src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEwSw_VdwLJSw_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_ElecEwSw_VdwLJSw_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_ElecEwSw_VdwLJSw_GeomW3P1_VF_sse2_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwioffset1;
  72     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  73     int              vdwioffset2;
  74     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  75     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  76     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  77     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  78     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  79     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  80     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     int              nvdwtype;
  83     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  84     int              *vdwtype;
  85     real             *vdwparam;
  86     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  87     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  88     __m128i          ewitab;
  89     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  90     real             *ewtab;
  91     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  92     real             rswitch_scalar,d_scalar;
  93     __m128           dummy_mask,cutoff_mask;
  94     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  95     __m128           one     = _mm_set1_ps(1.0);
  96     __m128           two     = _mm_set1_ps(2.0);
  97     x                = xx[0];
  98     f                = ff[0];
  99
 100     nri              = nlist->nri;
 101     iinr             = nlist->iinr;
 102     jindex           = nlist->jindex;
 103     jjnr             = nlist->jjnr;
 104     shiftidx         = nlist->shift;
 105     gid              = nlist->gid;
 106     shiftvec         = fr->shift_vec[0];
 107     fshift           = fr->fshift[0];
 108     facel            = _mm_set1_ps(fr->epsfac);
 109     charge           = mdatoms->chargeA;
 110     nvdwtype         = fr->ntype;
 111     vdwparam         = fr->nbfp;
 112     vdwtype          = mdatoms->typeA;
 113
 114     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 115     ewtab            = fr->ic->tabq_coul_FDV0;
 116     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 117     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 118
 119     /* Setup water-specific parameters */
 120     inr              = nlist->iinr[0];
 121     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 122     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 123     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 124     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 125
 126     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 127     rcutoff_scalar   = fr->rcoulomb;
 128     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 129     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 130
 131     rswitch_scalar   = fr->rcoulomb_switch;
 132     rswitch          = _mm_set1_ps(rswitch_scalar);
 133     /* Setup switch parameters */
 134     d_scalar         = rcutoff_scalar-rswitch_scalar;
 135     d                = _mm_set1_ps(d_scalar);
 136     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 137     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 138     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 139     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 140     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 141     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 142
 143     /* Avoid stupid compiler warnings */
 144     jnrA = jnrB = jnrC = jnrD = 0;
 145     j_coord_offsetA = 0;
 146     j_coord_offsetB = 0;
 147     j_coord_offsetC = 0;
 148     j_coord_offsetD = 0;
 149
 150     outeriter        = 0;
 151     inneriter        = 0;
 152
 153     for(iidx=0;iidx<4*DIM;iidx++)
 154     {
 155         scratch[iidx] = 0.0;
 156     }
 157
 158     /* Start outer loop over neighborlists */
 159     for(iidx=0; iidx<nri; iidx++)
 160     {
 161         /* Load shift vector for this list */
 162         i_shift_offset   = DIM*shiftidx[iidx];
 163
 164         /* Load limits for loop over neighbors */
 165         j_index_start    = jindex[iidx];
 166         j_index_end      = jindex[iidx+1];
 167
 168         /* Get outer coordinate index */
 169         inr              = iinr[iidx];
 170         i_coord_offset   = DIM*inr;
 171
 172         /* Load i particle coords and add shift vector */
 173         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 174                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 175
 176         fix0             = _mm_setzero_ps();
 177         fiy0             = _mm_setzero_ps();
 178         fiz0             = _mm_setzero_ps();
 179         fix1             = _mm_setzero_ps();
 180         fiy1             = _mm_setzero_ps();
 181         fiz1             = _mm_setzero_ps();
 182         fix2             = _mm_setzero_ps();
 183         fiy2             = _mm_setzero_ps();
 184         fiz2             = _mm_setzero_ps();
 185
 186         /* Reset potential sums */
 187         velecsum         = _mm_setzero_ps();
 188         vvdwsum          = _mm_setzero_ps();
 189
 190         /* Start inner kernel loop */
 191         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 192         {
 193
 194             /* Get j neighbor index, and coordinate index */
 195             jnrA             = jjnr[jidx];
 196             jnrB             = jjnr[jidx+1];
 197             jnrC             = jjnr[jidx+2];
 198             jnrD             = jjnr[jidx+3];
 199             j_coord_offsetA  = DIM*jnrA;
 200             j_coord_offsetB  = DIM*jnrB;
 201             j_coord_offsetC  = DIM*jnrC;
 202             j_coord_offsetD  = DIM*jnrD;
 203
 204             /* load j atom coordinates */
 205             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 206                                               x+j_coord_offsetC,x+j_coord_offsetD,
 207                                               &jx0,&jy0,&jz0);
 208
 209             /* Calculate displacement vector */
 210             dx00             = _mm_sub_ps(ix0,jx0);
 211             dy00             = _mm_sub_ps(iy0,jy0);
 212             dz00             = _mm_sub_ps(iz0,jz0);
 213             dx10             = _mm_sub_ps(ix1,jx0);
 214             dy10             = _mm_sub_ps(iy1,jy0);
 215             dz10             = _mm_sub_ps(iz1,jz0);
 216             dx20             = _mm_sub_ps(ix2,jx0);
 217             dy20             = _mm_sub_ps(iy2,jy0);
 218             dz20             = _mm_sub_ps(iz2,jz0);
 219
 220             /* Calculate squared distance and things based on it */
 221             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 222             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 223             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 224
 225             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 226             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 227             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 228
 229             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 230             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 231             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 232
 233             /* Load parameters for j particles */
 234             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 235                                                               charge+jnrC+0,charge+jnrD+0);
 236             vdwjidx0A        = 2*vdwtype[jnrA+0];
 237             vdwjidx0B        = 2*vdwtype[jnrB+0];
 238             vdwjidx0C        = 2*vdwtype[jnrC+0];
 239             vdwjidx0D        = 2*vdwtype[jnrD+0];
 240
 241             fjx0             = _mm_setzero_ps();
 242             fjy0             = _mm_setzero_ps();
 243             fjz0             = _mm_setzero_ps();
 244
 245             /**************************
 246              * CALCULATE INTERACTIONS *
 247              **************************/
 248
 249             if (gmx_mm_any_lt(rsq00,rcutoff2))
 250             {
 251
 252             r00              = _mm_mul_ps(rsq00,rinv00);
 253
 254             /* Compute parameters for interactions between i and j atoms */
 255             qq00             = _mm_mul_ps(iq0,jq0);
 256             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 257                                          vdwparam+vdwioffset0+vdwjidx0B,
 258                                          vdwparam+vdwioffset0+vdwjidx0C,
 259                                          vdwparam+vdwioffset0+vdwjidx0D,
 260                                          &c6_00,&c12_00);
 261
 262             /* EWALD ELECTROSTATICS */
 263
 264             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 265             ewrt             = _mm_mul_ps(r00,ewtabscale);
 266             ewitab           = _mm_cvttps_epi32(ewrt);
 267             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 268             ewitab           = _mm_slli_epi32(ewitab,2);
 269             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 270             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 271             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 272             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 273             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 274             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 275             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 276             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 277             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 278
 279             /* LENNARD-JONES DISPERSION/REPULSION */
 280
 281             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 282             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 283             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 284             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 285             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 286
 287             d                = _mm_sub_ps(r00,rswitch);
 288             d                = _mm_max_ps(d,_mm_setzero_ps());
 289             d2               = _mm_mul_ps(d,d);
 290             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 291
 292             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 293
 294             /* Evaluate switch function */
 295             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 296             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
 297             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 298             velec            = _mm_mul_ps(velec,sw);
 299             vvdw             = _mm_mul_ps(vvdw,sw);
 300             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 301
 302             /* Update potential sum for this i atom from the interaction with this j atom. */
 303             velec            = _mm_and_ps(velec,cutoff_mask);
 304             velecsum         = _mm_add_ps(velecsum,velec);
 305             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 306             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 307
 308             fscal            = _mm_add_ps(felec,fvdw);
 309
 310             fscal            = _mm_and_ps(fscal,cutoff_mask);
 311
 312             /* Calculate temporary vectorial force */
 313             tx               = _mm_mul_ps(fscal,dx00);
 314             ty               = _mm_mul_ps(fscal,dy00);
 315             tz               = _mm_mul_ps(fscal,dz00);
 316
 317             /* Update vectorial force */
 318             fix0             = _mm_add_ps(fix0,tx);
 319             fiy0             = _mm_add_ps(fiy0,ty);
 320             fiz0             = _mm_add_ps(fiz0,tz);
 321
 322             fjx0             = _mm_add_ps(fjx0,tx);
 323             fjy0             = _mm_add_ps(fjy0,ty);
 324             fjz0             = _mm_add_ps(fjz0,tz);
 325
 326             }
 327
 328             /**************************
 329              * CALCULATE INTERACTIONS *
 330              **************************/
 331
 332             if (gmx_mm_any_lt(rsq10,rcutoff2))
 333             {
 334
 335             r10              = _mm_mul_ps(rsq10,rinv10);
 336
 337             /* Compute parameters for interactions between i and j atoms */
 338             qq10             = _mm_mul_ps(iq1,jq0);
 339
 340             /* EWALD ELECTROSTATICS */
 341
 342             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 343             ewrt             = _mm_mul_ps(r10,ewtabscale);
 344             ewitab           = _mm_cvttps_epi32(ewrt);
 345             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 346             ewitab           = _mm_slli_epi32(ewitab,2);
 347             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 348             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 349             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 350             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 351             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 352             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 353             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 354             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 355             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 356
 357             d                = _mm_sub_ps(r10,rswitch);
 358             d                = _mm_max_ps(d,_mm_setzero_ps());
 359             d2               = _mm_mul_ps(d,d);
 360             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 361
 362             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 363
 364             /* Evaluate switch function */
 365             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 366             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
 367             velec            = _mm_mul_ps(velec,sw);
 368             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 369
 370             /* Update potential sum for this i atom from the interaction with this j atom. */
 371             velec            = _mm_and_ps(velec,cutoff_mask);
 372             velecsum         = _mm_add_ps(velecsum,velec);
 373
 374             fscal            = felec;
 375
 376             fscal            = _mm_and_ps(fscal,cutoff_mask);
 377
 378             /* Calculate temporary vectorial force */
 379             tx               = _mm_mul_ps(fscal,dx10);
 380             ty               = _mm_mul_ps(fscal,dy10);
 381             tz               = _mm_mul_ps(fscal,dz10);
 382
 383             /* Update vectorial force */
 384             fix1             = _mm_add_ps(fix1,tx);
 385             fiy1             = _mm_add_ps(fiy1,ty);
 386             fiz1             = _mm_add_ps(fiz1,tz);
 387
 388             fjx0             = _mm_add_ps(fjx0,tx);
 389             fjy0             = _mm_add_ps(fjy0,ty);
 390             fjz0             = _mm_add_ps(fjz0,tz);
 391
 392             }
 393
 394             /**************************
 395              * CALCULATE INTERACTIONS *
 396              **************************/
 397
 398             if (gmx_mm_any_lt(rsq20,rcutoff2))
 399             {
 400
 401             r20              = _mm_mul_ps(rsq20,rinv20);
 402
 403             /* Compute parameters for interactions between i and j atoms */
 404             qq20             = _mm_mul_ps(iq2,jq0);
 405
 406             /* EWALD ELECTROSTATICS */
 407
 408             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 409             ewrt             = _mm_mul_ps(r20,ewtabscale);
 410             ewitab           = _mm_cvttps_epi32(ewrt);
 411             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 412             ewitab           = _mm_slli_epi32(ewitab,2);
 413             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 414             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 415             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 416             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 417             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 418             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 419             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 420             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 421             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 422
 423             d                = _mm_sub_ps(r20,rswitch);
 424             d                = _mm_max_ps(d,_mm_setzero_ps());
 425             d2               = _mm_mul_ps(d,d);
 426             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 427
 428             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 429
 430             /* Evaluate switch function */
 431             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 432             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
 433             velec            = _mm_mul_ps(velec,sw);
 434             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 435
 436             /* Update potential sum for this i atom from the interaction with this j atom. */
 437             velec            = _mm_and_ps(velec,cutoff_mask);
 438             velecsum         = _mm_add_ps(velecsum,velec);
 439
 440             fscal            = felec;
 441
 442             fscal            = _mm_and_ps(fscal,cutoff_mask);
 443
 444             /* Calculate temporary vectorial force */
 445             tx               = _mm_mul_ps(fscal,dx20);
 446             ty               = _mm_mul_ps(fscal,dy20);
 447             tz               = _mm_mul_ps(fscal,dz20);
 448
 449             /* Update vectorial force */
 450             fix2             = _mm_add_ps(fix2,tx);
 451             fiy2             = _mm_add_ps(fiy2,ty);
 452             fiz2             = _mm_add_ps(fiz2,tz);
 453
 454             fjx0             = _mm_add_ps(fjx0,tx);
 455             fjy0             = _mm_add_ps(fjy0,ty);
 456             fjz0             = _mm_add_ps(fjz0,tz);
 457
 458             }
 459
 460             fjptrA             = f+j_coord_offsetA;
 461             fjptrB             = f+j_coord_offsetB;
 462             fjptrC             = f+j_coord_offsetC;
 463             fjptrD             = f+j_coord_offsetD;
 464
 465             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 466
 467             /* Inner loop uses 213 flops */
 468         }
 469
 470         if(jidx<j_index_end)
 471         {
 472
 473             /* Get j neighbor index, and coordinate index */
 474             jnrlistA         = jjnr[jidx];
 475             jnrlistB         = jjnr[jidx+1];
 476             jnrlistC         = jjnr[jidx+2];
 477             jnrlistD         = jjnr[jidx+3];
 478             /* Sign of each element will be negative for non-real atoms.
 479              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 480              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 481              */
 482             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 483             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 484             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 485             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 486             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 487             j_coord_offsetA  = DIM*jnrA;
 488             j_coord_offsetB  = DIM*jnrB;
 489             j_coord_offsetC  = DIM*jnrC;
 490             j_coord_offsetD  = DIM*jnrD;
 491
 492             /* load j atom coordinates */
 493             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 494                                               x+j_coord_offsetC,x+j_coord_offsetD,
 495                                               &jx0,&jy0,&jz0);
 496
 497             /* Calculate displacement vector */
 498             dx00             = _mm_sub_ps(ix0,jx0);
 499             dy00             = _mm_sub_ps(iy0,jy0);
 500             dz00             = _mm_sub_ps(iz0,jz0);
 501             dx10             = _mm_sub_ps(ix1,jx0);
 502             dy10             = _mm_sub_ps(iy1,jy0);
 503             dz10             = _mm_sub_ps(iz1,jz0);
 504             dx20             = _mm_sub_ps(ix2,jx0);
 505             dy20             = _mm_sub_ps(iy2,jy0);
 506             dz20             = _mm_sub_ps(iz2,jz0);
 507
 508             /* Calculate squared distance and things based on it */
 509             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 510             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 511             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 512
 513             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 514             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 515             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 516
 517             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 518             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 519             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 520
 521             /* Load parameters for j particles */
 522             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 523                                                               charge+jnrC+0,charge+jnrD+0);
 524             vdwjidx0A        = 2*vdwtype[jnrA+0];
 525             vdwjidx0B        = 2*vdwtype[jnrB+0];
 526             vdwjidx0C        = 2*vdwtype[jnrC+0];
 527             vdwjidx0D        = 2*vdwtype[jnrD+0];
 528
 529             fjx0             = _mm_setzero_ps();
 530             fjy0             = _mm_setzero_ps();
 531             fjz0             = _mm_setzero_ps();
 532
 533             /**************************
 534              * CALCULATE INTERACTIONS *
 535              **************************/
 536
 537             if (gmx_mm_any_lt(rsq00,rcutoff2))
 538             {
 539
 540             r00              = _mm_mul_ps(rsq00,rinv00);
 541             r00              = _mm_andnot_ps(dummy_mask,r00);
 542
 543             /* Compute parameters for interactions between i and j atoms */
 544             qq00             = _mm_mul_ps(iq0,jq0);
 545             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 546                                          vdwparam+vdwioffset0+vdwjidx0B,
 547                                          vdwparam+vdwioffset0+vdwjidx0C,
 548                                          vdwparam+vdwioffset0+vdwjidx0D,
 549                                          &c6_00,&c12_00);
 550
 551             /* EWALD ELECTROSTATICS */
 552
 553             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 554             ewrt             = _mm_mul_ps(r00,ewtabscale);
 555             ewitab           = _mm_cvttps_epi32(ewrt);
 556             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 557             ewitab           = _mm_slli_epi32(ewitab,2);
 558             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 559             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 560             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 561             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 562             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 563             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 564             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 565             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 566             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 567
 568             /* LENNARD-JONES DISPERSION/REPULSION */
 569
 570             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 571             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 572             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 573             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 574             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 575
 576             d                = _mm_sub_ps(r00,rswitch);
 577             d                = _mm_max_ps(d,_mm_setzero_ps());
 578             d2               = _mm_mul_ps(d,d);
 579             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 580
 581             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 582
 583             /* Evaluate switch function */
 584             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 585             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
 586             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 587             velec            = _mm_mul_ps(velec,sw);
 588             vvdw             = _mm_mul_ps(vvdw,sw);
 589             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 590
 591             /* Update potential sum for this i atom from the interaction with this j atom. */
 592             velec            = _mm_and_ps(velec,cutoff_mask);
 593             velec            = _mm_andnot_ps(dummy_mask,velec);
 594             velecsum         = _mm_add_ps(velecsum,velec);
 595             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 596             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 597             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 598
 599             fscal            = _mm_add_ps(felec,fvdw);
 600
 601             fscal            = _mm_and_ps(fscal,cutoff_mask);
 602
 603             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 604
 605             /* Calculate temporary vectorial force */
 606             tx               = _mm_mul_ps(fscal,dx00);
 607             ty               = _mm_mul_ps(fscal,dy00);
 608             tz               = _mm_mul_ps(fscal,dz00);
 609
 610             /* Update vectorial force */
 611             fix0             = _mm_add_ps(fix0,tx);
 612             fiy0             = _mm_add_ps(fiy0,ty);
 613             fiz0             = _mm_add_ps(fiz0,tz);
 614
 615             fjx0             = _mm_add_ps(fjx0,tx);
 616             fjy0             = _mm_add_ps(fjy0,ty);
 617             fjz0             = _mm_add_ps(fjz0,tz);
 618
 619             }
 620
 621             /**************************
 622              * CALCULATE INTERACTIONS *
 623              **************************/
 624
 625             if (gmx_mm_any_lt(rsq10,rcutoff2))
 626             {
 627
 628             r10              = _mm_mul_ps(rsq10,rinv10);
 629             r10              = _mm_andnot_ps(dummy_mask,r10);
 630
 631             /* Compute parameters for interactions between i and j atoms */
 632             qq10             = _mm_mul_ps(iq1,jq0);
 633
 634             /* EWALD ELECTROSTATICS */
 635
 636             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 637             ewrt             = _mm_mul_ps(r10,ewtabscale);
 638             ewitab           = _mm_cvttps_epi32(ewrt);
 639             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 640             ewitab           = _mm_slli_epi32(ewitab,2);
 641             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 642             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 643             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 644             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 645             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 646             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 647             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 648             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 649             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 650
 651             d                = _mm_sub_ps(r10,rswitch);
 652             d                = _mm_max_ps(d,_mm_setzero_ps());
 653             d2               = _mm_mul_ps(d,d);
 654             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 655
 656             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 657
 658             /* Evaluate switch function */
 659             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 660             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
 661             velec            = _mm_mul_ps(velec,sw);
 662             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 663
 664             /* Update potential sum for this i atom from the interaction with this j atom. */
 665             velec            = _mm_and_ps(velec,cutoff_mask);
 666             velec            = _mm_andnot_ps(dummy_mask,velec);
 667             velecsum         = _mm_add_ps(velecsum,velec);
 668
 669             fscal            = felec;
 670
 671             fscal            = _mm_and_ps(fscal,cutoff_mask);
 672
 673             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 674
 675             /* Calculate temporary vectorial force */
 676             tx               = _mm_mul_ps(fscal,dx10);
 677             ty               = _mm_mul_ps(fscal,dy10);
 678             tz               = _mm_mul_ps(fscal,dz10);
 679
 680             /* Update vectorial force */
 681             fix1             = _mm_add_ps(fix1,tx);
 682             fiy1             = _mm_add_ps(fiy1,ty);
 683             fiz1             = _mm_add_ps(fiz1,tz);
 684
 685             fjx0             = _mm_add_ps(fjx0,tx);
 686             fjy0             = _mm_add_ps(fjy0,ty);
 687             fjz0             = _mm_add_ps(fjz0,tz);
 688
 689             }
 690
 691             /**************************
 692              * CALCULATE INTERACTIONS *
 693              **************************/
 694
 695             if (gmx_mm_any_lt(rsq20,rcutoff2))
 696             {
 697
 698             r20              = _mm_mul_ps(rsq20,rinv20);
 699             r20              = _mm_andnot_ps(dummy_mask,r20);
 700
 701             /* Compute parameters for interactions between i and j atoms */
 702             qq20             = _mm_mul_ps(iq2,jq0);
 703
 704             /* EWALD ELECTROSTATICS */
 705
 706             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 707             ewrt             = _mm_mul_ps(r20,ewtabscale);
 708             ewitab           = _mm_cvttps_epi32(ewrt);
 709             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 710             ewitab           = _mm_slli_epi32(ewitab,2);
 711             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 712             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 713             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 714             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 715             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 716             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 717             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 718             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 719             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 720
 721             d                = _mm_sub_ps(r20,rswitch);
 722             d                = _mm_max_ps(d,_mm_setzero_ps());
 723             d2               = _mm_mul_ps(d,d);
 724             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 725
 726             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 727
 728             /* Evaluate switch function */
 729             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 730             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
 731             velec            = _mm_mul_ps(velec,sw);
 732             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 733
 734             /* Update potential sum for this i atom from the interaction with this j atom. */
 735             velec            = _mm_and_ps(velec,cutoff_mask);
 736             velec            = _mm_andnot_ps(dummy_mask,velec);
 737             velecsum         = _mm_add_ps(velecsum,velec);
 738
 739             fscal            = felec;
 740
 741             fscal            = _mm_and_ps(fscal,cutoff_mask);
 742
 743             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 744
 745             /* Calculate temporary vectorial force */
 746             tx               = _mm_mul_ps(fscal,dx20);
 747             ty               = _mm_mul_ps(fscal,dy20);
 748             tz               = _mm_mul_ps(fscal,dz20);
 749
 750             /* Update vectorial force */
 751             fix2             = _mm_add_ps(fix2,tx);
 752             fiy2             = _mm_add_ps(fiy2,ty);
 753             fiz2             = _mm_add_ps(fiz2,tz);
 754
 755             fjx0             = _mm_add_ps(fjx0,tx);
 756             fjy0             = _mm_add_ps(fjy0,ty);
 757             fjz0             = _mm_add_ps(fjz0,tz);
 758
 759             }
 760
 761             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 762             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 763             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 764             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 765
 766             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 767
 768             /* Inner loop uses 216 flops */
 769         }
 770
 771         /* End of innermost loop */
 772
 773         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 774                                               f+i_coord_offset,fshift+i_shift_offset);
 775
 776         ggid                        = gid[iidx];
 777         /* Update potential energies */
 778         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 779         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 780
 781         /* Increment number of inner iterations */
 782         inneriter                  += j_index_end - j_index_start;
 783
 784         /* Outer loop uses 20 flops */
 785     }
 786
 787     /* Increment number of outer iterations */
 788     outeriter        += nri;
 789
 790     /* Update outer/inner flops */
 791
 792     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*216);
 793 }
 794 /*
 795  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_F_sse2_single
 796  * Electrostatics interaction: Ewald
 797  * VdW interaction:            LennardJones
 798  * Geometry:                   Water3-Particle
 799  * Calculate force/pot:        Force
 800  */
 801 void
 802 nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_F_sse2_single
 803                     (t_nblist * gmx_restrict                nlist,
 804                      rvec * gmx_restrict                    xx,
 805                      rvec * gmx_restrict                    ff,
 806                      t_forcerec * gmx_restrict              fr,
 807                      t_mdatoms * gmx_restrict               mdatoms,
 808                      nb_kernel_data_t * gmx_restrict        kernel_data,
 809                      t_nrnb * gmx_restrict                  nrnb)
 810 {
 811     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 812      * just 0 for non-waters.
 813      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 814      * jnr indices corresponding to data put in the four positions in the SIMD register.
 815      */
 816     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 817     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 818     int              jnrA,jnrB,jnrC,jnrD;
 819     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 820     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 821     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 822     real             rcutoff_scalar;
 823     real             *shiftvec,*fshift,*x,*f;
 824     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 825     real             scratch[4*DIM];
 826     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 827     int              vdwioffset0;
 828     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 829     int              vdwioffset1;
 830     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 831     int              vdwioffset2;
 832     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 833     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 834     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 835     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 836     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 837     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 838     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 839     real             *charge;
 840     int              nvdwtype;
 841     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 842     int              *vdwtype;
 843     real             *vdwparam;
 844     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 845     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 846     __m128i          ewitab;
 847     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 848     real             *ewtab;
 849     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 850     real             rswitch_scalar,d_scalar;
 851     __m128           dummy_mask,cutoff_mask;
 852     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 853     __m128           one     = _mm_set1_ps(1.0);
 854     __m128           two     = _mm_set1_ps(2.0);
 855     x                = xx[0];
 856     f                = ff[0];
 857
 858     nri              = nlist->nri;
 859     iinr             = nlist->iinr;
 860     jindex           = nlist->jindex;
 861     jjnr             = nlist->jjnr;
 862     shiftidx         = nlist->shift;
 863     gid              = nlist->gid;
 864     shiftvec         = fr->shift_vec[0];
 865     fshift           = fr->fshift[0];
 866     facel            = _mm_set1_ps(fr->epsfac);
 867     charge           = mdatoms->chargeA;
 868     nvdwtype         = fr->ntype;
 869     vdwparam         = fr->nbfp;
 870     vdwtype          = mdatoms->typeA;
 871
 872     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 873     ewtab            = fr->ic->tabq_coul_FDV0;
 874     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 875     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 876
 877     /* Setup water-specific parameters */
 878     inr              = nlist->iinr[0];
 879     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 880     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 881     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 882     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 883
 884     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 885     rcutoff_scalar   = fr->rcoulomb;
 886     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 887     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 888
 889     rswitch_scalar   = fr->rcoulomb_switch;
 890     rswitch          = _mm_set1_ps(rswitch_scalar);
 891     /* Setup switch parameters */
 892     d_scalar         = rcutoff_scalar-rswitch_scalar;
 893     d                = _mm_set1_ps(d_scalar);
 894     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 895     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 896     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 897     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 898     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 899     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 900
 901     /* Avoid stupid compiler warnings */
 902     jnrA = jnrB = jnrC = jnrD = 0;
 903     j_coord_offsetA = 0;
 904     j_coord_offsetB = 0;
 905     j_coord_offsetC = 0;
 906     j_coord_offsetD = 0;
 907
 908     outeriter        = 0;
 909     inneriter        = 0;
 910
 911     for(iidx=0;iidx<4*DIM;iidx++)
 912     {
 913         scratch[iidx] = 0.0;
 914     }
 915
 916     /* Start outer loop over neighborlists */
 917     for(iidx=0; iidx<nri; iidx++)
 918     {
 919         /* Load shift vector for this list */
 920         i_shift_offset   = DIM*shiftidx[iidx];
 921
 922         /* Load limits for loop over neighbors */
 923         j_index_start    = jindex[iidx];
 924         j_index_end      = jindex[iidx+1];
 925
 926         /* Get outer coordinate index */
 927         inr              = iinr[iidx];
 928         i_coord_offset   = DIM*inr;
 929
 930         /* Load i particle coords and add shift vector */
 931         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 932                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 933
 934         fix0             = _mm_setzero_ps();
 935         fiy0             = _mm_setzero_ps();
 936         fiz0             = _mm_setzero_ps();
 937         fix1             = _mm_setzero_ps();
 938         fiy1             = _mm_setzero_ps();
 939         fiz1             = _mm_setzero_ps();
 940         fix2             = _mm_setzero_ps();
 941         fiy2             = _mm_setzero_ps();
 942         fiz2             = _mm_setzero_ps();
 943
 944         /* Start inner kernel loop */
 945         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 946         {
 947
 948             /* Get j neighbor index, and coordinate index */
 949             jnrA             = jjnr[jidx];
 950             jnrB             = jjnr[jidx+1];
 951             jnrC             = jjnr[jidx+2];
 952             jnrD             = jjnr[jidx+3];
 953             j_coord_offsetA  = DIM*jnrA;
 954             j_coord_offsetB  = DIM*jnrB;
 955             j_coord_offsetC  = DIM*jnrC;
 956             j_coord_offsetD  = DIM*jnrD;
 957
 958             /* load j atom coordinates */
 959             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 960                                               x+j_coord_offsetC,x+j_coord_offsetD,
 961                                               &jx0,&jy0,&jz0);
 962
 963             /* Calculate displacement vector */
 964             dx00             = _mm_sub_ps(ix0,jx0);
 965             dy00             = _mm_sub_ps(iy0,jy0);
 966             dz00             = _mm_sub_ps(iz0,jz0);
 967             dx10             = _mm_sub_ps(ix1,jx0);
 968             dy10             = _mm_sub_ps(iy1,jy0);
 969             dz10             = _mm_sub_ps(iz1,jz0);
 970             dx20             = _mm_sub_ps(ix2,jx0);
 971             dy20             = _mm_sub_ps(iy2,jy0);
 972             dz20             = _mm_sub_ps(iz2,jz0);
 973
 974             /* Calculate squared distance and things based on it */
 975             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 976             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 977             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 978
 979             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 980             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 981             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 982
 983             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 984             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 985             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 986
 987             /* Load parameters for j particles */
 988             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 989                                                               charge+jnrC+0,charge+jnrD+0);
 990             vdwjidx0A        = 2*vdwtype[jnrA+0];
 991             vdwjidx0B        = 2*vdwtype[jnrB+0];
 992             vdwjidx0C        = 2*vdwtype[jnrC+0];
 993             vdwjidx0D        = 2*vdwtype[jnrD+0];
 994
 995             fjx0             = _mm_setzero_ps();
 996             fjy0             = _mm_setzero_ps();
 997             fjz0             = _mm_setzero_ps();
 998
 999             /**************************
1000              * CALCULATE INTERACTIONS *
1001              **************************/
1002
1003             if (gmx_mm_any_lt(rsq00,rcutoff2))
1004             {
1005
1006             r00              = _mm_mul_ps(rsq00,rinv00);
1007
1008             /* Compute parameters for interactions between i and j atoms */
1009             qq00             = _mm_mul_ps(iq0,jq0);
1010             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1011                                          vdwparam+vdwioffset0+vdwjidx0B,
1012                                          vdwparam+vdwioffset0+vdwjidx0C,
1013                                          vdwparam+vdwioffset0+vdwjidx0D,
1014                                          &c6_00,&c12_00);
1015
1016             /* EWALD ELECTROSTATICS */
1017
1018             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1019             ewrt             = _mm_mul_ps(r00,ewtabscale);
1020             ewitab           = _mm_cvttps_epi32(ewrt);
1021             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1022             ewitab           = _mm_slli_epi32(ewitab,2);
1023             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1024             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1025             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1026             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1027             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1028             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1029             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1030             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
1031             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1032
1033             /* LENNARD-JONES DISPERSION/REPULSION */
1034
1035             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1036             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
1037             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
1038             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
1039             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
1040
1041             d                = _mm_sub_ps(r00,rswitch);
1042             d                = _mm_max_ps(d,_mm_setzero_ps());
1043             d2               = _mm_mul_ps(d,d);
1044             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1045
1046             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1047
1048             /* Evaluate switch function */
1049             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1050             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
1051             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1052             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1053
1054             fscal            = _mm_add_ps(felec,fvdw);
1055
1056             fscal            = _mm_and_ps(fscal,cutoff_mask);
1057
1058             /* Calculate temporary vectorial force */
1059             tx               = _mm_mul_ps(fscal,dx00);
1060             ty               = _mm_mul_ps(fscal,dy00);
1061             tz               = _mm_mul_ps(fscal,dz00);
1062
1063             /* Update vectorial force */
1064             fix0             = _mm_add_ps(fix0,tx);
1065             fiy0             = _mm_add_ps(fiy0,ty);
1066             fiz0             = _mm_add_ps(fiz0,tz);
1067
1068             fjx0             = _mm_add_ps(fjx0,tx);
1069             fjy0             = _mm_add_ps(fjy0,ty);
1070             fjz0             = _mm_add_ps(fjz0,tz);
1071
1072             }
1073
1074             /**************************
1075              * CALCULATE INTERACTIONS *
1076              **************************/
1077
1078             if (gmx_mm_any_lt(rsq10,rcutoff2))
1079             {
1080
1081             r10              = _mm_mul_ps(rsq10,rinv10);
1082
1083             /* Compute parameters for interactions between i and j atoms */
1084             qq10             = _mm_mul_ps(iq1,jq0);
1085
1086             /* EWALD ELECTROSTATICS */
1087
1088             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1089             ewrt             = _mm_mul_ps(r10,ewtabscale);
1090             ewitab           = _mm_cvttps_epi32(ewrt);
1091             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1092             ewitab           = _mm_slli_epi32(ewitab,2);
1093             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1094             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1095             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1096             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1097             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1098             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1099             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1100             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
1101             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1102
1103             d                = _mm_sub_ps(r10,rswitch);
1104             d                = _mm_max_ps(d,_mm_setzero_ps());
1105             d2               = _mm_mul_ps(d,d);
1106             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1107
1108             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1109
1110             /* Evaluate switch function */
1111             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1112             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
1113             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1114
1115             fscal            = felec;
1116
1117             fscal            = _mm_and_ps(fscal,cutoff_mask);
1118
1119             /* Calculate temporary vectorial force */
1120             tx               = _mm_mul_ps(fscal,dx10);
1121             ty               = _mm_mul_ps(fscal,dy10);
1122             tz               = _mm_mul_ps(fscal,dz10);
1123
1124             /* Update vectorial force */
1125             fix1             = _mm_add_ps(fix1,tx);
1126             fiy1             = _mm_add_ps(fiy1,ty);
1127             fiz1             = _mm_add_ps(fiz1,tz);
1128
1129             fjx0             = _mm_add_ps(fjx0,tx);
1130             fjy0             = _mm_add_ps(fjy0,ty);
1131             fjz0             = _mm_add_ps(fjz0,tz);
1132
1133             }
1134
1135             /**************************
1136              * CALCULATE INTERACTIONS *
1137              **************************/
1138
1139             if (gmx_mm_any_lt(rsq20,rcutoff2))
1140             {
1141
1142             r20              = _mm_mul_ps(rsq20,rinv20);
1143
1144             /* Compute parameters for interactions between i and j atoms */
1145             qq20             = _mm_mul_ps(iq2,jq0);
1146
1147             /* EWALD ELECTROSTATICS */
1148
1149             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1150             ewrt             = _mm_mul_ps(r20,ewtabscale);
1151             ewitab           = _mm_cvttps_epi32(ewrt);
1152             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1153             ewitab           = _mm_slli_epi32(ewitab,2);
1154             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1155             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1156             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1157             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1158             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1159             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1160             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1161             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
1162             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1163
1164             d                = _mm_sub_ps(r20,rswitch);
1165             d                = _mm_max_ps(d,_mm_setzero_ps());
1166             d2               = _mm_mul_ps(d,d);
1167             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1168
1169             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1170
1171             /* Evaluate switch function */
1172             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1173             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
1174             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1175
1176             fscal            = felec;
1177
1178             fscal            = _mm_and_ps(fscal,cutoff_mask);
1179
1180             /* Calculate temporary vectorial force */
1181             tx               = _mm_mul_ps(fscal,dx20);
1182             ty               = _mm_mul_ps(fscal,dy20);
1183             tz               = _mm_mul_ps(fscal,dz20);
1184
1185             /* Update vectorial force */
1186             fix2             = _mm_add_ps(fix2,tx);
1187             fiy2             = _mm_add_ps(fiy2,ty);
1188             fiz2             = _mm_add_ps(fiz2,tz);
1189
1190             fjx0             = _mm_add_ps(fjx0,tx);
1191             fjy0             = _mm_add_ps(fjy0,ty);
1192             fjz0             = _mm_add_ps(fjz0,tz);
1193
1194             }
1195
1196             fjptrA             = f+j_coord_offsetA;
1197             fjptrB             = f+j_coord_offsetB;
1198             fjptrC             = f+j_coord_offsetC;
1199             fjptrD             = f+j_coord_offsetD;
1200
1201             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1202
1203             /* Inner loop uses 201 flops */
1204         }
1205
1206         if(jidx<j_index_end)
1207         {
1208
1209             /* Get j neighbor index, and coordinate index */
1210             jnrlistA         = jjnr[jidx];
1211             jnrlistB         = jjnr[jidx+1];
1212             jnrlistC         = jjnr[jidx+2];
1213             jnrlistD         = jjnr[jidx+3];
1214             /* Sign of each element will be negative for non-real atoms.
1215              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1216              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1217              */
1218             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1219             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1220             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1221             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1222             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1223             j_coord_offsetA  = DIM*jnrA;
1224             j_coord_offsetB  = DIM*jnrB;
1225             j_coord_offsetC  = DIM*jnrC;
1226             j_coord_offsetD  = DIM*jnrD;
1227
1228             /* load j atom coordinates */
1229             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1230                                               x+j_coord_offsetC,x+j_coord_offsetD,
1231                                               &jx0,&jy0,&jz0);
1232
1233             /* Calculate displacement vector */
1234             dx00             = _mm_sub_ps(ix0,jx0);
1235             dy00             = _mm_sub_ps(iy0,jy0);
1236             dz00             = _mm_sub_ps(iz0,jz0);
1237             dx10             = _mm_sub_ps(ix1,jx0);
1238             dy10             = _mm_sub_ps(iy1,jy0);
1239             dz10             = _mm_sub_ps(iz1,jz0);
1240             dx20             = _mm_sub_ps(ix2,jx0);
1241             dy20             = _mm_sub_ps(iy2,jy0);
1242             dz20             = _mm_sub_ps(iz2,jz0);
1243
1244             /* Calculate squared distance and things based on it */
1245             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1246             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1247             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1248
1249             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1250             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1251             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1252
1253             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1254             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1255             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1256
1257             /* Load parameters for j particles */
1258             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1259                                                               charge+jnrC+0,charge+jnrD+0);
1260             vdwjidx0A        = 2*vdwtype[jnrA+0];
1261             vdwjidx0B        = 2*vdwtype[jnrB+0];
1262             vdwjidx0C        = 2*vdwtype[jnrC+0];
1263             vdwjidx0D        = 2*vdwtype[jnrD+0];
1264
1265             fjx0             = _mm_setzero_ps();
1266             fjy0             = _mm_setzero_ps();
1267             fjz0             = _mm_setzero_ps();
1268
1269             /**************************
1270              * CALCULATE INTERACTIONS *
1271              **************************/
1272
1273             if (gmx_mm_any_lt(rsq00,rcutoff2))
1274             {
1275
1276             r00              = _mm_mul_ps(rsq00,rinv00);
1277             r00              = _mm_andnot_ps(dummy_mask,r00);
1278
1279             /* Compute parameters for interactions between i and j atoms */
1280             qq00             = _mm_mul_ps(iq0,jq0);
1281             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1282                                          vdwparam+vdwioffset0+vdwjidx0B,
1283                                          vdwparam+vdwioffset0+vdwjidx0C,
1284                                          vdwparam+vdwioffset0+vdwjidx0D,
1285                                          &c6_00,&c12_00);
1286
1287             /* EWALD ELECTROSTATICS */
1288
1289             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1290             ewrt             = _mm_mul_ps(r00,ewtabscale);
1291             ewitab           = _mm_cvttps_epi32(ewrt);
1292             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1293             ewitab           = _mm_slli_epi32(ewitab,2);
1294             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1295             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1296             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1297             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1298             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1299             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1300             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1301             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
1302             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1303
1304             /* LENNARD-JONES DISPERSION/REPULSION */
1305
1306             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1307             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
1308             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
1309             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
1310             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
1311
1312             d                = _mm_sub_ps(r00,rswitch);
1313             d                = _mm_max_ps(d,_mm_setzero_ps());
1314             d2               = _mm_mul_ps(d,d);
1315             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1316
1317             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1318
1319             /* Evaluate switch function */
1320             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1321             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
1322             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1323             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1324
1325             fscal            = _mm_add_ps(felec,fvdw);
1326
1327             fscal            = _mm_and_ps(fscal,cutoff_mask);
1328
1329             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1330
1331             /* Calculate temporary vectorial force */
1332             tx               = _mm_mul_ps(fscal,dx00);
1333             ty               = _mm_mul_ps(fscal,dy00);
1334             tz               = _mm_mul_ps(fscal,dz00);
1335
1336             /* Update vectorial force */
1337             fix0             = _mm_add_ps(fix0,tx);
1338             fiy0             = _mm_add_ps(fiy0,ty);
1339             fiz0             = _mm_add_ps(fiz0,tz);
1340
1341             fjx0             = _mm_add_ps(fjx0,tx);
1342             fjy0             = _mm_add_ps(fjy0,ty);
1343             fjz0             = _mm_add_ps(fjz0,tz);
1344
1345             }
1346
1347             /**************************
1348              * CALCULATE INTERACTIONS *
1349              **************************/
1350
1351             if (gmx_mm_any_lt(rsq10,rcutoff2))
1352             {
1353
1354             r10              = _mm_mul_ps(rsq10,rinv10);
1355             r10              = _mm_andnot_ps(dummy_mask,r10);
1356
1357             /* Compute parameters for interactions between i and j atoms */
1358             qq10             = _mm_mul_ps(iq1,jq0);
1359
1360             /* EWALD ELECTROSTATICS */
1361
1362             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1363             ewrt             = _mm_mul_ps(r10,ewtabscale);
1364             ewitab           = _mm_cvttps_epi32(ewrt);
1365             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1366             ewitab           = _mm_slli_epi32(ewitab,2);
1367             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1368             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1369             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1370             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1371             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1372             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1373             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1374             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
1375             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1376
1377             d                = _mm_sub_ps(r10,rswitch);
1378             d                = _mm_max_ps(d,_mm_setzero_ps());
1379             d2               = _mm_mul_ps(d,d);
1380             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1381
1382             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1383
1384             /* Evaluate switch function */
1385             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1386             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
1387             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1388
1389             fscal            = felec;
1390
1391             fscal            = _mm_and_ps(fscal,cutoff_mask);
1392
1393             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1394
1395             /* Calculate temporary vectorial force */
1396             tx               = _mm_mul_ps(fscal,dx10);
1397             ty               = _mm_mul_ps(fscal,dy10);
1398             tz               = _mm_mul_ps(fscal,dz10);
1399
1400             /* Update vectorial force */
1401             fix1             = _mm_add_ps(fix1,tx);
1402             fiy1             = _mm_add_ps(fiy1,ty);
1403             fiz1             = _mm_add_ps(fiz1,tz);
1404
1405             fjx0             = _mm_add_ps(fjx0,tx);
1406             fjy0             = _mm_add_ps(fjy0,ty);
1407             fjz0             = _mm_add_ps(fjz0,tz);
1408
1409             }
1410
1411             /**************************
1412              * CALCULATE INTERACTIONS *
1413              **************************/
1414
1415             if (gmx_mm_any_lt(rsq20,rcutoff2))
1416             {
1417
1418             r20              = _mm_mul_ps(rsq20,rinv20);
1419             r20              = _mm_andnot_ps(dummy_mask,r20);
1420
1421             /* Compute parameters for interactions between i and j atoms */
1422             qq20             = _mm_mul_ps(iq2,jq0);
1423
1424             /* EWALD ELECTROSTATICS */
1425
1426             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1427             ewrt             = _mm_mul_ps(r20,ewtabscale);
1428             ewitab           = _mm_cvttps_epi32(ewrt);
1429             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1430             ewitab           = _mm_slli_epi32(ewitab,2);
1431             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1432             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1433             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1434             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1435             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1436             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1437             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1438             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
1439             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1440
1441             d                = _mm_sub_ps(r20,rswitch);
1442             d                = _mm_max_ps(d,_mm_setzero_ps());
1443             d2               = _mm_mul_ps(d,d);
1444             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1445
1446             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1447
1448             /* Evaluate switch function */
1449             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1450             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
1451             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1452
1453             fscal            = felec;
1454
1455             fscal            = _mm_and_ps(fscal,cutoff_mask);
1456
1457             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1458
1459             /* Calculate temporary vectorial force */
1460             tx               = _mm_mul_ps(fscal,dx20);
1461             ty               = _mm_mul_ps(fscal,dy20);
1462             tz               = _mm_mul_ps(fscal,dz20);
1463
1464             /* Update vectorial force */
1465             fix2             = _mm_add_ps(fix2,tx);
1466             fiy2             = _mm_add_ps(fiy2,ty);
1467             fiz2             = _mm_add_ps(fiz2,tz);
1468
1469             fjx0             = _mm_add_ps(fjx0,tx);
1470             fjy0             = _mm_add_ps(fjy0,ty);
1471             fjz0             = _mm_add_ps(fjz0,tz);
1472
1473             }
1474
1475             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1476             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1477             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1478             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1479
1480             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1481
1482             /* Inner loop uses 204 flops */
1483         }
1484
1485         /* End of innermost loop */
1486
1487         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1488                                               f+i_coord_offset,fshift+i_shift_offset);
1489
1490         /* Increment number of inner iterations */
1491         inneriter                  += j_index_end - j_index_start;
1492
1493         /* Outer loop uses 18 flops */
1494     }
1495
1496     /* Increment number of outer iterations */
1497     outeriter        += nri;
1498
1499     /* Update outer/inner flops */
1500
1501     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*204);
1502 }