src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEwSh_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_ElecEwSh_VdwLJSh_GeomW3P1_VF_sse2_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_sse2_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             shX,shY,shZ,rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwioffset1;
  69     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  70     int              vdwioffset2;
  71     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  72     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  73     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  74     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  75     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  76     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  77     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  78     real             *charge;
  79     int              nvdwtype;
  80     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  81     int              *vdwtype;
  82     real             *vdwparam;
  83     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  84     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  85     __m128i          ewitab;
  86     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  87     real             *ewtab;
  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     nvdwtype         = fr->ntype;
 106     vdwparam         = fr->nbfp;
 107     vdwtype          = mdatoms->typeA;
 108
 109     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 110     ewtab            = fr->ic->tabq_coul_FDV0;
 111     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 112     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 113
 114     /* Setup water-specific parameters */
 115     inr              = nlist->iinr[0];
 116     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 117     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 118     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 119     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 120
 121     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 122     rcutoff_scalar   = fr->rcoulomb;
 123     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 124     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 125
 126     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 127     rvdw             = _mm_set1_ps(fr->rvdw);
 128
 129     /* Avoid stupid compiler warnings */
 130     jnrA = jnrB = jnrC = jnrD = 0;
 131     j_coord_offsetA = 0;
 132     j_coord_offsetB = 0;
 133     j_coord_offsetC = 0;
 134     j_coord_offsetD = 0;
 135
 136     outeriter        = 0;
 137     inneriter        = 0;
 138
 139     /* Start outer loop over neighborlists */
 140     for(iidx=0; iidx<nri; iidx++)
 141     {
 142         /* Load shift vector for this list */
 143         i_shift_offset   = DIM*shiftidx[iidx];
 144         shX              = shiftvec[i_shift_offset+XX];
 145         shY              = shiftvec[i_shift_offset+YY];
 146         shZ              = shiftvec[i_shift_offset+ZZ];
 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         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 158         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 159         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 160         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 161         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 162         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 163         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 164         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 165         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 166
 167         fix0             = _mm_setzero_ps();
 168         fiy0             = _mm_setzero_ps();
 169         fiz0             = _mm_setzero_ps();
 170         fix1             = _mm_setzero_ps();
 171         fiy1             = _mm_setzero_ps();
 172         fiz1             = _mm_setzero_ps();
 173         fix2             = _mm_setzero_ps();
 174         fiy2             = _mm_setzero_ps();
 175         fiz2             = _mm_setzero_ps();
 176
 177         /* Reset potential sums */
 178         velecsum         = _mm_setzero_ps();
 179         vvdwsum          = _mm_setzero_ps();
 180
 181         /* Start inner kernel loop */
 182         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 183         {
 184
 185             /* Get j neighbor index, and coordinate index */
 186             jnrA             = jjnr[jidx];
 187             jnrB             = jjnr[jidx+1];
 188             jnrC             = jjnr[jidx+2];
 189             jnrD             = jjnr[jidx+3];
 190
 191             j_coord_offsetA  = DIM*jnrA;
 192             j_coord_offsetB  = DIM*jnrB;
 193             j_coord_offsetC  = DIM*jnrC;
 194             j_coord_offsetD  = DIM*jnrD;
 195
 196             /* load j atom coordinates */
 197             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 198                                               x+j_coord_offsetC,x+j_coord_offsetD,
 199                                               &jx0,&jy0,&jz0);
 200
 201             /* Calculate displacement vector */
 202             dx00             = _mm_sub_ps(ix0,jx0);
 203             dy00             = _mm_sub_ps(iy0,jy0);
 204             dz00             = _mm_sub_ps(iz0,jz0);
 205             dx10             = _mm_sub_ps(ix1,jx0);
 206             dy10             = _mm_sub_ps(iy1,jy0);
 207             dz10             = _mm_sub_ps(iz1,jz0);
 208             dx20             = _mm_sub_ps(ix2,jx0);
 209             dy20             = _mm_sub_ps(iy2,jy0);
 210             dz20             = _mm_sub_ps(iz2,jz0);
 211
 212             /* Calculate squared distance and things based on it */
 213             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 214             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 215             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 216
 217             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 218             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 219             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 220
 221             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 222             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 223             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 224
 225             /* Load parameters for j particles */
 226             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 227                                                               charge+jnrC+0,charge+jnrD+0);
 228             vdwjidx0A        = 2*vdwtype[jnrA+0];
 229             vdwjidx0B        = 2*vdwtype[jnrB+0];
 230             vdwjidx0C        = 2*vdwtype[jnrC+0];
 231             vdwjidx0D        = 2*vdwtype[jnrD+0];
 232
 233             /**************************
 234              * CALCULATE INTERACTIONS *
 235              **************************/
 236
 237             if (gmx_mm_any_lt(rsq00,rcutoff2))
 238             {
 239
 240             r00              = _mm_mul_ps(rsq00,rinv00);
 241
 242             /* Compute parameters for interactions between i and j atoms */
 243             qq00             = _mm_mul_ps(iq0,jq0);
 244             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 245                                          vdwparam+vdwioffset0+vdwjidx0B,
 246                                          vdwparam+vdwioffset0+vdwjidx0C,
 247                                          vdwparam+vdwioffset0+vdwjidx0D,
 248                                          &c6_00,&c12_00);
 249
 250             /* EWALD ELECTROSTATICS */
 251
 252             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 253             ewrt             = _mm_mul_ps(r00,ewtabscale);
 254             ewitab           = _mm_cvttps_epi32(ewrt);
 255             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 256             ewitab           = _mm_slli_epi32(ewitab,2);
 257             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 258             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 259             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 260             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 261             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 262             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 263             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 264             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 265             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 266
 267             /* LENNARD-JONES DISPERSION/REPULSION */
 268
 269             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 270             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 271             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 272             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) ,
 273                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 274             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 275
 276             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 277
 278             /* Update potential sum for this i atom from the interaction with this j atom. */
 279             velec            = _mm_and_ps(velec,cutoff_mask);
 280             velecsum         = _mm_add_ps(velecsum,velec);
 281             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 282             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 283
 284             fscal            = _mm_add_ps(felec,fvdw);
 285
 286             fscal            = _mm_and_ps(fscal,cutoff_mask);
 287
 288             /* Calculate temporary vectorial force */
 289             tx               = _mm_mul_ps(fscal,dx00);
 290             ty               = _mm_mul_ps(fscal,dy00);
 291             tz               = _mm_mul_ps(fscal,dz00);
 292
 293             /* Update vectorial force */
 294             fix0             = _mm_add_ps(fix0,tx);
 295             fiy0             = _mm_add_ps(fiy0,ty);
 296             fiz0             = _mm_add_ps(fiz0,tz);
 297
 298             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 299                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 300                                                    tx,ty,tz);
 301
 302             }
 303
 304             /**************************
 305              * CALCULATE INTERACTIONS *
 306              **************************/
 307
 308             if (gmx_mm_any_lt(rsq10,rcutoff2))
 309             {
 310
 311             r10              = _mm_mul_ps(rsq10,rinv10);
 312
 313             /* Compute parameters for interactions between i and j atoms */
 314             qq10             = _mm_mul_ps(iq1,jq0);
 315
 316             /* EWALD ELECTROSTATICS */
 317
 318             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 319             ewrt             = _mm_mul_ps(r10,ewtabscale);
 320             ewitab           = _mm_cvttps_epi32(ewrt);
 321             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 322             ewitab           = _mm_slli_epi32(ewitab,2);
 323             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 324             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 325             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 326             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 327             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 328             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 329             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 330             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 331             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 332
 333             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 334
 335             /* Update potential sum for this i atom from the interaction with this j atom. */
 336             velec            = _mm_and_ps(velec,cutoff_mask);
 337             velecsum         = _mm_add_ps(velecsum,velec);
 338
 339             fscal            = felec;
 340
 341             fscal            = _mm_and_ps(fscal,cutoff_mask);
 342
 343             /* Calculate temporary vectorial force */
 344             tx               = _mm_mul_ps(fscal,dx10);
 345             ty               = _mm_mul_ps(fscal,dy10);
 346             tz               = _mm_mul_ps(fscal,dz10);
 347
 348             /* Update vectorial force */
 349             fix1             = _mm_add_ps(fix1,tx);
 350             fiy1             = _mm_add_ps(fiy1,ty);
 351             fiz1             = _mm_add_ps(fiz1,tz);
 352
 353             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 354                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 355                                                    tx,ty,tz);
 356
 357             }
 358
 359             /**************************
 360              * CALCULATE INTERACTIONS *
 361              **************************/
 362
 363             if (gmx_mm_any_lt(rsq20,rcutoff2))
 364             {
 365
 366             r20              = _mm_mul_ps(rsq20,rinv20);
 367
 368             /* Compute parameters for interactions between i and j atoms */
 369             qq20             = _mm_mul_ps(iq2,jq0);
 370
 371             /* EWALD ELECTROSTATICS */
 372
 373             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 374             ewrt             = _mm_mul_ps(r20,ewtabscale);
 375             ewitab           = _mm_cvttps_epi32(ewrt);
 376             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 377             ewitab           = _mm_slli_epi32(ewitab,2);
 378             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 379             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 380             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 381             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 382             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 383             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 384             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 385             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 386             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 387
 388             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 389
 390             /* Update potential sum for this i atom from the interaction with this j atom. */
 391             velec            = _mm_and_ps(velec,cutoff_mask);
 392             velecsum         = _mm_add_ps(velecsum,velec);
 393
 394             fscal            = felec;
 395
 396             fscal            = _mm_and_ps(fscal,cutoff_mask);
 397
 398             /* Calculate temporary vectorial force */
 399             tx               = _mm_mul_ps(fscal,dx20);
 400             ty               = _mm_mul_ps(fscal,dy20);
 401             tz               = _mm_mul_ps(fscal,dz20);
 402
 403             /* Update vectorial force */
 404             fix2             = _mm_add_ps(fix2,tx);
 405             fiy2             = _mm_add_ps(fiy2,ty);
 406             fiz2             = _mm_add_ps(fiz2,tz);
 407
 408             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 409                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 410                                                    tx,ty,tz);
 411
 412             }
 413
 414             /* Inner loop uses 156 flops */
 415         }
 416
 417         if(jidx<j_index_end)
 418         {
 419
 420             /* Get j neighbor index, and coordinate index */
 421             jnrA             = jjnr[jidx];
 422             jnrB             = jjnr[jidx+1];
 423             jnrC             = jjnr[jidx+2];
 424             jnrD             = jjnr[jidx+3];
 425
 426             /* Sign of each element will be negative for non-real atoms.
 427              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 428              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 429              */
 430             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 431             jnrA       = (jnrA>=0) ? jnrA : 0;
 432             jnrB       = (jnrB>=0) ? jnrB : 0;
 433             jnrC       = (jnrC>=0) ? jnrC : 0;
 434             jnrD       = (jnrD>=0) ? jnrD : 0;
 435
 436             j_coord_offsetA  = DIM*jnrA;
 437             j_coord_offsetB  = DIM*jnrB;
 438             j_coord_offsetC  = DIM*jnrC;
 439             j_coord_offsetD  = DIM*jnrD;
 440
 441             /* load j atom coordinates */
 442             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 443                                               x+j_coord_offsetC,x+j_coord_offsetD,
 444                                               &jx0,&jy0,&jz0);
 445
 446             /* Calculate displacement vector */
 447             dx00             = _mm_sub_ps(ix0,jx0);
 448             dy00             = _mm_sub_ps(iy0,jy0);
 449             dz00             = _mm_sub_ps(iz0,jz0);
 450             dx10             = _mm_sub_ps(ix1,jx0);
 451             dy10             = _mm_sub_ps(iy1,jy0);
 452             dz10             = _mm_sub_ps(iz1,jz0);
 453             dx20             = _mm_sub_ps(ix2,jx0);
 454             dy20             = _mm_sub_ps(iy2,jy0);
 455             dz20             = _mm_sub_ps(iz2,jz0);
 456
 457             /* Calculate squared distance and things based on it */
 458             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 459             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 460             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 461
 462             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 463             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 464             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 465
 466             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 467             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 468             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 469
 470             /* Load parameters for j particles */
 471             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 472                                                               charge+jnrC+0,charge+jnrD+0);
 473             vdwjidx0A        = 2*vdwtype[jnrA+0];
 474             vdwjidx0B        = 2*vdwtype[jnrB+0];
 475             vdwjidx0C        = 2*vdwtype[jnrC+0];
 476             vdwjidx0D        = 2*vdwtype[jnrD+0];
 477
 478             /**************************
 479              * CALCULATE INTERACTIONS *
 480              **************************/
 481
 482             if (gmx_mm_any_lt(rsq00,rcutoff2))
 483             {
 484
 485             r00              = _mm_mul_ps(rsq00,rinv00);
 486             r00              = _mm_andnot_ps(dummy_mask,r00);
 487
 488             /* Compute parameters for interactions between i and j atoms */
 489             qq00             = _mm_mul_ps(iq0,jq0);
 490             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 491                                          vdwparam+vdwioffset0+vdwjidx0B,
 492                                          vdwparam+vdwioffset0+vdwjidx0C,
 493                                          vdwparam+vdwioffset0+vdwjidx0D,
 494                                          &c6_00,&c12_00);
 495
 496             /* EWALD ELECTROSTATICS */
 497
 498             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 499             ewrt             = _mm_mul_ps(r00,ewtabscale);
 500             ewitab           = _mm_cvttps_epi32(ewrt);
 501             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 502             ewitab           = _mm_slli_epi32(ewitab,2);
 503             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 504             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 505             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 506             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 507             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 508             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 509             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 510             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 511             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 512
 513             /* LENNARD-JONES DISPERSION/REPULSION */
 514
 515             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 516             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 517             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 518             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) ,
 519                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 520             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 521
 522             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 523
 524             /* Update potential sum for this i atom from the interaction with this j atom. */
 525             velec            = _mm_and_ps(velec,cutoff_mask);
 526             velec            = _mm_andnot_ps(dummy_mask,velec);
 527             velecsum         = _mm_add_ps(velecsum,velec);
 528             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 529             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 530             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 531
 532             fscal            = _mm_add_ps(felec,fvdw);
 533
 534             fscal            = _mm_and_ps(fscal,cutoff_mask);
 535
 536             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 537
 538             /* Calculate temporary vectorial force */
 539             tx               = _mm_mul_ps(fscal,dx00);
 540             ty               = _mm_mul_ps(fscal,dy00);
 541             tz               = _mm_mul_ps(fscal,dz00);
 542
 543             /* Update vectorial force */
 544             fix0             = _mm_add_ps(fix0,tx);
 545             fiy0             = _mm_add_ps(fiy0,ty);
 546             fiz0             = _mm_add_ps(fiz0,tz);
 547
 548             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 549                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 550                                                    tx,ty,tz);
 551
 552             }
 553
 554             /**************************
 555              * CALCULATE INTERACTIONS *
 556              **************************/
 557
 558             if (gmx_mm_any_lt(rsq10,rcutoff2))
 559             {
 560
 561             r10              = _mm_mul_ps(rsq10,rinv10);
 562             r10              = _mm_andnot_ps(dummy_mask,r10);
 563
 564             /* Compute parameters for interactions between i and j atoms */
 565             qq10             = _mm_mul_ps(iq1,jq0);
 566
 567             /* EWALD ELECTROSTATICS */
 568
 569             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 570             ewrt             = _mm_mul_ps(r10,ewtabscale);
 571             ewitab           = _mm_cvttps_epi32(ewrt);
 572             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 573             ewitab           = _mm_slli_epi32(ewitab,2);
 574             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 575             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 576             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 577             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 578             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 579             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 580             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 581             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 582             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 583
 584             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 585
 586             /* Update potential sum for this i atom from the interaction with this j atom. */
 587             velec            = _mm_and_ps(velec,cutoff_mask);
 588             velec            = _mm_andnot_ps(dummy_mask,velec);
 589             velecsum         = _mm_add_ps(velecsum,velec);
 590
 591             fscal            = felec;
 592
 593             fscal            = _mm_and_ps(fscal,cutoff_mask);
 594
 595             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 596
 597             /* Calculate temporary vectorial force */
 598             tx               = _mm_mul_ps(fscal,dx10);
 599             ty               = _mm_mul_ps(fscal,dy10);
 600             tz               = _mm_mul_ps(fscal,dz10);
 601
 602             /* Update vectorial force */
 603             fix1             = _mm_add_ps(fix1,tx);
 604             fiy1             = _mm_add_ps(fiy1,ty);
 605             fiz1             = _mm_add_ps(fiz1,tz);
 606
 607             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 608                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 609                                                    tx,ty,tz);
 610
 611             }
 612
 613             /**************************
 614              * CALCULATE INTERACTIONS *
 615              **************************/
 616
 617             if (gmx_mm_any_lt(rsq20,rcutoff2))
 618             {
 619
 620             r20              = _mm_mul_ps(rsq20,rinv20);
 621             r20              = _mm_andnot_ps(dummy_mask,r20);
 622
 623             /* Compute parameters for interactions between i and j atoms */
 624             qq20             = _mm_mul_ps(iq2,jq0);
 625
 626             /* EWALD ELECTROSTATICS */
 627
 628             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 629             ewrt             = _mm_mul_ps(r20,ewtabscale);
 630             ewitab           = _mm_cvttps_epi32(ewrt);
 631             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 632             ewitab           = _mm_slli_epi32(ewitab,2);
 633             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 634             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 635             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 636             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 637             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 638             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 639             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 640             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 641             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 642
 643             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 644
 645             /* Update potential sum for this i atom from the interaction with this j atom. */
 646             velec            = _mm_and_ps(velec,cutoff_mask);
 647             velec            = _mm_andnot_ps(dummy_mask,velec);
 648             velecsum         = _mm_add_ps(velecsum,velec);
 649
 650             fscal            = felec;
 651
 652             fscal            = _mm_and_ps(fscal,cutoff_mask);
 653
 654             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 655
 656             /* Calculate temporary vectorial force */
 657             tx               = _mm_mul_ps(fscal,dx20);
 658             ty               = _mm_mul_ps(fscal,dy20);
 659             tz               = _mm_mul_ps(fscal,dz20);
 660
 661             /* Update vectorial force */
 662             fix2             = _mm_add_ps(fix2,tx);
 663             fiy2             = _mm_add_ps(fiy2,ty);
 664             fiz2             = _mm_add_ps(fiz2,tz);
 665
 666             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 667                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 668                                                    tx,ty,tz);
 669
 670             }
 671
 672             /* Inner loop uses 159 flops */
 673         }
 674
 675         /* End of innermost loop */
 676
 677         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 678                                               f+i_coord_offset,fshift+i_shift_offset);
 679
 680         ggid                        = gid[iidx];
 681         /* Update potential energies */
 682         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 683         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 684
 685         /* Increment number of inner iterations */
 686         inneriter                  += j_index_end - j_index_start;
 687
 688         /* Outer loop uses 29 flops */
 689     }
 690
 691     /* Increment number of outer iterations */
 692     outeriter        += nri;
 693
 694     /* Update outer/inner flops */
 695
 696     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*29 + inneriter*159);
 697 }
 698 /*
 699  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_sse2_single
 700  * Electrostatics interaction: Ewald
 701  * VdW interaction:            LennardJones
 702  * Geometry:                   Water3-Particle
 703  * Calculate force/pot:        Force
 704  */
 705 void
 706 nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_sse2_single
 707                     (t_nblist * gmx_restrict                nlist,
 708                      rvec * gmx_restrict                    xx,
 709                      rvec * gmx_restrict                    ff,
 710                      t_forcerec * gmx_restrict              fr,
 711                      t_mdatoms * gmx_restrict               mdatoms,
 712                      nb_kernel_data_t * gmx_restrict        kernel_data,
 713                      t_nrnb * gmx_restrict                  nrnb)
 714 {
 715     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 716      * just 0 for non-waters.
 717      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 718      * jnr indices corresponding to data put in the four positions in the SIMD register.
 719      */
 720     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 721     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 722     int              jnrA,jnrB,jnrC,jnrD;
 723     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 724     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 725     real             shX,shY,shZ,rcutoff_scalar;
 726     real             *shiftvec,*fshift,*x,*f;
 727     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 728     int              vdwioffset0;
 729     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 730     int              vdwioffset1;
 731     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 732     int              vdwioffset2;
 733     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 734     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 735     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 736     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 737     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 738     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 739     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 740     real             *charge;
 741     int              nvdwtype;
 742     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 743     int              *vdwtype;
 744     real             *vdwparam;
 745     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 746     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 747     __m128i          ewitab;
 748     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 749     real             *ewtab;
 750     __m128           dummy_mask,cutoff_mask;
 751     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 752     __m128           one     = _mm_set1_ps(1.0);
 753     __m128           two     = _mm_set1_ps(2.0);
 754     x                = xx[0];
 755     f                = ff[0];
 756
 757     nri              = nlist->nri;
 758     iinr             = nlist->iinr;
 759     jindex           = nlist->jindex;
 760     jjnr             = nlist->jjnr;
 761     shiftidx         = nlist->shift;
 762     gid              = nlist->gid;
 763     shiftvec         = fr->shift_vec[0];
 764     fshift           = fr->fshift[0];
 765     facel            = _mm_set1_ps(fr->epsfac);
 766     charge           = mdatoms->chargeA;
 767     nvdwtype         = fr->ntype;
 768     vdwparam         = fr->nbfp;
 769     vdwtype          = mdatoms->typeA;
 770
 771     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 772     ewtab            = fr->ic->tabq_coul_F;
 773     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 774     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 775
 776     /* Setup water-specific parameters */
 777     inr              = nlist->iinr[0];
 778     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 779     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 780     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 781     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 782
 783     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 784     rcutoff_scalar   = fr->rcoulomb;
 785     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 786     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 787
 788     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 789     rvdw             = _mm_set1_ps(fr->rvdw);
 790
 791     /* Avoid stupid compiler warnings */
 792     jnrA = jnrB = jnrC = jnrD = 0;
 793     j_coord_offsetA = 0;
 794     j_coord_offsetB = 0;
 795     j_coord_offsetC = 0;
 796     j_coord_offsetD = 0;
 797
 798     outeriter        = 0;
 799     inneriter        = 0;
 800
 801     /* Start outer loop over neighborlists */
 802     for(iidx=0; iidx<nri; iidx++)
 803     {
 804         /* Load shift vector for this list */
 805         i_shift_offset   = DIM*shiftidx[iidx];
 806         shX              = shiftvec[i_shift_offset+XX];
 807         shY              = shiftvec[i_shift_offset+YY];
 808         shZ              = shiftvec[i_shift_offset+ZZ];
 809
 810         /* Load limits for loop over neighbors */
 811         j_index_start    = jindex[iidx];
 812         j_index_end      = jindex[iidx+1];
 813
 814         /* Get outer coordinate index */
 815         inr              = iinr[iidx];
 816         i_coord_offset   = DIM*inr;
 817
 818         /* Load i particle coords and add shift vector */
 819         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 820         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 821         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 822         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 823         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 824         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 825         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 826         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 827         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 828
 829         fix0             = _mm_setzero_ps();
 830         fiy0             = _mm_setzero_ps();
 831         fiz0             = _mm_setzero_ps();
 832         fix1             = _mm_setzero_ps();
 833         fiy1             = _mm_setzero_ps();
 834         fiz1             = _mm_setzero_ps();
 835         fix2             = _mm_setzero_ps();
 836         fiy2             = _mm_setzero_ps();
 837         fiz2             = _mm_setzero_ps();
 838
 839         /* Start inner kernel loop */
 840         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 841         {
 842
 843             /* Get j neighbor index, and coordinate index */
 844             jnrA             = jjnr[jidx];
 845             jnrB             = jjnr[jidx+1];
 846             jnrC             = jjnr[jidx+2];
 847             jnrD             = jjnr[jidx+3];
 848
 849             j_coord_offsetA  = DIM*jnrA;
 850             j_coord_offsetB  = DIM*jnrB;
 851             j_coord_offsetC  = DIM*jnrC;
 852             j_coord_offsetD  = DIM*jnrD;
 853
 854             /* load j atom coordinates */
 855             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 856                                               x+j_coord_offsetC,x+j_coord_offsetD,
 857                                               &jx0,&jy0,&jz0);
 858
 859             /* Calculate displacement vector */
 860             dx00             = _mm_sub_ps(ix0,jx0);
 861             dy00             = _mm_sub_ps(iy0,jy0);
 862             dz00             = _mm_sub_ps(iz0,jz0);
 863             dx10             = _mm_sub_ps(ix1,jx0);
 864             dy10             = _mm_sub_ps(iy1,jy0);
 865             dz10             = _mm_sub_ps(iz1,jz0);
 866             dx20             = _mm_sub_ps(ix2,jx0);
 867             dy20             = _mm_sub_ps(iy2,jy0);
 868             dz20             = _mm_sub_ps(iz2,jz0);
 869
 870             /* Calculate squared distance and things based on it */
 871             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 872             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 873             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 874
 875             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 876             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 877             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 878
 879             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 880             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 881             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 882
 883             /* Load parameters for j particles */
 884             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 885                                                               charge+jnrC+0,charge+jnrD+0);
 886             vdwjidx0A        = 2*vdwtype[jnrA+0];
 887             vdwjidx0B        = 2*vdwtype[jnrB+0];
 888             vdwjidx0C        = 2*vdwtype[jnrC+0];
 889             vdwjidx0D        = 2*vdwtype[jnrD+0];
 890
 891             /**************************
 892              * CALCULATE INTERACTIONS *
 893              **************************/
 894
 895             if (gmx_mm_any_lt(rsq00,rcutoff2))
 896             {
 897
 898             r00              = _mm_mul_ps(rsq00,rinv00);
 899
 900             /* Compute parameters for interactions between i and j atoms */
 901             qq00             = _mm_mul_ps(iq0,jq0);
 902             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 903                                          vdwparam+vdwioffset0+vdwjidx0B,
 904                                          vdwparam+vdwioffset0+vdwjidx0C,
 905                                          vdwparam+vdwioffset0+vdwjidx0D,
 906                                          &c6_00,&c12_00);
 907
 908             /* EWALD ELECTROSTATICS */
 909
 910             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 911             ewrt             = _mm_mul_ps(r00,ewtabscale);
 912             ewitab           = _mm_cvttps_epi32(ewrt);
 913             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 914             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 915                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 916                                          &ewtabF,&ewtabFn);
 917             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 918             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 919
 920             /* LENNARD-JONES DISPERSION/REPULSION */
 921
 922             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 923             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 924
 925             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 926
 927             fscal            = _mm_add_ps(felec,fvdw);
 928
 929             fscal            = _mm_and_ps(fscal,cutoff_mask);
 930
 931             /* Calculate temporary vectorial force */
 932             tx               = _mm_mul_ps(fscal,dx00);
 933             ty               = _mm_mul_ps(fscal,dy00);
 934             tz               = _mm_mul_ps(fscal,dz00);
 935
 936             /* Update vectorial force */
 937             fix0             = _mm_add_ps(fix0,tx);
 938             fiy0             = _mm_add_ps(fiy0,ty);
 939             fiz0             = _mm_add_ps(fiz0,tz);
 940
 941             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 942                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 943                                                    tx,ty,tz);
 944
 945             }
 946
 947             /**************************
 948              * CALCULATE INTERACTIONS *
 949              **************************/
 950
 951             if (gmx_mm_any_lt(rsq10,rcutoff2))
 952             {
 953
 954             r10              = _mm_mul_ps(rsq10,rinv10);
 955
 956             /* Compute parameters for interactions between i and j atoms */
 957             qq10             = _mm_mul_ps(iq1,jq0);
 958
 959             /* EWALD ELECTROSTATICS */
 960
 961             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 962             ewrt             = _mm_mul_ps(r10,ewtabscale);
 963             ewitab           = _mm_cvttps_epi32(ewrt);
 964             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 965             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 966                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 967                                          &ewtabF,&ewtabFn);
 968             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 969             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 970
 971             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 972
 973             fscal            = felec;
 974
 975             fscal            = _mm_and_ps(fscal,cutoff_mask);
 976
 977             /* Calculate temporary vectorial force */
 978             tx               = _mm_mul_ps(fscal,dx10);
 979             ty               = _mm_mul_ps(fscal,dy10);
 980             tz               = _mm_mul_ps(fscal,dz10);
 981
 982             /* Update vectorial force */
 983             fix1             = _mm_add_ps(fix1,tx);
 984             fiy1             = _mm_add_ps(fiy1,ty);
 985             fiz1             = _mm_add_ps(fiz1,tz);
 986
 987             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 988                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 989                                                    tx,ty,tz);
 990
 991             }
 992
 993             /**************************
 994              * CALCULATE INTERACTIONS *
 995              **************************/
 996
 997             if (gmx_mm_any_lt(rsq20,rcutoff2))
 998             {
 999
1000             r20              = _mm_mul_ps(rsq20,rinv20);
1001
1002             /* Compute parameters for interactions between i and j atoms */
1003             qq20             = _mm_mul_ps(iq2,jq0);
1004
1005             /* EWALD ELECTROSTATICS */
1006
1007             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1008             ewrt             = _mm_mul_ps(r20,ewtabscale);
1009             ewitab           = _mm_cvttps_epi32(ewrt);
1010             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1011             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1012                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1013                                          &ewtabF,&ewtabFn);
1014             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1015             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1016
1017             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1018
1019             fscal            = felec;
1020
1021             fscal            = _mm_and_ps(fscal,cutoff_mask);
1022
1023             /* Calculate temporary vectorial force */
1024             tx               = _mm_mul_ps(fscal,dx20);
1025             ty               = _mm_mul_ps(fscal,dy20);
1026             tz               = _mm_mul_ps(fscal,dz20);
1027
1028             /* Update vectorial force */
1029             fix2             = _mm_add_ps(fix2,tx);
1030             fiy2             = _mm_add_ps(fiy2,ty);
1031             fiz2             = _mm_add_ps(fiz2,tz);
1032
1033             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1034                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1035                                                    tx,ty,tz);
1036
1037             }
1038
1039             /* Inner loop uses 124 flops */
1040         }
1041
1042         if(jidx<j_index_end)
1043         {
1044
1045             /* Get j neighbor index, and coordinate index */
1046             jnrA             = jjnr[jidx];
1047             jnrB             = jjnr[jidx+1];
1048             jnrC             = jjnr[jidx+2];
1049             jnrD             = jjnr[jidx+3];
1050
1051             /* Sign of each element will be negative for non-real atoms.
1052              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1053              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1054              */
1055             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1056             jnrA       = (jnrA>=0) ? jnrA : 0;
1057             jnrB       = (jnrB>=0) ? jnrB : 0;
1058             jnrC       = (jnrC>=0) ? jnrC : 0;
1059             jnrD       = (jnrD>=0) ? jnrD : 0;
1060
1061             j_coord_offsetA  = DIM*jnrA;
1062             j_coord_offsetB  = DIM*jnrB;
1063             j_coord_offsetC  = DIM*jnrC;
1064             j_coord_offsetD  = DIM*jnrD;
1065
1066             /* load j atom coordinates */
1067             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1068                                               x+j_coord_offsetC,x+j_coord_offsetD,
1069                                               &jx0,&jy0,&jz0);
1070
1071             /* Calculate displacement vector */
1072             dx00             = _mm_sub_ps(ix0,jx0);
1073             dy00             = _mm_sub_ps(iy0,jy0);
1074             dz00             = _mm_sub_ps(iz0,jz0);
1075             dx10             = _mm_sub_ps(ix1,jx0);
1076             dy10             = _mm_sub_ps(iy1,jy0);
1077             dz10             = _mm_sub_ps(iz1,jz0);
1078             dx20             = _mm_sub_ps(ix2,jx0);
1079             dy20             = _mm_sub_ps(iy2,jy0);
1080             dz20             = _mm_sub_ps(iz2,jz0);
1081
1082             /* Calculate squared distance and things based on it */
1083             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1084             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1085             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1086
1087             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1088             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1089             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1090
1091             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1092             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1093             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1094
1095             /* Load parameters for j particles */
1096             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1097                                                               charge+jnrC+0,charge+jnrD+0);
1098             vdwjidx0A        = 2*vdwtype[jnrA+0];
1099             vdwjidx0B        = 2*vdwtype[jnrB+0];
1100             vdwjidx0C        = 2*vdwtype[jnrC+0];
1101             vdwjidx0D        = 2*vdwtype[jnrD+0];
1102
1103             /**************************
1104              * CALCULATE INTERACTIONS *
1105              **************************/
1106
1107             if (gmx_mm_any_lt(rsq00,rcutoff2))
1108             {
1109
1110             r00              = _mm_mul_ps(rsq00,rinv00);
1111             r00              = _mm_andnot_ps(dummy_mask,r00);
1112
1113             /* Compute parameters for interactions between i and j atoms */
1114             qq00             = _mm_mul_ps(iq0,jq0);
1115             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1116                                          vdwparam+vdwioffset0+vdwjidx0B,
1117                                          vdwparam+vdwioffset0+vdwjidx0C,
1118                                          vdwparam+vdwioffset0+vdwjidx0D,
1119                                          &c6_00,&c12_00);
1120
1121             /* EWALD ELECTROSTATICS */
1122
1123             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1124             ewrt             = _mm_mul_ps(r00,ewtabscale);
1125             ewitab           = _mm_cvttps_epi32(ewrt);
1126             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1127             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1128                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1129                                          &ewtabF,&ewtabFn);
1130             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1131             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1132
1133             /* LENNARD-JONES DISPERSION/REPULSION */
1134
1135             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1136             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1137
1138             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1139
1140             fscal            = _mm_add_ps(felec,fvdw);
1141
1142             fscal            = _mm_and_ps(fscal,cutoff_mask);
1143
1144             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1145
1146             /* Calculate temporary vectorial force */
1147             tx               = _mm_mul_ps(fscal,dx00);
1148             ty               = _mm_mul_ps(fscal,dy00);
1149             tz               = _mm_mul_ps(fscal,dz00);
1150
1151             /* Update vectorial force */
1152             fix0             = _mm_add_ps(fix0,tx);
1153             fiy0             = _mm_add_ps(fiy0,ty);
1154             fiz0             = _mm_add_ps(fiz0,tz);
1155
1156             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1157                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1158                                                    tx,ty,tz);
1159
1160             }
1161
1162             /**************************
1163              * CALCULATE INTERACTIONS *
1164              **************************/
1165
1166             if (gmx_mm_any_lt(rsq10,rcutoff2))
1167             {
1168
1169             r10              = _mm_mul_ps(rsq10,rinv10);
1170             r10              = _mm_andnot_ps(dummy_mask,r10);
1171
1172             /* Compute parameters for interactions between i and j atoms */
1173             qq10             = _mm_mul_ps(iq1,jq0);
1174
1175             /* EWALD ELECTROSTATICS */
1176
1177             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1178             ewrt             = _mm_mul_ps(r10,ewtabscale);
1179             ewitab           = _mm_cvttps_epi32(ewrt);
1180             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1181             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1182                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1183                                          &ewtabF,&ewtabFn);
1184             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1185             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1186
1187             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1188
1189             fscal            = felec;
1190
1191             fscal            = _mm_and_ps(fscal,cutoff_mask);
1192
1193             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1194
1195             /* Calculate temporary vectorial force */
1196             tx               = _mm_mul_ps(fscal,dx10);
1197             ty               = _mm_mul_ps(fscal,dy10);
1198             tz               = _mm_mul_ps(fscal,dz10);
1199
1200             /* Update vectorial force */
1201             fix1             = _mm_add_ps(fix1,tx);
1202             fiy1             = _mm_add_ps(fiy1,ty);
1203             fiz1             = _mm_add_ps(fiz1,tz);
1204
1205             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1206                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1207                                                    tx,ty,tz);
1208
1209             }
1210
1211             /**************************
1212              * CALCULATE INTERACTIONS *
1213              **************************/
1214
1215             if (gmx_mm_any_lt(rsq20,rcutoff2))
1216             {
1217
1218             r20              = _mm_mul_ps(rsq20,rinv20);
1219             r20              = _mm_andnot_ps(dummy_mask,r20);
1220
1221             /* Compute parameters for interactions between i and j atoms */
1222             qq20             = _mm_mul_ps(iq2,jq0);
1223
1224             /* EWALD ELECTROSTATICS */
1225
1226             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1227             ewrt             = _mm_mul_ps(r20,ewtabscale);
1228             ewitab           = _mm_cvttps_epi32(ewrt);
1229             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1230             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1231                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1232                                          &ewtabF,&ewtabFn);
1233             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1234             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1235
1236             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1237
1238             fscal            = felec;
1239
1240             fscal            = _mm_and_ps(fscal,cutoff_mask);
1241
1242             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1243
1244             /* Calculate temporary vectorial force */
1245             tx               = _mm_mul_ps(fscal,dx20);
1246             ty               = _mm_mul_ps(fscal,dy20);
1247             tz               = _mm_mul_ps(fscal,dz20);
1248
1249             /* Update vectorial force */
1250             fix2             = _mm_add_ps(fix2,tx);
1251             fiy2             = _mm_add_ps(fiy2,ty);
1252             fiz2             = _mm_add_ps(fiz2,tz);
1253
1254             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1255                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1256                                                    tx,ty,tz);
1257
1258             }
1259
1260             /* Inner loop uses 127 flops */
1261         }
1262
1263         /* End of innermost loop */
1264
1265         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1266                                               f+i_coord_offset,fshift+i_shift_offset);
1267
1268         /* Increment number of inner iterations */
1269         inneriter                  += j_index_end - j_index_start;
1270
1271         /* Outer loop uses 27 flops */
1272     }
1273
1274     /* Increment number of outer iterations */
1275     outeriter        += nri;
1276
1277     /* Update outer/inner flops */
1278
1279     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*27 + inneriter*127);
1280 }