src/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecCSTab_VdwCSTab_GeomW3W3_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_VdwCSTab_GeomW3W3_VF_avx_128_fma_single
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water3-Water3
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_VdwCSTab_GeomW3W3_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              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  76     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  77     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  78     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  79     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  80     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  81     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  82     __m128           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
  83     __m128           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
  84     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  85     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  86     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
  87     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  88     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
  89     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
  90     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     int              nvdwtype;
  93     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  94     int              *vdwtype;
  95     real             *vdwparam;
  96     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  97     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  98     __m128i          vfitab;
  99     __m128i          ifour       = _mm_set1_epi32(4);
 100     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 101     real             *vftab;
 102     __m128           dummy_mask,cutoff_mask;
 103     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 104     __m128           one     = _mm_set1_ps(1.0);
 105     __m128           two     = _mm_set1_ps(2.0);
 106     x                = xx[0];
 107     f                = ff[0];
 108
 109     nri              = nlist->nri;
 110     iinr             = nlist->iinr;
 111     jindex           = nlist->jindex;
 112     jjnr             = nlist->jjnr;
 113     shiftidx         = nlist->shift;
 114     gid              = nlist->gid;
 115     shiftvec         = fr->shift_vec[0];
 116     fshift           = fr->fshift[0];
 117     facel            = _mm_set1_ps(fr->epsfac);
 118     charge           = mdatoms->chargeA;
 119     nvdwtype         = fr->ntype;
 120     vdwparam         = fr->nbfp;
 121     vdwtype          = mdatoms->typeA;
 122
 123     vftab            = kernel_data->table_elec_vdw->data;
 124     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 125
 126     /* Setup water-specific parameters */
 127     inr              = nlist->iinr[0];
 128     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 129     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 130     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 131     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 132
 133     jq0              = _mm_set1_ps(charge[inr+0]);
 134     jq1              = _mm_set1_ps(charge[inr+1]);
 135     jq2              = _mm_set1_ps(charge[inr+2]);
 136     vdwjidx0A        = 2*vdwtype[inr+0];
 137     qq00             = _mm_mul_ps(iq0,jq0);
 138     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
 139     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
 140     qq01             = _mm_mul_ps(iq0,jq1);
 141     qq02             = _mm_mul_ps(iq0,jq2);
 142     qq10             = _mm_mul_ps(iq1,jq0);
 143     qq11             = _mm_mul_ps(iq1,jq1);
 144     qq12             = _mm_mul_ps(iq1,jq2);
 145     qq20             = _mm_mul_ps(iq2,jq0);
 146     qq21             = _mm_mul_ps(iq2,jq1);
 147     qq22             = _mm_mul_ps(iq2,jq2);
 148
 149     /* Avoid stupid compiler warnings */
 150     jnrA = jnrB = jnrC = jnrD = 0;
 151     j_coord_offsetA = 0;
 152     j_coord_offsetB = 0;
 153     j_coord_offsetC = 0;
 154     j_coord_offsetD = 0;
 155
 156     outeriter        = 0;
 157     inneriter        = 0;
 158
 159     for(iidx=0;iidx<4*DIM;iidx++)
 160     {
 161         scratch[iidx] = 0.0;
 162     }
 163
 164     /* Start outer loop over neighborlists */
 165     for(iidx=0; iidx<nri; iidx++)
 166     {
 167         /* Load shift vector for this list */
 168         i_shift_offset   = DIM*shiftidx[iidx];
 169
 170         /* Load limits for loop over neighbors */
 171         j_index_start    = jindex[iidx];
 172         j_index_end      = jindex[iidx+1];
 173
 174         /* Get outer coordinate index */
 175         inr              = iinr[iidx];
 176         i_coord_offset   = DIM*inr;
 177
 178         /* Load i particle coords and add shift vector */
 179         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 180                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 181
 182         fix0             = _mm_setzero_ps();
 183         fiy0             = _mm_setzero_ps();
 184         fiz0             = _mm_setzero_ps();
 185         fix1             = _mm_setzero_ps();
 186         fiy1             = _mm_setzero_ps();
 187         fiz1             = _mm_setzero_ps();
 188         fix2             = _mm_setzero_ps();
 189         fiy2             = _mm_setzero_ps();
 190         fiz2             = _mm_setzero_ps();
 191
 192         /* Reset potential sums */
 193         velecsum         = _mm_setzero_ps();
 194         vvdwsum          = _mm_setzero_ps();
 195
 196         /* Start inner kernel loop */
 197         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 198         {
 199
 200             /* Get j neighbor index, and coordinate index */
 201             jnrA             = jjnr[jidx];
 202             jnrB             = jjnr[jidx+1];
 203             jnrC             = jjnr[jidx+2];
 204             jnrD             = jjnr[jidx+3];
 205             j_coord_offsetA  = DIM*jnrA;
 206             j_coord_offsetB  = DIM*jnrB;
 207             j_coord_offsetC  = DIM*jnrC;
 208             j_coord_offsetD  = DIM*jnrD;
 209
 210             /* load j atom coordinates */
 211             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 212                                               x+j_coord_offsetC,x+j_coord_offsetD,
 213                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 214
 215             /* Calculate displacement vector */
 216             dx00             = _mm_sub_ps(ix0,jx0);
 217             dy00             = _mm_sub_ps(iy0,jy0);
 218             dz00             = _mm_sub_ps(iz0,jz0);
 219             dx01             = _mm_sub_ps(ix0,jx1);
 220             dy01             = _mm_sub_ps(iy0,jy1);
 221             dz01             = _mm_sub_ps(iz0,jz1);
 222             dx02             = _mm_sub_ps(ix0,jx2);
 223             dy02             = _mm_sub_ps(iy0,jy2);
 224             dz02             = _mm_sub_ps(iz0,jz2);
 225             dx10             = _mm_sub_ps(ix1,jx0);
 226             dy10             = _mm_sub_ps(iy1,jy0);
 227             dz10             = _mm_sub_ps(iz1,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             dx20             = _mm_sub_ps(ix2,jx0);
 235             dy20             = _mm_sub_ps(iy2,jy0);
 236             dz20             = _mm_sub_ps(iz2,jz0);
 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
 244             /* Calculate squared distance and things based on it */
 245             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 246             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
 247             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
 248             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 249             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 250             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 251             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 252             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 253             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 254
 255             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 256             rinv01           = gmx_mm_invsqrt_ps(rsq01);
 257             rinv02           = gmx_mm_invsqrt_ps(rsq02);
 258             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 259             rinv11           = gmx_mm_invsqrt_ps(rsq11);
 260             rinv12           = gmx_mm_invsqrt_ps(rsq12);
 261             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 262             rinv21           = gmx_mm_invsqrt_ps(rsq21);
 263             rinv22           = gmx_mm_invsqrt_ps(rsq22);
 264
 265             fjx0             = _mm_setzero_ps();
 266             fjy0             = _mm_setzero_ps();
 267             fjz0             = _mm_setzero_ps();
 268             fjx1             = _mm_setzero_ps();
 269             fjy1             = _mm_setzero_ps();
 270             fjz1             = _mm_setzero_ps();
 271             fjx2             = _mm_setzero_ps();
 272             fjy2             = _mm_setzero_ps();
 273             fjz2             = _mm_setzero_ps();
 274
 275             /**************************
 276              * CALCULATE INTERACTIONS *
 277              **************************/
 278
 279             r00              = _mm_mul_ps(rsq00,rinv00);
 280
 281             /* Calculate table index by multiplying r with table scale and truncate to integer */
 282             rt               = _mm_mul_ps(r00,vftabscale);
 283             vfitab           = _mm_cvttps_epi32(rt);
 284 #ifdef __XOP__
 285             vfeps            = _mm_frcz_ps(rt);
 286 #else
 287             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 288 #endif
 289             twovfeps         = _mm_add_ps(vfeps,vfeps);
 290             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 291
 292             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 293             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 294             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 295             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 296             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 297             _MM_TRANSPOSE4_PS(Y,F,G,H);
 298             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 299             VV               = _mm_macc_ps(vfeps,Fp,Y);
 300             velec            = _mm_mul_ps(qq00,VV);
 301             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 302             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 303
 304             /* CUBIC SPLINE TABLE DISPERSION */
 305             vfitab           = _mm_add_epi32(vfitab,ifour);
 306             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 307             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 308             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 309             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 310             _MM_TRANSPOSE4_PS(Y,F,G,H);
 311             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 312             VV               = _mm_macc_ps(vfeps,Fp,Y);
 313             vvdw6            = _mm_mul_ps(c6_00,VV);
 314             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 315             fvdw6            = _mm_mul_ps(c6_00,FF);
 316
 317             /* CUBIC SPLINE TABLE REPULSION */
 318             vfitab           = _mm_add_epi32(vfitab,ifour);
 319             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 320             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 321             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 322             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 323             _MM_TRANSPOSE4_PS(Y,F,G,H);
 324             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 325             VV               = _mm_macc_ps(vfeps,Fp,Y);
 326             vvdw12           = _mm_mul_ps(c12_00,VV);
 327             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 328             fvdw12           = _mm_mul_ps(c12_00,FF);
 329             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 330             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 331
 332             /* Update potential sum for this i atom from the interaction with this j atom. */
 333             velecsum         = _mm_add_ps(velecsum,velec);
 334             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 335
 336             fscal            = _mm_add_ps(felec,fvdw);
 337
 338              /* Update vectorial force */
 339             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 340             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 341             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 342
 343             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 344             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 345             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 346
 347             /**************************
 348              * CALCULATE INTERACTIONS *
 349              **************************/
 350
 351             r01              = _mm_mul_ps(rsq01,rinv01);
 352
 353             /* Calculate table index by multiplying r with table scale and truncate to integer */
 354             rt               = _mm_mul_ps(r01,vftabscale);
 355             vfitab           = _mm_cvttps_epi32(rt);
 356 #ifdef __XOP__
 357             vfeps            = _mm_frcz_ps(rt);
 358 #else
 359             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 360 #endif
 361             twovfeps         = _mm_add_ps(vfeps,vfeps);
 362             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 363
 364             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 365             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 366             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 367             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 368             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 369             _MM_TRANSPOSE4_PS(Y,F,G,H);
 370             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 371             VV               = _mm_macc_ps(vfeps,Fp,Y);
 372             velec            = _mm_mul_ps(qq01,VV);
 373             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 374             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq01,FF),_mm_mul_ps(vftabscale,rinv01)));
 375
 376             /* Update potential sum for this i atom from the interaction with this j atom. */
 377             velecsum         = _mm_add_ps(velecsum,velec);
 378
 379             fscal            = felec;
 380
 381              /* Update vectorial force */
 382             fix0             = _mm_macc_ps(dx01,fscal,fix0);
 383             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
 384             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
 385
 386             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
 387             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
 388             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
 389
 390             /**************************
 391              * CALCULATE INTERACTIONS *
 392              **************************/
 393
 394             r02              = _mm_mul_ps(rsq02,rinv02);
 395
 396             /* Calculate table index by multiplying r with table scale and truncate to integer */
 397             rt               = _mm_mul_ps(r02,vftabscale);
 398             vfitab           = _mm_cvttps_epi32(rt);
 399 #ifdef __XOP__
 400             vfeps            = _mm_frcz_ps(rt);
 401 #else
 402             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 403 #endif
 404             twovfeps         = _mm_add_ps(vfeps,vfeps);
 405             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 406
 407             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 408             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 409             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 410             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 411             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 412             _MM_TRANSPOSE4_PS(Y,F,G,H);
 413             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 414             VV               = _mm_macc_ps(vfeps,Fp,Y);
 415             velec            = _mm_mul_ps(qq02,VV);
 416             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 417             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq02,FF),_mm_mul_ps(vftabscale,rinv02)));
 418
 419             /* Update potential sum for this i atom from the interaction with this j atom. */
 420             velecsum         = _mm_add_ps(velecsum,velec);
 421
 422             fscal            = felec;
 423
 424              /* Update vectorial force */
 425             fix0             = _mm_macc_ps(dx02,fscal,fix0);
 426             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
 427             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
 428
 429             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
 430             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
 431             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
 432
 433             /**************************
 434              * CALCULATE INTERACTIONS *
 435              **************************/
 436
 437             r10              = _mm_mul_ps(rsq10,rinv10);
 438
 439             /* Calculate table index by multiplying r with table scale and truncate to integer */
 440             rt               = _mm_mul_ps(r10,vftabscale);
 441             vfitab           = _mm_cvttps_epi32(rt);
 442 #ifdef __XOP__
 443             vfeps            = _mm_frcz_ps(rt);
 444 #else
 445             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 446 #endif
 447             twovfeps         = _mm_add_ps(vfeps,vfeps);
 448             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 449
 450             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 451             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 452             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 453             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 454             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 455             _MM_TRANSPOSE4_PS(Y,F,G,H);
 456             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 457             VV               = _mm_macc_ps(vfeps,Fp,Y);
 458             velec            = _mm_mul_ps(qq10,VV);
 459             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 460             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 461
 462             /* Update potential sum for this i atom from the interaction with this j atom. */
 463             velecsum         = _mm_add_ps(velecsum,velec);
 464
 465             fscal            = felec;
 466
 467              /* Update vectorial force */
 468             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 469             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 470             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 471
 472             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 473             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 474             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 475
 476             /**************************
 477              * CALCULATE INTERACTIONS *
 478              **************************/
 479
 480             r11              = _mm_mul_ps(rsq11,rinv11);
 481
 482             /* Calculate table index by multiplying r with table scale and truncate to integer */
 483             rt               = _mm_mul_ps(r11,vftabscale);
 484             vfitab           = _mm_cvttps_epi32(rt);
 485 #ifdef __XOP__
 486             vfeps            = _mm_frcz_ps(rt);
 487 #else
 488             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 489 #endif
 490             twovfeps         = _mm_add_ps(vfeps,vfeps);
 491             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 492
 493             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 494             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 495             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 496             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 497             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 498             _MM_TRANSPOSE4_PS(Y,F,G,H);
 499             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 500             VV               = _mm_macc_ps(vfeps,Fp,Y);
 501             velec            = _mm_mul_ps(qq11,VV);
 502             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 503             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq11,FF),_mm_mul_ps(vftabscale,rinv11)));
 504
 505             /* Update potential sum for this i atom from the interaction with this j atom. */
 506             velecsum         = _mm_add_ps(velecsum,velec);
 507
 508             fscal            = felec;
 509
 510              /* Update vectorial force */
 511             fix1             = _mm_macc_ps(dx11,fscal,fix1);
 512             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
 513             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
 514
 515             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
 516             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
 517             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
 518
 519             /**************************
 520              * CALCULATE INTERACTIONS *
 521              **************************/
 522
 523             r12              = _mm_mul_ps(rsq12,rinv12);
 524
 525             /* Calculate table index by multiplying r with table scale and truncate to integer */
 526             rt               = _mm_mul_ps(r12,vftabscale);
 527             vfitab           = _mm_cvttps_epi32(rt);
 528 #ifdef __XOP__
 529             vfeps            = _mm_frcz_ps(rt);
 530 #else
 531             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 532 #endif
 533             twovfeps         = _mm_add_ps(vfeps,vfeps);
 534             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 535
 536             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 537             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 538             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 539             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 540             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 541             _MM_TRANSPOSE4_PS(Y,F,G,H);
 542             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 543             VV               = _mm_macc_ps(vfeps,Fp,Y);
 544             velec            = _mm_mul_ps(qq12,VV);
 545             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 546             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq12,FF),_mm_mul_ps(vftabscale,rinv12)));
 547
 548             /* Update potential sum for this i atom from the interaction with this j atom. */
 549             velecsum         = _mm_add_ps(velecsum,velec);
 550
 551             fscal            = felec;
 552
 553              /* Update vectorial force */
 554             fix1             = _mm_macc_ps(dx12,fscal,fix1);
 555             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
 556             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
 557
 558             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
 559             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
 560             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
 561
 562             /**************************
 563              * CALCULATE INTERACTIONS *
 564              **************************/
 565
 566             r20              = _mm_mul_ps(rsq20,rinv20);
 567
 568             /* Calculate table index by multiplying r with table scale and truncate to integer */
 569             rt               = _mm_mul_ps(r20,vftabscale);
 570             vfitab           = _mm_cvttps_epi32(rt);
 571 #ifdef __XOP__
 572             vfeps            = _mm_frcz_ps(rt);
 573 #else
 574             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 575 #endif
 576             twovfeps         = _mm_add_ps(vfeps,vfeps);
 577             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 578
 579             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 580             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 581             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 582             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 583             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 584             _MM_TRANSPOSE4_PS(Y,F,G,H);
 585             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 586             VV               = _mm_macc_ps(vfeps,Fp,Y);
 587             velec            = _mm_mul_ps(qq20,VV);
 588             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 589             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 590
 591             /* Update potential sum for this i atom from the interaction with this j atom. */
 592             velecsum         = _mm_add_ps(velecsum,velec);
 593
 594             fscal            = felec;
 595
 596              /* Update vectorial force */
 597             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 598             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 599             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 600
 601             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 602             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 603             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 604
 605             /**************************
 606              * CALCULATE INTERACTIONS *
 607              **************************/
 608
 609             r21              = _mm_mul_ps(rsq21,rinv21);
 610
 611             /* Calculate table index by multiplying r with table scale and truncate to integer */
 612             rt               = _mm_mul_ps(r21,vftabscale);
 613             vfitab           = _mm_cvttps_epi32(rt);
 614 #ifdef __XOP__
 615             vfeps            = _mm_frcz_ps(rt);
 616 #else
 617             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 618 #endif
 619             twovfeps         = _mm_add_ps(vfeps,vfeps);
 620             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 621
 622             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 623             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 624             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 625             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 626             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 627             _MM_TRANSPOSE4_PS(Y,F,G,H);
 628             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 629             VV               = _mm_macc_ps(vfeps,Fp,Y);
 630             velec            = _mm_mul_ps(qq21,VV);
 631             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 632             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq21,FF),_mm_mul_ps(vftabscale,rinv21)));
 633
 634             /* Update potential sum for this i atom from the interaction with this j atom. */
 635             velecsum         = _mm_add_ps(velecsum,velec);
 636
 637             fscal            = felec;
 638
 639              /* Update vectorial force */
 640             fix2             = _mm_macc_ps(dx21,fscal,fix2);
 641             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
 642             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
 643
 644             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
 645             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
 646             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
 647
 648             /**************************
 649              * CALCULATE INTERACTIONS *
 650              **************************/
 651
 652             r22              = _mm_mul_ps(rsq22,rinv22);
 653
 654             /* Calculate table index by multiplying r with table scale and truncate to integer */
 655             rt               = _mm_mul_ps(r22,vftabscale);
 656             vfitab           = _mm_cvttps_epi32(rt);
 657 #ifdef __XOP__
 658             vfeps            = _mm_frcz_ps(rt);
 659 #else
 660             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 661 #endif
 662             twovfeps         = _mm_add_ps(vfeps,vfeps);
 663             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 664
 665             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 666             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 667             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 668             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 669             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 670             _MM_TRANSPOSE4_PS(Y,F,G,H);
 671             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 672             VV               = _mm_macc_ps(vfeps,Fp,Y);
 673             velec            = _mm_mul_ps(qq22,VV);
 674             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 675             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq22,FF),_mm_mul_ps(vftabscale,rinv22)));
 676
 677             /* Update potential sum for this i atom from the interaction with this j atom. */
 678             velecsum         = _mm_add_ps(velecsum,velec);
 679
 680             fscal            = felec;
 681
 682              /* Update vectorial force */
 683             fix2             = _mm_macc_ps(dx22,fscal,fix2);
 684             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
 685             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
 686
 687             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
 688             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
 689             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
 690
 691             fjptrA             = f+j_coord_offsetA;
 692             fjptrB             = f+j_coord_offsetB;
 693             fjptrC             = f+j_coord_offsetC;
 694             fjptrD             = f+j_coord_offsetD;
 695
 696             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 697                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 698
 699             /* Inner loop uses 444 flops */
 700         }
 701
 702         if(jidx<j_index_end)
 703         {
 704
 705             /* Get j neighbor index, and coordinate index */
 706             jnrlistA         = jjnr[jidx];
 707             jnrlistB         = jjnr[jidx+1];
 708             jnrlistC         = jjnr[jidx+2];
 709             jnrlistD         = jjnr[jidx+3];
 710             /* Sign of each element will be negative for non-real atoms.
 711              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 712              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 713              */
 714             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 715             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 716             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 717             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 718             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 719             j_coord_offsetA  = DIM*jnrA;
 720             j_coord_offsetB  = DIM*jnrB;
 721             j_coord_offsetC  = DIM*jnrC;
 722             j_coord_offsetD  = DIM*jnrD;
 723
 724             /* load j atom coordinates */
 725             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 726                                               x+j_coord_offsetC,x+j_coord_offsetD,
 727                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 728
 729             /* Calculate displacement vector */
 730             dx00             = _mm_sub_ps(ix0,jx0);
 731             dy00             = _mm_sub_ps(iy0,jy0);
 732             dz00             = _mm_sub_ps(iz0,jz0);
 733             dx01             = _mm_sub_ps(ix0,jx1);
 734             dy01             = _mm_sub_ps(iy0,jy1);
 735             dz01             = _mm_sub_ps(iz0,jz1);
 736             dx02             = _mm_sub_ps(ix0,jx2);
 737             dy02             = _mm_sub_ps(iy0,jy2);
 738             dz02             = _mm_sub_ps(iz0,jz2);
 739             dx10             = _mm_sub_ps(ix1,jx0);
 740             dy10             = _mm_sub_ps(iy1,jy0);
 741             dz10             = _mm_sub_ps(iz1,jz0);
 742             dx11             = _mm_sub_ps(ix1,jx1);
 743             dy11             = _mm_sub_ps(iy1,jy1);
 744             dz11             = _mm_sub_ps(iz1,jz1);
 745             dx12             = _mm_sub_ps(ix1,jx2);
 746             dy12             = _mm_sub_ps(iy1,jy2);
 747             dz12             = _mm_sub_ps(iz1,jz2);
 748             dx20             = _mm_sub_ps(ix2,jx0);
 749             dy20             = _mm_sub_ps(iy2,jy0);
 750             dz20             = _mm_sub_ps(iz2,jz0);
 751             dx21             = _mm_sub_ps(ix2,jx1);
 752             dy21             = _mm_sub_ps(iy2,jy1);
 753             dz21             = _mm_sub_ps(iz2,jz1);
 754             dx22             = _mm_sub_ps(ix2,jx2);
 755             dy22             = _mm_sub_ps(iy2,jy2);
 756             dz22             = _mm_sub_ps(iz2,jz2);
 757
 758             /* Calculate squared distance and things based on it */
 759             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 760             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
 761             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
 762             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 763             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 764             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 765             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 766             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 767             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 768
 769             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 770             rinv01           = gmx_mm_invsqrt_ps(rsq01);
 771             rinv02           = gmx_mm_invsqrt_ps(rsq02);
 772             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 773             rinv11           = gmx_mm_invsqrt_ps(rsq11);
 774             rinv12           = gmx_mm_invsqrt_ps(rsq12);
 775             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 776             rinv21           = gmx_mm_invsqrt_ps(rsq21);
 777             rinv22           = gmx_mm_invsqrt_ps(rsq22);
 778
 779             fjx0             = _mm_setzero_ps();
 780             fjy0             = _mm_setzero_ps();
 781             fjz0             = _mm_setzero_ps();
 782             fjx1             = _mm_setzero_ps();
 783             fjy1             = _mm_setzero_ps();
 784             fjz1             = _mm_setzero_ps();
 785             fjx2             = _mm_setzero_ps();
 786             fjy2             = _mm_setzero_ps();
 787             fjz2             = _mm_setzero_ps();
 788
 789             /**************************
 790              * CALCULATE INTERACTIONS *
 791              **************************/
 792
 793             r00              = _mm_mul_ps(rsq00,rinv00);
 794             r00              = _mm_andnot_ps(dummy_mask,r00);
 795
 796             /* Calculate table index by multiplying r with table scale and truncate to integer */
 797             rt               = _mm_mul_ps(r00,vftabscale);
 798             vfitab           = _mm_cvttps_epi32(rt);
 799 #ifdef __XOP__
 800             vfeps            = _mm_frcz_ps(rt);
 801 #else
 802             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 803 #endif
 804             twovfeps         = _mm_add_ps(vfeps,vfeps);
 805             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 806
 807             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 808             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 809             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 810             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 811             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 812             _MM_TRANSPOSE4_PS(Y,F,G,H);
 813             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 814             VV               = _mm_macc_ps(vfeps,Fp,Y);
 815             velec            = _mm_mul_ps(qq00,VV);
 816             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 817             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 818
 819             /* CUBIC SPLINE TABLE DISPERSION */
 820             vfitab           = _mm_add_epi32(vfitab,ifour);
 821             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 822             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 823             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 824             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 825             _MM_TRANSPOSE4_PS(Y,F,G,H);
 826             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 827             VV               = _mm_macc_ps(vfeps,Fp,Y);
 828             vvdw6            = _mm_mul_ps(c6_00,VV);
 829             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 830             fvdw6            = _mm_mul_ps(c6_00,FF);
 831
 832             /* CUBIC SPLINE TABLE REPULSION */
 833             vfitab           = _mm_add_epi32(vfitab,ifour);
 834             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 835             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 836             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 837             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 838             _MM_TRANSPOSE4_PS(Y,F,G,H);
 839             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 840             VV               = _mm_macc_ps(vfeps,Fp,Y);
 841             vvdw12           = _mm_mul_ps(c12_00,VV);
 842             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 843             fvdw12           = _mm_mul_ps(c12_00,FF);
 844             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 845             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 846
 847             /* Update potential sum for this i atom from the interaction with this j atom. */
 848             velec            = _mm_andnot_ps(dummy_mask,velec);
 849             velecsum         = _mm_add_ps(velecsum,velec);
 850             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 851             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 852
 853             fscal            = _mm_add_ps(felec,fvdw);
 854
 855             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 856
 857              /* Update vectorial force */
 858             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 859             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 860             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 861
 862             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 863             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 864             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 865
 866             /**************************
 867              * CALCULATE INTERACTIONS *
 868              **************************/
 869
 870             r01              = _mm_mul_ps(rsq01,rinv01);
 871             r01              = _mm_andnot_ps(dummy_mask,r01);
 872
 873             /* Calculate table index by multiplying r with table scale and truncate to integer */
 874             rt               = _mm_mul_ps(r01,vftabscale);
 875             vfitab           = _mm_cvttps_epi32(rt);
 876 #ifdef __XOP__
 877             vfeps            = _mm_frcz_ps(rt);
 878 #else
 879             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 880 #endif
 881             twovfeps         = _mm_add_ps(vfeps,vfeps);
 882             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 883
 884             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 885             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 886             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 887             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 888             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 889             _MM_TRANSPOSE4_PS(Y,F,G,H);
 890             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 891             VV               = _mm_macc_ps(vfeps,Fp,Y);
 892             velec            = _mm_mul_ps(qq01,VV);
 893             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 894             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq01,FF),_mm_mul_ps(vftabscale,rinv01)));
 895
 896             /* Update potential sum for this i atom from the interaction with this j atom. */
 897             velec            = _mm_andnot_ps(dummy_mask,velec);
 898             velecsum         = _mm_add_ps(velecsum,velec);
 899
 900             fscal            = felec;
 901
 902             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 903
 904              /* Update vectorial force */
 905             fix0             = _mm_macc_ps(dx01,fscal,fix0);
 906             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
 907             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
 908
 909             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
 910             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
 911             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
 912
 913             /**************************
 914              * CALCULATE INTERACTIONS *
 915              **************************/
 916
 917             r02              = _mm_mul_ps(rsq02,rinv02);
 918             r02              = _mm_andnot_ps(dummy_mask,r02);
 919
 920             /* Calculate table index by multiplying r with table scale and truncate to integer */
 921             rt               = _mm_mul_ps(r02,vftabscale);
 922             vfitab           = _mm_cvttps_epi32(rt);
 923 #ifdef __XOP__
 924             vfeps            = _mm_frcz_ps(rt);
 925 #else
 926             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 927 #endif
 928             twovfeps         = _mm_add_ps(vfeps,vfeps);
 929             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 930
 931             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 932             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 933             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 934             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 935             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 936             _MM_TRANSPOSE4_PS(Y,F,G,H);
 937             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 938             VV               = _mm_macc_ps(vfeps,Fp,Y);
 939             velec            = _mm_mul_ps(qq02,VV);
 940             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 941             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq02,FF),_mm_mul_ps(vftabscale,rinv02)));
 942
 943             /* Update potential sum for this i atom from the interaction with this j atom. */
 944             velec            = _mm_andnot_ps(dummy_mask,velec);
 945             velecsum         = _mm_add_ps(velecsum,velec);
 946
 947             fscal            = felec;
 948
 949             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 950
 951              /* Update vectorial force */
 952             fix0             = _mm_macc_ps(dx02,fscal,fix0);
 953             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
 954             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
 955
 956             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
 957             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
 958             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
 959
 960             /**************************
 961              * CALCULATE INTERACTIONS *
 962              **************************/
 963
 964             r10              = _mm_mul_ps(rsq10,rinv10);
 965             r10              = _mm_andnot_ps(dummy_mask,r10);
 966
 967             /* Calculate table index by multiplying r with table scale and truncate to integer */
 968             rt               = _mm_mul_ps(r10,vftabscale);
 969             vfitab           = _mm_cvttps_epi32(rt);
 970 #ifdef __XOP__
 971             vfeps            = _mm_frcz_ps(rt);
 972 #else
 973             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 974 #endif
 975             twovfeps         = _mm_add_ps(vfeps,vfeps);
 976             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 977
 978             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 979             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 980             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 981             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 982             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 983             _MM_TRANSPOSE4_PS(Y,F,G,H);
 984             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 985             VV               = _mm_macc_ps(vfeps,Fp,Y);
 986             velec            = _mm_mul_ps(qq10,VV);
 987             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 988             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 989
 990             /* Update potential sum for this i atom from the interaction with this j atom. */
 991             velec            = _mm_andnot_ps(dummy_mask,velec);
 992             velecsum         = _mm_add_ps(velecsum,velec);
 993
 994             fscal            = felec;
 995
 996             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 997
 998              /* Update vectorial force */
 999             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1000             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1001             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1002
1003             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1004             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1005             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1006
1007             /**************************
1008              * CALCULATE INTERACTIONS *
1009              **************************/
1010
1011             r11              = _mm_mul_ps(rsq11,rinv11);
1012             r11              = _mm_andnot_ps(dummy_mask,r11);
1013
1014             /* Calculate table index by multiplying r with table scale and truncate to integer */
1015             rt               = _mm_mul_ps(r11,vftabscale);
1016             vfitab           = _mm_cvttps_epi32(rt);
1017 #ifdef __XOP__
1018             vfeps            = _mm_frcz_ps(rt);
1019 #else
1020             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1021 #endif
1022             twovfeps         = _mm_add_ps(vfeps,vfeps);
1023             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1024
1025             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1026             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1027             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1028             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1029             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1030             _MM_TRANSPOSE4_PS(Y,F,G,H);
1031             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1032             VV               = _mm_macc_ps(vfeps,Fp,Y);
1033             velec            = _mm_mul_ps(qq11,VV);
1034             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1035             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq11,FF),_mm_mul_ps(vftabscale,rinv11)));
1036
1037             /* Update potential sum for this i atom from the interaction with this j atom. */
1038             velec            = _mm_andnot_ps(dummy_mask,velec);
1039             velecsum         = _mm_add_ps(velecsum,velec);
1040
1041             fscal            = felec;
1042
1043             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1044
1045              /* Update vectorial force */
1046             fix1             = _mm_macc_ps(dx11,fscal,fix1);
1047             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
1048             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
1049
1050             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
1051             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
1052             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
1053
1054             /**************************
1055              * CALCULATE INTERACTIONS *
1056              **************************/
1057
1058             r12              = _mm_mul_ps(rsq12,rinv12);
1059             r12              = _mm_andnot_ps(dummy_mask,r12);
1060
1061             /* Calculate table index by multiplying r with table scale and truncate to integer */
1062             rt               = _mm_mul_ps(r12,vftabscale);
1063             vfitab           = _mm_cvttps_epi32(rt);
1064 #ifdef __XOP__
1065             vfeps            = _mm_frcz_ps(rt);
1066 #else
1067             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1068 #endif
1069             twovfeps         = _mm_add_ps(vfeps,vfeps);
1070             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1071
1072             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1073             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1074             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1075             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1076             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1077             _MM_TRANSPOSE4_PS(Y,F,G,H);
1078             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1079             VV               = _mm_macc_ps(vfeps,Fp,Y);
1080             velec            = _mm_mul_ps(qq12,VV);
1081             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1082             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq12,FF),_mm_mul_ps(vftabscale,rinv12)));
1083
1084             /* Update potential sum for this i atom from the interaction with this j atom. */
1085             velec            = _mm_andnot_ps(dummy_mask,velec);
1086             velecsum         = _mm_add_ps(velecsum,velec);
1087
1088             fscal            = felec;
1089
1090             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1091
1092              /* Update vectorial force */
1093             fix1             = _mm_macc_ps(dx12,fscal,fix1);
1094             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
1095             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
1096
1097             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
1098             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
1099             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
1100
1101             /**************************
1102              * CALCULATE INTERACTIONS *
1103              **************************/
1104
1105             r20              = _mm_mul_ps(rsq20,rinv20);
1106             r20              = _mm_andnot_ps(dummy_mask,r20);
1107
1108             /* Calculate table index by multiplying r with table scale and truncate to integer */
1109             rt               = _mm_mul_ps(r20,vftabscale);
1110             vfitab           = _mm_cvttps_epi32(rt);
1111 #ifdef __XOP__
1112             vfeps            = _mm_frcz_ps(rt);
1113 #else
1114             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1115 #endif
1116             twovfeps         = _mm_add_ps(vfeps,vfeps);
1117             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1118
1119             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1120             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1121             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1122             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1123             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1124             _MM_TRANSPOSE4_PS(Y,F,G,H);
1125             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1126             VV               = _mm_macc_ps(vfeps,Fp,Y);
1127             velec            = _mm_mul_ps(qq20,VV);
1128             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1129             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1130
1131             /* Update potential sum for this i atom from the interaction with this j atom. */
1132             velec            = _mm_andnot_ps(dummy_mask,velec);
1133             velecsum         = _mm_add_ps(velecsum,velec);
1134
1135             fscal            = felec;
1136
1137             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1138
1139              /* Update vectorial force */
1140             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1141             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1142             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1143
1144             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1145             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1146             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1147
1148             /**************************
1149              * CALCULATE INTERACTIONS *
1150              **************************/
1151
1152             r21              = _mm_mul_ps(rsq21,rinv21);
1153             r21              = _mm_andnot_ps(dummy_mask,r21);
1154
1155             /* Calculate table index by multiplying r with table scale and truncate to integer */
1156             rt               = _mm_mul_ps(r21,vftabscale);
1157             vfitab           = _mm_cvttps_epi32(rt);
1158 #ifdef __XOP__
1159             vfeps            = _mm_frcz_ps(rt);
1160 #else
1161             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1162 #endif
1163             twovfeps         = _mm_add_ps(vfeps,vfeps);
1164             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1165
1166             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1167             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1168             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1169             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1170             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1171             _MM_TRANSPOSE4_PS(Y,F,G,H);
1172             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1173             VV               = _mm_macc_ps(vfeps,Fp,Y);
1174             velec            = _mm_mul_ps(qq21,VV);
1175             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1176             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq21,FF),_mm_mul_ps(vftabscale,rinv21)));
1177
1178             /* Update potential sum for this i atom from the interaction with this j atom. */
1179             velec            = _mm_andnot_ps(dummy_mask,velec);
1180             velecsum         = _mm_add_ps(velecsum,velec);
1181
1182             fscal            = felec;
1183
1184             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1185
1186              /* Update vectorial force */
1187             fix2             = _mm_macc_ps(dx21,fscal,fix2);
1188             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
1189             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
1190
1191             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
1192             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
1193             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
1194
1195             /**************************
1196              * CALCULATE INTERACTIONS *
1197              **************************/
1198
1199             r22              = _mm_mul_ps(rsq22,rinv22);
1200             r22              = _mm_andnot_ps(dummy_mask,r22);
1201
1202             /* Calculate table index by multiplying r with table scale and truncate to integer */
1203             rt               = _mm_mul_ps(r22,vftabscale);
1204             vfitab           = _mm_cvttps_epi32(rt);
1205 #ifdef __XOP__
1206             vfeps            = _mm_frcz_ps(rt);
1207 #else
1208             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1209 #endif
1210             twovfeps         = _mm_add_ps(vfeps,vfeps);
1211             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1212
1213             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1214             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1215             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1216             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1217             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1218             _MM_TRANSPOSE4_PS(Y,F,G,H);
1219             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1220             VV               = _mm_macc_ps(vfeps,Fp,Y);
1221             velec            = _mm_mul_ps(qq22,VV);
1222             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1223             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq22,FF),_mm_mul_ps(vftabscale,rinv22)));
1224
1225             /* Update potential sum for this i atom from the interaction with this j atom. */
1226             velec            = _mm_andnot_ps(dummy_mask,velec);
1227             velecsum         = _mm_add_ps(velecsum,velec);
1228
1229             fscal            = felec;
1230
1231             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1232
1233              /* Update vectorial force */
1234             fix2             = _mm_macc_ps(dx22,fscal,fix2);
1235             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
1236             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
1237
1238             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
1239             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
1240             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
1241
1242             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1243             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1244             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1245             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1246
1247             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1248                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1249
1250             /* Inner loop uses 453 flops */
1251         }
1252
1253         /* End of innermost loop */
1254
1255         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1256                                               f+i_coord_offset,fshift+i_shift_offset);
1257
1258         ggid                        = gid[iidx];
1259         /* Update potential energies */
1260         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1261         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1262
1263         /* Increment number of inner iterations */
1264         inneriter                  += j_index_end - j_index_start;
1265
1266         /* Outer loop uses 20 flops */
1267     }
1268
1269     /* Increment number of outer iterations */
1270     outeriter        += nri;
1271
1272     /* Update outer/inner flops */
1273
1274     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*453);
1275 }
1276 /*
1277  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3W3_F_avx_128_fma_single
1278  * Electrostatics interaction: CubicSplineTable
1279  * VdW interaction:            CubicSplineTable
1280  * Geometry:                   Water3-Water3
1281  * Calculate force/pot:        Force
1282  */
1283 void
1284 nb_kernel_ElecCSTab_VdwCSTab_GeomW3W3_F_avx_128_fma_single
1285                     (t_nblist * gmx_restrict                nlist,
1286                      rvec * gmx_restrict                    xx,
1287                      rvec * gmx_restrict                    ff,
1288                      t_forcerec * gmx_restrict              fr,
1289                      t_mdatoms * gmx_restrict               mdatoms,
1290                      nb_kernel_data_t * gmx_restrict        kernel_data,
1291                      t_nrnb * gmx_restrict                  nrnb)
1292 {
1293     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1294      * just 0 for non-waters.
1295      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
1296      * jnr indices corresponding to data put in the four positions in the SIMD register.
1297      */
1298     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1299     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1300     int              jnrA,jnrB,jnrC,jnrD;
1301     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1302     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1303     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1304     real             rcutoff_scalar;
1305     real             *shiftvec,*fshift,*x,*f;
1306     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1307     real             scratch[4*DIM];
1308     __m128           fscal,rcutoff,rcutoff2,jidxall;
1309     int              vdwioffset0;
1310     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1311     int              vdwioffset1;
1312     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1313     int              vdwioffset2;
1314     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1315     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1316     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1317     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1318     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1319     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1320     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1321     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1322     __m128           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1323     __m128           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1324     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1325     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1326     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1327     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1328     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1329     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1330     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
1331     real             *charge;
1332     int              nvdwtype;
1333     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1334     int              *vdwtype;
1335     real             *vdwparam;
1336     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
1337     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
1338     __m128i          vfitab;
1339     __m128i          ifour       = _mm_set1_epi32(4);
1340     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
1341     real             *vftab;
1342     __m128           dummy_mask,cutoff_mask;
1343     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
1344     __m128           one     = _mm_set1_ps(1.0);
1345     __m128           two     = _mm_set1_ps(2.0);
1346     x                = xx[0];
1347     f                = ff[0];
1348
1349     nri              = nlist->nri;
1350     iinr             = nlist->iinr;
1351     jindex           = nlist->jindex;
1352     jjnr             = nlist->jjnr;
1353     shiftidx         = nlist->shift;
1354     gid              = nlist->gid;
1355     shiftvec         = fr->shift_vec[0];
1356     fshift           = fr->fshift[0];
1357     facel            = _mm_set1_ps(fr->epsfac);
1358     charge           = mdatoms->chargeA;
1359     nvdwtype         = fr->ntype;
1360     vdwparam         = fr->nbfp;
1361     vdwtype          = mdatoms->typeA;
1362
1363     vftab            = kernel_data->table_elec_vdw->data;
1364     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
1365
1366     /* Setup water-specific parameters */
1367     inr              = nlist->iinr[0];
1368     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
1369     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
1370     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
1371     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
1372
1373     jq0              = _mm_set1_ps(charge[inr+0]);
1374     jq1              = _mm_set1_ps(charge[inr+1]);
1375     jq2              = _mm_set1_ps(charge[inr+2]);
1376     vdwjidx0A        = 2*vdwtype[inr+0];
1377     qq00             = _mm_mul_ps(iq0,jq0);
1378     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
1379     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
1380     qq01             = _mm_mul_ps(iq0,jq1);
1381     qq02             = _mm_mul_ps(iq0,jq2);
1382     qq10             = _mm_mul_ps(iq1,jq0);
1383     qq11             = _mm_mul_ps(iq1,jq1);
1384     qq12             = _mm_mul_ps(iq1,jq2);
1385     qq20             = _mm_mul_ps(iq2,jq0);
1386     qq21             = _mm_mul_ps(iq2,jq1);
1387     qq22             = _mm_mul_ps(iq2,jq2);
1388
1389     /* Avoid stupid compiler warnings */
1390     jnrA = jnrB = jnrC = jnrD = 0;
1391     j_coord_offsetA = 0;
1392     j_coord_offsetB = 0;
1393     j_coord_offsetC = 0;
1394     j_coord_offsetD = 0;
1395
1396     outeriter        = 0;
1397     inneriter        = 0;
1398
1399     for(iidx=0;iidx<4*DIM;iidx++)
1400     {
1401         scratch[iidx] = 0.0;
1402     }
1403
1404     /* Start outer loop over neighborlists */
1405     for(iidx=0; iidx<nri; iidx++)
1406     {
1407         /* Load shift vector for this list */
1408         i_shift_offset   = DIM*shiftidx[iidx];
1409
1410         /* Load limits for loop over neighbors */
1411         j_index_start    = jindex[iidx];
1412         j_index_end      = jindex[iidx+1];
1413
1414         /* Get outer coordinate index */
1415         inr              = iinr[iidx];
1416         i_coord_offset   = DIM*inr;
1417
1418         /* Load i particle coords and add shift vector */
1419         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1420                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1421
1422         fix0             = _mm_setzero_ps();
1423         fiy0             = _mm_setzero_ps();
1424         fiz0             = _mm_setzero_ps();
1425         fix1             = _mm_setzero_ps();
1426         fiy1             = _mm_setzero_ps();
1427         fiz1             = _mm_setzero_ps();
1428         fix2             = _mm_setzero_ps();
1429         fiy2             = _mm_setzero_ps();
1430         fiz2             = _mm_setzero_ps();
1431
1432         /* Start inner kernel loop */
1433         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1434         {
1435
1436             /* Get j neighbor index, and coordinate index */
1437             jnrA             = jjnr[jidx];
1438             jnrB             = jjnr[jidx+1];
1439             jnrC             = jjnr[jidx+2];
1440             jnrD             = jjnr[jidx+3];
1441             j_coord_offsetA  = DIM*jnrA;
1442             j_coord_offsetB  = DIM*jnrB;
1443             j_coord_offsetC  = DIM*jnrC;
1444             j_coord_offsetD  = DIM*jnrD;
1445
1446             /* load j atom coordinates */
1447             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1448                                               x+j_coord_offsetC,x+j_coord_offsetD,
1449                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1450
1451             /* Calculate displacement vector */
1452             dx00             = _mm_sub_ps(ix0,jx0);
1453             dy00             = _mm_sub_ps(iy0,jy0);
1454             dz00             = _mm_sub_ps(iz0,jz0);
1455             dx01             = _mm_sub_ps(ix0,jx1);
1456             dy01             = _mm_sub_ps(iy0,jy1);
1457             dz01             = _mm_sub_ps(iz0,jz1);
1458             dx02             = _mm_sub_ps(ix0,jx2);
1459             dy02             = _mm_sub_ps(iy0,jy2);
1460             dz02             = _mm_sub_ps(iz0,jz2);
1461             dx10             = _mm_sub_ps(ix1,jx0);
1462             dy10             = _mm_sub_ps(iy1,jy0);
1463             dz10             = _mm_sub_ps(iz1,jz0);
1464             dx11             = _mm_sub_ps(ix1,jx1);
1465             dy11             = _mm_sub_ps(iy1,jy1);
1466             dz11             = _mm_sub_ps(iz1,jz1);
1467             dx12             = _mm_sub_ps(ix1,jx2);
1468             dy12             = _mm_sub_ps(iy1,jy2);
1469             dz12             = _mm_sub_ps(iz1,jz2);
1470             dx20             = _mm_sub_ps(ix2,jx0);
1471             dy20             = _mm_sub_ps(iy2,jy0);
1472             dz20             = _mm_sub_ps(iz2,jz0);
1473             dx21             = _mm_sub_ps(ix2,jx1);
1474             dy21             = _mm_sub_ps(iy2,jy1);
1475             dz21             = _mm_sub_ps(iz2,jz1);
1476             dx22             = _mm_sub_ps(ix2,jx2);
1477             dy22             = _mm_sub_ps(iy2,jy2);
1478             dz22             = _mm_sub_ps(iz2,jz2);
1479
1480             /* Calculate squared distance and things based on it */
1481             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1482             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
1483             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
1484             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1485             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1486             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1487             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1488             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1489             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1490
1491             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1492             rinv01           = gmx_mm_invsqrt_ps(rsq01);
1493             rinv02           = gmx_mm_invsqrt_ps(rsq02);
1494             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1495             rinv11           = gmx_mm_invsqrt_ps(rsq11);
1496             rinv12           = gmx_mm_invsqrt_ps(rsq12);
1497             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1498             rinv21           = gmx_mm_invsqrt_ps(rsq21);
1499             rinv22           = gmx_mm_invsqrt_ps(rsq22);
1500
1501             fjx0             = _mm_setzero_ps();
1502             fjy0             = _mm_setzero_ps();
1503             fjz0             = _mm_setzero_ps();
1504             fjx1             = _mm_setzero_ps();
1505             fjy1             = _mm_setzero_ps();
1506             fjz1             = _mm_setzero_ps();
1507             fjx2             = _mm_setzero_ps();
1508             fjy2             = _mm_setzero_ps();
1509             fjz2             = _mm_setzero_ps();
1510
1511             /**************************
1512              * CALCULATE INTERACTIONS *
1513              **************************/
1514
1515             r00              = _mm_mul_ps(rsq00,rinv00);
1516
1517             /* Calculate table index by multiplying r with table scale and truncate to integer */
1518             rt               = _mm_mul_ps(r00,vftabscale);
1519             vfitab           = _mm_cvttps_epi32(rt);
1520 #ifdef __XOP__
1521             vfeps            = _mm_frcz_ps(rt);
1522 #else
1523             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1524 #endif
1525             twovfeps         = _mm_add_ps(vfeps,vfeps);
1526             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1527
1528             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1529             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1530             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1531             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1532             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1533             _MM_TRANSPOSE4_PS(Y,F,G,H);
1534             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1535             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1536             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
1537
1538             /* CUBIC SPLINE TABLE DISPERSION */
1539             vfitab           = _mm_add_epi32(vfitab,ifour);
1540             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1541             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1542             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1543             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1544             _MM_TRANSPOSE4_PS(Y,F,G,H);
1545             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1546             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1547             fvdw6            = _mm_mul_ps(c6_00,FF);
1548
1549             /* CUBIC SPLINE TABLE REPULSION */
1550             vfitab           = _mm_add_epi32(vfitab,ifour);
1551             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1552             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1553             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1554             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1555             _MM_TRANSPOSE4_PS(Y,F,G,H);
1556             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1557             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1558             fvdw12           = _mm_mul_ps(c12_00,FF);
1559             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1560
1561             fscal            = _mm_add_ps(felec,fvdw);
1562
1563              /* Update vectorial force */
1564             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1565             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1566             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1567
1568             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1569             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1570             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1571
1572             /**************************
1573              * CALCULATE INTERACTIONS *
1574              **************************/
1575
1576             r01              = _mm_mul_ps(rsq01,rinv01);
1577
1578             /* Calculate table index by multiplying r with table scale and truncate to integer */
1579             rt               = _mm_mul_ps(r01,vftabscale);
1580             vfitab           = _mm_cvttps_epi32(rt);
1581 #ifdef __XOP__
1582             vfeps            = _mm_frcz_ps(rt);
1583 #else
1584             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1585 #endif
1586             twovfeps         = _mm_add_ps(vfeps,vfeps);
1587             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1588
1589             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1590             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1591             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1592             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1593             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1594             _MM_TRANSPOSE4_PS(Y,F,G,H);
1595             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1596             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1597             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq01,FF),_mm_mul_ps(vftabscale,rinv01)));
1598
1599             fscal            = felec;
1600
1601              /* Update vectorial force */
1602             fix0             = _mm_macc_ps(dx01,fscal,fix0);
1603             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
1604             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
1605
1606             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
1607             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
1608             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
1609
1610             /**************************
1611              * CALCULATE INTERACTIONS *
1612              **************************/
1613
1614             r02              = _mm_mul_ps(rsq02,rinv02);
1615
1616             /* Calculate table index by multiplying r with table scale and truncate to integer */
1617             rt               = _mm_mul_ps(r02,vftabscale);
1618             vfitab           = _mm_cvttps_epi32(rt);
1619 #ifdef __XOP__
1620             vfeps            = _mm_frcz_ps(rt);
1621 #else
1622             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1623 #endif
1624             twovfeps         = _mm_add_ps(vfeps,vfeps);
1625             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1626
1627             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1628             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1629             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1630             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1631             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1632             _MM_TRANSPOSE4_PS(Y,F,G,H);
1633             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1634             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1635             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq02,FF),_mm_mul_ps(vftabscale,rinv02)));
1636
1637             fscal            = felec;
1638
1639              /* Update vectorial force */
1640             fix0             = _mm_macc_ps(dx02,fscal,fix0);
1641             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
1642             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
1643
1644             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
1645             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
1646             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
1647
1648             /**************************
1649              * CALCULATE INTERACTIONS *
1650              **************************/
1651
1652             r10              = _mm_mul_ps(rsq10,rinv10);
1653
1654             /* Calculate table index by multiplying r with table scale and truncate to integer */
1655             rt               = _mm_mul_ps(r10,vftabscale);
1656             vfitab           = _mm_cvttps_epi32(rt);
1657 #ifdef __XOP__
1658             vfeps            = _mm_frcz_ps(rt);
1659 #else
1660             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1661 #endif
1662             twovfeps         = _mm_add_ps(vfeps,vfeps);
1663             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1664
1665             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1666             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1667             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1668             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1669             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1670             _MM_TRANSPOSE4_PS(Y,F,G,H);
1671             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1672             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1673             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1674
1675             fscal            = felec;
1676
1677              /* Update vectorial force */
1678             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1679             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1680             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1681
1682             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1683             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1684             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1685
1686             /**************************
1687              * CALCULATE INTERACTIONS *
1688              **************************/
1689
1690             r11              = _mm_mul_ps(rsq11,rinv11);
1691
1692             /* Calculate table index by multiplying r with table scale and truncate to integer */
1693             rt               = _mm_mul_ps(r11,vftabscale);
1694             vfitab           = _mm_cvttps_epi32(rt);
1695 #ifdef __XOP__
1696             vfeps            = _mm_frcz_ps(rt);
1697 #else
1698             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1699 #endif
1700             twovfeps         = _mm_add_ps(vfeps,vfeps);
1701             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1702
1703             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1704             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1705             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1706             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1707             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1708             _MM_TRANSPOSE4_PS(Y,F,G,H);
1709             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1710             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1711             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq11,FF),_mm_mul_ps(vftabscale,rinv11)));
1712
1713             fscal            = felec;
1714
1715              /* Update vectorial force */
1716             fix1             = _mm_macc_ps(dx11,fscal,fix1);
1717             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
1718             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
1719
1720             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
1721             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
1722             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
1723
1724             /**************************
1725              * CALCULATE INTERACTIONS *
1726              **************************/
1727
1728             r12              = _mm_mul_ps(rsq12,rinv12);
1729
1730             /* Calculate table index by multiplying r with table scale and truncate to integer */
1731             rt               = _mm_mul_ps(r12,vftabscale);
1732             vfitab           = _mm_cvttps_epi32(rt);
1733 #ifdef __XOP__
1734             vfeps            = _mm_frcz_ps(rt);
1735 #else
1736             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1737 #endif
1738             twovfeps         = _mm_add_ps(vfeps,vfeps);
1739             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1740
1741             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1742             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1743             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1744             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1745             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1746             _MM_TRANSPOSE4_PS(Y,F,G,H);
1747             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1748             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1749             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq12,FF),_mm_mul_ps(vftabscale,rinv12)));
1750
1751             fscal            = felec;
1752
1753              /* Update vectorial force */
1754             fix1             = _mm_macc_ps(dx12,fscal,fix1);
1755             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
1756             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
1757
1758             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
1759             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
1760             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
1761
1762             /**************************
1763              * CALCULATE INTERACTIONS *
1764              **************************/
1765
1766             r20              = _mm_mul_ps(rsq20,rinv20);
1767
1768             /* Calculate table index by multiplying r with table scale and truncate to integer */
1769             rt               = _mm_mul_ps(r20,vftabscale);
1770             vfitab           = _mm_cvttps_epi32(rt);
1771 #ifdef __XOP__
1772             vfeps            = _mm_frcz_ps(rt);
1773 #else
1774             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1775 #endif
1776             twovfeps         = _mm_add_ps(vfeps,vfeps);
1777             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1778
1779             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1780             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1781             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1782             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1783             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1784             _MM_TRANSPOSE4_PS(Y,F,G,H);
1785             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1786             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1787             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1788
1789             fscal            = felec;
1790
1791              /* Update vectorial force */
1792             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1793             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1794             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1795
1796             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1797             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1798             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1799
1800             /**************************
1801              * CALCULATE INTERACTIONS *
1802              **************************/
1803
1804             r21              = _mm_mul_ps(rsq21,rinv21);
1805
1806             /* Calculate table index by multiplying r with table scale and truncate to integer */
1807             rt               = _mm_mul_ps(r21,vftabscale);
1808             vfitab           = _mm_cvttps_epi32(rt);
1809 #ifdef __XOP__
1810             vfeps            = _mm_frcz_ps(rt);
1811 #else
1812             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1813 #endif
1814             twovfeps         = _mm_add_ps(vfeps,vfeps);
1815             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1816
1817             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1818             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1819             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1820             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1821             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1822             _MM_TRANSPOSE4_PS(Y,F,G,H);
1823             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1824             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1825             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq21,FF),_mm_mul_ps(vftabscale,rinv21)));
1826
1827             fscal            = felec;
1828
1829              /* Update vectorial force */
1830             fix2             = _mm_macc_ps(dx21,fscal,fix2);
1831             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
1832             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
1833
1834             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
1835             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
1836             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
1837
1838             /**************************
1839              * CALCULATE INTERACTIONS *
1840              **************************/
1841
1842             r22              = _mm_mul_ps(rsq22,rinv22);
1843
1844             /* Calculate table index by multiplying r with table scale and truncate to integer */
1845             rt               = _mm_mul_ps(r22,vftabscale);
1846             vfitab           = _mm_cvttps_epi32(rt);
1847 #ifdef __XOP__
1848             vfeps            = _mm_frcz_ps(rt);
1849 #else
1850             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1851 #endif
1852             twovfeps         = _mm_add_ps(vfeps,vfeps);
1853             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1854
1855             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1856             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1857             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1858             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1859             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1860             _MM_TRANSPOSE4_PS(Y,F,G,H);
1861             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1862             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1863             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq22,FF),_mm_mul_ps(vftabscale,rinv22)));
1864
1865             fscal            = felec;
1866
1867              /* Update vectorial force */
1868             fix2             = _mm_macc_ps(dx22,fscal,fix2);
1869             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
1870             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
1871
1872             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
1873             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
1874             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
1875
1876             fjptrA             = f+j_coord_offsetA;
1877             fjptrB             = f+j_coord_offsetB;
1878             fjptrC             = f+j_coord_offsetC;
1879             fjptrD             = f+j_coord_offsetD;
1880
1881             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1882                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1883
1884             /* Inner loop uses 400 flops */
1885         }
1886
1887         if(jidx<j_index_end)
1888         {
1889
1890             /* Get j neighbor index, and coordinate index */
1891             jnrlistA         = jjnr[jidx];
1892             jnrlistB         = jjnr[jidx+1];
1893             jnrlistC         = jjnr[jidx+2];
1894             jnrlistD         = jjnr[jidx+3];
1895             /* Sign of each element will be negative for non-real atoms.
1896              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1897              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1898              */
1899             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1900             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1901             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1902             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1903             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1904             j_coord_offsetA  = DIM*jnrA;
1905             j_coord_offsetB  = DIM*jnrB;
1906             j_coord_offsetC  = DIM*jnrC;
1907             j_coord_offsetD  = DIM*jnrD;
1908
1909             /* load j atom coordinates */
1910             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1911                                               x+j_coord_offsetC,x+j_coord_offsetD,
1912                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1913
1914             /* Calculate displacement vector */
1915             dx00             = _mm_sub_ps(ix0,jx0);
1916             dy00             = _mm_sub_ps(iy0,jy0);
1917             dz00             = _mm_sub_ps(iz0,jz0);
1918             dx01             = _mm_sub_ps(ix0,jx1);
1919             dy01             = _mm_sub_ps(iy0,jy1);
1920             dz01             = _mm_sub_ps(iz0,jz1);
1921             dx02             = _mm_sub_ps(ix0,jx2);
1922             dy02             = _mm_sub_ps(iy0,jy2);
1923             dz02             = _mm_sub_ps(iz0,jz2);
1924             dx10             = _mm_sub_ps(ix1,jx0);
1925             dy10             = _mm_sub_ps(iy1,jy0);
1926             dz10             = _mm_sub_ps(iz1,jz0);
1927             dx11             = _mm_sub_ps(ix1,jx1);
1928             dy11             = _mm_sub_ps(iy1,jy1);
1929             dz11             = _mm_sub_ps(iz1,jz1);
1930             dx12             = _mm_sub_ps(ix1,jx2);
1931             dy12             = _mm_sub_ps(iy1,jy2);
1932             dz12             = _mm_sub_ps(iz1,jz2);
1933             dx20             = _mm_sub_ps(ix2,jx0);
1934             dy20             = _mm_sub_ps(iy2,jy0);
1935             dz20             = _mm_sub_ps(iz2,jz0);
1936             dx21             = _mm_sub_ps(ix2,jx1);
1937             dy21             = _mm_sub_ps(iy2,jy1);
1938             dz21             = _mm_sub_ps(iz2,jz1);
1939             dx22             = _mm_sub_ps(ix2,jx2);
1940             dy22             = _mm_sub_ps(iy2,jy2);
1941             dz22             = _mm_sub_ps(iz2,jz2);
1942
1943             /* Calculate squared distance and things based on it */
1944             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1945             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
1946             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
1947             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1948             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1949             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1950             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1951             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1952             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1953
1954             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1955             rinv01           = gmx_mm_invsqrt_ps(rsq01);
1956             rinv02           = gmx_mm_invsqrt_ps(rsq02);
1957             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1958             rinv11           = gmx_mm_invsqrt_ps(rsq11);
1959             rinv12           = gmx_mm_invsqrt_ps(rsq12);
1960             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1961             rinv21           = gmx_mm_invsqrt_ps(rsq21);
1962             rinv22           = gmx_mm_invsqrt_ps(rsq22);
1963
1964             fjx0             = _mm_setzero_ps();
1965             fjy0             = _mm_setzero_ps();
1966             fjz0             = _mm_setzero_ps();
1967             fjx1             = _mm_setzero_ps();
1968             fjy1             = _mm_setzero_ps();
1969             fjz1             = _mm_setzero_ps();
1970             fjx2             = _mm_setzero_ps();
1971             fjy2             = _mm_setzero_ps();
1972             fjz2             = _mm_setzero_ps();
1973
1974             /**************************
1975              * CALCULATE INTERACTIONS *
1976              **************************/
1977
1978             r00              = _mm_mul_ps(rsq00,rinv00);
1979             r00              = _mm_andnot_ps(dummy_mask,r00);
1980
1981             /* Calculate table index by multiplying r with table scale and truncate to integer */
1982             rt               = _mm_mul_ps(r00,vftabscale);
1983             vfitab           = _mm_cvttps_epi32(rt);
1984 #ifdef __XOP__
1985             vfeps            = _mm_frcz_ps(rt);
1986 #else
1987             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1988 #endif
1989             twovfeps         = _mm_add_ps(vfeps,vfeps);
1990             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1991
1992             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1993             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1994             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1995             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1996             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1997             _MM_TRANSPOSE4_PS(Y,F,G,H);
1998             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1999             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
2000             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
2001
2002             /* CUBIC SPLINE TABLE DISPERSION */
2003             vfitab           = _mm_add_epi32(vfitab,ifour);
2004             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
2005             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
2006             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
2007             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
2008             _MM_TRANSPOSE4_PS(Y,F,G,H);
2009             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
2010             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
2011             fvdw6            = _mm_mul_ps(c6_00,FF);
2012
2013             /* CUBIC SPLINE TABLE REPULSION */
2014             vfitab           = _mm_add_epi32(vfitab,ifour);
2015             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
2016             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
2017             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
2018             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
2019             _MM_TRANSPOSE4_PS(Y,F,G,H);
2020             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
2021             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
2022             fvdw12           = _mm_mul_ps(c12_00,FF);
2023             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
2024
2025             fscal            = _mm_add_ps(felec,fvdw);
2026
2027             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2028
2029              /* Update vectorial force */
2030             fix0             = _mm_macc_ps(dx00,fscal,fix0);
2031             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
2032             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
2033
2034             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
2035             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
2036             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
2037
2038             /**************************
2039              * CALCULATE INTERACTIONS *
2040              **************************/
2041
2042             r01              = _mm_mul_ps(rsq01,rinv01);
2043             r01              = _mm_andnot_ps(dummy_mask,r01);
2044
2045             /* Calculate table index by multiplying r with table scale and truncate to integer */
2046             rt               = _mm_mul_ps(r01,vftabscale);
2047             vfitab           = _mm_cvttps_epi32(rt);
2048 #ifdef __XOP__
2049             vfeps            = _mm_frcz_ps(rt);
2050 #else
2051             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
2052 #endif
2053             twovfeps         = _mm_add_ps(vfeps,vfeps);
2054             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
2055
2056             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2057             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
2058             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
2059             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
2060             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
2061             _MM_TRANSPOSE4_PS(Y,F,G,H);
2062             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
2063             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
2064             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq01,FF),_mm_mul_ps(vftabscale,rinv01)));
2065
2066             fscal            = felec;
2067
2068             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2069
2070              /* Update vectorial force */
2071             fix0             = _mm_macc_ps(dx01,fscal,fix0);
2072             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
2073             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
2074
2075             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
2076             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
2077             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
2078
2079             /**************************
2080              * CALCULATE INTERACTIONS *
2081              **************************/
2082
2083             r02              = _mm_mul_ps(rsq02,rinv02);
2084             r02              = _mm_andnot_ps(dummy_mask,r02);
2085
2086             /* Calculate table index by multiplying r with table scale and truncate to integer */
2087             rt               = _mm_mul_ps(r02,vftabscale);
2088             vfitab           = _mm_cvttps_epi32(rt);
2089 #ifdef __XOP__
2090             vfeps            = _mm_frcz_ps(rt);
2091 #else
2092             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
2093 #endif
2094             twovfeps         = _mm_add_ps(vfeps,vfeps);
2095             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
2096
2097             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2098             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
2099             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
2100             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
2101             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
2102             _MM_TRANSPOSE4_PS(Y,F,G,H);
2103             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
2104             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
2105             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq02,FF),_mm_mul_ps(vftabscale,rinv02)));
2106
2107             fscal            = felec;
2108
2109             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2110
2111              /* Update vectorial force */
2112             fix0             = _mm_macc_ps(dx02,fscal,fix0);
2113             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
2114             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
2115
2116             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
2117             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
2118             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
2119
2120             /**************************
2121              * CALCULATE INTERACTIONS *
2122              **************************/
2123
2124             r10              = _mm_mul_ps(rsq10,rinv10);
2125             r10              = _mm_andnot_ps(dummy_mask,r10);
2126
2127             /* Calculate table index by multiplying r with table scale and truncate to integer */
2128             rt               = _mm_mul_ps(r10,vftabscale);
2129             vfitab           = _mm_cvttps_epi32(rt);
2130 #ifdef __XOP__
2131             vfeps            = _mm_frcz_ps(rt);
2132 #else
2133             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
2134 #endif
2135             twovfeps         = _mm_add_ps(vfeps,vfeps);
2136             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
2137
2138             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2139             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
2140             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
2141             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
2142             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
2143             _MM_TRANSPOSE4_PS(Y,F,G,H);
2144             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
2145             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
2146             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
2147
2148             fscal            = felec;
2149
2150             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2151
2152              /* Update vectorial force */
2153             fix1             = _mm_macc_ps(dx10,fscal,fix1);
2154             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
2155             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
2156
2157             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
2158             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
2159             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
2160
2161             /**************************
2162              * CALCULATE INTERACTIONS *
2163              **************************/
2164
2165             r11              = _mm_mul_ps(rsq11,rinv11);
2166             r11              = _mm_andnot_ps(dummy_mask,r11);
2167
2168             /* Calculate table index by multiplying r with table scale and truncate to integer */
2169             rt               = _mm_mul_ps(r11,vftabscale);
2170             vfitab           = _mm_cvttps_epi32(rt);
2171 #ifdef __XOP__
2172             vfeps            = _mm_frcz_ps(rt);
2173 #else
2174             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
2175 #endif
2176             twovfeps         = _mm_add_ps(vfeps,vfeps);
2177             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
2178
2179             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2180             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
2181             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
2182             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
2183             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
2184             _MM_TRANSPOSE4_PS(Y,F,G,H);
2185             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
2186             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
2187             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq11,FF),_mm_mul_ps(vftabscale,rinv11)));
2188
2189             fscal            = felec;
2190
2191             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2192
2193              /* Update vectorial force */
2194             fix1             = _mm_macc_ps(dx11,fscal,fix1);
2195             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
2196             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
2197
2198             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
2199             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
2200             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
2201
2202             /**************************
2203              * CALCULATE INTERACTIONS *
2204              **************************/
2205
2206             r12              = _mm_mul_ps(rsq12,rinv12);
2207             r12              = _mm_andnot_ps(dummy_mask,r12);
2208
2209             /* Calculate table index by multiplying r with table scale and truncate to integer */
2210             rt               = _mm_mul_ps(r12,vftabscale);
2211             vfitab           = _mm_cvttps_epi32(rt);
2212 #ifdef __XOP__
2213             vfeps            = _mm_frcz_ps(rt);
2214 #else
2215             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
2216 #endif
2217             twovfeps         = _mm_add_ps(vfeps,vfeps);
2218             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
2219
2220             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2221             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
2222             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
2223             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
2224             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
2225             _MM_TRANSPOSE4_PS(Y,F,G,H);
2226             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
2227             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
2228             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq12,FF),_mm_mul_ps(vftabscale,rinv12)));
2229
2230             fscal            = felec;
2231
2232             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2233
2234              /* Update vectorial force */
2235             fix1             = _mm_macc_ps(dx12,fscal,fix1);
2236             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
2237             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
2238
2239             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
2240             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
2241             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
2242
2243             /**************************
2244              * CALCULATE INTERACTIONS *
2245              **************************/
2246
2247             r20              = _mm_mul_ps(rsq20,rinv20);
2248             r20              = _mm_andnot_ps(dummy_mask,r20);
2249
2250             /* Calculate table index by multiplying r with table scale and truncate to integer */
2251             rt               = _mm_mul_ps(r20,vftabscale);
2252             vfitab           = _mm_cvttps_epi32(rt);
2253 #ifdef __XOP__
2254             vfeps            = _mm_frcz_ps(rt);
2255 #else
2256             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
2257 #endif
2258             twovfeps         = _mm_add_ps(vfeps,vfeps);
2259             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
2260
2261             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2262             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
2263             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
2264             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
2265             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
2266             _MM_TRANSPOSE4_PS(Y,F,G,H);
2267             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
2268             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
2269             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
2270
2271             fscal            = felec;
2272
2273             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2274
2275              /* Update vectorial force */
2276             fix2             = _mm_macc_ps(dx20,fscal,fix2);
2277             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
2278             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
2279
2280             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
2281             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
2282             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
2283
2284             /**************************
2285              * CALCULATE INTERACTIONS *
2286              **************************/
2287
2288             r21              = _mm_mul_ps(rsq21,rinv21);
2289             r21              = _mm_andnot_ps(dummy_mask,r21);
2290
2291             /* Calculate table index by multiplying r with table scale and truncate to integer */
2292             rt               = _mm_mul_ps(r21,vftabscale);
2293             vfitab           = _mm_cvttps_epi32(rt);
2294 #ifdef __XOP__
2295             vfeps            = _mm_frcz_ps(rt);
2296 #else
2297             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
2298 #endif
2299             twovfeps         = _mm_add_ps(vfeps,vfeps);
2300             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
2301
2302             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2303             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
2304             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
2305             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
2306             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
2307             _MM_TRANSPOSE4_PS(Y,F,G,H);
2308             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
2309             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
2310             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq21,FF),_mm_mul_ps(vftabscale,rinv21)));
2311
2312             fscal            = felec;
2313
2314             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2315
2316              /* Update vectorial force */
2317             fix2             = _mm_macc_ps(dx21,fscal,fix2);
2318             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
2319             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
2320
2321             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
2322             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
2323             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
2324
2325             /**************************
2326              * CALCULATE INTERACTIONS *
2327              **************************/
2328
2329             r22              = _mm_mul_ps(rsq22,rinv22);
2330             r22              = _mm_andnot_ps(dummy_mask,r22);
2331
2332             /* Calculate table index by multiplying r with table scale and truncate to integer */
2333             rt               = _mm_mul_ps(r22,vftabscale);
2334             vfitab           = _mm_cvttps_epi32(rt);
2335 #ifdef __XOP__
2336             vfeps            = _mm_frcz_ps(rt);
2337 #else
2338             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
2339 #endif
2340             twovfeps         = _mm_add_ps(vfeps,vfeps);
2341             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
2342
2343             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2344             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
2345             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
2346             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
2347             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
2348             _MM_TRANSPOSE4_PS(Y,F,G,H);
2349             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
2350             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
2351             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq22,FF),_mm_mul_ps(vftabscale,rinv22)));
2352
2353             fscal            = felec;
2354
2355             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2356
2357              /* Update vectorial force */
2358             fix2             = _mm_macc_ps(dx22,fscal,fix2);
2359             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
2360             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
2361
2362             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
2363             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
2364             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
2365
2366             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2367             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2368             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2369             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2370
2371             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
2372                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2373
2374             /* Inner loop uses 409 flops */
2375         }
2376
2377         /* End of innermost loop */
2378
2379         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2380                                               f+i_coord_offset,fshift+i_shift_offset);
2381
2382         /* Increment number of inner iterations */
2383         inneriter                  += j_index_end - j_index_start;
2384
2385         /* Outer loop uses 18 flops */
2386     }
2387
2388     /* Increment number of outer iterations */
2389     outeriter        += nri;
2390
2391     /* Update outer/inner flops */
2392
2393     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*409);
2394 }