src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEwSw_VdwLJSw_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_ElecEwSw_VdwLJSw_GeomW3P1_VF_sse4_1_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEwSw_VdwLJSw_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          ewitab;
  89     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  90     real             *ewtab;
  91     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  92     real             rswitch_scalar,d_scalar;
  93     __m128           dummy_mask,cutoff_mask;
  94     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  95     __m128           one     = _mm_set1_ps(1.0);
  96     __m128           two     = _mm_set1_ps(2.0);
  97     x                = xx[0];
  98     f                = ff[0];
  99
 100     nri              = nlist->nri;
 101     iinr             = nlist->iinr;
 102     jindex           = nlist->jindex;
 103     jjnr             = nlist->jjnr;
 104     shiftidx         = nlist->shift;
 105     gid              = nlist->gid;
 106     shiftvec         = fr->shift_vec[0];
 107     fshift           = fr->fshift[0];
 108     facel            = _mm_set1_ps(fr->epsfac);
 109     charge           = mdatoms->chargeA;
 110     nvdwtype         = fr->ntype;
 111     vdwparam         = fr->nbfp;
 112     vdwtype          = mdatoms->typeA;
 113
 114     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 115     ewtab            = fr->ic->tabq_coul_FDV0;
 116     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 117     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 118
 119     /* Setup water-specific parameters */
 120     inr              = nlist->iinr[0];
 121     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 122     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 123     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 124     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 125
 126     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 127     rcutoff_scalar   = fr->rcoulomb;
 128     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 129     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 130
 131     rswitch_scalar   = fr->rcoulomb_switch;
 132     rswitch          = _mm_set1_ps(rswitch_scalar);
 133     /* Setup switch parameters */
 134     d_scalar         = rcutoff_scalar-rswitch_scalar;
 135     d                = _mm_set1_ps(d_scalar);
 136     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 137     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 138     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 139     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 140     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 141     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 142
 143     /* Avoid stupid compiler warnings */
 144     jnrA = jnrB = jnrC = jnrD = 0;
 145     j_coord_offsetA = 0;
 146     j_coord_offsetB = 0;
 147     j_coord_offsetC = 0;
 148     j_coord_offsetD = 0;
 149
 150     outeriter        = 0;
 151     inneriter        = 0;
 152
 153     for(iidx=0;iidx<4*DIM;iidx++)
 154     {
 155         scratch[iidx] = 0.0;
 156     }
 157
 158     /* Start outer loop over neighborlists */
 159     for(iidx=0; iidx<nri; iidx++)
 160     {
 161         /* Load shift vector for this list */
 162         i_shift_offset   = DIM*shiftidx[iidx];
 163
 164         /* Load limits for loop over neighbors */
 165         j_index_start    = jindex[iidx];
 166         j_index_end      = jindex[iidx+1];
 167
 168         /* Get outer coordinate index */
 169         inr              = iinr[iidx];
 170         i_coord_offset   = DIM*inr;
 171
 172         /* Load i particle coords and add shift vector */
 173         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 174                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 175
 176         fix0             = _mm_setzero_ps();
 177         fiy0             = _mm_setzero_ps();
 178         fiz0             = _mm_setzero_ps();
 179         fix1             = _mm_setzero_ps();
 180         fiy1             = _mm_setzero_ps();
 181         fiz1             = _mm_setzero_ps();
 182         fix2             = _mm_setzero_ps();
 183         fiy2             = _mm_setzero_ps();
 184         fiz2             = _mm_setzero_ps();
 185
 186         /* Reset potential sums */
 187         velecsum         = _mm_setzero_ps();
 188         vvdwsum          = _mm_setzero_ps();
 189
 190         /* Start inner kernel loop */
 191         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 192         {
 193
 194             /* Get j neighbor index, and coordinate index */
 195             jnrA             = jjnr[jidx];
 196             jnrB             = jjnr[jidx+1];
 197             jnrC             = jjnr[jidx+2];
 198             jnrD             = jjnr[jidx+3];
 199             j_coord_offsetA  = DIM*jnrA;
 200             j_coord_offsetB  = DIM*jnrB;
 201             j_coord_offsetC  = DIM*jnrC;
 202             j_coord_offsetD  = DIM*jnrD;
 203
 204             /* load j atom coordinates */
 205             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 206                                               x+j_coord_offsetC,x+j_coord_offsetD,
 207                                               &jx0,&jy0,&jz0);
 208
 209             /* Calculate displacement vector */
 210             dx00             = _mm_sub_ps(ix0,jx0);
 211             dy00             = _mm_sub_ps(iy0,jy0);
 212             dz00             = _mm_sub_ps(iz0,jz0);
 213             dx10             = _mm_sub_ps(ix1,jx0);
 214             dy10             = _mm_sub_ps(iy1,jy0);
 215             dz10             = _mm_sub_ps(iz1,jz0);
 216             dx20             = _mm_sub_ps(ix2,jx0);
 217             dy20             = _mm_sub_ps(iy2,jy0);
 218             dz20             = _mm_sub_ps(iz2,jz0);
 219
 220             /* Calculate squared distance and things based on it */
 221             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 222             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 223             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 224
 225             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 226             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 227             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 228
 229             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 230             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 231             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 232
 233             /* Load parameters for j particles */
 234             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 235                                                               charge+jnrC+0,charge+jnrD+0);
 236             vdwjidx0A        = 2*vdwtype[jnrA+0];
 237             vdwjidx0B        = 2*vdwtype[jnrB+0];
 238             vdwjidx0C        = 2*vdwtype[jnrC+0];
 239             vdwjidx0D        = 2*vdwtype[jnrD+0];
 240
 241             /**************************
 242              * CALCULATE INTERACTIONS *
 243              **************************/
 244
 245             if (gmx_mm_any_lt(rsq00,rcutoff2))
 246             {
 247
 248             r00              = _mm_mul_ps(rsq00,rinv00);
 249
 250             /* Compute parameters for interactions between i and j atoms */
 251             qq00             = _mm_mul_ps(iq0,jq0);
 252             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 253                                          vdwparam+vdwioffset0+vdwjidx0B,
 254                                          vdwparam+vdwioffset0+vdwjidx0C,
 255                                          vdwparam+vdwioffset0+vdwjidx0D,
 256                                          &c6_00,&c12_00);
 257
 258             /* EWALD ELECTROSTATICS */
 259
 260             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 261             ewrt             = _mm_mul_ps(r00,ewtabscale);
 262             ewitab           = _mm_cvttps_epi32(ewrt);
 263             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 264             ewitab           = _mm_slli_epi32(ewitab,2);
 265             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 266             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 267             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 268             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 269             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 270             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 271             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 272             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 273             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 274
 275             /* LENNARD-JONES DISPERSION/REPULSION */
 276
 277             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 278             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 279             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 280             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 281             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 282
 283             d                = _mm_sub_ps(r00,rswitch);
 284             d                = _mm_max_ps(d,_mm_setzero_ps());
 285             d2               = _mm_mul_ps(d,d);
 286             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 287
 288             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 289
 290             /* Evaluate switch function */
 291             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 292             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
 293             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 294             velec            = _mm_mul_ps(velec,sw);
 295             vvdw             = _mm_mul_ps(vvdw,sw);
 296             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 297
 298             /* Update potential sum for this i atom from the interaction with this j atom. */
 299             velec            = _mm_and_ps(velec,cutoff_mask);
 300             velecsum         = _mm_add_ps(velecsum,velec);
 301             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 302             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 303
 304             fscal            = _mm_add_ps(felec,fvdw);
 305
 306             fscal            = _mm_and_ps(fscal,cutoff_mask);
 307
 308             /* Calculate temporary vectorial force */
 309             tx               = _mm_mul_ps(fscal,dx00);
 310             ty               = _mm_mul_ps(fscal,dy00);
 311             tz               = _mm_mul_ps(fscal,dz00);
 312
 313             /* Update vectorial force */
 314             fix0             = _mm_add_ps(fix0,tx);
 315             fiy0             = _mm_add_ps(fiy0,ty);
 316             fiz0             = _mm_add_ps(fiz0,tz);
 317
 318             fjptrA             = f+j_coord_offsetA;
 319             fjptrB             = f+j_coord_offsetB;
 320             fjptrC             = f+j_coord_offsetC;
 321             fjptrD             = f+j_coord_offsetD;
 322             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 323
 324             }
 325
 326             /**************************
 327              * CALCULATE INTERACTIONS *
 328              **************************/
 329
 330             if (gmx_mm_any_lt(rsq10,rcutoff2))
 331             {
 332
 333             r10              = _mm_mul_ps(rsq10,rinv10);
 334
 335             /* Compute parameters for interactions between i and j atoms */
 336             qq10             = _mm_mul_ps(iq1,jq0);
 337
 338             /* EWALD ELECTROSTATICS */
 339
 340             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 341             ewrt             = _mm_mul_ps(r10,ewtabscale);
 342             ewitab           = _mm_cvttps_epi32(ewrt);
 343             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 344             ewitab           = _mm_slli_epi32(ewitab,2);
 345             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 346             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 347             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 348             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 349             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 350             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 351             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 352             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 353             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 354
 355             d                = _mm_sub_ps(r10,rswitch);
 356             d                = _mm_max_ps(d,_mm_setzero_ps());
 357             d2               = _mm_mul_ps(d,d);
 358             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 359
 360             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 361
 362             /* Evaluate switch function */
 363             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 364             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
 365             velec            = _mm_mul_ps(velec,sw);
 366             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 367
 368             /* Update potential sum for this i atom from the interaction with this j atom. */
 369             velec            = _mm_and_ps(velec,cutoff_mask);
 370             velecsum         = _mm_add_ps(velecsum,velec);
 371
 372             fscal            = felec;
 373
 374             fscal            = _mm_and_ps(fscal,cutoff_mask);
 375
 376             /* Calculate temporary vectorial force */
 377             tx               = _mm_mul_ps(fscal,dx10);
 378             ty               = _mm_mul_ps(fscal,dy10);
 379             tz               = _mm_mul_ps(fscal,dz10);
 380
 381             /* Update vectorial force */
 382             fix1             = _mm_add_ps(fix1,tx);
 383             fiy1             = _mm_add_ps(fiy1,ty);
 384             fiz1             = _mm_add_ps(fiz1,tz);
 385
 386             fjptrA             = f+j_coord_offsetA;
 387             fjptrB             = f+j_coord_offsetB;
 388             fjptrC             = f+j_coord_offsetC;
 389             fjptrD             = f+j_coord_offsetD;
 390             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 391
 392             }
 393
 394             /**************************
 395              * CALCULATE INTERACTIONS *
 396              **************************/
 397
 398             if (gmx_mm_any_lt(rsq20,rcutoff2))
 399             {
 400
 401             r20              = _mm_mul_ps(rsq20,rinv20);
 402
 403             /* Compute parameters for interactions between i and j atoms */
 404             qq20             = _mm_mul_ps(iq2,jq0);
 405
 406             /* EWALD ELECTROSTATICS */
 407
 408             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 409             ewrt             = _mm_mul_ps(r20,ewtabscale);
 410             ewitab           = _mm_cvttps_epi32(ewrt);
 411             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 412             ewitab           = _mm_slli_epi32(ewitab,2);
 413             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 414             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 415             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 416             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 417             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 418             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 419             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 420             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 421             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 422
 423             d                = _mm_sub_ps(r20,rswitch);
 424             d                = _mm_max_ps(d,_mm_setzero_ps());
 425             d2               = _mm_mul_ps(d,d);
 426             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 427
 428             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 429
 430             /* Evaluate switch function */
 431             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 432             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
 433             velec            = _mm_mul_ps(velec,sw);
 434             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 435
 436             /* Update potential sum for this i atom from the interaction with this j atom. */
 437             velec            = _mm_and_ps(velec,cutoff_mask);
 438             velecsum         = _mm_add_ps(velecsum,velec);
 439
 440             fscal            = felec;
 441
 442             fscal            = _mm_and_ps(fscal,cutoff_mask);
 443
 444             /* Calculate temporary vectorial force */
 445             tx               = _mm_mul_ps(fscal,dx20);
 446             ty               = _mm_mul_ps(fscal,dy20);
 447             tz               = _mm_mul_ps(fscal,dz20);
 448
 449             /* Update vectorial force */
 450             fix2             = _mm_add_ps(fix2,tx);
 451             fiy2             = _mm_add_ps(fiy2,ty);
 452             fiz2             = _mm_add_ps(fiz2,tz);
 453
 454             fjptrA             = f+j_coord_offsetA;
 455             fjptrB             = f+j_coord_offsetB;
 456             fjptrC             = f+j_coord_offsetC;
 457             fjptrD             = f+j_coord_offsetD;
 458             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 459
 460             }
 461
 462             /* Inner loop uses 213 flops */
 463         }
 464
 465         if(jidx<j_index_end)
 466         {
 467
 468             /* Get j neighbor index, and coordinate index */
 469             jnrlistA         = jjnr[jidx];
 470             jnrlistB         = jjnr[jidx+1];
 471             jnrlistC         = jjnr[jidx+2];
 472             jnrlistD         = jjnr[jidx+3];
 473             /* Sign of each element will be negative for non-real atoms.
 474              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 475              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 476              */
 477             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 478             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 479             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 480             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 481             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 482             j_coord_offsetA  = DIM*jnrA;
 483             j_coord_offsetB  = DIM*jnrB;
 484             j_coord_offsetC  = DIM*jnrC;
 485             j_coord_offsetD  = DIM*jnrD;
 486
 487             /* load j atom coordinates */
 488             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 489                                               x+j_coord_offsetC,x+j_coord_offsetD,
 490                                               &jx0,&jy0,&jz0);
 491
 492             /* Calculate displacement vector */
 493             dx00             = _mm_sub_ps(ix0,jx0);
 494             dy00             = _mm_sub_ps(iy0,jy0);
 495             dz00             = _mm_sub_ps(iz0,jz0);
 496             dx10             = _mm_sub_ps(ix1,jx0);
 497             dy10             = _mm_sub_ps(iy1,jy0);
 498             dz10             = _mm_sub_ps(iz1,jz0);
 499             dx20             = _mm_sub_ps(ix2,jx0);
 500             dy20             = _mm_sub_ps(iy2,jy0);
 501             dz20             = _mm_sub_ps(iz2,jz0);
 502
 503             /* Calculate squared distance and things based on it */
 504             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 505             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 506             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 507
 508             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 509             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 510             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 511
 512             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 513             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 514             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 515
 516             /* Load parameters for j particles */
 517             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 518                                                               charge+jnrC+0,charge+jnrD+0);
 519             vdwjidx0A        = 2*vdwtype[jnrA+0];
 520             vdwjidx0B        = 2*vdwtype[jnrB+0];
 521             vdwjidx0C        = 2*vdwtype[jnrC+0];
 522             vdwjidx0D        = 2*vdwtype[jnrD+0];
 523
 524             /**************************
 525              * CALCULATE INTERACTIONS *
 526              **************************/
 527
 528             if (gmx_mm_any_lt(rsq00,rcutoff2))
 529             {
 530
 531             r00              = _mm_mul_ps(rsq00,rinv00);
 532             r00              = _mm_andnot_ps(dummy_mask,r00);
 533
 534             /* Compute parameters for interactions between i and j atoms */
 535             qq00             = _mm_mul_ps(iq0,jq0);
 536             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 537                                          vdwparam+vdwioffset0+vdwjidx0B,
 538                                          vdwparam+vdwioffset0+vdwjidx0C,
 539                                          vdwparam+vdwioffset0+vdwjidx0D,
 540                                          &c6_00,&c12_00);
 541
 542             /* EWALD ELECTROSTATICS */
 543
 544             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 545             ewrt             = _mm_mul_ps(r00,ewtabscale);
 546             ewitab           = _mm_cvttps_epi32(ewrt);
 547             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 548             ewitab           = _mm_slli_epi32(ewitab,2);
 549             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 550             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 551             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 552             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 553             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 554             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 555             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 556             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 557             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 558
 559             /* LENNARD-JONES DISPERSION/REPULSION */
 560
 561             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 562             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 563             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 564             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 565             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 566
 567             d                = _mm_sub_ps(r00,rswitch);
 568             d                = _mm_max_ps(d,_mm_setzero_ps());
 569             d2               = _mm_mul_ps(d,d);
 570             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 571
 572             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 573
 574             /* Evaluate switch function */
 575             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 576             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
 577             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 578             velec            = _mm_mul_ps(velec,sw);
 579             vvdw             = _mm_mul_ps(vvdw,sw);
 580             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 581
 582             /* Update potential sum for this i atom from the interaction with this j atom. */
 583             velec            = _mm_and_ps(velec,cutoff_mask);
 584             velec            = _mm_andnot_ps(dummy_mask,velec);
 585             velecsum         = _mm_add_ps(velecsum,velec);
 586             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 587             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 588             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 589
 590             fscal            = _mm_add_ps(felec,fvdw);
 591
 592             fscal            = _mm_and_ps(fscal,cutoff_mask);
 593
 594             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 595
 596             /* Calculate temporary vectorial force */
 597             tx               = _mm_mul_ps(fscal,dx00);
 598             ty               = _mm_mul_ps(fscal,dy00);
 599             tz               = _mm_mul_ps(fscal,dz00);
 600
 601             /* Update vectorial force */
 602             fix0             = _mm_add_ps(fix0,tx);
 603             fiy0             = _mm_add_ps(fiy0,ty);
 604             fiz0             = _mm_add_ps(fiz0,tz);
 605
 606             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 607             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 608             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 609             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 610             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 611
 612             }
 613
 614             /**************************
 615              * CALCULATE INTERACTIONS *
 616              **************************/
 617
 618             if (gmx_mm_any_lt(rsq10,rcutoff2))
 619             {
 620
 621             r10              = _mm_mul_ps(rsq10,rinv10);
 622             r10              = _mm_andnot_ps(dummy_mask,r10);
 623
 624             /* Compute parameters for interactions between i and j atoms */
 625             qq10             = _mm_mul_ps(iq1,jq0);
 626
 627             /* EWALD ELECTROSTATICS */
 628
 629             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 630             ewrt             = _mm_mul_ps(r10,ewtabscale);
 631             ewitab           = _mm_cvttps_epi32(ewrt);
 632             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 633             ewitab           = _mm_slli_epi32(ewitab,2);
 634             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 635             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 636             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 637             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 638             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 639             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 640             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 641             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 642             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 643
 644             d                = _mm_sub_ps(r10,rswitch);
 645             d                = _mm_max_ps(d,_mm_setzero_ps());
 646             d2               = _mm_mul_ps(d,d);
 647             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 648
 649             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 650
 651             /* Evaluate switch function */
 652             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 653             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
 654             velec            = _mm_mul_ps(velec,sw);
 655             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 656
 657             /* Update potential sum for this i atom from the interaction with this j atom. */
 658             velec            = _mm_and_ps(velec,cutoff_mask);
 659             velec            = _mm_andnot_ps(dummy_mask,velec);
 660             velecsum         = _mm_add_ps(velecsum,velec);
 661
 662             fscal            = felec;
 663
 664             fscal            = _mm_and_ps(fscal,cutoff_mask);
 665
 666             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 667
 668             /* Calculate temporary vectorial force */
 669             tx               = _mm_mul_ps(fscal,dx10);
 670             ty               = _mm_mul_ps(fscal,dy10);
 671             tz               = _mm_mul_ps(fscal,dz10);
 672
 673             /* Update vectorial force */
 674             fix1             = _mm_add_ps(fix1,tx);
 675             fiy1             = _mm_add_ps(fiy1,ty);
 676             fiz1             = _mm_add_ps(fiz1,tz);
 677
 678             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 679             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 680             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 681             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 682             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 683
 684             }
 685
 686             /**************************
 687              * CALCULATE INTERACTIONS *
 688              **************************/
 689
 690             if (gmx_mm_any_lt(rsq20,rcutoff2))
 691             {
 692
 693             r20              = _mm_mul_ps(rsq20,rinv20);
 694             r20              = _mm_andnot_ps(dummy_mask,r20);
 695
 696             /* Compute parameters for interactions between i and j atoms */
 697             qq20             = _mm_mul_ps(iq2,jq0);
 698
 699             /* EWALD ELECTROSTATICS */
 700
 701             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 702             ewrt             = _mm_mul_ps(r20,ewtabscale);
 703             ewitab           = _mm_cvttps_epi32(ewrt);
 704             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 705             ewitab           = _mm_slli_epi32(ewitab,2);
 706             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 707             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 708             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 709             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 710             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 711             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 712             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 713             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 714             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 715
 716             d                = _mm_sub_ps(r20,rswitch);
 717             d                = _mm_max_ps(d,_mm_setzero_ps());
 718             d2               = _mm_mul_ps(d,d);
 719             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 720
 721             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 722
 723             /* Evaluate switch function */
 724             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 725             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
 726             velec            = _mm_mul_ps(velec,sw);
 727             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 728
 729             /* Update potential sum for this i atom from the interaction with this j atom. */
 730             velec            = _mm_and_ps(velec,cutoff_mask);
 731             velec            = _mm_andnot_ps(dummy_mask,velec);
 732             velecsum         = _mm_add_ps(velecsum,velec);
 733
 734             fscal            = felec;
 735
 736             fscal            = _mm_and_ps(fscal,cutoff_mask);
 737
 738             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 739
 740             /* Calculate temporary vectorial force */
 741             tx               = _mm_mul_ps(fscal,dx20);
 742             ty               = _mm_mul_ps(fscal,dy20);
 743             tz               = _mm_mul_ps(fscal,dz20);
 744
 745             /* Update vectorial force */
 746             fix2             = _mm_add_ps(fix2,tx);
 747             fiy2             = _mm_add_ps(fiy2,ty);
 748             fiz2             = _mm_add_ps(fiz2,tz);
 749
 750             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 751             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 752             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 753             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 754             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 755
 756             }
 757
 758             /* Inner loop uses 216 flops */
 759         }
 760
 761         /* End of innermost loop */
 762
 763         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 764                                               f+i_coord_offset,fshift+i_shift_offset);
 765
 766         ggid                        = gid[iidx];
 767         /* Update potential energies */
 768         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 769         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 770
 771         /* Increment number of inner iterations */
 772         inneriter                  += j_index_end - j_index_start;
 773
 774         /* Outer loop uses 20 flops */
 775     }
 776
 777     /* Increment number of outer iterations */
 778     outeriter        += nri;
 779
 780     /* Update outer/inner flops */
 781
 782     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*216);
 783 }
 784 /*
 785  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_F_sse4_1_single
 786  * Electrostatics interaction: Ewald
 787  * VdW interaction:            LennardJones
 788  * Geometry:                   Water3-Particle
 789  * Calculate force/pot:        Force
 790  */
 791 void
 792 nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_F_sse4_1_single
 793                     (t_nblist * gmx_restrict                nlist,
 794                      rvec * gmx_restrict                    xx,
 795                      rvec * gmx_restrict                    ff,
 796                      t_forcerec * gmx_restrict              fr,
 797                      t_mdatoms * gmx_restrict               mdatoms,
 798                      nb_kernel_data_t * gmx_restrict        kernel_data,
 799                      t_nrnb * gmx_restrict                  nrnb)
 800 {
 801     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 802      * just 0 for non-waters.
 803      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 804      * jnr indices corresponding to data put in the four positions in the SIMD register.
 805      */
 806     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 807     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 808     int              jnrA,jnrB,jnrC,jnrD;
 809     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 810     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 811     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 812     real             rcutoff_scalar;
 813     real             *shiftvec,*fshift,*x,*f;
 814     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 815     real             scratch[4*DIM];
 816     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 817     int              vdwioffset0;
 818     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 819     int              vdwioffset1;
 820     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 821     int              vdwioffset2;
 822     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 823     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 824     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 825     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 826     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 827     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 828     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 829     real             *charge;
 830     int              nvdwtype;
 831     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 832     int              *vdwtype;
 833     real             *vdwparam;
 834     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 835     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 836     __m128i          ewitab;
 837     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 838     real             *ewtab;
 839     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 840     real             rswitch_scalar,d_scalar;
 841     __m128           dummy_mask,cutoff_mask;
 842     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 843     __m128           one     = _mm_set1_ps(1.0);
 844     __m128           two     = _mm_set1_ps(2.0);
 845     x                = xx[0];
 846     f                = ff[0];
 847
 848     nri              = nlist->nri;
 849     iinr             = nlist->iinr;
 850     jindex           = nlist->jindex;
 851     jjnr             = nlist->jjnr;
 852     shiftidx         = nlist->shift;
 853     gid              = nlist->gid;
 854     shiftvec         = fr->shift_vec[0];
 855     fshift           = fr->fshift[0];
 856     facel            = _mm_set1_ps(fr->epsfac);
 857     charge           = mdatoms->chargeA;
 858     nvdwtype         = fr->ntype;
 859     vdwparam         = fr->nbfp;
 860     vdwtype          = mdatoms->typeA;
 861
 862     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 863     ewtab            = fr->ic->tabq_coul_FDV0;
 864     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 865     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 866
 867     /* Setup water-specific parameters */
 868     inr              = nlist->iinr[0];
 869     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 870     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 871     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 872     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 873
 874     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 875     rcutoff_scalar   = fr->rcoulomb;
 876     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 877     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 878
 879     rswitch_scalar   = fr->rcoulomb_switch;
 880     rswitch          = _mm_set1_ps(rswitch_scalar);
 881     /* Setup switch parameters */
 882     d_scalar         = rcutoff_scalar-rswitch_scalar;
 883     d                = _mm_set1_ps(d_scalar);
 884     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 885     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 886     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 887     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 888     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 889     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 890
 891     /* Avoid stupid compiler warnings */
 892     jnrA = jnrB = jnrC = jnrD = 0;
 893     j_coord_offsetA = 0;
 894     j_coord_offsetB = 0;
 895     j_coord_offsetC = 0;
 896     j_coord_offsetD = 0;
 897
 898     outeriter        = 0;
 899     inneriter        = 0;
 900
 901     for(iidx=0;iidx<4*DIM;iidx++)
 902     {
 903         scratch[iidx] = 0.0;
 904     }
 905
 906     /* Start outer loop over neighborlists */
 907     for(iidx=0; iidx<nri; iidx++)
 908     {
 909         /* Load shift vector for this list */
 910         i_shift_offset   = DIM*shiftidx[iidx];
 911
 912         /* Load limits for loop over neighbors */
 913         j_index_start    = jindex[iidx];
 914         j_index_end      = jindex[iidx+1];
 915
 916         /* Get outer coordinate index */
 917         inr              = iinr[iidx];
 918         i_coord_offset   = DIM*inr;
 919
 920         /* Load i particle coords and add shift vector */
 921         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 922                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 923
 924         fix0             = _mm_setzero_ps();
 925         fiy0             = _mm_setzero_ps();
 926         fiz0             = _mm_setzero_ps();
 927         fix1             = _mm_setzero_ps();
 928         fiy1             = _mm_setzero_ps();
 929         fiz1             = _mm_setzero_ps();
 930         fix2             = _mm_setzero_ps();
 931         fiy2             = _mm_setzero_ps();
 932         fiz2             = _mm_setzero_ps();
 933
 934         /* Start inner kernel loop */
 935         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 936         {
 937
 938             /* Get j neighbor index, and coordinate index */
 939             jnrA             = jjnr[jidx];
 940             jnrB             = jjnr[jidx+1];
 941             jnrC             = jjnr[jidx+2];
 942             jnrD             = jjnr[jidx+3];
 943             j_coord_offsetA  = DIM*jnrA;
 944             j_coord_offsetB  = DIM*jnrB;
 945             j_coord_offsetC  = DIM*jnrC;
 946             j_coord_offsetD  = DIM*jnrD;
 947
 948             /* load j atom coordinates */
 949             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 950                                               x+j_coord_offsetC,x+j_coord_offsetD,
 951                                               &jx0,&jy0,&jz0);
 952
 953             /* Calculate displacement vector */
 954             dx00             = _mm_sub_ps(ix0,jx0);
 955             dy00             = _mm_sub_ps(iy0,jy0);
 956             dz00             = _mm_sub_ps(iz0,jz0);
 957             dx10             = _mm_sub_ps(ix1,jx0);
 958             dy10             = _mm_sub_ps(iy1,jy0);
 959             dz10             = _mm_sub_ps(iz1,jz0);
 960             dx20             = _mm_sub_ps(ix2,jx0);
 961             dy20             = _mm_sub_ps(iy2,jy0);
 962             dz20             = _mm_sub_ps(iz2,jz0);
 963
 964             /* Calculate squared distance and things based on it */
 965             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 966             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 967             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 968
 969             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 970             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 971             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 972
 973             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 974             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 975             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 976
 977             /* Load parameters for j particles */
 978             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 979                                                               charge+jnrC+0,charge+jnrD+0);
 980             vdwjidx0A        = 2*vdwtype[jnrA+0];
 981             vdwjidx0B        = 2*vdwtype[jnrB+0];
 982             vdwjidx0C        = 2*vdwtype[jnrC+0];
 983             vdwjidx0D        = 2*vdwtype[jnrD+0];
 984
 985             /**************************
 986              * CALCULATE INTERACTIONS *
 987              **************************/
 988
 989             if (gmx_mm_any_lt(rsq00,rcutoff2))
 990             {
 991
 992             r00              = _mm_mul_ps(rsq00,rinv00);
 993
 994             /* Compute parameters for interactions between i and j atoms */
 995             qq00             = _mm_mul_ps(iq0,jq0);
 996             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 997                                          vdwparam+vdwioffset0+vdwjidx0B,
 998                                          vdwparam+vdwioffset0+vdwjidx0C,
 999                                          vdwparam+vdwioffset0+vdwjidx0D,
1000                                          &c6_00,&c12_00);
1001
1002             /* EWALD ELECTROSTATICS */
1003
1004             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1005             ewrt             = _mm_mul_ps(r00,ewtabscale);
1006             ewitab           = _mm_cvttps_epi32(ewrt);
1007             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1008             ewitab           = _mm_slli_epi32(ewitab,2);
1009             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1010             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1011             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1012             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1013             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1014             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1015             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1016             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
1017             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1018
1019             /* LENNARD-JONES DISPERSION/REPULSION */
1020
1021             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1022             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
1023             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
1024             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
1025             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
1026
1027             d                = _mm_sub_ps(r00,rswitch);
1028             d                = _mm_max_ps(d,_mm_setzero_ps());
1029             d2               = _mm_mul_ps(d,d);
1030             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1031
1032             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1033
1034             /* Evaluate switch function */
1035             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1036             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
1037             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1038             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1039
1040             fscal            = _mm_add_ps(felec,fvdw);
1041
1042             fscal            = _mm_and_ps(fscal,cutoff_mask);
1043
1044             /* Calculate temporary vectorial force */
1045             tx               = _mm_mul_ps(fscal,dx00);
1046             ty               = _mm_mul_ps(fscal,dy00);
1047             tz               = _mm_mul_ps(fscal,dz00);
1048
1049             /* Update vectorial force */
1050             fix0             = _mm_add_ps(fix0,tx);
1051             fiy0             = _mm_add_ps(fiy0,ty);
1052             fiz0             = _mm_add_ps(fiz0,tz);
1053
1054             fjptrA             = f+j_coord_offsetA;
1055             fjptrB             = f+j_coord_offsetB;
1056             fjptrC             = f+j_coord_offsetC;
1057             fjptrD             = f+j_coord_offsetD;
1058             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1059
1060             }
1061
1062             /**************************
1063              * CALCULATE INTERACTIONS *
1064              **************************/
1065
1066             if (gmx_mm_any_lt(rsq10,rcutoff2))
1067             {
1068
1069             r10              = _mm_mul_ps(rsq10,rinv10);
1070
1071             /* Compute parameters for interactions between i and j atoms */
1072             qq10             = _mm_mul_ps(iq1,jq0);
1073
1074             /* EWALD ELECTROSTATICS */
1075
1076             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1077             ewrt             = _mm_mul_ps(r10,ewtabscale);
1078             ewitab           = _mm_cvttps_epi32(ewrt);
1079             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1080             ewitab           = _mm_slli_epi32(ewitab,2);
1081             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1082             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1083             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1084             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1085             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1086             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1087             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1088             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
1089             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1090
1091             d                = _mm_sub_ps(r10,rswitch);
1092             d                = _mm_max_ps(d,_mm_setzero_ps());
1093             d2               = _mm_mul_ps(d,d);
1094             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1095
1096             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1097
1098             /* Evaluate switch function */
1099             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1100             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
1101             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1102
1103             fscal            = felec;
1104
1105             fscal            = _mm_and_ps(fscal,cutoff_mask);
1106
1107             /* Calculate temporary vectorial force */
1108             tx               = _mm_mul_ps(fscal,dx10);
1109             ty               = _mm_mul_ps(fscal,dy10);
1110             tz               = _mm_mul_ps(fscal,dz10);
1111
1112             /* Update vectorial force */
1113             fix1             = _mm_add_ps(fix1,tx);
1114             fiy1             = _mm_add_ps(fiy1,ty);
1115             fiz1             = _mm_add_ps(fiz1,tz);
1116
1117             fjptrA             = f+j_coord_offsetA;
1118             fjptrB             = f+j_coord_offsetB;
1119             fjptrC             = f+j_coord_offsetC;
1120             fjptrD             = f+j_coord_offsetD;
1121             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1122
1123             }
1124
1125             /**************************
1126              * CALCULATE INTERACTIONS *
1127              **************************/
1128
1129             if (gmx_mm_any_lt(rsq20,rcutoff2))
1130             {
1131
1132             r20              = _mm_mul_ps(rsq20,rinv20);
1133
1134             /* Compute parameters for interactions between i and j atoms */
1135             qq20             = _mm_mul_ps(iq2,jq0);
1136
1137             /* EWALD ELECTROSTATICS */
1138
1139             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1140             ewrt             = _mm_mul_ps(r20,ewtabscale);
1141             ewitab           = _mm_cvttps_epi32(ewrt);
1142             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1143             ewitab           = _mm_slli_epi32(ewitab,2);
1144             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1145             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1146             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1147             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1148             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1149             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1150             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1151             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
1152             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1153
1154             d                = _mm_sub_ps(r20,rswitch);
1155             d                = _mm_max_ps(d,_mm_setzero_ps());
1156             d2               = _mm_mul_ps(d,d);
1157             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1158
1159             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1160
1161             /* Evaluate switch function */
1162             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1163             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
1164             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1165
1166             fscal            = felec;
1167
1168             fscal            = _mm_and_ps(fscal,cutoff_mask);
1169
1170             /* Calculate temporary vectorial force */
1171             tx               = _mm_mul_ps(fscal,dx20);
1172             ty               = _mm_mul_ps(fscal,dy20);
1173             tz               = _mm_mul_ps(fscal,dz20);
1174
1175             /* Update vectorial force */
1176             fix2             = _mm_add_ps(fix2,tx);
1177             fiy2             = _mm_add_ps(fiy2,ty);
1178             fiz2             = _mm_add_ps(fiz2,tz);
1179
1180             fjptrA             = f+j_coord_offsetA;
1181             fjptrB             = f+j_coord_offsetB;
1182             fjptrC             = f+j_coord_offsetC;
1183             fjptrD             = f+j_coord_offsetD;
1184             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1185
1186             }
1187
1188             /* Inner loop uses 201 flops */
1189         }
1190
1191         if(jidx<j_index_end)
1192         {
1193
1194             /* Get j neighbor index, and coordinate index */
1195             jnrlistA         = jjnr[jidx];
1196             jnrlistB         = jjnr[jidx+1];
1197             jnrlistC         = jjnr[jidx+2];
1198             jnrlistD         = jjnr[jidx+3];
1199             /* Sign of each element will be negative for non-real atoms.
1200              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1201              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1202              */
1203             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1204             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1205             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1206             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1207             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1208             j_coord_offsetA  = DIM*jnrA;
1209             j_coord_offsetB  = DIM*jnrB;
1210             j_coord_offsetC  = DIM*jnrC;
1211             j_coord_offsetD  = DIM*jnrD;
1212
1213             /* load j atom coordinates */
1214             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1215                                               x+j_coord_offsetC,x+j_coord_offsetD,
1216                                               &jx0,&jy0,&jz0);
1217
1218             /* Calculate displacement vector */
1219             dx00             = _mm_sub_ps(ix0,jx0);
1220             dy00             = _mm_sub_ps(iy0,jy0);
1221             dz00             = _mm_sub_ps(iz0,jz0);
1222             dx10             = _mm_sub_ps(ix1,jx0);
1223             dy10             = _mm_sub_ps(iy1,jy0);
1224             dz10             = _mm_sub_ps(iz1,jz0);
1225             dx20             = _mm_sub_ps(ix2,jx0);
1226             dy20             = _mm_sub_ps(iy2,jy0);
1227             dz20             = _mm_sub_ps(iz2,jz0);
1228
1229             /* Calculate squared distance and things based on it */
1230             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1231             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1232             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1233
1234             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1235             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1236             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1237
1238             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1239             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1240             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1241
1242             /* Load parameters for j particles */
1243             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1244                                                               charge+jnrC+0,charge+jnrD+0);
1245             vdwjidx0A        = 2*vdwtype[jnrA+0];
1246             vdwjidx0B        = 2*vdwtype[jnrB+0];
1247             vdwjidx0C        = 2*vdwtype[jnrC+0];
1248             vdwjidx0D        = 2*vdwtype[jnrD+0];
1249
1250             /**************************
1251              * CALCULATE INTERACTIONS *
1252              **************************/
1253
1254             if (gmx_mm_any_lt(rsq00,rcutoff2))
1255             {
1256
1257             r00              = _mm_mul_ps(rsq00,rinv00);
1258             r00              = _mm_andnot_ps(dummy_mask,r00);
1259
1260             /* Compute parameters for interactions between i and j atoms */
1261             qq00             = _mm_mul_ps(iq0,jq0);
1262             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1263                                          vdwparam+vdwioffset0+vdwjidx0B,
1264                                          vdwparam+vdwioffset0+vdwjidx0C,
1265                                          vdwparam+vdwioffset0+vdwjidx0D,
1266                                          &c6_00,&c12_00);
1267
1268             /* EWALD ELECTROSTATICS */
1269
1270             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1271             ewrt             = _mm_mul_ps(r00,ewtabscale);
1272             ewitab           = _mm_cvttps_epi32(ewrt);
1273             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1274             ewitab           = _mm_slli_epi32(ewitab,2);
1275             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1276             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1277             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1278             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1279             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1280             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1281             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1282             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
1283             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1284
1285             /* LENNARD-JONES DISPERSION/REPULSION */
1286
1287             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1288             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
1289             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
1290             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
1291             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
1292
1293             d                = _mm_sub_ps(r00,rswitch);
1294             d                = _mm_max_ps(d,_mm_setzero_ps());
1295             d2               = _mm_mul_ps(d,d);
1296             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1297
1298             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1299
1300             /* Evaluate switch function */
1301             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1302             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
1303             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1304             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1305
1306             fscal            = _mm_add_ps(felec,fvdw);
1307
1308             fscal            = _mm_and_ps(fscal,cutoff_mask);
1309
1310             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1311
1312             /* Calculate temporary vectorial force */
1313             tx               = _mm_mul_ps(fscal,dx00);
1314             ty               = _mm_mul_ps(fscal,dy00);
1315             tz               = _mm_mul_ps(fscal,dz00);
1316
1317             /* Update vectorial force */
1318             fix0             = _mm_add_ps(fix0,tx);
1319             fiy0             = _mm_add_ps(fiy0,ty);
1320             fiz0             = _mm_add_ps(fiz0,tz);
1321
1322             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1323             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1324             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1325             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1326             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1327
1328             }
1329
1330             /**************************
1331              * CALCULATE INTERACTIONS *
1332              **************************/
1333
1334             if (gmx_mm_any_lt(rsq10,rcutoff2))
1335             {
1336
1337             r10              = _mm_mul_ps(rsq10,rinv10);
1338             r10              = _mm_andnot_ps(dummy_mask,r10);
1339
1340             /* Compute parameters for interactions between i and j atoms */
1341             qq10             = _mm_mul_ps(iq1,jq0);
1342
1343             /* EWALD ELECTROSTATICS */
1344
1345             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1346             ewrt             = _mm_mul_ps(r10,ewtabscale);
1347             ewitab           = _mm_cvttps_epi32(ewrt);
1348             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1349             ewitab           = _mm_slli_epi32(ewitab,2);
1350             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1351             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1352             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1353             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1354             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1355             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1356             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1357             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
1358             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1359
1360             d                = _mm_sub_ps(r10,rswitch);
1361             d                = _mm_max_ps(d,_mm_setzero_ps());
1362             d2               = _mm_mul_ps(d,d);
1363             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1364
1365             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1366
1367             /* Evaluate switch function */
1368             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1369             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
1370             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1371
1372             fscal            = felec;
1373
1374             fscal            = _mm_and_ps(fscal,cutoff_mask);
1375
1376             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1377
1378             /* Calculate temporary vectorial force */
1379             tx               = _mm_mul_ps(fscal,dx10);
1380             ty               = _mm_mul_ps(fscal,dy10);
1381             tz               = _mm_mul_ps(fscal,dz10);
1382
1383             /* Update vectorial force */
1384             fix1             = _mm_add_ps(fix1,tx);
1385             fiy1             = _mm_add_ps(fiy1,ty);
1386             fiz1             = _mm_add_ps(fiz1,tz);
1387
1388             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1389             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1390             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1391             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1392             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1393
1394             }
1395
1396             /**************************
1397              * CALCULATE INTERACTIONS *
1398              **************************/
1399
1400             if (gmx_mm_any_lt(rsq20,rcutoff2))
1401             {
1402
1403             r20              = _mm_mul_ps(rsq20,rinv20);
1404             r20              = _mm_andnot_ps(dummy_mask,r20);
1405
1406             /* Compute parameters for interactions between i and j atoms */
1407             qq20             = _mm_mul_ps(iq2,jq0);
1408
1409             /* EWALD ELECTROSTATICS */
1410
1411             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1412             ewrt             = _mm_mul_ps(r20,ewtabscale);
1413             ewitab           = _mm_cvttps_epi32(ewrt);
1414             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1415             ewitab           = _mm_slli_epi32(ewitab,2);
1416             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1417             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1418             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1419             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1420             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1421             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1422             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1423             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
1424             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1425
1426             d                = _mm_sub_ps(r20,rswitch);
1427             d                = _mm_max_ps(d,_mm_setzero_ps());
1428             d2               = _mm_mul_ps(d,d);
1429             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1430
1431             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1432
1433             /* Evaluate switch function */
1434             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1435             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
1436             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1437
1438             fscal            = felec;
1439
1440             fscal            = _mm_and_ps(fscal,cutoff_mask);
1441
1442             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1443
1444             /* Calculate temporary vectorial force */
1445             tx               = _mm_mul_ps(fscal,dx20);
1446             ty               = _mm_mul_ps(fscal,dy20);
1447             tz               = _mm_mul_ps(fscal,dz20);
1448
1449             /* Update vectorial force */
1450             fix2             = _mm_add_ps(fix2,tx);
1451             fiy2             = _mm_add_ps(fiy2,ty);
1452             fiz2             = _mm_add_ps(fiz2,tz);
1453
1454             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1455             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1456             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1457             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1458             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1459
1460             }
1461
1462             /* Inner loop uses 204 flops */
1463         }
1464
1465         /* End of innermost loop */
1466
1467         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1468                                               f+i_coord_offset,fshift+i_shift_offset);
1469
1470         /* Increment number of inner iterations */
1471         inneriter                  += j_index_end - j_index_start;
1472
1473         /* Outer loop uses 18 flops */
1474     }
1475
1476     /* Increment number of outer iterations */
1477     outeriter        += nri;
1478
1479     /* Update outer/inner flops */
1480
1481     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*204);
1482 }