src/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_sse4_1_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse4_1_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_sse4_1_single.h"
  34 #include "kernelutil_x86_sse4_1_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_VF_sse4_1_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_VF_sse4_1_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwioffset1;
  72     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  73     int              vdwioffset2;
  74     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  75     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  76     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  77     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  78     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  79     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  80     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     int              nvdwtype;
  83     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  84     int              *vdwtype;
  85     real             *vdwparam;
  86     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  87     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  88     __m128i          ewitab;
  89     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  90     real             *ewtab;
  91     __m128           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     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 120     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 121     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 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     for(iidx=0;iidx<4*DIM;iidx++)
 143     {
 144         scratch[iidx] = 0.0;
 145     }
 146
 147     /* Start outer loop over neighborlists */
 148     for(iidx=0; iidx<nri; iidx++)
 149     {
 150         /* Load shift vector for this list */
 151         i_shift_offset   = DIM*shiftidx[iidx];
 152
 153         /* Load limits for loop over neighbors */
 154         j_index_start    = jindex[iidx];
 155         j_index_end      = jindex[iidx+1];
 156
 157         /* Get outer coordinate index */
 158         inr              = iinr[iidx];
 159         i_coord_offset   = DIM*inr;
 160
 161         /* Load i particle coords and add shift vector */
 162         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 163                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 164
 165         fix0             = _mm_setzero_ps();
 166         fiy0             = _mm_setzero_ps();
 167         fiz0             = _mm_setzero_ps();
 168         fix1             = _mm_setzero_ps();
 169         fiy1             = _mm_setzero_ps();
 170         fiz1             = _mm_setzero_ps();
 171         fix2             = _mm_setzero_ps();
 172         fiy2             = _mm_setzero_ps();
 173         fiz2             = _mm_setzero_ps();
 174
 175         /* Reset potential sums */
 176         velecsum         = _mm_setzero_ps();
 177         vvdwsum          = _mm_setzero_ps();
 178
 179         /* Start inner kernel loop */
 180         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 181         {
 182
 183             /* Get j neighbor index, and coordinate index */
 184             jnrA             = jjnr[jidx];
 185             jnrB             = jjnr[jidx+1];
 186             jnrC             = jjnr[jidx+2];
 187             jnrD             = jjnr[jidx+3];
 188             j_coord_offsetA  = DIM*jnrA;
 189             j_coord_offsetB  = DIM*jnrB;
 190             j_coord_offsetC  = DIM*jnrC;
 191             j_coord_offsetD  = DIM*jnrD;
 192
 193             /* load j atom coordinates */
 194             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 195                                               x+j_coord_offsetC,x+j_coord_offsetD,
 196                                               &jx0,&jy0,&jz0);
 197
 198             /* Calculate displacement vector */
 199             dx00             = _mm_sub_ps(ix0,jx0);
 200             dy00             = _mm_sub_ps(iy0,jy0);
 201             dz00             = _mm_sub_ps(iz0,jz0);
 202             dx10             = _mm_sub_ps(ix1,jx0);
 203             dy10             = _mm_sub_ps(iy1,jy0);
 204             dz10             = _mm_sub_ps(iz1,jz0);
 205             dx20             = _mm_sub_ps(ix2,jx0);
 206             dy20             = _mm_sub_ps(iy2,jy0);
 207             dz20             = _mm_sub_ps(iz2,jz0);
 208
 209             /* Calculate squared distance and things based on it */
 210             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 211             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 212             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 213
 214             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 215             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 216             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 217
 218             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 219             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 220             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 221
 222             /* Load parameters for j particles */
 223             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 224                                                               charge+jnrC+0,charge+jnrD+0);
 225             vdwjidx0A        = 2*vdwtype[jnrA+0];
 226             vdwjidx0B        = 2*vdwtype[jnrB+0];
 227             vdwjidx0C        = 2*vdwtype[jnrC+0];
 228             vdwjidx0D        = 2*vdwtype[jnrD+0];
 229
 230             fjx0             = _mm_setzero_ps();
 231             fjy0             = _mm_setzero_ps();
 232             fjz0             = _mm_setzero_ps();
 233
 234             /**************************
 235              * CALCULATE INTERACTIONS *
 236              **************************/
 237
 238             if (gmx_mm_any_lt(rsq00,rcutoff2))
 239             {
 240
 241             r00              = _mm_mul_ps(rsq00,rinv00);
 242
 243             /* Compute parameters for interactions between i and j atoms */
 244             qq00             = _mm_mul_ps(iq0,jq0);
 245             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 246                                          vdwparam+vdwioffset0+vdwjidx0B,
 247                                          vdwparam+vdwioffset0+vdwjidx0C,
 248                                          vdwparam+vdwioffset0+vdwjidx0D,
 249                                          &c6_00,&c12_00);
 250
 251             /* EWALD ELECTROSTATICS */
 252
 253             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 254             ewrt             = _mm_mul_ps(r00,ewtabscale);
 255             ewitab           = _mm_cvttps_epi32(ewrt);
 256             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 257             ewitab           = _mm_slli_epi32(ewitab,2);
 258             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 259             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 260             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 261             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 262             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 263             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 264             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 265             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 266             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 267
 268             /* LENNARD-JONES DISPERSION/REPULSION */
 269
 270             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 271             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 272             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 273             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) ,
 274                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 275             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 276
 277             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 278
 279             /* Update potential sum for this i atom from the interaction with this j atom. */
 280             velec            = _mm_and_ps(velec,cutoff_mask);
 281             velecsum         = _mm_add_ps(velecsum,velec);
 282             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 283             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 284
 285             fscal            = _mm_add_ps(felec,fvdw);
 286
 287             fscal            = _mm_and_ps(fscal,cutoff_mask);
 288
 289             /* Calculate temporary vectorial force */
 290             tx               = _mm_mul_ps(fscal,dx00);
 291             ty               = _mm_mul_ps(fscal,dy00);
 292             tz               = _mm_mul_ps(fscal,dz00);
 293
 294             /* Update vectorial force */
 295             fix0             = _mm_add_ps(fix0,tx);
 296             fiy0             = _mm_add_ps(fiy0,ty);
 297             fiz0             = _mm_add_ps(fiz0,tz);
 298
 299             fjx0             = _mm_add_ps(fjx0,tx);
 300             fjy0             = _mm_add_ps(fjy0,ty);
 301             fjz0             = _mm_add_ps(fjz0,tz);
 302
 303             }
 304
 305             /**************************
 306              * CALCULATE INTERACTIONS *
 307              **************************/
 308
 309             if (gmx_mm_any_lt(rsq10,rcutoff2))
 310             {
 311
 312             r10              = _mm_mul_ps(rsq10,rinv10);
 313
 314             /* Compute parameters for interactions between i and j atoms */
 315             qq10             = _mm_mul_ps(iq1,jq0);
 316
 317             /* EWALD ELECTROSTATICS */
 318
 319             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 320             ewrt             = _mm_mul_ps(r10,ewtabscale);
 321             ewitab           = _mm_cvttps_epi32(ewrt);
 322             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 323             ewitab           = _mm_slli_epi32(ewitab,2);
 324             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 325             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 326             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 327             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 328             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 329             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 330             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 331             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 332             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 333
 334             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 335
 336             /* Update potential sum for this i atom from the interaction with this j atom. */
 337             velec            = _mm_and_ps(velec,cutoff_mask);
 338             velecsum         = _mm_add_ps(velecsum,velec);
 339
 340             fscal            = felec;
 341
 342             fscal            = _mm_and_ps(fscal,cutoff_mask);
 343
 344             /* Calculate temporary vectorial force */
 345             tx               = _mm_mul_ps(fscal,dx10);
 346             ty               = _mm_mul_ps(fscal,dy10);
 347             tz               = _mm_mul_ps(fscal,dz10);
 348
 349             /* Update vectorial force */
 350             fix1             = _mm_add_ps(fix1,tx);
 351             fiy1             = _mm_add_ps(fiy1,ty);
 352             fiz1             = _mm_add_ps(fiz1,tz);
 353
 354             fjx0             = _mm_add_ps(fjx0,tx);
 355             fjy0             = _mm_add_ps(fjy0,ty);
 356             fjz0             = _mm_add_ps(fjz0,tz);
 357
 358             }
 359
 360             /**************************
 361              * CALCULATE INTERACTIONS *
 362              **************************/
 363
 364             if (gmx_mm_any_lt(rsq20,rcutoff2))
 365             {
 366
 367             r20              = _mm_mul_ps(rsq20,rinv20);
 368
 369             /* Compute parameters for interactions between i and j atoms */
 370             qq20             = _mm_mul_ps(iq2,jq0);
 371
 372             /* EWALD ELECTROSTATICS */
 373
 374             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 375             ewrt             = _mm_mul_ps(r20,ewtabscale);
 376             ewitab           = _mm_cvttps_epi32(ewrt);
 377             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 378             ewitab           = _mm_slli_epi32(ewitab,2);
 379             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 380             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 381             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 382             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 383             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 384             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 385             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 386             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 387             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 388
 389             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 390
 391             /* Update potential sum for this i atom from the interaction with this j atom. */
 392             velec            = _mm_and_ps(velec,cutoff_mask);
 393             velecsum         = _mm_add_ps(velecsum,velec);
 394
 395             fscal            = felec;
 396
 397             fscal            = _mm_and_ps(fscal,cutoff_mask);
 398
 399             /* Calculate temporary vectorial force */
 400             tx               = _mm_mul_ps(fscal,dx20);
 401             ty               = _mm_mul_ps(fscal,dy20);
 402             tz               = _mm_mul_ps(fscal,dz20);
 403
 404             /* Update vectorial force */
 405             fix2             = _mm_add_ps(fix2,tx);
 406             fiy2             = _mm_add_ps(fiy2,ty);
 407             fiz2             = _mm_add_ps(fiz2,tz);
 408
 409             fjx0             = _mm_add_ps(fjx0,tx);
 410             fjy0             = _mm_add_ps(fjy0,ty);
 411             fjz0             = _mm_add_ps(fjz0,tz);
 412
 413             }
 414
 415             fjptrA             = f+j_coord_offsetA;
 416             fjptrB             = f+j_coord_offsetB;
 417             fjptrC             = f+j_coord_offsetC;
 418             fjptrD             = f+j_coord_offsetD;
 419
 420             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 421
 422             /* Inner loop uses 156 flops */
 423         }
 424
 425         if(jidx<j_index_end)
 426         {
 427
 428             /* Get j neighbor index, and coordinate index */
 429             jnrlistA         = jjnr[jidx];
 430             jnrlistB         = jjnr[jidx+1];
 431             jnrlistC         = jjnr[jidx+2];
 432             jnrlistD         = jjnr[jidx+3];
 433             /* Sign of each element will be negative for non-real atoms.
 434              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 435              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 436              */
 437             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 438             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 439             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 440             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 441             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 442             j_coord_offsetA  = DIM*jnrA;
 443             j_coord_offsetB  = DIM*jnrB;
 444             j_coord_offsetC  = DIM*jnrC;
 445             j_coord_offsetD  = DIM*jnrD;
 446
 447             /* load j atom coordinates */
 448             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 449                                               x+j_coord_offsetC,x+j_coord_offsetD,
 450                                               &jx0,&jy0,&jz0);
 451
 452             /* Calculate displacement vector */
 453             dx00             = _mm_sub_ps(ix0,jx0);
 454             dy00             = _mm_sub_ps(iy0,jy0);
 455             dz00             = _mm_sub_ps(iz0,jz0);
 456             dx10             = _mm_sub_ps(ix1,jx0);
 457             dy10             = _mm_sub_ps(iy1,jy0);
 458             dz10             = _mm_sub_ps(iz1,jz0);
 459             dx20             = _mm_sub_ps(ix2,jx0);
 460             dy20             = _mm_sub_ps(iy2,jy0);
 461             dz20             = _mm_sub_ps(iz2,jz0);
 462
 463             /* Calculate squared distance and things based on it */
 464             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 465             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 466             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 467
 468             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 469             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 470             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 471
 472             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 473             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 474             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 475
 476             /* Load parameters for j particles */
 477             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 478                                                               charge+jnrC+0,charge+jnrD+0);
 479             vdwjidx0A        = 2*vdwtype[jnrA+0];
 480             vdwjidx0B        = 2*vdwtype[jnrB+0];
 481             vdwjidx0C        = 2*vdwtype[jnrC+0];
 482             vdwjidx0D        = 2*vdwtype[jnrD+0];
 483
 484             fjx0             = _mm_setzero_ps();
 485             fjy0             = _mm_setzero_ps();
 486             fjz0             = _mm_setzero_ps();
 487
 488             /**************************
 489              * CALCULATE INTERACTIONS *
 490              **************************/
 491
 492             if (gmx_mm_any_lt(rsq00,rcutoff2))
 493             {
 494
 495             r00              = _mm_mul_ps(rsq00,rinv00);
 496             r00              = _mm_andnot_ps(dummy_mask,r00);
 497
 498             /* Compute parameters for interactions between i and j atoms */
 499             qq00             = _mm_mul_ps(iq0,jq0);
 500             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 501                                          vdwparam+vdwioffset0+vdwjidx0B,
 502                                          vdwparam+vdwioffset0+vdwjidx0C,
 503                                          vdwparam+vdwioffset0+vdwjidx0D,
 504                                          &c6_00,&c12_00);
 505
 506             /* EWALD ELECTROSTATICS */
 507
 508             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 509             ewrt             = _mm_mul_ps(r00,ewtabscale);
 510             ewitab           = _mm_cvttps_epi32(ewrt);
 511             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 512             ewitab           = _mm_slli_epi32(ewitab,2);
 513             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 514             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 515             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 516             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 517             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 518             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 519             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 520             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 521             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 522
 523             /* LENNARD-JONES DISPERSION/REPULSION */
 524
 525             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 526             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 527             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 528             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) ,
 529                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 530             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 531
 532             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 533
 534             /* Update potential sum for this i atom from the interaction with this j atom. */
 535             velec            = _mm_and_ps(velec,cutoff_mask);
 536             velec            = _mm_andnot_ps(dummy_mask,velec);
 537             velecsum         = _mm_add_ps(velecsum,velec);
 538             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 539             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 540             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 541
 542             fscal            = _mm_add_ps(felec,fvdw);
 543
 544             fscal            = _mm_and_ps(fscal,cutoff_mask);
 545
 546             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 547
 548             /* Calculate temporary vectorial force */
 549             tx               = _mm_mul_ps(fscal,dx00);
 550             ty               = _mm_mul_ps(fscal,dy00);
 551             tz               = _mm_mul_ps(fscal,dz00);
 552
 553             /* Update vectorial force */
 554             fix0             = _mm_add_ps(fix0,tx);
 555             fiy0             = _mm_add_ps(fiy0,ty);
 556             fiz0             = _mm_add_ps(fiz0,tz);
 557
 558             fjx0             = _mm_add_ps(fjx0,tx);
 559             fjy0             = _mm_add_ps(fjy0,ty);
 560             fjz0             = _mm_add_ps(fjz0,tz);
 561
 562             }
 563
 564             /**************************
 565              * CALCULATE INTERACTIONS *
 566              **************************/
 567
 568             if (gmx_mm_any_lt(rsq10,rcutoff2))
 569             {
 570
 571             r10              = _mm_mul_ps(rsq10,rinv10);
 572             r10              = _mm_andnot_ps(dummy_mask,r10);
 573
 574             /* Compute parameters for interactions between i and j atoms */
 575             qq10             = _mm_mul_ps(iq1,jq0);
 576
 577             /* EWALD ELECTROSTATICS */
 578
 579             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 580             ewrt             = _mm_mul_ps(r10,ewtabscale);
 581             ewitab           = _mm_cvttps_epi32(ewrt);
 582             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 583             ewitab           = _mm_slli_epi32(ewitab,2);
 584             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 585             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 586             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 587             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 588             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 589             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 590             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 591             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 592             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 593
 594             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 595
 596             /* Update potential sum for this i atom from the interaction with this j atom. */
 597             velec            = _mm_and_ps(velec,cutoff_mask);
 598             velec            = _mm_andnot_ps(dummy_mask,velec);
 599             velecsum         = _mm_add_ps(velecsum,velec);
 600
 601             fscal            = felec;
 602
 603             fscal            = _mm_and_ps(fscal,cutoff_mask);
 604
 605             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 606
 607             /* Calculate temporary vectorial force */
 608             tx               = _mm_mul_ps(fscal,dx10);
 609             ty               = _mm_mul_ps(fscal,dy10);
 610             tz               = _mm_mul_ps(fscal,dz10);
 611
 612             /* Update vectorial force */
 613             fix1             = _mm_add_ps(fix1,tx);
 614             fiy1             = _mm_add_ps(fiy1,ty);
 615             fiz1             = _mm_add_ps(fiz1,tz);
 616
 617             fjx0             = _mm_add_ps(fjx0,tx);
 618             fjy0             = _mm_add_ps(fjy0,ty);
 619             fjz0             = _mm_add_ps(fjz0,tz);
 620
 621             }
 622
 623             /**************************
 624              * CALCULATE INTERACTIONS *
 625              **************************/
 626
 627             if (gmx_mm_any_lt(rsq20,rcutoff2))
 628             {
 629
 630             r20              = _mm_mul_ps(rsq20,rinv20);
 631             r20              = _mm_andnot_ps(dummy_mask,r20);
 632
 633             /* Compute parameters for interactions between i and j atoms */
 634             qq20             = _mm_mul_ps(iq2,jq0);
 635
 636             /* EWALD ELECTROSTATICS */
 637
 638             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 639             ewrt             = _mm_mul_ps(r20,ewtabscale);
 640             ewitab           = _mm_cvttps_epi32(ewrt);
 641             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 642             ewitab           = _mm_slli_epi32(ewitab,2);
 643             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 644             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 645             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 646             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 647             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 648             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 649             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 650             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 651             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 652
 653             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 654
 655             /* Update potential sum for this i atom from the interaction with this j atom. */
 656             velec            = _mm_and_ps(velec,cutoff_mask);
 657             velec            = _mm_andnot_ps(dummy_mask,velec);
 658             velecsum         = _mm_add_ps(velecsum,velec);
 659
 660             fscal            = felec;
 661
 662             fscal            = _mm_and_ps(fscal,cutoff_mask);
 663
 664             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 665
 666             /* Calculate temporary vectorial force */
 667             tx               = _mm_mul_ps(fscal,dx20);
 668             ty               = _mm_mul_ps(fscal,dy20);
 669             tz               = _mm_mul_ps(fscal,dz20);
 670
 671             /* Update vectorial force */
 672             fix2             = _mm_add_ps(fix2,tx);
 673             fiy2             = _mm_add_ps(fiy2,ty);
 674             fiz2             = _mm_add_ps(fiz2,tz);
 675
 676             fjx0             = _mm_add_ps(fjx0,tx);
 677             fjy0             = _mm_add_ps(fjy0,ty);
 678             fjz0             = _mm_add_ps(fjz0,tz);
 679
 680             }
 681
 682             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 683             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 684             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 685             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 686
 687             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 688
 689             /* Inner loop uses 159 flops */
 690         }
 691
 692         /* End of innermost loop */
 693
 694         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 695                                               f+i_coord_offset,fshift+i_shift_offset);
 696
 697         ggid                        = gid[iidx];
 698         /* Update potential energies */
 699         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 700         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 701
 702         /* Increment number of inner iterations */
 703         inneriter                  += j_index_end - j_index_start;
 704
 705         /* Outer loop uses 20 flops */
 706     }
 707
 708     /* Increment number of outer iterations */
 709     outeriter        += nri;
 710
 711     /* Update outer/inner flops */
 712
 713     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*159);
 714 }
 715 /*
 716  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_sse4_1_single
 717  * Electrostatics interaction: Ewald
 718  * VdW interaction:            LennardJones
 719  * Geometry:                   Water3-Particle
 720  * Calculate force/pot:        Force
 721  */
 722 void
 723 nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_sse4_1_single
 724                     (t_nblist * gmx_restrict                nlist,
 725                      rvec * gmx_restrict                    xx,
 726                      rvec * gmx_restrict                    ff,
 727                      t_forcerec * gmx_restrict              fr,
 728                      t_mdatoms * gmx_restrict               mdatoms,
 729                      nb_kernel_data_t * gmx_restrict        kernel_data,
 730                      t_nrnb * gmx_restrict                  nrnb)
 731 {
 732     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 733      * just 0 for non-waters.
 734      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 735      * jnr indices corresponding to data put in the four positions in the SIMD register.
 736      */
 737     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 738     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 739     int              jnrA,jnrB,jnrC,jnrD;
 740     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 741     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 742     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 743     real             rcutoff_scalar;
 744     real             *shiftvec,*fshift,*x,*f;
 745     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 746     real             scratch[4*DIM];
 747     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 748     int              vdwioffset0;
 749     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 750     int              vdwioffset1;
 751     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 752     int              vdwioffset2;
 753     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 754     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 755     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 756     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 757     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 758     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 759     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 760     real             *charge;
 761     int              nvdwtype;
 762     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 763     int              *vdwtype;
 764     real             *vdwparam;
 765     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 766     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 767     __m128i          ewitab;
 768     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 769     real             *ewtab;
 770     __m128           dummy_mask,cutoff_mask;
 771     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 772     __m128           one     = _mm_set1_ps(1.0);
 773     __m128           two     = _mm_set1_ps(2.0);
 774     x                = xx[0];
 775     f                = ff[0];
 776
 777     nri              = nlist->nri;
 778     iinr             = nlist->iinr;
 779     jindex           = nlist->jindex;
 780     jjnr             = nlist->jjnr;
 781     shiftidx         = nlist->shift;
 782     gid              = nlist->gid;
 783     shiftvec         = fr->shift_vec[0];
 784     fshift           = fr->fshift[0];
 785     facel            = _mm_set1_ps(fr->epsfac);
 786     charge           = mdatoms->chargeA;
 787     nvdwtype         = fr->ntype;
 788     vdwparam         = fr->nbfp;
 789     vdwtype          = mdatoms->typeA;
 790
 791     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 792     ewtab            = fr->ic->tabq_coul_F;
 793     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 794     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 795
 796     /* Setup water-specific parameters */
 797     inr              = nlist->iinr[0];
 798     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 799     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 800     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 801     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 802
 803     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 804     rcutoff_scalar   = fr->rcoulomb;
 805     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 806     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 807
 808     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 809     rvdw             = _mm_set1_ps(fr->rvdw);
 810
 811     /* Avoid stupid compiler warnings */
 812     jnrA = jnrB = jnrC = jnrD = 0;
 813     j_coord_offsetA = 0;
 814     j_coord_offsetB = 0;
 815     j_coord_offsetC = 0;
 816     j_coord_offsetD = 0;
 817
 818     outeriter        = 0;
 819     inneriter        = 0;
 820
 821     for(iidx=0;iidx<4*DIM;iidx++)
 822     {
 823         scratch[iidx] = 0.0;
 824     }
 825
 826     /* Start outer loop over neighborlists */
 827     for(iidx=0; iidx<nri; iidx++)
 828     {
 829         /* Load shift vector for this list */
 830         i_shift_offset   = DIM*shiftidx[iidx];
 831
 832         /* Load limits for loop over neighbors */
 833         j_index_start    = jindex[iidx];
 834         j_index_end      = jindex[iidx+1];
 835
 836         /* Get outer coordinate index */
 837         inr              = iinr[iidx];
 838         i_coord_offset   = DIM*inr;
 839
 840         /* Load i particle coords and add shift vector */
 841         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 842                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 843
 844         fix0             = _mm_setzero_ps();
 845         fiy0             = _mm_setzero_ps();
 846         fiz0             = _mm_setzero_ps();
 847         fix1             = _mm_setzero_ps();
 848         fiy1             = _mm_setzero_ps();
 849         fiz1             = _mm_setzero_ps();
 850         fix2             = _mm_setzero_ps();
 851         fiy2             = _mm_setzero_ps();
 852         fiz2             = _mm_setzero_ps();
 853
 854         /* Start inner kernel loop */
 855         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 856         {
 857
 858             /* Get j neighbor index, and coordinate index */
 859             jnrA             = jjnr[jidx];
 860             jnrB             = jjnr[jidx+1];
 861             jnrC             = jjnr[jidx+2];
 862             jnrD             = jjnr[jidx+3];
 863             j_coord_offsetA  = DIM*jnrA;
 864             j_coord_offsetB  = DIM*jnrB;
 865             j_coord_offsetC  = DIM*jnrC;
 866             j_coord_offsetD  = DIM*jnrD;
 867
 868             /* load j atom coordinates */
 869             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 870                                               x+j_coord_offsetC,x+j_coord_offsetD,
 871                                               &jx0,&jy0,&jz0);
 872
 873             /* Calculate displacement vector */
 874             dx00             = _mm_sub_ps(ix0,jx0);
 875             dy00             = _mm_sub_ps(iy0,jy0);
 876             dz00             = _mm_sub_ps(iz0,jz0);
 877             dx10             = _mm_sub_ps(ix1,jx0);
 878             dy10             = _mm_sub_ps(iy1,jy0);
 879             dz10             = _mm_sub_ps(iz1,jz0);
 880             dx20             = _mm_sub_ps(ix2,jx0);
 881             dy20             = _mm_sub_ps(iy2,jy0);
 882             dz20             = _mm_sub_ps(iz2,jz0);
 883
 884             /* Calculate squared distance and things based on it */
 885             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 886             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 887             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 888
 889             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 890             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 891             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 892
 893             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 894             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 895             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 896
 897             /* Load parameters for j particles */
 898             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 899                                                               charge+jnrC+0,charge+jnrD+0);
 900             vdwjidx0A        = 2*vdwtype[jnrA+0];
 901             vdwjidx0B        = 2*vdwtype[jnrB+0];
 902             vdwjidx0C        = 2*vdwtype[jnrC+0];
 903             vdwjidx0D        = 2*vdwtype[jnrD+0];
 904
 905             fjx0             = _mm_setzero_ps();
 906             fjy0             = _mm_setzero_ps();
 907             fjz0             = _mm_setzero_ps();
 908
 909             /**************************
 910              * CALCULATE INTERACTIONS *
 911              **************************/
 912
 913             if (gmx_mm_any_lt(rsq00,rcutoff2))
 914             {
 915
 916             r00              = _mm_mul_ps(rsq00,rinv00);
 917
 918             /* Compute parameters for interactions between i and j atoms */
 919             qq00             = _mm_mul_ps(iq0,jq0);
 920             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 921                                          vdwparam+vdwioffset0+vdwjidx0B,
 922                                          vdwparam+vdwioffset0+vdwjidx0C,
 923                                          vdwparam+vdwioffset0+vdwjidx0D,
 924                                          &c6_00,&c12_00);
 925
 926             /* EWALD ELECTROSTATICS */
 927
 928             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 929             ewrt             = _mm_mul_ps(r00,ewtabscale);
 930             ewitab           = _mm_cvttps_epi32(ewrt);
 931             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 932             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 933                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 934                                          &ewtabF,&ewtabFn);
 935             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 936             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 937
 938             /* LENNARD-JONES DISPERSION/REPULSION */
 939
 940             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 941             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 942
 943             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 944
 945             fscal            = _mm_add_ps(felec,fvdw);
 946
 947             fscal            = _mm_and_ps(fscal,cutoff_mask);
 948
 949             /* Calculate temporary vectorial force */
 950             tx               = _mm_mul_ps(fscal,dx00);
 951             ty               = _mm_mul_ps(fscal,dy00);
 952             tz               = _mm_mul_ps(fscal,dz00);
 953
 954             /* Update vectorial force */
 955             fix0             = _mm_add_ps(fix0,tx);
 956             fiy0             = _mm_add_ps(fiy0,ty);
 957             fiz0             = _mm_add_ps(fiz0,tz);
 958
 959             fjx0             = _mm_add_ps(fjx0,tx);
 960             fjy0             = _mm_add_ps(fjy0,ty);
 961             fjz0             = _mm_add_ps(fjz0,tz);
 962
 963             }
 964
 965             /**************************
 966              * CALCULATE INTERACTIONS *
 967              **************************/
 968
 969             if (gmx_mm_any_lt(rsq10,rcutoff2))
 970             {
 971
 972             r10              = _mm_mul_ps(rsq10,rinv10);
 973
 974             /* Compute parameters for interactions between i and j atoms */
 975             qq10             = _mm_mul_ps(iq1,jq0);
 976
 977             /* EWALD ELECTROSTATICS */
 978
 979             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 980             ewrt             = _mm_mul_ps(r10,ewtabscale);
 981             ewitab           = _mm_cvttps_epi32(ewrt);
 982             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 983             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 984                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 985                                          &ewtabF,&ewtabFn);
 986             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 987             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 988
 989             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 990
 991             fscal            = felec;
 992
 993             fscal            = _mm_and_ps(fscal,cutoff_mask);
 994
 995             /* Calculate temporary vectorial force */
 996             tx               = _mm_mul_ps(fscal,dx10);
 997             ty               = _mm_mul_ps(fscal,dy10);
 998             tz               = _mm_mul_ps(fscal,dz10);
 999
1000             /* Update vectorial force */
1001             fix1             = _mm_add_ps(fix1,tx);
1002             fiy1             = _mm_add_ps(fiy1,ty);
1003             fiz1             = _mm_add_ps(fiz1,tz);
1004
1005             fjx0             = _mm_add_ps(fjx0,tx);
1006             fjy0             = _mm_add_ps(fjy0,ty);
1007             fjz0             = _mm_add_ps(fjz0,tz);
1008
1009             }
1010
1011             /**************************
1012              * CALCULATE INTERACTIONS *
1013              **************************/
1014
1015             if (gmx_mm_any_lt(rsq20,rcutoff2))
1016             {
1017
1018             r20              = _mm_mul_ps(rsq20,rinv20);
1019
1020             /* Compute parameters for interactions between i and j atoms */
1021             qq20             = _mm_mul_ps(iq2,jq0);
1022
1023             /* EWALD ELECTROSTATICS */
1024
1025             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1026             ewrt             = _mm_mul_ps(r20,ewtabscale);
1027             ewitab           = _mm_cvttps_epi32(ewrt);
1028             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1029             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1030                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1031                                          &ewtabF,&ewtabFn);
1032             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1033             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1034
1035             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1036
1037             fscal            = felec;
1038
1039             fscal            = _mm_and_ps(fscal,cutoff_mask);
1040
1041             /* Calculate temporary vectorial force */
1042             tx               = _mm_mul_ps(fscal,dx20);
1043             ty               = _mm_mul_ps(fscal,dy20);
1044             tz               = _mm_mul_ps(fscal,dz20);
1045
1046             /* Update vectorial force */
1047             fix2             = _mm_add_ps(fix2,tx);
1048             fiy2             = _mm_add_ps(fiy2,ty);
1049             fiz2             = _mm_add_ps(fiz2,tz);
1050
1051             fjx0             = _mm_add_ps(fjx0,tx);
1052             fjy0             = _mm_add_ps(fjy0,ty);
1053             fjz0             = _mm_add_ps(fjz0,tz);
1054
1055             }
1056
1057             fjptrA             = f+j_coord_offsetA;
1058             fjptrB             = f+j_coord_offsetB;
1059             fjptrC             = f+j_coord_offsetC;
1060             fjptrD             = f+j_coord_offsetD;
1061
1062             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1063
1064             /* Inner loop uses 124 flops */
1065         }
1066
1067         if(jidx<j_index_end)
1068         {
1069
1070             /* Get j neighbor index, and coordinate index */
1071             jnrlistA         = jjnr[jidx];
1072             jnrlistB         = jjnr[jidx+1];
1073             jnrlistC         = jjnr[jidx+2];
1074             jnrlistD         = jjnr[jidx+3];
1075             /* Sign of each element will be negative for non-real atoms.
1076              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1077              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1078              */
1079             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1080             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1081             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1082             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1083             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1084             j_coord_offsetA  = DIM*jnrA;
1085             j_coord_offsetB  = DIM*jnrB;
1086             j_coord_offsetC  = DIM*jnrC;
1087             j_coord_offsetD  = DIM*jnrD;
1088
1089             /* load j atom coordinates */
1090             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1091                                               x+j_coord_offsetC,x+j_coord_offsetD,
1092                                               &jx0,&jy0,&jz0);
1093
1094             /* Calculate displacement vector */
1095             dx00             = _mm_sub_ps(ix0,jx0);
1096             dy00             = _mm_sub_ps(iy0,jy0);
1097             dz00             = _mm_sub_ps(iz0,jz0);
1098             dx10             = _mm_sub_ps(ix1,jx0);
1099             dy10             = _mm_sub_ps(iy1,jy0);
1100             dz10             = _mm_sub_ps(iz1,jz0);
1101             dx20             = _mm_sub_ps(ix2,jx0);
1102             dy20             = _mm_sub_ps(iy2,jy0);
1103             dz20             = _mm_sub_ps(iz2,jz0);
1104
1105             /* Calculate squared distance and things based on it */
1106             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1107             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1108             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1109
1110             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1111             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1112             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1113
1114             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1115             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1116             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1117
1118             /* Load parameters for j particles */
1119             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1120                                                               charge+jnrC+0,charge+jnrD+0);
1121             vdwjidx0A        = 2*vdwtype[jnrA+0];
1122             vdwjidx0B        = 2*vdwtype[jnrB+0];
1123             vdwjidx0C        = 2*vdwtype[jnrC+0];
1124             vdwjidx0D        = 2*vdwtype[jnrD+0];
1125
1126             fjx0             = _mm_setzero_ps();
1127             fjy0             = _mm_setzero_ps();
1128             fjz0             = _mm_setzero_ps();
1129
1130             /**************************
1131              * CALCULATE INTERACTIONS *
1132              **************************/
1133
1134             if (gmx_mm_any_lt(rsq00,rcutoff2))
1135             {
1136
1137             r00              = _mm_mul_ps(rsq00,rinv00);
1138             r00              = _mm_andnot_ps(dummy_mask,r00);
1139
1140             /* Compute parameters for interactions between i and j atoms */
1141             qq00             = _mm_mul_ps(iq0,jq0);
1142             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1143                                          vdwparam+vdwioffset0+vdwjidx0B,
1144                                          vdwparam+vdwioffset0+vdwjidx0C,
1145                                          vdwparam+vdwioffset0+vdwjidx0D,
1146                                          &c6_00,&c12_00);
1147
1148             /* EWALD ELECTROSTATICS */
1149
1150             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1151             ewrt             = _mm_mul_ps(r00,ewtabscale);
1152             ewitab           = _mm_cvttps_epi32(ewrt);
1153             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1154             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1155                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1156                                          &ewtabF,&ewtabFn);
1157             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1158             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1159
1160             /* LENNARD-JONES DISPERSION/REPULSION */
1161
1162             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1163             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1164
1165             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1166
1167             fscal            = _mm_add_ps(felec,fvdw);
1168
1169             fscal            = _mm_and_ps(fscal,cutoff_mask);
1170
1171             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1172
1173             /* Calculate temporary vectorial force */
1174             tx               = _mm_mul_ps(fscal,dx00);
1175             ty               = _mm_mul_ps(fscal,dy00);
1176             tz               = _mm_mul_ps(fscal,dz00);
1177
1178             /* Update vectorial force */
1179             fix0             = _mm_add_ps(fix0,tx);
1180             fiy0             = _mm_add_ps(fiy0,ty);
1181             fiz0             = _mm_add_ps(fiz0,tz);
1182
1183             fjx0             = _mm_add_ps(fjx0,tx);
1184             fjy0             = _mm_add_ps(fjy0,ty);
1185             fjz0             = _mm_add_ps(fjz0,tz);
1186
1187             }
1188
1189             /**************************
1190              * CALCULATE INTERACTIONS *
1191              **************************/
1192
1193             if (gmx_mm_any_lt(rsq10,rcutoff2))
1194             {
1195
1196             r10              = _mm_mul_ps(rsq10,rinv10);
1197             r10              = _mm_andnot_ps(dummy_mask,r10);
1198
1199             /* Compute parameters for interactions between i and j atoms */
1200             qq10             = _mm_mul_ps(iq1,jq0);
1201
1202             /* EWALD ELECTROSTATICS */
1203
1204             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1205             ewrt             = _mm_mul_ps(r10,ewtabscale);
1206             ewitab           = _mm_cvttps_epi32(ewrt);
1207             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1208             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1209                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1210                                          &ewtabF,&ewtabFn);
1211             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1212             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1213
1214             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1215
1216             fscal            = felec;
1217
1218             fscal            = _mm_and_ps(fscal,cutoff_mask);
1219
1220             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1221
1222             /* Calculate temporary vectorial force */
1223             tx               = _mm_mul_ps(fscal,dx10);
1224             ty               = _mm_mul_ps(fscal,dy10);
1225             tz               = _mm_mul_ps(fscal,dz10);
1226
1227             /* Update vectorial force */
1228             fix1             = _mm_add_ps(fix1,tx);
1229             fiy1             = _mm_add_ps(fiy1,ty);
1230             fiz1             = _mm_add_ps(fiz1,tz);
1231
1232             fjx0             = _mm_add_ps(fjx0,tx);
1233             fjy0             = _mm_add_ps(fjy0,ty);
1234             fjz0             = _mm_add_ps(fjz0,tz);
1235
1236             }
1237
1238             /**************************
1239              * CALCULATE INTERACTIONS *
1240              **************************/
1241
1242             if (gmx_mm_any_lt(rsq20,rcutoff2))
1243             {
1244
1245             r20              = _mm_mul_ps(rsq20,rinv20);
1246             r20              = _mm_andnot_ps(dummy_mask,r20);
1247
1248             /* Compute parameters for interactions between i and j atoms */
1249             qq20             = _mm_mul_ps(iq2,jq0);
1250
1251             /* EWALD ELECTROSTATICS */
1252
1253             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1254             ewrt             = _mm_mul_ps(r20,ewtabscale);
1255             ewitab           = _mm_cvttps_epi32(ewrt);
1256             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1257             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1258                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1259                                          &ewtabF,&ewtabFn);
1260             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1261             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1262
1263             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1264
1265             fscal            = felec;
1266
1267             fscal            = _mm_and_ps(fscal,cutoff_mask);
1268
1269             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1270
1271             /* Calculate temporary vectorial force */
1272             tx               = _mm_mul_ps(fscal,dx20);
1273             ty               = _mm_mul_ps(fscal,dy20);
1274             tz               = _mm_mul_ps(fscal,dz20);
1275
1276             /* Update vectorial force */
1277             fix2             = _mm_add_ps(fix2,tx);
1278             fiy2             = _mm_add_ps(fiy2,ty);
1279             fiz2             = _mm_add_ps(fiz2,tz);
1280
1281             fjx0             = _mm_add_ps(fjx0,tx);
1282             fjy0             = _mm_add_ps(fjy0,ty);
1283             fjz0             = _mm_add_ps(fjz0,tz);
1284
1285             }
1286
1287             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1288             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1289             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1290             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1291
1292             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1293
1294             /* Inner loop uses 127 flops */
1295         }
1296
1297         /* End of innermost loop */
1298
1299         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1300                                               f+i_coord_offset,fshift+i_shift_offset);
1301
1302         /* Increment number of inner iterations */
1303         inneriter                  += j_index_end - j_index_start;
1304
1305         /* Outer loop uses 18 flops */
1306     }
1307
1308     /* Increment number of outer iterations */
1309     outeriter        += nri;
1310
1311     /* Update outer/inner flops */
1312
1313     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*127);
1314 }