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

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse2_single.h"
  48 #include "kernelutil_x86_sse2_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_VF_sse2_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_VF_sse2_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwioffset1;
  86     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     int              vdwioffset2;
  88     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwioffset3;
  90     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  97     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     int              nvdwtype;
 100     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 101     int              *vdwtype;
 102     real             *vdwparam;
 103     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 104     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 105     __m128i          ewitab;
 106     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 107     real             *ewtab;
 108     __m128           dummy_mask,cutoff_mask;
 109     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 110     __m128           one     = _mm_set1_ps(1.0);
 111     __m128           two     = _mm_set1_ps(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm_set1_ps(fr->epsfac);
 124     charge           = mdatoms->chargeA;
 125     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128
 129     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 130     ewtab            = fr->ic->tabq_coul_FDV0;
 131     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 132     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 133
 134     /* Setup water-specific parameters */
 135     inr              = nlist->iinr[0];
 136     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 137     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 138     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 139     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 140
 141     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 142     rcutoff_scalar   = fr->rcoulomb;
 143     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 144     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 145
 146     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 147     rvdw             = _mm_set1_ps(fr->rvdw);
 148
 149     /* Avoid stupid compiler warnings */
 150     jnrA = jnrB = jnrC = jnrD = 0;
 151     j_coord_offsetA = 0;
 152     j_coord_offsetB = 0;
 153     j_coord_offsetC = 0;
 154     j_coord_offsetD = 0;
 155
 156     outeriter        = 0;
 157     inneriter        = 0;
 158
 159     for(iidx=0;iidx<4*DIM;iidx++)
 160     {
 161         scratch[iidx] = 0.0;
 162     }
 163
 164     /* Start outer loop over neighborlists */
 165     for(iidx=0; iidx<nri; iidx++)
 166     {
 167         /* Load shift vector for this list */
 168         i_shift_offset   = DIM*shiftidx[iidx];
 169
 170         /* Load limits for loop over neighbors */
 171         j_index_start    = jindex[iidx];
 172         j_index_end      = jindex[iidx+1];
 173
 174         /* Get outer coordinate index */
 175         inr              = iinr[iidx];
 176         i_coord_offset   = DIM*inr;
 177
 178         /* Load i particle coords and add shift vector */
 179         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 180                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 181
 182         fix0             = _mm_setzero_ps();
 183         fiy0             = _mm_setzero_ps();
 184         fiz0             = _mm_setzero_ps();
 185         fix1             = _mm_setzero_ps();
 186         fiy1             = _mm_setzero_ps();
 187         fiz1             = _mm_setzero_ps();
 188         fix2             = _mm_setzero_ps();
 189         fiy2             = _mm_setzero_ps();
 190         fiz2             = _mm_setzero_ps();
 191         fix3             = _mm_setzero_ps();
 192         fiy3             = _mm_setzero_ps();
 193         fiz3             = _mm_setzero_ps();
 194
 195         /* Reset potential sums */
 196         velecsum         = _mm_setzero_ps();
 197         vvdwsum          = _mm_setzero_ps();
 198
 199         /* Start inner kernel loop */
 200         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 201         {
 202
 203             /* Get j neighbor index, and coordinate index */
 204             jnrA             = jjnr[jidx];
 205             jnrB             = jjnr[jidx+1];
 206             jnrC             = jjnr[jidx+2];
 207             jnrD             = jjnr[jidx+3];
 208             j_coord_offsetA  = DIM*jnrA;
 209             j_coord_offsetB  = DIM*jnrB;
 210             j_coord_offsetC  = DIM*jnrC;
 211             j_coord_offsetD  = DIM*jnrD;
 212
 213             /* load j atom coordinates */
 214             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 215                                               x+j_coord_offsetC,x+j_coord_offsetD,
 216                                               &jx0,&jy0,&jz0);
 217
 218             /* Calculate displacement vector */
 219             dx00             = _mm_sub_ps(ix0,jx0);
 220             dy00             = _mm_sub_ps(iy0,jy0);
 221             dz00             = _mm_sub_ps(iz0,jz0);
 222             dx10             = _mm_sub_ps(ix1,jx0);
 223             dy10             = _mm_sub_ps(iy1,jy0);
 224             dz10             = _mm_sub_ps(iz1,jz0);
 225             dx20             = _mm_sub_ps(ix2,jx0);
 226             dy20             = _mm_sub_ps(iy2,jy0);
 227             dz20             = _mm_sub_ps(iz2,jz0);
 228             dx30             = _mm_sub_ps(ix3,jx0);
 229             dy30             = _mm_sub_ps(iy3,jy0);
 230             dz30             = _mm_sub_ps(iz3,jz0);
 231
 232             /* Calculate squared distance and things based on it */
 233             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 234             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 235             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 236             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 237
 238             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 239             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 240             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 241
 242             rinvsq00         = gmx_mm_inv_ps(rsq00);
 243             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 244             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 245             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 246
 247             /* Load parameters for j particles */
 248             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 249                                                               charge+jnrC+0,charge+jnrD+0);
 250             vdwjidx0A        = 2*vdwtype[jnrA+0];
 251             vdwjidx0B        = 2*vdwtype[jnrB+0];
 252             vdwjidx0C        = 2*vdwtype[jnrC+0];
 253             vdwjidx0D        = 2*vdwtype[jnrD+0];
 254
 255             fjx0             = _mm_setzero_ps();
 256             fjy0             = _mm_setzero_ps();
 257             fjz0             = _mm_setzero_ps();
 258
 259             /**************************
 260              * CALCULATE INTERACTIONS *
 261              **************************/
 262
 263             if (gmx_mm_any_lt(rsq00,rcutoff2))
 264             {
 265
 266             /* Compute parameters for interactions between i and j atoms */
 267             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 268                                          vdwparam+vdwioffset0+vdwjidx0B,
 269                                          vdwparam+vdwioffset0+vdwjidx0C,
 270                                          vdwparam+vdwioffset0+vdwjidx0D,
 271                                          &c6_00,&c12_00);
 272
 273             /* LENNARD-JONES DISPERSION/REPULSION */
 274
 275             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 276             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 277             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 278             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) ,
 279                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 280             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 281
 282             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 283
 284             /* Update potential sum for this i atom from the interaction with this j atom. */
 285             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 286             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 287
 288             fscal            = fvdw;
 289
 290             fscal            = _mm_and_ps(fscal,cutoff_mask);
 291
 292             /* Calculate temporary vectorial force */
 293             tx               = _mm_mul_ps(fscal,dx00);
 294             ty               = _mm_mul_ps(fscal,dy00);
 295             tz               = _mm_mul_ps(fscal,dz00);
 296
 297             /* Update vectorial force */
 298             fix0             = _mm_add_ps(fix0,tx);
 299             fiy0             = _mm_add_ps(fiy0,ty);
 300             fiz0             = _mm_add_ps(fiz0,tz);
 301
 302             fjx0             = _mm_add_ps(fjx0,tx);
 303             fjy0             = _mm_add_ps(fjy0,ty);
 304             fjz0             = _mm_add_ps(fjz0,tz);
 305
 306             }
 307
 308             /**************************
 309              * CALCULATE INTERACTIONS *
 310              **************************/
 311
 312             if (gmx_mm_any_lt(rsq10,rcutoff2))
 313             {
 314
 315             r10              = _mm_mul_ps(rsq10,rinv10);
 316
 317             /* Compute parameters for interactions between i and j atoms */
 318             qq10             = _mm_mul_ps(iq1,jq0);
 319
 320             /* EWALD ELECTROSTATICS */
 321
 322             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 323             ewrt             = _mm_mul_ps(r10,ewtabscale);
 324             ewitab           = _mm_cvttps_epi32(ewrt);
 325             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 326             ewitab           = _mm_slli_epi32(ewitab,2);
 327             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 328             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 329             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 330             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 331             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 332             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 333             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 334             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 335             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 336
 337             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 338
 339             /* Update potential sum for this i atom from the interaction with this j atom. */
 340             velec            = _mm_and_ps(velec,cutoff_mask);
 341             velecsum         = _mm_add_ps(velecsum,velec);
 342
 343             fscal            = felec;
 344
 345             fscal            = _mm_and_ps(fscal,cutoff_mask);
 346
 347             /* Calculate temporary vectorial force */
 348             tx               = _mm_mul_ps(fscal,dx10);
 349             ty               = _mm_mul_ps(fscal,dy10);
 350             tz               = _mm_mul_ps(fscal,dz10);
 351
 352             /* Update vectorial force */
 353             fix1             = _mm_add_ps(fix1,tx);
 354             fiy1             = _mm_add_ps(fiy1,ty);
 355             fiz1             = _mm_add_ps(fiz1,tz);
 356
 357             fjx0             = _mm_add_ps(fjx0,tx);
 358             fjy0             = _mm_add_ps(fjy0,ty);
 359             fjz0             = _mm_add_ps(fjz0,tz);
 360
 361             }
 362
 363             /**************************
 364              * CALCULATE INTERACTIONS *
 365              **************************/
 366
 367             if (gmx_mm_any_lt(rsq20,rcutoff2))
 368             {
 369
 370             r20              = _mm_mul_ps(rsq20,rinv20);
 371
 372             /* Compute parameters for interactions between i and j atoms */
 373             qq20             = _mm_mul_ps(iq2,jq0);
 374
 375             /* EWALD ELECTROSTATICS */
 376
 377             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 378             ewrt             = _mm_mul_ps(r20,ewtabscale);
 379             ewitab           = _mm_cvttps_epi32(ewrt);
 380             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 381             ewitab           = _mm_slli_epi32(ewitab,2);
 382             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 383             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 384             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 385             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 386             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 387             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 388             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 389             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 390             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 391
 392             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 393
 394             /* Update potential sum for this i atom from the interaction with this j atom. */
 395             velec            = _mm_and_ps(velec,cutoff_mask);
 396             velecsum         = _mm_add_ps(velecsum,velec);
 397
 398             fscal            = felec;
 399
 400             fscal            = _mm_and_ps(fscal,cutoff_mask);
 401
 402             /* Calculate temporary vectorial force */
 403             tx               = _mm_mul_ps(fscal,dx20);
 404             ty               = _mm_mul_ps(fscal,dy20);
 405             tz               = _mm_mul_ps(fscal,dz20);
 406
 407             /* Update vectorial force */
 408             fix2             = _mm_add_ps(fix2,tx);
 409             fiy2             = _mm_add_ps(fiy2,ty);
 410             fiz2             = _mm_add_ps(fiz2,tz);
 411
 412             fjx0             = _mm_add_ps(fjx0,tx);
 413             fjy0             = _mm_add_ps(fjy0,ty);
 414             fjz0             = _mm_add_ps(fjz0,tz);
 415
 416             }
 417
 418             /**************************
 419              * CALCULATE INTERACTIONS *
 420              **************************/
 421
 422             if (gmx_mm_any_lt(rsq30,rcutoff2))
 423             {
 424
 425             r30              = _mm_mul_ps(rsq30,rinv30);
 426
 427             /* Compute parameters for interactions between i and j atoms */
 428             qq30             = _mm_mul_ps(iq3,jq0);
 429
 430             /* EWALD ELECTROSTATICS */
 431
 432             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 433             ewrt             = _mm_mul_ps(r30,ewtabscale);
 434             ewitab           = _mm_cvttps_epi32(ewrt);
 435             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 436             ewitab           = _mm_slli_epi32(ewitab,2);
 437             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 438             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 439             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 440             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 441             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 442             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 443             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 444             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_sub_ps(rinv30,sh_ewald),velec));
 445             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 446
 447             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 448
 449             /* Update potential sum for this i atom from the interaction with this j atom. */
 450             velec            = _mm_and_ps(velec,cutoff_mask);
 451             velecsum         = _mm_add_ps(velecsum,velec);
 452
 453             fscal            = felec;
 454
 455             fscal            = _mm_and_ps(fscal,cutoff_mask);
 456
 457             /* Calculate temporary vectorial force */
 458             tx               = _mm_mul_ps(fscal,dx30);
 459             ty               = _mm_mul_ps(fscal,dy30);
 460             tz               = _mm_mul_ps(fscal,dz30);
 461
 462             /* Update vectorial force */
 463             fix3             = _mm_add_ps(fix3,tx);
 464             fiy3             = _mm_add_ps(fiy3,ty);
 465             fiz3             = _mm_add_ps(fiz3,tz);
 466
 467             fjx0             = _mm_add_ps(fjx0,tx);
 468             fjy0             = _mm_add_ps(fjy0,ty);
 469             fjz0             = _mm_add_ps(fjz0,tz);
 470
 471             }
 472
 473             fjptrA             = f+j_coord_offsetA;
 474             fjptrB             = f+j_coord_offsetB;
 475             fjptrC             = f+j_coord_offsetC;
 476             fjptrD             = f+j_coord_offsetD;
 477
 478             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 479
 480             /* Inner loop uses 179 flops */
 481         }
 482
 483         if(jidx<j_index_end)
 484         {
 485
 486             /* Get j neighbor index, and coordinate index */
 487             jnrlistA         = jjnr[jidx];
 488             jnrlistB         = jjnr[jidx+1];
 489             jnrlistC         = jjnr[jidx+2];
 490             jnrlistD         = jjnr[jidx+3];
 491             /* Sign of each element will be negative for non-real atoms.
 492              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 493              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 494              */
 495             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 496             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 497             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 498             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 499             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 500             j_coord_offsetA  = DIM*jnrA;
 501             j_coord_offsetB  = DIM*jnrB;
 502             j_coord_offsetC  = DIM*jnrC;
 503             j_coord_offsetD  = DIM*jnrD;
 504
 505             /* load j atom coordinates */
 506             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 507                                               x+j_coord_offsetC,x+j_coord_offsetD,
 508                                               &jx0,&jy0,&jz0);
 509
 510             /* Calculate displacement vector */
 511             dx00             = _mm_sub_ps(ix0,jx0);
 512             dy00             = _mm_sub_ps(iy0,jy0);
 513             dz00             = _mm_sub_ps(iz0,jz0);
 514             dx10             = _mm_sub_ps(ix1,jx0);
 515             dy10             = _mm_sub_ps(iy1,jy0);
 516             dz10             = _mm_sub_ps(iz1,jz0);
 517             dx20             = _mm_sub_ps(ix2,jx0);
 518             dy20             = _mm_sub_ps(iy2,jy0);
 519             dz20             = _mm_sub_ps(iz2,jz0);
 520             dx30             = _mm_sub_ps(ix3,jx0);
 521             dy30             = _mm_sub_ps(iy3,jy0);
 522             dz30             = _mm_sub_ps(iz3,jz0);
 523
 524             /* Calculate squared distance and things based on it */
 525             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 526             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 527             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 528             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 529
 530             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 531             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 532             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 533
 534             rinvsq00         = gmx_mm_inv_ps(rsq00);
 535             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 536             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 537             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 538
 539             /* Load parameters for j particles */
 540             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 541                                                               charge+jnrC+0,charge+jnrD+0);
 542             vdwjidx0A        = 2*vdwtype[jnrA+0];
 543             vdwjidx0B        = 2*vdwtype[jnrB+0];
 544             vdwjidx0C        = 2*vdwtype[jnrC+0];
 545             vdwjidx0D        = 2*vdwtype[jnrD+0];
 546
 547             fjx0             = _mm_setzero_ps();
 548             fjy0             = _mm_setzero_ps();
 549             fjz0             = _mm_setzero_ps();
 550
 551             /**************************
 552              * CALCULATE INTERACTIONS *
 553              **************************/
 554
 555             if (gmx_mm_any_lt(rsq00,rcutoff2))
 556             {
 557
 558             /* Compute parameters for interactions between i and j atoms */
 559             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 560                                          vdwparam+vdwioffset0+vdwjidx0B,
 561                                          vdwparam+vdwioffset0+vdwjidx0C,
 562                                          vdwparam+vdwioffset0+vdwjidx0D,
 563                                          &c6_00,&c12_00);
 564
 565             /* LENNARD-JONES DISPERSION/REPULSION */
 566
 567             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 568             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 569             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 570             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) ,
 571                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 572             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 573
 574             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 575
 576             /* Update potential sum for this i atom from the interaction with this j atom. */
 577             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 578             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 579             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 580
 581             fscal            = fvdw;
 582
 583             fscal            = _mm_and_ps(fscal,cutoff_mask);
 584
 585             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 586
 587             /* Calculate temporary vectorial force */
 588             tx               = _mm_mul_ps(fscal,dx00);
 589             ty               = _mm_mul_ps(fscal,dy00);
 590             tz               = _mm_mul_ps(fscal,dz00);
 591
 592             /* Update vectorial force */
 593             fix0             = _mm_add_ps(fix0,tx);
 594             fiy0             = _mm_add_ps(fiy0,ty);
 595             fiz0             = _mm_add_ps(fiz0,tz);
 596
 597             fjx0             = _mm_add_ps(fjx0,tx);
 598             fjy0             = _mm_add_ps(fjy0,ty);
 599             fjz0             = _mm_add_ps(fjz0,tz);
 600
 601             }
 602
 603             /**************************
 604              * CALCULATE INTERACTIONS *
 605              **************************/
 606
 607             if (gmx_mm_any_lt(rsq10,rcutoff2))
 608             {
 609
 610             r10              = _mm_mul_ps(rsq10,rinv10);
 611             r10              = _mm_andnot_ps(dummy_mask,r10);
 612
 613             /* Compute parameters for interactions between i and j atoms */
 614             qq10             = _mm_mul_ps(iq1,jq0);
 615
 616             /* EWALD ELECTROSTATICS */
 617
 618             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 619             ewrt             = _mm_mul_ps(r10,ewtabscale);
 620             ewitab           = _mm_cvttps_epi32(ewrt);
 621             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 622             ewitab           = _mm_slli_epi32(ewitab,2);
 623             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 624             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 625             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 626             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 627             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 628             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 629             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 630             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 631             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 632
 633             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 634
 635             /* Update potential sum for this i atom from the interaction with this j atom. */
 636             velec            = _mm_and_ps(velec,cutoff_mask);
 637             velec            = _mm_andnot_ps(dummy_mask,velec);
 638             velecsum         = _mm_add_ps(velecsum,velec);
 639
 640             fscal            = felec;
 641
 642             fscal            = _mm_and_ps(fscal,cutoff_mask);
 643
 644             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 645
 646             /* Calculate temporary vectorial force */
 647             tx               = _mm_mul_ps(fscal,dx10);
 648             ty               = _mm_mul_ps(fscal,dy10);
 649             tz               = _mm_mul_ps(fscal,dz10);
 650
 651             /* Update vectorial force */
 652             fix1             = _mm_add_ps(fix1,tx);
 653             fiy1             = _mm_add_ps(fiy1,ty);
 654             fiz1             = _mm_add_ps(fiz1,tz);
 655
 656             fjx0             = _mm_add_ps(fjx0,tx);
 657             fjy0             = _mm_add_ps(fjy0,ty);
 658             fjz0             = _mm_add_ps(fjz0,tz);
 659
 660             }
 661
 662             /**************************
 663              * CALCULATE INTERACTIONS *
 664              **************************/
 665
 666             if (gmx_mm_any_lt(rsq20,rcutoff2))
 667             {
 668
 669             r20              = _mm_mul_ps(rsq20,rinv20);
 670             r20              = _mm_andnot_ps(dummy_mask,r20);
 671
 672             /* Compute parameters for interactions between i and j atoms */
 673             qq20             = _mm_mul_ps(iq2,jq0);
 674
 675             /* EWALD ELECTROSTATICS */
 676
 677             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 678             ewrt             = _mm_mul_ps(r20,ewtabscale);
 679             ewitab           = _mm_cvttps_epi32(ewrt);
 680             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 681             ewitab           = _mm_slli_epi32(ewitab,2);
 682             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 683             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 684             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 685             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 686             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 687             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 688             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 689             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 690             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 691
 692             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 693
 694             /* Update potential sum for this i atom from the interaction with this j atom. */
 695             velec            = _mm_and_ps(velec,cutoff_mask);
 696             velec            = _mm_andnot_ps(dummy_mask,velec);
 697             velecsum         = _mm_add_ps(velecsum,velec);
 698
 699             fscal            = felec;
 700
 701             fscal            = _mm_and_ps(fscal,cutoff_mask);
 702
 703             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 704
 705             /* Calculate temporary vectorial force */
 706             tx               = _mm_mul_ps(fscal,dx20);
 707             ty               = _mm_mul_ps(fscal,dy20);
 708             tz               = _mm_mul_ps(fscal,dz20);
 709
 710             /* Update vectorial force */
 711             fix2             = _mm_add_ps(fix2,tx);
 712             fiy2             = _mm_add_ps(fiy2,ty);
 713             fiz2             = _mm_add_ps(fiz2,tz);
 714
 715             fjx0             = _mm_add_ps(fjx0,tx);
 716             fjy0             = _mm_add_ps(fjy0,ty);
 717             fjz0             = _mm_add_ps(fjz0,tz);
 718
 719             }
 720
 721             /**************************
 722              * CALCULATE INTERACTIONS *
 723              **************************/
 724
 725             if (gmx_mm_any_lt(rsq30,rcutoff2))
 726             {
 727
 728             r30              = _mm_mul_ps(rsq30,rinv30);
 729             r30              = _mm_andnot_ps(dummy_mask,r30);
 730
 731             /* Compute parameters for interactions between i and j atoms */
 732             qq30             = _mm_mul_ps(iq3,jq0);
 733
 734             /* EWALD ELECTROSTATICS */
 735
 736             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 737             ewrt             = _mm_mul_ps(r30,ewtabscale);
 738             ewitab           = _mm_cvttps_epi32(ewrt);
 739             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 740             ewitab           = _mm_slli_epi32(ewitab,2);
 741             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 742             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 743             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 744             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 745             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 746             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 747             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 748             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_sub_ps(rinv30,sh_ewald),velec));
 749             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 750
 751             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 752
 753             /* Update potential sum for this i atom from the interaction with this j atom. */
 754             velec            = _mm_and_ps(velec,cutoff_mask);
 755             velec            = _mm_andnot_ps(dummy_mask,velec);
 756             velecsum         = _mm_add_ps(velecsum,velec);
 757
 758             fscal            = felec;
 759
 760             fscal            = _mm_and_ps(fscal,cutoff_mask);
 761
 762             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 763
 764             /* Calculate temporary vectorial force */
 765             tx               = _mm_mul_ps(fscal,dx30);
 766             ty               = _mm_mul_ps(fscal,dy30);
 767             tz               = _mm_mul_ps(fscal,dz30);
 768
 769             /* Update vectorial force */
 770             fix3             = _mm_add_ps(fix3,tx);
 771             fiy3             = _mm_add_ps(fiy3,ty);
 772             fiz3             = _mm_add_ps(fiz3,tz);
 773
 774             fjx0             = _mm_add_ps(fjx0,tx);
 775             fjy0             = _mm_add_ps(fjy0,ty);
 776             fjz0             = _mm_add_ps(fjz0,tz);
 777
 778             }
 779
 780             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 781             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 782             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 783             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 784
 785             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 786
 787             /* Inner loop uses 182 flops */
 788         }
 789
 790         /* End of innermost loop */
 791
 792         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 793                                               f+i_coord_offset,fshift+i_shift_offset);
 794
 795         ggid                        = gid[iidx];
 796         /* Update potential energies */
 797         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 798         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 799
 800         /* Increment number of inner iterations */
 801         inneriter                  += j_index_end - j_index_start;
 802
 803         /* Outer loop uses 26 flops */
 804     }
 805
 806     /* Increment number of outer iterations */
 807     outeriter        += nri;
 808
 809     /* Update outer/inner flops */
 810
 811     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*182);
 812 }
 813 /*
 814  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_sse2_single
 815  * Electrostatics interaction: Ewald
 816  * VdW interaction:            LennardJones
 817  * Geometry:                   Water4-Particle
 818  * Calculate force/pot:        Force
 819  */
 820 void
 821 nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_sse2_single
 822                     (t_nblist                    * gmx_restrict       nlist,
 823                      rvec                        * gmx_restrict          xx,
 824                      rvec                        * gmx_restrict          ff,
 825                      t_forcerec                  * gmx_restrict          fr,
 826                      t_mdatoms                   * gmx_restrict     mdatoms,
 827                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 828                      t_nrnb                      * gmx_restrict        nrnb)
 829 {
 830     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 831      * just 0 for non-waters.
 832      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 833      * jnr indices corresponding to data put in the four positions in the SIMD register.
 834      */
 835     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 836     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 837     int              jnrA,jnrB,jnrC,jnrD;
 838     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 839     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 840     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 841     real             rcutoff_scalar;
 842     real             *shiftvec,*fshift,*x,*f;
 843     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 844     real             scratch[4*DIM];
 845     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 846     int              vdwioffset0;
 847     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 848     int              vdwioffset1;
 849     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 850     int              vdwioffset2;
 851     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 852     int              vdwioffset3;
 853     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 854     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 855     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 856     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 857     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 858     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 859     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 860     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 861     real             *charge;
 862     int              nvdwtype;
 863     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 864     int              *vdwtype;
 865     real             *vdwparam;
 866     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 867     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 868     __m128i          ewitab;
 869     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 870     real             *ewtab;
 871     __m128           dummy_mask,cutoff_mask;
 872     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 873     __m128           one     = _mm_set1_ps(1.0);
 874     __m128           two     = _mm_set1_ps(2.0);
 875     x                = xx[0];
 876     f                = ff[0];
 877
 878     nri              = nlist->nri;
 879     iinr             = nlist->iinr;
 880     jindex           = nlist->jindex;
 881     jjnr             = nlist->jjnr;
 882     shiftidx         = nlist->shift;
 883     gid              = nlist->gid;
 884     shiftvec         = fr->shift_vec[0];
 885     fshift           = fr->fshift[0];
 886     facel            = _mm_set1_ps(fr->epsfac);
 887     charge           = mdatoms->chargeA;
 888     nvdwtype         = fr->ntype;
 889     vdwparam         = fr->nbfp;
 890     vdwtype          = mdatoms->typeA;
 891
 892     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 893     ewtab            = fr->ic->tabq_coul_F;
 894     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 895     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 896
 897     /* Setup water-specific parameters */
 898     inr              = nlist->iinr[0];
 899     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 900     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 901     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 902     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 903
 904     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 905     rcutoff_scalar   = fr->rcoulomb;
 906     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 907     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 908
 909     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 910     rvdw             = _mm_set1_ps(fr->rvdw);
 911
 912     /* Avoid stupid compiler warnings */
 913     jnrA = jnrB = jnrC = jnrD = 0;
 914     j_coord_offsetA = 0;
 915     j_coord_offsetB = 0;
 916     j_coord_offsetC = 0;
 917     j_coord_offsetD = 0;
 918
 919     outeriter        = 0;
 920     inneriter        = 0;
 921
 922     for(iidx=0;iidx<4*DIM;iidx++)
 923     {
 924         scratch[iidx] = 0.0;
 925     }
 926
 927     /* Start outer loop over neighborlists */
 928     for(iidx=0; iidx<nri; iidx++)
 929     {
 930         /* Load shift vector for this list */
 931         i_shift_offset   = DIM*shiftidx[iidx];
 932
 933         /* Load limits for loop over neighbors */
 934         j_index_start    = jindex[iidx];
 935         j_index_end      = jindex[iidx+1];
 936
 937         /* Get outer coordinate index */
 938         inr              = iinr[iidx];
 939         i_coord_offset   = DIM*inr;
 940
 941         /* Load i particle coords and add shift vector */
 942         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 943                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 944
 945         fix0             = _mm_setzero_ps();
 946         fiy0             = _mm_setzero_ps();
 947         fiz0             = _mm_setzero_ps();
 948         fix1             = _mm_setzero_ps();
 949         fiy1             = _mm_setzero_ps();
 950         fiz1             = _mm_setzero_ps();
 951         fix2             = _mm_setzero_ps();
 952         fiy2             = _mm_setzero_ps();
 953         fiz2             = _mm_setzero_ps();
 954         fix3             = _mm_setzero_ps();
 955         fiy3             = _mm_setzero_ps();
 956         fiz3             = _mm_setzero_ps();
 957
 958         /* Start inner kernel loop */
 959         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 960         {
 961
 962             /* Get j neighbor index, and coordinate index */
 963             jnrA             = jjnr[jidx];
 964             jnrB             = jjnr[jidx+1];
 965             jnrC             = jjnr[jidx+2];
 966             jnrD             = jjnr[jidx+3];
 967             j_coord_offsetA  = DIM*jnrA;
 968             j_coord_offsetB  = DIM*jnrB;
 969             j_coord_offsetC  = DIM*jnrC;
 970             j_coord_offsetD  = DIM*jnrD;
 971
 972             /* load j atom coordinates */
 973             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 974                                               x+j_coord_offsetC,x+j_coord_offsetD,
 975                                               &jx0,&jy0,&jz0);
 976
 977             /* Calculate displacement vector */
 978             dx00             = _mm_sub_ps(ix0,jx0);
 979             dy00             = _mm_sub_ps(iy0,jy0);
 980             dz00             = _mm_sub_ps(iz0,jz0);
 981             dx10             = _mm_sub_ps(ix1,jx0);
 982             dy10             = _mm_sub_ps(iy1,jy0);
 983             dz10             = _mm_sub_ps(iz1,jz0);
 984             dx20             = _mm_sub_ps(ix2,jx0);
 985             dy20             = _mm_sub_ps(iy2,jy0);
 986             dz20             = _mm_sub_ps(iz2,jz0);
 987             dx30             = _mm_sub_ps(ix3,jx0);
 988             dy30             = _mm_sub_ps(iy3,jy0);
 989             dz30             = _mm_sub_ps(iz3,jz0);
 990
 991             /* Calculate squared distance and things based on it */
 992             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 993             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 994             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 995             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 996
 997             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 998             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 999             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1000
1001             rinvsq00         = gmx_mm_inv_ps(rsq00);
1002             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1003             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1004             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1005
1006             /* Load parameters for j particles */
1007             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1008                                                               charge+jnrC+0,charge+jnrD+0);
1009             vdwjidx0A        = 2*vdwtype[jnrA+0];
1010             vdwjidx0B        = 2*vdwtype[jnrB+0];
1011             vdwjidx0C        = 2*vdwtype[jnrC+0];
1012             vdwjidx0D        = 2*vdwtype[jnrD+0];
1013
1014             fjx0             = _mm_setzero_ps();
1015             fjy0             = _mm_setzero_ps();
1016             fjz0             = _mm_setzero_ps();
1017
1018             /**************************
1019              * CALCULATE INTERACTIONS *
1020              **************************/
1021
1022             if (gmx_mm_any_lt(rsq00,rcutoff2))
1023             {
1024
1025             /* Compute parameters for interactions between i and j atoms */
1026             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1027                                          vdwparam+vdwioffset0+vdwjidx0B,
1028                                          vdwparam+vdwioffset0+vdwjidx0C,
1029                                          vdwparam+vdwioffset0+vdwjidx0D,
1030                                          &c6_00,&c12_00);
1031
1032             /* LENNARD-JONES DISPERSION/REPULSION */
1033
1034             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1035             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1036
1037             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1038
1039             fscal            = fvdw;
1040
1041             fscal            = _mm_and_ps(fscal,cutoff_mask);
1042
1043             /* Calculate temporary vectorial force */
1044             tx               = _mm_mul_ps(fscal,dx00);
1045             ty               = _mm_mul_ps(fscal,dy00);
1046             tz               = _mm_mul_ps(fscal,dz00);
1047
1048             /* Update vectorial force */
1049             fix0             = _mm_add_ps(fix0,tx);
1050             fiy0             = _mm_add_ps(fiy0,ty);
1051             fiz0             = _mm_add_ps(fiz0,tz);
1052
1053             fjx0             = _mm_add_ps(fjx0,tx);
1054             fjy0             = _mm_add_ps(fjy0,ty);
1055             fjz0             = _mm_add_ps(fjz0,tz);
1056
1057             }
1058
1059             /**************************
1060              * CALCULATE INTERACTIONS *
1061              **************************/
1062
1063             if (gmx_mm_any_lt(rsq10,rcutoff2))
1064             {
1065
1066             r10              = _mm_mul_ps(rsq10,rinv10);
1067
1068             /* Compute parameters for interactions between i and j atoms */
1069             qq10             = _mm_mul_ps(iq1,jq0);
1070
1071             /* EWALD ELECTROSTATICS */
1072
1073             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1074             ewrt             = _mm_mul_ps(r10,ewtabscale);
1075             ewitab           = _mm_cvttps_epi32(ewrt);
1076             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1077             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1078                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1079                                          &ewtabF,&ewtabFn);
1080             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1081             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1082
1083             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1084
1085             fscal            = felec;
1086
1087             fscal            = _mm_and_ps(fscal,cutoff_mask);
1088
1089             /* Calculate temporary vectorial force */
1090             tx               = _mm_mul_ps(fscal,dx10);
1091             ty               = _mm_mul_ps(fscal,dy10);
1092             tz               = _mm_mul_ps(fscal,dz10);
1093
1094             /* Update vectorial force */
1095             fix1             = _mm_add_ps(fix1,tx);
1096             fiy1             = _mm_add_ps(fiy1,ty);
1097             fiz1             = _mm_add_ps(fiz1,tz);
1098
1099             fjx0             = _mm_add_ps(fjx0,tx);
1100             fjy0             = _mm_add_ps(fjy0,ty);
1101             fjz0             = _mm_add_ps(fjz0,tz);
1102
1103             }
1104
1105             /**************************
1106              * CALCULATE INTERACTIONS *
1107              **************************/
1108
1109             if (gmx_mm_any_lt(rsq20,rcutoff2))
1110             {
1111
1112             r20              = _mm_mul_ps(rsq20,rinv20);
1113
1114             /* Compute parameters for interactions between i and j atoms */
1115             qq20             = _mm_mul_ps(iq2,jq0);
1116
1117             /* EWALD ELECTROSTATICS */
1118
1119             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1120             ewrt             = _mm_mul_ps(r20,ewtabscale);
1121             ewitab           = _mm_cvttps_epi32(ewrt);
1122             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1123             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1124                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1125                                          &ewtabF,&ewtabFn);
1126             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1127             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1128
1129             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1130
1131             fscal            = felec;
1132
1133             fscal            = _mm_and_ps(fscal,cutoff_mask);
1134
1135             /* Calculate temporary vectorial force */
1136             tx               = _mm_mul_ps(fscal,dx20);
1137             ty               = _mm_mul_ps(fscal,dy20);
1138             tz               = _mm_mul_ps(fscal,dz20);
1139
1140             /* Update vectorial force */
1141             fix2             = _mm_add_ps(fix2,tx);
1142             fiy2             = _mm_add_ps(fiy2,ty);
1143             fiz2             = _mm_add_ps(fiz2,tz);
1144
1145             fjx0             = _mm_add_ps(fjx0,tx);
1146             fjy0             = _mm_add_ps(fjy0,ty);
1147             fjz0             = _mm_add_ps(fjz0,tz);
1148
1149             }
1150
1151             /**************************
1152              * CALCULATE INTERACTIONS *
1153              **************************/
1154
1155             if (gmx_mm_any_lt(rsq30,rcutoff2))
1156             {
1157
1158             r30              = _mm_mul_ps(rsq30,rinv30);
1159
1160             /* Compute parameters for interactions between i and j atoms */
1161             qq30             = _mm_mul_ps(iq3,jq0);
1162
1163             /* EWALD ELECTROSTATICS */
1164
1165             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1166             ewrt             = _mm_mul_ps(r30,ewtabscale);
1167             ewitab           = _mm_cvttps_epi32(ewrt);
1168             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1169             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1170                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1171                                          &ewtabF,&ewtabFn);
1172             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1173             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1174
1175             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1176
1177             fscal            = felec;
1178
1179             fscal            = _mm_and_ps(fscal,cutoff_mask);
1180
1181             /* Calculate temporary vectorial force */
1182             tx               = _mm_mul_ps(fscal,dx30);
1183             ty               = _mm_mul_ps(fscal,dy30);
1184             tz               = _mm_mul_ps(fscal,dz30);
1185
1186             /* Update vectorial force */
1187             fix3             = _mm_add_ps(fix3,tx);
1188             fiy3             = _mm_add_ps(fiy3,ty);
1189             fiz3             = _mm_add_ps(fiz3,tz);
1190
1191             fjx0             = _mm_add_ps(fjx0,tx);
1192             fjy0             = _mm_add_ps(fjy0,ty);
1193             fjz0             = _mm_add_ps(fjz0,tz);
1194
1195             }
1196
1197             fjptrA             = f+j_coord_offsetA;
1198             fjptrB             = f+j_coord_offsetB;
1199             fjptrC             = f+j_coord_offsetC;
1200             fjptrD             = f+j_coord_offsetD;
1201
1202             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1203
1204             /* Inner loop uses 147 flops */
1205         }
1206
1207         if(jidx<j_index_end)
1208         {
1209
1210             /* Get j neighbor index, and coordinate index */
1211             jnrlistA         = jjnr[jidx];
1212             jnrlistB         = jjnr[jidx+1];
1213             jnrlistC         = jjnr[jidx+2];
1214             jnrlistD         = jjnr[jidx+3];
1215             /* Sign of each element will be negative for non-real atoms.
1216              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1217              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1218              */
1219             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1220             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1221             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1222             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1223             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1224             j_coord_offsetA  = DIM*jnrA;
1225             j_coord_offsetB  = DIM*jnrB;
1226             j_coord_offsetC  = DIM*jnrC;
1227             j_coord_offsetD  = DIM*jnrD;
1228
1229             /* load j atom coordinates */
1230             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1231                                               x+j_coord_offsetC,x+j_coord_offsetD,
1232                                               &jx0,&jy0,&jz0);
1233
1234             /* Calculate displacement vector */
1235             dx00             = _mm_sub_ps(ix0,jx0);
1236             dy00             = _mm_sub_ps(iy0,jy0);
1237             dz00             = _mm_sub_ps(iz0,jz0);
1238             dx10             = _mm_sub_ps(ix1,jx0);
1239             dy10             = _mm_sub_ps(iy1,jy0);
1240             dz10             = _mm_sub_ps(iz1,jz0);
1241             dx20             = _mm_sub_ps(ix2,jx0);
1242             dy20             = _mm_sub_ps(iy2,jy0);
1243             dz20             = _mm_sub_ps(iz2,jz0);
1244             dx30             = _mm_sub_ps(ix3,jx0);
1245             dy30             = _mm_sub_ps(iy3,jy0);
1246             dz30             = _mm_sub_ps(iz3,jz0);
1247
1248             /* Calculate squared distance and things based on it */
1249             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1250             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1251             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1252             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1253
1254             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1255             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1256             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1257
1258             rinvsq00         = gmx_mm_inv_ps(rsq00);
1259             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1260             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1261             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1262
1263             /* Load parameters for j particles */
1264             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1265                                                               charge+jnrC+0,charge+jnrD+0);
1266             vdwjidx0A        = 2*vdwtype[jnrA+0];
1267             vdwjidx0B        = 2*vdwtype[jnrB+0];
1268             vdwjidx0C        = 2*vdwtype[jnrC+0];
1269             vdwjidx0D        = 2*vdwtype[jnrD+0];
1270
1271             fjx0             = _mm_setzero_ps();
1272             fjy0             = _mm_setzero_ps();
1273             fjz0             = _mm_setzero_ps();
1274
1275             /**************************
1276              * CALCULATE INTERACTIONS *
1277              **************************/
1278
1279             if (gmx_mm_any_lt(rsq00,rcutoff2))
1280             {
1281
1282             /* Compute parameters for interactions between i and j atoms */
1283             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1284                                          vdwparam+vdwioffset0+vdwjidx0B,
1285                                          vdwparam+vdwioffset0+vdwjidx0C,
1286                                          vdwparam+vdwioffset0+vdwjidx0D,
1287                                          &c6_00,&c12_00);
1288
1289             /* LENNARD-JONES DISPERSION/REPULSION */
1290
1291             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1292             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1293
1294             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1295
1296             fscal            = fvdw;
1297
1298             fscal            = _mm_and_ps(fscal,cutoff_mask);
1299
1300             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1301
1302             /* Calculate temporary vectorial force */
1303             tx               = _mm_mul_ps(fscal,dx00);
1304             ty               = _mm_mul_ps(fscal,dy00);
1305             tz               = _mm_mul_ps(fscal,dz00);
1306
1307             /* Update vectorial force */
1308             fix0             = _mm_add_ps(fix0,tx);
1309             fiy0             = _mm_add_ps(fiy0,ty);
1310             fiz0             = _mm_add_ps(fiz0,tz);
1311
1312             fjx0             = _mm_add_ps(fjx0,tx);
1313             fjy0             = _mm_add_ps(fjy0,ty);
1314             fjz0             = _mm_add_ps(fjz0,tz);
1315
1316             }
1317
1318             /**************************
1319              * CALCULATE INTERACTIONS *
1320              **************************/
1321
1322             if (gmx_mm_any_lt(rsq10,rcutoff2))
1323             {
1324
1325             r10              = _mm_mul_ps(rsq10,rinv10);
1326             r10              = _mm_andnot_ps(dummy_mask,r10);
1327
1328             /* Compute parameters for interactions between i and j atoms */
1329             qq10             = _mm_mul_ps(iq1,jq0);
1330
1331             /* EWALD ELECTROSTATICS */
1332
1333             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1334             ewrt             = _mm_mul_ps(r10,ewtabscale);
1335             ewitab           = _mm_cvttps_epi32(ewrt);
1336             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1337             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1338                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1339                                          &ewtabF,&ewtabFn);
1340             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1341             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1342
1343             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1344
1345             fscal            = felec;
1346
1347             fscal            = _mm_and_ps(fscal,cutoff_mask);
1348
1349             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1350
1351             /* Calculate temporary vectorial force */
1352             tx               = _mm_mul_ps(fscal,dx10);
1353             ty               = _mm_mul_ps(fscal,dy10);
1354             tz               = _mm_mul_ps(fscal,dz10);
1355
1356             /* Update vectorial force */
1357             fix1             = _mm_add_ps(fix1,tx);
1358             fiy1             = _mm_add_ps(fiy1,ty);
1359             fiz1             = _mm_add_ps(fiz1,tz);
1360
1361             fjx0             = _mm_add_ps(fjx0,tx);
1362             fjy0             = _mm_add_ps(fjy0,ty);
1363             fjz0             = _mm_add_ps(fjz0,tz);
1364
1365             }
1366
1367             /**************************
1368              * CALCULATE INTERACTIONS *
1369              **************************/
1370
1371             if (gmx_mm_any_lt(rsq20,rcutoff2))
1372             {
1373
1374             r20              = _mm_mul_ps(rsq20,rinv20);
1375             r20              = _mm_andnot_ps(dummy_mask,r20);
1376
1377             /* Compute parameters for interactions between i and j atoms */
1378             qq20             = _mm_mul_ps(iq2,jq0);
1379
1380             /* EWALD ELECTROSTATICS */
1381
1382             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1383             ewrt             = _mm_mul_ps(r20,ewtabscale);
1384             ewitab           = _mm_cvttps_epi32(ewrt);
1385             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1386             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1387                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1388                                          &ewtabF,&ewtabFn);
1389             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1390             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1391
1392             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1393
1394             fscal            = felec;
1395
1396             fscal            = _mm_and_ps(fscal,cutoff_mask);
1397
1398             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1399
1400             /* Calculate temporary vectorial force */
1401             tx               = _mm_mul_ps(fscal,dx20);
1402             ty               = _mm_mul_ps(fscal,dy20);
1403             tz               = _mm_mul_ps(fscal,dz20);
1404
1405             /* Update vectorial force */
1406             fix2             = _mm_add_ps(fix2,tx);
1407             fiy2             = _mm_add_ps(fiy2,ty);
1408             fiz2             = _mm_add_ps(fiz2,tz);
1409
1410             fjx0             = _mm_add_ps(fjx0,tx);
1411             fjy0             = _mm_add_ps(fjy0,ty);
1412             fjz0             = _mm_add_ps(fjz0,tz);
1413
1414             }
1415
1416             /**************************
1417              * CALCULATE INTERACTIONS *
1418              **************************/
1419
1420             if (gmx_mm_any_lt(rsq30,rcutoff2))
1421             {
1422
1423             r30              = _mm_mul_ps(rsq30,rinv30);
1424             r30              = _mm_andnot_ps(dummy_mask,r30);
1425
1426             /* Compute parameters for interactions between i and j atoms */
1427             qq30             = _mm_mul_ps(iq3,jq0);
1428
1429             /* EWALD ELECTROSTATICS */
1430
1431             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1432             ewrt             = _mm_mul_ps(r30,ewtabscale);
1433             ewitab           = _mm_cvttps_epi32(ewrt);
1434             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1435             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1436                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1437                                          &ewtabF,&ewtabFn);
1438             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1439             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1440
1441             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1442
1443             fscal            = felec;
1444
1445             fscal            = _mm_and_ps(fscal,cutoff_mask);
1446
1447             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1448
1449             /* Calculate temporary vectorial force */
1450             tx               = _mm_mul_ps(fscal,dx30);
1451             ty               = _mm_mul_ps(fscal,dy30);
1452             tz               = _mm_mul_ps(fscal,dz30);
1453
1454             /* Update vectorial force */
1455             fix3             = _mm_add_ps(fix3,tx);
1456             fiy3             = _mm_add_ps(fiy3,ty);
1457             fiz3             = _mm_add_ps(fiz3,tz);
1458
1459             fjx0             = _mm_add_ps(fjx0,tx);
1460             fjy0             = _mm_add_ps(fjy0,ty);
1461             fjz0             = _mm_add_ps(fjz0,tz);
1462
1463             }
1464
1465             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1466             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1467             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1468             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1469
1470             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1471
1472             /* Inner loop uses 150 flops */
1473         }
1474
1475         /* End of innermost loop */
1476
1477         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1478                                               f+i_coord_offset,fshift+i_shift_offset);
1479
1480         /* Increment number of inner iterations */
1481         inneriter                  += j_index_end - j_index_start;
1482
1483         /* Outer loop uses 24 flops */
1484     }
1485
1486     /* Increment number of outer iterations */
1487     outeriter        += nri;
1488
1489     /* Update outer/inner flops */
1490
1491     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*150);
1492 }