src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEwSw_VdwNone_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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse2_single.h"
  50 #include "kernelutil_x86_sse2_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomW4P1_VF_sse2_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            None
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSw_VdwNone_GeomW4P1_VF_sse2_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  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           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  96     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     __m128i          ewitab;
  99     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 100     real             *ewtab;
 101     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 102     real             rswitch_scalar,d_scalar;
 103     __m128           dummy_mask,cutoff_mask;
 104     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 105     __m128           one     = _mm_set1_ps(1.0);
 106     __m128           two     = _mm_set1_ps(2.0);
 107     x                = xx[0];
 108     f                = ff[0];
 109
 110     nri              = nlist->nri;
 111     iinr             = nlist->iinr;
 112     jindex           = nlist->jindex;
 113     jjnr             = nlist->jjnr;
 114     shiftidx         = nlist->shift;
 115     gid              = nlist->gid;
 116     shiftvec         = fr->shift_vec[0];
 117     fshift           = fr->fshift[0];
 118     facel            = _mm_set1_ps(fr->epsfac);
 119     charge           = mdatoms->chargeA;
 120
 121     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 122     ewtab            = fr->ic->tabq_coul_FDV0;
 123     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 124     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 125
 126     /* Setup water-specific parameters */
 127     inr              = nlist->iinr[0];
 128     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 129     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 130     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 131
 132     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 133     rcutoff_scalar   = fr->rcoulomb;
 134     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 135     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 136
 137     rswitch_scalar   = fr->rcoulomb_switch;
 138     rswitch          = _mm_set1_ps(rswitch_scalar);
 139     /* Setup switch parameters */
 140     d_scalar         = rcutoff_scalar-rswitch_scalar;
 141     d                = _mm_set1_ps(d_scalar);
 142     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 143     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 144     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 145     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 146     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 147     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 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_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 180                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 181
 182         fix1             = _mm_setzero_ps();
 183         fiy1             = _mm_setzero_ps();
 184         fiz1             = _mm_setzero_ps();
 185         fix2             = _mm_setzero_ps();
 186         fiy2             = _mm_setzero_ps();
 187         fiz2             = _mm_setzero_ps();
 188         fix3             = _mm_setzero_ps();
 189         fiy3             = _mm_setzero_ps();
 190         fiz3             = _mm_setzero_ps();
 191
 192         /* Reset potential sums */
 193         velecsum         = _mm_setzero_ps();
 194
 195         /* Start inner kernel loop */
 196         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 197         {
 198
 199             /* Get j neighbor index, and coordinate index */
 200             jnrA             = jjnr[jidx];
 201             jnrB             = jjnr[jidx+1];
 202             jnrC             = jjnr[jidx+2];
 203             jnrD             = jjnr[jidx+3];
 204             j_coord_offsetA  = DIM*jnrA;
 205             j_coord_offsetB  = DIM*jnrB;
 206             j_coord_offsetC  = DIM*jnrC;
 207             j_coord_offsetD  = DIM*jnrD;
 208
 209             /* load j atom coordinates */
 210             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 211                                               x+j_coord_offsetC,x+j_coord_offsetD,
 212                                               &jx0,&jy0,&jz0);
 213
 214             /* Calculate displacement vector */
 215             dx10             = _mm_sub_ps(ix1,jx0);
 216             dy10             = _mm_sub_ps(iy1,jy0);
 217             dz10             = _mm_sub_ps(iz1,jz0);
 218             dx20             = _mm_sub_ps(ix2,jx0);
 219             dy20             = _mm_sub_ps(iy2,jy0);
 220             dz20             = _mm_sub_ps(iz2,jz0);
 221             dx30             = _mm_sub_ps(ix3,jx0);
 222             dy30             = _mm_sub_ps(iy3,jy0);
 223             dz30             = _mm_sub_ps(iz3,jz0);
 224
 225             /* Calculate squared distance and things based on it */
 226             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 227             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 228             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 229
 230             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 231             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 232             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 233
 234             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 235             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 236             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 237
 238             /* Load parameters for j particles */
 239             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 240                                                               charge+jnrC+0,charge+jnrD+0);
 241
 242             fjx0             = _mm_setzero_ps();
 243             fjy0             = _mm_setzero_ps();
 244             fjz0             = _mm_setzero_ps();
 245
 246             /**************************
 247              * CALCULATE INTERACTIONS *
 248              **************************/
 249
 250             if (gmx_mm_any_lt(rsq10,rcutoff2))
 251             {
 252
 253             r10              = _mm_mul_ps(rsq10,rinv10);
 254
 255             /* Compute parameters for interactions between i and j atoms */
 256             qq10             = _mm_mul_ps(iq1,jq0);
 257
 258             /* EWALD ELECTROSTATICS */
 259
 260             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 261             ewrt             = _mm_mul_ps(r10,ewtabscale);
 262             ewitab           = _mm_cvttps_epi32(ewrt);
 263             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 264             ewitab           = _mm_slli_epi32(ewitab,2);
 265             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 266             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 267             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 268             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 269             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 270             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 271             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 272             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 273             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 274
 275             d                = _mm_sub_ps(r10,rswitch);
 276             d                = _mm_max_ps(d,_mm_setzero_ps());
 277             d2               = _mm_mul_ps(d,d);
 278             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)))))));
 279
 280             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 281
 282             /* Evaluate switch function */
 283             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 284             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
 285             velec            = _mm_mul_ps(velec,sw);
 286             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 287
 288             /* Update potential sum for this i atom from the interaction with this j atom. */
 289             velec            = _mm_and_ps(velec,cutoff_mask);
 290             velecsum         = _mm_add_ps(velecsum,velec);
 291
 292             fscal            = felec;
 293
 294             fscal            = _mm_and_ps(fscal,cutoff_mask);
 295
 296             /* Calculate temporary vectorial force */
 297             tx               = _mm_mul_ps(fscal,dx10);
 298             ty               = _mm_mul_ps(fscal,dy10);
 299             tz               = _mm_mul_ps(fscal,dz10);
 300
 301             /* Update vectorial force */
 302             fix1             = _mm_add_ps(fix1,tx);
 303             fiy1             = _mm_add_ps(fiy1,ty);
 304             fiz1             = _mm_add_ps(fiz1,tz);
 305
 306             fjx0             = _mm_add_ps(fjx0,tx);
 307             fjy0             = _mm_add_ps(fjy0,ty);
 308             fjz0             = _mm_add_ps(fjz0,tz);
 309
 310             }
 311
 312             /**************************
 313              * CALCULATE INTERACTIONS *
 314              **************************/
 315
 316             if (gmx_mm_any_lt(rsq20,rcutoff2))
 317             {
 318
 319             r20              = _mm_mul_ps(rsq20,rinv20);
 320
 321             /* Compute parameters for interactions between i and j atoms */
 322             qq20             = _mm_mul_ps(iq2,jq0);
 323
 324             /* EWALD ELECTROSTATICS */
 325
 326             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 327             ewrt             = _mm_mul_ps(r20,ewtabscale);
 328             ewitab           = _mm_cvttps_epi32(ewrt);
 329             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 330             ewitab           = _mm_slli_epi32(ewitab,2);
 331             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 332             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 333             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 334             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 335             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 336             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 337             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 338             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 339             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 340
 341             d                = _mm_sub_ps(r20,rswitch);
 342             d                = _mm_max_ps(d,_mm_setzero_ps());
 343             d2               = _mm_mul_ps(d,d);
 344             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)))))));
 345
 346             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 347
 348             /* Evaluate switch function */
 349             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 350             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
 351             velec            = _mm_mul_ps(velec,sw);
 352             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 353
 354             /* Update potential sum for this i atom from the interaction with this j atom. */
 355             velec            = _mm_and_ps(velec,cutoff_mask);
 356             velecsum         = _mm_add_ps(velecsum,velec);
 357
 358             fscal            = felec;
 359
 360             fscal            = _mm_and_ps(fscal,cutoff_mask);
 361
 362             /* Calculate temporary vectorial force */
 363             tx               = _mm_mul_ps(fscal,dx20);
 364             ty               = _mm_mul_ps(fscal,dy20);
 365             tz               = _mm_mul_ps(fscal,dz20);
 366
 367             /* Update vectorial force */
 368             fix2             = _mm_add_ps(fix2,tx);
 369             fiy2             = _mm_add_ps(fiy2,ty);
 370             fiz2             = _mm_add_ps(fiz2,tz);
 371
 372             fjx0             = _mm_add_ps(fjx0,tx);
 373             fjy0             = _mm_add_ps(fjy0,ty);
 374             fjz0             = _mm_add_ps(fjz0,tz);
 375
 376             }
 377
 378             /**************************
 379              * CALCULATE INTERACTIONS *
 380              **************************/
 381
 382             if (gmx_mm_any_lt(rsq30,rcutoff2))
 383             {
 384
 385             r30              = _mm_mul_ps(rsq30,rinv30);
 386
 387             /* Compute parameters for interactions between i and j atoms */
 388             qq30             = _mm_mul_ps(iq3,jq0);
 389
 390             /* EWALD ELECTROSTATICS */
 391
 392             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 393             ewrt             = _mm_mul_ps(r30,ewtabscale);
 394             ewitab           = _mm_cvttps_epi32(ewrt);
 395             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 396             ewitab           = _mm_slli_epi32(ewitab,2);
 397             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 398             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 399             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 400             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 401             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 402             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 403             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 404             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 405             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 406
 407             d                = _mm_sub_ps(r30,rswitch);
 408             d                = _mm_max_ps(d,_mm_setzero_ps());
 409             d2               = _mm_mul_ps(d,d);
 410             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)))))));
 411
 412             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 413
 414             /* Evaluate switch function */
 415             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 416             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv30,_mm_mul_ps(velec,dsw)) );
 417             velec            = _mm_mul_ps(velec,sw);
 418             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 419
 420             /* Update potential sum for this i atom from the interaction with this j atom. */
 421             velec            = _mm_and_ps(velec,cutoff_mask);
 422             velecsum         = _mm_add_ps(velecsum,velec);
 423
 424             fscal            = felec;
 425
 426             fscal            = _mm_and_ps(fscal,cutoff_mask);
 427
 428             /* Calculate temporary vectorial force */
 429             tx               = _mm_mul_ps(fscal,dx30);
 430             ty               = _mm_mul_ps(fscal,dy30);
 431             tz               = _mm_mul_ps(fscal,dz30);
 432
 433             /* Update vectorial force */
 434             fix3             = _mm_add_ps(fix3,tx);
 435             fiy3             = _mm_add_ps(fiy3,ty);
 436             fiz3             = _mm_add_ps(fiz3,tz);
 437
 438             fjx0             = _mm_add_ps(fjx0,tx);
 439             fjy0             = _mm_add_ps(fjy0,ty);
 440             fjz0             = _mm_add_ps(fjz0,tz);
 441
 442             }
 443
 444             fjptrA             = f+j_coord_offsetA;
 445             fjptrB             = f+j_coord_offsetB;
 446             fjptrC             = f+j_coord_offsetC;
 447             fjptrD             = f+j_coord_offsetD;
 448
 449             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 450
 451             /* Inner loop uses 195 flops */
 452         }
 453
 454         if(jidx<j_index_end)
 455         {
 456
 457             /* Get j neighbor index, and coordinate index */
 458             jnrlistA         = jjnr[jidx];
 459             jnrlistB         = jjnr[jidx+1];
 460             jnrlistC         = jjnr[jidx+2];
 461             jnrlistD         = jjnr[jidx+3];
 462             /* Sign of each element will be negative for non-real atoms.
 463              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 464              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 465              */
 466             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 467             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 468             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 469             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 470             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 471             j_coord_offsetA  = DIM*jnrA;
 472             j_coord_offsetB  = DIM*jnrB;
 473             j_coord_offsetC  = DIM*jnrC;
 474             j_coord_offsetD  = DIM*jnrD;
 475
 476             /* load j atom coordinates */
 477             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 478                                               x+j_coord_offsetC,x+j_coord_offsetD,
 479                                               &jx0,&jy0,&jz0);
 480
 481             /* Calculate displacement vector */
 482             dx10             = _mm_sub_ps(ix1,jx0);
 483             dy10             = _mm_sub_ps(iy1,jy0);
 484             dz10             = _mm_sub_ps(iz1,jz0);
 485             dx20             = _mm_sub_ps(ix2,jx0);
 486             dy20             = _mm_sub_ps(iy2,jy0);
 487             dz20             = _mm_sub_ps(iz2,jz0);
 488             dx30             = _mm_sub_ps(ix3,jx0);
 489             dy30             = _mm_sub_ps(iy3,jy0);
 490             dz30             = _mm_sub_ps(iz3,jz0);
 491
 492             /* Calculate squared distance and things based on it */
 493             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 494             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 495             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 496
 497             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 498             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 499             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 500
 501             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 502             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 503             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 504
 505             /* Load parameters for j particles */
 506             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 507                                                               charge+jnrC+0,charge+jnrD+0);
 508
 509             fjx0             = _mm_setzero_ps();
 510             fjy0             = _mm_setzero_ps();
 511             fjz0             = _mm_setzero_ps();
 512
 513             /**************************
 514              * CALCULATE INTERACTIONS *
 515              **************************/
 516
 517             if (gmx_mm_any_lt(rsq10,rcutoff2))
 518             {
 519
 520             r10              = _mm_mul_ps(rsq10,rinv10);
 521             r10              = _mm_andnot_ps(dummy_mask,r10);
 522
 523             /* Compute parameters for interactions between i and j atoms */
 524             qq10             = _mm_mul_ps(iq1,jq0);
 525
 526             /* EWALD ELECTROSTATICS */
 527
 528             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 529             ewrt             = _mm_mul_ps(r10,ewtabscale);
 530             ewitab           = _mm_cvttps_epi32(ewrt);
 531             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 532             ewitab           = _mm_slli_epi32(ewitab,2);
 533             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 534             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 535             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 536             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 537             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 538             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 539             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 540             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 541             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 542
 543             d                = _mm_sub_ps(r10,rswitch);
 544             d                = _mm_max_ps(d,_mm_setzero_ps());
 545             d2               = _mm_mul_ps(d,d);
 546             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)))))));
 547
 548             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 549
 550             /* Evaluate switch function */
 551             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 552             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
 553             velec            = _mm_mul_ps(velec,sw);
 554             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 555
 556             /* Update potential sum for this i atom from the interaction with this j atom. */
 557             velec            = _mm_and_ps(velec,cutoff_mask);
 558             velec            = _mm_andnot_ps(dummy_mask,velec);
 559             velecsum         = _mm_add_ps(velecsum,velec);
 560
 561             fscal            = felec;
 562
 563             fscal            = _mm_and_ps(fscal,cutoff_mask);
 564
 565             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 566
 567             /* Calculate temporary vectorial force */
 568             tx               = _mm_mul_ps(fscal,dx10);
 569             ty               = _mm_mul_ps(fscal,dy10);
 570             tz               = _mm_mul_ps(fscal,dz10);
 571
 572             /* Update vectorial force */
 573             fix1             = _mm_add_ps(fix1,tx);
 574             fiy1             = _mm_add_ps(fiy1,ty);
 575             fiz1             = _mm_add_ps(fiz1,tz);
 576
 577             fjx0             = _mm_add_ps(fjx0,tx);
 578             fjy0             = _mm_add_ps(fjy0,ty);
 579             fjz0             = _mm_add_ps(fjz0,tz);
 580
 581             }
 582
 583             /**************************
 584              * CALCULATE INTERACTIONS *
 585              **************************/
 586
 587             if (gmx_mm_any_lt(rsq20,rcutoff2))
 588             {
 589
 590             r20              = _mm_mul_ps(rsq20,rinv20);
 591             r20              = _mm_andnot_ps(dummy_mask,r20);
 592
 593             /* Compute parameters for interactions between i and j atoms */
 594             qq20             = _mm_mul_ps(iq2,jq0);
 595
 596             /* EWALD ELECTROSTATICS */
 597
 598             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 599             ewrt             = _mm_mul_ps(r20,ewtabscale);
 600             ewitab           = _mm_cvttps_epi32(ewrt);
 601             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 602             ewitab           = _mm_slli_epi32(ewitab,2);
 603             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 604             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 605             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 606             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 607             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 608             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 609             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 610             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 611             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 612
 613             d                = _mm_sub_ps(r20,rswitch);
 614             d                = _mm_max_ps(d,_mm_setzero_ps());
 615             d2               = _mm_mul_ps(d,d);
 616             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)))))));
 617
 618             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 619
 620             /* Evaluate switch function */
 621             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 622             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
 623             velec            = _mm_mul_ps(velec,sw);
 624             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 625
 626             /* Update potential sum for this i atom from the interaction with this j atom. */
 627             velec            = _mm_and_ps(velec,cutoff_mask);
 628             velec            = _mm_andnot_ps(dummy_mask,velec);
 629             velecsum         = _mm_add_ps(velecsum,velec);
 630
 631             fscal            = felec;
 632
 633             fscal            = _mm_and_ps(fscal,cutoff_mask);
 634
 635             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 636
 637             /* Calculate temporary vectorial force */
 638             tx               = _mm_mul_ps(fscal,dx20);
 639             ty               = _mm_mul_ps(fscal,dy20);
 640             tz               = _mm_mul_ps(fscal,dz20);
 641
 642             /* Update vectorial force */
 643             fix2             = _mm_add_ps(fix2,tx);
 644             fiy2             = _mm_add_ps(fiy2,ty);
 645             fiz2             = _mm_add_ps(fiz2,tz);
 646
 647             fjx0             = _mm_add_ps(fjx0,tx);
 648             fjy0             = _mm_add_ps(fjy0,ty);
 649             fjz0             = _mm_add_ps(fjz0,tz);
 650
 651             }
 652
 653             /**************************
 654              * CALCULATE INTERACTIONS *
 655              **************************/
 656
 657             if (gmx_mm_any_lt(rsq30,rcutoff2))
 658             {
 659
 660             r30              = _mm_mul_ps(rsq30,rinv30);
 661             r30              = _mm_andnot_ps(dummy_mask,r30);
 662
 663             /* Compute parameters for interactions between i and j atoms */
 664             qq30             = _mm_mul_ps(iq3,jq0);
 665
 666             /* EWALD ELECTROSTATICS */
 667
 668             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 669             ewrt             = _mm_mul_ps(r30,ewtabscale);
 670             ewitab           = _mm_cvttps_epi32(ewrt);
 671             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 672             ewitab           = _mm_slli_epi32(ewitab,2);
 673             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 674             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 675             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 676             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 677             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 678             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 679             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 680             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 681             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 682
 683             d                = _mm_sub_ps(r30,rswitch);
 684             d                = _mm_max_ps(d,_mm_setzero_ps());
 685             d2               = _mm_mul_ps(d,d);
 686             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)))))));
 687
 688             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 689
 690             /* Evaluate switch function */
 691             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 692             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv30,_mm_mul_ps(velec,dsw)) );
 693             velec            = _mm_mul_ps(velec,sw);
 694             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 695
 696             /* Update potential sum for this i atom from the interaction with this j atom. */
 697             velec            = _mm_and_ps(velec,cutoff_mask);
 698             velec            = _mm_andnot_ps(dummy_mask,velec);
 699             velecsum         = _mm_add_ps(velecsum,velec);
 700
 701             fscal            = felec;
 702
 703             fscal            = _mm_and_ps(fscal,cutoff_mask);
 704
 705             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 706
 707             /* Calculate temporary vectorial force */
 708             tx               = _mm_mul_ps(fscal,dx30);
 709             ty               = _mm_mul_ps(fscal,dy30);
 710             tz               = _mm_mul_ps(fscal,dz30);
 711
 712             /* Update vectorial force */
 713             fix3             = _mm_add_ps(fix3,tx);
 714             fiy3             = _mm_add_ps(fiy3,ty);
 715             fiz3             = _mm_add_ps(fiz3,tz);
 716
 717             fjx0             = _mm_add_ps(fjx0,tx);
 718             fjy0             = _mm_add_ps(fjy0,ty);
 719             fjz0             = _mm_add_ps(fjz0,tz);
 720
 721             }
 722
 723             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 724             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 725             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 726             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 727
 728             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 729
 730             /* Inner loop uses 198 flops */
 731         }
 732
 733         /* End of innermost loop */
 734
 735         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 736                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 737
 738         ggid                        = gid[iidx];
 739         /* Update potential energies */
 740         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 741
 742         /* Increment number of inner iterations */
 743         inneriter                  += j_index_end - j_index_start;
 744
 745         /* Outer loop uses 19 flops */
 746     }
 747
 748     /* Increment number of outer iterations */
 749     outeriter        += nri;
 750
 751     /* Update outer/inner flops */
 752
 753     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*198);
 754 }
 755 /*
 756  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomW4P1_F_sse2_single
 757  * Electrostatics interaction: Ewald
 758  * VdW interaction:            None
 759  * Geometry:                   Water4-Particle
 760  * Calculate force/pot:        Force
 761  */
 762 void
 763 nb_kernel_ElecEwSw_VdwNone_GeomW4P1_F_sse2_single
 764                     (t_nblist                    * gmx_restrict       nlist,
 765                      rvec                        * gmx_restrict          xx,
 766                      rvec                        * gmx_restrict          ff,
 767                      t_forcerec                  * gmx_restrict          fr,
 768                      t_mdatoms                   * gmx_restrict     mdatoms,
 769                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 770                      t_nrnb                      * gmx_restrict        nrnb)
 771 {
 772     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 773      * just 0 for non-waters.
 774      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 775      * jnr indices corresponding to data put in the four positions in the SIMD register.
 776      */
 777     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 778     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 779     int              jnrA,jnrB,jnrC,jnrD;
 780     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 781     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 782     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 783     real             rcutoff_scalar;
 784     real             *shiftvec,*fshift,*x,*f;
 785     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 786     real             scratch[4*DIM];
 787     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 788     int              vdwioffset1;
 789     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 790     int              vdwioffset2;
 791     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 792     int              vdwioffset3;
 793     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 794     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 795     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 796     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 797     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 798     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 799     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 800     real             *charge;
 801     __m128i          ewitab;
 802     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 803     real             *ewtab;
 804     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 805     real             rswitch_scalar,d_scalar;
 806     __m128           dummy_mask,cutoff_mask;
 807     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 808     __m128           one     = _mm_set1_ps(1.0);
 809     __m128           two     = _mm_set1_ps(2.0);
 810     x                = xx[0];
 811     f                = ff[0];
 812
 813     nri              = nlist->nri;
 814     iinr             = nlist->iinr;
 815     jindex           = nlist->jindex;
 816     jjnr             = nlist->jjnr;
 817     shiftidx         = nlist->shift;
 818     gid              = nlist->gid;
 819     shiftvec         = fr->shift_vec[0];
 820     fshift           = fr->fshift[0];
 821     facel            = _mm_set1_ps(fr->epsfac);
 822     charge           = mdatoms->chargeA;
 823
 824     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 825     ewtab            = fr->ic->tabq_coul_FDV0;
 826     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 827     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 828
 829     /* Setup water-specific parameters */
 830     inr              = nlist->iinr[0];
 831     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 832     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 833     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 834
 835     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 836     rcutoff_scalar   = fr->rcoulomb;
 837     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 838     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 839
 840     rswitch_scalar   = fr->rcoulomb_switch;
 841     rswitch          = _mm_set1_ps(rswitch_scalar);
 842     /* Setup switch parameters */
 843     d_scalar         = rcutoff_scalar-rswitch_scalar;
 844     d                = _mm_set1_ps(d_scalar);
 845     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 846     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 847     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 848     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 849     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 850     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 851
 852     /* Avoid stupid compiler warnings */
 853     jnrA = jnrB = jnrC = jnrD = 0;
 854     j_coord_offsetA = 0;
 855     j_coord_offsetB = 0;
 856     j_coord_offsetC = 0;
 857     j_coord_offsetD = 0;
 858
 859     outeriter        = 0;
 860     inneriter        = 0;
 861
 862     for(iidx=0;iidx<4*DIM;iidx++)
 863     {
 864         scratch[iidx] = 0.0;
 865     }
 866
 867     /* Start outer loop over neighborlists */
 868     for(iidx=0; iidx<nri; iidx++)
 869     {
 870         /* Load shift vector for this list */
 871         i_shift_offset   = DIM*shiftidx[iidx];
 872
 873         /* Load limits for loop over neighbors */
 874         j_index_start    = jindex[iidx];
 875         j_index_end      = jindex[iidx+1];
 876
 877         /* Get outer coordinate index */
 878         inr              = iinr[iidx];
 879         i_coord_offset   = DIM*inr;
 880
 881         /* Load i particle coords and add shift vector */
 882         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 883                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 884
 885         fix1             = _mm_setzero_ps();
 886         fiy1             = _mm_setzero_ps();
 887         fiz1             = _mm_setzero_ps();
 888         fix2             = _mm_setzero_ps();
 889         fiy2             = _mm_setzero_ps();
 890         fiz2             = _mm_setzero_ps();
 891         fix3             = _mm_setzero_ps();
 892         fiy3             = _mm_setzero_ps();
 893         fiz3             = _mm_setzero_ps();
 894
 895         /* Start inner kernel loop */
 896         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 897         {
 898
 899             /* Get j neighbor index, and coordinate index */
 900             jnrA             = jjnr[jidx];
 901             jnrB             = jjnr[jidx+1];
 902             jnrC             = jjnr[jidx+2];
 903             jnrD             = jjnr[jidx+3];
 904             j_coord_offsetA  = DIM*jnrA;
 905             j_coord_offsetB  = DIM*jnrB;
 906             j_coord_offsetC  = DIM*jnrC;
 907             j_coord_offsetD  = DIM*jnrD;
 908
 909             /* load j atom coordinates */
 910             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 911                                               x+j_coord_offsetC,x+j_coord_offsetD,
 912                                               &jx0,&jy0,&jz0);
 913
 914             /* Calculate displacement vector */
 915             dx10             = _mm_sub_ps(ix1,jx0);
 916             dy10             = _mm_sub_ps(iy1,jy0);
 917             dz10             = _mm_sub_ps(iz1,jz0);
 918             dx20             = _mm_sub_ps(ix2,jx0);
 919             dy20             = _mm_sub_ps(iy2,jy0);
 920             dz20             = _mm_sub_ps(iz2,jz0);
 921             dx30             = _mm_sub_ps(ix3,jx0);
 922             dy30             = _mm_sub_ps(iy3,jy0);
 923             dz30             = _mm_sub_ps(iz3,jz0);
 924
 925             /* Calculate squared distance and things based on it */
 926             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 927             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 928             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 929
 930             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 931             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 932             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 933
 934             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 935             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 936             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 937
 938             /* Load parameters for j particles */
 939             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 940                                                               charge+jnrC+0,charge+jnrD+0);
 941
 942             fjx0             = _mm_setzero_ps();
 943             fjy0             = _mm_setzero_ps();
 944             fjz0             = _mm_setzero_ps();
 945
 946             /**************************
 947              * CALCULATE INTERACTIONS *
 948              **************************/
 949
 950             if (gmx_mm_any_lt(rsq10,rcutoff2))
 951             {
 952
 953             r10              = _mm_mul_ps(rsq10,rinv10);
 954
 955             /* Compute parameters for interactions between i and j atoms */
 956             qq10             = _mm_mul_ps(iq1,jq0);
 957
 958             /* EWALD ELECTROSTATICS */
 959
 960             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 961             ewrt             = _mm_mul_ps(r10,ewtabscale);
 962             ewitab           = _mm_cvttps_epi32(ewrt);
 963             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 964             ewitab           = _mm_slli_epi32(ewitab,2);
 965             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 966             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 967             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 968             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 969             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 970             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 971             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 972             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 973             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 974
 975             d                = _mm_sub_ps(r10,rswitch);
 976             d                = _mm_max_ps(d,_mm_setzero_ps());
 977             d2               = _mm_mul_ps(d,d);
 978             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)))))));
 979
 980             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 981
 982             /* Evaluate switch function */
 983             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 984             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
 985             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 986
 987             fscal            = felec;
 988
 989             fscal            = _mm_and_ps(fscal,cutoff_mask);
 990
 991             /* Calculate temporary vectorial force */
 992             tx               = _mm_mul_ps(fscal,dx10);
 993             ty               = _mm_mul_ps(fscal,dy10);
 994             tz               = _mm_mul_ps(fscal,dz10);
 995
 996             /* Update vectorial force */
 997             fix1             = _mm_add_ps(fix1,tx);
 998             fiy1             = _mm_add_ps(fiy1,ty);
 999             fiz1             = _mm_add_ps(fiz1,tz);
1000
1001             fjx0             = _mm_add_ps(fjx0,tx);
1002             fjy0             = _mm_add_ps(fjy0,ty);
1003             fjz0             = _mm_add_ps(fjz0,tz);
1004
1005             }
1006
1007             /**************************
1008              * CALCULATE INTERACTIONS *
1009              **************************/
1010
1011             if (gmx_mm_any_lt(rsq20,rcutoff2))
1012             {
1013
1014             r20              = _mm_mul_ps(rsq20,rinv20);
1015
1016             /* Compute parameters for interactions between i and j atoms */
1017             qq20             = _mm_mul_ps(iq2,jq0);
1018
1019             /* EWALD ELECTROSTATICS */
1020
1021             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1022             ewrt             = _mm_mul_ps(r20,ewtabscale);
1023             ewitab           = _mm_cvttps_epi32(ewrt);
1024             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1025             ewitab           = _mm_slli_epi32(ewitab,2);
1026             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1027             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1028             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1029             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1030             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1031             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1032             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1033             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
1034             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1035
1036             d                = _mm_sub_ps(r20,rswitch);
1037             d                = _mm_max_ps(d,_mm_setzero_ps());
1038             d2               = _mm_mul_ps(d,d);
1039             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)))))));
1040
1041             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1042
1043             /* Evaluate switch function */
1044             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1045             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
1046             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1047
1048             fscal            = felec;
1049
1050             fscal            = _mm_and_ps(fscal,cutoff_mask);
1051
1052             /* Calculate temporary vectorial force */
1053             tx               = _mm_mul_ps(fscal,dx20);
1054             ty               = _mm_mul_ps(fscal,dy20);
1055             tz               = _mm_mul_ps(fscal,dz20);
1056
1057             /* Update vectorial force */
1058             fix2             = _mm_add_ps(fix2,tx);
1059             fiy2             = _mm_add_ps(fiy2,ty);
1060             fiz2             = _mm_add_ps(fiz2,tz);
1061
1062             fjx0             = _mm_add_ps(fjx0,tx);
1063             fjy0             = _mm_add_ps(fjy0,ty);
1064             fjz0             = _mm_add_ps(fjz0,tz);
1065
1066             }
1067
1068             /**************************
1069              * CALCULATE INTERACTIONS *
1070              **************************/
1071
1072             if (gmx_mm_any_lt(rsq30,rcutoff2))
1073             {
1074
1075             r30              = _mm_mul_ps(rsq30,rinv30);
1076
1077             /* Compute parameters for interactions between i and j atoms */
1078             qq30             = _mm_mul_ps(iq3,jq0);
1079
1080             /* EWALD ELECTROSTATICS */
1081
1082             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1083             ewrt             = _mm_mul_ps(r30,ewtabscale);
1084             ewitab           = _mm_cvttps_epi32(ewrt);
1085             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1086             ewitab           = _mm_slli_epi32(ewitab,2);
1087             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1088             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1089             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1090             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1091             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1092             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1093             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1094             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
1095             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1096
1097             d                = _mm_sub_ps(r30,rswitch);
1098             d                = _mm_max_ps(d,_mm_setzero_ps());
1099             d2               = _mm_mul_ps(d,d);
1100             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)))))));
1101
1102             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1103
1104             /* Evaluate switch function */
1105             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1106             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv30,_mm_mul_ps(velec,dsw)) );
1107             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1108
1109             fscal            = felec;
1110
1111             fscal            = _mm_and_ps(fscal,cutoff_mask);
1112
1113             /* Calculate temporary vectorial force */
1114             tx               = _mm_mul_ps(fscal,dx30);
1115             ty               = _mm_mul_ps(fscal,dy30);
1116             tz               = _mm_mul_ps(fscal,dz30);
1117
1118             /* Update vectorial force */
1119             fix3             = _mm_add_ps(fix3,tx);
1120             fiy3             = _mm_add_ps(fiy3,ty);
1121             fiz3             = _mm_add_ps(fiz3,tz);
1122
1123             fjx0             = _mm_add_ps(fjx0,tx);
1124             fjy0             = _mm_add_ps(fjy0,ty);
1125             fjz0             = _mm_add_ps(fjz0,tz);
1126
1127             }
1128
1129             fjptrA             = f+j_coord_offsetA;
1130             fjptrB             = f+j_coord_offsetB;
1131             fjptrC             = f+j_coord_offsetC;
1132             fjptrD             = f+j_coord_offsetD;
1133
1134             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1135
1136             /* Inner loop uses 186 flops */
1137         }
1138
1139         if(jidx<j_index_end)
1140         {
1141
1142             /* Get j neighbor index, and coordinate index */
1143             jnrlistA         = jjnr[jidx];
1144             jnrlistB         = jjnr[jidx+1];
1145             jnrlistC         = jjnr[jidx+2];
1146             jnrlistD         = jjnr[jidx+3];
1147             /* Sign of each element will be negative for non-real atoms.
1148              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1149              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1150              */
1151             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1152             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1153             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1154             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1155             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1156             j_coord_offsetA  = DIM*jnrA;
1157             j_coord_offsetB  = DIM*jnrB;
1158             j_coord_offsetC  = DIM*jnrC;
1159             j_coord_offsetD  = DIM*jnrD;
1160
1161             /* load j atom coordinates */
1162             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1163                                               x+j_coord_offsetC,x+j_coord_offsetD,
1164                                               &jx0,&jy0,&jz0);
1165
1166             /* Calculate displacement vector */
1167             dx10             = _mm_sub_ps(ix1,jx0);
1168             dy10             = _mm_sub_ps(iy1,jy0);
1169             dz10             = _mm_sub_ps(iz1,jz0);
1170             dx20             = _mm_sub_ps(ix2,jx0);
1171             dy20             = _mm_sub_ps(iy2,jy0);
1172             dz20             = _mm_sub_ps(iz2,jz0);
1173             dx30             = _mm_sub_ps(ix3,jx0);
1174             dy30             = _mm_sub_ps(iy3,jy0);
1175             dz30             = _mm_sub_ps(iz3,jz0);
1176
1177             /* Calculate squared distance and things based on it */
1178             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1179             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1180             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1181
1182             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1183             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1184             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1185
1186             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1187             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1188             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1189
1190             /* Load parameters for j particles */
1191             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1192                                                               charge+jnrC+0,charge+jnrD+0);
1193
1194             fjx0             = _mm_setzero_ps();
1195             fjy0             = _mm_setzero_ps();
1196             fjz0             = _mm_setzero_ps();
1197
1198             /**************************
1199              * CALCULATE INTERACTIONS *
1200              **************************/
1201
1202             if (gmx_mm_any_lt(rsq10,rcutoff2))
1203             {
1204
1205             r10              = _mm_mul_ps(rsq10,rinv10);
1206             r10              = _mm_andnot_ps(dummy_mask,r10);
1207
1208             /* Compute parameters for interactions between i and j atoms */
1209             qq10             = _mm_mul_ps(iq1,jq0);
1210
1211             /* EWALD ELECTROSTATICS */
1212
1213             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1214             ewrt             = _mm_mul_ps(r10,ewtabscale);
1215             ewitab           = _mm_cvttps_epi32(ewrt);
1216             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1217             ewitab           = _mm_slli_epi32(ewitab,2);
1218             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1219             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1220             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1221             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1222             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1223             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1224             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1225             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
1226             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1227
1228             d                = _mm_sub_ps(r10,rswitch);
1229             d                = _mm_max_ps(d,_mm_setzero_ps());
1230             d2               = _mm_mul_ps(d,d);
1231             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)))))));
1232
1233             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1234
1235             /* Evaluate switch function */
1236             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1237             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
1238             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1239
1240             fscal            = felec;
1241
1242             fscal            = _mm_and_ps(fscal,cutoff_mask);
1243
1244             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1245
1246             /* Calculate temporary vectorial force */
1247             tx               = _mm_mul_ps(fscal,dx10);
1248             ty               = _mm_mul_ps(fscal,dy10);
1249             tz               = _mm_mul_ps(fscal,dz10);
1250
1251             /* Update vectorial force */
1252             fix1             = _mm_add_ps(fix1,tx);
1253             fiy1             = _mm_add_ps(fiy1,ty);
1254             fiz1             = _mm_add_ps(fiz1,tz);
1255
1256             fjx0             = _mm_add_ps(fjx0,tx);
1257             fjy0             = _mm_add_ps(fjy0,ty);
1258             fjz0             = _mm_add_ps(fjz0,tz);
1259
1260             }
1261
1262             /**************************
1263              * CALCULATE INTERACTIONS *
1264              **************************/
1265
1266             if (gmx_mm_any_lt(rsq20,rcutoff2))
1267             {
1268
1269             r20              = _mm_mul_ps(rsq20,rinv20);
1270             r20              = _mm_andnot_ps(dummy_mask,r20);
1271
1272             /* Compute parameters for interactions between i and j atoms */
1273             qq20             = _mm_mul_ps(iq2,jq0);
1274
1275             /* EWALD ELECTROSTATICS */
1276
1277             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1278             ewrt             = _mm_mul_ps(r20,ewtabscale);
1279             ewitab           = _mm_cvttps_epi32(ewrt);
1280             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1281             ewitab           = _mm_slli_epi32(ewitab,2);
1282             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1283             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1284             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1285             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1286             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1287             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1288             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1289             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
1290             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1291
1292             d                = _mm_sub_ps(r20,rswitch);
1293             d                = _mm_max_ps(d,_mm_setzero_ps());
1294             d2               = _mm_mul_ps(d,d);
1295             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)))))));
1296
1297             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1298
1299             /* Evaluate switch function */
1300             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1301             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
1302             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1303
1304             fscal            = felec;
1305
1306             fscal            = _mm_and_ps(fscal,cutoff_mask);
1307
1308             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1309
1310             /* Calculate temporary vectorial force */
1311             tx               = _mm_mul_ps(fscal,dx20);
1312             ty               = _mm_mul_ps(fscal,dy20);
1313             tz               = _mm_mul_ps(fscal,dz20);
1314
1315             /* Update vectorial force */
1316             fix2             = _mm_add_ps(fix2,tx);
1317             fiy2             = _mm_add_ps(fiy2,ty);
1318             fiz2             = _mm_add_ps(fiz2,tz);
1319
1320             fjx0             = _mm_add_ps(fjx0,tx);
1321             fjy0             = _mm_add_ps(fjy0,ty);
1322             fjz0             = _mm_add_ps(fjz0,tz);
1323
1324             }
1325
1326             /**************************
1327              * CALCULATE INTERACTIONS *
1328              **************************/
1329
1330             if (gmx_mm_any_lt(rsq30,rcutoff2))
1331             {
1332
1333             r30              = _mm_mul_ps(rsq30,rinv30);
1334             r30              = _mm_andnot_ps(dummy_mask,r30);
1335
1336             /* Compute parameters for interactions between i and j atoms */
1337             qq30             = _mm_mul_ps(iq3,jq0);
1338
1339             /* EWALD ELECTROSTATICS */
1340
1341             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1342             ewrt             = _mm_mul_ps(r30,ewtabscale);
1343             ewitab           = _mm_cvttps_epi32(ewrt);
1344             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1345             ewitab           = _mm_slli_epi32(ewitab,2);
1346             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1347             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1348             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1349             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1350             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1351             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1352             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1353             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
1354             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1355
1356             d                = _mm_sub_ps(r30,rswitch);
1357             d                = _mm_max_ps(d,_mm_setzero_ps());
1358             d2               = _mm_mul_ps(d,d);
1359             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)))))));
1360
1361             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1362
1363             /* Evaluate switch function */
1364             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1365             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv30,_mm_mul_ps(velec,dsw)) );
1366             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1367
1368             fscal            = felec;
1369
1370             fscal            = _mm_and_ps(fscal,cutoff_mask);
1371
1372             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1373
1374             /* Calculate temporary vectorial force */
1375             tx               = _mm_mul_ps(fscal,dx30);
1376             ty               = _mm_mul_ps(fscal,dy30);
1377             tz               = _mm_mul_ps(fscal,dz30);
1378
1379             /* Update vectorial force */
1380             fix3             = _mm_add_ps(fix3,tx);
1381             fiy3             = _mm_add_ps(fiy3,ty);
1382             fiz3             = _mm_add_ps(fiz3,tz);
1383
1384             fjx0             = _mm_add_ps(fjx0,tx);
1385             fjy0             = _mm_add_ps(fjy0,ty);
1386             fjz0             = _mm_add_ps(fjz0,tz);
1387
1388             }
1389
1390             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1391             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1392             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1393             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1394
1395             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1396
1397             /* Inner loop uses 189 flops */
1398         }
1399
1400         /* End of innermost loop */
1401
1402         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1403                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
1404
1405         /* Increment number of inner iterations */
1406         inneriter                  += j_index_end - j_index_start;
1407
1408         /* Outer loop uses 18 flops */
1409     }
1410
1411     /* Increment number of outer iterations */
1412     outeriter        += nri;
1413
1414     /* Update outer/inner flops */
1415
1416     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*189);
1417 }