src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_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_ElecRFCut_VdwLJSh_GeomW3P1_VF_sse2_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_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              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     __m128           dummy_mask,cutoff_mask;
  89     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  90     __m128           one     = _mm_set1_ps(1.0);
  91     __m128           two     = _mm_set1_ps(2.0);
  92     x                = xx[0];
  93     f                = ff[0];
  94
  95     nri              = nlist->nri;
  96     iinr             = nlist->iinr;
  97     jindex           = nlist->jindex;
  98     jjnr             = nlist->jjnr;
  99     shiftidx         = nlist->shift;
 100     gid              = nlist->gid;
 101     shiftvec         = fr->shift_vec[0];
 102     fshift           = fr->fshift[0];
 103     facel            = _mm_set1_ps(fr->epsfac);
 104     charge           = mdatoms->chargeA;
 105     krf              = _mm_set1_ps(fr->ic->k_rf);
 106     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 107     crf              = _mm_set1_ps(fr->ic->c_rf);
 108     nvdwtype         = fr->ntype;
 109     vdwparam         = fr->nbfp;
 110     vdwtype          = mdatoms->typeA;
 111
 112     /* Setup water-specific parameters */
 113     inr              = nlist->iinr[0];
 114     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 115     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 116     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 117     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 118
 119     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 120     rcutoff_scalar   = fr->rcoulomb;
 121     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 122     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 123
 124     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 125     rvdw             = _mm_set1_ps(fr->rvdw);
 126
 127     /* Avoid stupid compiler warnings */
 128     jnrA = jnrB = jnrC = jnrD = 0;
 129     j_coord_offsetA = 0;
 130     j_coord_offsetB = 0;
 131     j_coord_offsetC = 0;
 132     j_coord_offsetD = 0;
 133
 134     outeriter        = 0;
 135     inneriter        = 0;
 136
 137     for(iidx=0;iidx<4*DIM;iidx++)
 138     {
 139         scratch[iidx] = 0.0;
 140     }
 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
 148         /* Load limits for loop over neighbors */
 149         j_index_start    = jindex[iidx];
 150         j_index_end      = jindex[iidx+1];
 151
 152         /* Get outer coordinate index */
 153         inr              = iinr[iidx];
 154         i_coord_offset   = DIM*inr;
 155
 156         /* Load i particle coords and add shift vector */
 157         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 158                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 159
 160         fix0             = _mm_setzero_ps();
 161         fiy0             = _mm_setzero_ps();
 162         fiz0             = _mm_setzero_ps();
 163         fix1             = _mm_setzero_ps();
 164         fiy1             = _mm_setzero_ps();
 165         fiz1             = _mm_setzero_ps();
 166         fix2             = _mm_setzero_ps();
 167         fiy2             = _mm_setzero_ps();
 168         fiz2             = _mm_setzero_ps();
 169
 170         /* Reset potential sums */
 171         velecsum         = _mm_setzero_ps();
 172         vvdwsum          = _mm_setzero_ps();
 173
 174         /* Start inner kernel loop */
 175         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 176         {
 177
 178             /* Get j neighbor index, and coordinate index */
 179             jnrA             = jjnr[jidx];
 180             jnrB             = jjnr[jidx+1];
 181             jnrC             = jjnr[jidx+2];
 182             jnrD             = jjnr[jidx+3];
 183             j_coord_offsetA  = DIM*jnrA;
 184             j_coord_offsetB  = DIM*jnrB;
 185             j_coord_offsetC  = DIM*jnrC;
 186             j_coord_offsetD  = DIM*jnrD;
 187
 188             /* load j atom coordinates */
 189             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 190                                               x+j_coord_offsetC,x+j_coord_offsetD,
 191                                               &jx0,&jy0,&jz0);
 192
 193             /* Calculate displacement vector */
 194             dx00             = _mm_sub_ps(ix0,jx0);
 195             dy00             = _mm_sub_ps(iy0,jy0);
 196             dz00             = _mm_sub_ps(iz0,jz0);
 197             dx10             = _mm_sub_ps(ix1,jx0);
 198             dy10             = _mm_sub_ps(iy1,jy0);
 199             dz10             = _mm_sub_ps(iz1,jz0);
 200             dx20             = _mm_sub_ps(ix2,jx0);
 201             dy20             = _mm_sub_ps(iy2,jy0);
 202             dz20             = _mm_sub_ps(iz2,jz0);
 203
 204             /* Calculate squared distance and things based on it */
 205             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 206             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 207             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 208
 209             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 210             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 211             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 212
 213             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 214             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 215             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 216
 217             /* Load parameters for j particles */
 218             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 219                                                               charge+jnrC+0,charge+jnrD+0);
 220             vdwjidx0A        = 2*vdwtype[jnrA+0];
 221             vdwjidx0B        = 2*vdwtype[jnrB+0];
 222             vdwjidx0C        = 2*vdwtype[jnrC+0];
 223             vdwjidx0D        = 2*vdwtype[jnrD+0];
 224
 225             /**************************
 226              * CALCULATE INTERACTIONS *
 227              **************************/
 228
 229             if (gmx_mm_any_lt(rsq00,rcutoff2))
 230             {
 231
 232             /* Compute parameters for interactions between i and j atoms */
 233             qq00             = _mm_mul_ps(iq0,jq0);
 234             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 235                                          vdwparam+vdwioffset0+vdwjidx0B,
 236                                          vdwparam+vdwioffset0+vdwjidx0C,
 237                                          vdwparam+vdwioffset0+vdwjidx0D,
 238                                          &c6_00,&c12_00);
 239
 240             /* REACTION-FIELD ELECTROSTATICS */
 241             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 242             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 243
 244             /* LENNARD-JONES DISPERSION/REPULSION */
 245
 246             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 247             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 248             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 249             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) ,
 250                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 251             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 252
 253             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 254
 255             /* Update potential sum for this i atom from the interaction with this j atom. */
 256             velec            = _mm_and_ps(velec,cutoff_mask);
 257             velecsum         = _mm_add_ps(velecsum,velec);
 258             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 259             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 260
 261             fscal            = _mm_add_ps(felec,fvdw);
 262
 263             fscal            = _mm_and_ps(fscal,cutoff_mask);
 264
 265             /* Calculate temporary vectorial force */
 266             tx               = _mm_mul_ps(fscal,dx00);
 267             ty               = _mm_mul_ps(fscal,dy00);
 268             tz               = _mm_mul_ps(fscal,dz00);
 269
 270             /* Update vectorial force */
 271             fix0             = _mm_add_ps(fix0,tx);
 272             fiy0             = _mm_add_ps(fiy0,ty);
 273             fiz0             = _mm_add_ps(fiz0,tz);
 274
 275             fjptrA             = f+j_coord_offsetA;
 276             fjptrB             = f+j_coord_offsetB;
 277             fjptrC             = f+j_coord_offsetC;
 278             fjptrD             = f+j_coord_offsetD;
 279             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 280
 281             }
 282
 283             /**************************
 284              * CALCULATE INTERACTIONS *
 285              **************************/
 286
 287             if (gmx_mm_any_lt(rsq10,rcutoff2))
 288             {
 289
 290             /* Compute parameters for interactions between i and j atoms */
 291             qq10             = _mm_mul_ps(iq1,jq0);
 292
 293             /* REACTION-FIELD ELECTROSTATICS */
 294             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 295             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 296
 297             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 298
 299             /* Update potential sum for this i atom from the interaction with this j atom. */
 300             velec            = _mm_and_ps(velec,cutoff_mask);
 301             velecsum         = _mm_add_ps(velecsum,velec);
 302
 303             fscal            = felec;
 304
 305             fscal            = _mm_and_ps(fscal,cutoff_mask);
 306
 307             /* Calculate temporary vectorial force */
 308             tx               = _mm_mul_ps(fscal,dx10);
 309             ty               = _mm_mul_ps(fscal,dy10);
 310             tz               = _mm_mul_ps(fscal,dz10);
 311
 312             /* Update vectorial force */
 313             fix1             = _mm_add_ps(fix1,tx);
 314             fiy1             = _mm_add_ps(fiy1,ty);
 315             fiz1             = _mm_add_ps(fiz1,tz);
 316
 317             fjptrA             = f+j_coord_offsetA;
 318             fjptrB             = f+j_coord_offsetB;
 319             fjptrC             = f+j_coord_offsetC;
 320             fjptrD             = f+j_coord_offsetD;
 321             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 322
 323             }
 324
 325             /**************************
 326              * CALCULATE INTERACTIONS *
 327              **************************/
 328
 329             if (gmx_mm_any_lt(rsq20,rcutoff2))
 330             {
 331
 332             /* Compute parameters for interactions between i and j atoms */
 333             qq20             = _mm_mul_ps(iq2,jq0);
 334
 335             /* REACTION-FIELD ELECTROSTATICS */
 336             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 337             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 338
 339             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 340
 341             /* Update potential sum for this i atom from the interaction with this j atom. */
 342             velec            = _mm_and_ps(velec,cutoff_mask);
 343             velecsum         = _mm_add_ps(velecsum,velec);
 344
 345             fscal            = felec;
 346
 347             fscal            = _mm_and_ps(fscal,cutoff_mask);
 348
 349             /* Calculate temporary vectorial force */
 350             tx               = _mm_mul_ps(fscal,dx20);
 351             ty               = _mm_mul_ps(fscal,dy20);
 352             tz               = _mm_mul_ps(fscal,dz20);
 353
 354             /* Update vectorial force */
 355             fix2             = _mm_add_ps(fix2,tx);
 356             fiy2             = _mm_add_ps(fiy2,ty);
 357             fiz2             = _mm_add_ps(fiz2,tz);
 358
 359             fjptrA             = f+j_coord_offsetA;
 360             fjptrB             = f+j_coord_offsetB;
 361             fjptrC             = f+j_coord_offsetC;
 362             fjptrD             = f+j_coord_offsetD;
 363             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 364
 365             }
 366
 367             /* Inner loop uses 126 flops */
 368         }
 369
 370         if(jidx<j_index_end)
 371         {
 372
 373             /* Get j neighbor index, and coordinate index */
 374             jnrlistA         = jjnr[jidx];
 375             jnrlistB         = jjnr[jidx+1];
 376             jnrlistC         = jjnr[jidx+2];
 377             jnrlistD         = jjnr[jidx+3];
 378             /* Sign of each element will be negative for non-real atoms.
 379              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 380              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 381              */
 382             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 383             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 384             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 385             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 386             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 387             j_coord_offsetA  = DIM*jnrA;
 388             j_coord_offsetB  = DIM*jnrB;
 389             j_coord_offsetC  = DIM*jnrC;
 390             j_coord_offsetD  = DIM*jnrD;
 391
 392             /* load j atom coordinates */
 393             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 394                                               x+j_coord_offsetC,x+j_coord_offsetD,
 395                                               &jx0,&jy0,&jz0);
 396
 397             /* Calculate displacement vector */
 398             dx00             = _mm_sub_ps(ix0,jx0);
 399             dy00             = _mm_sub_ps(iy0,jy0);
 400             dz00             = _mm_sub_ps(iz0,jz0);
 401             dx10             = _mm_sub_ps(ix1,jx0);
 402             dy10             = _mm_sub_ps(iy1,jy0);
 403             dz10             = _mm_sub_ps(iz1,jz0);
 404             dx20             = _mm_sub_ps(ix2,jx0);
 405             dy20             = _mm_sub_ps(iy2,jy0);
 406             dz20             = _mm_sub_ps(iz2,jz0);
 407
 408             /* Calculate squared distance and things based on it */
 409             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 410             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 411             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 412
 413             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 414             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 415             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 416
 417             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 418             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 419             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 420
 421             /* Load parameters for j particles */
 422             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 423                                                               charge+jnrC+0,charge+jnrD+0);
 424             vdwjidx0A        = 2*vdwtype[jnrA+0];
 425             vdwjidx0B        = 2*vdwtype[jnrB+0];
 426             vdwjidx0C        = 2*vdwtype[jnrC+0];
 427             vdwjidx0D        = 2*vdwtype[jnrD+0];
 428
 429             /**************************
 430              * CALCULATE INTERACTIONS *
 431              **************************/
 432
 433             if (gmx_mm_any_lt(rsq00,rcutoff2))
 434             {
 435
 436             /* Compute parameters for interactions between i and j atoms */
 437             qq00             = _mm_mul_ps(iq0,jq0);
 438             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 439                                          vdwparam+vdwioffset0+vdwjidx0B,
 440                                          vdwparam+vdwioffset0+vdwjidx0C,
 441                                          vdwparam+vdwioffset0+vdwjidx0D,
 442                                          &c6_00,&c12_00);
 443
 444             /* REACTION-FIELD ELECTROSTATICS */
 445             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 446             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 447
 448             /* LENNARD-JONES DISPERSION/REPULSION */
 449
 450             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 451             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 452             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 453             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) ,
 454                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 455             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 456
 457             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 458
 459             /* Update potential sum for this i atom from the interaction with this j atom. */
 460             velec            = _mm_and_ps(velec,cutoff_mask);
 461             velec            = _mm_andnot_ps(dummy_mask,velec);
 462             velecsum         = _mm_add_ps(velecsum,velec);
 463             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 464             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 465             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 466
 467             fscal            = _mm_add_ps(felec,fvdw);
 468
 469             fscal            = _mm_and_ps(fscal,cutoff_mask);
 470
 471             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 472
 473             /* Calculate temporary vectorial force */
 474             tx               = _mm_mul_ps(fscal,dx00);
 475             ty               = _mm_mul_ps(fscal,dy00);
 476             tz               = _mm_mul_ps(fscal,dz00);
 477
 478             /* Update vectorial force */
 479             fix0             = _mm_add_ps(fix0,tx);
 480             fiy0             = _mm_add_ps(fiy0,ty);
 481             fiz0             = _mm_add_ps(fiz0,tz);
 482
 483             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 484             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 485             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 486             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 487             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 488
 489             }
 490
 491             /**************************
 492              * CALCULATE INTERACTIONS *
 493              **************************/
 494
 495             if (gmx_mm_any_lt(rsq10,rcutoff2))
 496             {
 497
 498             /* Compute parameters for interactions between i and j atoms */
 499             qq10             = _mm_mul_ps(iq1,jq0);
 500
 501             /* REACTION-FIELD ELECTROSTATICS */
 502             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 503             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 504
 505             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 506
 507             /* Update potential sum for this i atom from the interaction with this j atom. */
 508             velec            = _mm_and_ps(velec,cutoff_mask);
 509             velec            = _mm_andnot_ps(dummy_mask,velec);
 510             velecsum         = _mm_add_ps(velecsum,velec);
 511
 512             fscal            = felec;
 513
 514             fscal            = _mm_and_ps(fscal,cutoff_mask);
 515
 516             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 517
 518             /* Calculate temporary vectorial force */
 519             tx               = _mm_mul_ps(fscal,dx10);
 520             ty               = _mm_mul_ps(fscal,dy10);
 521             tz               = _mm_mul_ps(fscal,dz10);
 522
 523             /* Update vectorial force */
 524             fix1             = _mm_add_ps(fix1,tx);
 525             fiy1             = _mm_add_ps(fiy1,ty);
 526             fiz1             = _mm_add_ps(fiz1,tz);
 527
 528             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 529             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 530             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 531             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 532             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 533
 534             }
 535
 536             /**************************
 537              * CALCULATE INTERACTIONS *
 538              **************************/
 539
 540             if (gmx_mm_any_lt(rsq20,rcutoff2))
 541             {
 542
 543             /* Compute parameters for interactions between i and j atoms */
 544             qq20             = _mm_mul_ps(iq2,jq0);
 545
 546             /* REACTION-FIELD ELECTROSTATICS */
 547             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 548             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 549
 550             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 551
 552             /* Update potential sum for this i atom from the interaction with this j atom. */
 553             velec            = _mm_and_ps(velec,cutoff_mask);
 554             velec            = _mm_andnot_ps(dummy_mask,velec);
 555             velecsum         = _mm_add_ps(velecsum,velec);
 556
 557             fscal            = felec;
 558
 559             fscal            = _mm_and_ps(fscal,cutoff_mask);
 560
 561             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 562
 563             /* Calculate temporary vectorial force */
 564             tx               = _mm_mul_ps(fscal,dx20);
 565             ty               = _mm_mul_ps(fscal,dy20);
 566             tz               = _mm_mul_ps(fscal,dz20);
 567
 568             /* Update vectorial force */
 569             fix2             = _mm_add_ps(fix2,tx);
 570             fiy2             = _mm_add_ps(fiy2,ty);
 571             fiz2             = _mm_add_ps(fiz2,tz);
 572
 573             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 574             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 575             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 576             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 577             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 578
 579             }
 580
 581             /* Inner loop uses 126 flops */
 582         }
 583
 584         /* End of innermost loop */
 585
 586         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 587                                               f+i_coord_offset,fshift+i_shift_offset);
 588
 589         ggid                        = gid[iidx];
 590         /* Update potential energies */
 591         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 592         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 593
 594         /* Increment number of inner iterations */
 595         inneriter                  += j_index_end - j_index_start;
 596
 597         /* Outer loop uses 20 flops */
 598     }
 599
 600     /* Increment number of outer iterations */
 601     outeriter        += nri;
 602
 603     /* Update outer/inner flops */
 604
 605     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*126);
 606 }
 607 /*
 608  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_sse2_single
 609  * Electrostatics interaction: ReactionField
 610  * VdW interaction:            LennardJones
 611  * Geometry:                   Water3-Particle
 612  * Calculate force/pot:        Force
 613  */
 614 void
 615 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_sse2_single
 616                     (t_nblist * gmx_restrict                nlist,
 617                      rvec * gmx_restrict                    xx,
 618                      rvec * gmx_restrict                    ff,
 619                      t_forcerec * gmx_restrict              fr,
 620                      t_mdatoms * gmx_restrict               mdatoms,
 621                      nb_kernel_data_t * gmx_restrict        kernel_data,
 622                      t_nrnb * gmx_restrict                  nrnb)
 623 {
 624     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 625      * just 0 for non-waters.
 626      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 627      * jnr indices corresponding to data put in the four positions in the SIMD register.
 628      */
 629     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 630     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 631     int              jnrA,jnrB,jnrC,jnrD;
 632     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 633     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 634     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 635     real             rcutoff_scalar;
 636     real             *shiftvec,*fshift,*x,*f;
 637     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 638     real             scratch[4*DIM];
 639     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 640     int              vdwioffset0;
 641     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 642     int              vdwioffset1;
 643     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 644     int              vdwioffset2;
 645     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 646     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 647     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 648     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 649     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 650     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 651     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 652     real             *charge;
 653     int              nvdwtype;
 654     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 655     int              *vdwtype;
 656     real             *vdwparam;
 657     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 658     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 659     __m128           dummy_mask,cutoff_mask;
 660     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 661     __m128           one     = _mm_set1_ps(1.0);
 662     __m128           two     = _mm_set1_ps(2.0);
 663     x                = xx[0];
 664     f                = ff[0];
 665
 666     nri              = nlist->nri;
 667     iinr             = nlist->iinr;
 668     jindex           = nlist->jindex;
 669     jjnr             = nlist->jjnr;
 670     shiftidx         = nlist->shift;
 671     gid              = nlist->gid;
 672     shiftvec         = fr->shift_vec[0];
 673     fshift           = fr->fshift[0];
 674     facel            = _mm_set1_ps(fr->epsfac);
 675     charge           = mdatoms->chargeA;
 676     krf              = _mm_set1_ps(fr->ic->k_rf);
 677     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 678     crf              = _mm_set1_ps(fr->ic->c_rf);
 679     nvdwtype         = fr->ntype;
 680     vdwparam         = fr->nbfp;
 681     vdwtype          = mdatoms->typeA;
 682
 683     /* Setup water-specific parameters */
 684     inr              = nlist->iinr[0];
 685     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 686     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 687     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 688     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 689
 690     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 691     rcutoff_scalar   = fr->rcoulomb;
 692     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 693     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 694
 695     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 696     rvdw             = _mm_set1_ps(fr->rvdw);
 697
 698     /* Avoid stupid compiler warnings */
 699     jnrA = jnrB = jnrC = jnrD = 0;
 700     j_coord_offsetA = 0;
 701     j_coord_offsetB = 0;
 702     j_coord_offsetC = 0;
 703     j_coord_offsetD = 0;
 704
 705     outeriter        = 0;
 706     inneriter        = 0;
 707
 708     for(iidx=0;iidx<4*DIM;iidx++)
 709     {
 710         scratch[iidx] = 0.0;
 711     }
 712
 713     /* Start outer loop over neighborlists */
 714     for(iidx=0; iidx<nri; iidx++)
 715     {
 716         /* Load shift vector for this list */
 717         i_shift_offset   = DIM*shiftidx[iidx];
 718
 719         /* Load limits for loop over neighbors */
 720         j_index_start    = jindex[iidx];
 721         j_index_end      = jindex[iidx+1];
 722
 723         /* Get outer coordinate index */
 724         inr              = iinr[iidx];
 725         i_coord_offset   = DIM*inr;
 726
 727         /* Load i particle coords and add shift vector */
 728         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 729                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 730
 731         fix0             = _mm_setzero_ps();
 732         fiy0             = _mm_setzero_ps();
 733         fiz0             = _mm_setzero_ps();
 734         fix1             = _mm_setzero_ps();
 735         fiy1             = _mm_setzero_ps();
 736         fiz1             = _mm_setzero_ps();
 737         fix2             = _mm_setzero_ps();
 738         fiy2             = _mm_setzero_ps();
 739         fiz2             = _mm_setzero_ps();
 740
 741         /* Start inner kernel loop */
 742         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 743         {
 744
 745             /* Get j neighbor index, and coordinate index */
 746             jnrA             = jjnr[jidx];
 747             jnrB             = jjnr[jidx+1];
 748             jnrC             = jjnr[jidx+2];
 749             jnrD             = jjnr[jidx+3];
 750             j_coord_offsetA  = DIM*jnrA;
 751             j_coord_offsetB  = DIM*jnrB;
 752             j_coord_offsetC  = DIM*jnrC;
 753             j_coord_offsetD  = DIM*jnrD;
 754
 755             /* load j atom coordinates */
 756             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 757                                               x+j_coord_offsetC,x+j_coord_offsetD,
 758                                               &jx0,&jy0,&jz0);
 759
 760             /* Calculate displacement vector */
 761             dx00             = _mm_sub_ps(ix0,jx0);
 762             dy00             = _mm_sub_ps(iy0,jy0);
 763             dz00             = _mm_sub_ps(iz0,jz0);
 764             dx10             = _mm_sub_ps(ix1,jx0);
 765             dy10             = _mm_sub_ps(iy1,jy0);
 766             dz10             = _mm_sub_ps(iz1,jz0);
 767             dx20             = _mm_sub_ps(ix2,jx0);
 768             dy20             = _mm_sub_ps(iy2,jy0);
 769             dz20             = _mm_sub_ps(iz2,jz0);
 770
 771             /* Calculate squared distance and things based on it */
 772             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 773             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 774             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 775
 776             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 777             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 778             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 779
 780             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 781             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 782             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 783
 784             /* Load parameters for j particles */
 785             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 786                                                               charge+jnrC+0,charge+jnrD+0);
 787             vdwjidx0A        = 2*vdwtype[jnrA+0];
 788             vdwjidx0B        = 2*vdwtype[jnrB+0];
 789             vdwjidx0C        = 2*vdwtype[jnrC+0];
 790             vdwjidx0D        = 2*vdwtype[jnrD+0];
 791
 792             /**************************
 793              * CALCULATE INTERACTIONS *
 794              **************************/
 795
 796             if (gmx_mm_any_lt(rsq00,rcutoff2))
 797             {
 798
 799             /* Compute parameters for interactions between i and j atoms */
 800             qq00             = _mm_mul_ps(iq0,jq0);
 801             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 802                                          vdwparam+vdwioffset0+vdwjidx0B,
 803                                          vdwparam+vdwioffset0+vdwjidx0C,
 804                                          vdwparam+vdwioffset0+vdwjidx0D,
 805                                          &c6_00,&c12_00);
 806
 807             /* REACTION-FIELD ELECTROSTATICS */
 808             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 809
 810             /* LENNARD-JONES DISPERSION/REPULSION */
 811
 812             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 813             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 814
 815             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 816
 817             fscal            = _mm_add_ps(felec,fvdw);
 818
 819             fscal            = _mm_and_ps(fscal,cutoff_mask);
 820
 821             /* Calculate temporary vectorial force */
 822             tx               = _mm_mul_ps(fscal,dx00);
 823             ty               = _mm_mul_ps(fscal,dy00);
 824             tz               = _mm_mul_ps(fscal,dz00);
 825
 826             /* Update vectorial force */
 827             fix0             = _mm_add_ps(fix0,tx);
 828             fiy0             = _mm_add_ps(fiy0,ty);
 829             fiz0             = _mm_add_ps(fiz0,tz);
 830
 831             fjptrA             = f+j_coord_offsetA;
 832             fjptrB             = f+j_coord_offsetB;
 833             fjptrC             = f+j_coord_offsetC;
 834             fjptrD             = f+j_coord_offsetD;
 835             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 836
 837             }
 838
 839             /**************************
 840              * CALCULATE INTERACTIONS *
 841              **************************/
 842
 843             if (gmx_mm_any_lt(rsq10,rcutoff2))
 844             {
 845
 846             /* Compute parameters for interactions between i and j atoms */
 847             qq10             = _mm_mul_ps(iq1,jq0);
 848
 849             /* REACTION-FIELD ELECTROSTATICS */
 850             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 851
 852             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 853
 854             fscal            = felec;
 855
 856             fscal            = _mm_and_ps(fscal,cutoff_mask);
 857
 858             /* Calculate temporary vectorial force */
 859             tx               = _mm_mul_ps(fscal,dx10);
 860             ty               = _mm_mul_ps(fscal,dy10);
 861             tz               = _mm_mul_ps(fscal,dz10);
 862
 863             /* Update vectorial force */
 864             fix1             = _mm_add_ps(fix1,tx);
 865             fiy1             = _mm_add_ps(fiy1,ty);
 866             fiz1             = _mm_add_ps(fiz1,tz);
 867
 868             fjptrA             = f+j_coord_offsetA;
 869             fjptrB             = f+j_coord_offsetB;
 870             fjptrC             = f+j_coord_offsetC;
 871             fjptrD             = f+j_coord_offsetD;
 872             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 873
 874             }
 875
 876             /**************************
 877              * CALCULATE INTERACTIONS *
 878              **************************/
 879
 880             if (gmx_mm_any_lt(rsq20,rcutoff2))
 881             {
 882
 883             /* Compute parameters for interactions between i and j atoms */
 884             qq20             = _mm_mul_ps(iq2,jq0);
 885
 886             /* REACTION-FIELD ELECTROSTATICS */
 887             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 888
 889             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 890
 891             fscal            = felec;
 892
 893             fscal            = _mm_and_ps(fscal,cutoff_mask);
 894
 895             /* Calculate temporary vectorial force */
 896             tx               = _mm_mul_ps(fscal,dx20);
 897             ty               = _mm_mul_ps(fscal,dy20);
 898             tz               = _mm_mul_ps(fscal,dz20);
 899
 900             /* Update vectorial force */
 901             fix2             = _mm_add_ps(fix2,tx);
 902             fiy2             = _mm_add_ps(fiy2,ty);
 903             fiz2             = _mm_add_ps(fiz2,tz);
 904
 905             fjptrA             = f+j_coord_offsetA;
 906             fjptrB             = f+j_coord_offsetB;
 907             fjptrC             = f+j_coord_offsetC;
 908             fjptrD             = f+j_coord_offsetD;
 909             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 910
 911             }
 912
 913             /* Inner loop uses 97 flops */
 914         }
 915
 916         if(jidx<j_index_end)
 917         {
 918
 919             /* Get j neighbor index, and coordinate index */
 920             jnrlistA         = jjnr[jidx];
 921             jnrlistB         = jjnr[jidx+1];
 922             jnrlistC         = jjnr[jidx+2];
 923             jnrlistD         = jjnr[jidx+3];
 924             /* Sign of each element will be negative for non-real atoms.
 925              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 926              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 927              */
 928             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 929             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 930             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 931             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 932             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 933             j_coord_offsetA  = DIM*jnrA;
 934             j_coord_offsetB  = DIM*jnrB;
 935             j_coord_offsetC  = DIM*jnrC;
 936             j_coord_offsetD  = DIM*jnrD;
 937
 938             /* load j atom coordinates */
 939             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 940                                               x+j_coord_offsetC,x+j_coord_offsetD,
 941                                               &jx0,&jy0,&jz0);
 942
 943             /* Calculate displacement vector */
 944             dx00             = _mm_sub_ps(ix0,jx0);
 945             dy00             = _mm_sub_ps(iy0,jy0);
 946             dz00             = _mm_sub_ps(iz0,jz0);
 947             dx10             = _mm_sub_ps(ix1,jx0);
 948             dy10             = _mm_sub_ps(iy1,jy0);
 949             dz10             = _mm_sub_ps(iz1,jz0);
 950             dx20             = _mm_sub_ps(ix2,jx0);
 951             dy20             = _mm_sub_ps(iy2,jy0);
 952             dz20             = _mm_sub_ps(iz2,jz0);
 953
 954             /* Calculate squared distance and things based on it */
 955             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 956             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 957             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 958
 959             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 960             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 961             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 962
 963             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 964             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 965             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 966
 967             /* Load parameters for j particles */
 968             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 969                                                               charge+jnrC+0,charge+jnrD+0);
 970             vdwjidx0A        = 2*vdwtype[jnrA+0];
 971             vdwjidx0B        = 2*vdwtype[jnrB+0];
 972             vdwjidx0C        = 2*vdwtype[jnrC+0];
 973             vdwjidx0D        = 2*vdwtype[jnrD+0];
 974
 975             /**************************
 976              * CALCULATE INTERACTIONS *
 977              **************************/
 978
 979             if (gmx_mm_any_lt(rsq00,rcutoff2))
 980             {
 981
 982             /* Compute parameters for interactions between i and j atoms */
 983             qq00             = _mm_mul_ps(iq0,jq0);
 984             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 985                                          vdwparam+vdwioffset0+vdwjidx0B,
 986                                          vdwparam+vdwioffset0+vdwjidx0C,
 987                                          vdwparam+vdwioffset0+vdwjidx0D,
 988                                          &c6_00,&c12_00);
 989
 990             /* REACTION-FIELD ELECTROSTATICS */
 991             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 992
 993             /* LENNARD-JONES DISPERSION/REPULSION */
 994
 995             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 996             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 997
 998             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 999
1000             fscal            = _mm_add_ps(felec,fvdw);
1001
1002             fscal            = _mm_and_ps(fscal,cutoff_mask);
1003
1004             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1005
1006             /* Calculate temporary vectorial force */
1007             tx               = _mm_mul_ps(fscal,dx00);
1008             ty               = _mm_mul_ps(fscal,dy00);
1009             tz               = _mm_mul_ps(fscal,dz00);
1010
1011             /* Update vectorial force */
1012             fix0             = _mm_add_ps(fix0,tx);
1013             fiy0             = _mm_add_ps(fiy0,ty);
1014             fiz0             = _mm_add_ps(fiz0,tz);
1015
1016             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1017             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1018             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1019             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1020             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1021
1022             }
1023
1024             /**************************
1025              * CALCULATE INTERACTIONS *
1026              **************************/
1027
1028             if (gmx_mm_any_lt(rsq10,rcutoff2))
1029             {
1030
1031             /* Compute parameters for interactions between i and j atoms */
1032             qq10             = _mm_mul_ps(iq1,jq0);
1033
1034             /* REACTION-FIELD ELECTROSTATICS */
1035             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1036
1037             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1038
1039             fscal            = felec;
1040
1041             fscal            = _mm_and_ps(fscal,cutoff_mask);
1042
1043             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1044
1045             /* Calculate temporary vectorial force */
1046             tx               = _mm_mul_ps(fscal,dx10);
1047             ty               = _mm_mul_ps(fscal,dy10);
1048             tz               = _mm_mul_ps(fscal,dz10);
1049
1050             /* Update vectorial force */
1051             fix1             = _mm_add_ps(fix1,tx);
1052             fiy1             = _mm_add_ps(fiy1,ty);
1053             fiz1             = _mm_add_ps(fiz1,tz);
1054
1055             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1056             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1057             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1058             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1059             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1060
1061             }
1062
1063             /**************************
1064              * CALCULATE INTERACTIONS *
1065              **************************/
1066
1067             if (gmx_mm_any_lt(rsq20,rcutoff2))
1068             {
1069
1070             /* Compute parameters for interactions between i and j atoms */
1071             qq20             = _mm_mul_ps(iq2,jq0);
1072
1073             /* REACTION-FIELD ELECTROSTATICS */
1074             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1075
1076             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1077
1078             fscal            = felec;
1079
1080             fscal            = _mm_and_ps(fscal,cutoff_mask);
1081
1082             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1083
1084             /* Calculate temporary vectorial force */
1085             tx               = _mm_mul_ps(fscal,dx20);
1086             ty               = _mm_mul_ps(fscal,dy20);
1087             tz               = _mm_mul_ps(fscal,dz20);
1088
1089             /* Update vectorial force */
1090             fix2             = _mm_add_ps(fix2,tx);
1091             fiy2             = _mm_add_ps(fiy2,ty);
1092             fiz2             = _mm_add_ps(fiz2,tz);
1093
1094             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1095             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1096             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1097             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1098             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1099
1100             }
1101
1102             /* Inner loop uses 97 flops */
1103         }
1104
1105         /* End of innermost loop */
1106
1107         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1108                                               f+i_coord_offset,fshift+i_shift_offset);
1109
1110         /* Increment number of inner iterations */
1111         inneriter                  += j_index_end - j_index_start;
1112
1113         /* Outer loop uses 18 flops */
1114     }
1115
1116     /* Increment number of outer iterations */
1117     outeriter        += nri;
1118
1119     /* Update outer/inner flops */
1120
1121     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*97);
1122 }