src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecCSTab_VdwLJ_GeomW4W4_avx_128_fma_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 avx_128_fma_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  50 #include "kernelutil_x86_avx_128_fma_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW4W4_VF_avx_128_fma_single
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water4-Water4
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_VdwLJ_GeomW4W4_VF_avx_128_fma_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwioffset3;
  92     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  93     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  94     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  95     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  96     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  97     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  98     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  99     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
 100     __m128           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
 101     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 102     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 103     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 104     __m128           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
 105     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 106     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 107     __m128           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 108     __m128           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 109     __m128           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 110     __m128           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 111     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 112     real             *charge;
 113     int              nvdwtype;
 114     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 115     int              *vdwtype;
 116     real             *vdwparam;
 117     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 118     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 119     __m128i          vfitab;
 120     __m128i          ifour       = _mm_set1_epi32(4);
 121     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 122     real             *vftab;
 123     __m128           dummy_mask,cutoff_mask;
 124     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 125     __m128           one     = _mm_set1_ps(1.0);
 126     __m128           two     = _mm_set1_ps(2.0);
 127     x                = xx[0];
 128     f                = ff[0];
 129
 130     nri              = nlist->nri;
 131     iinr             = nlist->iinr;
 132     jindex           = nlist->jindex;
 133     jjnr             = nlist->jjnr;
 134     shiftidx         = nlist->shift;
 135     gid              = nlist->gid;
 136     shiftvec         = fr->shift_vec[0];
 137     fshift           = fr->fshift[0];
 138     facel            = _mm_set1_ps(fr->epsfac);
 139     charge           = mdatoms->chargeA;
 140     nvdwtype         = fr->ntype;
 141     vdwparam         = fr->nbfp;
 142     vdwtype          = mdatoms->typeA;
 143
 144     vftab            = kernel_data->table_elec->data;
 145     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 146
 147     /* Setup water-specific parameters */
 148     inr              = nlist->iinr[0];
 149     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 150     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 151     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 152     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 153
 154     jq1              = _mm_set1_ps(charge[inr+1]);
 155     jq2              = _mm_set1_ps(charge[inr+2]);
 156     jq3              = _mm_set1_ps(charge[inr+3]);
 157     vdwjidx0A        = 2*vdwtype[inr+0];
 158     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
 159     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
 160     qq11             = _mm_mul_ps(iq1,jq1);
 161     qq12             = _mm_mul_ps(iq1,jq2);
 162     qq13             = _mm_mul_ps(iq1,jq3);
 163     qq21             = _mm_mul_ps(iq2,jq1);
 164     qq22             = _mm_mul_ps(iq2,jq2);
 165     qq23             = _mm_mul_ps(iq2,jq3);
 166     qq31             = _mm_mul_ps(iq3,jq1);
 167     qq32             = _mm_mul_ps(iq3,jq2);
 168     qq33             = _mm_mul_ps(iq3,jq3);
 169
 170     /* Avoid stupid compiler warnings */
 171     jnrA = jnrB = jnrC = jnrD = 0;
 172     j_coord_offsetA = 0;
 173     j_coord_offsetB = 0;
 174     j_coord_offsetC = 0;
 175     j_coord_offsetD = 0;
 176
 177     outeriter        = 0;
 178     inneriter        = 0;
 179
 180     for(iidx=0;iidx<4*DIM;iidx++)
 181     {
 182         scratch[iidx] = 0.0;
 183     }
 184
 185     /* Start outer loop over neighborlists */
 186     for(iidx=0; iidx<nri; iidx++)
 187     {
 188         /* Load shift vector for this list */
 189         i_shift_offset   = DIM*shiftidx[iidx];
 190
 191         /* Load limits for loop over neighbors */
 192         j_index_start    = jindex[iidx];
 193         j_index_end      = jindex[iidx+1];
 194
 195         /* Get outer coordinate index */
 196         inr              = iinr[iidx];
 197         i_coord_offset   = DIM*inr;
 198
 199         /* Load i particle coords and add shift vector */
 200         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 201                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 202
 203         fix0             = _mm_setzero_ps();
 204         fiy0             = _mm_setzero_ps();
 205         fiz0             = _mm_setzero_ps();
 206         fix1             = _mm_setzero_ps();
 207         fiy1             = _mm_setzero_ps();
 208         fiz1             = _mm_setzero_ps();
 209         fix2             = _mm_setzero_ps();
 210         fiy2             = _mm_setzero_ps();
 211         fiz2             = _mm_setzero_ps();
 212         fix3             = _mm_setzero_ps();
 213         fiy3             = _mm_setzero_ps();
 214         fiz3             = _mm_setzero_ps();
 215
 216         /* Reset potential sums */
 217         velecsum         = _mm_setzero_ps();
 218         vvdwsum          = _mm_setzero_ps();
 219
 220         /* Start inner kernel loop */
 221         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 222         {
 223
 224             /* Get j neighbor index, and coordinate index */
 225             jnrA             = jjnr[jidx];
 226             jnrB             = jjnr[jidx+1];
 227             jnrC             = jjnr[jidx+2];
 228             jnrD             = jjnr[jidx+3];
 229             j_coord_offsetA  = DIM*jnrA;
 230             j_coord_offsetB  = DIM*jnrB;
 231             j_coord_offsetC  = DIM*jnrC;
 232             j_coord_offsetD  = DIM*jnrD;
 233
 234             /* load j atom coordinates */
 235             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 236                                               x+j_coord_offsetC,x+j_coord_offsetD,
 237                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 238                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 239
 240             /* Calculate displacement vector */
 241             dx00             = _mm_sub_ps(ix0,jx0);
 242             dy00             = _mm_sub_ps(iy0,jy0);
 243             dz00             = _mm_sub_ps(iz0,jz0);
 244             dx11             = _mm_sub_ps(ix1,jx1);
 245             dy11             = _mm_sub_ps(iy1,jy1);
 246             dz11             = _mm_sub_ps(iz1,jz1);
 247             dx12             = _mm_sub_ps(ix1,jx2);
 248             dy12             = _mm_sub_ps(iy1,jy2);
 249             dz12             = _mm_sub_ps(iz1,jz2);
 250             dx13             = _mm_sub_ps(ix1,jx3);
 251             dy13             = _mm_sub_ps(iy1,jy3);
 252             dz13             = _mm_sub_ps(iz1,jz3);
 253             dx21             = _mm_sub_ps(ix2,jx1);
 254             dy21             = _mm_sub_ps(iy2,jy1);
 255             dz21             = _mm_sub_ps(iz2,jz1);
 256             dx22             = _mm_sub_ps(ix2,jx2);
 257             dy22             = _mm_sub_ps(iy2,jy2);
 258             dz22             = _mm_sub_ps(iz2,jz2);
 259             dx23             = _mm_sub_ps(ix2,jx3);
 260             dy23             = _mm_sub_ps(iy2,jy3);
 261             dz23             = _mm_sub_ps(iz2,jz3);
 262             dx31             = _mm_sub_ps(ix3,jx1);
 263             dy31             = _mm_sub_ps(iy3,jy1);
 264             dz31             = _mm_sub_ps(iz3,jz1);
 265             dx32             = _mm_sub_ps(ix3,jx2);
 266             dy32             = _mm_sub_ps(iy3,jy2);
 267             dz32             = _mm_sub_ps(iz3,jz2);
 268             dx33             = _mm_sub_ps(ix3,jx3);
 269             dy33             = _mm_sub_ps(iy3,jy3);
 270             dz33             = _mm_sub_ps(iz3,jz3);
 271
 272             /* Calculate squared distance and things based on it */
 273             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 274             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 275             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 276             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
 277             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 278             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 279             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
 280             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
 281             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
 282             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
 283
 284             rinv11           = gmx_mm_invsqrt_ps(rsq11);
 285             rinv12           = gmx_mm_invsqrt_ps(rsq12);
 286             rinv13           = gmx_mm_invsqrt_ps(rsq13);
 287             rinv21           = gmx_mm_invsqrt_ps(rsq21);
 288             rinv22           = gmx_mm_invsqrt_ps(rsq22);
 289             rinv23           = gmx_mm_invsqrt_ps(rsq23);
 290             rinv31           = gmx_mm_invsqrt_ps(rsq31);
 291             rinv32           = gmx_mm_invsqrt_ps(rsq32);
 292             rinv33           = gmx_mm_invsqrt_ps(rsq33);
 293
 294             rinvsq00         = gmx_mm_inv_ps(rsq00);
 295
 296             fjx0             = _mm_setzero_ps();
 297             fjy0             = _mm_setzero_ps();
 298             fjz0             = _mm_setzero_ps();
 299             fjx1             = _mm_setzero_ps();
 300             fjy1             = _mm_setzero_ps();
 301             fjz1             = _mm_setzero_ps();
 302             fjx2             = _mm_setzero_ps();
 303             fjy2             = _mm_setzero_ps();
 304             fjz2             = _mm_setzero_ps();
 305             fjx3             = _mm_setzero_ps();
 306             fjy3             = _mm_setzero_ps();
 307             fjz3             = _mm_setzero_ps();
 308
 309             /**************************
 310              * CALCULATE INTERACTIONS *
 311              **************************/
 312
 313             /* LENNARD-JONES DISPERSION/REPULSION */
 314
 315             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 316             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 317             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 318             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 319             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 320
 321             /* Update potential sum for this i atom from the interaction with this j atom. */
 322             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 323
 324             fscal            = fvdw;
 325
 326              /* Update vectorial force */
 327             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 328             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 329             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 330
 331             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 332             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 333             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 334
 335             /**************************
 336              * CALCULATE INTERACTIONS *
 337              **************************/
 338
 339             r11              = _mm_mul_ps(rsq11,rinv11);
 340
 341             /* Calculate table index by multiplying r with table scale and truncate to integer */
 342             rt               = _mm_mul_ps(r11,vftabscale);
 343             vfitab           = _mm_cvttps_epi32(rt);
 344 #ifdef __XOP__
 345             vfeps            = _mm_frcz_ps(rt);
 346 #else
 347             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 348 #endif
 349             twovfeps         = _mm_add_ps(vfeps,vfeps);
 350             vfitab           = _mm_slli_epi32(vfitab,2);
 351
 352             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 353             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 354             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 355             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 356             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 357             _MM_TRANSPOSE4_PS(Y,F,G,H);
 358             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 359             VV               = _mm_macc_ps(vfeps,Fp,Y);
 360             velec            = _mm_mul_ps(qq11,VV);
 361             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 362             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq11,FF),_mm_mul_ps(vftabscale,rinv11)));
 363
 364             /* Update potential sum for this i atom from the interaction with this j atom. */
 365             velecsum         = _mm_add_ps(velecsum,velec);
 366
 367             fscal            = felec;
 368
 369              /* Update vectorial force */
 370             fix1             = _mm_macc_ps(dx11,fscal,fix1);
 371             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
 372             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
 373
 374             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
 375             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
 376             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
 377
 378             /**************************
 379              * CALCULATE INTERACTIONS *
 380              **************************/
 381
 382             r12              = _mm_mul_ps(rsq12,rinv12);
 383
 384             /* Calculate table index by multiplying r with table scale and truncate to integer */
 385             rt               = _mm_mul_ps(r12,vftabscale);
 386             vfitab           = _mm_cvttps_epi32(rt);
 387 #ifdef __XOP__
 388             vfeps            = _mm_frcz_ps(rt);
 389 #else
 390             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 391 #endif
 392             twovfeps         = _mm_add_ps(vfeps,vfeps);
 393             vfitab           = _mm_slli_epi32(vfitab,2);
 394
 395             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 396             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 397             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 398             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 399             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 400             _MM_TRANSPOSE4_PS(Y,F,G,H);
 401             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 402             VV               = _mm_macc_ps(vfeps,Fp,Y);
 403             velec            = _mm_mul_ps(qq12,VV);
 404             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 405             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq12,FF),_mm_mul_ps(vftabscale,rinv12)));
 406
 407             /* Update potential sum for this i atom from the interaction with this j atom. */
 408             velecsum         = _mm_add_ps(velecsum,velec);
 409
 410             fscal            = felec;
 411
 412              /* Update vectorial force */
 413             fix1             = _mm_macc_ps(dx12,fscal,fix1);
 414             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
 415             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
 416
 417             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
 418             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
 419             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
 420
 421             /**************************
 422              * CALCULATE INTERACTIONS *
 423              **************************/
 424
 425             r13              = _mm_mul_ps(rsq13,rinv13);
 426
 427             /* Calculate table index by multiplying r with table scale and truncate to integer */
 428             rt               = _mm_mul_ps(r13,vftabscale);
 429             vfitab           = _mm_cvttps_epi32(rt);
 430 #ifdef __XOP__
 431             vfeps            = _mm_frcz_ps(rt);
 432 #else
 433             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 434 #endif
 435             twovfeps         = _mm_add_ps(vfeps,vfeps);
 436             vfitab           = _mm_slli_epi32(vfitab,2);
 437
 438             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 439             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 440             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 441             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 442             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 443             _MM_TRANSPOSE4_PS(Y,F,G,H);
 444             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 445             VV               = _mm_macc_ps(vfeps,Fp,Y);
 446             velec            = _mm_mul_ps(qq13,VV);
 447             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 448             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq13,FF),_mm_mul_ps(vftabscale,rinv13)));
 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              /* Update vectorial force */
 456             fix1             = _mm_macc_ps(dx13,fscal,fix1);
 457             fiy1             = _mm_macc_ps(dy13,fscal,fiy1);
 458             fiz1             = _mm_macc_ps(dz13,fscal,fiz1);
 459
 460             fjx3             = _mm_macc_ps(dx13,fscal,fjx3);
 461             fjy3             = _mm_macc_ps(dy13,fscal,fjy3);
 462             fjz3             = _mm_macc_ps(dz13,fscal,fjz3);
 463
 464             /**************************
 465              * CALCULATE INTERACTIONS *
 466              **************************/
 467
 468             r21              = _mm_mul_ps(rsq21,rinv21);
 469
 470             /* Calculate table index by multiplying r with table scale and truncate to integer */
 471             rt               = _mm_mul_ps(r21,vftabscale);
 472             vfitab           = _mm_cvttps_epi32(rt);
 473 #ifdef __XOP__
 474             vfeps            = _mm_frcz_ps(rt);
 475 #else
 476             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 477 #endif
 478             twovfeps         = _mm_add_ps(vfeps,vfeps);
 479             vfitab           = _mm_slli_epi32(vfitab,2);
 480
 481             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 482             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 483             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 484             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 485             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 486             _MM_TRANSPOSE4_PS(Y,F,G,H);
 487             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 488             VV               = _mm_macc_ps(vfeps,Fp,Y);
 489             velec            = _mm_mul_ps(qq21,VV);
 490             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 491             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq21,FF),_mm_mul_ps(vftabscale,rinv21)));
 492
 493             /* Update potential sum for this i atom from the interaction with this j atom. */
 494             velecsum         = _mm_add_ps(velecsum,velec);
 495
 496             fscal            = felec;
 497
 498              /* Update vectorial force */
 499             fix2             = _mm_macc_ps(dx21,fscal,fix2);
 500             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
 501             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
 502
 503             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
 504             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
 505             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
 506
 507             /**************************
 508              * CALCULATE INTERACTIONS *
 509              **************************/
 510
 511             r22              = _mm_mul_ps(rsq22,rinv22);
 512
 513             /* Calculate table index by multiplying r with table scale and truncate to integer */
 514             rt               = _mm_mul_ps(r22,vftabscale);
 515             vfitab           = _mm_cvttps_epi32(rt);
 516 #ifdef __XOP__
 517             vfeps            = _mm_frcz_ps(rt);
 518 #else
 519             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 520 #endif
 521             twovfeps         = _mm_add_ps(vfeps,vfeps);
 522             vfitab           = _mm_slli_epi32(vfitab,2);
 523
 524             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 525             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 526             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 527             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 528             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 529             _MM_TRANSPOSE4_PS(Y,F,G,H);
 530             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 531             VV               = _mm_macc_ps(vfeps,Fp,Y);
 532             velec            = _mm_mul_ps(qq22,VV);
 533             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 534             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq22,FF),_mm_mul_ps(vftabscale,rinv22)));
 535
 536             /* Update potential sum for this i atom from the interaction with this j atom. */
 537             velecsum         = _mm_add_ps(velecsum,velec);
 538
 539             fscal            = felec;
 540
 541              /* Update vectorial force */
 542             fix2             = _mm_macc_ps(dx22,fscal,fix2);
 543             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
 544             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
 545
 546             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
 547             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
 548             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
 549
 550             /**************************
 551              * CALCULATE INTERACTIONS *
 552              **************************/
 553
 554             r23              = _mm_mul_ps(rsq23,rinv23);
 555
 556             /* Calculate table index by multiplying r with table scale and truncate to integer */
 557             rt               = _mm_mul_ps(r23,vftabscale);
 558             vfitab           = _mm_cvttps_epi32(rt);
 559 #ifdef __XOP__
 560             vfeps            = _mm_frcz_ps(rt);
 561 #else
 562             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 563 #endif
 564             twovfeps         = _mm_add_ps(vfeps,vfeps);
 565             vfitab           = _mm_slli_epi32(vfitab,2);
 566
 567             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 568             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 569             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 570             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 571             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 572             _MM_TRANSPOSE4_PS(Y,F,G,H);
 573             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 574             VV               = _mm_macc_ps(vfeps,Fp,Y);
 575             velec            = _mm_mul_ps(qq23,VV);
 576             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 577             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq23,FF),_mm_mul_ps(vftabscale,rinv23)));
 578
 579             /* Update potential sum for this i atom from the interaction with this j atom. */
 580             velecsum         = _mm_add_ps(velecsum,velec);
 581
 582             fscal            = felec;
 583
 584              /* Update vectorial force */
 585             fix2             = _mm_macc_ps(dx23,fscal,fix2);
 586             fiy2             = _mm_macc_ps(dy23,fscal,fiy2);
 587             fiz2             = _mm_macc_ps(dz23,fscal,fiz2);
 588
 589             fjx3             = _mm_macc_ps(dx23,fscal,fjx3);
 590             fjy3             = _mm_macc_ps(dy23,fscal,fjy3);
 591             fjz3             = _mm_macc_ps(dz23,fscal,fjz3);
 592
 593             /**************************
 594              * CALCULATE INTERACTIONS *
 595              **************************/
 596
 597             r31              = _mm_mul_ps(rsq31,rinv31);
 598
 599             /* Calculate table index by multiplying r with table scale and truncate to integer */
 600             rt               = _mm_mul_ps(r31,vftabscale);
 601             vfitab           = _mm_cvttps_epi32(rt);
 602 #ifdef __XOP__
 603             vfeps            = _mm_frcz_ps(rt);
 604 #else
 605             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 606 #endif
 607             twovfeps         = _mm_add_ps(vfeps,vfeps);
 608             vfitab           = _mm_slli_epi32(vfitab,2);
 609
 610             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 611             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 612             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 613             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 614             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 615             _MM_TRANSPOSE4_PS(Y,F,G,H);
 616             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 617             VV               = _mm_macc_ps(vfeps,Fp,Y);
 618             velec            = _mm_mul_ps(qq31,VV);
 619             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 620             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq31,FF),_mm_mul_ps(vftabscale,rinv31)));
 621
 622             /* Update potential sum for this i atom from the interaction with this j atom. */
 623             velecsum         = _mm_add_ps(velecsum,velec);
 624
 625             fscal            = felec;
 626
 627              /* Update vectorial force */
 628             fix3             = _mm_macc_ps(dx31,fscal,fix3);
 629             fiy3             = _mm_macc_ps(dy31,fscal,fiy3);
 630             fiz3             = _mm_macc_ps(dz31,fscal,fiz3);
 631
 632             fjx1             = _mm_macc_ps(dx31,fscal,fjx1);
 633             fjy1             = _mm_macc_ps(dy31,fscal,fjy1);
 634             fjz1             = _mm_macc_ps(dz31,fscal,fjz1);
 635
 636             /**************************
 637              * CALCULATE INTERACTIONS *
 638              **************************/
 639
 640             r32              = _mm_mul_ps(rsq32,rinv32);
 641
 642             /* Calculate table index by multiplying r with table scale and truncate to integer */
 643             rt               = _mm_mul_ps(r32,vftabscale);
 644             vfitab           = _mm_cvttps_epi32(rt);
 645 #ifdef __XOP__
 646             vfeps            = _mm_frcz_ps(rt);
 647 #else
 648             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 649 #endif
 650             twovfeps         = _mm_add_ps(vfeps,vfeps);
 651             vfitab           = _mm_slli_epi32(vfitab,2);
 652
 653             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 654             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 655             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 656             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 657             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 658             _MM_TRANSPOSE4_PS(Y,F,G,H);
 659             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 660             VV               = _mm_macc_ps(vfeps,Fp,Y);
 661             velec            = _mm_mul_ps(qq32,VV);
 662             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 663             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq32,FF),_mm_mul_ps(vftabscale,rinv32)));
 664
 665             /* Update potential sum for this i atom from the interaction with this j atom. */
 666             velecsum         = _mm_add_ps(velecsum,velec);
 667
 668             fscal            = felec;
 669
 670              /* Update vectorial force */
 671             fix3             = _mm_macc_ps(dx32,fscal,fix3);
 672             fiy3             = _mm_macc_ps(dy32,fscal,fiy3);
 673             fiz3             = _mm_macc_ps(dz32,fscal,fiz3);
 674
 675             fjx2             = _mm_macc_ps(dx32,fscal,fjx2);
 676             fjy2             = _mm_macc_ps(dy32,fscal,fjy2);
 677             fjz2             = _mm_macc_ps(dz32,fscal,fjz2);
 678
 679             /**************************
 680              * CALCULATE INTERACTIONS *
 681              **************************/
 682
 683             r33              = _mm_mul_ps(rsq33,rinv33);
 684
 685             /* Calculate table index by multiplying r with table scale and truncate to integer */
 686             rt               = _mm_mul_ps(r33,vftabscale);
 687             vfitab           = _mm_cvttps_epi32(rt);
 688 #ifdef __XOP__
 689             vfeps            = _mm_frcz_ps(rt);
 690 #else
 691             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 692 #endif
 693             twovfeps         = _mm_add_ps(vfeps,vfeps);
 694             vfitab           = _mm_slli_epi32(vfitab,2);
 695
 696             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 697             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 698             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 699             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 700             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 701             _MM_TRANSPOSE4_PS(Y,F,G,H);
 702             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 703             VV               = _mm_macc_ps(vfeps,Fp,Y);
 704             velec            = _mm_mul_ps(qq33,VV);
 705             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 706             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq33,FF),_mm_mul_ps(vftabscale,rinv33)));
 707
 708             /* Update potential sum for this i atom from the interaction with this j atom. */
 709             velecsum         = _mm_add_ps(velecsum,velec);
 710
 711             fscal            = felec;
 712
 713              /* Update vectorial force */
 714             fix3             = _mm_macc_ps(dx33,fscal,fix3);
 715             fiy3             = _mm_macc_ps(dy33,fscal,fiy3);
 716             fiz3             = _mm_macc_ps(dz33,fscal,fiz3);
 717
 718             fjx3             = _mm_macc_ps(dx33,fscal,fjx3);
 719             fjy3             = _mm_macc_ps(dy33,fscal,fjy3);
 720             fjz3             = _mm_macc_ps(dz33,fscal,fjz3);
 721
 722             fjptrA             = f+j_coord_offsetA;
 723             fjptrB             = f+j_coord_offsetB;
 724             fjptrC             = f+j_coord_offsetC;
 725             fjptrD             = f+j_coord_offsetD;
 726
 727             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 728                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
 729                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 730
 731             /* Inner loop uses 452 flops */
 732         }
 733
 734         if(jidx<j_index_end)
 735         {
 736
 737             /* Get j neighbor index, and coordinate index */
 738             jnrlistA         = jjnr[jidx];
 739             jnrlistB         = jjnr[jidx+1];
 740             jnrlistC         = jjnr[jidx+2];
 741             jnrlistD         = jjnr[jidx+3];
 742             /* Sign of each element will be negative for non-real atoms.
 743              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 744              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 745              */
 746             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 747             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 748             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 749             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 750             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 751             j_coord_offsetA  = DIM*jnrA;
 752             j_coord_offsetB  = DIM*jnrB;
 753             j_coord_offsetC  = DIM*jnrC;
 754             j_coord_offsetD  = DIM*jnrD;
 755
 756             /* load j atom coordinates */
 757             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 758                                               x+j_coord_offsetC,x+j_coord_offsetD,
 759                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 760                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 761
 762             /* Calculate displacement vector */
 763             dx00             = _mm_sub_ps(ix0,jx0);
 764             dy00             = _mm_sub_ps(iy0,jy0);
 765             dz00             = _mm_sub_ps(iz0,jz0);
 766             dx11             = _mm_sub_ps(ix1,jx1);
 767             dy11             = _mm_sub_ps(iy1,jy1);
 768             dz11             = _mm_sub_ps(iz1,jz1);
 769             dx12             = _mm_sub_ps(ix1,jx2);
 770             dy12             = _mm_sub_ps(iy1,jy2);
 771             dz12             = _mm_sub_ps(iz1,jz2);
 772             dx13             = _mm_sub_ps(ix1,jx3);
 773             dy13             = _mm_sub_ps(iy1,jy3);
 774             dz13             = _mm_sub_ps(iz1,jz3);
 775             dx21             = _mm_sub_ps(ix2,jx1);
 776             dy21             = _mm_sub_ps(iy2,jy1);
 777             dz21             = _mm_sub_ps(iz2,jz1);
 778             dx22             = _mm_sub_ps(ix2,jx2);
 779             dy22             = _mm_sub_ps(iy2,jy2);
 780             dz22             = _mm_sub_ps(iz2,jz2);
 781             dx23             = _mm_sub_ps(ix2,jx3);
 782             dy23             = _mm_sub_ps(iy2,jy3);
 783             dz23             = _mm_sub_ps(iz2,jz3);
 784             dx31             = _mm_sub_ps(ix3,jx1);
 785             dy31             = _mm_sub_ps(iy3,jy1);
 786             dz31             = _mm_sub_ps(iz3,jz1);
 787             dx32             = _mm_sub_ps(ix3,jx2);
 788             dy32             = _mm_sub_ps(iy3,jy2);
 789             dz32             = _mm_sub_ps(iz3,jz2);
 790             dx33             = _mm_sub_ps(ix3,jx3);
 791             dy33             = _mm_sub_ps(iy3,jy3);
 792             dz33             = _mm_sub_ps(iz3,jz3);
 793
 794             /* Calculate squared distance and things based on it */
 795             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 796             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 797             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 798             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
 799             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 800             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 801             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
 802             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
 803             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
 804             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
 805
 806             rinv11           = gmx_mm_invsqrt_ps(rsq11);
 807             rinv12           = gmx_mm_invsqrt_ps(rsq12);
 808             rinv13           = gmx_mm_invsqrt_ps(rsq13);
 809             rinv21           = gmx_mm_invsqrt_ps(rsq21);
 810             rinv22           = gmx_mm_invsqrt_ps(rsq22);
 811             rinv23           = gmx_mm_invsqrt_ps(rsq23);
 812             rinv31           = gmx_mm_invsqrt_ps(rsq31);
 813             rinv32           = gmx_mm_invsqrt_ps(rsq32);
 814             rinv33           = gmx_mm_invsqrt_ps(rsq33);
 815
 816             rinvsq00         = gmx_mm_inv_ps(rsq00);
 817
 818             fjx0             = _mm_setzero_ps();
 819             fjy0             = _mm_setzero_ps();
 820             fjz0             = _mm_setzero_ps();
 821             fjx1             = _mm_setzero_ps();
 822             fjy1             = _mm_setzero_ps();
 823             fjz1             = _mm_setzero_ps();
 824             fjx2             = _mm_setzero_ps();
 825             fjy2             = _mm_setzero_ps();
 826             fjz2             = _mm_setzero_ps();
 827             fjx3             = _mm_setzero_ps();
 828             fjy3             = _mm_setzero_ps();
 829             fjz3             = _mm_setzero_ps();
 830
 831             /**************************
 832              * CALCULATE INTERACTIONS *
 833              **************************/
 834
 835             /* LENNARD-JONES DISPERSION/REPULSION */
 836
 837             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 838             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 839             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 840             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 841             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 842
 843             /* Update potential sum for this i atom from the interaction with this j atom. */
 844             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 845             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 846
 847             fscal            = fvdw;
 848
 849             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 850
 851              /* Update vectorial force */
 852             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 853             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 854             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 855
 856             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 857             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 858             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 859
 860             /**************************
 861              * CALCULATE INTERACTIONS *
 862              **************************/
 863
 864             r11              = _mm_mul_ps(rsq11,rinv11);
 865             r11              = _mm_andnot_ps(dummy_mask,r11);
 866
 867             /* Calculate table index by multiplying r with table scale and truncate to integer */
 868             rt               = _mm_mul_ps(r11,vftabscale);
 869             vfitab           = _mm_cvttps_epi32(rt);
 870 #ifdef __XOP__
 871             vfeps            = _mm_frcz_ps(rt);
 872 #else
 873             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 874 #endif
 875             twovfeps         = _mm_add_ps(vfeps,vfeps);
 876             vfitab           = _mm_slli_epi32(vfitab,2);
 877
 878             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 879             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 880             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 881             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 882             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 883             _MM_TRANSPOSE4_PS(Y,F,G,H);
 884             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 885             VV               = _mm_macc_ps(vfeps,Fp,Y);
 886             velec            = _mm_mul_ps(qq11,VV);
 887             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 888             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq11,FF),_mm_mul_ps(vftabscale,rinv11)));
 889
 890             /* Update potential sum for this i atom from the interaction with this j atom. */
 891             velec            = _mm_andnot_ps(dummy_mask,velec);
 892             velecsum         = _mm_add_ps(velecsum,velec);
 893
 894             fscal            = felec;
 895
 896             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 897
 898              /* Update vectorial force */
 899             fix1             = _mm_macc_ps(dx11,fscal,fix1);
 900             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
 901             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
 902
 903             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
 904             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
 905             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
 906
 907             /**************************
 908              * CALCULATE INTERACTIONS *
 909              **************************/
 910
 911             r12              = _mm_mul_ps(rsq12,rinv12);
 912             r12              = _mm_andnot_ps(dummy_mask,r12);
 913
 914             /* Calculate table index by multiplying r with table scale and truncate to integer */
 915             rt               = _mm_mul_ps(r12,vftabscale);
 916             vfitab           = _mm_cvttps_epi32(rt);
 917 #ifdef __XOP__
 918             vfeps            = _mm_frcz_ps(rt);
 919 #else
 920             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 921 #endif
 922             twovfeps         = _mm_add_ps(vfeps,vfeps);
 923             vfitab           = _mm_slli_epi32(vfitab,2);
 924
 925             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 926             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 927             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 928             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 929             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 930             _MM_TRANSPOSE4_PS(Y,F,G,H);
 931             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 932             VV               = _mm_macc_ps(vfeps,Fp,Y);
 933             velec            = _mm_mul_ps(qq12,VV);
 934             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 935             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq12,FF),_mm_mul_ps(vftabscale,rinv12)));
 936
 937             /* Update potential sum for this i atom from the interaction with this j atom. */
 938             velec            = _mm_andnot_ps(dummy_mask,velec);
 939             velecsum         = _mm_add_ps(velecsum,velec);
 940
 941             fscal            = felec;
 942
 943             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 944
 945              /* Update vectorial force */
 946             fix1             = _mm_macc_ps(dx12,fscal,fix1);
 947             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
 948             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
 949
 950             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
 951             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
 952             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
 953
 954             /**************************
 955              * CALCULATE INTERACTIONS *
 956              **************************/
 957
 958             r13              = _mm_mul_ps(rsq13,rinv13);
 959             r13              = _mm_andnot_ps(dummy_mask,r13);
 960
 961             /* Calculate table index by multiplying r with table scale and truncate to integer */
 962             rt               = _mm_mul_ps(r13,vftabscale);
 963             vfitab           = _mm_cvttps_epi32(rt);
 964 #ifdef __XOP__
 965             vfeps            = _mm_frcz_ps(rt);
 966 #else
 967             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 968 #endif
 969             twovfeps         = _mm_add_ps(vfeps,vfeps);
 970             vfitab           = _mm_slli_epi32(vfitab,2);
 971
 972             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 973             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 974             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 975             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 976             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 977             _MM_TRANSPOSE4_PS(Y,F,G,H);
 978             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 979             VV               = _mm_macc_ps(vfeps,Fp,Y);
 980             velec            = _mm_mul_ps(qq13,VV);
 981             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 982             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq13,FF),_mm_mul_ps(vftabscale,rinv13)));
 983
 984             /* Update potential sum for this i atom from the interaction with this j atom. */
 985             velec            = _mm_andnot_ps(dummy_mask,velec);
 986             velecsum         = _mm_add_ps(velecsum,velec);
 987
 988             fscal            = felec;
 989
 990             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 991
 992              /* Update vectorial force */
 993             fix1             = _mm_macc_ps(dx13,fscal,fix1);
 994             fiy1             = _mm_macc_ps(dy13,fscal,fiy1);
 995             fiz1             = _mm_macc_ps(dz13,fscal,fiz1);
 996
 997             fjx3             = _mm_macc_ps(dx13,fscal,fjx3);
 998             fjy3             = _mm_macc_ps(dy13,fscal,fjy3);
 999             fjz3             = _mm_macc_ps(dz13,fscal,fjz3);
1000
1001             /**************************
1002              * CALCULATE INTERACTIONS *
1003              **************************/
1004
1005             r21              = _mm_mul_ps(rsq21,rinv21);
1006             r21              = _mm_andnot_ps(dummy_mask,r21);
1007
1008             /* Calculate table index by multiplying r with table scale and truncate to integer */
1009             rt               = _mm_mul_ps(r21,vftabscale);
1010             vfitab           = _mm_cvttps_epi32(rt);
1011 #ifdef __XOP__
1012             vfeps            = _mm_frcz_ps(rt);
1013 #else
1014             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1015 #endif
1016             twovfeps         = _mm_add_ps(vfeps,vfeps);
1017             vfitab           = _mm_slli_epi32(vfitab,2);
1018
1019             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1020             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1021             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1022             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1023             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1024             _MM_TRANSPOSE4_PS(Y,F,G,H);
1025             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1026             VV               = _mm_macc_ps(vfeps,Fp,Y);
1027             velec            = _mm_mul_ps(qq21,VV);
1028             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1029             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq21,FF),_mm_mul_ps(vftabscale,rinv21)));
1030
1031             /* Update potential sum for this i atom from the interaction with this j atom. */
1032             velec            = _mm_andnot_ps(dummy_mask,velec);
1033             velecsum         = _mm_add_ps(velecsum,velec);
1034
1035             fscal            = felec;
1036
1037             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1038
1039              /* Update vectorial force */
1040             fix2             = _mm_macc_ps(dx21,fscal,fix2);
1041             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
1042             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
1043
1044             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
1045             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
1046             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
1047
1048             /**************************
1049              * CALCULATE INTERACTIONS *
1050              **************************/
1051
1052             r22              = _mm_mul_ps(rsq22,rinv22);
1053             r22              = _mm_andnot_ps(dummy_mask,r22);
1054
1055             /* Calculate table index by multiplying r with table scale and truncate to integer */
1056             rt               = _mm_mul_ps(r22,vftabscale);
1057             vfitab           = _mm_cvttps_epi32(rt);
1058 #ifdef __XOP__
1059             vfeps            = _mm_frcz_ps(rt);
1060 #else
1061             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1062 #endif
1063             twovfeps         = _mm_add_ps(vfeps,vfeps);
1064             vfitab           = _mm_slli_epi32(vfitab,2);
1065
1066             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1067             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1068             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1069             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1070             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1071             _MM_TRANSPOSE4_PS(Y,F,G,H);
1072             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1073             VV               = _mm_macc_ps(vfeps,Fp,Y);
1074             velec            = _mm_mul_ps(qq22,VV);
1075             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1076             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq22,FF),_mm_mul_ps(vftabscale,rinv22)));
1077
1078             /* Update potential sum for this i atom from the interaction with this j atom. */
1079             velec            = _mm_andnot_ps(dummy_mask,velec);
1080             velecsum         = _mm_add_ps(velecsum,velec);
1081
1082             fscal            = felec;
1083
1084             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1085
1086              /* Update vectorial force */
1087             fix2             = _mm_macc_ps(dx22,fscal,fix2);
1088             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
1089             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
1090
1091             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
1092             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
1093             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
1094
1095             /**************************
1096              * CALCULATE INTERACTIONS *
1097              **************************/
1098
1099             r23              = _mm_mul_ps(rsq23,rinv23);
1100             r23              = _mm_andnot_ps(dummy_mask,r23);
1101
1102             /* Calculate table index by multiplying r with table scale and truncate to integer */
1103             rt               = _mm_mul_ps(r23,vftabscale);
1104             vfitab           = _mm_cvttps_epi32(rt);
1105 #ifdef __XOP__
1106             vfeps            = _mm_frcz_ps(rt);
1107 #else
1108             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1109 #endif
1110             twovfeps         = _mm_add_ps(vfeps,vfeps);
1111             vfitab           = _mm_slli_epi32(vfitab,2);
1112
1113             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1114             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1115             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1116             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1117             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1118             _MM_TRANSPOSE4_PS(Y,F,G,H);
1119             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1120             VV               = _mm_macc_ps(vfeps,Fp,Y);
1121             velec            = _mm_mul_ps(qq23,VV);
1122             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1123             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq23,FF),_mm_mul_ps(vftabscale,rinv23)));
1124
1125             /* Update potential sum for this i atom from the interaction with this j atom. */
1126             velec            = _mm_andnot_ps(dummy_mask,velec);
1127             velecsum         = _mm_add_ps(velecsum,velec);
1128
1129             fscal            = felec;
1130
1131             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1132
1133              /* Update vectorial force */
1134             fix2             = _mm_macc_ps(dx23,fscal,fix2);
1135             fiy2             = _mm_macc_ps(dy23,fscal,fiy2);
1136             fiz2             = _mm_macc_ps(dz23,fscal,fiz2);
1137
1138             fjx3             = _mm_macc_ps(dx23,fscal,fjx3);
1139             fjy3             = _mm_macc_ps(dy23,fscal,fjy3);
1140             fjz3             = _mm_macc_ps(dz23,fscal,fjz3);
1141
1142             /**************************
1143              * CALCULATE INTERACTIONS *
1144              **************************/
1145
1146             r31              = _mm_mul_ps(rsq31,rinv31);
1147             r31              = _mm_andnot_ps(dummy_mask,r31);
1148
1149             /* Calculate table index by multiplying r with table scale and truncate to integer */
1150             rt               = _mm_mul_ps(r31,vftabscale);
1151             vfitab           = _mm_cvttps_epi32(rt);
1152 #ifdef __XOP__
1153             vfeps            = _mm_frcz_ps(rt);
1154 #else
1155             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1156 #endif
1157             twovfeps         = _mm_add_ps(vfeps,vfeps);
1158             vfitab           = _mm_slli_epi32(vfitab,2);
1159
1160             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1161             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1162             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1163             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1164             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1165             _MM_TRANSPOSE4_PS(Y,F,G,H);
1166             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1167             VV               = _mm_macc_ps(vfeps,Fp,Y);
1168             velec            = _mm_mul_ps(qq31,VV);
1169             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1170             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq31,FF),_mm_mul_ps(vftabscale,rinv31)));
1171
1172             /* Update potential sum for this i atom from the interaction with this j atom. */
1173             velec            = _mm_andnot_ps(dummy_mask,velec);
1174             velecsum         = _mm_add_ps(velecsum,velec);
1175
1176             fscal            = felec;
1177
1178             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1179
1180              /* Update vectorial force */
1181             fix3             = _mm_macc_ps(dx31,fscal,fix3);
1182             fiy3             = _mm_macc_ps(dy31,fscal,fiy3);
1183             fiz3             = _mm_macc_ps(dz31,fscal,fiz3);
1184
1185             fjx1             = _mm_macc_ps(dx31,fscal,fjx1);
1186             fjy1             = _mm_macc_ps(dy31,fscal,fjy1);
1187             fjz1             = _mm_macc_ps(dz31,fscal,fjz1);
1188
1189             /**************************
1190              * CALCULATE INTERACTIONS *
1191              **************************/
1192
1193             r32              = _mm_mul_ps(rsq32,rinv32);
1194             r32              = _mm_andnot_ps(dummy_mask,r32);
1195
1196             /* Calculate table index by multiplying r with table scale and truncate to integer */
1197             rt               = _mm_mul_ps(r32,vftabscale);
1198             vfitab           = _mm_cvttps_epi32(rt);
1199 #ifdef __XOP__
1200             vfeps            = _mm_frcz_ps(rt);
1201 #else
1202             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1203 #endif
1204             twovfeps         = _mm_add_ps(vfeps,vfeps);
1205             vfitab           = _mm_slli_epi32(vfitab,2);
1206
1207             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1208             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1209             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1210             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1211             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1212             _MM_TRANSPOSE4_PS(Y,F,G,H);
1213             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1214             VV               = _mm_macc_ps(vfeps,Fp,Y);
1215             velec            = _mm_mul_ps(qq32,VV);
1216             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1217             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq32,FF),_mm_mul_ps(vftabscale,rinv32)));
1218
1219             /* Update potential sum for this i atom from the interaction with this j atom. */
1220             velec            = _mm_andnot_ps(dummy_mask,velec);
1221             velecsum         = _mm_add_ps(velecsum,velec);
1222
1223             fscal            = felec;
1224
1225             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1226
1227              /* Update vectorial force */
1228             fix3             = _mm_macc_ps(dx32,fscal,fix3);
1229             fiy3             = _mm_macc_ps(dy32,fscal,fiy3);
1230             fiz3             = _mm_macc_ps(dz32,fscal,fiz3);
1231
1232             fjx2             = _mm_macc_ps(dx32,fscal,fjx2);
1233             fjy2             = _mm_macc_ps(dy32,fscal,fjy2);
1234             fjz2             = _mm_macc_ps(dz32,fscal,fjz2);
1235
1236             /**************************
1237              * CALCULATE INTERACTIONS *
1238              **************************/
1239
1240             r33              = _mm_mul_ps(rsq33,rinv33);
1241             r33              = _mm_andnot_ps(dummy_mask,r33);
1242
1243             /* Calculate table index by multiplying r with table scale and truncate to integer */
1244             rt               = _mm_mul_ps(r33,vftabscale);
1245             vfitab           = _mm_cvttps_epi32(rt);
1246 #ifdef __XOP__
1247             vfeps            = _mm_frcz_ps(rt);
1248 #else
1249             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1250 #endif
1251             twovfeps         = _mm_add_ps(vfeps,vfeps);
1252             vfitab           = _mm_slli_epi32(vfitab,2);
1253
1254             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1255             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1256             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1257             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1258             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1259             _MM_TRANSPOSE4_PS(Y,F,G,H);
1260             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1261             VV               = _mm_macc_ps(vfeps,Fp,Y);
1262             velec            = _mm_mul_ps(qq33,VV);
1263             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1264             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq33,FF),_mm_mul_ps(vftabscale,rinv33)));
1265
1266             /* Update potential sum for this i atom from the interaction with this j atom. */
1267             velec            = _mm_andnot_ps(dummy_mask,velec);
1268             velecsum         = _mm_add_ps(velecsum,velec);
1269
1270             fscal            = felec;
1271
1272             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1273
1274              /* Update vectorial force */
1275             fix3             = _mm_macc_ps(dx33,fscal,fix3);
1276             fiy3             = _mm_macc_ps(dy33,fscal,fiy3);
1277             fiz3             = _mm_macc_ps(dz33,fscal,fiz3);
1278
1279             fjx3             = _mm_macc_ps(dx33,fscal,fjx3);
1280             fjy3             = _mm_macc_ps(dy33,fscal,fjy3);
1281             fjz3             = _mm_macc_ps(dz33,fscal,fjz3);
1282
1283             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1284             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1285             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1286             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1287
1288             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1289                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1290                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1291
1292             /* Inner loop uses 461 flops */
1293         }
1294
1295         /* End of innermost loop */
1296
1297         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1298                                               f+i_coord_offset,fshift+i_shift_offset);
1299
1300         ggid                        = gid[iidx];
1301         /* Update potential energies */
1302         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1303         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1304
1305         /* Increment number of inner iterations */
1306         inneriter                  += j_index_end - j_index_start;
1307
1308         /* Outer loop uses 26 flops */
1309     }
1310
1311     /* Increment number of outer iterations */
1312     outeriter        += nri;
1313
1314     /* Update outer/inner flops */
1315
1316     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*461);
1317 }
1318 /*
1319  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW4W4_F_avx_128_fma_single
1320  * Electrostatics interaction: CubicSplineTable
1321  * VdW interaction:            LennardJones
1322  * Geometry:                   Water4-Water4
1323  * Calculate force/pot:        Force
1324  */
1325 void
1326 nb_kernel_ElecCSTab_VdwLJ_GeomW4W4_F_avx_128_fma_single
1327                     (t_nblist                    * gmx_restrict       nlist,
1328                      rvec                        * gmx_restrict          xx,
1329                      rvec                        * gmx_restrict          ff,
1330                      t_forcerec                  * gmx_restrict          fr,
1331                      t_mdatoms                   * gmx_restrict     mdatoms,
1332                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1333                      t_nrnb                      * gmx_restrict        nrnb)
1334 {
1335     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1336      * just 0 for non-waters.
1337      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
1338      * jnr indices corresponding to data put in the four positions in the SIMD register.
1339      */
1340     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1341     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1342     int              jnrA,jnrB,jnrC,jnrD;
1343     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1344     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1345     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1346     real             rcutoff_scalar;
1347     real             *shiftvec,*fshift,*x,*f;
1348     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1349     real             scratch[4*DIM];
1350     __m128           fscal,rcutoff,rcutoff2,jidxall;
1351     int              vdwioffset0;
1352     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1353     int              vdwioffset1;
1354     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1355     int              vdwioffset2;
1356     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1357     int              vdwioffset3;
1358     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1359     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1360     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1361     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1362     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1363     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1364     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1365     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
1366     __m128           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1367     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1368     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1369     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1370     __m128           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1371     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1372     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1373     __m128           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1374     __m128           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1375     __m128           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1376     __m128           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1377     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
1378     real             *charge;
1379     int              nvdwtype;
1380     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1381     int              *vdwtype;
1382     real             *vdwparam;
1383     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
1384     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
1385     __m128i          vfitab;
1386     __m128i          ifour       = _mm_set1_epi32(4);
1387     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
1388     real             *vftab;
1389     __m128           dummy_mask,cutoff_mask;
1390     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
1391     __m128           one     = _mm_set1_ps(1.0);
1392     __m128           two     = _mm_set1_ps(2.0);
1393     x                = xx[0];
1394     f                = ff[0];
1395
1396     nri              = nlist->nri;
1397     iinr             = nlist->iinr;
1398     jindex           = nlist->jindex;
1399     jjnr             = nlist->jjnr;
1400     shiftidx         = nlist->shift;
1401     gid              = nlist->gid;
1402     shiftvec         = fr->shift_vec[0];
1403     fshift           = fr->fshift[0];
1404     facel            = _mm_set1_ps(fr->epsfac);
1405     charge           = mdatoms->chargeA;
1406     nvdwtype         = fr->ntype;
1407     vdwparam         = fr->nbfp;
1408     vdwtype          = mdatoms->typeA;
1409
1410     vftab            = kernel_data->table_elec->data;
1411     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
1412
1413     /* Setup water-specific parameters */
1414     inr              = nlist->iinr[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     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
1418     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
1419
1420     jq1              = _mm_set1_ps(charge[inr+1]);
1421     jq2              = _mm_set1_ps(charge[inr+2]);
1422     jq3              = _mm_set1_ps(charge[inr+3]);
1423     vdwjidx0A        = 2*vdwtype[inr+0];
1424     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
1425     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
1426     qq11             = _mm_mul_ps(iq1,jq1);
1427     qq12             = _mm_mul_ps(iq1,jq2);
1428     qq13             = _mm_mul_ps(iq1,jq3);
1429     qq21             = _mm_mul_ps(iq2,jq1);
1430     qq22             = _mm_mul_ps(iq2,jq2);
1431     qq23             = _mm_mul_ps(iq2,jq3);
1432     qq31             = _mm_mul_ps(iq3,jq1);
1433     qq32             = _mm_mul_ps(iq3,jq2);
1434     qq33             = _mm_mul_ps(iq3,jq3);
1435
1436     /* Avoid stupid compiler warnings */
1437     jnrA = jnrB = jnrC = jnrD = 0;
1438     j_coord_offsetA = 0;
1439     j_coord_offsetB = 0;
1440     j_coord_offsetC = 0;
1441     j_coord_offsetD = 0;
1442
1443     outeriter        = 0;
1444     inneriter        = 0;
1445
1446     for(iidx=0;iidx<4*DIM;iidx++)
1447     {
1448         scratch[iidx] = 0.0;
1449     }
1450
1451     /* Start outer loop over neighborlists */
1452     for(iidx=0; iidx<nri; iidx++)
1453     {
1454         /* Load shift vector for this list */
1455         i_shift_offset   = DIM*shiftidx[iidx];
1456
1457         /* Load limits for loop over neighbors */
1458         j_index_start    = jindex[iidx];
1459         j_index_end      = jindex[iidx+1];
1460
1461         /* Get outer coordinate index */
1462         inr              = iinr[iidx];
1463         i_coord_offset   = DIM*inr;
1464
1465         /* Load i particle coords and add shift vector */
1466         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1467                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1468
1469         fix0             = _mm_setzero_ps();
1470         fiy0             = _mm_setzero_ps();
1471         fiz0             = _mm_setzero_ps();
1472         fix1             = _mm_setzero_ps();
1473         fiy1             = _mm_setzero_ps();
1474         fiz1             = _mm_setzero_ps();
1475         fix2             = _mm_setzero_ps();
1476         fiy2             = _mm_setzero_ps();
1477         fiz2             = _mm_setzero_ps();
1478         fix3             = _mm_setzero_ps();
1479         fiy3             = _mm_setzero_ps();
1480         fiz3             = _mm_setzero_ps();
1481
1482         /* Start inner kernel loop */
1483         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1484         {
1485
1486             /* Get j neighbor index, and coordinate index */
1487             jnrA             = jjnr[jidx];
1488             jnrB             = jjnr[jidx+1];
1489             jnrC             = jjnr[jidx+2];
1490             jnrD             = jjnr[jidx+3];
1491             j_coord_offsetA  = DIM*jnrA;
1492             j_coord_offsetB  = DIM*jnrB;
1493             j_coord_offsetC  = DIM*jnrC;
1494             j_coord_offsetD  = DIM*jnrD;
1495
1496             /* load j atom coordinates */
1497             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1498                                               x+j_coord_offsetC,x+j_coord_offsetD,
1499                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1500                                               &jy2,&jz2,&jx3,&jy3,&jz3);
1501
1502             /* Calculate displacement vector */
1503             dx00             = _mm_sub_ps(ix0,jx0);
1504             dy00             = _mm_sub_ps(iy0,jy0);
1505             dz00             = _mm_sub_ps(iz0,jz0);
1506             dx11             = _mm_sub_ps(ix1,jx1);
1507             dy11             = _mm_sub_ps(iy1,jy1);
1508             dz11             = _mm_sub_ps(iz1,jz1);
1509             dx12             = _mm_sub_ps(ix1,jx2);
1510             dy12             = _mm_sub_ps(iy1,jy2);
1511             dz12             = _mm_sub_ps(iz1,jz2);
1512             dx13             = _mm_sub_ps(ix1,jx3);
1513             dy13             = _mm_sub_ps(iy1,jy3);
1514             dz13             = _mm_sub_ps(iz1,jz3);
1515             dx21             = _mm_sub_ps(ix2,jx1);
1516             dy21             = _mm_sub_ps(iy2,jy1);
1517             dz21             = _mm_sub_ps(iz2,jz1);
1518             dx22             = _mm_sub_ps(ix2,jx2);
1519             dy22             = _mm_sub_ps(iy2,jy2);
1520             dz22             = _mm_sub_ps(iz2,jz2);
1521             dx23             = _mm_sub_ps(ix2,jx3);
1522             dy23             = _mm_sub_ps(iy2,jy3);
1523             dz23             = _mm_sub_ps(iz2,jz3);
1524             dx31             = _mm_sub_ps(ix3,jx1);
1525             dy31             = _mm_sub_ps(iy3,jy1);
1526             dz31             = _mm_sub_ps(iz3,jz1);
1527             dx32             = _mm_sub_ps(ix3,jx2);
1528             dy32             = _mm_sub_ps(iy3,jy2);
1529             dz32             = _mm_sub_ps(iz3,jz2);
1530             dx33             = _mm_sub_ps(ix3,jx3);
1531             dy33             = _mm_sub_ps(iy3,jy3);
1532             dz33             = _mm_sub_ps(iz3,jz3);
1533
1534             /* Calculate squared distance and things based on it */
1535             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1536             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1537             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1538             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
1539             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1540             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1541             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
1542             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
1543             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
1544             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
1545
1546             rinv11           = gmx_mm_invsqrt_ps(rsq11);
1547             rinv12           = gmx_mm_invsqrt_ps(rsq12);
1548             rinv13           = gmx_mm_invsqrt_ps(rsq13);
1549             rinv21           = gmx_mm_invsqrt_ps(rsq21);
1550             rinv22           = gmx_mm_invsqrt_ps(rsq22);
1551             rinv23           = gmx_mm_invsqrt_ps(rsq23);
1552             rinv31           = gmx_mm_invsqrt_ps(rsq31);
1553             rinv32           = gmx_mm_invsqrt_ps(rsq32);
1554             rinv33           = gmx_mm_invsqrt_ps(rsq33);
1555
1556             rinvsq00         = gmx_mm_inv_ps(rsq00);
1557
1558             fjx0             = _mm_setzero_ps();
1559             fjy0             = _mm_setzero_ps();
1560             fjz0             = _mm_setzero_ps();
1561             fjx1             = _mm_setzero_ps();
1562             fjy1             = _mm_setzero_ps();
1563             fjz1             = _mm_setzero_ps();
1564             fjx2             = _mm_setzero_ps();
1565             fjy2             = _mm_setzero_ps();
1566             fjz2             = _mm_setzero_ps();
1567             fjx3             = _mm_setzero_ps();
1568             fjy3             = _mm_setzero_ps();
1569             fjz3             = _mm_setzero_ps();
1570
1571             /**************************
1572              * CALCULATE INTERACTIONS *
1573              **************************/
1574
1575             /* LENNARD-JONES DISPERSION/REPULSION */
1576
1577             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1578             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1579
1580             fscal            = fvdw;
1581
1582              /* Update vectorial force */
1583             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1584             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1585             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1586
1587             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1588             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1589             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1590
1591             /**************************
1592              * CALCULATE INTERACTIONS *
1593              **************************/
1594
1595             r11              = _mm_mul_ps(rsq11,rinv11);
1596
1597             /* Calculate table index by multiplying r with table scale and truncate to integer */
1598             rt               = _mm_mul_ps(r11,vftabscale);
1599             vfitab           = _mm_cvttps_epi32(rt);
1600 #ifdef __XOP__
1601             vfeps            = _mm_frcz_ps(rt);
1602 #else
1603             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1604 #endif
1605             twovfeps         = _mm_add_ps(vfeps,vfeps);
1606             vfitab           = _mm_slli_epi32(vfitab,2);
1607
1608             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1609             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1610             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1611             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1612             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1613             _MM_TRANSPOSE4_PS(Y,F,G,H);
1614             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1615             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1616             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq11,FF),_mm_mul_ps(vftabscale,rinv11)));
1617
1618             fscal            = felec;
1619
1620              /* Update vectorial force */
1621             fix1             = _mm_macc_ps(dx11,fscal,fix1);
1622             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
1623             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
1624
1625             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
1626             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
1627             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
1628
1629             /**************************
1630              * CALCULATE INTERACTIONS *
1631              **************************/
1632
1633             r12              = _mm_mul_ps(rsq12,rinv12);
1634
1635             /* Calculate table index by multiplying r with table scale and truncate to integer */
1636             rt               = _mm_mul_ps(r12,vftabscale);
1637             vfitab           = _mm_cvttps_epi32(rt);
1638 #ifdef __XOP__
1639             vfeps            = _mm_frcz_ps(rt);
1640 #else
1641             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1642 #endif
1643             twovfeps         = _mm_add_ps(vfeps,vfeps);
1644             vfitab           = _mm_slli_epi32(vfitab,2);
1645
1646             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1647             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1648             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1649             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1650             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1651             _MM_TRANSPOSE4_PS(Y,F,G,H);
1652             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1653             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1654             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq12,FF),_mm_mul_ps(vftabscale,rinv12)));
1655
1656             fscal            = felec;
1657
1658              /* Update vectorial force */
1659             fix1             = _mm_macc_ps(dx12,fscal,fix1);
1660             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
1661             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
1662
1663             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
1664             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
1665             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
1666
1667             /**************************
1668              * CALCULATE INTERACTIONS *
1669              **************************/
1670
1671             r13              = _mm_mul_ps(rsq13,rinv13);
1672
1673             /* Calculate table index by multiplying r with table scale and truncate to integer */
1674             rt               = _mm_mul_ps(r13,vftabscale);
1675             vfitab           = _mm_cvttps_epi32(rt);
1676 #ifdef __XOP__
1677             vfeps            = _mm_frcz_ps(rt);
1678 #else
1679             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1680 #endif
1681             twovfeps         = _mm_add_ps(vfeps,vfeps);
1682             vfitab           = _mm_slli_epi32(vfitab,2);
1683
1684             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1685             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1686             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1687             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1688             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1689             _MM_TRANSPOSE4_PS(Y,F,G,H);
1690             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1691             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1692             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq13,FF),_mm_mul_ps(vftabscale,rinv13)));
1693
1694             fscal            = felec;
1695
1696              /* Update vectorial force */
1697             fix1             = _mm_macc_ps(dx13,fscal,fix1);
1698             fiy1             = _mm_macc_ps(dy13,fscal,fiy1);
1699             fiz1             = _mm_macc_ps(dz13,fscal,fiz1);
1700
1701             fjx3             = _mm_macc_ps(dx13,fscal,fjx3);
1702             fjy3             = _mm_macc_ps(dy13,fscal,fjy3);
1703             fjz3             = _mm_macc_ps(dz13,fscal,fjz3);
1704
1705             /**************************
1706              * CALCULATE INTERACTIONS *
1707              **************************/
1708
1709             r21              = _mm_mul_ps(rsq21,rinv21);
1710
1711             /* Calculate table index by multiplying r with table scale and truncate to integer */
1712             rt               = _mm_mul_ps(r21,vftabscale);
1713             vfitab           = _mm_cvttps_epi32(rt);
1714 #ifdef __XOP__
1715             vfeps            = _mm_frcz_ps(rt);
1716 #else
1717             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1718 #endif
1719             twovfeps         = _mm_add_ps(vfeps,vfeps);
1720             vfitab           = _mm_slli_epi32(vfitab,2);
1721
1722             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1723             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1724             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1725             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1726             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1727             _MM_TRANSPOSE4_PS(Y,F,G,H);
1728             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1729             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1730             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq21,FF),_mm_mul_ps(vftabscale,rinv21)));
1731
1732             fscal            = felec;
1733
1734              /* Update vectorial force */
1735             fix2             = _mm_macc_ps(dx21,fscal,fix2);
1736             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
1737             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
1738
1739             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
1740             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
1741             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
1742
1743             /**************************
1744              * CALCULATE INTERACTIONS *
1745              **************************/
1746
1747             r22              = _mm_mul_ps(rsq22,rinv22);
1748
1749             /* Calculate table index by multiplying r with table scale and truncate to integer */
1750             rt               = _mm_mul_ps(r22,vftabscale);
1751             vfitab           = _mm_cvttps_epi32(rt);
1752 #ifdef __XOP__
1753             vfeps            = _mm_frcz_ps(rt);
1754 #else
1755             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1756 #endif
1757             twovfeps         = _mm_add_ps(vfeps,vfeps);
1758             vfitab           = _mm_slli_epi32(vfitab,2);
1759
1760             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1761             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1762             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1763             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1764             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1765             _MM_TRANSPOSE4_PS(Y,F,G,H);
1766             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1767             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1768             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq22,FF),_mm_mul_ps(vftabscale,rinv22)));
1769
1770             fscal            = felec;
1771
1772              /* Update vectorial force */
1773             fix2             = _mm_macc_ps(dx22,fscal,fix2);
1774             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
1775             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
1776
1777             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
1778             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
1779             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
1780
1781             /**************************
1782              * CALCULATE INTERACTIONS *
1783              **************************/
1784
1785             r23              = _mm_mul_ps(rsq23,rinv23);
1786
1787             /* Calculate table index by multiplying r with table scale and truncate to integer */
1788             rt               = _mm_mul_ps(r23,vftabscale);
1789             vfitab           = _mm_cvttps_epi32(rt);
1790 #ifdef __XOP__
1791             vfeps            = _mm_frcz_ps(rt);
1792 #else
1793             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1794 #endif
1795             twovfeps         = _mm_add_ps(vfeps,vfeps);
1796             vfitab           = _mm_slli_epi32(vfitab,2);
1797
1798             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1799             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1800             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1801             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1802             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1803             _MM_TRANSPOSE4_PS(Y,F,G,H);
1804             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1805             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1806             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq23,FF),_mm_mul_ps(vftabscale,rinv23)));
1807
1808             fscal            = felec;
1809
1810              /* Update vectorial force */
1811             fix2             = _mm_macc_ps(dx23,fscal,fix2);
1812             fiy2             = _mm_macc_ps(dy23,fscal,fiy2);
1813             fiz2             = _mm_macc_ps(dz23,fscal,fiz2);
1814
1815             fjx3             = _mm_macc_ps(dx23,fscal,fjx3);
1816             fjy3             = _mm_macc_ps(dy23,fscal,fjy3);
1817             fjz3             = _mm_macc_ps(dz23,fscal,fjz3);
1818
1819             /**************************
1820              * CALCULATE INTERACTIONS *
1821              **************************/
1822
1823             r31              = _mm_mul_ps(rsq31,rinv31);
1824
1825             /* Calculate table index by multiplying r with table scale and truncate to integer */
1826             rt               = _mm_mul_ps(r31,vftabscale);
1827             vfitab           = _mm_cvttps_epi32(rt);
1828 #ifdef __XOP__
1829             vfeps            = _mm_frcz_ps(rt);
1830 #else
1831             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1832 #endif
1833             twovfeps         = _mm_add_ps(vfeps,vfeps);
1834             vfitab           = _mm_slli_epi32(vfitab,2);
1835
1836             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1837             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1838             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1839             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1840             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1841             _MM_TRANSPOSE4_PS(Y,F,G,H);
1842             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1843             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1844             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq31,FF),_mm_mul_ps(vftabscale,rinv31)));
1845
1846             fscal            = felec;
1847
1848              /* Update vectorial force */
1849             fix3             = _mm_macc_ps(dx31,fscal,fix3);
1850             fiy3             = _mm_macc_ps(dy31,fscal,fiy3);
1851             fiz3             = _mm_macc_ps(dz31,fscal,fiz3);
1852
1853             fjx1             = _mm_macc_ps(dx31,fscal,fjx1);
1854             fjy1             = _mm_macc_ps(dy31,fscal,fjy1);
1855             fjz1             = _mm_macc_ps(dz31,fscal,fjz1);
1856
1857             /**************************
1858              * CALCULATE INTERACTIONS *
1859              **************************/
1860
1861             r32              = _mm_mul_ps(rsq32,rinv32);
1862
1863             /* Calculate table index by multiplying r with table scale and truncate to integer */
1864             rt               = _mm_mul_ps(r32,vftabscale);
1865             vfitab           = _mm_cvttps_epi32(rt);
1866 #ifdef __XOP__
1867             vfeps            = _mm_frcz_ps(rt);
1868 #else
1869             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1870 #endif
1871             twovfeps         = _mm_add_ps(vfeps,vfeps);
1872             vfitab           = _mm_slli_epi32(vfitab,2);
1873
1874             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1875             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1876             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1877             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1878             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1879             _MM_TRANSPOSE4_PS(Y,F,G,H);
1880             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1881             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1882             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq32,FF),_mm_mul_ps(vftabscale,rinv32)));
1883
1884             fscal            = felec;
1885
1886              /* Update vectorial force */
1887             fix3             = _mm_macc_ps(dx32,fscal,fix3);
1888             fiy3             = _mm_macc_ps(dy32,fscal,fiy3);
1889             fiz3             = _mm_macc_ps(dz32,fscal,fiz3);
1890
1891             fjx2             = _mm_macc_ps(dx32,fscal,fjx2);
1892             fjy2             = _mm_macc_ps(dy32,fscal,fjy2);
1893             fjz2             = _mm_macc_ps(dz32,fscal,fjz2);
1894
1895             /**************************
1896              * CALCULATE INTERACTIONS *
1897              **************************/
1898
1899             r33              = _mm_mul_ps(rsq33,rinv33);
1900
1901             /* Calculate table index by multiplying r with table scale and truncate to integer */
1902             rt               = _mm_mul_ps(r33,vftabscale);
1903             vfitab           = _mm_cvttps_epi32(rt);
1904 #ifdef __XOP__
1905             vfeps            = _mm_frcz_ps(rt);
1906 #else
1907             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1908 #endif
1909             twovfeps         = _mm_add_ps(vfeps,vfeps);
1910             vfitab           = _mm_slli_epi32(vfitab,2);
1911
1912             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1913             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1914             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1915             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1916             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1917             _MM_TRANSPOSE4_PS(Y,F,G,H);
1918             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1919             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1920             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq33,FF),_mm_mul_ps(vftabscale,rinv33)));
1921
1922             fscal            = felec;
1923
1924              /* Update vectorial force */
1925             fix3             = _mm_macc_ps(dx33,fscal,fix3);
1926             fiy3             = _mm_macc_ps(dy33,fscal,fiy3);
1927             fiz3             = _mm_macc_ps(dz33,fscal,fiz3);
1928
1929             fjx3             = _mm_macc_ps(dx33,fscal,fjx3);
1930             fjy3             = _mm_macc_ps(dy33,fscal,fjy3);
1931             fjz3             = _mm_macc_ps(dz33,fscal,fjz3);
1932
1933             fjptrA             = f+j_coord_offsetA;
1934             fjptrB             = f+j_coord_offsetB;
1935             fjptrC             = f+j_coord_offsetC;
1936             fjptrD             = f+j_coord_offsetD;
1937
1938             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1939                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1940                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1941
1942             /* Inner loop uses 411 flops */
1943         }
1944
1945         if(jidx<j_index_end)
1946         {
1947
1948             /* Get j neighbor index, and coordinate index */
1949             jnrlistA         = jjnr[jidx];
1950             jnrlistB         = jjnr[jidx+1];
1951             jnrlistC         = jjnr[jidx+2];
1952             jnrlistD         = jjnr[jidx+3];
1953             /* Sign of each element will be negative for non-real atoms.
1954              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1955              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1956              */
1957             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1958             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1959             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1960             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1961             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1962             j_coord_offsetA  = DIM*jnrA;
1963             j_coord_offsetB  = DIM*jnrB;
1964             j_coord_offsetC  = DIM*jnrC;
1965             j_coord_offsetD  = DIM*jnrD;
1966
1967             /* load j atom coordinates */
1968             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1969                                               x+j_coord_offsetC,x+j_coord_offsetD,
1970                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1971                                               &jy2,&jz2,&jx3,&jy3,&jz3);
1972
1973             /* Calculate displacement vector */
1974             dx00             = _mm_sub_ps(ix0,jx0);
1975             dy00             = _mm_sub_ps(iy0,jy0);
1976             dz00             = _mm_sub_ps(iz0,jz0);
1977             dx11             = _mm_sub_ps(ix1,jx1);
1978             dy11             = _mm_sub_ps(iy1,jy1);
1979             dz11             = _mm_sub_ps(iz1,jz1);
1980             dx12             = _mm_sub_ps(ix1,jx2);
1981             dy12             = _mm_sub_ps(iy1,jy2);
1982             dz12             = _mm_sub_ps(iz1,jz2);
1983             dx13             = _mm_sub_ps(ix1,jx3);
1984             dy13             = _mm_sub_ps(iy1,jy3);
1985             dz13             = _mm_sub_ps(iz1,jz3);
1986             dx21             = _mm_sub_ps(ix2,jx1);
1987             dy21             = _mm_sub_ps(iy2,jy1);
1988             dz21             = _mm_sub_ps(iz2,jz1);
1989             dx22             = _mm_sub_ps(ix2,jx2);
1990             dy22             = _mm_sub_ps(iy2,jy2);
1991             dz22             = _mm_sub_ps(iz2,jz2);
1992             dx23             = _mm_sub_ps(ix2,jx3);
1993             dy23             = _mm_sub_ps(iy2,jy3);
1994             dz23             = _mm_sub_ps(iz2,jz3);
1995             dx31             = _mm_sub_ps(ix3,jx1);
1996             dy31             = _mm_sub_ps(iy3,jy1);
1997             dz31             = _mm_sub_ps(iz3,jz1);
1998             dx32             = _mm_sub_ps(ix3,jx2);
1999             dy32             = _mm_sub_ps(iy3,jy2);
2000             dz32             = _mm_sub_ps(iz3,jz2);
2001             dx33             = _mm_sub_ps(ix3,jx3);
2002             dy33             = _mm_sub_ps(iy3,jy3);
2003             dz33             = _mm_sub_ps(iz3,jz3);
2004
2005             /* Calculate squared distance and things based on it */
2006             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
2007             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
2008             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
2009             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
2010             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
2011             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
2012             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
2013             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
2014             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
2015             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
2016
2017             rinv11           = gmx_mm_invsqrt_ps(rsq11);
2018             rinv12           = gmx_mm_invsqrt_ps(rsq12);
2019             rinv13           = gmx_mm_invsqrt_ps(rsq13);
2020             rinv21           = gmx_mm_invsqrt_ps(rsq21);
2021             rinv22           = gmx_mm_invsqrt_ps(rsq22);
2022             rinv23           = gmx_mm_invsqrt_ps(rsq23);
2023             rinv31           = gmx_mm_invsqrt_ps(rsq31);
2024             rinv32           = gmx_mm_invsqrt_ps(rsq32);
2025             rinv33           = gmx_mm_invsqrt_ps(rsq33);
2026
2027             rinvsq00         = gmx_mm_inv_ps(rsq00);
2028
2029             fjx0             = _mm_setzero_ps();
2030             fjy0             = _mm_setzero_ps();
2031             fjz0             = _mm_setzero_ps();
2032             fjx1             = _mm_setzero_ps();
2033             fjy1             = _mm_setzero_ps();
2034             fjz1             = _mm_setzero_ps();
2035             fjx2             = _mm_setzero_ps();
2036             fjy2             = _mm_setzero_ps();
2037             fjz2             = _mm_setzero_ps();
2038             fjx3             = _mm_setzero_ps();
2039             fjy3             = _mm_setzero_ps();
2040             fjz3             = _mm_setzero_ps();
2041
2042             /**************************
2043              * CALCULATE INTERACTIONS *
2044              **************************/
2045
2046             /* LENNARD-JONES DISPERSION/REPULSION */
2047
2048             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
2049             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
2050
2051             fscal            = fvdw;
2052
2053             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2054
2055              /* Update vectorial force */
2056             fix0             = _mm_macc_ps(dx00,fscal,fix0);
2057             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
2058             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
2059
2060             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
2061             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
2062             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
2063
2064             /**************************
2065              * CALCULATE INTERACTIONS *
2066              **************************/
2067
2068             r11              = _mm_mul_ps(rsq11,rinv11);
2069             r11              = _mm_andnot_ps(dummy_mask,r11);
2070
2071             /* Calculate table index by multiplying r with table scale and truncate to integer */
2072             rt               = _mm_mul_ps(r11,vftabscale);
2073             vfitab           = _mm_cvttps_epi32(rt);
2074 #ifdef __XOP__
2075             vfeps            = _mm_frcz_ps(rt);
2076 #else
2077             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
2078 #endif
2079             twovfeps         = _mm_add_ps(vfeps,vfeps);
2080             vfitab           = _mm_slli_epi32(vfitab,2);
2081
2082             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2083             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
2084             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
2085             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
2086             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
2087             _MM_TRANSPOSE4_PS(Y,F,G,H);
2088             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
2089             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
2090             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq11,FF),_mm_mul_ps(vftabscale,rinv11)));
2091
2092             fscal            = felec;
2093
2094             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2095
2096              /* Update vectorial force */
2097             fix1             = _mm_macc_ps(dx11,fscal,fix1);
2098             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
2099             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
2100
2101             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
2102             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
2103             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
2104
2105             /**************************
2106              * CALCULATE INTERACTIONS *
2107              **************************/
2108
2109             r12              = _mm_mul_ps(rsq12,rinv12);
2110             r12              = _mm_andnot_ps(dummy_mask,r12);
2111
2112             /* Calculate table index by multiplying r with table scale and truncate to integer */
2113             rt               = _mm_mul_ps(r12,vftabscale);
2114             vfitab           = _mm_cvttps_epi32(rt);
2115 #ifdef __XOP__
2116             vfeps            = _mm_frcz_ps(rt);
2117 #else
2118             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
2119 #endif
2120             twovfeps         = _mm_add_ps(vfeps,vfeps);
2121             vfitab           = _mm_slli_epi32(vfitab,2);
2122
2123             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2124             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
2125             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
2126             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
2127             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
2128             _MM_TRANSPOSE4_PS(Y,F,G,H);
2129             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
2130             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
2131             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq12,FF),_mm_mul_ps(vftabscale,rinv12)));
2132
2133             fscal            = felec;
2134
2135             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2136
2137              /* Update vectorial force */
2138             fix1             = _mm_macc_ps(dx12,fscal,fix1);
2139             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
2140             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
2141
2142             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
2143             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
2144             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
2145
2146             /**************************
2147              * CALCULATE INTERACTIONS *
2148              **************************/
2149
2150             r13              = _mm_mul_ps(rsq13,rinv13);
2151             r13              = _mm_andnot_ps(dummy_mask,r13);
2152
2153             /* Calculate table index by multiplying r with table scale and truncate to integer */
2154             rt               = _mm_mul_ps(r13,vftabscale);
2155             vfitab           = _mm_cvttps_epi32(rt);
2156 #ifdef __XOP__
2157             vfeps            = _mm_frcz_ps(rt);
2158 #else
2159             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
2160 #endif
2161             twovfeps         = _mm_add_ps(vfeps,vfeps);
2162             vfitab           = _mm_slli_epi32(vfitab,2);
2163
2164             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2165             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
2166             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
2167             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
2168             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
2169             _MM_TRANSPOSE4_PS(Y,F,G,H);
2170             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
2171             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
2172             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq13,FF),_mm_mul_ps(vftabscale,rinv13)));
2173
2174             fscal            = felec;
2175
2176             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2177
2178              /* Update vectorial force */
2179             fix1             = _mm_macc_ps(dx13,fscal,fix1);
2180             fiy1             = _mm_macc_ps(dy13,fscal,fiy1);
2181             fiz1             = _mm_macc_ps(dz13,fscal,fiz1);
2182
2183             fjx3             = _mm_macc_ps(dx13,fscal,fjx3);
2184             fjy3             = _mm_macc_ps(dy13,fscal,fjy3);
2185             fjz3             = _mm_macc_ps(dz13,fscal,fjz3);
2186
2187             /**************************
2188              * CALCULATE INTERACTIONS *
2189              **************************/
2190
2191             r21              = _mm_mul_ps(rsq21,rinv21);
2192             r21              = _mm_andnot_ps(dummy_mask,r21);
2193
2194             /* Calculate table index by multiplying r with table scale and truncate to integer */
2195             rt               = _mm_mul_ps(r21,vftabscale);
2196             vfitab           = _mm_cvttps_epi32(rt);
2197 #ifdef __XOP__
2198             vfeps            = _mm_frcz_ps(rt);
2199 #else
2200             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
2201 #endif
2202             twovfeps         = _mm_add_ps(vfeps,vfeps);
2203             vfitab           = _mm_slli_epi32(vfitab,2);
2204
2205             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2206             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
2207             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
2208             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
2209             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
2210             _MM_TRANSPOSE4_PS(Y,F,G,H);
2211             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
2212             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
2213             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq21,FF),_mm_mul_ps(vftabscale,rinv21)));
2214
2215             fscal            = felec;
2216
2217             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2218
2219              /* Update vectorial force */
2220             fix2             = _mm_macc_ps(dx21,fscal,fix2);
2221             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
2222             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
2223
2224             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
2225             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
2226             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
2227
2228             /**************************
2229              * CALCULATE INTERACTIONS *
2230              **************************/
2231
2232             r22              = _mm_mul_ps(rsq22,rinv22);
2233             r22              = _mm_andnot_ps(dummy_mask,r22);
2234
2235             /* Calculate table index by multiplying r with table scale and truncate to integer */
2236             rt               = _mm_mul_ps(r22,vftabscale);
2237             vfitab           = _mm_cvttps_epi32(rt);
2238 #ifdef __XOP__
2239             vfeps            = _mm_frcz_ps(rt);
2240 #else
2241             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
2242 #endif
2243             twovfeps         = _mm_add_ps(vfeps,vfeps);
2244             vfitab           = _mm_slli_epi32(vfitab,2);
2245
2246             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2247             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
2248             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
2249             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
2250             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
2251             _MM_TRANSPOSE4_PS(Y,F,G,H);
2252             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
2253             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
2254             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq22,FF),_mm_mul_ps(vftabscale,rinv22)));
2255
2256             fscal            = felec;
2257
2258             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2259
2260              /* Update vectorial force */
2261             fix2             = _mm_macc_ps(dx22,fscal,fix2);
2262             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
2263             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
2264
2265             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
2266             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
2267             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
2268
2269             /**************************
2270              * CALCULATE INTERACTIONS *
2271              **************************/
2272
2273             r23              = _mm_mul_ps(rsq23,rinv23);
2274             r23              = _mm_andnot_ps(dummy_mask,r23);
2275
2276             /* Calculate table index by multiplying r with table scale and truncate to integer */
2277             rt               = _mm_mul_ps(r23,vftabscale);
2278             vfitab           = _mm_cvttps_epi32(rt);
2279 #ifdef __XOP__
2280             vfeps            = _mm_frcz_ps(rt);
2281 #else
2282             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
2283 #endif
2284             twovfeps         = _mm_add_ps(vfeps,vfeps);
2285             vfitab           = _mm_slli_epi32(vfitab,2);
2286
2287             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2288             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
2289             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
2290             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
2291             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
2292             _MM_TRANSPOSE4_PS(Y,F,G,H);
2293             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
2294             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
2295             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq23,FF),_mm_mul_ps(vftabscale,rinv23)));
2296
2297             fscal            = felec;
2298
2299             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2300
2301              /* Update vectorial force */
2302             fix2             = _mm_macc_ps(dx23,fscal,fix2);
2303             fiy2             = _mm_macc_ps(dy23,fscal,fiy2);
2304             fiz2             = _mm_macc_ps(dz23,fscal,fiz2);
2305
2306             fjx3             = _mm_macc_ps(dx23,fscal,fjx3);
2307             fjy3             = _mm_macc_ps(dy23,fscal,fjy3);
2308             fjz3             = _mm_macc_ps(dz23,fscal,fjz3);
2309
2310             /**************************
2311              * CALCULATE INTERACTIONS *
2312              **************************/
2313
2314             r31              = _mm_mul_ps(rsq31,rinv31);
2315             r31              = _mm_andnot_ps(dummy_mask,r31);
2316
2317             /* Calculate table index by multiplying r with table scale and truncate to integer */
2318             rt               = _mm_mul_ps(r31,vftabscale);
2319             vfitab           = _mm_cvttps_epi32(rt);
2320 #ifdef __XOP__
2321             vfeps            = _mm_frcz_ps(rt);
2322 #else
2323             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
2324 #endif
2325             twovfeps         = _mm_add_ps(vfeps,vfeps);
2326             vfitab           = _mm_slli_epi32(vfitab,2);
2327
2328             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2329             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
2330             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
2331             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
2332             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
2333             _MM_TRANSPOSE4_PS(Y,F,G,H);
2334             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
2335             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
2336             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq31,FF),_mm_mul_ps(vftabscale,rinv31)));
2337
2338             fscal            = felec;
2339
2340             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2341
2342              /* Update vectorial force */
2343             fix3             = _mm_macc_ps(dx31,fscal,fix3);
2344             fiy3             = _mm_macc_ps(dy31,fscal,fiy3);
2345             fiz3             = _mm_macc_ps(dz31,fscal,fiz3);
2346
2347             fjx1             = _mm_macc_ps(dx31,fscal,fjx1);
2348             fjy1             = _mm_macc_ps(dy31,fscal,fjy1);
2349             fjz1             = _mm_macc_ps(dz31,fscal,fjz1);
2350
2351             /**************************
2352              * CALCULATE INTERACTIONS *
2353              **************************/
2354
2355             r32              = _mm_mul_ps(rsq32,rinv32);
2356             r32              = _mm_andnot_ps(dummy_mask,r32);
2357
2358             /* Calculate table index by multiplying r with table scale and truncate to integer */
2359             rt               = _mm_mul_ps(r32,vftabscale);
2360             vfitab           = _mm_cvttps_epi32(rt);
2361 #ifdef __XOP__
2362             vfeps            = _mm_frcz_ps(rt);
2363 #else
2364             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
2365 #endif
2366             twovfeps         = _mm_add_ps(vfeps,vfeps);
2367             vfitab           = _mm_slli_epi32(vfitab,2);
2368
2369             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2370             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
2371             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
2372             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
2373             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
2374             _MM_TRANSPOSE4_PS(Y,F,G,H);
2375             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
2376             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
2377             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq32,FF),_mm_mul_ps(vftabscale,rinv32)));
2378
2379             fscal            = felec;
2380
2381             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2382
2383              /* Update vectorial force */
2384             fix3             = _mm_macc_ps(dx32,fscal,fix3);
2385             fiy3             = _mm_macc_ps(dy32,fscal,fiy3);
2386             fiz3             = _mm_macc_ps(dz32,fscal,fiz3);
2387
2388             fjx2             = _mm_macc_ps(dx32,fscal,fjx2);
2389             fjy2             = _mm_macc_ps(dy32,fscal,fjy2);
2390             fjz2             = _mm_macc_ps(dz32,fscal,fjz2);
2391
2392             /**************************
2393              * CALCULATE INTERACTIONS *
2394              **************************/
2395
2396             r33              = _mm_mul_ps(rsq33,rinv33);
2397             r33              = _mm_andnot_ps(dummy_mask,r33);
2398
2399             /* Calculate table index by multiplying r with table scale and truncate to integer */
2400             rt               = _mm_mul_ps(r33,vftabscale);
2401             vfitab           = _mm_cvttps_epi32(rt);
2402 #ifdef __XOP__
2403             vfeps            = _mm_frcz_ps(rt);
2404 #else
2405             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
2406 #endif
2407             twovfeps         = _mm_add_ps(vfeps,vfeps);
2408             vfitab           = _mm_slli_epi32(vfitab,2);
2409
2410             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2411             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
2412             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
2413             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
2414             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
2415             _MM_TRANSPOSE4_PS(Y,F,G,H);
2416             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
2417             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
2418             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq33,FF),_mm_mul_ps(vftabscale,rinv33)));
2419
2420             fscal            = felec;
2421
2422             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2423
2424              /* Update vectorial force */
2425             fix3             = _mm_macc_ps(dx33,fscal,fix3);
2426             fiy3             = _mm_macc_ps(dy33,fscal,fiy3);
2427             fiz3             = _mm_macc_ps(dz33,fscal,fiz3);
2428
2429             fjx3             = _mm_macc_ps(dx33,fscal,fjx3);
2430             fjy3             = _mm_macc_ps(dy33,fscal,fjy3);
2431             fjz3             = _mm_macc_ps(dz33,fscal,fjz3);
2432
2433             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2434             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2435             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2436             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2437
2438             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
2439                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2440                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2441
2442             /* Inner loop uses 420 flops */
2443         }
2444
2445         /* End of innermost loop */
2446
2447         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2448                                               f+i_coord_offset,fshift+i_shift_offset);
2449
2450         /* Increment number of inner iterations */
2451         inneriter                  += j_index_end - j_index_start;
2452
2453         /* Outer loop uses 24 flops */
2454     }
2455
2456     /* Increment number of outer iterations */
2457     outeriter        += nri;
2458
2459     /* Update outer/inner flops */
2460
2461     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*420);
2462 }