src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEw_VdwCSTab_GeomW3W3_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 "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "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_ElecEw_VdwCSTab_GeomW3W3_VF_sse2_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water3-Water3
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwCSTab_GeomW3W3_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              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  92     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  93     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  94     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  95     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  96     __m128           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
  97     __m128           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
  98     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  99     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 100     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 101     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 102     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 103     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 104     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 105     real             *charge;
 106     int              nvdwtype;
 107     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 108     int              *vdwtype;
 109     real             *vdwparam;
 110     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 111     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 112     __m128i          vfitab;
 113     __m128i          ifour       = _mm_set1_epi32(4);
 114     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 115     real             *vftab;
 116     __m128i          ewitab;
 117     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 118     real             *ewtab;
 119     __m128           dummy_mask,cutoff_mask;
 120     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 121     __m128           one     = _mm_set1_ps(1.0);
 122     __m128           two     = _mm_set1_ps(2.0);
 123     x                = xx[0];
 124     f                = ff[0];
 125
 126     nri              = nlist->nri;
 127     iinr             = nlist->iinr;
 128     jindex           = nlist->jindex;
 129     jjnr             = nlist->jjnr;
 130     shiftidx         = nlist->shift;
 131     gid              = nlist->gid;
 132     shiftvec         = fr->shift_vec[0];
 133     fshift           = fr->fshift[0];
 134     facel            = _mm_set1_ps(fr->epsfac);
 135     charge           = mdatoms->chargeA;
 136     nvdwtype         = fr->ntype;
 137     vdwparam         = fr->nbfp;
 138     vdwtype          = mdatoms->typeA;
 139
 140     vftab            = kernel_data->table_vdw->data;
 141     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 142
 143     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 144     ewtab            = fr->ic->tabq_coul_FDV0;
 145     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 146     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 147
 148     /* Setup water-specific parameters */
 149     inr              = nlist->iinr[0];
 150     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 151     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 152     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 153     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 154
 155     jq0              = _mm_set1_ps(charge[inr+0]);
 156     jq1              = _mm_set1_ps(charge[inr+1]);
 157     jq2              = _mm_set1_ps(charge[inr+2]);
 158     vdwjidx0A        = 2*vdwtype[inr+0];
 159     qq00             = _mm_mul_ps(iq0,jq0);
 160     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
 161     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
 162     qq01             = _mm_mul_ps(iq0,jq1);
 163     qq02             = _mm_mul_ps(iq0,jq2);
 164     qq10             = _mm_mul_ps(iq1,jq0);
 165     qq11             = _mm_mul_ps(iq1,jq1);
 166     qq12             = _mm_mul_ps(iq1,jq2);
 167     qq20             = _mm_mul_ps(iq2,jq0);
 168     qq21             = _mm_mul_ps(iq2,jq1);
 169     qq22             = _mm_mul_ps(iq2,jq2);
 170
 171     /* Avoid stupid compiler warnings */
 172     jnrA = jnrB = jnrC = jnrD = 0;
 173     j_coord_offsetA = 0;
 174     j_coord_offsetB = 0;
 175     j_coord_offsetC = 0;
 176     j_coord_offsetD = 0;
 177
 178     outeriter        = 0;
 179     inneriter        = 0;
 180
 181     for(iidx=0;iidx<4*DIM;iidx++)
 182     {
 183         scratch[iidx] = 0.0;
 184     }
 185
 186     /* Start outer loop over neighborlists */
 187     for(iidx=0; iidx<nri; iidx++)
 188     {
 189         /* Load shift vector for this list */
 190         i_shift_offset   = DIM*shiftidx[iidx];
 191
 192         /* Load limits for loop over neighbors */
 193         j_index_start    = jindex[iidx];
 194         j_index_end      = jindex[iidx+1];
 195
 196         /* Get outer coordinate index */
 197         inr              = iinr[iidx];
 198         i_coord_offset   = DIM*inr;
 199
 200         /* Load i particle coords and add shift vector */
 201         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 202                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 203
 204         fix0             = _mm_setzero_ps();
 205         fiy0             = _mm_setzero_ps();
 206         fiz0             = _mm_setzero_ps();
 207         fix1             = _mm_setzero_ps();
 208         fiy1             = _mm_setzero_ps();
 209         fiz1             = _mm_setzero_ps();
 210         fix2             = _mm_setzero_ps();
 211         fiy2             = _mm_setzero_ps();
 212         fiz2             = _mm_setzero_ps();
 213
 214         /* Reset potential sums */
 215         velecsum         = _mm_setzero_ps();
 216         vvdwsum          = _mm_setzero_ps();
 217
 218         /* Start inner kernel loop */
 219         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 220         {
 221
 222             /* Get j neighbor index, and coordinate index */
 223             jnrA             = jjnr[jidx];
 224             jnrB             = jjnr[jidx+1];
 225             jnrC             = jjnr[jidx+2];
 226             jnrD             = jjnr[jidx+3];
 227             j_coord_offsetA  = DIM*jnrA;
 228             j_coord_offsetB  = DIM*jnrB;
 229             j_coord_offsetC  = DIM*jnrC;
 230             j_coord_offsetD  = DIM*jnrD;
 231
 232             /* load j atom coordinates */
 233             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 234                                               x+j_coord_offsetC,x+j_coord_offsetD,
 235                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 236
 237             /* Calculate displacement vector */
 238             dx00             = _mm_sub_ps(ix0,jx0);
 239             dy00             = _mm_sub_ps(iy0,jy0);
 240             dz00             = _mm_sub_ps(iz0,jz0);
 241             dx01             = _mm_sub_ps(ix0,jx1);
 242             dy01             = _mm_sub_ps(iy0,jy1);
 243             dz01             = _mm_sub_ps(iz0,jz1);
 244             dx02             = _mm_sub_ps(ix0,jx2);
 245             dy02             = _mm_sub_ps(iy0,jy2);
 246             dz02             = _mm_sub_ps(iz0,jz2);
 247             dx10             = _mm_sub_ps(ix1,jx0);
 248             dy10             = _mm_sub_ps(iy1,jy0);
 249             dz10             = _mm_sub_ps(iz1,jz0);
 250             dx11             = _mm_sub_ps(ix1,jx1);
 251             dy11             = _mm_sub_ps(iy1,jy1);
 252             dz11             = _mm_sub_ps(iz1,jz1);
 253             dx12             = _mm_sub_ps(ix1,jx2);
 254             dy12             = _mm_sub_ps(iy1,jy2);
 255             dz12             = _mm_sub_ps(iz1,jz2);
 256             dx20             = _mm_sub_ps(ix2,jx0);
 257             dy20             = _mm_sub_ps(iy2,jy0);
 258             dz20             = _mm_sub_ps(iz2,jz0);
 259             dx21             = _mm_sub_ps(ix2,jx1);
 260             dy21             = _mm_sub_ps(iy2,jy1);
 261             dz21             = _mm_sub_ps(iz2,jz1);
 262             dx22             = _mm_sub_ps(ix2,jx2);
 263             dy22             = _mm_sub_ps(iy2,jy2);
 264             dz22             = _mm_sub_ps(iz2,jz2);
 265
 266             /* Calculate squared distance and things based on it */
 267             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 268             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
 269             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
 270             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 271             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 272             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 273             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 274             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 275             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 276
 277             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 278             rinv01           = gmx_mm_invsqrt_ps(rsq01);
 279             rinv02           = gmx_mm_invsqrt_ps(rsq02);
 280             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 281             rinv11           = gmx_mm_invsqrt_ps(rsq11);
 282             rinv12           = gmx_mm_invsqrt_ps(rsq12);
 283             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 284             rinv21           = gmx_mm_invsqrt_ps(rsq21);
 285             rinv22           = gmx_mm_invsqrt_ps(rsq22);
 286
 287             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 288             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
 289             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
 290             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 291             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 292             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 293             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 294             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 295             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 296
 297             fjx0             = _mm_setzero_ps();
 298             fjy0             = _mm_setzero_ps();
 299             fjz0             = _mm_setzero_ps();
 300             fjx1             = _mm_setzero_ps();
 301             fjy1             = _mm_setzero_ps();
 302             fjz1             = _mm_setzero_ps();
 303             fjx2             = _mm_setzero_ps();
 304             fjy2             = _mm_setzero_ps();
 305             fjz2             = _mm_setzero_ps();
 306
 307             /**************************
 308              * CALCULATE INTERACTIONS *
 309              **************************/
 310
 311             r00              = _mm_mul_ps(rsq00,rinv00);
 312
 313             /* Calculate table index by multiplying r with table scale and truncate to integer */
 314             rt               = _mm_mul_ps(r00,vftabscale);
 315             vfitab           = _mm_cvttps_epi32(rt);
 316             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 317             vfitab           = _mm_slli_epi32(vfitab,3);
 318
 319             /* EWALD ELECTROSTATICS */
 320
 321             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 322             ewrt             = _mm_mul_ps(r00,ewtabscale);
 323             ewitab           = _mm_cvttps_epi32(ewrt);
 324             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 325             ewitab           = _mm_slli_epi32(ewitab,2);
 326             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 327             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 328             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 329             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 330             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 331             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 332             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 333             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 334             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 335
 336             /* CUBIC SPLINE TABLE DISPERSION */
 337             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 338             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 339             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 340             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 341             _MM_TRANSPOSE4_PS(Y,F,G,H);
 342             Heps             = _mm_mul_ps(vfeps,H);
 343             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 344             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 345             vvdw6            = _mm_mul_ps(c6_00,VV);
 346             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 347             fvdw6            = _mm_mul_ps(c6_00,FF);
 348
 349             /* CUBIC SPLINE TABLE REPULSION */
 350             vfitab           = _mm_add_epi32(vfitab,ifour);
 351             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 352             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 353             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 354             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 355             _MM_TRANSPOSE4_PS(Y,F,G,H);
 356             Heps             = _mm_mul_ps(vfeps,H);
 357             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 358             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 359             vvdw12           = _mm_mul_ps(c12_00,VV);
 360             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 361             fvdw12           = _mm_mul_ps(c12_00,FF);
 362             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 363             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 364
 365             /* Update potential sum for this i atom from the interaction with this j atom. */
 366             velecsum         = _mm_add_ps(velecsum,velec);
 367             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 368
 369             fscal            = _mm_add_ps(felec,fvdw);
 370
 371             /* Calculate temporary vectorial force */
 372             tx               = _mm_mul_ps(fscal,dx00);
 373             ty               = _mm_mul_ps(fscal,dy00);
 374             tz               = _mm_mul_ps(fscal,dz00);
 375
 376             /* Update vectorial force */
 377             fix0             = _mm_add_ps(fix0,tx);
 378             fiy0             = _mm_add_ps(fiy0,ty);
 379             fiz0             = _mm_add_ps(fiz0,tz);
 380
 381             fjx0             = _mm_add_ps(fjx0,tx);
 382             fjy0             = _mm_add_ps(fjy0,ty);
 383             fjz0             = _mm_add_ps(fjz0,tz);
 384
 385             /**************************
 386              * CALCULATE INTERACTIONS *
 387              **************************/
 388
 389             r01              = _mm_mul_ps(rsq01,rinv01);
 390
 391             /* EWALD ELECTROSTATICS */
 392
 393             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 394             ewrt             = _mm_mul_ps(r01,ewtabscale);
 395             ewitab           = _mm_cvttps_epi32(ewrt);
 396             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 397             ewitab           = _mm_slli_epi32(ewitab,2);
 398             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 399             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 400             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 401             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 402             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 403             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 404             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 405             velec            = _mm_mul_ps(qq01,_mm_sub_ps(rinv01,velec));
 406             felec            = _mm_mul_ps(_mm_mul_ps(qq01,rinv01),_mm_sub_ps(rinvsq01,felec));
 407
 408             /* Update potential sum for this i atom from the interaction with this j atom. */
 409             velecsum         = _mm_add_ps(velecsum,velec);
 410
 411             fscal            = felec;
 412
 413             /* Calculate temporary vectorial force */
 414             tx               = _mm_mul_ps(fscal,dx01);
 415             ty               = _mm_mul_ps(fscal,dy01);
 416             tz               = _mm_mul_ps(fscal,dz01);
 417
 418             /* Update vectorial force */
 419             fix0             = _mm_add_ps(fix0,tx);
 420             fiy0             = _mm_add_ps(fiy0,ty);
 421             fiz0             = _mm_add_ps(fiz0,tz);
 422
 423             fjx1             = _mm_add_ps(fjx1,tx);
 424             fjy1             = _mm_add_ps(fjy1,ty);
 425             fjz1             = _mm_add_ps(fjz1,tz);
 426
 427             /**************************
 428              * CALCULATE INTERACTIONS *
 429              **************************/
 430
 431             r02              = _mm_mul_ps(rsq02,rinv02);
 432
 433             /* EWALD ELECTROSTATICS */
 434
 435             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 436             ewrt             = _mm_mul_ps(r02,ewtabscale);
 437             ewitab           = _mm_cvttps_epi32(ewrt);
 438             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 439             ewitab           = _mm_slli_epi32(ewitab,2);
 440             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 441             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 442             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 443             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 444             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 445             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 446             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 447             velec            = _mm_mul_ps(qq02,_mm_sub_ps(rinv02,velec));
 448             felec            = _mm_mul_ps(_mm_mul_ps(qq02,rinv02),_mm_sub_ps(rinvsq02,felec));
 449
 450             /* Update potential sum for this i atom from the interaction with this j atom. */
 451             velecsum         = _mm_add_ps(velecsum,velec);
 452
 453             fscal            = felec;
 454
 455             /* Calculate temporary vectorial force */
 456             tx               = _mm_mul_ps(fscal,dx02);
 457             ty               = _mm_mul_ps(fscal,dy02);
 458             tz               = _mm_mul_ps(fscal,dz02);
 459
 460             /* Update vectorial force */
 461             fix0             = _mm_add_ps(fix0,tx);
 462             fiy0             = _mm_add_ps(fiy0,ty);
 463             fiz0             = _mm_add_ps(fiz0,tz);
 464
 465             fjx2             = _mm_add_ps(fjx2,tx);
 466             fjy2             = _mm_add_ps(fjy2,ty);
 467             fjz2             = _mm_add_ps(fjz2,tz);
 468
 469             /**************************
 470              * CALCULATE INTERACTIONS *
 471              **************************/
 472
 473             r10              = _mm_mul_ps(rsq10,rinv10);
 474
 475             /* EWALD ELECTROSTATICS */
 476
 477             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 478             ewrt             = _mm_mul_ps(r10,ewtabscale);
 479             ewitab           = _mm_cvttps_epi32(ewrt);
 480             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 481             ewitab           = _mm_slli_epi32(ewitab,2);
 482             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 483             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 484             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 485             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 486             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 487             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 488             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 489             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 490             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 491
 492             /* Update potential sum for this i atom from the interaction with this j atom. */
 493             velecsum         = _mm_add_ps(velecsum,velec);
 494
 495             fscal            = felec;
 496
 497             /* Calculate temporary vectorial force */
 498             tx               = _mm_mul_ps(fscal,dx10);
 499             ty               = _mm_mul_ps(fscal,dy10);
 500             tz               = _mm_mul_ps(fscal,dz10);
 501
 502             /* Update vectorial force */
 503             fix1             = _mm_add_ps(fix1,tx);
 504             fiy1             = _mm_add_ps(fiy1,ty);
 505             fiz1             = _mm_add_ps(fiz1,tz);
 506
 507             fjx0             = _mm_add_ps(fjx0,tx);
 508             fjy0             = _mm_add_ps(fjy0,ty);
 509             fjz0             = _mm_add_ps(fjz0,tz);
 510
 511             /**************************
 512              * CALCULATE INTERACTIONS *
 513              **************************/
 514
 515             r11              = _mm_mul_ps(rsq11,rinv11);
 516
 517             /* EWALD ELECTROSTATICS */
 518
 519             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 520             ewrt             = _mm_mul_ps(r11,ewtabscale);
 521             ewitab           = _mm_cvttps_epi32(ewrt);
 522             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 523             ewitab           = _mm_slli_epi32(ewitab,2);
 524             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 525             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 526             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 527             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 528             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 529             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 530             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 531             velec            = _mm_mul_ps(qq11,_mm_sub_ps(rinv11,velec));
 532             felec            = _mm_mul_ps(_mm_mul_ps(qq11,rinv11),_mm_sub_ps(rinvsq11,felec));
 533
 534             /* Update potential sum for this i atom from the interaction with this j atom. */
 535             velecsum         = _mm_add_ps(velecsum,velec);
 536
 537             fscal            = felec;
 538
 539             /* Calculate temporary vectorial force */
 540             tx               = _mm_mul_ps(fscal,dx11);
 541             ty               = _mm_mul_ps(fscal,dy11);
 542             tz               = _mm_mul_ps(fscal,dz11);
 543
 544             /* Update vectorial force */
 545             fix1             = _mm_add_ps(fix1,tx);
 546             fiy1             = _mm_add_ps(fiy1,ty);
 547             fiz1             = _mm_add_ps(fiz1,tz);
 548
 549             fjx1             = _mm_add_ps(fjx1,tx);
 550             fjy1             = _mm_add_ps(fjy1,ty);
 551             fjz1             = _mm_add_ps(fjz1,tz);
 552
 553             /**************************
 554              * CALCULATE INTERACTIONS *
 555              **************************/
 556
 557             r12              = _mm_mul_ps(rsq12,rinv12);
 558
 559             /* EWALD ELECTROSTATICS */
 560
 561             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 562             ewrt             = _mm_mul_ps(r12,ewtabscale);
 563             ewitab           = _mm_cvttps_epi32(ewrt);
 564             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 565             ewitab           = _mm_slli_epi32(ewitab,2);
 566             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 567             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 568             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 569             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 570             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 571             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 572             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 573             velec            = _mm_mul_ps(qq12,_mm_sub_ps(rinv12,velec));
 574             felec            = _mm_mul_ps(_mm_mul_ps(qq12,rinv12),_mm_sub_ps(rinvsq12,felec));
 575
 576             /* Update potential sum for this i atom from the interaction with this j atom. */
 577             velecsum         = _mm_add_ps(velecsum,velec);
 578
 579             fscal            = felec;
 580
 581             /* Calculate temporary vectorial force */
 582             tx               = _mm_mul_ps(fscal,dx12);
 583             ty               = _mm_mul_ps(fscal,dy12);
 584             tz               = _mm_mul_ps(fscal,dz12);
 585
 586             /* Update vectorial force */
 587             fix1             = _mm_add_ps(fix1,tx);
 588             fiy1             = _mm_add_ps(fiy1,ty);
 589             fiz1             = _mm_add_ps(fiz1,tz);
 590
 591             fjx2             = _mm_add_ps(fjx2,tx);
 592             fjy2             = _mm_add_ps(fjy2,ty);
 593             fjz2             = _mm_add_ps(fjz2,tz);
 594
 595             /**************************
 596              * CALCULATE INTERACTIONS *
 597              **************************/
 598
 599             r20              = _mm_mul_ps(rsq20,rinv20);
 600
 601             /* EWALD ELECTROSTATICS */
 602
 603             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 604             ewrt             = _mm_mul_ps(r20,ewtabscale);
 605             ewitab           = _mm_cvttps_epi32(ewrt);
 606             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 607             ewitab           = _mm_slli_epi32(ewitab,2);
 608             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 609             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 610             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 611             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 612             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 613             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 614             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 615             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 616             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 617
 618             /* Update potential sum for this i atom from the interaction with this j atom. */
 619             velecsum         = _mm_add_ps(velecsum,velec);
 620
 621             fscal            = felec;
 622
 623             /* Calculate temporary vectorial force */
 624             tx               = _mm_mul_ps(fscal,dx20);
 625             ty               = _mm_mul_ps(fscal,dy20);
 626             tz               = _mm_mul_ps(fscal,dz20);
 627
 628             /* Update vectorial force */
 629             fix2             = _mm_add_ps(fix2,tx);
 630             fiy2             = _mm_add_ps(fiy2,ty);
 631             fiz2             = _mm_add_ps(fiz2,tz);
 632
 633             fjx0             = _mm_add_ps(fjx0,tx);
 634             fjy0             = _mm_add_ps(fjy0,ty);
 635             fjz0             = _mm_add_ps(fjz0,tz);
 636
 637             /**************************
 638              * CALCULATE INTERACTIONS *
 639              **************************/
 640
 641             r21              = _mm_mul_ps(rsq21,rinv21);
 642
 643             /* EWALD ELECTROSTATICS */
 644
 645             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 646             ewrt             = _mm_mul_ps(r21,ewtabscale);
 647             ewitab           = _mm_cvttps_epi32(ewrt);
 648             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 649             ewitab           = _mm_slli_epi32(ewitab,2);
 650             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 651             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 652             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 653             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 654             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 655             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 656             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 657             velec            = _mm_mul_ps(qq21,_mm_sub_ps(rinv21,velec));
 658             felec            = _mm_mul_ps(_mm_mul_ps(qq21,rinv21),_mm_sub_ps(rinvsq21,felec));
 659
 660             /* Update potential sum for this i atom from the interaction with this j atom. */
 661             velecsum         = _mm_add_ps(velecsum,velec);
 662
 663             fscal            = felec;
 664
 665             /* Calculate temporary vectorial force */
 666             tx               = _mm_mul_ps(fscal,dx21);
 667             ty               = _mm_mul_ps(fscal,dy21);
 668             tz               = _mm_mul_ps(fscal,dz21);
 669
 670             /* Update vectorial force */
 671             fix2             = _mm_add_ps(fix2,tx);
 672             fiy2             = _mm_add_ps(fiy2,ty);
 673             fiz2             = _mm_add_ps(fiz2,tz);
 674
 675             fjx1             = _mm_add_ps(fjx1,tx);
 676             fjy1             = _mm_add_ps(fjy1,ty);
 677             fjz1             = _mm_add_ps(fjz1,tz);
 678
 679             /**************************
 680              * CALCULATE INTERACTIONS *
 681              **************************/
 682
 683             r22              = _mm_mul_ps(rsq22,rinv22);
 684
 685             /* EWALD ELECTROSTATICS */
 686
 687             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 688             ewrt             = _mm_mul_ps(r22,ewtabscale);
 689             ewitab           = _mm_cvttps_epi32(ewrt);
 690             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 691             ewitab           = _mm_slli_epi32(ewitab,2);
 692             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 693             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 694             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 695             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 696             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 697             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 698             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 699             velec            = _mm_mul_ps(qq22,_mm_sub_ps(rinv22,velec));
 700             felec            = _mm_mul_ps(_mm_mul_ps(qq22,rinv22),_mm_sub_ps(rinvsq22,felec));
 701
 702             /* Update potential sum for this i atom from the interaction with this j atom. */
 703             velecsum         = _mm_add_ps(velecsum,velec);
 704
 705             fscal            = felec;
 706
 707             /* Calculate temporary vectorial force */
 708             tx               = _mm_mul_ps(fscal,dx22);
 709             ty               = _mm_mul_ps(fscal,dy22);
 710             tz               = _mm_mul_ps(fscal,dz22);
 711
 712             /* Update vectorial force */
 713             fix2             = _mm_add_ps(fix2,tx);
 714             fiy2             = _mm_add_ps(fiy2,ty);
 715             fiz2             = _mm_add_ps(fiz2,tz);
 716
 717             fjx2             = _mm_add_ps(fjx2,tx);
 718             fjy2             = _mm_add_ps(fjy2,ty);
 719             fjz2             = _mm_add_ps(fjz2,tz);
 720
 721             fjptrA             = f+j_coord_offsetA;
 722             fjptrB             = f+j_coord_offsetB;
 723             fjptrC             = f+j_coord_offsetC;
 724             fjptrD             = f+j_coord_offsetD;
 725
 726             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 727                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 728
 729             /* Inner loop uses 403 flops */
 730         }
 731
 732         if(jidx<j_index_end)
 733         {
 734
 735             /* Get j neighbor index, and coordinate index */
 736             jnrlistA         = jjnr[jidx];
 737             jnrlistB         = jjnr[jidx+1];
 738             jnrlistC         = jjnr[jidx+2];
 739             jnrlistD         = jjnr[jidx+3];
 740             /* Sign of each element will be negative for non-real atoms.
 741              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 742              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 743              */
 744             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 745             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 746             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 747             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 748             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 749             j_coord_offsetA  = DIM*jnrA;
 750             j_coord_offsetB  = DIM*jnrB;
 751             j_coord_offsetC  = DIM*jnrC;
 752             j_coord_offsetD  = DIM*jnrD;
 753
 754             /* load j atom coordinates */
 755             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 756                                               x+j_coord_offsetC,x+j_coord_offsetD,
 757                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 758
 759             /* Calculate displacement vector */
 760             dx00             = _mm_sub_ps(ix0,jx0);
 761             dy00             = _mm_sub_ps(iy0,jy0);
 762             dz00             = _mm_sub_ps(iz0,jz0);
 763             dx01             = _mm_sub_ps(ix0,jx1);
 764             dy01             = _mm_sub_ps(iy0,jy1);
 765             dz01             = _mm_sub_ps(iz0,jz1);
 766             dx02             = _mm_sub_ps(ix0,jx2);
 767             dy02             = _mm_sub_ps(iy0,jy2);
 768             dz02             = _mm_sub_ps(iz0,jz2);
 769             dx10             = _mm_sub_ps(ix1,jx0);
 770             dy10             = _mm_sub_ps(iy1,jy0);
 771             dz10             = _mm_sub_ps(iz1,jz0);
 772             dx11             = _mm_sub_ps(ix1,jx1);
 773             dy11             = _mm_sub_ps(iy1,jy1);
 774             dz11             = _mm_sub_ps(iz1,jz1);
 775             dx12             = _mm_sub_ps(ix1,jx2);
 776             dy12             = _mm_sub_ps(iy1,jy2);
 777             dz12             = _mm_sub_ps(iz1,jz2);
 778             dx20             = _mm_sub_ps(ix2,jx0);
 779             dy20             = _mm_sub_ps(iy2,jy0);
 780             dz20             = _mm_sub_ps(iz2,jz0);
 781             dx21             = _mm_sub_ps(ix2,jx1);
 782             dy21             = _mm_sub_ps(iy2,jy1);
 783             dz21             = _mm_sub_ps(iz2,jz1);
 784             dx22             = _mm_sub_ps(ix2,jx2);
 785             dy22             = _mm_sub_ps(iy2,jy2);
 786             dz22             = _mm_sub_ps(iz2,jz2);
 787
 788             /* Calculate squared distance and things based on it */
 789             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 790             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
 791             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
 792             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 793             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 794             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 795             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 796             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 797             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 798
 799             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 800             rinv01           = gmx_mm_invsqrt_ps(rsq01);
 801             rinv02           = gmx_mm_invsqrt_ps(rsq02);
 802             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 803             rinv11           = gmx_mm_invsqrt_ps(rsq11);
 804             rinv12           = gmx_mm_invsqrt_ps(rsq12);
 805             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 806             rinv21           = gmx_mm_invsqrt_ps(rsq21);
 807             rinv22           = gmx_mm_invsqrt_ps(rsq22);
 808
 809             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 810             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
 811             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
 812             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 813             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 814             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 815             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 816             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 817             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 818
 819             fjx0             = _mm_setzero_ps();
 820             fjy0             = _mm_setzero_ps();
 821             fjz0             = _mm_setzero_ps();
 822             fjx1             = _mm_setzero_ps();
 823             fjy1             = _mm_setzero_ps();
 824             fjz1             = _mm_setzero_ps();
 825             fjx2             = _mm_setzero_ps();
 826             fjy2             = _mm_setzero_ps();
 827             fjz2             = _mm_setzero_ps();
 828
 829             /**************************
 830              * CALCULATE INTERACTIONS *
 831              **************************/
 832
 833             r00              = _mm_mul_ps(rsq00,rinv00);
 834             r00              = _mm_andnot_ps(dummy_mask,r00);
 835
 836             /* Calculate table index by multiplying r with table scale and truncate to integer */
 837             rt               = _mm_mul_ps(r00,vftabscale);
 838             vfitab           = _mm_cvttps_epi32(rt);
 839             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 840             vfitab           = _mm_slli_epi32(vfitab,3);
 841
 842             /* EWALD ELECTROSTATICS */
 843
 844             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 845             ewrt             = _mm_mul_ps(r00,ewtabscale);
 846             ewitab           = _mm_cvttps_epi32(ewrt);
 847             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 848             ewitab           = _mm_slli_epi32(ewitab,2);
 849             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 850             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 851             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 852             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 853             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 854             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 855             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 856             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 857             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 858
 859             /* CUBIC SPLINE TABLE DISPERSION */
 860             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 861             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 862             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 863             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 864             _MM_TRANSPOSE4_PS(Y,F,G,H);
 865             Heps             = _mm_mul_ps(vfeps,H);
 866             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 867             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 868             vvdw6            = _mm_mul_ps(c6_00,VV);
 869             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 870             fvdw6            = _mm_mul_ps(c6_00,FF);
 871
 872             /* CUBIC SPLINE TABLE REPULSION */
 873             vfitab           = _mm_add_epi32(vfitab,ifour);
 874             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 875             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 876             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 877             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 878             _MM_TRANSPOSE4_PS(Y,F,G,H);
 879             Heps             = _mm_mul_ps(vfeps,H);
 880             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 881             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 882             vvdw12           = _mm_mul_ps(c12_00,VV);
 883             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 884             fvdw12           = _mm_mul_ps(c12_00,FF);
 885             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 886             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 887
 888             /* Update potential sum for this i atom from the interaction with this j atom. */
 889             velec            = _mm_andnot_ps(dummy_mask,velec);
 890             velecsum         = _mm_add_ps(velecsum,velec);
 891             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 892             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 893
 894             fscal            = _mm_add_ps(felec,fvdw);
 895
 896             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 897
 898             /* Calculate temporary vectorial force */
 899             tx               = _mm_mul_ps(fscal,dx00);
 900             ty               = _mm_mul_ps(fscal,dy00);
 901             tz               = _mm_mul_ps(fscal,dz00);
 902
 903             /* Update vectorial force */
 904             fix0             = _mm_add_ps(fix0,tx);
 905             fiy0             = _mm_add_ps(fiy0,ty);
 906             fiz0             = _mm_add_ps(fiz0,tz);
 907
 908             fjx0             = _mm_add_ps(fjx0,tx);
 909             fjy0             = _mm_add_ps(fjy0,ty);
 910             fjz0             = _mm_add_ps(fjz0,tz);
 911
 912             /**************************
 913              * CALCULATE INTERACTIONS *
 914              **************************/
 915
 916             r01              = _mm_mul_ps(rsq01,rinv01);
 917             r01              = _mm_andnot_ps(dummy_mask,r01);
 918
 919             /* EWALD ELECTROSTATICS */
 920
 921             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 922             ewrt             = _mm_mul_ps(r01,ewtabscale);
 923             ewitab           = _mm_cvttps_epi32(ewrt);
 924             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 925             ewitab           = _mm_slli_epi32(ewitab,2);
 926             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 927             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 928             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 929             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 930             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 931             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 932             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 933             velec            = _mm_mul_ps(qq01,_mm_sub_ps(rinv01,velec));
 934             felec            = _mm_mul_ps(_mm_mul_ps(qq01,rinv01),_mm_sub_ps(rinvsq01,felec));
 935
 936             /* Update potential sum for this i atom from the interaction with this j atom. */
 937             velec            = _mm_andnot_ps(dummy_mask,velec);
 938             velecsum         = _mm_add_ps(velecsum,velec);
 939
 940             fscal            = felec;
 941
 942             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 943
 944             /* Calculate temporary vectorial force */
 945             tx               = _mm_mul_ps(fscal,dx01);
 946             ty               = _mm_mul_ps(fscal,dy01);
 947             tz               = _mm_mul_ps(fscal,dz01);
 948
 949             /* Update vectorial force */
 950             fix0             = _mm_add_ps(fix0,tx);
 951             fiy0             = _mm_add_ps(fiy0,ty);
 952             fiz0             = _mm_add_ps(fiz0,tz);
 953
 954             fjx1             = _mm_add_ps(fjx1,tx);
 955             fjy1             = _mm_add_ps(fjy1,ty);
 956             fjz1             = _mm_add_ps(fjz1,tz);
 957
 958             /**************************
 959              * CALCULATE INTERACTIONS *
 960              **************************/
 961
 962             r02              = _mm_mul_ps(rsq02,rinv02);
 963             r02              = _mm_andnot_ps(dummy_mask,r02);
 964
 965             /* EWALD ELECTROSTATICS */
 966
 967             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 968             ewrt             = _mm_mul_ps(r02,ewtabscale);
 969             ewitab           = _mm_cvttps_epi32(ewrt);
 970             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 971             ewitab           = _mm_slli_epi32(ewitab,2);
 972             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 973             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 974             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 975             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 976             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 977             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 978             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 979             velec            = _mm_mul_ps(qq02,_mm_sub_ps(rinv02,velec));
 980             felec            = _mm_mul_ps(_mm_mul_ps(qq02,rinv02),_mm_sub_ps(rinvsq02,felec));
 981
 982             /* Update potential sum for this i atom from the interaction with this j atom. */
 983             velec            = _mm_andnot_ps(dummy_mask,velec);
 984             velecsum         = _mm_add_ps(velecsum,velec);
 985
 986             fscal            = felec;
 987
 988             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 989
 990             /* Calculate temporary vectorial force */
 991             tx               = _mm_mul_ps(fscal,dx02);
 992             ty               = _mm_mul_ps(fscal,dy02);
 993             tz               = _mm_mul_ps(fscal,dz02);
 994
 995             /* Update vectorial force */
 996             fix0             = _mm_add_ps(fix0,tx);
 997             fiy0             = _mm_add_ps(fiy0,ty);
 998             fiz0             = _mm_add_ps(fiz0,tz);
 999
1000             fjx2             = _mm_add_ps(fjx2,tx);
1001             fjy2             = _mm_add_ps(fjy2,ty);
1002             fjz2             = _mm_add_ps(fjz2,tz);
1003
1004             /**************************
1005              * CALCULATE INTERACTIONS *
1006              **************************/
1007
1008             r10              = _mm_mul_ps(rsq10,rinv10);
1009             r10              = _mm_andnot_ps(dummy_mask,r10);
1010
1011             /* EWALD ELECTROSTATICS */
1012
1013             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1014             ewrt             = _mm_mul_ps(r10,ewtabscale);
1015             ewitab           = _mm_cvttps_epi32(ewrt);
1016             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1017             ewitab           = _mm_slli_epi32(ewitab,2);
1018             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1019             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1020             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1021             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1022             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1023             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1024             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1025             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
1026             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1027
1028             /* Update potential sum for this i atom from the interaction with this j atom. */
1029             velec            = _mm_andnot_ps(dummy_mask,velec);
1030             velecsum         = _mm_add_ps(velecsum,velec);
1031
1032             fscal            = felec;
1033
1034             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1035
1036             /* Calculate temporary vectorial force */
1037             tx               = _mm_mul_ps(fscal,dx10);
1038             ty               = _mm_mul_ps(fscal,dy10);
1039             tz               = _mm_mul_ps(fscal,dz10);
1040
1041             /* Update vectorial force */
1042             fix1             = _mm_add_ps(fix1,tx);
1043             fiy1             = _mm_add_ps(fiy1,ty);
1044             fiz1             = _mm_add_ps(fiz1,tz);
1045
1046             fjx0             = _mm_add_ps(fjx0,tx);
1047             fjy0             = _mm_add_ps(fjy0,ty);
1048             fjz0             = _mm_add_ps(fjz0,tz);
1049
1050             /**************************
1051              * CALCULATE INTERACTIONS *
1052              **************************/
1053
1054             r11              = _mm_mul_ps(rsq11,rinv11);
1055             r11              = _mm_andnot_ps(dummy_mask,r11);
1056
1057             /* EWALD ELECTROSTATICS */
1058
1059             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1060             ewrt             = _mm_mul_ps(r11,ewtabscale);
1061             ewitab           = _mm_cvttps_epi32(ewrt);
1062             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1063             ewitab           = _mm_slli_epi32(ewitab,2);
1064             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1065             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1066             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1067             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1068             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1069             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1070             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1071             velec            = _mm_mul_ps(qq11,_mm_sub_ps(rinv11,velec));
1072             felec            = _mm_mul_ps(_mm_mul_ps(qq11,rinv11),_mm_sub_ps(rinvsq11,felec));
1073
1074             /* Update potential sum for this i atom from the interaction with this j atom. */
1075             velec            = _mm_andnot_ps(dummy_mask,velec);
1076             velecsum         = _mm_add_ps(velecsum,velec);
1077
1078             fscal            = felec;
1079
1080             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1081
1082             /* Calculate temporary vectorial force */
1083             tx               = _mm_mul_ps(fscal,dx11);
1084             ty               = _mm_mul_ps(fscal,dy11);
1085             tz               = _mm_mul_ps(fscal,dz11);
1086
1087             /* Update vectorial force */
1088             fix1             = _mm_add_ps(fix1,tx);
1089             fiy1             = _mm_add_ps(fiy1,ty);
1090             fiz1             = _mm_add_ps(fiz1,tz);
1091
1092             fjx1             = _mm_add_ps(fjx1,tx);
1093             fjy1             = _mm_add_ps(fjy1,ty);
1094             fjz1             = _mm_add_ps(fjz1,tz);
1095
1096             /**************************
1097              * CALCULATE INTERACTIONS *
1098              **************************/
1099
1100             r12              = _mm_mul_ps(rsq12,rinv12);
1101             r12              = _mm_andnot_ps(dummy_mask,r12);
1102
1103             /* EWALD ELECTROSTATICS */
1104
1105             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1106             ewrt             = _mm_mul_ps(r12,ewtabscale);
1107             ewitab           = _mm_cvttps_epi32(ewrt);
1108             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1109             ewitab           = _mm_slli_epi32(ewitab,2);
1110             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1111             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1112             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1113             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1114             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1115             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1116             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1117             velec            = _mm_mul_ps(qq12,_mm_sub_ps(rinv12,velec));
1118             felec            = _mm_mul_ps(_mm_mul_ps(qq12,rinv12),_mm_sub_ps(rinvsq12,felec));
1119
1120             /* Update potential sum for this i atom from the interaction with this j atom. */
1121             velec            = _mm_andnot_ps(dummy_mask,velec);
1122             velecsum         = _mm_add_ps(velecsum,velec);
1123
1124             fscal            = felec;
1125
1126             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1127
1128             /* Calculate temporary vectorial force */
1129             tx               = _mm_mul_ps(fscal,dx12);
1130             ty               = _mm_mul_ps(fscal,dy12);
1131             tz               = _mm_mul_ps(fscal,dz12);
1132
1133             /* Update vectorial force */
1134             fix1             = _mm_add_ps(fix1,tx);
1135             fiy1             = _mm_add_ps(fiy1,ty);
1136             fiz1             = _mm_add_ps(fiz1,tz);
1137
1138             fjx2             = _mm_add_ps(fjx2,tx);
1139             fjy2             = _mm_add_ps(fjy2,ty);
1140             fjz2             = _mm_add_ps(fjz2,tz);
1141
1142             /**************************
1143              * CALCULATE INTERACTIONS *
1144              **************************/
1145
1146             r20              = _mm_mul_ps(rsq20,rinv20);
1147             r20              = _mm_andnot_ps(dummy_mask,r20);
1148
1149             /* EWALD ELECTROSTATICS */
1150
1151             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1152             ewrt             = _mm_mul_ps(r20,ewtabscale);
1153             ewitab           = _mm_cvttps_epi32(ewrt);
1154             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1155             ewitab           = _mm_slli_epi32(ewitab,2);
1156             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1157             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1158             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1159             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1160             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1161             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1162             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1163             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
1164             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1165
1166             /* Update potential sum for this i atom from the interaction with this j atom. */
1167             velec            = _mm_andnot_ps(dummy_mask,velec);
1168             velecsum         = _mm_add_ps(velecsum,velec);
1169
1170             fscal            = felec;
1171
1172             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1173
1174             /* Calculate temporary vectorial force */
1175             tx               = _mm_mul_ps(fscal,dx20);
1176             ty               = _mm_mul_ps(fscal,dy20);
1177             tz               = _mm_mul_ps(fscal,dz20);
1178
1179             /* Update vectorial force */
1180             fix2             = _mm_add_ps(fix2,tx);
1181             fiy2             = _mm_add_ps(fiy2,ty);
1182             fiz2             = _mm_add_ps(fiz2,tz);
1183
1184             fjx0             = _mm_add_ps(fjx0,tx);
1185             fjy0             = _mm_add_ps(fjy0,ty);
1186             fjz0             = _mm_add_ps(fjz0,tz);
1187
1188             /**************************
1189              * CALCULATE INTERACTIONS *
1190              **************************/
1191
1192             r21              = _mm_mul_ps(rsq21,rinv21);
1193             r21              = _mm_andnot_ps(dummy_mask,r21);
1194
1195             /* EWALD ELECTROSTATICS */
1196
1197             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1198             ewrt             = _mm_mul_ps(r21,ewtabscale);
1199             ewitab           = _mm_cvttps_epi32(ewrt);
1200             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1201             ewitab           = _mm_slli_epi32(ewitab,2);
1202             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1203             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1204             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1205             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1206             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1207             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1208             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1209             velec            = _mm_mul_ps(qq21,_mm_sub_ps(rinv21,velec));
1210             felec            = _mm_mul_ps(_mm_mul_ps(qq21,rinv21),_mm_sub_ps(rinvsq21,felec));
1211
1212             /* Update potential sum for this i atom from the interaction with this j atom. */
1213             velec            = _mm_andnot_ps(dummy_mask,velec);
1214             velecsum         = _mm_add_ps(velecsum,velec);
1215
1216             fscal            = felec;
1217
1218             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1219
1220             /* Calculate temporary vectorial force */
1221             tx               = _mm_mul_ps(fscal,dx21);
1222             ty               = _mm_mul_ps(fscal,dy21);
1223             tz               = _mm_mul_ps(fscal,dz21);
1224
1225             /* Update vectorial force */
1226             fix2             = _mm_add_ps(fix2,tx);
1227             fiy2             = _mm_add_ps(fiy2,ty);
1228             fiz2             = _mm_add_ps(fiz2,tz);
1229
1230             fjx1             = _mm_add_ps(fjx1,tx);
1231             fjy1             = _mm_add_ps(fjy1,ty);
1232             fjz1             = _mm_add_ps(fjz1,tz);
1233
1234             /**************************
1235              * CALCULATE INTERACTIONS *
1236              **************************/
1237
1238             r22              = _mm_mul_ps(rsq22,rinv22);
1239             r22              = _mm_andnot_ps(dummy_mask,r22);
1240
1241             /* EWALD ELECTROSTATICS */
1242
1243             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1244             ewrt             = _mm_mul_ps(r22,ewtabscale);
1245             ewitab           = _mm_cvttps_epi32(ewrt);
1246             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1247             ewitab           = _mm_slli_epi32(ewitab,2);
1248             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1249             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1250             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1251             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1252             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1253             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1254             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1255             velec            = _mm_mul_ps(qq22,_mm_sub_ps(rinv22,velec));
1256             felec            = _mm_mul_ps(_mm_mul_ps(qq22,rinv22),_mm_sub_ps(rinvsq22,felec));
1257
1258             /* Update potential sum for this i atom from the interaction with this j atom. */
1259             velec            = _mm_andnot_ps(dummy_mask,velec);
1260             velecsum         = _mm_add_ps(velecsum,velec);
1261
1262             fscal            = felec;
1263
1264             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1265
1266             /* Calculate temporary vectorial force */
1267             tx               = _mm_mul_ps(fscal,dx22);
1268             ty               = _mm_mul_ps(fscal,dy22);
1269             tz               = _mm_mul_ps(fscal,dz22);
1270
1271             /* Update vectorial force */
1272             fix2             = _mm_add_ps(fix2,tx);
1273             fiy2             = _mm_add_ps(fiy2,ty);
1274             fiz2             = _mm_add_ps(fiz2,tz);
1275
1276             fjx2             = _mm_add_ps(fjx2,tx);
1277             fjy2             = _mm_add_ps(fjy2,ty);
1278             fjz2             = _mm_add_ps(fjz2,tz);
1279
1280             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1281             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1282             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1283             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1284
1285             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1286                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1287
1288             /* Inner loop uses 412 flops */
1289         }
1290
1291         /* End of innermost loop */
1292
1293         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1294                                               f+i_coord_offset,fshift+i_shift_offset);
1295
1296         ggid                        = gid[iidx];
1297         /* Update potential energies */
1298         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1299         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1300
1301         /* Increment number of inner iterations */
1302         inneriter                  += j_index_end - j_index_start;
1303
1304         /* Outer loop uses 20 flops */
1305     }
1306
1307     /* Increment number of outer iterations */
1308     outeriter        += nri;
1309
1310     /* Update outer/inner flops */
1311
1312     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*412);
1313 }
1314 /*
1315  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3W3_F_sse2_single
1316  * Electrostatics interaction: Ewald
1317  * VdW interaction:            CubicSplineTable
1318  * Geometry:                   Water3-Water3
1319  * Calculate force/pot:        Force
1320  */
1321 void
1322 nb_kernel_ElecEw_VdwCSTab_GeomW3W3_F_sse2_single
1323                     (t_nblist                    * gmx_restrict       nlist,
1324                      rvec                        * gmx_restrict          xx,
1325                      rvec                        * gmx_restrict          ff,
1326                      t_forcerec                  * gmx_restrict          fr,
1327                      t_mdatoms                   * gmx_restrict     mdatoms,
1328                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1329                      t_nrnb                      * gmx_restrict        nrnb)
1330 {
1331     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1332      * just 0 for non-waters.
1333      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
1334      * jnr indices corresponding to data put in the four positions in the SIMD register.
1335      */
1336     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1337     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1338     int              jnrA,jnrB,jnrC,jnrD;
1339     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1340     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1341     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1342     real             rcutoff_scalar;
1343     real             *shiftvec,*fshift,*x,*f;
1344     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1345     real             scratch[4*DIM];
1346     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1347     int              vdwioffset0;
1348     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1349     int              vdwioffset1;
1350     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1351     int              vdwioffset2;
1352     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1353     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1354     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1355     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1356     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1357     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1358     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1359     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1360     __m128           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1361     __m128           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1362     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1363     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1364     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1365     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1366     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1367     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1368     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
1369     real             *charge;
1370     int              nvdwtype;
1371     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1372     int              *vdwtype;
1373     real             *vdwparam;
1374     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
1375     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
1376     __m128i          vfitab;
1377     __m128i          ifour       = _mm_set1_epi32(4);
1378     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1379     real             *vftab;
1380     __m128i          ewitab;
1381     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1382     real             *ewtab;
1383     __m128           dummy_mask,cutoff_mask;
1384     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
1385     __m128           one     = _mm_set1_ps(1.0);
1386     __m128           two     = _mm_set1_ps(2.0);
1387     x                = xx[0];
1388     f                = ff[0];
1389
1390     nri              = nlist->nri;
1391     iinr             = nlist->iinr;
1392     jindex           = nlist->jindex;
1393     jjnr             = nlist->jjnr;
1394     shiftidx         = nlist->shift;
1395     gid              = nlist->gid;
1396     shiftvec         = fr->shift_vec[0];
1397     fshift           = fr->fshift[0];
1398     facel            = _mm_set1_ps(fr->epsfac);
1399     charge           = mdatoms->chargeA;
1400     nvdwtype         = fr->ntype;
1401     vdwparam         = fr->nbfp;
1402     vdwtype          = mdatoms->typeA;
1403
1404     vftab            = kernel_data->table_vdw->data;
1405     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
1406
1407     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
1408     ewtab            = fr->ic->tabq_coul_F;
1409     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
1410     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
1411
1412     /* Setup water-specific parameters */
1413     inr              = nlist->iinr[0];
1414     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
1415     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
1416     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
1417     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
1418
1419     jq0              = _mm_set1_ps(charge[inr+0]);
1420     jq1              = _mm_set1_ps(charge[inr+1]);
1421     jq2              = _mm_set1_ps(charge[inr+2]);
1422     vdwjidx0A        = 2*vdwtype[inr+0];
1423     qq00             = _mm_mul_ps(iq0,jq0);
1424     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
1425     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
1426     qq01             = _mm_mul_ps(iq0,jq1);
1427     qq02             = _mm_mul_ps(iq0,jq2);
1428     qq10             = _mm_mul_ps(iq1,jq0);
1429     qq11             = _mm_mul_ps(iq1,jq1);
1430     qq12             = _mm_mul_ps(iq1,jq2);
1431     qq20             = _mm_mul_ps(iq2,jq0);
1432     qq21             = _mm_mul_ps(iq2,jq1);
1433     qq22             = _mm_mul_ps(iq2,jq2);
1434
1435     /* Avoid stupid compiler warnings */
1436     jnrA = jnrB = jnrC = jnrD = 0;
1437     j_coord_offsetA = 0;
1438     j_coord_offsetB = 0;
1439     j_coord_offsetC = 0;
1440     j_coord_offsetD = 0;
1441
1442     outeriter        = 0;
1443     inneriter        = 0;
1444
1445     for(iidx=0;iidx<4*DIM;iidx++)
1446     {
1447         scratch[iidx] = 0.0;
1448     }
1449
1450     /* Start outer loop over neighborlists */
1451     for(iidx=0; iidx<nri; iidx++)
1452     {
1453         /* Load shift vector for this list */
1454         i_shift_offset   = DIM*shiftidx[iidx];
1455
1456         /* Load limits for loop over neighbors */
1457         j_index_start    = jindex[iidx];
1458         j_index_end      = jindex[iidx+1];
1459
1460         /* Get outer coordinate index */
1461         inr              = iinr[iidx];
1462         i_coord_offset   = DIM*inr;
1463
1464         /* Load i particle coords and add shift vector */
1465         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1466                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1467
1468         fix0             = _mm_setzero_ps();
1469         fiy0             = _mm_setzero_ps();
1470         fiz0             = _mm_setzero_ps();
1471         fix1             = _mm_setzero_ps();
1472         fiy1             = _mm_setzero_ps();
1473         fiz1             = _mm_setzero_ps();
1474         fix2             = _mm_setzero_ps();
1475         fiy2             = _mm_setzero_ps();
1476         fiz2             = _mm_setzero_ps();
1477
1478         /* Start inner kernel loop */
1479         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1480         {
1481
1482             /* Get j neighbor index, and coordinate index */
1483             jnrA             = jjnr[jidx];
1484             jnrB             = jjnr[jidx+1];
1485             jnrC             = jjnr[jidx+2];
1486             jnrD             = jjnr[jidx+3];
1487             j_coord_offsetA  = DIM*jnrA;
1488             j_coord_offsetB  = DIM*jnrB;
1489             j_coord_offsetC  = DIM*jnrC;
1490             j_coord_offsetD  = DIM*jnrD;
1491
1492             /* load j atom coordinates */
1493             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1494                                               x+j_coord_offsetC,x+j_coord_offsetD,
1495                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1496
1497             /* Calculate displacement vector */
1498             dx00             = _mm_sub_ps(ix0,jx0);
1499             dy00             = _mm_sub_ps(iy0,jy0);
1500             dz00             = _mm_sub_ps(iz0,jz0);
1501             dx01             = _mm_sub_ps(ix0,jx1);
1502             dy01             = _mm_sub_ps(iy0,jy1);
1503             dz01             = _mm_sub_ps(iz0,jz1);
1504             dx02             = _mm_sub_ps(ix0,jx2);
1505             dy02             = _mm_sub_ps(iy0,jy2);
1506             dz02             = _mm_sub_ps(iz0,jz2);
1507             dx10             = _mm_sub_ps(ix1,jx0);
1508             dy10             = _mm_sub_ps(iy1,jy0);
1509             dz10             = _mm_sub_ps(iz1,jz0);
1510             dx11             = _mm_sub_ps(ix1,jx1);
1511             dy11             = _mm_sub_ps(iy1,jy1);
1512             dz11             = _mm_sub_ps(iz1,jz1);
1513             dx12             = _mm_sub_ps(ix1,jx2);
1514             dy12             = _mm_sub_ps(iy1,jy2);
1515             dz12             = _mm_sub_ps(iz1,jz2);
1516             dx20             = _mm_sub_ps(ix2,jx0);
1517             dy20             = _mm_sub_ps(iy2,jy0);
1518             dz20             = _mm_sub_ps(iz2,jz0);
1519             dx21             = _mm_sub_ps(ix2,jx1);
1520             dy21             = _mm_sub_ps(iy2,jy1);
1521             dz21             = _mm_sub_ps(iz2,jz1);
1522             dx22             = _mm_sub_ps(ix2,jx2);
1523             dy22             = _mm_sub_ps(iy2,jy2);
1524             dz22             = _mm_sub_ps(iz2,jz2);
1525
1526             /* Calculate squared distance and things based on it */
1527             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1528             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
1529             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
1530             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1531             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1532             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1533             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1534             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1535             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1536
1537             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1538             rinv01           = gmx_mm_invsqrt_ps(rsq01);
1539             rinv02           = gmx_mm_invsqrt_ps(rsq02);
1540             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1541             rinv11           = gmx_mm_invsqrt_ps(rsq11);
1542             rinv12           = gmx_mm_invsqrt_ps(rsq12);
1543             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1544             rinv21           = gmx_mm_invsqrt_ps(rsq21);
1545             rinv22           = gmx_mm_invsqrt_ps(rsq22);
1546
1547             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1548             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
1549             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
1550             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1551             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
1552             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
1553             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1554             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
1555             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
1556
1557             fjx0             = _mm_setzero_ps();
1558             fjy0             = _mm_setzero_ps();
1559             fjz0             = _mm_setzero_ps();
1560             fjx1             = _mm_setzero_ps();
1561             fjy1             = _mm_setzero_ps();
1562             fjz1             = _mm_setzero_ps();
1563             fjx2             = _mm_setzero_ps();
1564             fjy2             = _mm_setzero_ps();
1565             fjz2             = _mm_setzero_ps();
1566
1567             /**************************
1568              * CALCULATE INTERACTIONS *
1569              **************************/
1570
1571             r00              = _mm_mul_ps(rsq00,rinv00);
1572
1573             /* Calculate table index by multiplying r with table scale and truncate to integer */
1574             rt               = _mm_mul_ps(r00,vftabscale);
1575             vfitab           = _mm_cvttps_epi32(rt);
1576             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1577             vfitab           = _mm_slli_epi32(vfitab,3);
1578
1579             /* EWALD ELECTROSTATICS */
1580
1581             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1582             ewrt             = _mm_mul_ps(r00,ewtabscale);
1583             ewitab           = _mm_cvttps_epi32(ewrt);
1584             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1585             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1586                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1587                                          &ewtabF,&ewtabFn);
1588             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1589             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1590
1591             /* CUBIC SPLINE TABLE DISPERSION */
1592             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1593             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1594             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1595             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1596             _MM_TRANSPOSE4_PS(Y,F,G,H);
1597             Heps             = _mm_mul_ps(vfeps,H);
1598             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1599             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1600             fvdw6            = _mm_mul_ps(c6_00,FF);
1601
1602             /* CUBIC SPLINE TABLE REPULSION */
1603             vfitab           = _mm_add_epi32(vfitab,ifour);
1604             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1605             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1606             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1607             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1608             _MM_TRANSPOSE4_PS(Y,F,G,H);
1609             Heps             = _mm_mul_ps(vfeps,H);
1610             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1611             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1612             fvdw12           = _mm_mul_ps(c12_00,FF);
1613             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1614
1615             fscal            = _mm_add_ps(felec,fvdw);
1616
1617             /* Calculate temporary vectorial force */
1618             tx               = _mm_mul_ps(fscal,dx00);
1619             ty               = _mm_mul_ps(fscal,dy00);
1620             tz               = _mm_mul_ps(fscal,dz00);
1621
1622             /* Update vectorial force */
1623             fix0             = _mm_add_ps(fix0,tx);
1624             fiy0             = _mm_add_ps(fiy0,ty);
1625             fiz0             = _mm_add_ps(fiz0,tz);
1626
1627             fjx0             = _mm_add_ps(fjx0,tx);
1628             fjy0             = _mm_add_ps(fjy0,ty);
1629             fjz0             = _mm_add_ps(fjz0,tz);
1630
1631             /**************************
1632              * CALCULATE INTERACTIONS *
1633              **************************/
1634
1635             r01              = _mm_mul_ps(rsq01,rinv01);
1636
1637             /* EWALD ELECTROSTATICS */
1638
1639             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1640             ewrt             = _mm_mul_ps(r01,ewtabscale);
1641             ewitab           = _mm_cvttps_epi32(ewrt);
1642             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1643             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1644                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1645                                          &ewtabF,&ewtabFn);
1646             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1647             felec            = _mm_mul_ps(_mm_mul_ps(qq01,rinv01),_mm_sub_ps(rinvsq01,felec));
1648
1649             fscal            = felec;
1650
1651             /* Calculate temporary vectorial force */
1652             tx               = _mm_mul_ps(fscal,dx01);
1653             ty               = _mm_mul_ps(fscal,dy01);
1654             tz               = _mm_mul_ps(fscal,dz01);
1655
1656             /* Update vectorial force */
1657             fix0             = _mm_add_ps(fix0,tx);
1658             fiy0             = _mm_add_ps(fiy0,ty);
1659             fiz0             = _mm_add_ps(fiz0,tz);
1660
1661             fjx1             = _mm_add_ps(fjx1,tx);
1662             fjy1             = _mm_add_ps(fjy1,ty);
1663             fjz1             = _mm_add_ps(fjz1,tz);
1664
1665             /**************************
1666              * CALCULATE INTERACTIONS *
1667              **************************/
1668
1669             r02              = _mm_mul_ps(rsq02,rinv02);
1670
1671             /* EWALD ELECTROSTATICS */
1672
1673             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1674             ewrt             = _mm_mul_ps(r02,ewtabscale);
1675             ewitab           = _mm_cvttps_epi32(ewrt);
1676             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1677             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1678                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1679                                          &ewtabF,&ewtabFn);
1680             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1681             felec            = _mm_mul_ps(_mm_mul_ps(qq02,rinv02),_mm_sub_ps(rinvsq02,felec));
1682
1683             fscal            = felec;
1684
1685             /* Calculate temporary vectorial force */
1686             tx               = _mm_mul_ps(fscal,dx02);
1687             ty               = _mm_mul_ps(fscal,dy02);
1688             tz               = _mm_mul_ps(fscal,dz02);
1689
1690             /* Update vectorial force */
1691             fix0             = _mm_add_ps(fix0,tx);
1692             fiy0             = _mm_add_ps(fiy0,ty);
1693             fiz0             = _mm_add_ps(fiz0,tz);
1694
1695             fjx2             = _mm_add_ps(fjx2,tx);
1696             fjy2             = _mm_add_ps(fjy2,ty);
1697             fjz2             = _mm_add_ps(fjz2,tz);
1698
1699             /**************************
1700              * CALCULATE INTERACTIONS *
1701              **************************/
1702
1703             r10              = _mm_mul_ps(rsq10,rinv10);
1704
1705             /* EWALD ELECTROSTATICS */
1706
1707             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1708             ewrt             = _mm_mul_ps(r10,ewtabscale);
1709             ewitab           = _mm_cvttps_epi32(ewrt);
1710             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1711             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1712                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1713                                          &ewtabF,&ewtabFn);
1714             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1715             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1716
1717             fscal            = felec;
1718
1719             /* Calculate temporary vectorial force */
1720             tx               = _mm_mul_ps(fscal,dx10);
1721             ty               = _mm_mul_ps(fscal,dy10);
1722             tz               = _mm_mul_ps(fscal,dz10);
1723
1724             /* Update vectorial force */
1725             fix1             = _mm_add_ps(fix1,tx);
1726             fiy1             = _mm_add_ps(fiy1,ty);
1727             fiz1             = _mm_add_ps(fiz1,tz);
1728
1729             fjx0             = _mm_add_ps(fjx0,tx);
1730             fjy0             = _mm_add_ps(fjy0,ty);
1731             fjz0             = _mm_add_ps(fjz0,tz);
1732
1733             /**************************
1734              * CALCULATE INTERACTIONS *
1735              **************************/
1736
1737             r11              = _mm_mul_ps(rsq11,rinv11);
1738
1739             /* EWALD ELECTROSTATICS */
1740
1741             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1742             ewrt             = _mm_mul_ps(r11,ewtabscale);
1743             ewitab           = _mm_cvttps_epi32(ewrt);
1744             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1745             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1746                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1747                                          &ewtabF,&ewtabFn);
1748             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1749             felec            = _mm_mul_ps(_mm_mul_ps(qq11,rinv11),_mm_sub_ps(rinvsq11,felec));
1750
1751             fscal            = felec;
1752
1753             /* Calculate temporary vectorial force */
1754             tx               = _mm_mul_ps(fscal,dx11);
1755             ty               = _mm_mul_ps(fscal,dy11);
1756             tz               = _mm_mul_ps(fscal,dz11);
1757
1758             /* Update vectorial force */
1759             fix1             = _mm_add_ps(fix1,tx);
1760             fiy1             = _mm_add_ps(fiy1,ty);
1761             fiz1             = _mm_add_ps(fiz1,tz);
1762
1763             fjx1             = _mm_add_ps(fjx1,tx);
1764             fjy1             = _mm_add_ps(fjy1,ty);
1765             fjz1             = _mm_add_ps(fjz1,tz);
1766
1767             /**************************
1768              * CALCULATE INTERACTIONS *
1769              **************************/
1770
1771             r12              = _mm_mul_ps(rsq12,rinv12);
1772
1773             /* EWALD ELECTROSTATICS */
1774
1775             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1776             ewrt             = _mm_mul_ps(r12,ewtabscale);
1777             ewitab           = _mm_cvttps_epi32(ewrt);
1778             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1779             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1780                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1781                                          &ewtabF,&ewtabFn);
1782             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1783             felec            = _mm_mul_ps(_mm_mul_ps(qq12,rinv12),_mm_sub_ps(rinvsq12,felec));
1784
1785             fscal            = felec;
1786
1787             /* Calculate temporary vectorial force */
1788             tx               = _mm_mul_ps(fscal,dx12);
1789             ty               = _mm_mul_ps(fscal,dy12);
1790             tz               = _mm_mul_ps(fscal,dz12);
1791
1792             /* Update vectorial force */
1793             fix1             = _mm_add_ps(fix1,tx);
1794             fiy1             = _mm_add_ps(fiy1,ty);
1795             fiz1             = _mm_add_ps(fiz1,tz);
1796
1797             fjx2             = _mm_add_ps(fjx2,tx);
1798             fjy2             = _mm_add_ps(fjy2,ty);
1799             fjz2             = _mm_add_ps(fjz2,tz);
1800
1801             /**************************
1802              * CALCULATE INTERACTIONS *
1803              **************************/
1804
1805             r20              = _mm_mul_ps(rsq20,rinv20);
1806
1807             /* EWALD ELECTROSTATICS */
1808
1809             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1810             ewrt             = _mm_mul_ps(r20,ewtabscale);
1811             ewitab           = _mm_cvttps_epi32(ewrt);
1812             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1813             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1814                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1815                                          &ewtabF,&ewtabFn);
1816             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1817             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1818
1819             fscal            = felec;
1820
1821             /* Calculate temporary vectorial force */
1822             tx               = _mm_mul_ps(fscal,dx20);
1823             ty               = _mm_mul_ps(fscal,dy20);
1824             tz               = _mm_mul_ps(fscal,dz20);
1825
1826             /* Update vectorial force */
1827             fix2             = _mm_add_ps(fix2,tx);
1828             fiy2             = _mm_add_ps(fiy2,ty);
1829             fiz2             = _mm_add_ps(fiz2,tz);
1830
1831             fjx0             = _mm_add_ps(fjx0,tx);
1832             fjy0             = _mm_add_ps(fjy0,ty);
1833             fjz0             = _mm_add_ps(fjz0,tz);
1834
1835             /**************************
1836              * CALCULATE INTERACTIONS *
1837              **************************/
1838
1839             r21              = _mm_mul_ps(rsq21,rinv21);
1840
1841             /* EWALD ELECTROSTATICS */
1842
1843             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1844             ewrt             = _mm_mul_ps(r21,ewtabscale);
1845             ewitab           = _mm_cvttps_epi32(ewrt);
1846             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1847             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1848                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1849                                          &ewtabF,&ewtabFn);
1850             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1851             felec            = _mm_mul_ps(_mm_mul_ps(qq21,rinv21),_mm_sub_ps(rinvsq21,felec));
1852
1853             fscal            = felec;
1854
1855             /* Calculate temporary vectorial force */
1856             tx               = _mm_mul_ps(fscal,dx21);
1857             ty               = _mm_mul_ps(fscal,dy21);
1858             tz               = _mm_mul_ps(fscal,dz21);
1859
1860             /* Update vectorial force */
1861             fix2             = _mm_add_ps(fix2,tx);
1862             fiy2             = _mm_add_ps(fiy2,ty);
1863             fiz2             = _mm_add_ps(fiz2,tz);
1864
1865             fjx1             = _mm_add_ps(fjx1,tx);
1866             fjy1             = _mm_add_ps(fjy1,ty);
1867             fjz1             = _mm_add_ps(fjz1,tz);
1868
1869             /**************************
1870              * CALCULATE INTERACTIONS *
1871              **************************/
1872
1873             r22              = _mm_mul_ps(rsq22,rinv22);
1874
1875             /* EWALD ELECTROSTATICS */
1876
1877             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1878             ewrt             = _mm_mul_ps(r22,ewtabscale);
1879             ewitab           = _mm_cvttps_epi32(ewrt);
1880             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1881             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1882                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1883                                          &ewtabF,&ewtabFn);
1884             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1885             felec            = _mm_mul_ps(_mm_mul_ps(qq22,rinv22),_mm_sub_ps(rinvsq22,felec));
1886
1887             fscal            = felec;
1888
1889             /* Calculate temporary vectorial force */
1890             tx               = _mm_mul_ps(fscal,dx22);
1891             ty               = _mm_mul_ps(fscal,dy22);
1892             tz               = _mm_mul_ps(fscal,dz22);
1893
1894             /* Update vectorial force */
1895             fix2             = _mm_add_ps(fix2,tx);
1896             fiy2             = _mm_add_ps(fiy2,ty);
1897             fiz2             = _mm_add_ps(fiz2,tz);
1898
1899             fjx2             = _mm_add_ps(fjx2,tx);
1900             fjy2             = _mm_add_ps(fjy2,ty);
1901             fjz2             = _mm_add_ps(fjz2,tz);
1902
1903             fjptrA             = f+j_coord_offsetA;
1904             fjptrB             = f+j_coord_offsetB;
1905             fjptrC             = f+j_coord_offsetC;
1906             fjptrD             = f+j_coord_offsetD;
1907
1908             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1909                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1910
1911             /* Inner loop uses 350 flops */
1912         }
1913
1914         if(jidx<j_index_end)
1915         {
1916
1917             /* Get j neighbor index, and coordinate index */
1918             jnrlistA         = jjnr[jidx];
1919             jnrlistB         = jjnr[jidx+1];
1920             jnrlistC         = jjnr[jidx+2];
1921             jnrlistD         = jjnr[jidx+3];
1922             /* Sign of each element will be negative for non-real atoms.
1923              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1924              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1925              */
1926             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1927             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1928             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1929             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1930             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1931             j_coord_offsetA  = DIM*jnrA;
1932             j_coord_offsetB  = DIM*jnrB;
1933             j_coord_offsetC  = DIM*jnrC;
1934             j_coord_offsetD  = DIM*jnrD;
1935
1936             /* load j atom coordinates */
1937             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1938                                               x+j_coord_offsetC,x+j_coord_offsetD,
1939                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1940
1941             /* Calculate displacement vector */
1942             dx00             = _mm_sub_ps(ix0,jx0);
1943             dy00             = _mm_sub_ps(iy0,jy0);
1944             dz00             = _mm_sub_ps(iz0,jz0);
1945             dx01             = _mm_sub_ps(ix0,jx1);
1946             dy01             = _mm_sub_ps(iy0,jy1);
1947             dz01             = _mm_sub_ps(iz0,jz1);
1948             dx02             = _mm_sub_ps(ix0,jx2);
1949             dy02             = _mm_sub_ps(iy0,jy2);
1950             dz02             = _mm_sub_ps(iz0,jz2);
1951             dx10             = _mm_sub_ps(ix1,jx0);
1952             dy10             = _mm_sub_ps(iy1,jy0);
1953             dz10             = _mm_sub_ps(iz1,jz0);
1954             dx11             = _mm_sub_ps(ix1,jx1);
1955             dy11             = _mm_sub_ps(iy1,jy1);
1956             dz11             = _mm_sub_ps(iz1,jz1);
1957             dx12             = _mm_sub_ps(ix1,jx2);
1958             dy12             = _mm_sub_ps(iy1,jy2);
1959             dz12             = _mm_sub_ps(iz1,jz2);
1960             dx20             = _mm_sub_ps(ix2,jx0);
1961             dy20             = _mm_sub_ps(iy2,jy0);
1962             dz20             = _mm_sub_ps(iz2,jz0);
1963             dx21             = _mm_sub_ps(ix2,jx1);
1964             dy21             = _mm_sub_ps(iy2,jy1);
1965             dz21             = _mm_sub_ps(iz2,jz1);
1966             dx22             = _mm_sub_ps(ix2,jx2);
1967             dy22             = _mm_sub_ps(iy2,jy2);
1968             dz22             = _mm_sub_ps(iz2,jz2);
1969
1970             /* Calculate squared distance and things based on it */
1971             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1972             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
1973             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
1974             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1975             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1976             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1977             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1978             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1979             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1980
1981             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1982             rinv01           = gmx_mm_invsqrt_ps(rsq01);
1983             rinv02           = gmx_mm_invsqrt_ps(rsq02);
1984             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1985             rinv11           = gmx_mm_invsqrt_ps(rsq11);
1986             rinv12           = gmx_mm_invsqrt_ps(rsq12);
1987             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1988             rinv21           = gmx_mm_invsqrt_ps(rsq21);
1989             rinv22           = gmx_mm_invsqrt_ps(rsq22);
1990
1991             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1992             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
1993             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
1994             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1995             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
1996             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
1997             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1998             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
1999             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
2000
2001             fjx0             = _mm_setzero_ps();
2002             fjy0             = _mm_setzero_ps();
2003             fjz0             = _mm_setzero_ps();
2004             fjx1             = _mm_setzero_ps();
2005             fjy1             = _mm_setzero_ps();
2006             fjz1             = _mm_setzero_ps();
2007             fjx2             = _mm_setzero_ps();
2008             fjy2             = _mm_setzero_ps();
2009             fjz2             = _mm_setzero_ps();
2010
2011             /**************************
2012              * CALCULATE INTERACTIONS *
2013              **************************/
2014
2015             r00              = _mm_mul_ps(rsq00,rinv00);
2016             r00              = _mm_andnot_ps(dummy_mask,r00);
2017
2018             /* Calculate table index by multiplying r with table scale and truncate to integer */
2019             rt               = _mm_mul_ps(r00,vftabscale);
2020             vfitab           = _mm_cvttps_epi32(rt);
2021             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
2022             vfitab           = _mm_slli_epi32(vfitab,3);
2023
2024             /* EWALD ELECTROSTATICS */
2025
2026             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2027             ewrt             = _mm_mul_ps(r00,ewtabscale);
2028             ewitab           = _mm_cvttps_epi32(ewrt);
2029             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2030             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
2031                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
2032                                          &ewtabF,&ewtabFn);
2033             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2034             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
2035
2036             /* CUBIC SPLINE TABLE DISPERSION */
2037             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
2038             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
2039             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
2040             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
2041             _MM_TRANSPOSE4_PS(Y,F,G,H);
2042             Heps             = _mm_mul_ps(vfeps,H);
2043             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
2044             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
2045             fvdw6            = _mm_mul_ps(c6_00,FF);
2046
2047             /* CUBIC SPLINE TABLE REPULSION */
2048             vfitab           = _mm_add_epi32(vfitab,ifour);
2049             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
2050             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
2051             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
2052             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
2053             _MM_TRANSPOSE4_PS(Y,F,G,H);
2054             Heps             = _mm_mul_ps(vfeps,H);
2055             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
2056             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
2057             fvdw12           = _mm_mul_ps(c12_00,FF);
2058             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
2059
2060             fscal            = _mm_add_ps(felec,fvdw);
2061
2062             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2063
2064             /* Calculate temporary vectorial force */
2065             tx               = _mm_mul_ps(fscal,dx00);
2066             ty               = _mm_mul_ps(fscal,dy00);
2067             tz               = _mm_mul_ps(fscal,dz00);
2068
2069             /* Update vectorial force */
2070             fix0             = _mm_add_ps(fix0,tx);
2071             fiy0             = _mm_add_ps(fiy0,ty);
2072             fiz0             = _mm_add_ps(fiz0,tz);
2073
2074             fjx0             = _mm_add_ps(fjx0,tx);
2075             fjy0             = _mm_add_ps(fjy0,ty);
2076             fjz0             = _mm_add_ps(fjz0,tz);
2077
2078             /**************************
2079              * CALCULATE INTERACTIONS *
2080              **************************/
2081
2082             r01              = _mm_mul_ps(rsq01,rinv01);
2083             r01              = _mm_andnot_ps(dummy_mask,r01);
2084
2085             /* EWALD ELECTROSTATICS */
2086
2087             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2088             ewrt             = _mm_mul_ps(r01,ewtabscale);
2089             ewitab           = _mm_cvttps_epi32(ewrt);
2090             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2091             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
2092                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
2093                                          &ewtabF,&ewtabFn);
2094             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2095             felec            = _mm_mul_ps(_mm_mul_ps(qq01,rinv01),_mm_sub_ps(rinvsq01,felec));
2096
2097             fscal            = felec;
2098
2099             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2100
2101             /* Calculate temporary vectorial force */
2102             tx               = _mm_mul_ps(fscal,dx01);
2103             ty               = _mm_mul_ps(fscal,dy01);
2104             tz               = _mm_mul_ps(fscal,dz01);
2105
2106             /* Update vectorial force */
2107             fix0             = _mm_add_ps(fix0,tx);
2108             fiy0             = _mm_add_ps(fiy0,ty);
2109             fiz0             = _mm_add_ps(fiz0,tz);
2110
2111             fjx1             = _mm_add_ps(fjx1,tx);
2112             fjy1             = _mm_add_ps(fjy1,ty);
2113             fjz1             = _mm_add_ps(fjz1,tz);
2114
2115             /**************************
2116              * CALCULATE INTERACTIONS *
2117              **************************/
2118
2119             r02              = _mm_mul_ps(rsq02,rinv02);
2120             r02              = _mm_andnot_ps(dummy_mask,r02);
2121
2122             /* EWALD ELECTROSTATICS */
2123
2124             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2125             ewrt             = _mm_mul_ps(r02,ewtabscale);
2126             ewitab           = _mm_cvttps_epi32(ewrt);
2127             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2128             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
2129                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
2130                                          &ewtabF,&ewtabFn);
2131             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2132             felec            = _mm_mul_ps(_mm_mul_ps(qq02,rinv02),_mm_sub_ps(rinvsq02,felec));
2133
2134             fscal            = felec;
2135
2136             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2137
2138             /* Calculate temporary vectorial force */
2139             tx               = _mm_mul_ps(fscal,dx02);
2140             ty               = _mm_mul_ps(fscal,dy02);
2141             tz               = _mm_mul_ps(fscal,dz02);
2142
2143             /* Update vectorial force */
2144             fix0             = _mm_add_ps(fix0,tx);
2145             fiy0             = _mm_add_ps(fiy0,ty);
2146             fiz0             = _mm_add_ps(fiz0,tz);
2147
2148             fjx2             = _mm_add_ps(fjx2,tx);
2149             fjy2             = _mm_add_ps(fjy2,ty);
2150             fjz2             = _mm_add_ps(fjz2,tz);
2151
2152             /**************************
2153              * CALCULATE INTERACTIONS *
2154              **************************/
2155
2156             r10              = _mm_mul_ps(rsq10,rinv10);
2157             r10              = _mm_andnot_ps(dummy_mask,r10);
2158
2159             /* EWALD ELECTROSTATICS */
2160
2161             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2162             ewrt             = _mm_mul_ps(r10,ewtabscale);
2163             ewitab           = _mm_cvttps_epi32(ewrt);
2164             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2165             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
2166                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
2167                                          &ewtabF,&ewtabFn);
2168             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2169             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
2170
2171             fscal            = felec;
2172
2173             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2174
2175             /* Calculate temporary vectorial force */
2176             tx               = _mm_mul_ps(fscal,dx10);
2177             ty               = _mm_mul_ps(fscal,dy10);
2178             tz               = _mm_mul_ps(fscal,dz10);
2179
2180             /* Update vectorial force */
2181             fix1             = _mm_add_ps(fix1,tx);
2182             fiy1             = _mm_add_ps(fiy1,ty);
2183             fiz1             = _mm_add_ps(fiz1,tz);
2184
2185             fjx0             = _mm_add_ps(fjx0,tx);
2186             fjy0             = _mm_add_ps(fjy0,ty);
2187             fjz0             = _mm_add_ps(fjz0,tz);
2188
2189             /**************************
2190              * CALCULATE INTERACTIONS *
2191              **************************/
2192
2193             r11              = _mm_mul_ps(rsq11,rinv11);
2194             r11              = _mm_andnot_ps(dummy_mask,r11);
2195
2196             /* EWALD ELECTROSTATICS */
2197
2198             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2199             ewrt             = _mm_mul_ps(r11,ewtabscale);
2200             ewitab           = _mm_cvttps_epi32(ewrt);
2201             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2202             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
2203                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
2204                                          &ewtabF,&ewtabFn);
2205             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2206             felec            = _mm_mul_ps(_mm_mul_ps(qq11,rinv11),_mm_sub_ps(rinvsq11,felec));
2207
2208             fscal            = felec;
2209
2210             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2211
2212             /* Calculate temporary vectorial force */
2213             tx               = _mm_mul_ps(fscal,dx11);
2214             ty               = _mm_mul_ps(fscal,dy11);
2215             tz               = _mm_mul_ps(fscal,dz11);
2216
2217             /* Update vectorial force */
2218             fix1             = _mm_add_ps(fix1,tx);
2219             fiy1             = _mm_add_ps(fiy1,ty);
2220             fiz1             = _mm_add_ps(fiz1,tz);
2221
2222             fjx1             = _mm_add_ps(fjx1,tx);
2223             fjy1             = _mm_add_ps(fjy1,ty);
2224             fjz1             = _mm_add_ps(fjz1,tz);
2225
2226             /**************************
2227              * CALCULATE INTERACTIONS *
2228              **************************/
2229
2230             r12              = _mm_mul_ps(rsq12,rinv12);
2231             r12              = _mm_andnot_ps(dummy_mask,r12);
2232
2233             /* EWALD ELECTROSTATICS */
2234
2235             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2236             ewrt             = _mm_mul_ps(r12,ewtabscale);
2237             ewitab           = _mm_cvttps_epi32(ewrt);
2238             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2239             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
2240                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
2241                                          &ewtabF,&ewtabFn);
2242             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2243             felec            = _mm_mul_ps(_mm_mul_ps(qq12,rinv12),_mm_sub_ps(rinvsq12,felec));
2244
2245             fscal            = felec;
2246
2247             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2248
2249             /* Calculate temporary vectorial force */
2250             tx               = _mm_mul_ps(fscal,dx12);
2251             ty               = _mm_mul_ps(fscal,dy12);
2252             tz               = _mm_mul_ps(fscal,dz12);
2253
2254             /* Update vectorial force */
2255             fix1             = _mm_add_ps(fix1,tx);
2256             fiy1             = _mm_add_ps(fiy1,ty);
2257             fiz1             = _mm_add_ps(fiz1,tz);
2258
2259             fjx2             = _mm_add_ps(fjx2,tx);
2260             fjy2             = _mm_add_ps(fjy2,ty);
2261             fjz2             = _mm_add_ps(fjz2,tz);
2262
2263             /**************************
2264              * CALCULATE INTERACTIONS *
2265              **************************/
2266
2267             r20              = _mm_mul_ps(rsq20,rinv20);
2268             r20              = _mm_andnot_ps(dummy_mask,r20);
2269
2270             /* EWALD ELECTROSTATICS */
2271
2272             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2273             ewrt             = _mm_mul_ps(r20,ewtabscale);
2274             ewitab           = _mm_cvttps_epi32(ewrt);
2275             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2276             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
2277                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
2278                                          &ewtabF,&ewtabFn);
2279             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2280             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
2281
2282             fscal            = felec;
2283
2284             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2285
2286             /* Calculate temporary vectorial force */
2287             tx               = _mm_mul_ps(fscal,dx20);
2288             ty               = _mm_mul_ps(fscal,dy20);
2289             tz               = _mm_mul_ps(fscal,dz20);
2290
2291             /* Update vectorial force */
2292             fix2             = _mm_add_ps(fix2,tx);
2293             fiy2             = _mm_add_ps(fiy2,ty);
2294             fiz2             = _mm_add_ps(fiz2,tz);
2295
2296             fjx0             = _mm_add_ps(fjx0,tx);
2297             fjy0             = _mm_add_ps(fjy0,ty);
2298             fjz0             = _mm_add_ps(fjz0,tz);
2299
2300             /**************************
2301              * CALCULATE INTERACTIONS *
2302              **************************/
2303
2304             r21              = _mm_mul_ps(rsq21,rinv21);
2305             r21              = _mm_andnot_ps(dummy_mask,r21);
2306
2307             /* EWALD ELECTROSTATICS */
2308
2309             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2310             ewrt             = _mm_mul_ps(r21,ewtabscale);
2311             ewitab           = _mm_cvttps_epi32(ewrt);
2312             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2313             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
2314                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
2315                                          &ewtabF,&ewtabFn);
2316             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2317             felec            = _mm_mul_ps(_mm_mul_ps(qq21,rinv21),_mm_sub_ps(rinvsq21,felec));
2318
2319             fscal            = felec;
2320
2321             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2322
2323             /* Calculate temporary vectorial force */
2324             tx               = _mm_mul_ps(fscal,dx21);
2325             ty               = _mm_mul_ps(fscal,dy21);
2326             tz               = _mm_mul_ps(fscal,dz21);
2327
2328             /* Update vectorial force */
2329             fix2             = _mm_add_ps(fix2,tx);
2330             fiy2             = _mm_add_ps(fiy2,ty);
2331             fiz2             = _mm_add_ps(fiz2,tz);
2332
2333             fjx1             = _mm_add_ps(fjx1,tx);
2334             fjy1             = _mm_add_ps(fjy1,ty);
2335             fjz1             = _mm_add_ps(fjz1,tz);
2336
2337             /**************************
2338              * CALCULATE INTERACTIONS *
2339              **************************/
2340
2341             r22              = _mm_mul_ps(rsq22,rinv22);
2342             r22              = _mm_andnot_ps(dummy_mask,r22);
2343
2344             /* EWALD ELECTROSTATICS */
2345
2346             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2347             ewrt             = _mm_mul_ps(r22,ewtabscale);
2348             ewitab           = _mm_cvttps_epi32(ewrt);
2349             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2350             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
2351                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
2352                                          &ewtabF,&ewtabFn);
2353             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2354             felec            = _mm_mul_ps(_mm_mul_ps(qq22,rinv22),_mm_sub_ps(rinvsq22,felec));
2355
2356             fscal            = felec;
2357
2358             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2359
2360             /* Calculate temporary vectorial force */
2361             tx               = _mm_mul_ps(fscal,dx22);
2362             ty               = _mm_mul_ps(fscal,dy22);
2363             tz               = _mm_mul_ps(fscal,dz22);
2364
2365             /* Update vectorial force */
2366             fix2             = _mm_add_ps(fix2,tx);
2367             fiy2             = _mm_add_ps(fiy2,ty);
2368             fiz2             = _mm_add_ps(fiz2,tz);
2369
2370             fjx2             = _mm_add_ps(fjx2,tx);
2371             fjy2             = _mm_add_ps(fjy2,ty);
2372             fjz2             = _mm_add_ps(fjz2,tz);
2373
2374             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2375             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2376             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2377             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2378
2379             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
2380                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2381
2382             /* Inner loop uses 359 flops */
2383         }
2384
2385         /* End of innermost loop */
2386
2387         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2388                                               f+i_coord_offset,fshift+i_shift_offset);
2389
2390         /* Increment number of inner iterations */
2391         inneriter                  += j_index_end - j_index_start;
2392
2393         /* Outer loop uses 18 flops */
2394     }
2395
2396     /* Increment number of outer iterations */
2397     outeriter        += nri;
2398
2399     /* Update outer/inner flops */
2400
2401     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*359);
2402 }