src/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecCSTab_VdwLJ_GeomW4W4_avx_128_fma_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_128_fma_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_128_fma_single.h"
  34 #include "kernelutil_x86_avx_128_fma_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW4W4_VF_avx_128_fma_single
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water4-Water4
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_VdwLJ_GeomW4W4_VF_avx_128_fma_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwioffset1;
  72     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  73     int              vdwioffset2;
  74     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  75     int              vdwioffset3;
  76     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  77     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  78     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  79     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  80     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  81     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  82     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  83     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
  84     __m128           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
  85     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  86     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  87     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
  88     __m128           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
  89     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
  90     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
  91     __m128           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
  92     __m128           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
  93     __m128           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
  94     __m128           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
  95     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  96     real             *charge;
  97     int              nvdwtype;
  98     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  99     int              *vdwtype;
 100     real             *vdwparam;
 101     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 102     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 103     __m128i          vfitab;
 104     __m128i          ifour       = _mm_set1_epi32(4);
 105     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 106     real             *vftab;
 107     __m128           dummy_mask,cutoff_mask;
 108     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 109     __m128           one     = _mm_set1_ps(1.0);
 110     __m128           two     = _mm_set1_ps(2.0);
 111     x                = xx[0];
 112     f                = ff[0];
 113
 114     nri              = nlist->nri;
 115     iinr             = nlist->iinr;
 116     jindex           = nlist->jindex;
 117     jjnr             = nlist->jjnr;
 118     shiftidx         = nlist->shift;
 119     gid              = nlist->gid;
 120     shiftvec         = fr->shift_vec[0];
 121     fshift           = fr->fshift[0];
 122     facel            = _mm_set1_ps(fr->epsfac);
 123     charge           = mdatoms->chargeA;
 124     nvdwtype         = fr->ntype;
 125     vdwparam         = fr->nbfp;
 126     vdwtype          = mdatoms->typeA;
 127
 128     vftab            = kernel_data->table_elec->data;
 129     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 130
 131     /* Setup water-specific parameters */
 132     inr              = nlist->iinr[0];
 133     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 134     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 135     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 136     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 137
 138     jq1              = _mm_set1_ps(charge[inr+1]);
 139     jq2              = _mm_set1_ps(charge[inr+2]);
 140     jq3              = _mm_set1_ps(charge[inr+3]);
 141     vdwjidx0A        = 2*vdwtype[inr+0];
 142     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
 143     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
 144     qq11             = _mm_mul_ps(iq1,jq1);
 145     qq12             = _mm_mul_ps(iq1,jq2);
 146     qq13             = _mm_mul_ps(iq1,jq3);
 147     qq21             = _mm_mul_ps(iq2,jq1);
 148     qq22             = _mm_mul_ps(iq2,jq2);
 149     qq23             = _mm_mul_ps(iq2,jq3);
 150     qq31             = _mm_mul_ps(iq3,jq1);
 151     qq32             = _mm_mul_ps(iq3,jq2);
 152     qq33             = _mm_mul_ps(iq3,jq3);
 153
 154     /* Avoid stupid compiler warnings */
 155     jnrA = jnrB = jnrC = jnrD = 0;
 156     j_coord_offsetA = 0;
 157     j_coord_offsetB = 0;
 158     j_coord_offsetC = 0;
 159     j_coord_offsetD = 0;
 160
 161     outeriter        = 0;
 162     inneriter        = 0;
 163
 164     for(iidx=0;iidx<4*DIM;iidx++)
 165     {
 166         scratch[iidx] = 0.0;
 167     }
 168
 169     /* Start outer loop over neighborlists */
 170     for(iidx=0; iidx<nri; iidx++)
 171     {
 172         /* Load shift vector for this list */
 173         i_shift_offset   = DIM*shiftidx[iidx];
 174
 175         /* Load limits for loop over neighbors */
 176         j_index_start    = jindex[iidx];
 177         j_index_end      = jindex[iidx+1];
 178
 179         /* Get outer coordinate index */
 180         inr              = iinr[iidx];
 181         i_coord_offset   = DIM*inr;
 182
 183         /* Load i particle coords and add shift vector */
 184         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 185                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 186
 187         fix0             = _mm_setzero_ps();
 188         fiy0             = _mm_setzero_ps();
 189         fiz0             = _mm_setzero_ps();
 190         fix1             = _mm_setzero_ps();
 191         fiy1             = _mm_setzero_ps();
 192         fiz1             = _mm_setzero_ps();
 193         fix2             = _mm_setzero_ps();
 194         fiy2             = _mm_setzero_ps();
 195         fiz2             = _mm_setzero_ps();
 196         fix3             = _mm_setzero_ps();
 197         fiy3             = _mm_setzero_ps();
 198         fiz3             = _mm_setzero_ps();
 199
 200         /* Reset potential sums */
 201         velecsum         = _mm_setzero_ps();
 202         vvdwsum          = _mm_setzero_ps();
 203
 204         /* Start inner kernel loop */
 205         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 206         {
 207
 208             /* Get j neighbor index, and coordinate index */
 209             jnrA             = jjnr[jidx];
 210             jnrB             = jjnr[jidx+1];
 211             jnrC             = jjnr[jidx+2];
 212             jnrD             = jjnr[jidx+3];
 213             j_coord_offsetA  = DIM*jnrA;
 214             j_coord_offsetB  = DIM*jnrB;
 215             j_coord_offsetC  = DIM*jnrC;
 216             j_coord_offsetD  = DIM*jnrD;
 217
 218             /* load j atom coordinates */
 219             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 220                                               x+j_coord_offsetC,x+j_coord_offsetD,
 221                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 222                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 223
 224             /* Calculate displacement vector */
 225             dx00             = _mm_sub_ps(ix0,jx0);
 226             dy00             = _mm_sub_ps(iy0,jy0);
 227             dz00             = _mm_sub_ps(iz0,jz0);
 228             dx11             = _mm_sub_ps(ix1,jx1);
 229             dy11             = _mm_sub_ps(iy1,jy1);
 230             dz11             = _mm_sub_ps(iz1,jz1);
 231             dx12             = _mm_sub_ps(ix1,jx2);
 232             dy12             = _mm_sub_ps(iy1,jy2);
 233             dz12             = _mm_sub_ps(iz1,jz2);
 234             dx13             = _mm_sub_ps(ix1,jx3);
 235             dy13             = _mm_sub_ps(iy1,jy3);
 236             dz13             = _mm_sub_ps(iz1,jz3);
 237             dx21             = _mm_sub_ps(ix2,jx1);
 238             dy21             = _mm_sub_ps(iy2,jy1);
 239             dz21             = _mm_sub_ps(iz2,jz1);
 240             dx22             = _mm_sub_ps(ix2,jx2);
 241             dy22             = _mm_sub_ps(iy2,jy2);
 242             dz22             = _mm_sub_ps(iz2,jz2);
 243             dx23             = _mm_sub_ps(ix2,jx3);
 244             dy23             = _mm_sub_ps(iy2,jy3);
 245             dz23             = _mm_sub_ps(iz2,jz3);
 246             dx31             = _mm_sub_ps(ix3,jx1);
 247             dy31             = _mm_sub_ps(iy3,jy1);
 248             dz31             = _mm_sub_ps(iz3,jz1);
 249             dx32             = _mm_sub_ps(ix3,jx2);
 250             dy32             = _mm_sub_ps(iy3,jy2);
 251             dz32             = _mm_sub_ps(iz3,jz2);
 252             dx33             = _mm_sub_ps(ix3,jx3);
 253             dy33             = _mm_sub_ps(iy3,jy3);
 254             dz33             = _mm_sub_ps(iz3,jz3);
 255
 256             /* Calculate squared distance and things based on it */
 257             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 258             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 259             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 260             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
 261             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 262             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 263             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
 264             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
 265             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
 266             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
 267
 268             rinv11           = gmx_mm_invsqrt_ps(rsq11);
 269             rinv12           = gmx_mm_invsqrt_ps(rsq12);
 270             rinv13           = gmx_mm_invsqrt_ps(rsq13);
 271             rinv21           = gmx_mm_invsqrt_ps(rsq21);
 272             rinv22           = gmx_mm_invsqrt_ps(rsq22);
 273             rinv23           = gmx_mm_invsqrt_ps(rsq23);
 274             rinv31           = gmx_mm_invsqrt_ps(rsq31);
 275             rinv32           = gmx_mm_invsqrt_ps(rsq32);
 276             rinv33           = gmx_mm_invsqrt_ps(rsq33);
 277
 278             rinvsq00         = gmx_mm_inv_ps(rsq00);
 279
 280             fjx0             = _mm_setzero_ps();
 281             fjy0             = _mm_setzero_ps();
 282             fjz0             = _mm_setzero_ps();
 283             fjx1             = _mm_setzero_ps();
 284             fjy1             = _mm_setzero_ps();
 285             fjz1             = _mm_setzero_ps();
 286             fjx2             = _mm_setzero_ps();
 287             fjy2             = _mm_setzero_ps();
 288             fjz2             = _mm_setzero_ps();
 289             fjx3             = _mm_setzero_ps();
 290             fjy3             = _mm_setzero_ps();
 291             fjz3             = _mm_setzero_ps();
 292
 293             /**************************
 294              * CALCULATE INTERACTIONS *
 295              **************************/
 296
 297             /* LENNARD-JONES DISPERSION/REPULSION */
 298
 299             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 300             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 301             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 302             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 303             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 304
 305             /* Update potential sum for this i atom from the interaction with this j atom. */
 306             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 307
 308             fscal            = fvdw;
 309
 310              /* Update vectorial force */
 311             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 312             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 313             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 314
 315             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 316             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 317             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 318
 319             /**************************
 320              * CALCULATE INTERACTIONS *
 321              **************************/
 322
 323             r11              = _mm_mul_ps(rsq11,rinv11);
 324
 325             /* Calculate table index by multiplying r with table scale and truncate to integer */
 326             rt               = _mm_mul_ps(r11,vftabscale);
 327             vfitab           = _mm_cvttps_epi32(rt);
 328 #ifdef __XOP__
 329             vfeps            = _mm_frcz_ps(rt);
 330 #else
 331             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 332 #endif
 333             twovfeps         = _mm_add_ps(vfeps,vfeps);
 334             vfitab           = _mm_slli_epi32(vfitab,2);
 335
 336             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 337             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 338             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 339             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 340             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 341             _MM_TRANSPOSE4_PS(Y,F,G,H);
 342             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 343             VV               = _mm_macc_ps(vfeps,Fp,Y);
 344             velec            = _mm_mul_ps(qq11,VV);
 345             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 346             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq11,FF),_mm_mul_ps(vftabscale,rinv11)));
 347
 348             /* Update potential sum for this i atom from the interaction with this j atom. */
 349             velecsum         = _mm_add_ps(velecsum,velec);
 350
 351             fscal            = felec;
 352
 353              /* Update vectorial force */
 354             fix1             = _mm_macc_ps(dx11,fscal,fix1);
 355             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
 356             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
 357
 358             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
 359             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
 360             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
 361
 362             /**************************
 363              * CALCULATE INTERACTIONS *
 364              **************************/
 365
 366             r12              = _mm_mul_ps(rsq12,rinv12);
 367
 368             /* Calculate table index by multiplying r with table scale and truncate to integer */
 369             rt               = _mm_mul_ps(r12,vftabscale);
 370             vfitab           = _mm_cvttps_epi32(rt);
 371 #ifdef __XOP__
 372             vfeps            = _mm_frcz_ps(rt);
 373 #else
 374             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 375 #endif
 376             twovfeps         = _mm_add_ps(vfeps,vfeps);
 377             vfitab           = _mm_slli_epi32(vfitab,2);
 378
 379             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 380             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 381             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 382             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 383             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 384             _MM_TRANSPOSE4_PS(Y,F,G,H);
 385             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 386             VV               = _mm_macc_ps(vfeps,Fp,Y);
 387             velec            = _mm_mul_ps(qq12,VV);
 388             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 389             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq12,FF),_mm_mul_ps(vftabscale,rinv12)));
 390
 391             /* Update potential sum for this i atom from the interaction with this j atom. */
 392             velecsum         = _mm_add_ps(velecsum,velec);
 393
 394             fscal            = felec;
 395
 396              /* Update vectorial force */
 397             fix1             = _mm_macc_ps(dx12,fscal,fix1);
 398             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
 399             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
 400
 401             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
 402             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
 403             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
 404
 405             /**************************
 406              * CALCULATE INTERACTIONS *
 407              **************************/
 408
 409             r13              = _mm_mul_ps(rsq13,rinv13);
 410
 411             /* Calculate table index by multiplying r with table scale and truncate to integer */
 412             rt               = _mm_mul_ps(r13,vftabscale);
 413             vfitab           = _mm_cvttps_epi32(rt);
 414 #ifdef __XOP__
 415             vfeps            = _mm_frcz_ps(rt);
 416 #else
 417             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 418 #endif
 419             twovfeps         = _mm_add_ps(vfeps,vfeps);
 420             vfitab           = _mm_slli_epi32(vfitab,2);
 421
 422             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 423             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 424             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 425             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 426             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 427             _MM_TRANSPOSE4_PS(Y,F,G,H);
 428             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 429             VV               = _mm_macc_ps(vfeps,Fp,Y);
 430             velec            = _mm_mul_ps(qq13,VV);
 431             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 432             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq13,FF),_mm_mul_ps(vftabscale,rinv13)));
 433
 434             /* Update potential sum for this i atom from the interaction with this j atom. */
 435             velecsum         = _mm_add_ps(velecsum,velec);
 436
 437             fscal            = felec;
 438
 439              /* Update vectorial force */
 440             fix1             = _mm_macc_ps(dx13,fscal,fix1);
 441             fiy1             = _mm_macc_ps(dy13,fscal,fiy1);
 442             fiz1             = _mm_macc_ps(dz13,fscal,fiz1);
 443
 444             fjx3             = _mm_macc_ps(dx13,fscal,fjx3);
 445             fjy3             = _mm_macc_ps(dy13,fscal,fjy3);
 446             fjz3             = _mm_macc_ps(dz13,fscal,fjz3);
 447
 448             /**************************
 449              * CALCULATE INTERACTIONS *
 450              **************************/
 451
 452             r21              = _mm_mul_ps(rsq21,rinv21);
 453
 454             /* Calculate table index by multiplying r with table scale and truncate to integer */
 455             rt               = _mm_mul_ps(r21,vftabscale);
 456             vfitab           = _mm_cvttps_epi32(rt);
 457 #ifdef __XOP__
 458             vfeps            = _mm_frcz_ps(rt);
 459 #else
 460             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 461 #endif
 462             twovfeps         = _mm_add_ps(vfeps,vfeps);
 463             vfitab           = _mm_slli_epi32(vfitab,2);
 464
 465             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 466             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 467             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 468             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 469             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 470             _MM_TRANSPOSE4_PS(Y,F,G,H);
 471             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 472             VV               = _mm_macc_ps(vfeps,Fp,Y);
 473             velec            = _mm_mul_ps(qq21,VV);
 474             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 475             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq21,FF),_mm_mul_ps(vftabscale,rinv21)));
 476
 477             /* Update potential sum for this i atom from the interaction with this j atom. */
 478             velecsum         = _mm_add_ps(velecsum,velec);
 479
 480             fscal            = felec;
 481
 482              /* Update vectorial force */
 483             fix2             = _mm_macc_ps(dx21,fscal,fix2);
 484             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
 485             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
 486
 487             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
 488             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
 489             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
 490
 491             /**************************
 492              * CALCULATE INTERACTIONS *
 493              **************************/
 494
 495             r22              = _mm_mul_ps(rsq22,rinv22);
 496
 497             /* Calculate table index by multiplying r with table scale and truncate to integer */
 498             rt               = _mm_mul_ps(r22,vftabscale);
 499             vfitab           = _mm_cvttps_epi32(rt);
 500 #ifdef __XOP__
 501             vfeps            = _mm_frcz_ps(rt);
 502 #else
 503             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 504 #endif
 505             twovfeps         = _mm_add_ps(vfeps,vfeps);
 506             vfitab           = _mm_slli_epi32(vfitab,2);
 507
 508             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 509             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 510             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 511             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 512             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 513             _MM_TRANSPOSE4_PS(Y,F,G,H);
 514             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 515             VV               = _mm_macc_ps(vfeps,Fp,Y);
 516             velec            = _mm_mul_ps(qq22,VV);
 517             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 518             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq22,FF),_mm_mul_ps(vftabscale,rinv22)));
 519
 520             /* Update potential sum for this i atom from the interaction with this j atom. */
 521             velecsum         = _mm_add_ps(velecsum,velec);
 522
 523             fscal            = felec;
 524
 525              /* Update vectorial force */
 526             fix2             = _mm_macc_ps(dx22,fscal,fix2);
 527             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
 528             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
 529
 530             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
 531             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
 532             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
 533
 534             /**************************
 535              * CALCULATE INTERACTIONS *
 536              **************************/
 537
 538             r23              = _mm_mul_ps(rsq23,rinv23);
 539
 540             /* Calculate table index by multiplying r with table scale and truncate to integer */
 541             rt               = _mm_mul_ps(r23,vftabscale);
 542             vfitab           = _mm_cvttps_epi32(rt);
 543 #ifdef __XOP__
 544             vfeps            = _mm_frcz_ps(rt);
 545 #else
 546             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 547 #endif
 548             twovfeps         = _mm_add_ps(vfeps,vfeps);
 549             vfitab           = _mm_slli_epi32(vfitab,2);
 550
 551             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 552             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 553             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 554             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 555             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 556             _MM_TRANSPOSE4_PS(Y,F,G,H);
 557             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 558             VV               = _mm_macc_ps(vfeps,Fp,Y);
 559             velec            = _mm_mul_ps(qq23,VV);
 560             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 561             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq23,FF),_mm_mul_ps(vftabscale,rinv23)));
 562
 563             /* Update potential sum for this i atom from the interaction with this j atom. */
 564             velecsum         = _mm_add_ps(velecsum,velec);
 565
 566             fscal            = felec;
 567
 568              /* Update vectorial force */
 569             fix2             = _mm_macc_ps(dx23,fscal,fix2);
 570             fiy2             = _mm_macc_ps(dy23,fscal,fiy2);
 571             fiz2             = _mm_macc_ps(dz23,fscal,fiz2);
 572
 573             fjx3             = _mm_macc_ps(dx23,fscal,fjx3);
 574             fjy3             = _mm_macc_ps(dy23,fscal,fjy3);
 575             fjz3             = _mm_macc_ps(dz23,fscal,fjz3);
 576
 577             /**************************
 578              * CALCULATE INTERACTIONS *
 579              **************************/
 580
 581             r31              = _mm_mul_ps(rsq31,rinv31);
 582
 583             /* Calculate table index by multiplying r with table scale and truncate to integer */
 584             rt               = _mm_mul_ps(r31,vftabscale);
 585             vfitab           = _mm_cvttps_epi32(rt);
 586 #ifdef __XOP__
 587             vfeps            = _mm_frcz_ps(rt);
 588 #else
 589             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 590 #endif
 591             twovfeps         = _mm_add_ps(vfeps,vfeps);
 592             vfitab           = _mm_slli_epi32(vfitab,2);
 593
 594             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 595             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 596             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 597             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 598             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 599             _MM_TRANSPOSE4_PS(Y,F,G,H);
 600             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 601             VV               = _mm_macc_ps(vfeps,Fp,Y);
 602             velec            = _mm_mul_ps(qq31,VV);
 603             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 604             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq31,FF),_mm_mul_ps(vftabscale,rinv31)));
 605
 606             /* Update potential sum for this i atom from the interaction with this j atom. */
 607             velecsum         = _mm_add_ps(velecsum,velec);
 608
 609             fscal            = felec;
 610
 611              /* Update vectorial force */
 612             fix3             = _mm_macc_ps(dx31,fscal,fix3);
 613             fiy3             = _mm_macc_ps(dy31,fscal,fiy3);
 614             fiz3             = _mm_macc_ps(dz31,fscal,fiz3);
 615
 616             fjx1             = _mm_macc_ps(dx31,fscal,fjx1);
 617             fjy1             = _mm_macc_ps(dy31,fscal,fjy1);
 618             fjz1             = _mm_macc_ps(dz31,fscal,fjz1);
 619
 620             /**************************
 621              * CALCULATE INTERACTIONS *
 622              **************************/
 623
 624             r32              = _mm_mul_ps(rsq32,rinv32);
 625
 626             /* Calculate table index by multiplying r with table scale and truncate to integer */
 627             rt               = _mm_mul_ps(r32,vftabscale);
 628             vfitab           = _mm_cvttps_epi32(rt);
 629 #ifdef __XOP__
 630             vfeps            = _mm_frcz_ps(rt);
 631 #else
 632             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 633 #endif
 634             twovfeps         = _mm_add_ps(vfeps,vfeps);
 635             vfitab           = _mm_slli_epi32(vfitab,2);
 636
 637             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 638             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 639             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 640             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 641             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 642             _MM_TRANSPOSE4_PS(Y,F,G,H);
 643             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 644             VV               = _mm_macc_ps(vfeps,Fp,Y);
 645             velec            = _mm_mul_ps(qq32,VV);
 646             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 647             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq32,FF),_mm_mul_ps(vftabscale,rinv32)));
 648
 649             /* Update potential sum for this i atom from the interaction with this j atom. */
 650             velecsum         = _mm_add_ps(velecsum,velec);
 651
 652             fscal            = felec;
 653
 654              /* Update vectorial force */
 655             fix3             = _mm_macc_ps(dx32,fscal,fix3);
 656             fiy3             = _mm_macc_ps(dy32,fscal,fiy3);
 657             fiz3             = _mm_macc_ps(dz32,fscal,fiz3);
 658
 659             fjx2             = _mm_macc_ps(dx32,fscal,fjx2);
 660             fjy2             = _mm_macc_ps(dy32,fscal,fjy2);
 661             fjz2             = _mm_macc_ps(dz32,fscal,fjz2);
 662
 663             /**************************
 664              * CALCULATE INTERACTIONS *
 665              **************************/
 666
 667             r33              = _mm_mul_ps(rsq33,rinv33);
 668
 669             /* Calculate table index by multiplying r with table scale and truncate to integer */
 670             rt               = _mm_mul_ps(r33,vftabscale);
 671             vfitab           = _mm_cvttps_epi32(rt);
 672 #ifdef __XOP__
 673             vfeps            = _mm_frcz_ps(rt);
 674 #else
 675             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 676 #endif
 677             twovfeps         = _mm_add_ps(vfeps,vfeps);
 678             vfitab           = _mm_slli_epi32(vfitab,2);
 679
 680             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 681             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 682             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 683             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 684             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 685             _MM_TRANSPOSE4_PS(Y,F,G,H);
 686             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 687             VV               = _mm_macc_ps(vfeps,Fp,Y);
 688             velec            = _mm_mul_ps(qq33,VV);
 689             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 690             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq33,FF),_mm_mul_ps(vftabscale,rinv33)));
 691
 692             /* Update potential sum for this i atom from the interaction with this j atom. */
 693             velecsum         = _mm_add_ps(velecsum,velec);
 694
 695             fscal            = felec;
 696
 697              /* Update vectorial force */
 698             fix3             = _mm_macc_ps(dx33,fscal,fix3);
 699             fiy3             = _mm_macc_ps(dy33,fscal,fiy3);
 700             fiz3             = _mm_macc_ps(dz33,fscal,fiz3);
 701
 702             fjx3             = _mm_macc_ps(dx33,fscal,fjx3);
 703             fjy3             = _mm_macc_ps(dy33,fscal,fjy3);
 704             fjz3             = _mm_macc_ps(dz33,fscal,fjz3);
 705
 706             fjptrA             = f+j_coord_offsetA;
 707             fjptrB             = f+j_coord_offsetB;
 708             fjptrC             = f+j_coord_offsetC;
 709             fjptrD             = f+j_coord_offsetD;
 710
 711             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 712                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
 713                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 714
 715             /* Inner loop uses 452 flops */
 716         }
 717
 718         if(jidx<j_index_end)
 719         {
 720
 721             /* Get j neighbor index, and coordinate index */
 722             jnrlistA         = jjnr[jidx];
 723             jnrlistB         = jjnr[jidx+1];
 724             jnrlistC         = jjnr[jidx+2];
 725             jnrlistD         = jjnr[jidx+3];
 726             /* Sign of each element will be negative for non-real atoms.
 727              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 728              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 729              */
 730             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 731             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 732             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 733             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 734             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 735             j_coord_offsetA  = DIM*jnrA;
 736             j_coord_offsetB  = DIM*jnrB;
 737             j_coord_offsetC  = DIM*jnrC;
 738             j_coord_offsetD  = DIM*jnrD;
 739
 740             /* load j atom coordinates */
 741             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 742                                               x+j_coord_offsetC,x+j_coord_offsetD,
 743                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 744                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 745
 746             /* Calculate displacement vector */
 747             dx00             = _mm_sub_ps(ix0,jx0);
 748             dy00             = _mm_sub_ps(iy0,jy0);
 749             dz00             = _mm_sub_ps(iz0,jz0);
 750             dx11             = _mm_sub_ps(ix1,jx1);
 751             dy11             = _mm_sub_ps(iy1,jy1);
 752             dz11             = _mm_sub_ps(iz1,jz1);
 753             dx12             = _mm_sub_ps(ix1,jx2);
 754             dy12             = _mm_sub_ps(iy1,jy2);
 755             dz12             = _mm_sub_ps(iz1,jz2);
 756             dx13             = _mm_sub_ps(ix1,jx3);
 757             dy13             = _mm_sub_ps(iy1,jy3);
 758             dz13             = _mm_sub_ps(iz1,jz3);
 759             dx21             = _mm_sub_ps(ix2,jx1);
 760             dy21             = _mm_sub_ps(iy2,jy1);
 761             dz21             = _mm_sub_ps(iz2,jz1);
 762             dx22             = _mm_sub_ps(ix2,jx2);
 763             dy22             = _mm_sub_ps(iy2,jy2);
 764             dz22             = _mm_sub_ps(iz2,jz2);
 765             dx23             = _mm_sub_ps(ix2,jx3);
 766             dy23             = _mm_sub_ps(iy2,jy3);
 767             dz23             = _mm_sub_ps(iz2,jz3);
 768             dx31             = _mm_sub_ps(ix3,jx1);
 769             dy31             = _mm_sub_ps(iy3,jy1);
 770             dz31             = _mm_sub_ps(iz3,jz1);
 771             dx32             = _mm_sub_ps(ix3,jx2);
 772             dy32             = _mm_sub_ps(iy3,jy2);
 773             dz32             = _mm_sub_ps(iz3,jz2);
 774             dx33             = _mm_sub_ps(ix3,jx3);
 775             dy33             = _mm_sub_ps(iy3,jy3);
 776             dz33             = _mm_sub_ps(iz3,jz3);
 777
 778             /* Calculate squared distance and things based on it */
 779             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 780             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 781             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 782             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
 783             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 784             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 785             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
 786             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
 787             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
 788             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
 789
 790             rinv11           = gmx_mm_invsqrt_ps(rsq11);
 791             rinv12           = gmx_mm_invsqrt_ps(rsq12);
 792             rinv13           = gmx_mm_invsqrt_ps(rsq13);
 793             rinv21           = gmx_mm_invsqrt_ps(rsq21);
 794             rinv22           = gmx_mm_invsqrt_ps(rsq22);
 795             rinv23           = gmx_mm_invsqrt_ps(rsq23);
 796             rinv31           = gmx_mm_invsqrt_ps(rsq31);
 797             rinv32           = gmx_mm_invsqrt_ps(rsq32);
 798             rinv33           = gmx_mm_invsqrt_ps(rsq33);
 799
 800             rinvsq00         = gmx_mm_inv_ps(rsq00);
 801
 802             fjx0             = _mm_setzero_ps();
 803             fjy0             = _mm_setzero_ps();
 804             fjz0             = _mm_setzero_ps();
 805             fjx1             = _mm_setzero_ps();
 806             fjy1             = _mm_setzero_ps();
 807             fjz1             = _mm_setzero_ps();
 808             fjx2             = _mm_setzero_ps();
 809             fjy2             = _mm_setzero_ps();
 810             fjz2             = _mm_setzero_ps();
 811             fjx3             = _mm_setzero_ps();
 812             fjy3             = _mm_setzero_ps();
 813             fjz3             = _mm_setzero_ps();
 814
 815             /**************************
 816              * CALCULATE INTERACTIONS *
 817              **************************/
 818
 819             /* LENNARD-JONES DISPERSION/REPULSION */
 820
 821             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 822             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 823             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 824             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 825             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 826
 827             /* Update potential sum for this i atom from the interaction with this j atom. */
 828             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 829             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 830
 831             fscal            = fvdw;
 832
 833             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 834
 835              /* Update vectorial force */
 836             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 837             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 838             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 839
 840             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 841             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 842             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 843
 844             /**************************
 845              * CALCULATE INTERACTIONS *
 846              **************************/
 847
 848             r11              = _mm_mul_ps(rsq11,rinv11);
 849             r11              = _mm_andnot_ps(dummy_mask,r11);
 850
 851             /* Calculate table index by multiplying r with table scale and truncate to integer */
 852             rt               = _mm_mul_ps(r11,vftabscale);
 853             vfitab           = _mm_cvttps_epi32(rt);
 854 #ifdef __XOP__
 855             vfeps            = _mm_frcz_ps(rt);
 856 #else
 857             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 858 #endif
 859             twovfeps         = _mm_add_ps(vfeps,vfeps);
 860             vfitab           = _mm_slli_epi32(vfitab,2);
 861
 862             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 863             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 864             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 865             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 866             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 867             _MM_TRANSPOSE4_PS(Y,F,G,H);
 868             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 869             VV               = _mm_macc_ps(vfeps,Fp,Y);
 870             velec            = _mm_mul_ps(qq11,VV);
 871             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 872             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq11,FF),_mm_mul_ps(vftabscale,rinv11)));
 873
 874             /* Update potential sum for this i atom from the interaction with this j atom. */
 875             velec            = _mm_andnot_ps(dummy_mask,velec);
 876             velecsum         = _mm_add_ps(velecsum,velec);
 877
 878             fscal            = felec;
 879
 880             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 881
 882              /* Update vectorial force */
 883             fix1             = _mm_macc_ps(dx11,fscal,fix1);
 884             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
 885             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
 886
 887             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
 888             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
 889             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
 890
 891             /**************************
 892              * CALCULATE INTERACTIONS *
 893              **************************/
 894
 895             r12              = _mm_mul_ps(rsq12,rinv12);
 896             r12              = _mm_andnot_ps(dummy_mask,r12);
 897
 898             /* Calculate table index by multiplying r with table scale and truncate to integer */
 899             rt               = _mm_mul_ps(r12,vftabscale);
 900             vfitab           = _mm_cvttps_epi32(rt);
 901 #ifdef __XOP__
 902             vfeps            = _mm_frcz_ps(rt);
 903 #else
 904             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 905 #endif
 906             twovfeps         = _mm_add_ps(vfeps,vfeps);
 907             vfitab           = _mm_slli_epi32(vfitab,2);
 908
 909             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 910             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 911             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 912             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 913             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 914             _MM_TRANSPOSE4_PS(Y,F,G,H);
 915             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 916             VV               = _mm_macc_ps(vfeps,Fp,Y);
 917             velec            = _mm_mul_ps(qq12,VV);
 918             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 919             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq12,FF),_mm_mul_ps(vftabscale,rinv12)));
 920
 921             /* Update potential sum for this i atom from the interaction with this j atom. */
 922             velec            = _mm_andnot_ps(dummy_mask,velec);
 923             velecsum         = _mm_add_ps(velecsum,velec);
 924
 925             fscal            = felec;
 926
 927             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 928
 929              /* Update vectorial force */
 930             fix1             = _mm_macc_ps(dx12,fscal,fix1);
 931             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
 932             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
 933
 934             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
 935             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
 936             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
 937
 938             /**************************
 939              * CALCULATE INTERACTIONS *
 940              **************************/
 941
 942             r13              = _mm_mul_ps(rsq13,rinv13);
 943             r13              = _mm_andnot_ps(dummy_mask,r13);
 944
 945             /* Calculate table index by multiplying r with table scale and truncate to integer */
 946             rt               = _mm_mul_ps(r13,vftabscale);
 947             vfitab           = _mm_cvttps_epi32(rt);
 948 #ifdef __XOP__
 949             vfeps            = _mm_frcz_ps(rt);
 950 #else
 951             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 952 #endif
 953             twovfeps         = _mm_add_ps(vfeps,vfeps);
 954             vfitab           = _mm_slli_epi32(vfitab,2);
 955
 956             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 957             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 958             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 959             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 960             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 961             _MM_TRANSPOSE4_PS(Y,F,G,H);
 962             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 963             VV               = _mm_macc_ps(vfeps,Fp,Y);
 964             velec            = _mm_mul_ps(qq13,VV);
 965             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 966             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq13,FF),_mm_mul_ps(vftabscale,rinv13)));
 967
 968             /* Update potential sum for this i atom from the interaction with this j atom. */
 969             velec            = _mm_andnot_ps(dummy_mask,velec);
 970             velecsum         = _mm_add_ps(velecsum,velec);
 971
 972             fscal            = felec;
 973
 974             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 975
 976              /* Update vectorial force */
 977             fix1             = _mm_macc_ps(dx13,fscal,fix1);
 978             fiy1             = _mm_macc_ps(dy13,fscal,fiy1);
 979             fiz1             = _mm_macc_ps(dz13,fscal,fiz1);
 980
 981             fjx3             = _mm_macc_ps(dx13,fscal,fjx3);
 982             fjy3             = _mm_macc_ps(dy13,fscal,fjy3);
 983             fjz3             = _mm_macc_ps(dz13,fscal,fjz3);
 984
 985             /**************************
 986              * CALCULATE INTERACTIONS *
 987              **************************/
 988
 989             r21              = _mm_mul_ps(rsq21,rinv21);
 990             r21              = _mm_andnot_ps(dummy_mask,r21);
 991
 992             /* Calculate table index by multiplying r with table scale and truncate to integer */
 993             rt               = _mm_mul_ps(r21,vftabscale);
 994             vfitab           = _mm_cvttps_epi32(rt);
 995 #ifdef __XOP__
 996             vfeps            = _mm_frcz_ps(rt);
 997 #else
 998             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 999 #endif
1000             twovfeps         = _mm_add_ps(vfeps,vfeps);
1001             vfitab           = _mm_slli_epi32(vfitab,2);
1002
1003             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1004             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1005             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1006             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1007             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1008             _MM_TRANSPOSE4_PS(Y,F,G,H);
1009             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1010             VV               = _mm_macc_ps(vfeps,Fp,Y);
1011             velec            = _mm_mul_ps(qq21,VV);
1012             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1013             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq21,FF),_mm_mul_ps(vftabscale,rinv21)));
1014
1015             /* Update potential sum for this i atom from the interaction with this j atom. */
1016             velec            = _mm_andnot_ps(dummy_mask,velec);
1017             velecsum         = _mm_add_ps(velecsum,velec);
1018
1019             fscal            = felec;
1020
1021             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1022
1023              /* Update vectorial force */
1024             fix2             = _mm_macc_ps(dx21,fscal,fix2);
1025             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
1026             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
1027
1028             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
1029             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
1030             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
1031
1032             /**************************
1033              * CALCULATE INTERACTIONS *
1034              **************************/
1035
1036             r22              = _mm_mul_ps(rsq22,rinv22);
1037             r22              = _mm_andnot_ps(dummy_mask,r22);
1038
1039             /* Calculate table index by multiplying r with table scale and truncate to integer */
1040             rt               = _mm_mul_ps(r22,vftabscale);
1041             vfitab           = _mm_cvttps_epi32(rt);
1042 #ifdef __XOP__
1043             vfeps            = _mm_frcz_ps(rt);
1044 #else
1045             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1046 #endif
1047             twovfeps         = _mm_add_ps(vfeps,vfeps);
1048             vfitab           = _mm_slli_epi32(vfitab,2);
1049
1050             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1051             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1052             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1053             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1054             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1055             _MM_TRANSPOSE4_PS(Y,F,G,H);
1056             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1057             VV               = _mm_macc_ps(vfeps,Fp,Y);
1058             velec            = _mm_mul_ps(qq22,VV);
1059             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1060             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq22,FF),_mm_mul_ps(vftabscale,rinv22)));
1061
1062             /* Update potential sum for this i atom from the interaction with this j atom. */
1063             velec            = _mm_andnot_ps(dummy_mask,velec);
1064             velecsum         = _mm_add_ps(velecsum,velec);
1065
1066             fscal            = felec;
1067
1068             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1069
1070              /* Update vectorial force */
1071             fix2             = _mm_macc_ps(dx22,fscal,fix2);
1072             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
1073             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
1074
1075             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
1076             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
1077             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
1078
1079             /**************************
1080              * CALCULATE INTERACTIONS *
1081              **************************/
1082
1083             r23              = _mm_mul_ps(rsq23,rinv23);
1084             r23              = _mm_andnot_ps(dummy_mask,r23);
1085
1086             /* Calculate table index by multiplying r with table scale and truncate to integer */
1087             rt               = _mm_mul_ps(r23,vftabscale);
1088             vfitab           = _mm_cvttps_epi32(rt);
1089 #ifdef __XOP__
1090             vfeps            = _mm_frcz_ps(rt);
1091 #else
1092             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1093 #endif
1094             twovfeps         = _mm_add_ps(vfeps,vfeps);
1095             vfitab           = _mm_slli_epi32(vfitab,2);
1096
1097             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1098             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1099             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1100             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1101             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1102             _MM_TRANSPOSE4_PS(Y,F,G,H);
1103             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1104             VV               = _mm_macc_ps(vfeps,Fp,Y);
1105             velec            = _mm_mul_ps(qq23,VV);
1106             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1107             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq23,FF),_mm_mul_ps(vftabscale,rinv23)));
1108
1109             /* Update potential sum for this i atom from the interaction with this j atom. */
1110             velec            = _mm_andnot_ps(dummy_mask,velec);
1111             velecsum         = _mm_add_ps(velecsum,velec);
1112
1113             fscal            = felec;
1114
1115             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1116
1117              /* Update vectorial force */
1118             fix2             = _mm_macc_ps(dx23,fscal,fix2);
1119             fiy2             = _mm_macc_ps(dy23,fscal,fiy2);
1120             fiz2             = _mm_macc_ps(dz23,fscal,fiz2);
1121
1122             fjx3             = _mm_macc_ps(dx23,fscal,fjx3);
1123             fjy3             = _mm_macc_ps(dy23,fscal,fjy3);
1124             fjz3             = _mm_macc_ps(dz23,fscal,fjz3);
1125
1126             /**************************
1127              * CALCULATE INTERACTIONS *
1128              **************************/
1129
1130             r31              = _mm_mul_ps(rsq31,rinv31);
1131             r31              = _mm_andnot_ps(dummy_mask,r31);
1132
1133             /* Calculate table index by multiplying r with table scale and truncate to integer */
1134             rt               = _mm_mul_ps(r31,vftabscale);
1135             vfitab           = _mm_cvttps_epi32(rt);
1136 #ifdef __XOP__
1137             vfeps            = _mm_frcz_ps(rt);
1138 #else
1139             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1140 #endif
1141             twovfeps         = _mm_add_ps(vfeps,vfeps);
1142             vfitab           = _mm_slli_epi32(vfitab,2);
1143
1144             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1145             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1146             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1147             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1148             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1149             _MM_TRANSPOSE4_PS(Y,F,G,H);
1150             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1151             VV               = _mm_macc_ps(vfeps,Fp,Y);
1152             velec            = _mm_mul_ps(qq31,VV);
1153             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1154             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq31,FF),_mm_mul_ps(vftabscale,rinv31)));
1155
1156             /* Update potential sum for this i atom from the interaction with this j atom. */
1157             velec            = _mm_andnot_ps(dummy_mask,velec);
1158             velecsum         = _mm_add_ps(velecsum,velec);
1159
1160             fscal            = felec;
1161
1162             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1163
1164              /* Update vectorial force */
1165             fix3             = _mm_macc_ps(dx31,fscal,fix3);
1166             fiy3             = _mm_macc_ps(dy31,fscal,fiy3);
1167             fiz3             = _mm_macc_ps(dz31,fscal,fiz3);
1168
1169             fjx1             = _mm_macc_ps(dx31,fscal,fjx1);
1170             fjy1             = _mm_macc_ps(dy31,fscal,fjy1);
1171             fjz1             = _mm_macc_ps(dz31,fscal,fjz1);
1172
1173             /**************************
1174              * CALCULATE INTERACTIONS *
1175              **************************/
1176
1177             r32              = _mm_mul_ps(rsq32,rinv32);
1178             r32              = _mm_andnot_ps(dummy_mask,r32);
1179
1180             /* Calculate table index by multiplying r with table scale and truncate to integer */
1181             rt               = _mm_mul_ps(r32,vftabscale);
1182             vfitab           = _mm_cvttps_epi32(rt);
1183 #ifdef __XOP__
1184             vfeps            = _mm_frcz_ps(rt);
1185 #else
1186             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1187 #endif
1188             twovfeps         = _mm_add_ps(vfeps,vfeps);
1189             vfitab           = _mm_slli_epi32(vfitab,2);
1190
1191             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1192             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1193             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1194             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1195             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1196             _MM_TRANSPOSE4_PS(Y,F,G,H);
1197             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1198             VV               = _mm_macc_ps(vfeps,Fp,Y);
1199             velec            = _mm_mul_ps(qq32,VV);
1200             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1201             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq32,FF),_mm_mul_ps(vftabscale,rinv32)));
1202
1203             /* Update potential sum for this i atom from the interaction with this j atom. */
1204             velec            = _mm_andnot_ps(dummy_mask,velec);
1205             velecsum         = _mm_add_ps(velecsum,velec);
1206
1207             fscal            = felec;
1208
1209             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1210
1211              /* Update vectorial force */
1212             fix3             = _mm_macc_ps(dx32,fscal,fix3);
1213             fiy3             = _mm_macc_ps(dy32,fscal,fiy3);
1214             fiz3             = _mm_macc_ps(dz32,fscal,fiz3);
1215
1216             fjx2             = _mm_macc_ps(dx32,fscal,fjx2);
1217             fjy2             = _mm_macc_ps(dy32,fscal,fjy2);
1218             fjz2             = _mm_macc_ps(dz32,fscal,fjz2);
1219
1220             /**************************
1221              * CALCULATE INTERACTIONS *
1222              **************************/
1223
1224             r33              = _mm_mul_ps(rsq33,rinv33);
1225             r33              = _mm_andnot_ps(dummy_mask,r33);
1226
1227             /* Calculate table index by multiplying r with table scale and truncate to integer */
1228             rt               = _mm_mul_ps(r33,vftabscale);
1229             vfitab           = _mm_cvttps_epi32(rt);
1230 #ifdef __XOP__
1231             vfeps            = _mm_frcz_ps(rt);
1232 #else
1233             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1234 #endif
1235             twovfeps         = _mm_add_ps(vfeps,vfeps);
1236             vfitab           = _mm_slli_epi32(vfitab,2);
1237
1238             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1239             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1240             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1241             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1242             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1243             _MM_TRANSPOSE4_PS(Y,F,G,H);
1244             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1245             VV               = _mm_macc_ps(vfeps,Fp,Y);
1246             velec            = _mm_mul_ps(qq33,VV);
1247             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1248             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq33,FF),_mm_mul_ps(vftabscale,rinv33)));
1249
1250             /* Update potential sum for this i atom from the interaction with this j atom. */
1251             velec            = _mm_andnot_ps(dummy_mask,velec);
1252             velecsum         = _mm_add_ps(velecsum,velec);
1253
1254             fscal            = felec;
1255
1256             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1257
1258              /* Update vectorial force */
1259             fix3             = _mm_macc_ps(dx33,fscal,fix3);
1260             fiy3             = _mm_macc_ps(dy33,fscal,fiy3);
1261             fiz3             = _mm_macc_ps(dz33,fscal,fiz3);
1262
1263             fjx3             = _mm_macc_ps(dx33,fscal,fjx3);
1264             fjy3             = _mm_macc_ps(dy33,fscal,fjy3);
1265             fjz3             = _mm_macc_ps(dz33,fscal,fjz3);
1266
1267             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1268             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1269             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1270             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1271
1272             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1273                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1274                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1275
1276             /* Inner loop uses 461 flops */
1277         }
1278
1279         /* End of innermost loop */
1280
1281         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1282                                               f+i_coord_offset,fshift+i_shift_offset);
1283
1284         ggid                        = gid[iidx];
1285         /* Update potential energies */
1286         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1287         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1288
1289         /* Increment number of inner iterations */
1290         inneriter                  += j_index_end - j_index_start;
1291
1292         /* Outer loop uses 26 flops */
1293     }
1294
1295     /* Increment number of outer iterations */
1296     outeriter        += nri;
1297
1298     /* Update outer/inner flops */
1299
1300     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*461);
1301 }
1302 /*
1303  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW4W4_F_avx_128_fma_single
1304  * Electrostatics interaction: CubicSplineTable
1305  * VdW interaction:            LennardJones
1306  * Geometry:                   Water4-Water4
1307  * Calculate force/pot:        Force
1308  */
1309 void
1310 nb_kernel_ElecCSTab_VdwLJ_GeomW4W4_F_avx_128_fma_single
1311                     (t_nblist * gmx_restrict                nlist,
1312                      rvec * gmx_restrict                    xx,
1313                      rvec * gmx_restrict                    ff,
1314                      t_forcerec * gmx_restrict              fr,
1315                      t_mdatoms * gmx_restrict               mdatoms,
1316                      nb_kernel_data_t * gmx_restrict        kernel_data,
1317                      t_nrnb * gmx_restrict                  nrnb)
1318 {
1319     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1320      * just 0 for non-waters.
1321      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
1322      * jnr indices corresponding to data put in the four positions in the SIMD register.
1323      */
1324     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1325     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1326     int              jnrA,jnrB,jnrC,jnrD;
1327     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1328     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1329     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1330     real             rcutoff_scalar;
1331     real             *shiftvec,*fshift,*x,*f;
1332     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1333     real             scratch[4*DIM];
1334     __m128           fscal,rcutoff,rcutoff2,jidxall;
1335     int              vdwioffset0;
1336     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1337     int              vdwioffset1;
1338     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1339     int              vdwioffset2;
1340     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1341     int              vdwioffset3;
1342     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1343     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1344     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1345     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1346     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1347     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1348     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1349     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
1350     __m128           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1351     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1352     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1353     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1354     __m128           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1355     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1356     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1357     __m128           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1358     __m128           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1359     __m128           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1360     __m128           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1361     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
1362     real             *charge;
1363     int              nvdwtype;
1364     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1365     int              *vdwtype;
1366     real             *vdwparam;
1367     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
1368     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
1369     __m128i          vfitab;
1370     __m128i          ifour       = _mm_set1_epi32(4);
1371     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
1372     real             *vftab;
1373     __m128           dummy_mask,cutoff_mask;
1374     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
1375     __m128           one     = _mm_set1_ps(1.0);
1376     __m128           two     = _mm_set1_ps(2.0);
1377     x                = xx[0];
1378     f                = ff[0];
1379
1380     nri              = nlist->nri;
1381     iinr             = nlist->iinr;
1382     jindex           = nlist->jindex;
1383     jjnr             = nlist->jjnr;
1384     shiftidx         = nlist->shift;
1385     gid              = nlist->gid;
1386     shiftvec         = fr->shift_vec[0];
1387     fshift           = fr->fshift[0];
1388     facel            = _mm_set1_ps(fr->epsfac);
1389     charge           = mdatoms->chargeA;
1390     nvdwtype         = fr->ntype;
1391     vdwparam         = fr->nbfp;
1392     vdwtype          = mdatoms->typeA;
1393
1394     vftab            = kernel_data->table_elec->data;
1395     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
1396
1397     /* Setup water-specific parameters */
1398     inr              = nlist->iinr[0];
1399     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
1400     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
1401     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
1402     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
1403
1404     jq1              = _mm_set1_ps(charge[inr+1]);
1405     jq2              = _mm_set1_ps(charge[inr+2]);
1406     jq3              = _mm_set1_ps(charge[inr+3]);
1407     vdwjidx0A        = 2*vdwtype[inr+0];
1408     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
1409     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
1410     qq11             = _mm_mul_ps(iq1,jq1);
1411     qq12             = _mm_mul_ps(iq1,jq2);
1412     qq13             = _mm_mul_ps(iq1,jq3);
1413     qq21             = _mm_mul_ps(iq2,jq1);
1414     qq22             = _mm_mul_ps(iq2,jq2);
1415     qq23             = _mm_mul_ps(iq2,jq3);
1416     qq31             = _mm_mul_ps(iq3,jq1);
1417     qq32             = _mm_mul_ps(iq3,jq2);
1418     qq33             = _mm_mul_ps(iq3,jq3);
1419
1420     /* Avoid stupid compiler warnings */
1421     jnrA = jnrB = jnrC = jnrD = 0;
1422     j_coord_offsetA = 0;
1423     j_coord_offsetB = 0;
1424     j_coord_offsetC = 0;
1425     j_coord_offsetD = 0;
1426
1427     outeriter        = 0;
1428     inneriter        = 0;
1429
1430     for(iidx=0;iidx<4*DIM;iidx++)
1431     {
1432         scratch[iidx] = 0.0;
1433     }
1434
1435     /* Start outer loop over neighborlists */
1436     for(iidx=0; iidx<nri; iidx++)
1437     {
1438         /* Load shift vector for this list */
1439         i_shift_offset   = DIM*shiftidx[iidx];
1440
1441         /* Load limits for loop over neighbors */
1442         j_index_start    = jindex[iidx];
1443         j_index_end      = jindex[iidx+1];
1444
1445         /* Get outer coordinate index */
1446         inr              = iinr[iidx];
1447         i_coord_offset   = DIM*inr;
1448
1449         /* Load i particle coords and add shift vector */
1450         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1451                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1452
1453         fix0             = _mm_setzero_ps();
1454         fiy0             = _mm_setzero_ps();
1455         fiz0             = _mm_setzero_ps();
1456         fix1             = _mm_setzero_ps();
1457         fiy1             = _mm_setzero_ps();
1458         fiz1             = _mm_setzero_ps();
1459         fix2             = _mm_setzero_ps();
1460         fiy2             = _mm_setzero_ps();
1461         fiz2             = _mm_setzero_ps();
1462         fix3             = _mm_setzero_ps();
1463         fiy3             = _mm_setzero_ps();
1464         fiz3             = _mm_setzero_ps();
1465
1466         /* Start inner kernel loop */
1467         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1468         {
1469
1470             /* Get j neighbor index, and coordinate index */
1471             jnrA             = jjnr[jidx];
1472             jnrB             = jjnr[jidx+1];
1473             jnrC             = jjnr[jidx+2];
1474             jnrD             = jjnr[jidx+3];
1475             j_coord_offsetA  = DIM*jnrA;
1476             j_coord_offsetB  = DIM*jnrB;
1477             j_coord_offsetC  = DIM*jnrC;
1478             j_coord_offsetD  = DIM*jnrD;
1479
1480             /* load j atom coordinates */
1481             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1482                                               x+j_coord_offsetC,x+j_coord_offsetD,
1483                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1484                                               &jy2,&jz2,&jx3,&jy3,&jz3);
1485
1486             /* Calculate displacement vector */
1487             dx00             = _mm_sub_ps(ix0,jx0);
1488             dy00             = _mm_sub_ps(iy0,jy0);
1489             dz00             = _mm_sub_ps(iz0,jz0);
1490             dx11             = _mm_sub_ps(ix1,jx1);
1491             dy11             = _mm_sub_ps(iy1,jy1);
1492             dz11             = _mm_sub_ps(iz1,jz1);
1493             dx12             = _mm_sub_ps(ix1,jx2);
1494             dy12             = _mm_sub_ps(iy1,jy2);
1495             dz12             = _mm_sub_ps(iz1,jz2);
1496             dx13             = _mm_sub_ps(ix1,jx3);
1497             dy13             = _mm_sub_ps(iy1,jy3);
1498             dz13             = _mm_sub_ps(iz1,jz3);
1499             dx21             = _mm_sub_ps(ix2,jx1);
1500             dy21             = _mm_sub_ps(iy2,jy1);
1501             dz21             = _mm_sub_ps(iz2,jz1);
1502             dx22             = _mm_sub_ps(ix2,jx2);
1503             dy22             = _mm_sub_ps(iy2,jy2);
1504             dz22             = _mm_sub_ps(iz2,jz2);
1505             dx23             = _mm_sub_ps(ix2,jx3);
1506             dy23             = _mm_sub_ps(iy2,jy3);
1507             dz23             = _mm_sub_ps(iz2,jz3);
1508             dx31             = _mm_sub_ps(ix3,jx1);
1509             dy31             = _mm_sub_ps(iy3,jy1);
1510             dz31             = _mm_sub_ps(iz3,jz1);
1511             dx32             = _mm_sub_ps(ix3,jx2);
1512             dy32             = _mm_sub_ps(iy3,jy2);
1513             dz32             = _mm_sub_ps(iz3,jz2);
1514             dx33             = _mm_sub_ps(ix3,jx3);
1515             dy33             = _mm_sub_ps(iy3,jy3);
1516             dz33             = _mm_sub_ps(iz3,jz3);
1517
1518             /* Calculate squared distance and things based on it */
1519             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1520             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1521             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1522             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
1523             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1524             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1525             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
1526             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
1527             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
1528             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
1529
1530             rinv11           = gmx_mm_invsqrt_ps(rsq11);
1531             rinv12           = gmx_mm_invsqrt_ps(rsq12);
1532             rinv13           = gmx_mm_invsqrt_ps(rsq13);
1533             rinv21           = gmx_mm_invsqrt_ps(rsq21);
1534             rinv22           = gmx_mm_invsqrt_ps(rsq22);
1535             rinv23           = gmx_mm_invsqrt_ps(rsq23);
1536             rinv31           = gmx_mm_invsqrt_ps(rsq31);
1537             rinv32           = gmx_mm_invsqrt_ps(rsq32);
1538             rinv33           = gmx_mm_invsqrt_ps(rsq33);
1539
1540             rinvsq00         = gmx_mm_inv_ps(rsq00);
1541
1542             fjx0             = _mm_setzero_ps();
1543             fjy0             = _mm_setzero_ps();
1544             fjz0             = _mm_setzero_ps();
1545             fjx1             = _mm_setzero_ps();
1546             fjy1             = _mm_setzero_ps();
1547             fjz1             = _mm_setzero_ps();
1548             fjx2             = _mm_setzero_ps();
1549             fjy2             = _mm_setzero_ps();
1550             fjz2             = _mm_setzero_ps();
1551             fjx3             = _mm_setzero_ps();
1552             fjy3             = _mm_setzero_ps();
1553             fjz3             = _mm_setzero_ps();
1554
1555             /**************************
1556              * CALCULATE INTERACTIONS *
1557              **************************/
1558
1559             /* LENNARD-JONES DISPERSION/REPULSION */
1560
1561             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1562             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1563
1564             fscal            = fvdw;
1565
1566              /* Update vectorial force */
1567             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1568             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1569             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1570
1571             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1572             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1573             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1574
1575             /**************************
1576              * CALCULATE INTERACTIONS *
1577              **************************/
1578
1579             r11              = _mm_mul_ps(rsq11,rinv11);
1580
1581             /* Calculate table index by multiplying r with table scale and truncate to integer */
1582             rt               = _mm_mul_ps(r11,vftabscale);
1583             vfitab           = _mm_cvttps_epi32(rt);
1584 #ifdef __XOP__
1585             vfeps            = _mm_frcz_ps(rt);
1586 #else
1587             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1588 #endif
1589             twovfeps         = _mm_add_ps(vfeps,vfeps);
1590             vfitab           = _mm_slli_epi32(vfitab,2);
1591
1592             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1593             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1594             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1595             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1596             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1597             _MM_TRANSPOSE4_PS(Y,F,G,H);
1598             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1599             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1600             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq11,FF),_mm_mul_ps(vftabscale,rinv11)));
1601
1602             fscal            = felec;
1603
1604              /* Update vectorial force */
1605             fix1             = _mm_macc_ps(dx11,fscal,fix1);
1606             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
1607             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
1608
1609             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
1610             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
1611             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
1612
1613             /**************************
1614              * CALCULATE INTERACTIONS *
1615              **************************/
1616
1617             r12              = _mm_mul_ps(rsq12,rinv12);
1618
1619             /* Calculate table index by multiplying r with table scale and truncate to integer */
1620             rt               = _mm_mul_ps(r12,vftabscale);
1621             vfitab           = _mm_cvttps_epi32(rt);
1622 #ifdef __XOP__
1623             vfeps            = _mm_frcz_ps(rt);
1624 #else
1625             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1626 #endif
1627             twovfeps         = _mm_add_ps(vfeps,vfeps);
1628             vfitab           = _mm_slli_epi32(vfitab,2);
1629
1630             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1631             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1632             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1633             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1634             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1635             _MM_TRANSPOSE4_PS(Y,F,G,H);
1636             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1637             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1638             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq12,FF),_mm_mul_ps(vftabscale,rinv12)));
1639
1640             fscal            = felec;
1641
1642              /* Update vectorial force */
1643             fix1             = _mm_macc_ps(dx12,fscal,fix1);
1644             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
1645             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
1646
1647             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
1648             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
1649             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
1650
1651             /**************************
1652              * CALCULATE INTERACTIONS *
1653              **************************/
1654
1655             r13              = _mm_mul_ps(rsq13,rinv13);
1656
1657             /* Calculate table index by multiplying r with table scale and truncate to integer */
1658             rt               = _mm_mul_ps(r13,vftabscale);
1659             vfitab           = _mm_cvttps_epi32(rt);
1660 #ifdef __XOP__
1661             vfeps            = _mm_frcz_ps(rt);
1662 #else
1663             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1664 #endif
1665             twovfeps         = _mm_add_ps(vfeps,vfeps);
1666             vfitab           = _mm_slli_epi32(vfitab,2);
1667
1668             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1669             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1670             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1671             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1672             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1673             _MM_TRANSPOSE4_PS(Y,F,G,H);
1674             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1675             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1676             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq13,FF),_mm_mul_ps(vftabscale,rinv13)));
1677
1678             fscal            = felec;
1679
1680              /* Update vectorial force */
1681             fix1             = _mm_macc_ps(dx13,fscal,fix1);
1682             fiy1             = _mm_macc_ps(dy13,fscal,fiy1);
1683             fiz1             = _mm_macc_ps(dz13,fscal,fiz1);
1684
1685             fjx3             = _mm_macc_ps(dx13,fscal,fjx3);
1686             fjy3             = _mm_macc_ps(dy13,fscal,fjy3);
1687             fjz3             = _mm_macc_ps(dz13,fscal,fjz3);
1688
1689             /**************************
1690              * CALCULATE INTERACTIONS *
1691              **************************/
1692
1693             r21              = _mm_mul_ps(rsq21,rinv21);
1694
1695             /* Calculate table index by multiplying r with table scale and truncate to integer */
1696             rt               = _mm_mul_ps(r21,vftabscale);
1697             vfitab           = _mm_cvttps_epi32(rt);
1698 #ifdef __XOP__
1699             vfeps            = _mm_frcz_ps(rt);
1700 #else
1701             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1702 #endif
1703             twovfeps         = _mm_add_ps(vfeps,vfeps);
1704             vfitab           = _mm_slli_epi32(vfitab,2);
1705
1706             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1707             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1708             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1709             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1710             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1711             _MM_TRANSPOSE4_PS(Y,F,G,H);
1712             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1713             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1714             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq21,FF),_mm_mul_ps(vftabscale,rinv21)));
1715
1716             fscal            = felec;
1717
1718              /* Update vectorial force */
1719             fix2             = _mm_macc_ps(dx21,fscal,fix2);
1720             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
1721             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
1722
1723             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
1724             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
1725             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
1726
1727             /**************************
1728              * CALCULATE INTERACTIONS *
1729              **************************/
1730
1731             r22              = _mm_mul_ps(rsq22,rinv22);
1732
1733             /* Calculate table index by multiplying r with table scale and truncate to integer */
1734             rt               = _mm_mul_ps(r22,vftabscale);
1735             vfitab           = _mm_cvttps_epi32(rt);
1736 #ifdef __XOP__
1737             vfeps            = _mm_frcz_ps(rt);
1738 #else
1739             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1740 #endif
1741             twovfeps         = _mm_add_ps(vfeps,vfeps);
1742             vfitab           = _mm_slli_epi32(vfitab,2);
1743
1744             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1745             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1746             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1747             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1748             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1749             _MM_TRANSPOSE4_PS(Y,F,G,H);
1750             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1751             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1752             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq22,FF),_mm_mul_ps(vftabscale,rinv22)));
1753
1754             fscal            = felec;
1755
1756              /* Update vectorial force */
1757             fix2             = _mm_macc_ps(dx22,fscal,fix2);
1758             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
1759             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
1760
1761             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
1762             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
1763             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
1764
1765             /**************************
1766              * CALCULATE INTERACTIONS *
1767              **************************/
1768
1769             r23              = _mm_mul_ps(rsq23,rinv23);
1770
1771             /* Calculate table index by multiplying r with table scale and truncate to integer */
1772             rt               = _mm_mul_ps(r23,vftabscale);
1773             vfitab           = _mm_cvttps_epi32(rt);
1774 #ifdef __XOP__
1775             vfeps            = _mm_frcz_ps(rt);
1776 #else
1777             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1778 #endif
1779             twovfeps         = _mm_add_ps(vfeps,vfeps);
1780             vfitab           = _mm_slli_epi32(vfitab,2);
1781
1782             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1783             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1784             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1785             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1786             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1787             _MM_TRANSPOSE4_PS(Y,F,G,H);
1788             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1789             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1790             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq23,FF),_mm_mul_ps(vftabscale,rinv23)));
1791
1792             fscal            = felec;
1793
1794              /* Update vectorial force */
1795             fix2             = _mm_macc_ps(dx23,fscal,fix2);
1796             fiy2             = _mm_macc_ps(dy23,fscal,fiy2);
1797             fiz2             = _mm_macc_ps(dz23,fscal,fiz2);
1798
1799             fjx3             = _mm_macc_ps(dx23,fscal,fjx3);
1800             fjy3             = _mm_macc_ps(dy23,fscal,fjy3);
1801             fjz3             = _mm_macc_ps(dz23,fscal,fjz3);
1802
1803             /**************************
1804              * CALCULATE INTERACTIONS *
1805              **************************/
1806
1807             r31              = _mm_mul_ps(rsq31,rinv31);
1808
1809             /* Calculate table index by multiplying r with table scale and truncate to integer */
1810             rt               = _mm_mul_ps(r31,vftabscale);
1811             vfitab           = _mm_cvttps_epi32(rt);
1812 #ifdef __XOP__
1813             vfeps            = _mm_frcz_ps(rt);
1814 #else
1815             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1816 #endif
1817             twovfeps         = _mm_add_ps(vfeps,vfeps);
1818             vfitab           = _mm_slli_epi32(vfitab,2);
1819
1820             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1821             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1822             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1823             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1824             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1825             _MM_TRANSPOSE4_PS(Y,F,G,H);
1826             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1827             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1828             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq31,FF),_mm_mul_ps(vftabscale,rinv31)));
1829
1830             fscal            = felec;
1831
1832              /* Update vectorial force */
1833             fix3             = _mm_macc_ps(dx31,fscal,fix3);
1834             fiy3             = _mm_macc_ps(dy31,fscal,fiy3);
1835             fiz3             = _mm_macc_ps(dz31,fscal,fiz3);
1836
1837             fjx1             = _mm_macc_ps(dx31,fscal,fjx1);
1838             fjy1             = _mm_macc_ps(dy31,fscal,fjy1);
1839             fjz1             = _mm_macc_ps(dz31,fscal,fjz1);
1840
1841             /**************************
1842              * CALCULATE INTERACTIONS *
1843              **************************/
1844
1845             r32              = _mm_mul_ps(rsq32,rinv32);
1846
1847             /* Calculate table index by multiplying r with table scale and truncate to integer */
1848             rt               = _mm_mul_ps(r32,vftabscale);
1849             vfitab           = _mm_cvttps_epi32(rt);
1850 #ifdef __XOP__
1851             vfeps            = _mm_frcz_ps(rt);
1852 #else
1853             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1854 #endif
1855             twovfeps         = _mm_add_ps(vfeps,vfeps);
1856             vfitab           = _mm_slli_epi32(vfitab,2);
1857
1858             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1859             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1860             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1861             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1862             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1863             _MM_TRANSPOSE4_PS(Y,F,G,H);
1864             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1865             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1866             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq32,FF),_mm_mul_ps(vftabscale,rinv32)));
1867
1868             fscal            = felec;
1869
1870              /* Update vectorial force */
1871             fix3             = _mm_macc_ps(dx32,fscal,fix3);
1872             fiy3             = _mm_macc_ps(dy32,fscal,fiy3);
1873             fiz3             = _mm_macc_ps(dz32,fscal,fiz3);
1874
1875             fjx2             = _mm_macc_ps(dx32,fscal,fjx2);
1876             fjy2             = _mm_macc_ps(dy32,fscal,fjy2);
1877             fjz2             = _mm_macc_ps(dz32,fscal,fjz2);
1878
1879             /**************************
1880              * CALCULATE INTERACTIONS *
1881              **************************/
1882
1883             r33              = _mm_mul_ps(rsq33,rinv33);
1884
1885             /* Calculate table index by multiplying r with table scale and truncate to integer */
1886             rt               = _mm_mul_ps(r33,vftabscale);
1887             vfitab           = _mm_cvttps_epi32(rt);
1888 #ifdef __XOP__
1889             vfeps            = _mm_frcz_ps(rt);
1890 #else
1891             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1892 #endif
1893             twovfeps         = _mm_add_ps(vfeps,vfeps);
1894             vfitab           = _mm_slli_epi32(vfitab,2);
1895
1896             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1897             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1898             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1899             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1900             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1901             _MM_TRANSPOSE4_PS(Y,F,G,H);
1902             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1903             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1904             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq33,FF),_mm_mul_ps(vftabscale,rinv33)));
1905
1906             fscal            = felec;
1907
1908              /* Update vectorial force */
1909             fix3             = _mm_macc_ps(dx33,fscal,fix3);
1910             fiy3             = _mm_macc_ps(dy33,fscal,fiy3);
1911             fiz3             = _mm_macc_ps(dz33,fscal,fiz3);
1912
1913             fjx3             = _mm_macc_ps(dx33,fscal,fjx3);
1914             fjy3             = _mm_macc_ps(dy33,fscal,fjy3);
1915             fjz3             = _mm_macc_ps(dz33,fscal,fjz3);
1916
1917             fjptrA             = f+j_coord_offsetA;
1918             fjptrB             = f+j_coord_offsetB;
1919             fjptrC             = f+j_coord_offsetC;
1920             fjptrD             = f+j_coord_offsetD;
1921
1922             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1923                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1924                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1925
1926             /* Inner loop uses 411 flops */
1927         }
1928
1929         if(jidx<j_index_end)
1930         {
1931
1932             /* Get j neighbor index, and coordinate index */
1933             jnrlistA         = jjnr[jidx];
1934             jnrlistB         = jjnr[jidx+1];
1935             jnrlistC         = jjnr[jidx+2];
1936             jnrlistD         = jjnr[jidx+3];
1937             /* Sign of each element will be negative for non-real atoms.
1938              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1939              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1940              */
1941             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1942             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1943             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1944             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1945             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1946             j_coord_offsetA  = DIM*jnrA;
1947             j_coord_offsetB  = DIM*jnrB;
1948             j_coord_offsetC  = DIM*jnrC;
1949             j_coord_offsetD  = DIM*jnrD;
1950
1951             /* load j atom coordinates */
1952             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1953                                               x+j_coord_offsetC,x+j_coord_offsetD,
1954                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1955                                               &jy2,&jz2,&jx3,&jy3,&jz3);
1956
1957             /* Calculate displacement vector */
1958             dx00             = _mm_sub_ps(ix0,jx0);
1959             dy00             = _mm_sub_ps(iy0,jy0);
1960             dz00             = _mm_sub_ps(iz0,jz0);
1961             dx11             = _mm_sub_ps(ix1,jx1);
1962             dy11             = _mm_sub_ps(iy1,jy1);
1963             dz11             = _mm_sub_ps(iz1,jz1);
1964             dx12             = _mm_sub_ps(ix1,jx2);
1965             dy12             = _mm_sub_ps(iy1,jy2);
1966             dz12             = _mm_sub_ps(iz1,jz2);
1967             dx13             = _mm_sub_ps(ix1,jx3);
1968             dy13             = _mm_sub_ps(iy1,jy3);
1969             dz13             = _mm_sub_ps(iz1,jz3);
1970             dx21             = _mm_sub_ps(ix2,jx1);
1971             dy21             = _mm_sub_ps(iy2,jy1);
1972             dz21             = _mm_sub_ps(iz2,jz1);
1973             dx22             = _mm_sub_ps(ix2,jx2);
1974             dy22             = _mm_sub_ps(iy2,jy2);
1975             dz22             = _mm_sub_ps(iz2,jz2);
1976             dx23             = _mm_sub_ps(ix2,jx3);
1977             dy23             = _mm_sub_ps(iy2,jy3);
1978             dz23             = _mm_sub_ps(iz2,jz3);
1979             dx31             = _mm_sub_ps(ix3,jx1);
1980             dy31             = _mm_sub_ps(iy3,jy1);
1981             dz31             = _mm_sub_ps(iz3,jz1);
1982             dx32             = _mm_sub_ps(ix3,jx2);
1983             dy32             = _mm_sub_ps(iy3,jy2);
1984             dz32             = _mm_sub_ps(iz3,jz2);
1985             dx33             = _mm_sub_ps(ix3,jx3);
1986             dy33             = _mm_sub_ps(iy3,jy3);
1987             dz33             = _mm_sub_ps(iz3,jz3);
1988
1989             /* Calculate squared distance and things based on it */
1990             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1991             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1992             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1993             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
1994             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1995             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1996             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
1997             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
1998             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
1999             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
2000
2001             rinv11           = gmx_mm_invsqrt_ps(rsq11);
2002             rinv12           = gmx_mm_invsqrt_ps(rsq12);
2003             rinv13           = gmx_mm_invsqrt_ps(rsq13);
2004             rinv21           = gmx_mm_invsqrt_ps(rsq21);
2005             rinv22           = gmx_mm_invsqrt_ps(rsq22);
2006             rinv23           = gmx_mm_invsqrt_ps(rsq23);
2007             rinv31           = gmx_mm_invsqrt_ps(rsq31);
2008             rinv32           = gmx_mm_invsqrt_ps(rsq32);
2009             rinv33           = gmx_mm_invsqrt_ps(rsq33);
2010
2011             rinvsq00         = gmx_mm_inv_ps(rsq00);
2012
2013             fjx0             = _mm_setzero_ps();
2014             fjy0             = _mm_setzero_ps();
2015             fjz0             = _mm_setzero_ps();
2016             fjx1             = _mm_setzero_ps();
2017             fjy1             = _mm_setzero_ps();
2018             fjz1             = _mm_setzero_ps();
2019             fjx2             = _mm_setzero_ps();
2020             fjy2             = _mm_setzero_ps();
2021             fjz2             = _mm_setzero_ps();
2022             fjx3             = _mm_setzero_ps();
2023             fjy3             = _mm_setzero_ps();
2024             fjz3             = _mm_setzero_ps();
2025
2026             /**************************
2027              * CALCULATE INTERACTIONS *
2028              **************************/
2029
2030             /* LENNARD-JONES DISPERSION/REPULSION */
2031
2032             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
2033             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
2034
2035             fscal            = fvdw;
2036
2037             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2038
2039              /* Update vectorial force */
2040             fix0             = _mm_macc_ps(dx00,fscal,fix0);
2041             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
2042             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
2043
2044             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
2045             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
2046             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
2047
2048             /**************************
2049              * CALCULATE INTERACTIONS *
2050              **************************/
2051
2052             r11              = _mm_mul_ps(rsq11,rinv11);
2053             r11              = _mm_andnot_ps(dummy_mask,r11);
2054
2055             /* Calculate table index by multiplying r with table scale and truncate to integer */
2056             rt               = _mm_mul_ps(r11,vftabscale);
2057             vfitab           = _mm_cvttps_epi32(rt);
2058 #ifdef __XOP__
2059             vfeps            = _mm_frcz_ps(rt);
2060 #else
2061             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
2062 #endif
2063             twovfeps         = _mm_add_ps(vfeps,vfeps);
2064             vfitab           = _mm_slli_epi32(vfitab,2);
2065
2066             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2067             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
2068             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
2069             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
2070             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
2071             _MM_TRANSPOSE4_PS(Y,F,G,H);
2072             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
2073             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
2074             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq11,FF),_mm_mul_ps(vftabscale,rinv11)));
2075
2076             fscal            = felec;
2077
2078             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2079
2080              /* Update vectorial force */
2081             fix1             = _mm_macc_ps(dx11,fscal,fix1);
2082             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
2083             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
2084
2085             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
2086             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
2087             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
2088
2089             /**************************
2090              * CALCULATE INTERACTIONS *
2091              **************************/
2092
2093             r12              = _mm_mul_ps(rsq12,rinv12);
2094             r12              = _mm_andnot_ps(dummy_mask,r12);
2095
2096             /* Calculate table index by multiplying r with table scale and truncate to integer */
2097             rt               = _mm_mul_ps(r12,vftabscale);
2098             vfitab           = _mm_cvttps_epi32(rt);
2099 #ifdef __XOP__
2100             vfeps            = _mm_frcz_ps(rt);
2101 #else
2102             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
2103 #endif
2104             twovfeps         = _mm_add_ps(vfeps,vfeps);
2105             vfitab           = _mm_slli_epi32(vfitab,2);
2106
2107             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2108             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
2109             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
2110             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
2111             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
2112             _MM_TRANSPOSE4_PS(Y,F,G,H);
2113             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
2114             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
2115             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq12,FF),_mm_mul_ps(vftabscale,rinv12)));
2116
2117             fscal            = felec;
2118
2119             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2120
2121              /* Update vectorial force */
2122             fix1             = _mm_macc_ps(dx12,fscal,fix1);
2123             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
2124             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
2125
2126             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
2127             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
2128             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
2129
2130             /**************************
2131              * CALCULATE INTERACTIONS *
2132              **************************/
2133
2134             r13              = _mm_mul_ps(rsq13,rinv13);
2135             r13              = _mm_andnot_ps(dummy_mask,r13);
2136
2137             /* Calculate table index by multiplying r with table scale and truncate to integer */
2138             rt               = _mm_mul_ps(r13,vftabscale);
2139             vfitab           = _mm_cvttps_epi32(rt);
2140 #ifdef __XOP__
2141             vfeps            = _mm_frcz_ps(rt);
2142 #else
2143             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
2144 #endif
2145             twovfeps         = _mm_add_ps(vfeps,vfeps);
2146             vfitab           = _mm_slli_epi32(vfitab,2);
2147
2148             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2149             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
2150             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
2151             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
2152             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
2153             _MM_TRANSPOSE4_PS(Y,F,G,H);
2154             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
2155             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
2156             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq13,FF),_mm_mul_ps(vftabscale,rinv13)));
2157
2158             fscal            = felec;
2159
2160             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2161
2162              /* Update vectorial force */
2163             fix1             = _mm_macc_ps(dx13,fscal,fix1);
2164             fiy1             = _mm_macc_ps(dy13,fscal,fiy1);
2165             fiz1             = _mm_macc_ps(dz13,fscal,fiz1);
2166
2167             fjx3             = _mm_macc_ps(dx13,fscal,fjx3);
2168             fjy3             = _mm_macc_ps(dy13,fscal,fjy3);
2169             fjz3             = _mm_macc_ps(dz13,fscal,fjz3);
2170
2171             /**************************
2172              * CALCULATE INTERACTIONS *
2173              **************************/
2174
2175             r21              = _mm_mul_ps(rsq21,rinv21);
2176             r21              = _mm_andnot_ps(dummy_mask,r21);
2177
2178             /* Calculate table index by multiplying r with table scale and truncate to integer */
2179             rt               = _mm_mul_ps(r21,vftabscale);
2180             vfitab           = _mm_cvttps_epi32(rt);
2181 #ifdef __XOP__
2182             vfeps            = _mm_frcz_ps(rt);
2183 #else
2184             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
2185 #endif
2186             twovfeps         = _mm_add_ps(vfeps,vfeps);
2187             vfitab           = _mm_slli_epi32(vfitab,2);
2188
2189             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2190             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
2191             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
2192             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
2193             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
2194             _MM_TRANSPOSE4_PS(Y,F,G,H);
2195             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
2196             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
2197             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq21,FF),_mm_mul_ps(vftabscale,rinv21)));
2198
2199             fscal            = felec;
2200
2201             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2202
2203              /* Update vectorial force */
2204             fix2             = _mm_macc_ps(dx21,fscal,fix2);
2205             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
2206             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
2207
2208             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
2209             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
2210             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
2211
2212             /**************************
2213              * CALCULATE INTERACTIONS *
2214              **************************/
2215
2216             r22              = _mm_mul_ps(rsq22,rinv22);
2217             r22              = _mm_andnot_ps(dummy_mask,r22);
2218
2219             /* Calculate table index by multiplying r with table scale and truncate to integer */
2220             rt               = _mm_mul_ps(r22,vftabscale);
2221             vfitab           = _mm_cvttps_epi32(rt);
2222 #ifdef __XOP__
2223             vfeps            = _mm_frcz_ps(rt);
2224 #else
2225             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
2226 #endif
2227             twovfeps         = _mm_add_ps(vfeps,vfeps);
2228             vfitab           = _mm_slli_epi32(vfitab,2);
2229
2230             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2231             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
2232             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
2233             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
2234             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
2235             _MM_TRANSPOSE4_PS(Y,F,G,H);
2236             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
2237             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
2238             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq22,FF),_mm_mul_ps(vftabscale,rinv22)));
2239
2240             fscal            = felec;
2241
2242             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2243
2244              /* Update vectorial force */
2245             fix2             = _mm_macc_ps(dx22,fscal,fix2);
2246             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
2247             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
2248
2249             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
2250             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
2251             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
2252
2253             /**************************
2254              * CALCULATE INTERACTIONS *
2255              **************************/
2256
2257             r23              = _mm_mul_ps(rsq23,rinv23);
2258             r23              = _mm_andnot_ps(dummy_mask,r23);
2259
2260             /* Calculate table index by multiplying r with table scale and truncate to integer */
2261             rt               = _mm_mul_ps(r23,vftabscale);
2262             vfitab           = _mm_cvttps_epi32(rt);
2263 #ifdef __XOP__
2264             vfeps            = _mm_frcz_ps(rt);
2265 #else
2266             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
2267 #endif
2268             twovfeps         = _mm_add_ps(vfeps,vfeps);
2269             vfitab           = _mm_slli_epi32(vfitab,2);
2270
2271             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2272             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
2273             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
2274             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
2275             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
2276             _MM_TRANSPOSE4_PS(Y,F,G,H);
2277             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
2278             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
2279             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq23,FF),_mm_mul_ps(vftabscale,rinv23)));
2280
2281             fscal            = felec;
2282
2283             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2284
2285              /* Update vectorial force */
2286             fix2             = _mm_macc_ps(dx23,fscal,fix2);
2287             fiy2             = _mm_macc_ps(dy23,fscal,fiy2);
2288             fiz2             = _mm_macc_ps(dz23,fscal,fiz2);
2289
2290             fjx3             = _mm_macc_ps(dx23,fscal,fjx3);
2291             fjy3             = _mm_macc_ps(dy23,fscal,fjy3);
2292             fjz3             = _mm_macc_ps(dz23,fscal,fjz3);
2293
2294             /**************************
2295              * CALCULATE INTERACTIONS *
2296              **************************/
2297
2298             r31              = _mm_mul_ps(rsq31,rinv31);
2299             r31              = _mm_andnot_ps(dummy_mask,r31);
2300
2301             /* Calculate table index by multiplying r with table scale and truncate to integer */
2302             rt               = _mm_mul_ps(r31,vftabscale);
2303             vfitab           = _mm_cvttps_epi32(rt);
2304 #ifdef __XOP__
2305             vfeps            = _mm_frcz_ps(rt);
2306 #else
2307             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
2308 #endif
2309             twovfeps         = _mm_add_ps(vfeps,vfeps);
2310             vfitab           = _mm_slli_epi32(vfitab,2);
2311
2312             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2313             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
2314             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
2315             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
2316             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
2317             _MM_TRANSPOSE4_PS(Y,F,G,H);
2318             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
2319             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
2320             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq31,FF),_mm_mul_ps(vftabscale,rinv31)));
2321
2322             fscal            = felec;
2323
2324             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2325
2326              /* Update vectorial force */
2327             fix3             = _mm_macc_ps(dx31,fscal,fix3);
2328             fiy3             = _mm_macc_ps(dy31,fscal,fiy3);
2329             fiz3             = _mm_macc_ps(dz31,fscal,fiz3);
2330
2331             fjx1             = _mm_macc_ps(dx31,fscal,fjx1);
2332             fjy1             = _mm_macc_ps(dy31,fscal,fjy1);
2333             fjz1             = _mm_macc_ps(dz31,fscal,fjz1);
2334
2335             /**************************
2336              * CALCULATE INTERACTIONS *
2337              **************************/
2338
2339             r32              = _mm_mul_ps(rsq32,rinv32);
2340             r32              = _mm_andnot_ps(dummy_mask,r32);
2341
2342             /* Calculate table index by multiplying r with table scale and truncate to integer */
2343             rt               = _mm_mul_ps(r32,vftabscale);
2344             vfitab           = _mm_cvttps_epi32(rt);
2345 #ifdef __XOP__
2346             vfeps            = _mm_frcz_ps(rt);
2347 #else
2348             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
2349 #endif
2350             twovfeps         = _mm_add_ps(vfeps,vfeps);
2351             vfitab           = _mm_slli_epi32(vfitab,2);
2352
2353             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2354             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
2355             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
2356             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
2357             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
2358             _MM_TRANSPOSE4_PS(Y,F,G,H);
2359             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
2360             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
2361             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq32,FF),_mm_mul_ps(vftabscale,rinv32)));
2362
2363             fscal            = felec;
2364
2365             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2366
2367              /* Update vectorial force */
2368             fix3             = _mm_macc_ps(dx32,fscal,fix3);
2369             fiy3             = _mm_macc_ps(dy32,fscal,fiy3);
2370             fiz3             = _mm_macc_ps(dz32,fscal,fiz3);
2371
2372             fjx2             = _mm_macc_ps(dx32,fscal,fjx2);
2373             fjy2             = _mm_macc_ps(dy32,fscal,fjy2);
2374             fjz2             = _mm_macc_ps(dz32,fscal,fjz2);
2375
2376             /**************************
2377              * CALCULATE INTERACTIONS *
2378              **************************/
2379
2380             r33              = _mm_mul_ps(rsq33,rinv33);
2381             r33              = _mm_andnot_ps(dummy_mask,r33);
2382
2383             /* Calculate table index by multiplying r with table scale and truncate to integer */
2384             rt               = _mm_mul_ps(r33,vftabscale);
2385             vfitab           = _mm_cvttps_epi32(rt);
2386 #ifdef __XOP__
2387             vfeps            = _mm_frcz_ps(rt);
2388 #else
2389             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
2390 #endif
2391             twovfeps         = _mm_add_ps(vfeps,vfeps);
2392             vfitab           = _mm_slli_epi32(vfitab,2);
2393
2394             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2395             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
2396             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
2397             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
2398             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
2399             _MM_TRANSPOSE4_PS(Y,F,G,H);
2400             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
2401             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
2402             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq33,FF),_mm_mul_ps(vftabscale,rinv33)));
2403
2404             fscal            = felec;
2405
2406             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2407
2408              /* Update vectorial force */
2409             fix3             = _mm_macc_ps(dx33,fscal,fix3);
2410             fiy3             = _mm_macc_ps(dy33,fscal,fiy3);
2411             fiz3             = _mm_macc_ps(dz33,fscal,fiz3);
2412
2413             fjx3             = _mm_macc_ps(dx33,fscal,fjx3);
2414             fjy3             = _mm_macc_ps(dy33,fscal,fjy3);
2415             fjz3             = _mm_macc_ps(dz33,fscal,fjz3);
2416
2417             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2418             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2419             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2420             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2421
2422             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
2423                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2424                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2425
2426             /* Inner loop uses 420 flops */
2427         }
2428
2429         /* End of innermost loop */
2430
2431         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2432                                               f+i_coord_offset,fshift+i_shift_offset);
2433
2434         /* Increment number of inner iterations */
2435         inneriter                  += j_index_end - j_index_start;
2436
2437         /* Outer loop uses 24 flops */
2438     }
2439
2440     /* Increment number of outer iterations */
2441     outeriter        += nri;
2442
2443     /* Update outer/inner flops */
2444
2445     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*420);
2446 }