src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_sse2_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse2_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_sse2_single.h"
  34 #include "kernelutil_x86_sse2_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_VF_sse2_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_VF_sse2_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              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             shX,shY,shZ,rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwioffset1;
  69     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  70     int              vdwioffset2;
  71     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  72     int              vdwioffset3;
  73     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  74     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  75     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  76     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  77     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  78     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  79     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  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           dummy_mask,cutoff_mask;
  92     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  93     __m128           one     = _mm_set1_ps(1.0);
  94     __m128           two     = _mm_set1_ps(2.0);
  95     x                = xx[0];
  96     f                = ff[0];
  97
  98     nri              = nlist->nri;
  99     iinr             = nlist->iinr;
 100     jindex           = nlist->jindex;
 101     jjnr             = nlist->jjnr;
 102     shiftidx         = nlist->shift;
 103     gid              = nlist->gid;
 104     shiftvec         = fr->shift_vec[0];
 105     fshift           = fr->fshift[0];
 106     facel            = _mm_set1_ps(fr->epsfac);
 107     charge           = mdatoms->chargeA;
 108     nvdwtype         = fr->ntype;
 109     vdwparam         = fr->nbfp;
 110     vdwtype          = mdatoms->typeA;
 111
 112     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 113     ewtab            = fr->ic->tabq_coul_FDV0;
 114     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 115     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 116
 117     /* Setup water-specific parameters */
 118     inr              = nlist->iinr[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     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 122     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 123
 124     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 125     rcutoff_scalar   = fr->rcoulomb;
 126     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 127     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 128
 129     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 130     rvdw             = _mm_set1_ps(fr->rvdw);
 131
 132     /* Avoid stupid compiler warnings */
 133     jnrA = jnrB = jnrC = jnrD = 0;
 134     j_coord_offsetA = 0;
 135     j_coord_offsetB = 0;
 136     j_coord_offsetC = 0;
 137     j_coord_offsetD = 0;
 138
 139     outeriter        = 0;
 140     inneriter        = 0;
 141
 142     /* Start outer loop over neighborlists */
 143     for(iidx=0; iidx<nri; iidx++)
 144     {
 145         /* Load shift vector for this list */
 146         i_shift_offset   = DIM*shiftidx[iidx];
 147         shX              = shiftvec[i_shift_offset+XX];
 148         shY              = shiftvec[i_shift_offset+YY];
 149         shZ              = shiftvec[i_shift_offset+ZZ];
 150
 151         /* Load limits for loop over neighbors */
 152         j_index_start    = jindex[iidx];
 153         j_index_end      = jindex[iidx+1];
 154
 155         /* Get outer coordinate index */
 156         inr              = iinr[iidx];
 157         i_coord_offset   = DIM*inr;
 158
 159         /* Load i particle coords and add shift vector */
 160         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 161         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 162         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 163         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 164         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 165         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 166         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 167         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 168         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 169         ix3              = _mm_set1_ps(shX + x[i_coord_offset+DIM*3+XX]);
 170         iy3              = _mm_set1_ps(shY + x[i_coord_offset+DIM*3+YY]);
 171         iz3              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*3+ZZ]);
 172
 173         fix0             = _mm_setzero_ps();
 174         fiy0             = _mm_setzero_ps();
 175         fiz0             = _mm_setzero_ps();
 176         fix1             = _mm_setzero_ps();
 177         fiy1             = _mm_setzero_ps();
 178         fiz1             = _mm_setzero_ps();
 179         fix2             = _mm_setzero_ps();
 180         fiy2             = _mm_setzero_ps();
 181         fiz2             = _mm_setzero_ps();
 182         fix3             = _mm_setzero_ps();
 183         fiy3             = _mm_setzero_ps();
 184         fiz3             = _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
 200             j_coord_offsetA  = DIM*jnrA;
 201             j_coord_offsetB  = DIM*jnrB;
 202             j_coord_offsetC  = DIM*jnrC;
 203             j_coord_offsetD  = DIM*jnrD;
 204
 205             /* load j atom coordinates */
 206             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 207                                               x+j_coord_offsetC,x+j_coord_offsetD,
 208                                               &jx0,&jy0,&jz0);
 209
 210             /* Calculate displacement vector */
 211             dx00             = _mm_sub_ps(ix0,jx0);
 212             dy00             = _mm_sub_ps(iy0,jy0);
 213             dz00             = _mm_sub_ps(iz0,jz0);
 214             dx10             = _mm_sub_ps(ix1,jx0);
 215             dy10             = _mm_sub_ps(iy1,jy0);
 216             dz10             = _mm_sub_ps(iz1,jz0);
 217             dx20             = _mm_sub_ps(ix2,jx0);
 218             dy20             = _mm_sub_ps(iy2,jy0);
 219             dz20             = _mm_sub_ps(iz2,jz0);
 220             dx30             = _mm_sub_ps(ix3,jx0);
 221             dy30             = _mm_sub_ps(iy3,jy0);
 222             dz30             = _mm_sub_ps(iz3,jz0);
 223
 224             /* Calculate squared distance and things based on it */
 225             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 226             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 227             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 228             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 229
 230             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 231             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 232             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 233
 234             rinvsq00         = gmx_mm_inv_ps(rsq00);
 235             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 236             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 237             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 238
 239             /* Load parameters for j particles */
 240             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 241                                                               charge+jnrC+0,charge+jnrD+0);
 242             vdwjidx0A        = 2*vdwtype[jnrA+0];
 243             vdwjidx0B        = 2*vdwtype[jnrB+0];
 244             vdwjidx0C        = 2*vdwtype[jnrC+0];
 245             vdwjidx0D        = 2*vdwtype[jnrD+0];
 246
 247             /**************************
 248              * CALCULATE INTERACTIONS *
 249              **************************/
 250
 251             if (gmx_mm_any_lt(rsq00,rcutoff2))
 252             {
 253
 254             /* Compute parameters for interactions between i and j atoms */
 255             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 256                                          vdwparam+vdwioffset0+vdwjidx0B,
 257                                          vdwparam+vdwioffset0+vdwjidx0C,
 258                                          vdwparam+vdwioffset0+vdwjidx0D,
 259                                          &c6_00,&c12_00);
 260
 261             /* LENNARD-JONES DISPERSION/REPULSION */
 262
 263             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 264             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 265             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 266             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 267                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 268             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 269
 270             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 271
 272             /* Update potential sum for this i atom from the interaction with this j atom. */
 273             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 274             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 275
 276             fscal            = fvdw;
 277
 278             fscal            = _mm_and_ps(fscal,cutoff_mask);
 279
 280             /* Calculate temporary vectorial force */
 281             tx               = _mm_mul_ps(fscal,dx00);
 282             ty               = _mm_mul_ps(fscal,dy00);
 283             tz               = _mm_mul_ps(fscal,dz00);
 284
 285             /* Update vectorial force */
 286             fix0             = _mm_add_ps(fix0,tx);
 287             fiy0             = _mm_add_ps(fiy0,ty);
 288             fiz0             = _mm_add_ps(fiz0,tz);
 289
 290             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 291                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 292                                                    tx,ty,tz);
 293
 294             }
 295
 296             /**************************
 297              * CALCULATE INTERACTIONS *
 298              **************************/
 299
 300             if (gmx_mm_any_lt(rsq10,rcutoff2))
 301             {
 302
 303             r10              = _mm_mul_ps(rsq10,rinv10);
 304
 305             /* Compute parameters for interactions between i and j atoms */
 306             qq10             = _mm_mul_ps(iq1,jq0);
 307
 308             /* EWALD ELECTROSTATICS */
 309
 310             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 311             ewrt             = _mm_mul_ps(r10,ewtabscale);
 312             ewitab           = _mm_cvttps_epi32(ewrt);
 313             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 314             ewitab           = _mm_slli_epi32(ewitab,2);
 315             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 316             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 317             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 318             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 319             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 320             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 321             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 322             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 323             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 324
 325             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 326
 327             /* Update potential sum for this i atom from the interaction with this j atom. */
 328             velec            = _mm_and_ps(velec,cutoff_mask);
 329             velecsum         = _mm_add_ps(velecsum,velec);
 330
 331             fscal            = felec;
 332
 333             fscal            = _mm_and_ps(fscal,cutoff_mask);
 334
 335             /* Calculate temporary vectorial force */
 336             tx               = _mm_mul_ps(fscal,dx10);
 337             ty               = _mm_mul_ps(fscal,dy10);
 338             tz               = _mm_mul_ps(fscal,dz10);
 339
 340             /* Update vectorial force */
 341             fix1             = _mm_add_ps(fix1,tx);
 342             fiy1             = _mm_add_ps(fiy1,ty);
 343             fiz1             = _mm_add_ps(fiz1,tz);
 344
 345             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 346                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 347                                                    tx,ty,tz);
 348
 349             }
 350
 351             /**************************
 352              * CALCULATE INTERACTIONS *
 353              **************************/
 354
 355             if (gmx_mm_any_lt(rsq20,rcutoff2))
 356             {
 357
 358             r20              = _mm_mul_ps(rsq20,rinv20);
 359
 360             /* Compute parameters for interactions between i and j atoms */
 361             qq20             = _mm_mul_ps(iq2,jq0);
 362
 363             /* EWALD ELECTROSTATICS */
 364
 365             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 366             ewrt             = _mm_mul_ps(r20,ewtabscale);
 367             ewitab           = _mm_cvttps_epi32(ewrt);
 368             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 369             ewitab           = _mm_slli_epi32(ewitab,2);
 370             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 371             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 372             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 373             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 374             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 375             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 376             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 377             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 378             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 379
 380             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 381
 382             /* Update potential sum for this i atom from the interaction with this j atom. */
 383             velec            = _mm_and_ps(velec,cutoff_mask);
 384             velecsum         = _mm_add_ps(velecsum,velec);
 385
 386             fscal            = felec;
 387
 388             fscal            = _mm_and_ps(fscal,cutoff_mask);
 389
 390             /* Calculate temporary vectorial force */
 391             tx               = _mm_mul_ps(fscal,dx20);
 392             ty               = _mm_mul_ps(fscal,dy20);
 393             tz               = _mm_mul_ps(fscal,dz20);
 394
 395             /* Update vectorial force */
 396             fix2             = _mm_add_ps(fix2,tx);
 397             fiy2             = _mm_add_ps(fiy2,ty);
 398             fiz2             = _mm_add_ps(fiz2,tz);
 399
 400             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 401                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 402                                                    tx,ty,tz);
 403
 404             }
 405
 406             /**************************
 407              * CALCULATE INTERACTIONS *
 408              **************************/
 409
 410             if (gmx_mm_any_lt(rsq30,rcutoff2))
 411             {
 412
 413             r30              = _mm_mul_ps(rsq30,rinv30);
 414
 415             /* Compute parameters for interactions between i and j atoms */
 416             qq30             = _mm_mul_ps(iq3,jq0);
 417
 418             /* EWALD ELECTROSTATICS */
 419
 420             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 421             ewrt             = _mm_mul_ps(r30,ewtabscale);
 422             ewitab           = _mm_cvttps_epi32(ewrt);
 423             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 424             ewitab           = _mm_slli_epi32(ewitab,2);
 425             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 426             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 427             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 428             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 429             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 430             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 431             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 432             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_sub_ps(rinv30,sh_ewald),velec));
 433             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 434
 435             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 436
 437             /* Update potential sum for this i atom from the interaction with this j atom. */
 438             velec            = _mm_and_ps(velec,cutoff_mask);
 439             velecsum         = _mm_add_ps(velecsum,velec);
 440
 441             fscal            = felec;
 442
 443             fscal            = _mm_and_ps(fscal,cutoff_mask);
 444
 445             /* Calculate temporary vectorial force */
 446             tx               = _mm_mul_ps(fscal,dx30);
 447             ty               = _mm_mul_ps(fscal,dy30);
 448             tz               = _mm_mul_ps(fscal,dz30);
 449
 450             /* Update vectorial force */
 451             fix3             = _mm_add_ps(fix3,tx);
 452             fiy3             = _mm_add_ps(fiy3,ty);
 453             fiz3             = _mm_add_ps(fiz3,tz);
 454
 455             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 456                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 457                                                    tx,ty,tz);
 458
 459             }
 460
 461             /* Inner loop uses 179 flops */
 462         }
 463
 464         if(jidx<j_index_end)
 465         {
 466
 467             /* Get j neighbor index, and coordinate index */
 468             jnrA             = jjnr[jidx];
 469             jnrB             = jjnr[jidx+1];
 470             jnrC             = jjnr[jidx+2];
 471             jnrD             = jjnr[jidx+3];
 472
 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       = (jnrA>=0) ? jnrA : 0;
 479             jnrB       = (jnrB>=0) ? jnrB : 0;
 480             jnrC       = (jnrC>=0) ? jnrC : 0;
 481             jnrD       = (jnrD>=0) ? jnrD : 0;
 482
 483             j_coord_offsetA  = DIM*jnrA;
 484             j_coord_offsetB  = DIM*jnrB;
 485             j_coord_offsetC  = DIM*jnrC;
 486             j_coord_offsetD  = DIM*jnrD;
 487
 488             /* load j atom coordinates */
 489             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 490                                               x+j_coord_offsetC,x+j_coord_offsetD,
 491                                               &jx0,&jy0,&jz0);
 492
 493             /* Calculate displacement vector */
 494             dx00             = _mm_sub_ps(ix0,jx0);
 495             dy00             = _mm_sub_ps(iy0,jy0);
 496             dz00             = _mm_sub_ps(iz0,jz0);
 497             dx10             = _mm_sub_ps(ix1,jx0);
 498             dy10             = _mm_sub_ps(iy1,jy0);
 499             dz10             = _mm_sub_ps(iz1,jz0);
 500             dx20             = _mm_sub_ps(ix2,jx0);
 501             dy20             = _mm_sub_ps(iy2,jy0);
 502             dz20             = _mm_sub_ps(iz2,jz0);
 503             dx30             = _mm_sub_ps(ix3,jx0);
 504             dy30             = _mm_sub_ps(iy3,jy0);
 505             dz30             = _mm_sub_ps(iz3,jz0);
 506
 507             /* Calculate squared distance and things based on it */
 508             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 509             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 510             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 511             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 512
 513             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 514             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 515             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 516
 517             rinvsq00         = gmx_mm_inv_ps(rsq00);
 518             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 519             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 520             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 521
 522             /* Load parameters for j particles */
 523             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 524                                                               charge+jnrC+0,charge+jnrD+0);
 525             vdwjidx0A        = 2*vdwtype[jnrA+0];
 526             vdwjidx0B        = 2*vdwtype[jnrB+0];
 527             vdwjidx0C        = 2*vdwtype[jnrC+0];
 528             vdwjidx0D        = 2*vdwtype[jnrD+0];
 529
 530             /**************************
 531              * CALCULATE INTERACTIONS *
 532              **************************/
 533
 534             if (gmx_mm_any_lt(rsq00,rcutoff2))
 535             {
 536
 537             /* Compute parameters for interactions between i and j atoms */
 538             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 539                                          vdwparam+vdwioffset0+vdwjidx0B,
 540                                          vdwparam+vdwioffset0+vdwjidx0C,
 541                                          vdwparam+vdwioffset0+vdwjidx0D,
 542                                          &c6_00,&c12_00);
 543
 544             /* LENNARD-JONES DISPERSION/REPULSION */
 545
 546             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 547             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 548             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 549             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 550                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 551             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 552
 553             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 554
 555             /* Update potential sum for this i atom from the interaction with this j atom. */
 556             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 557             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 558             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 559
 560             fscal            = fvdw;
 561
 562             fscal            = _mm_and_ps(fscal,cutoff_mask);
 563
 564             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 565
 566             /* Calculate temporary vectorial force */
 567             tx               = _mm_mul_ps(fscal,dx00);
 568             ty               = _mm_mul_ps(fscal,dy00);
 569             tz               = _mm_mul_ps(fscal,dz00);
 570
 571             /* Update vectorial force */
 572             fix0             = _mm_add_ps(fix0,tx);
 573             fiy0             = _mm_add_ps(fiy0,ty);
 574             fiz0             = _mm_add_ps(fiz0,tz);
 575
 576             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 577                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 578                                                    tx,ty,tz);
 579
 580             }
 581
 582             /**************************
 583              * CALCULATE INTERACTIONS *
 584              **************************/
 585
 586             if (gmx_mm_any_lt(rsq10,rcutoff2))
 587             {
 588
 589             r10              = _mm_mul_ps(rsq10,rinv10);
 590             r10              = _mm_andnot_ps(dummy_mask,r10);
 591
 592             /* Compute parameters for interactions between i and j atoms */
 593             qq10             = _mm_mul_ps(iq1,jq0);
 594
 595             /* EWALD ELECTROSTATICS */
 596
 597             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 598             ewrt             = _mm_mul_ps(r10,ewtabscale);
 599             ewitab           = _mm_cvttps_epi32(ewrt);
 600             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 601             ewitab           = _mm_slli_epi32(ewitab,2);
 602             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 603             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 604             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 605             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 606             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 607             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 608             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 609             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 610             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 611
 612             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 613
 614             /* Update potential sum for this i atom from the interaction with this j atom. */
 615             velec            = _mm_and_ps(velec,cutoff_mask);
 616             velec            = _mm_andnot_ps(dummy_mask,velec);
 617             velecsum         = _mm_add_ps(velecsum,velec);
 618
 619             fscal            = felec;
 620
 621             fscal            = _mm_and_ps(fscal,cutoff_mask);
 622
 623             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 624
 625             /* Calculate temporary vectorial force */
 626             tx               = _mm_mul_ps(fscal,dx10);
 627             ty               = _mm_mul_ps(fscal,dy10);
 628             tz               = _mm_mul_ps(fscal,dz10);
 629
 630             /* Update vectorial force */
 631             fix1             = _mm_add_ps(fix1,tx);
 632             fiy1             = _mm_add_ps(fiy1,ty);
 633             fiz1             = _mm_add_ps(fiz1,tz);
 634
 635             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 636                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 637                                                    tx,ty,tz);
 638
 639             }
 640
 641             /**************************
 642              * CALCULATE INTERACTIONS *
 643              **************************/
 644
 645             if (gmx_mm_any_lt(rsq20,rcutoff2))
 646             {
 647
 648             r20              = _mm_mul_ps(rsq20,rinv20);
 649             r20              = _mm_andnot_ps(dummy_mask,r20);
 650
 651             /* Compute parameters for interactions between i and j atoms */
 652             qq20             = _mm_mul_ps(iq2,jq0);
 653
 654             /* EWALD ELECTROSTATICS */
 655
 656             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 657             ewrt             = _mm_mul_ps(r20,ewtabscale);
 658             ewitab           = _mm_cvttps_epi32(ewrt);
 659             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 660             ewitab           = _mm_slli_epi32(ewitab,2);
 661             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 662             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 663             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 664             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 665             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 666             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 667             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 668             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 669             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 670
 671             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 672
 673             /* Update potential sum for this i atom from the interaction with this j atom. */
 674             velec            = _mm_and_ps(velec,cutoff_mask);
 675             velec            = _mm_andnot_ps(dummy_mask,velec);
 676             velecsum         = _mm_add_ps(velecsum,velec);
 677
 678             fscal            = felec;
 679
 680             fscal            = _mm_and_ps(fscal,cutoff_mask);
 681
 682             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 683
 684             /* Calculate temporary vectorial force */
 685             tx               = _mm_mul_ps(fscal,dx20);
 686             ty               = _mm_mul_ps(fscal,dy20);
 687             tz               = _mm_mul_ps(fscal,dz20);
 688
 689             /* Update vectorial force */
 690             fix2             = _mm_add_ps(fix2,tx);
 691             fiy2             = _mm_add_ps(fiy2,ty);
 692             fiz2             = _mm_add_ps(fiz2,tz);
 693
 694             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 695                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 696                                                    tx,ty,tz);
 697
 698             }
 699
 700             /**************************
 701              * CALCULATE INTERACTIONS *
 702              **************************/
 703
 704             if (gmx_mm_any_lt(rsq30,rcutoff2))
 705             {
 706
 707             r30              = _mm_mul_ps(rsq30,rinv30);
 708             r30              = _mm_andnot_ps(dummy_mask,r30);
 709
 710             /* Compute parameters for interactions between i and j atoms */
 711             qq30             = _mm_mul_ps(iq3,jq0);
 712
 713             /* EWALD ELECTROSTATICS */
 714
 715             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 716             ewrt             = _mm_mul_ps(r30,ewtabscale);
 717             ewitab           = _mm_cvttps_epi32(ewrt);
 718             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 719             ewitab           = _mm_slli_epi32(ewitab,2);
 720             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 721             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 722             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 723             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 724             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 725             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 726             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 727             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_sub_ps(rinv30,sh_ewald),velec));
 728             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 729
 730             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 731
 732             /* Update potential sum for this i atom from the interaction with this j atom. */
 733             velec            = _mm_and_ps(velec,cutoff_mask);
 734             velec            = _mm_andnot_ps(dummy_mask,velec);
 735             velecsum         = _mm_add_ps(velecsum,velec);
 736
 737             fscal            = felec;
 738
 739             fscal            = _mm_and_ps(fscal,cutoff_mask);
 740
 741             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 742
 743             /* Calculate temporary vectorial force */
 744             tx               = _mm_mul_ps(fscal,dx30);
 745             ty               = _mm_mul_ps(fscal,dy30);
 746             tz               = _mm_mul_ps(fscal,dz30);
 747
 748             /* Update vectorial force */
 749             fix3             = _mm_add_ps(fix3,tx);
 750             fiy3             = _mm_add_ps(fiy3,ty);
 751             fiz3             = _mm_add_ps(fiz3,tz);
 752
 753             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 754                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 755                                                    tx,ty,tz);
 756
 757             }
 758
 759             /* Inner loop uses 182 flops */
 760         }
 761
 762         /* End of innermost loop */
 763
 764         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 765                                               f+i_coord_offset,fshift+i_shift_offset);
 766
 767         ggid                        = gid[iidx];
 768         /* Update potential energies */
 769         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 770         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 771
 772         /* Increment number of inner iterations */
 773         inneriter                  += j_index_end - j_index_start;
 774
 775         /* Outer loop uses 38 flops */
 776     }
 777
 778     /* Increment number of outer iterations */
 779     outeriter        += nri;
 780
 781     /* Update outer/inner flops */
 782
 783     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*38 + inneriter*182);
 784 }
 785 /*
 786  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_sse2_single
 787  * Electrostatics interaction: Ewald
 788  * VdW interaction:            LennardJones
 789  * Geometry:                   Water4-Particle
 790  * Calculate force/pot:        Force
 791  */
 792 void
 793 nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_sse2_single
 794                     (t_nblist * gmx_restrict                nlist,
 795                      rvec * gmx_restrict                    xx,
 796                      rvec * gmx_restrict                    ff,
 797                      t_forcerec * gmx_restrict              fr,
 798                      t_mdatoms * gmx_restrict               mdatoms,
 799                      nb_kernel_data_t * gmx_restrict        kernel_data,
 800                      t_nrnb * gmx_restrict                  nrnb)
 801 {
 802     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 803      * just 0 for non-waters.
 804      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 805      * jnr indices corresponding to data put in the four positions in the SIMD register.
 806      */
 807     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 808     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 809     int              jnrA,jnrB,jnrC,jnrD;
 810     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 811     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 812     real             shX,shY,shZ,rcutoff_scalar;
 813     real             *shiftvec,*fshift,*x,*f;
 814     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 815     int              vdwioffset0;
 816     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 817     int              vdwioffset1;
 818     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 819     int              vdwioffset2;
 820     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 821     int              vdwioffset3;
 822     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 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           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 829     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 830     real             *charge;
 831     int              nvdwtype;
 832     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 833     int              *vdwtype;
 834     real             *vdwparam;
 835     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 836     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 837     __m128i          ewitab;
 838     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 839     real             *ewtab;
 840     __m128           dummy_mask,cutoff_mask;
 841     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 842     __m128           one     = _mm_set1_ps(1.0);
 843     __m128           two     = _mm_set1_ps(2.0);
 844     x                = xx[0];
 845     f                = ff[0];
 846
 847     nri              = nlist->nri;
 848     iinr             = nlist->iinr;
 849     jindex           = nlist->jindex;
 850     jjnr             = nlist->jjnr;
 851     shiftidx         = nlist->shift;
 852     gid              = nlist->gid;
 853     shiftvec         = fr->shift_vec[0];
 854     fshift           = fr->fshift[0];
 855     facel            = _mm_set1_ps(fr->epsfac);
 856     charge           = mdatoms->chargeA;
 857     nvdwtype         = fr->ntype;
 858     vdwparam         = fr->nbfp;
 859     vdwtype          = mdatoms->typeA;
 860
 861     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 862     ewtab            = fr->ic->tabq_coul_F;
 863     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 864     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 865
 866     /* Setup water-specific parameters */
 867     inr              = nlist->iinr[0];
 868     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 869     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 870     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 871     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 872
 873     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 874     rcutoff_scalar   = fr->rcoulomb;
 875     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 876     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 877
 878     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 879     rvdw             = _mm_set1_ps(fr->rvdw);
 880
 881     /* Avoid stupid compiler warnings */
 882     jnrA = jnrB = jnrC = jnrD = 0;
 883     j_coord_offsetA = 0;
 884     j_coord_offsetB = 0;
 885     j_coord_offsetC = 0;
 886     j_coord_offsetD = 0;
 887
 888     outeriter        = 0;
 889     inneriter        = 0;
 890
 891     /* Start outer loop over neighborlists */
 892     for(iidx=0; iidx<nri; iidx++)
 893     {
 894         /* Load shift vector for this list */
 895         i_shift_offset   = DIM*shiftidx[iidx];
 896         shX              = shiftvec[i_shift_offset+XX];
 897         shY              = shiftvec[i_shift_offset+YY];
 898         shZ              = shiftvec[i_shift_offset+ZZ];
 899
 900         /* Load limits for loop over neighbors */
 901         j_index_start    = jindex[iidx];
 902         j_index_end      = jindex[iidx+1];
 903
 904         /* Get outer coordinate index */
 905         inr              = iinr[iidx];
 906         i_coord_offset   = DIM*inr;
 907
 908         /* Load i particle coords and add shift vector */
 909         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 910         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 911         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 912         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 913         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 914         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 915         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 916         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 917         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 918         ix3              = _mm_set1_ps(shX + x[i_coord_offset+DIM*3+XX]);
 919         iy3              = _mm_set1_ps(shY + x[i_coord_offset+DIM*3+YY]);
 920         iz3              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*3+ZZ]);
 921
 922         fix0             = _mm_setzero_ps();
 923         fiy0             = _mm_setzero_ps();
 924         fiz0             = _mm_setzero_ps();
 925         fix1             = _mm_setzero_ps();
 926         fiy1             = _mm_setzero_ps();
 927         fiz1             = _mm_setzero_ps();
 928         fix2             = _mm_setzero_ps();
 929         fiy2             = _mm_setzero_ps();
 930         fiz2             = _mm_setzero_ps();
 931         fix3             = _mm_setzero_ps();
 932         fiy3             = _mm_setzero_ps();
 933         fiz3             = _mm_setzero_ps();
 934
 935         /* Start inner kernel loop */
 936         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 937         {
 938
 939             /* Get j neighbor index, and coordinate index */
 940             jnrA             = jjnr[jidx];
 941             jnrB             = jjnr[jidx+1];
 942             jnrC             = jjnr[jidx+2];
 943             jnrD             = jjnr[jidx+3];
 944
 945             j_coord_offsetA  = DIM*jnrA;
 946             j_coord_offsetB  = DIM*jnrB;
 947             j_coord_offsetC  = DIM*jnrC;
 948             j_coord_offsetD  = DIM*jnrD;
 949
 950             /* load j atom coordinates */
 951             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 952                                               x+j_coord_offsetC,x+j_coord_offsetD,
 953                                               &jx0,&jy0,&jz0);
 954
 955             /* Calculate displacement vector */
 956             dx00             = _mm_sub_ps(ix0,jx0);
 957             dy00             = _mm_sub_ps(iy0,jy0);
 958             dz00             = _mm_sub_ps(iz0,jz0);
 959             dx10             = _mm_sub_ps(ix1,jx0);
 960             dy10             = _mm_sub_ps(iy1,jy0);
 961             dz10             = _mm_sub_ps(iz1,jz0);
 962             dx20             = _mm_sub_ps(ix2,jx0);
 963             dy20             = _mm_sub_ps(iy2,jy0);
 964             dz20             = _mm_sub_ps(iz2,jz0);
 965             dx30             = _mm_sub_ps(ix3,jx0);
 966             dy30             = _mm_sub_ps(iy3,jy0);
 967             dz30             = _mm_sub_ps(iz3,jz0);
 968
 969             /* Calculate squared distance and things based on it */
 970             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 971             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 972             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 973             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 974
 975             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 976             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 977             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 978
 979             rinvsq00         = gmx_mm_inv_ps(rsq00);
 980             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 981             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 982             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 983
 984             /* Load parameters for j particles */
 985             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 986                                                               charge+jnrC+0,charge+jnrD+0);
 987             vdwjidx0A        = 2*vdwtype[jnrA+0];
 988             vdwjidx0B        = 2*vdwtype[jnrB+0];
 989             vdwjidx0C        = 2*vdwtype[jnrC+0];
 990             vdwjidx0D        = 2*vdwtype[jnrD+0];
 991
 992             /**************************
 993              * CALCULATE INTERACTIONS *
 994              **************************/
 995
 996             if (gmx_mm_any_lt(rsq00,rcutoff2))
 997             {
 998
 999             /* Compute parameters for interactions between i and j atoms */
1000             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1001                                          vdwparam+vdwioffset0+vdwjidx0B,
1002                                          vdwparam+vdwioffset0+vdwjidx0C,
1003                                          vdwparam+vdwioffset0+vdwjidx0D,
1004                                          &c6_00,&c12_00);
1005
1006             /* LENNARD-JONES DISPERSION/REPULSION */
1007
1008             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1009             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1010
1011             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1012
1013             fscal            = fvdw;
1014
1015             fscal            = _mm_and_ps(fscal,cutoff_mask);
1016
1017             /* Calculate temporary vectorial force */
1018             tx               = _mm_mul_ps(fscal,dx00);
1019             ty               = _mm_mul_ps(fscal,dy00);
1020             tz               = _mm_mul_ps(fscal,dz00);
1021
1022             /* Update vectorial force */
1023             fix0             = _mm_add_ps(fix0,tx);
1024             fiy0             = _mm_add_ps(fiy0,ty);
1025             fiz0             = _mm_add_ps(fiz0,tz);
1026
1027             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1028                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1029                                                    tx,ty,tz);
1030
1031             }
1032
1033             /**************************
1034              * CALCULATE INTERACTIONS *
1035              **************************/
1036
1037             if (gmx_mm_any_lt(rsq10,rcutoff2))
1038             {
1039
1040             r10              = _mm_mul_ps(rsq10,rinv10);
1041
1042             /* Compute parameters for interactions between i and j atoms */
1043             qq10             = _mm_mul_ps(iq1,jq0);
1044
1045             /* EWALD ELECTROSTATICS */
1046
1047             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1048             ewrt             = _mm_mul_ps(r10,ewtabscale);
1049             ewitab           = _mm_cvttps_epi32(ewrt);
1050             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1051             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1052                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1053                                          &ewtabF,&ewtabFn);
1054             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1055             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1056
1057             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1058
1059             fscal            = felec;
1060
1061             fscal            = _mm_and_ps(fscal,cutoff_mask);
1062
1063             /* Calculate temporary vectorial force */
1064             tx               = _mm_mul_ps(fscal,dx10);
1065             ty               = _mm_mul_ps(fscal,dy10);
1066             tz               = _mm_mul_ps(fscal,dz10);
1067
1068             /* Update vectorial force */
1069             fix1             = _mm_add_ps(fix1,tx);
1070             fiy1             = _mm_add_ps(fiy1,ty);
1071             fiz1             = _mm_add_ps(fiz1,tz);
1072
1073             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1074                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1075                                                    tx,ty,tz);
1076
1077             }
1078
1079             /**************************
1080              * CALCULATE INTERACTIONS *
1081              **************************/
1082
1083             if (gmx_mm_any_lt(rsq20,rcutoff2))
1084             {
1085
1086             r20              = _mm_mul_ps(rsq20,rinv20);
1087
1088             /* Compute parameters for interactions between i and j atoms */
1089             qq20             = _mm_mul_ps(iq2,jq0);
1090
1091             /* EWALD ELECTROSTATICS */
1092
1093             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1094             ewrt             = _mm_mul_ps(r20,ewtabscale);
1095             ewitab           = _mm_cvttps_epi32(ewrt);
1096             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1097             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1098                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1099                                          &ewtabF,&ewtabFn);
1100             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1101             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1102
1103             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1104
1105             fscal            = felec;
1106
1107             fscal            = _mm_and_ps(fscal,cutoff_mask);
1108
1109             /* Calculate temporary vectorial force */
1110             tx               = _mm_mul_ps(fscal,dx20);
1111             ty               = _mm_mul_ps(fscal,dy20);
1112             tz               = _mm_mul_ps(fscal,dz20);
1113
1114             /* Update vectorial force */
1115             fix2             = _mm_add_ps(fix2,tx);
1116             fiy2             = _mm_add_ps(fiy2,ty);
1117             fiz2             = _mm_add_ps(fiz2,tz);
1118
1119             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1120                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1121                                                    tx,ty,tz);
1122
1123             }
1124
1125             /**************************
1126              * CALCULATE INTERACTIONS *
1127              **************************/
1128
1129             if (gmx_mm_any_lt(rsq30,rcutoff2))
1130             {
1131
1132             r30              = _mm_mul_ps(rsq30,rinv30);
1133
1134             /* Compute parameters for interactions between i and j atoms */
1135             qq30             = _mm_mul_ps(iq3,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(r30,ewtabscale);
1141             ewitab           = _mm_cvttps_epi32(ewrt);
1142             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1143             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1144                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1145                                          &ewtabF,&ewtabFn);
1146             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1147             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1148
1149             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1150
1151             fscal            = felec;
1152
1153             fscal            = _mm_and_ps(fscal,cutoff_mask);
1154
1155             /* Calculate temporary vectorial force */
1156             tx               = _mm_mul_ps(fscal,dx30);
1157             ty               = _mm_mul_ps(fscal,dy30);
1158             tz               = _mm_mul_ps(fscal,dz30);
1159
1160             /* Update vectorial force */
1161             fix3             = _mm_add_ps(fix3,tx);
1162             fiy3             = _mm_add_ps(fiy3,ty);
1163             fiz3             = _mm_add_ps(fiz3,tz);
1164
1165             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1166                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1167                                                    tx,ty,tz);
1168
1169             }
1170
1171             /* Inner loop uses 147 flops */
1172         }
1173
1174         if(jidx<j_index_end)
1175         {
1176
1177             /* Get j neighbor index, and coordinate index */
1178             jnrA             = jjnr[jidx];
1179             jnrB             = jjnr[jidx+1];
1180             jnrC             = jjnr[jidx+2];
1181             jnrD             = jjnr[jidx+3];
1182
1183             /* Sign of each element will be negative for non-real atoms.
1184              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1185              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1186              */
1187             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1188             jnrA       = (jnrA>=0) ? jnrA : 0;
1189             jnrB       = (jnrB>=0) ? jnrB : 0;
1190             jnrC       = (jnrC>=0) ? jnrC : 0;
1191             jnrD       = (jnrD>=0) ? jnrD : 0;
1192
1193             j_coord_offsetA  = DIM*jnrA;
1194             j_coord_offsetB  = DIM*jnrB;
1195             j_coord_offsetC  = DIM*jnrC;
1196             j_coord_offsetD  = DIM*jnrD;
1197
1198             /* load j atom coordinates */
1199             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1200                                               x+j_coord_offsetC,x+j_coord_offsetD,
1201                                               &jx0,&jy0,&jz0);
1202
1203             /* Calculate displacement vector */
1204             dx00             = _mm_sub_ps(ix0,jx0);
1205             dy00             = _mm_sub_ps(iy0,jy0);
1206             dz00             = _mm_sub_ps(iz0,jz0);
1207             dx10             = _mm_sub_ps(ix1,jx0);
1208             dy10             = _mm_sub_ps(iy1,jy0);
1209             dz10             = _mm_sub_ps(iz1,jz0);
1210             dx20             = _mm_sub_ps(ix2,jx0);
1211             dy20             = _mm_sub_ps(iy2,jy0);
1212             dz20             = _mm_sub_ps(iz2,jz0);
1213             dx30             = _mm_sub_ps(ix3,jx0);
1214             dy30             = _mm_sub_ps(iy3,jy0);
1215             dz30             = _mm_sub_ps(iz3,jz0);
1216
1217             /* Calculate squared distance and things based on it */
1218             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1219             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1220             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1221             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1222
1223             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1224             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1225             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1226
1227             rinvsq00         = gmx_mm_inv_ps(rsq00);
1228             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1229             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1230             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1231
1232             /* Load parameters for j particles */
1233             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1234                                                               charge+jnrC+0,charge+jnrD+0);
1235             vdwjidx0A        = 2*vdwtype[jnrA+0];
1236             vdwjidx0B        = 2*vdwtype[jnrB+0];
1237             vdwjidx0C        = 2*vdwtype[jnrC+0];
1238             vdwjidx0D        = 2*vdwtype[jnrD+0];
1239
1240             /**************************
1241              * CALCULATE INTERACTIONS *
1242              **************************/
1243
1244             if (gmx_mm_any_lt(rsq00,rcutoff2))
1245             {
1246
1247             /* Compute parameters for interactions between i and j atoms */
1248             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1249                                          vdwparam+vdwioffset0+vdwjidx0B,
1250                                          vdwparam+vdwioffset0+vdwjidx0C,
1251                                          vdwparam+vdwioffset0+vdwjidx0D,
1252                                          &c6_00,&c12_00);
1253
1254             /* LENNARD-JONES DISPERSION/REPULSION */
1255
1256             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1257             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1258
1259             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1260
1261             fscal            = fvdw;
1262
1263             fscal            = _mm_and_ps(fscal,cutoff_mask);
1264
1265             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1266
1267             /* Calculate temporary vectorial force */
1268             tx               = _mm_mul_ps(fscal,dx00);
1269             ty               = _mm_mul_ps(fscal,dy00);
1270             tz               = _mm_mul_ps(fscal,dz00);
1271
1272             /* Update vectorial force */
1273             fix0             = _mm_add_ps(fix0,tx);
1274             fiy0             = _mm_add_ps(fiy0,ty);
1275             fiz0             = _mm_add_ps(fiz0,tz);
1276
1277             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1278                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1279                                                    tx,ty,tz);
1280
1281             }
1282
1283             /**************************
1284              * CALCULATE INTERACTIONS *
1285              **************************/
1286
1287             if (gmx_mm_any_lt(rsq10,rcutoff2))
1288             {
1289
1290             r10              = _mm_mul_ps(rsq10,rinv10);
1291             r10              = _mm_andnot_ps(dummy_mask,r10);
1292
1293             /* Compute parameters for interactions between i and j atoms */
1294             qq10             = _mm_mul_ps(iq1,jq0);
1295
1296             /* EWALD ELECTROSTATICS */
1297
1298             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1299             ewrt             = _mm_mul_ps(r10,ewtabscale);
1300             ewitab           = _mm_cvttps_epi32(ewrt);
1301             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1302             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1303                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1304                                          &ewtabF,&ewtabFn);
1305             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1306             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1307
1308             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1309
1310             fscal            = felec;
1311
1312             fscal            = _mm_and_ps(fscal,cutoff_mask);
1313
1314             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1315
1316             /* Calculate temporary vectorial force */
1317             tx               = _mm_mul_ps(fscal,dx10);
1318             ty               = _mm_mul_ps(fscal,dy10);
1319             tz               = _mm_mul_ps(fscal,dz10);
1320
1321             /* Update vectorial force */
1322             fix1             = _mm_add_ps(fix1,tx);
1323             fiy1             = _mm_add_ps(fiy1,ty);
1324             fiz1             = _mm_add_ps(fiz1,tz);
1325
1326             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1327                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1328                                                    tx,ty,tz);
1329
1330             }
1331
1332             /**************************
1333              * CALCULATE INTERACTIONS *
1334              **************************/
1335
1336             if (gmx_mm_any_lt(rsq20,rcutoff2))
1337             {
1338
1339             r20              = _mm_mul_ps(rsq20,rinv20);
1340             r20              = _mm_andnot_ps(dummy_mask,r20);
1341
1342             /* Compute parameters for interactions between i and j atoms */
1343             qq20             = _mm_mul_ps(iq2,jq0);
1344
1345             /* EWALD ELECTROSTATICS */
1346
1347             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1348             ewrt             = _mm_mul_ps(r20,ewtabscale);
1349             ewitab           = _mm_cvttps_epi32(ewrt);
1350             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1351             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1352                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1353                                          &ewtabF,&ewtabFn);
1354             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1355             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1356
1357             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1358
1359             fscal            = felec;
1360
1361             fscal            = _mm_and_ps(fscal,cutoff_mask);
1362
1363             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1364
1365             /* Calculate temporary vectorial force */
1366             tx               = _mm_mul_ps(fscal,dx20);
1367             ty               = _mm_mul_ps(fscal,dy20);
1368             tz               = _mm_mul_ps(fscal,dz20);
1369
1370             /* Update vectorial force */
1371             fix2             = _mm_add_ps(fix2,tx);
1372             fiy2             = _mm_add_ps(fiy2,ty);
1373             fiz2             = _mm_add_ps(fiz2,tz);
1374
1375             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1376                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1377                                                    tx,ty,tz);
1378
1379             }
1380
1381             /**************************
1382              * CALCULATE INTERACTIONS *
1383              **************************/
1384
1385             if (gmx_mm_any_lt(rsq30,rcutoff2))
1386             {
1387
1388             r30              = _mm_mul_ps(rsq30,rinv30);
1389             r30              = _mm_andnot_ps(dummy_mask,r30);
1390
1391             /* Compute parameters for interactions between i and j atoms */
1392             qq30             = _mm_mul_ps(iq3,jq0);
1393
1394             /* EWALD ELECTROSTATICS */
1395
1396             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1397             ewrt             = _mm_mul_ps(r30,ewtabscale);
1398             ewitab           = _mm_cvttps_epi32(ewrt);
1399             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1400             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1401                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1402                                          &ewtabF,&ewtabFn);
1403             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1404             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1405
1406             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1407
1408             fscal            = felec;
1409
1410             fscal            = _mm_and_ps(fscal,cutoff_mask);
1411
1412             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1413
1414             /* Calculate temporary vectorial force */
1415             tx               = _mm_mul_ps(fscal,dx30);
1416             ty               = _mm_mul_ps(fscal,dy30);
1417             tz               = _mm_mul_ps(fscal,dz30);
1418
1419             /* Update vectorial force */
1420             fix3             = _mm_add_ps(fix3,tx);
1421             fiy3             = _mm_add_ps(fiy3,ty);
1422             fiz3             = _mm_add_ps(fiz3,tz);
1423
1424             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1425                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1426                                                    tx,ty,tz);
1427
1428             }
1429
1430             /* Inner loop uses 150 flops */
1431         }
1432
1433         /* End of innermost loop */
1434
1435         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1436                                               f+i_coord_offset,fshift+i_shift_offset);
1437
1438         /* Increment number of inner iterations */
1439         inneriter                  += j_index_end - j_index_start;
1440
1441         /* Outer loop uses 36 flops */
1442     }
1443
1444     /* Increment number of outer iterations */
1445     outeriter        += nri;
1446
1447     /* Update outer/inner flops */
1448
1449     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*36 + inneriter*150);
1450 }