vectorial force */
 291             tx               = _mm_mul_ps(fscal,dx00);
 292             ty               = _mm_mul_ps(fscal,dy00);
 293             tz               = _mm_mul_ps(fscal,dz00);
 294
 295             /* Update vectorial force */
 296             fix0             = _mm_add_ps(fix0,tx);
 297             fiy0             = _mm_add_ps(fiy0,ty);
 298             fiz0             = _mm_add_ps(fiz0,tz);
 299
 300             fjx0             = _mm_add_ps(fjx0,tx);
 301             fjy0             = _mm_add_ps(fjy0,ty);
 302             fjz0             = _mm_add_ps(fjz0,tz);
 303
 304             /**************************
 305              * CALCULATE INTERACTIONS *
 306              **************************/
 307
 308             r10              = _mm_mul_ps(rsq10,rinv10);
 309
 310             /* Compute parameters for interactions between i and j atoms */
 311             qq10             = _mm_mul_ps(iq1,jq0);
 312
 313             /* Calculate table index by multiplying r with table scale and truncate to integer */
 314             rt               = _mm_mul_ps(r10,vftabscale);
 315             vfitab           = _mm_cvttps_epi32(rt);
 316             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 317             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 318
 319             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 320             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 321             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 322             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 323             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 324             _MM_TRANSPOSE4_PS(Y,F,G,H);
 325             Heps             = _mm_mul_ps(vfeps,H);
 326             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 327             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 328             velec            = _mm_mul_ps(qq10,VV);
 329             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 330             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 331
 332             /* Update potential sum for this i atom from the interaction with this j atom. */
 333             velecsum         = _mm_add_ps(velecsum,velec);
 334
 335             fscal            = felec;
 336
 337             /* Calculate temporary vectorial force */
 338             tx               = _mm_mul_ps(fscal,dx10);
 339             ty               = _mm_mul_ps(fscal,dy10);
 340             tz               = _mm_mul_ps(fscal,dz10);
 341
 342             /* Update vectorial force */
 343             fix1             = _mm_add_ps(fix1,tx);
 344             fiy1             = _mm_add_ps(fiy1,ty);
 345             fiz1             = _mm_add_ps(fiz1,tz);
 346
 347             fjx0             = _mm_add_ps(fjx0,tx);
 348             fjy0             = _mm_add_ps(fjy0,ty);
 349             fjz0             = _mm_add_ps(fjz0,tz);
 350
 351             /**************************
 352              * CALCULATE INTERACTIONS *
 353              **************************/
 354
 355             r20              = _mm_mul_ps(rsq20,rinv20);
 356
 357             /* Compute parameters for interactions between i and j atoms */
 358             qq20             = _mm_mul_ps(iq2,jq0);
 359
 360             /* Calculate table index by multiplying r with table scale and truncate to integer */
 361             rt               = _mm_mul_ps(r20,vftabscale);
 362             vfitab           = _mm_cvttps_epi32(rt);
 363             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 364             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 365
 366             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 367             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 368             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 369             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 370             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 371             _MM_TRANSPOSE4_PS(Y,F,G,H);
 372             Heps             = _mm_mul_ps(vfeps,H);
 373             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 374             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 375             velec            = _mm_mul_ps(qq20,VV);
 376             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 377             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 378
 379             /* Update potential sum for this i atom from the interaction with this j atom. */
 380             velecsum         = _mm_add_ps(velecsum,velec);
 381
 382             fscal            = felec;
 383
 384             /* Calculate temporary vectorial force */
 385             tx               = _mm_mul_ps(fscal,dx20);
 386             ty               = _mm_mul_ps(fscal,dy20);
 387             tz               = _mm_mul_ps(fscal,dz20);
 388
 389             /* Update vectorial force */
 390             fix2             = _mm_add_ps(fix2,tx);
 391             fiy2             = _mm_add_ps(fiy2,ty);
 392             fiz2             = _mm_add_ps(fiz2,tz);
 393
 394             fjx0             = _mm_add_ps(fjx0,tx);
 395             fjy0             = _mm_add_ps(fjy0,ty);
 396             fjz0             = _mm_add_ps(fjz0,tz);
 397
 398             fjptrA             = f+j_coord_offsetA;
 399             fjptrB             = f+j_coord_offsetB;
 400             fjptrC             = f+j_coord_offsetC;
 401             fjptrD             = f+j_coord_offsetD;
 402
 403             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 404
 405             /* Inner loop uses 159 flops */
 406         }
 407
 408         if(jidx<j_index_end)
 409         {
 410
 411             /* Get j neighbor index, and coordinate index */
 412             jnrlistA         = jjnr[jidx];
 413             jnrlistB         = jjnr[jidx+1];
 414             jnrlistC         = jjnr[jidx+2];
 415             jnrlistD         = jjnr[jidx+3];
 416             /* Sign of each element will be negative for non-real atoms.
 417              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 418              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 419              */
 420             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 421             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 422             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 423             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 424             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 425             j_coord_offsetA  = DIM*jnrA;
 426             j_coord_offsetB  = DIM*jnrB;
 427             j_coord_offsetC  = DIM*jnrC;
 428             j_coord_offsetD  = DIM*jnrD;
 429
 430             /* load j atom coordinates */
 431             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 432                                               x+j_coord_offsetC,x+j_coord_offsetD,
 433                                               &jx0,&jy0,&jz0);
 434
 435             /* Calculate displacement vector */
 436             dx00             = _mm_sub_ps(ix0,jx0);
 437             dy00             = _mm_sub_ps(iy0,jy0);
 438             dz00             = _mm_sub_ps(iz0,jz0);
 439             dx10             = _mm_sub_ps(ix1,jx0);
 440             dy10             = _mm_sub_ps(iy1,jy0);
 441             dz10             = _mm_sub_ps(iz1,jz0);
 442             dx20             = _mm_sub_ps(ix2,jx0);
 443             dy20             = _mm_sub_ps(iy2,jy0);
 444             dz20             = _mm_sub_ps(iz2,jz0);
 445
 446             /* Calculate squared distance and things based on it */
 447             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 448             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 449             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 450
 451             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 452             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 453             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 454
 455             /* Load parameters for j particles */
 456             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 457                                                               charge+jnrC+0,charge+jnrD+0);
 458             vdwjidx0A        = 2*vdwtype[jnrA+0];
 459             vdwjidx0B        = 2*vdwtype[jnrB+0];
 460             vdwjidx0C        = 2*vdwtype[jnrC+0];
 461             vdwjidx0D        = 2*vdwtype[jnrD+0];
 462
 463             fjx0             = _mm_setzero_ps();
 464             fjy0             = _mm_setzero_ps();
 465             fjz0             = _mm_setzero_ps();
 466
 467             /**************************
 468              * CALCULATE INTERACTIONS *
 469              **************************/
 470
 471             r00              = _mm_mul_ps(rsq00,rinv00);
 472             r00              = _mm_andnot_ps(dummy_mask,r00);
 473
 474             /* Compute parameters for interactions between i and j atoms */
 475             qq00             = _mm_mul_ps(iq0,jq0);
 476             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 477                                          vdwparam+vdwioffset0+vdwjidx0B,
 478                                          vdwparam+vdwioffset0+vdwjidx0C,
 479                                          vdwparam+vdwioffset0+vdwjidx0D,
 480                                          &c6_00,&c12_00);
 481
 482             /* Calculate table index by multiplying r with table scale and truncate to integer */
 483             rt               = _mm_mul_ps(r00,vftabscale);
 484             vfitab           = _mm_cvttps_epi32(rt);
 485             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 486             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 487
 488             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 489             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 490             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 491             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 492             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 493             _MM_TRANSPOSE4_PS(Y,F,G,H);
 494             Heps             = _mm_mul_ps(vfeps,H);
 495             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 496             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 497             velec            = _mm_mul_ps(qq00,VV);
 498             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 499             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 500
 501             /* CUBIC SPLINE TABLE DISPERSION */
 502             vfitab           = _mm_add_epi32(vfitab,ifour);
 503             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 504             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 505             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 506             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 507             _MM_TRANSPOSE4_PS(Y,F,G,H);
 508             Heps             = _mm_mul_ps(vfeps,H);
 509             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 510             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 511             vvdw6            = _mm_mul_ps(c6_00,VV);
 512             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 513             fvdw6            = _mm_mul_ps(c6_00,FF);
 514
 515             /* CUBIC SPLINE TABLE REPULSION */
 516             vfitab           = _mm_add_epi32(vfitab,ifour);
 517             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 518             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 519             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 520             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 521             _MM_TRANSPOSE4_PS(Y,F,G,H);
 522             Heps             = _mm_mul_ps(vfeps,H);
 523             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 524             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 525             vvdw12           = _mm_mul_ps(c12_00,VV);
 526             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 527             fvdw12           = _mm_mul_ps(c12_00,FF);
 528             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 529             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 530
 531             /* Update potential sum for this i atom from the interaction with this j atom. */
 532             velec            = _mm_andnot_ps(dummy_mask,velec);
 533             velecsum         = _mm_add_ps(velecsum,velec);
 534             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 535             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 536
 537             fscal            = _mm_add_ps(felec,fvdw);
 538
 539             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 540
 541             /* Calculate temporary vectorial force */
 542             tx               = _mm_mul_ps(fscal,dx00);
 543             ty               = _mm_mul_ps(fscal,dy00);
 544             tz               = _mm_mul_ps(fscal,dz00);
 545
 546             /* Update vectorial force */
 547             fix0             = _mm_add_ps(fix0,tx);
 548             fiy0             = _mm_add_ps(fiy0,ty);
 549             fiz0             = _mm_add_ps(fiz0,tz);
 550
 551             fjx0             = _mm_add_ps(fjx0,tx);
 552             fjy0             = _mm_add_ps(fjy0,ty);
 553             fjz0             = _mm_add_ps(fjz0,tz);
 554
 555             /**************************
 556              * CALCULATE INTERACTIONS *
 557              **************************/
 558
 559             r10              = _mm_mul_ps(rsq10,rinv10);
 560             r10              = _mm_andnot_ps(dummy_mask,r10);
 561
 562             /* Compute parameters for interactions between i and j atoms */
 563             qq10             = _mm_mul_ps(iq1,jq0);
 564
 565             /* Calculate table index by multiplying r with table scale and truncate to integer */
 566             rt               = _mm_mul_ps(r10,vftabscale);
 567             vfitab           = _mm_cvttps_epi32(rt);
 568             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 569             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 570
 571             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 572             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 573             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 574             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 575             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 576             _MM_TRANSPOSE4_PS(Y,F,G,H);
 577             Heps             = _mm_mul_ps(vfeps,H);
 578             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 579             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 580             velec            = _mm_mul_ps(qq10,VV);
 581             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 582             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 583
 584             /* Update potential sum for this i atom from the interaction with this j atom. */
 585             velec            = _mm_andnot_ps(dummy_mask,velec);
 586             velecsum         = _mm_add_ps(velecsum,velec);
 587
 588             fscal            = felec;
 589
 590             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 591
 592             /* Calculate temporary vectorial force */
 593             tx               = _mm_mul_ps(fscal,dx10);
 594             ty               = _mm_mul_ps(fscal,dy10);
 595             tz               = _mm_mul_ps(fscal,dz10);
 596
 597             /* Update vectorial force */
 598             fix1             = _mm_add_ps(fix1,tx);
 599             fiy1             = _mm_add_ps(fiy1,ty);
 600             fiz1             = _mm_add_ps(fiz1,tz);
 601
 602             fjx0             = _mm_add_ps(fjx0,tx);
 603             fjy0             = _mm_add_ps(fjy0,ty);
 604             fjz0             = _mm_add_ps(fjz0,tz);
 605
 606             /**************************
 607              * CALCULATE INTERACTIONS *
 608              **************************/
 609
 610             r20              = _mm_mul_ps(rsq20,rinv20);
 611             r20              = _mm_andnot_ps(dummy_mask,r20);
 612
 613             /* Compute parameters for interactions between i and j atoms */
 614             qq20             = _mm_mul_ps(iq2,jq0);
 615
 616             /* Calculate table index by multiplying r with table scale and truncate to integer */
 617             rt               = _mm_mul_ps(r20,vftabscale);
 618             vfitab           = _mm_cvttps_epi32(rt);
 619             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 620             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 621
 622             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 623             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 624             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 625             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 626             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 627             _MM_TRANSPOSE4_PS(Y,F,G,H);
 628             Heps             = _mm_mul_ps(vfeps,H);
 629             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 630             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 631             velec            = _mm_mul_ps(qq20,VV);
 632             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 633             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 634
 635             /* Update potential sum for this i atom from the interaction with this j atom. */
 636             velec            = _mm_andnot_ps(dummy_mask,velec);
 637             velecsum         = _mm_add_ps(velecsum,velec);
 638
 639             fscal            = felec;
 640
 641             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 642
 643             /* Calculate temporary vectorial force */
 644             tx               = _mm_mul_ps(fscal,dx20);
 645             ty               = _mm_mul_ps(fscal,dy20);
 646             tz               = _mm_mul_ps(fscal,dz20);
 647
 648             /* Update vectorial force */
 649             fix2             = _mm_add_ps(fix2,tx);
 650             fiy2             = _mm_add_ps(fiy2,ty);
 651             fiz2             = _mm_add_ps(fiz2,tz);
 652
 653             fjx0             = _mm_add_ps(fjx0,tx);
 654             fjy0             = _mm_add_ps(fjy0,ty);
 655             fjz0             = _mm_add_ps(fjz0,tz);
 656
 657             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 658             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 659             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 660             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 661
 662             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 663
 664             /* Inner loop uses 162 flops */
 665         }
 666
 667         /* End of innermost loop */
 668
 669         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 670                                               f+i_coord_offset,fshift+i_shift_offset);
 671
 672         ggid                        = gid[iidx];
 673         /* Update potential energies */
 674         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 675         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 676
 677         /* Increment number of inner iterations */
 678         inneriter                  += j_index_end - j_index_start;
 679
 680         /* Outer loop uses 20 flops */
 681     }
 682
 683     /* Increment number of outer iterations */
 684     outeriter        += nri;
 685
 686     /* Update outer/inner flops */
 687
 688     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*162);
 689 }
 690 /*
 691  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_sse4_1_single
 692  * Electrostatics interaction: CubicSplineTable
 693  * VdW interaction:            CubicSplineTable
 694  * Geometry:                   Water3-Particle
 695  * Calculate force/pot:        Force
 696  */
 697 void
 698 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_sse4_1_single
 699                     (t_nblist * gmx_restrict                nlist,
 700                      rvec * gmx_restrict                    xx,
 701                      rvec * gmx_restrict                    ff,
 702                      t_forcerec * gmx_restrict              fr,
 703                      t_mdatoms * gmx_restrict               mdatoms,
 704                      nb_kernel_data_t * gmx_restrict        kernel_data,
 705                      t_nrnb * gmx_restrict                  nrnb)
 706 {
 707     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 708      * just 0 for non-waters.
 709      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 710      * jnr indices corresponding to data put in the four positions in the SIMD register.
 711      */
 712     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 713     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 714     int              jnrA,jnrB,jnrC,jnrD;
 715     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 716     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 717     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 718     real             rcutoff_scalar;
 719     real             *shiftvec,*fshift,*x,*f;
 720     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 721     real             scratch[4*DIM];
 722     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 723     int              vdwioffset0;
 724     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 725     int              vdwioffset1;
 726     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 727     int              vdwioffset2;
 728     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 729     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 730     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 731     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 732     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 733     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 734     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 735     real             *charge;
 736     int              nvdwtype;
 737     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 738     int              *vdwtype;
 739     real             *vdwparam;
 740     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 741     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 742     __m128i          vfitab;
 743     __m128i          ifour       = _mm_set1_epi32(4);
 744     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 745     real             *vftab;
 746     __m128           dummy_mask,cutoff_mask;
 747     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 748     __m128           one     = _mm_set1_ps(1.0);
 749     __m128           two     = _mm_set1_ps(2.0);
 750     x                = xx[0];
 751     f                = ff[0];
 752
 753     nri              = nlist->nri;
 754     iinr             = nlist->iinr;
 755     jindex           = nlist->jindex;
 756     jjnr             = nlist->jjnr;
 757     shiftidx         = nlist->shift;
 758     gid              = nlist->gid;
 759     shiftvec         = fr->shift_vec[0];
 760     fshift           = fr->fshift[0];
 761     facel            = _mm_set1_ps(fr->epsfac);
 762     charge           = mdatoms->chargeA;
 763     nvdwtype         = fr->ntype;
 764     vdwparam         = fr->nbfp;
 765     vdwtype          = mdatoms->typeA;
 766
 767     vftab            = kernel_data->table_elec_vdw->data;
 768     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 769
 770     /* Setup water-specific parameters */
 771     inr              = nlist->iinr[0];
 772     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 773     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 774     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 775     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 776
 777     /* Avoid stupid compiler warnings */
 778     jnrA = jnrB = jnrC = jnrD = 0;
 779     j_coord_offsetA = 0;
 780     j_coord_offsetB = 0;
 781     j_coord_offsetC = 0;
 782     j_coord_offsetD = 0;
 783
 784     outeriter        = 0;
 785     inneriter        = 0;
 786
 787     for(iidx=0;iidx<4*DIM;iidx++)
 788     {
 789         scratch[iidx] = 0.0;
 790     }
 791
 792     /* Start outer loop over neighborlists */
 793     for(iidx=0; iidx<nri; iidx++)
 794     {
 795         /* Load shift vector for this list */
 796         i_shift_offset   = DIM*shiftidx[iidx];
 797
 798         /* Load limits for loop over neighbors */
 799         j_index_start    = jindex[iidx];
 800         j_index_end      = jindex[iidx+1];
 801
 802         /* Get outer coordinate index */
 803         inr              = iinr[iidx];
 804         i_coord_offset   = DIM*inr;
 805
 806         /* Load i particle coords and add shift vector */
 807         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 808                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 809
 810         fix0             = _mm_setzero_ps();
 811         fiy0             = _mm_setzero_ps();
 812         fiz0             = _mm_setzero_ps();
 813         fix1             = _mm_setzero_ps();
 814         fiy1             = _mm_setzero_ps();
 815         fiz1             = _mm_setzero_ps();
 816         fix2             = _mm_setzero_ps();
 817         fiy2             = _mm_setzero_ps();
 818         fiz2             = _mm_setzero_ps();
 819
 820         /* Start inner kernel loop */
 821         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 822         {
 823
 824             /* Get j neighbor index, and coordinate index */
 825             jnrA             = jjnr[jidx];
 826             jnrB             = jjnr[jidx+1];
 827             jnrC             = jjnr[jidx+2];
 828             jnrD             = jjnr[jidx+3];
 829             j_coord_offsetA  = DIM*jnrA;
 830             j_coord_offsetB  = DIM*jnrB;
 831             j_coord_offsetC  = DIM*jnrC;
 832             j_coord_offsetD  = DIM*jnrD;
 833
 834             /* load j atom coordinates */
 835             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 836                                               x+j_coord_offsetC,x+j_coord_offsetD,
 837                                               &jx0,&jy0,&jz0);
 838
 839             /* Calculate displacement vector */
 840             dx00             = _mm_sub_ps(ix0,jx0);
 841             dy00             = _mm_sub_ps(iy0,jy0);
 842             dz00             = _mm_sub_ps(iz0,jz0);
 843             dx10             = _mm_sub_ps(ix1,jx0);
 844             dy10             = _mm_sub_ps(iy1,jy0);
 845             dz10             = _mm_sub_ps(iz1,jz0);
 846             dx20             = _mm_sub_ps(ix2,jx0);
 847             dy20             = _mm_sub_ps(iy2,jy0);
 848             dz20             = _mm_sub_ps(iz2,jz0);
 849
 850             /* Calculate squared distance and things based on it */
 851             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 852             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 853             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 854
 855             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 856             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 857             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 858
 859             /* Load parameters for j particles */
 860             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 861                                                               charge+jnrC+0,charge+jnrD+0);
 862             vdwjidx0A        = 2*vdwtype[jnrA+0];
 863             vdwjidx0B        = 2*vdwtype[jnrB+0];
 864             vdwjidx0C        = 2*vdwtype[jnrC+0];
 865             vdwjidx0D        = 2*vdwtype[jnrD+0];
 866
 867             fjx0             = _mm_setzero_ps();
 868             fjy0             = _mm_setzero_ps();
 869             fjz0             = _mm_setzero_ps();
 870
 871             /**************************
 872              * CALCULATE INTERACTIONS *
 873              **************************/
 874
 875             r00              = _mm_mul_ps(rsq00,rinv00);
 876
 877             /* Compute parameters for interactions between i and j atoms */
 878             qq00             = _mm_mul_ps(iq0,jq0);
 879             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 880                                          vdwparam+vdwioffset0+vdwjidx0B,
 881                                          vdwparam+vdwioffset0+vdwjidx0C,
 882                                          vdwparam+vdwioffset0+vdwjidx0D,
 883                                          &c6_00,&c12_00);
 884
 885             /* Calculate table index by multiplying r with table scale and truncate to integer */
 886             rt               = _mm_mul_ps(r00,vftabscale);
 887             vfitab           = _mm_cvttps_epi32(rt);
 888             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 889             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 890
 891             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 892             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 893             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 894             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 895             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 896             _MM_TRANSPOSE4_PS(Y,F,G,H);
 897             Heps             = _mm_mul_ps(vfeps,H);
 898             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 899             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 900             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 901
 902             /* CUBIC SPLINE TABLE DISPERSION */
 903             vfitab           = _mm_add_epi32(vfitab,ifour);
 904             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 905             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 906             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 907             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 908             _MM_TRANSPOSE4_PS(Y,F,G,H);
 909             Heps             = _mm_mul_ps(vfeps,H);
 910             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 911             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 912             fvdw6            = _mm_mul_ps(c6_00,FF);
 913
 914             /* CUBIC SPLINE TABLE REPULSION */
 915             vfitab           = _mm_add_epi32(vfitab,ifour);
 916             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 917             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 918             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 919             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 920             _MM_TRANSPOSE4_PS(Y,F,G,H);
 921             Heps             = _mm_mul_ps(vfeps,H);
 922             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 923             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 924             fvdw12           = _mm_mul_ps(c12_00,FF);
 925             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 926
 927             fscal            = _mm_add_ps(felec,fvdw);
 928
 929             /* Calculate temporary vectorial force */
 930             tx               = _mm_mul_ps(fscal,dx00);
 931             ty               = _mm_mul_ps(fscal,dy00);
 932             tz               = _mm_mul_ps(fscal,dz00);
 933
 934             /* Update vectorial force */
 935             fix0             = _mm_add_ps(fix0,tx);
 936             fiy0             = _mm_add_ps(fiy0,ty);
 937             fiz0             = _mm_add_ps(fiz0,tz);
 938
 939             fjx0             = _mm_add_ps(fjx0,tx);
 940             fjy0             = _mm_add_ps(fjy0,ty);
 941             fjz0             = _mm_add_ps(fjz0,tz);
 942
 943             /**************************
 944              * CALCULATE INTERACTIONS *
 945              **************************/
 946
 947             r10              = _mm_mul_ps(rsq10,rinv10);
 948
 949             /* Compute parameters for interactions between i and j atoms */
 950             qq10             = _mm_mul_ps(iq1,jq0);
 951
 952             /* Calculate table index by multiplying r with table scale and truncate to integer */
 953             rt               = _mm_mul_ps(r10,vftabscale);
 954             vfitab           = _mm_cvttps_epi32(rt);
 955             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 956             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 957
 958             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 959             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 960             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 961             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 962             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 963             _MM_TRANSPOSE4_PS(Y,F,G,H);
 964             Heps             = _mm_mul_ps(vfeps,H);
 965             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 966             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 967             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 968
 969             fscal            = felec;
 970
 971             /* Calculate temporary vectorial force */
 972             tx               = _mm_mul_ps(fscal,dx10);
 973             ty               = _mm_mul_ps(fscal,dy10);
 974             tz               = _mm_mul_ps(fscal,dz10);
 975
 976             /* Update vectorial force */
 977             fix1             = _mm_add_ps(fix1,tx);
 978             fiy1             = _mm_add_ps(fiy1,ty);
 979             fiz1             = _mm_add_ps(fiz1,tz);
 980
 981             fjx0             = _mm_add_ps(fjx0,tx);
 982             fjy0             = _mm_add_ps(fjy0,ty);
 983             fjz0             = _mm_add_ps(fjz0,tz);
 984
 985             /**************************
 986              * CALCULATE INTERACTIONS *
 987              **************************/
 988
 989             r20              = _mm_mul_ps(rsq20,rinv20);
 990
 991             /* Compute parameters for interactions between i and j atoms */
 992             qq20             = _mm_mul_ps(iq2,jq0);
 993
 994             /* Calculate table index by multiplying r with table scale and truncate to integer */
 995             rt               = _mm_mul_ps(r20,vftabscale);
 996             vfitab           = _mm_cvttps_epi32(rt);
 997             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 998             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 999
1000             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1001             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1002             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1003             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1004             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1005             _MM_TRANSPOSE4_PS(Y,F,G,H);
1006             Heps             = _mm_mul_ps(vfeps,H);
1007             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1008             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1009             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1010
1011             fscal            = felec;
1012
1013             /* Calculate temporary vectorial force */
1014             tx               = _mm_mul_ps(fscal,dx20);
1015             ty               = _mm_mul_ps(fscal,dy20);
1016             tz               = _mm_mul_ps(fscal,dz20);
1017
1018             /* Update vectorial force */
1019             fix2             = _mm_add_ps(fix2,tx);
1020             fiy2             = _mm_add_ps(fiy2,ty);
1021             fiz2             = _mm_add_ps(fiz2,tz);
1022
1023             fjx0             = _mm_add_ps(fjx0,tx);
1024             fjy0             = _mm_add_ps(fjy0,ty);
1025             fjz0             = _mm_add_ps(fjz0,tz);
1026
1027             fjptrA             = f+j_coord_offsetA;
1028             fjptrB             = f+j_coord_offsetB;
1029             fjptrC             = f+j_coord_offsetC;
1030             fjptrD             = f+j_coord_offsetD;
1031
1032             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1033
1034             /* Inner loop uses 139 flops */
1035         }
1036
1037         if(jidx<j_index_end)
1038         {
1039
1040             /* Get j neighbor index, and coordinate index */
1041             jnrlistA         = jjnr[jidx];
1042             jnrlistB         = jjnr[jidx+1];
1043             jnrlistC         = jjnr[jidx+2];
1044             jnrlistD         = jjnr[jidx+3];
1045             /* Sign of each element will be negative for non-real atoms.
1046              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1047              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1048              */
1049             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1050             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1051             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1052             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1053             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1054             j_coord_offsetA  = DIM*jnrA;
1055             j_coord_offsetB  = DIM*jnrB;
1056             j_coord_offsetC  = DIM*jnrC;
1057             j_coord_offsetD  = DIM*jnrD;
1058
1059             /* load j atom coordinates */
1060             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1061                                               x+j_coord_offsetC,x+j_coord_offsetD,
1062                                               &jx0,&jy0,&jz0);
1063
1064             /* Calculate displacement vector */
1065             dx00             = _mm_sub_ps(ix0,jx0);
1066             dy00             = _mm_sub_ps(iy0,jy0);
1067             dz00             = _mm_sub_ps(iz0,jz0);
1068             dx10             = _mm_sub_ps(ix1,jx0);
1069             dy10             = _mm_sub_ps(iy1,jy0);
1070             dz10             = _mm_sub_ps(iz1,jz0);
1071             dx20             = _mm_sub_ps(ix2,jx0);
1072             dy20             = _mm_sub_ps(iy2,jy0);
1073             dz20             = _mm_sub_ps(iz2,jz0);
1074
1075             /* Calculate squared distance and things based on it */
1076             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1077             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1078             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1079
1080             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1081             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1082             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1083
1084             /* Load parameters for j particles */
1085             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1086                                                               charge+jnrC+0,charge+jnrD+0);
1087             vdwjidx0A        = 2*vdwtype[jnrA+0];
1088             vdwjidx0B        = 2*vdwtype[jnrB+0];
1089             vdwjidx0C        = 2*vdwtype[jnrC+0];
1090             vdwjidx0D        = 2*vdwtype[jnrD+0];
1091
1092             fjx0             = _mm_setzero_ps();
1093             fjy0             = _mm_setzero_ps();
1094             fjz0             = _mm_setzero_ps();
1095
1096             /**************************
1097              * CALCULATE INTERACTIONS *
1098              **************************/
1099
1100             r00              = _mm_mul_ps(rsq00,rinv00);
1101             r00              = _mm_andnot_ps(dummy_mask,r00);
1102
1103             /* Compute parameters for interactions between i and j atoms */
1104             qq00             = _mm_mul_ps(iq0,jq0);
1105             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1106                                          vdwparam+vdwioffset0+vdwjidx0B,
1107                                          vdwparam+vdwioffset0+vdwjidx0C,
1108                                          vdwparam+vdwioffset0+vdwjidx0D,
1109                                          &c6_00,&c12_00);
1110
1111             /* Calculate table index by multiplying r with table scale and truncate to integer */
1112             rt               = _mm_mul_ps(r00,vftabscale);
1113             vfitab           = _mm_cvttps_epi32(rt);
1114             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1115             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1116
1117             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1118             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1119             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1120             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1121             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1122             _MM_TRANSPOSE4_PS(Y,F,G,H);
1123             Heps             = _mm_mul_ps(vfeps,H);
1124             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1125             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1126             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
1127
1128             /* CUBIC SPLINE TABLE DISPERSION */
1129             vfitab           = _mm_add_epi32(vfitab,ifour);
1130             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1131             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1132             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1133             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1134             _MM_TRANSPOSE4_PS(Y,F,G,H);
1135             Heps             = _mm_mul_ps(vfeps,H);
1136             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1137             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1138             fvdw6            = _mm_mul_ps(c6_00,FF);
1139
1140             /* CUBIC SPLINE TABLE REPULSION */
1141             vfitab           = _mm_add_epi32(vfitab,ifour);
1142             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1143             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1144             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1145             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1146             _MM_TRANSPOSE4_PS(Y,F,G,H);
1147             Heps             = _mm_mul_ps(vfeps,H);
1148             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1149             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1150             fvdw12           = _mm_mul_ps(c12_00,FF);
1151             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1152
1153             fscal            = _mm_add_ps(felec,fvdw);
1154
1155             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1156
1157             /* Calculate temporary vectorial force */
1158             tx               = _mm_mul_ps(fscal,dx00);
1159             ty               = _mm_mul_ps(fscal,dy00);
1160             tz               = _mm_mul_ps(fscal,dz00);
1161
1162             /* Update vectorial force */
1163             fix0             = _mm_add_ps(fix0,tx);
1164             fiy0             = _mm_add_ps(fiy0,ty);
1165             fiz0             = _mm_add_ps(fiz0,tz);
1166
1167             fjx0             = _mm_add_ps(fjx0,tx);
1168             fjy0             = _mm_add_ps(fjy0,ty);
1169             fjz0             = _mm_add_ps(fjz0,tz);
1170
1171             /**************************
1172              * CALCULATE INTERACTIONS *
1173              **************************/
1174
1175             r10              = _mm_mul_ps(rsq10,rinv10);
1176             r10              = _mm_andnot_ps(dummy_mask,r10);
1177
1178             /* Compute parameters for interactions between i and j atoms */
1179             qq10             = _mm_mul_ps(iq1,jq0);
1180
1181             /* Calculate table index by multiplying r with table scale and truncate to integer */
1182             rt               = _mm_mul_ps(r10,vftabscale);
1183             vfitab           = _mm_cvttps_epi32(rt);
1184             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1185             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1186
1187             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1188             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1189             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1190             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1191             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1192             _MM_TRANSPOSE4_PS(Y,F,G,H);
1193             Heps             = _mm_mul_ps(vfeps,H);
1194             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1195             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1196             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1197
1198             fscal            = felec;
1199
1200             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1201
1202             /* Calculate temporary vectorial force */
1203             tx               = _mm_mul_ps(fscal,dx10);
1204             ty               = _mm_mul_ps(fscal,dy10);
1205             tz               = _mm_mul_ps(fscal,dz10);
1206
1207             /* Update vectorial force */
1208             fix1             = _mm_add_ps(fix1,tx);
1209             fiy1             = _mm_add_ps(fiy1,ty);
1210             fiz1             = _mm_add_ps(fiz1,tz);
1211
1212             fjx0             = _mm_add_ps(fjx0,tx);
1213             fjy0             = _mm_add_ps(fjy0,ty);
1214             fjz0             = _mm_add_ps(fjz0,tz);
1215
1216             /**************************
1217              * CALCULATE INTERACTIONS *
1218              **************************/
1219
1220             r20              = _mm_mul_ps(rsq20,rinv20);
1221             r20              = _mm_andnot_ps(dummy_mask,r20);
1222
1223             /* Compute parameters for interactions between i and j atoms */
1224             qq20             = _mm_mul_ps(iq2,jq0);
1225
1226             /* Calculate table index by multiplying r with table scale and truncate to integer */
1227             rt               = _mm_mul_ps(r20,vftabscale);
1228             vfitab           = _mm_cvttps_epi32(rt);
1229             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1230             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1231
1232             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1233             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1234             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1235             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1236             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1237             _MM_TRANSPOSE4_PS(Y,F,G,H);
1238             Heps             = _mm_mul_ps(vfeps,H);
1239             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1240             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1241             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1242
1243             fscal            = felec;
1244
1245             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1246
1247             /* Calculate temporary vectorial force */
1248             tx               = _mm_mul_ps(fscal,dx20);
1249             ty               = _mm_mul_ps(fscal,dy20);
1250             tz               = _mm_mul_ps(fscal,dz20);
1251
1252             /* Update vectorial force */
1253             fix2             = _mm_add_ps(fix2,tx);
1254             fiy2             = _mm_add_ps(fiy2,ty);
1255             fiz2             = _mm_add_ps(fiz2,tz);
1256
1257             fjx0             = _mm_add_ps(fjx0,tx);
1258             fjy0             = _mm_add_ps(fjy0,ty);
1259             fjz0             = _mm_add_ps(fjz0,tz);
1260
1261             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1262             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1263             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1264             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1265
1266             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1267
1268             /* Inner loop uses 142 flops */
1269         }
1270
1271         /* End of innermost loop */
1272
1273         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1274                                               f+i_coord_offset,fshift+i_shift_offset);
1275
1276         /* Increment number of inner iterations */
1277         inneriter                  += j_index_end - j_index_start;
1278
1279         /* Outer loop uses 18 flops */
1280     }
1281
1282     /* Increment number of outer iterations */
1283     outeriter        += nri;
1284
1285     /* Update outer/inner flops */
1286
1287     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*142);
1288 }