src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_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_GeomW3P1_VF_avx_128_fma_single
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_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     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  78     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  79     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  80     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     int              nvdwtype;
  83     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  84     int              *vdwtype;
  85     real             *vdwparam;
  86     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  87     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  88     __m128i          vfitab;
  89     __m128i          ifour       = _mm_set1_epi32(4);
  90     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
  91     real             *vftab;
  92     __m128           dummy_mask,cutoff_mask;
  93     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  94     __m128           one     = _mm_set1_ps(1.0);
  95     __m128           two     = _mm_set1_ps(2.0);
  96     x                = xx[0];
  97     f                = ff[0];
  98
  99     nri              = nlist->nri;
 100     iinr             = nlist->iinr;
 101     jindex           = nlist->jindex;
 102     jjnr             = nlist->jjnr;
 103     shiftidx         = nlist->shift;
 104     gid              = nlist->gid;
 105     shiftvec         = fr->shift_vec[0];
 106     fshift           = fr->fshift[0];
 107     facel            = _mm_set1_ps(fr->epsfac);
 108     charge           = mdatoms->chargeA;
 109     nvdwtype         = fr->ntype;
 110     vdwparam         = fr->nbfp;
 111     vdwtype          = mdatoms->typeA;
 112
 113     vftab            = kernel_data->table_elec_vdw->data;
 114     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 115
 116     /* Setup water-specific parameters */
 117     inr              = nlist->iinr[0];
 118     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 119     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 120     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 121     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 122
 123     /* Avoid stupid compiler warnings */
 124     jnrA = jnrB = jnrC = jnrD = 0;
 125     j_coord_offsetA = 0;
 126     j_coord_offsetB = 0;
 127     j_coord_offsetC = 0;
 128     j_coord_offsetD = 0;
 129
 130     outeriter        = 0;
 131     inneriter        = 0;
 132
 133     for(iidx=0;iidx<4*DIM;iidx++)
 134     {
 135         scratch[iidx] = 0.0;
 136     }
 137
 138     /* Start outer loop over neighborlists */
 139     for(iidx=0; iidx<nri; iidx++)
 140     {
 141         /* Load shift vector for this list */
 142         i_shift_offset   = DIM*shiftidx[iidx];
 143
 144         /* Load limits for loop over neighbors */
 145         j_index_start    = jindex[iidx];
 146         j_index_end      = jindex[iidx+1];
 147
 148         /* Get outer coordinate index */
 149         inr              = iinr[iidx];
 150         i_coord_offset   = DIM*inr;
 151
 152         /* Load i particle coords and add shift vector */
 153         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 154                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 155
 156         fix0             = _mm_setzero_ps();
 157         fiy0             = _mm_setzero_ps();
 158         fiz0             = _mm_setzero_ps();
 159         fix1             = _mm_setzero_ps();
 160         fiy1             = _mm_setzero_ps();
 161         fiz1             = _mm_setzero_ps();
 162         fix2             = _mm_setzero_ps();
 163         fiy2             = _mm_setzero_ps();
 164         fiz2             = _mm_setzero_ps();
 165
 166         /* Reset potential sums */
 167         velecsum         = _mm_setzero_ps();
 168         vvdwsum          = _mm_setzero_ps();
 169
 170         /* Start inner kernel loop */
 171         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 172         {
 173
 174             /* Get j neighbor index, and coordinate index */
 175             jnrA             = jjnr[jidx];
 176             jnrB             = jjnr[jidx+1];
 177             jnrC             = jjnr[jidx+2];
 178             jnrD             = jjnr[jidx+3];
 179             j_coord_offsetA  = DIM*jnrA;
 180             j_coord_offsetB  = DIM*jnrB;
 181             j_coord_offsetC  = DIM*jnrC;
 182             j_coord_offsetD  = DIM*jnrD;
 183
 184             /* load j atom coordinates */
 185             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 186                                               x+j_coord_offsetC,x+j_coord_offsetD,
 187                                               &jx0,&jy0,&jz0);
 188
 189             /* Calculate displacement vector */
 190             dx00             = _mm_sub_ps(ix0,jx0);
 191             dy00             = _mm_sub_ps(iy0,jy0);
 192             dz00             = _mm_sub_ps(iz0,jz0);
 193             dx10             = _mm_sub_ps(ix1,jx0);
 194             dy10             = _mm_sub_ps(iy1,jy0);
 195             dz10             = _mm_sub_ps(iz1,jz0);
 196             dx20             = _mm_sub_ps(ix2,jx0);
 197             dy20             = _mm_sub_ps(iy2,jy0);
 198             dz20             = _mm_sub_ps(iz2,jz0);
 199
 200             /* Calculate squared distance and things based on it */
 201             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 202             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 203             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 204
 205             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 206             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 207             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 208
 209             /* Load parameters for j particles */
 210             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 211                                                               charge+jnrC+0,charge+jnrD+0);
 212             vdwjidx0A        = 2*vdwtype[jnrA+0];
 213             vdwjidx0B        = 2*vdwtype[jnrB+0];
 214             vdwjidx0C        = 2*vdwtype[jnrC+0];
 215             vdwjidx0D        = 2*vdwtype[jnrD+0];
 216
 217             fjx0             = _mm_setzero_ps();
 218             fjy0             = _mm_setzero_ps();
 219             fjz0             = _mm_setzero_ps();
 220
 221             /**************************
 222              * CALCULATE INTERACTIONS *
 223              **************************/
 224
 225             r00              = _mm_mul_ps(rsq00,rinv00);
 226
 227             /* Compute parameters for interactions between i and j atoms */
 228             qq00             = _mm_mul_ps(iq0,jq0);
 229             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 230                                          vdwparam+vdwioffset0+vdwjidx0B,
 231                                          vdwparam+vdwioffset0+vdwjidx0C,
 232                                          vdwparam+vdwioffset0+vdwjidx0D,
 233                                          &c6_00,&c12_00);
 234
 235             /* Calculate table index by multiplying r with table scale and truncate to integer */
 236             rt               = _mm_mul_ps(r00,vftabscale);
 237             vfitab           = _mm_cvttps_epi32(rt);
 238 #ifdef __XOP__
 239             vfeps            = _mm_frcz_ps(rt);
 240 #else
 241             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 242 #endif
 243             twovfeps         = _mm_add_ps(vfeps,vfeps);
 244             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 245
 246             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 247             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 248             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 249             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 250             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 251             _MM_TRANSPOSE4_PS(Y,F,G,H);
 252             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 253             VV               = _mm_macc_ps(vfeps,Fp,Y);
 254             velec            = _mm_mul_ps(qq00,VV);
 255             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 256             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 257
 258             /* CUBIC SPLINE TABLE DISPERSION */
 259             vfitab           = _mm_add_epi32(vfitab,ifour);
 260             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 261             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 262             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 263             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 264             _MM_TRANSPOSE4_PS(Y,F,G,H);
 265             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 266             VV               = _mm_macc_ps(vfeps,Fp,Y);
 267             vvdw6            = _mm_mul_ps(c6_00,VV);
 268             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 269             fvdw6            = _mm_mul_ps(c6_00,FF);
 270
 271             /* CUBIC SPLINE TABLE REPULSION */
 272             vfitab           = _mm_add_epi32(vfitab,ifour);
 273             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 274             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 275             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 276             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 277             _MM_TRANSPOSE4_PS(Y,F,G,H);
 278             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 279             VV               = _mm_macc_ps(vfeps,Fp,Y);
 280             vvdw12           = _mm_mul_ps(c12_00,VV);
 281             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 282             fvdw12           = _mm_mul_ps(c12_00,FF);
 283             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 284             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 285
 286             /* Update potential sum for this i atom from the interaction with this j atom. */
 287             velecsum         = _mm_add_ps(velecsum,velec);
 288             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 289
 290             fscal            = _mm_add_ps(felec,fvdw);
 291
 292              /* Update vectorial force */
 293             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 294             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 295             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 296
 297             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 298             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 299             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 300
 301             /**************************
 302              * CALCULATE INTERACTIONS *
 303              **************************/
 304
 305             r10              = _mm_mul_ps(rsq10,rinv10);
 306
 307             /* Compute parameters for interactions between i and j atoms */
 308             qq10             = _mm_mul_ps(iq1,jq0);
 309
 310             /* Calculate table index by multiplying r with table scale and truncate to integer */
 311             rt               = _mm_mul_ps(r10,vftabscale);
 312             vfitab           = _mm_cvttps_epi32(rt);
 313 #ifdef __XOP__
 314             vfeps            = _mm_frcz_ps(rt);
 315 #else
 316             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 317 #endif
 318             twovfeps         = _mm_add_ps(vfeps,vfeps);
 319             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 320
 321             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 322             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 323             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 324             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 325             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 326             _MM_TRANSPOSE4_PS(Y,F,G,H);
 327             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 328             VV               = _mm_macc_ps(vfeps,Fp,Y);
 329             velec            = _mm_mul_ps(qq10,VV);
 330             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 331             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 332
 333             /* Update potential sum for this i atom from the interaction with this j atom. */
 334             velecsum         = _mm_add_ps(velecsum,velec);
 335
 336             fscal            = felec;
 337
 338              /* Update vectorial force */
 339             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 340             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 341             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 342
 343             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 344             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 345             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 346
 347             /**************************
 348              * CALCULATE INTERACTIONS *
 349              **************************/
 350
 351             r20              = _mm_mul_ps(rsq20,rinv20);
 352
 353             /* Compute parameters for interactions between i and j atoms */
 354             qq20             = _mm_mul_ps(iq2,jq0);
 355
 356             /* Calculate table index by multiplying r with table scale and truncate to integer */
 357             rt               = _mm_mul_ps(r20,vftabscale);
 358             vfitab           = _mm_cvttps_epi32(rt);
 359 #ifdef __XOP__
 360             vfeps            = _mm_frcz_ps(rt);
 361 #else
 362             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 363 #endif
 364             twovfeps         = _mm_add_ps(vfeps,vfeps);
 365             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 366
 367             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 368             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 369             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 370             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 371             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 372             _MM_TRANSPOSE4_PS(Y,F,G,H);
 373             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 374             VV               = _mm_macc_ps(vfeps,Fp,Y);
 375             velec            = _mm_mul_ps(qq20,VV);
 376             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 377             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 378
 379             /* Update potential sum for this i atom from the interaction with this j atom. */
 380             velecsum         = _mm_add_ps(velecsum,velec);
 381
 382             fscal            = felec;
 383
 384              /* Update vectorial force */
 385             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 386             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 387             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 388
 389             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 390             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 391             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 392
 393             fjptrA             = f+j_coord_offsetA;
 394             fjptrB             = f+j_coord_offsetB;
 395             fjptrC             = f+j_coord_offsetC;
 396             fjptrD             = f+j_coord_offsetD;
 397
 398             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 399
 400             /* Inner loop uses 168 flops */
 401         }
 402
 403         if(jidx<j_index_end)
 404         {
 405
 406             /* Get j neighbor index, and coordinate index */
 407             jnrlistA         = jjnr[jidx];
 408             jnrlistB         = jjnr[jidx+1];
 409             jnrlistC         = jjnr[jidx+2];
 410             jnrlistD         = jjnr[jidx+3];
 411             /* Sign of each element will be negative for non-real atoms.
 412              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 413              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 414              */
 415             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 416             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 417             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 418             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 419             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 420             j_coord_offsetA  = DIM*jnrA;
 421             j_coord_offsetB  = DIM*jnrB;
 422             j_coord_offsetC  = DIM*jnrC;
 423             j_coord_offsetD  = DIM*jnrD;
 424
 425             /* load j atom coordinates */
 426             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 427                                               x+j_coord_offsetC,x+j_coord_offsetD,
 428                                               &jx0,&jy0,&jz0);
 429
 430             /* Calculate displacement vector */
 431             dx00             = _mm_sub_ps(ix0,jx0);
 432             dy00             = _mm_sub_ps(iy0,jy0);
 433             dz00             = _mm_sub_ps(iz0,jz0);
 434             dx10             = _mm_sub_ps(ix1,jx0);
 435             dy10             = _mm_sub_ps(iy1,jy0);
 436             dz10             = _mm_sub_ps(iz1,jz0);
 437             dx20             = _mm_sub_ps(ix2,jx0);
 438             dy20             = _mm_sub_ps(iy2,jy0);
 439             dz20             = _mm_sub_ps(iz2,jz0);
 440
 441             /* Calculate squared distance and things based on it */
 442             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 443             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 444             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 445
 446             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 447             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 448             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 449
 450             /* Load parameters for j particles */
 451             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 452                                                               charge+jnrC+0,charge+jnrD+0);
 453             vdwjidx0A        = 2*vdwtype[jnrA+0];
 454             vdwjidx0B        = 2*vdwtype[jnrB+0];
 455             vdwjidx0C        = 2*vdwtype[jnrC+0];
 456             vdwjidx0D        = 2*vdwtype[jnrD+0];
 457
 458             fjx0             = _mm_setzero_ps();
 459             fjy0             = _mm_setzero_ps();
 460             fjz0             = _mm_setzero_ps();
 461
 462             /**************************
 463              * CALCULATE INTERACTIONS *
 464              **************************/
 465
 466             r00              = _mm_mul_ps(rsq00,rinv00);
 467             r00              = _mm_andnot_ps(dummy_mask,r00);
 468
 469             /* Compute parameters for interactions between i and j atoms */
 470             qq00             = _mm_mul_ps(iq0,jq0);
 471             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 472                                          vdwparam+vdwioffset0+vdwjidx0B,
 473                                          vdwparam+vdwioffset0+vdwjidx0C,
 474                                          vdwparam+vdwioffset0+vdwjidx0D,
 475                                          &c6_00,&c12_00);
 476
 477             /* Calculate table index by multiplying r with table scale and truncate to integer */
 478             rt               = _mm_mul_ps(r00,vftabscale);
 479             vfitab           = _mm_cvttps_epi32(rt);
 480 #ifdef __XOP__
 481             vfeps            = _mm_frcz_ps(rt);
 482 #else
 483             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 484 #endif
 485             twovfeps         = _mm_add_ps(vfeps,vfeps);
 486             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 487
 488             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 489             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 490             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 491             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 492             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 493             _MM_TRANSPOSE4_PS(Y,F,G,H);
 494             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 495             VV               = _mm_macc_ps(vfeps,Fp,Y);
 496             velec            = _mm_mul_ps(qq00,VV);
 497             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 498             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 499
 500             /* CUBIC SPLINE TABLE DISPERSION */
 501             vfitab           = _mm_add_epi32(vfitab,ifour);
 502             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 503             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 504             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 505             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 506             _MM_TRANSPOSE4_PS(Y,F,G,H);
 507             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 508             VV               = _mm_macc_ps(vfeps,Fp,Y);
 509             vvdw6            = _mm_mul_ps(c6_00,VV);
 510             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 511             fvdw6            = _mm_mul_ps(c6_00,FF);
 512
 513             /* CUBIC SPLINE TABLE REPULSION */
 514             vfitab           = _mm_add_epi32(vfitab,ifour);
 515             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 516             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 517             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 518             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 519             _MM_TRANSPOSE4_PS(Y,F,G,H);
 520             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 521             VV               = _mm_macc_ps(vfeps,Fp,Y);
 522             vvdw12           = _mm_mul_ps(c12_00,VV);
 523             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 524             fvdw12           = _mm_mul_ps(c12_00,FF);
 525             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 526             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 527
 528             /* Update potential sum for this i atom from the interaction with this j atom. */
 529             velec            = _mm_andnot_ps(dummy_mask,velec);
 530             velecsum         = _mm_add_ps(velecsum,velec);
 531             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 532             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 533
 534             fscal            = _mm_add_ps(felec,fvdw);
 535
 536             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 537
 538              /* Update vectorial force */
 539             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 540             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 541             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 542
 543             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 544             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 545             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 546
 547             /**************************
 548              * CALCULATE INTERACTIONS *
 549              **************************/
 550
 551             r10              = _mm_mul_ps(rsq10,rinv10);
 552             r10              = _mm_andnot_ps(dummy_mask,r10);
 553
 554             /* Compute parameters for interactions between i and j atoms */
 555             qq10             = _mm_mul_ps(iq1,jq0);
 556
 557             /* Calculate table index by multiplying r with table scale and truncate to integer */
 558             rt               = _mm_mul_ps(r10,vftabscale);
 559             vfitab           = _mm_cvttps_epi32(rt);
 560 #ifdef __XOP__
 561             vfeps            = _mm_frcz_ps(rt);
 562 #else
 563             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 564 #endif
 565             twovfeps         = _mm_add_ps(vfeps,vfeps);
 566             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 567
 568             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 569             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 570             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 571             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 572             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 573             _MM_TRANSPOSE4_PS(Y,F,G,H);
 574             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 575             VV               = _mm_macc_ps(vfeps,Fp,Y);
 576             velec            = _mm_mul_ps(qq10,VV);
 577             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 578             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 579
 580             /* Update potential sum for this i atom from the interaction with this j atom. */
 581             velec            = _mm_andnot_ps(dummy_mask,velec);
 582             velecsum         = _mm_add_ps(velecsum,velec);
 583
 584             fscal            = felec;
 585
 586             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 587
 588              /* Update vectorial force */
 589             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 590             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 591             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 592
 593             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 594             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 595             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 596
 597             /**************************
 598              * CALCULATE INTERACTIONS *
 599              **************************/
 600
 601             r20              = _mm_mul_ps(rsq20,rinv20);
 602             r20              = _mm_andnot_ps(dummy_mask,r20);
 603
 604             /* Compute parameters for interactions between i and j atoms */
 605             qq20             = _mm_mul_ps(iq2,jq0);
 606
 607             /* Calculate table index by multiplying r with table scale and truncate to integer */
 608             rt               = _mm_mul_ps(r20,vftabscale);
 609             vfitab           = _mm_cvttps_epi32(rt);
 610 #ifdef __XOP__
 611             vfeps            = _mm_frcz_ps(rt);
 612 #else
 613             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 614 #endif
 615             twovfeps         = _mm_add_ps(vfeps,vfeps);
 616             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 617
 618             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 619             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 620             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 621             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 622             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 623             _MM_TRANSPOSE4_PS(Y,F,G,H);
 624             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 625             VV               = _mm_macc_ps(vfeps,Fp,Y);
 626             velec            = _mm_mul_ps(qq20,VV);
 627             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 628             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 629
 630             /* Update potential sum for this i atom from the interaction with this j atom. */
 631             velec            = _mm_andnot_ps(dummy_mask,velec);
 632             velecsum         = _mm_add_ps(velecsum,velec);
 633
 634             fscal            = felec;
 635
 636             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 637
 638              /* Update vectorial force */
 639             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 640             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 641             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 642
 643             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 644             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 645             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 646
 647             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 648             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 649             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 650             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 651
 652             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 653
 654             /* Inner loop uses 171 flops */
 655         }
 656
 657         /* End of innermost loop */
 658
 659         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 660                                               f+i_coord_offset,fshift+i_shift_offset);
 661
 662         ggid                        = gid[iidx];
 663         /* Update potential energies */
 664         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 665         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 666
 667         /* Increment number of inner iterations */
 668         inneriter                  += j_index_end - j_index_start;
 669
 670         /* Outer loop uses 20 flops */
 671     }
 672
 673     /* Increment number of outer iterations */
 674     outeriter        += nri;
 675
 676     /* Update outer/inner flops */
 677
 678     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*171);
 679 }
 680 /*
 681  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_avx_128_fma_single
 682  * Electrostatics interaction: CubicSplineTable
 683  * VdW interaction:            CubicSplineTable
 684  * Geometry:                   Water3-Particle
 685  * Calculate force/pot:        Force
 686  */
 687 void
 688 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_avx_128_fma_single
 689                     (t_nblist * gmx_restrict                nlist,
 690                      rvec * gmx_restrict                    xx,
 691                      rvec * gmx_restrict                    ff,
 692                      t_forcerec * gmx_restrict              fr,
 693                      t_mdatoms * gmx_restrict               mdatoms,
 694                      nb_kernel_data_t * gmx_restrict        kernel_data,
 695                      t_nrnb * gmx_restrict                  nrnb)
 696 {
 697     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 698      * just 0 for non-waters.
 699      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 700      * jnr indices corresponding to data put in the four positions in the SIMD register.
 701      */
 702     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 703     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 704     int              jnrA,jnrB,jnrC,jnrD;
 705     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 706     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 707     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 708     real             rcutoff_scalar;
 709     real             *shiftvec,*fshift,*x,*f;
 710     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 711     real             scratch[4*DIM];
 712     __m128           fscal,rcutoff,rcutoff2,jidxall;
 713     int              vdwioffset0;
 714     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 715     int              vdwioffset1;
 716     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 717     int              vdwioffset2;
 718     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 719     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 720     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 721     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 722     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 723     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 724     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 725     real             *charge;
 726     int              nvdwtype;
 727     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 728     int              *vdwtype;
 729     real             *vdwparam;
 730     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 731     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 732     __m128i          vfitab;
 733     __m128i          ifour       = _mm_set1_epi32(4);
 734     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 735     real             *vftab;
 736     __m128           dummy_mask,cutoff_mask;
 737     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 738     __m128           one     = _mm_set1_ps(1.0);
 739     __m128           two     = _mm_set1_ps(2.0);
 740     x                = xx[0];
 741     f                = ff[0];
 742
 743     nri              = nlist->nri;
 744     iinr             = nlist->iinr;
 745     jindex           = nlist->jindex;
 746     jjnr             = nlist->jjnr;
 747     shiftidx         = nlist->shift;
 748     gid              = nlist->gid;
 749     shiftvec         = fr->shift_vec[0];
 750     fshift           = fr->fshift[0];
 751     facel            = _mm_set1_ps(fr->epsfac);
 752     charge           = mdatoms->chargeA;
 753     nvdwtype         = fr->ntype;
 754     vdwparam         = fr->nbfp;
 755     vdwtype          = mdatoms->typeA;
 756
 757     vftab            = kernel_data->table_elec_vdw->data;
 758     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 759
 760     /* Setup water-specific parameters */
 761     inr              = nlist->iinr[0];
 762     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 763     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 764     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 765     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 766
 767     /* Avoid stupid compiler warnings */
 768     jnrA = jnrB = jnrC = jnrD = 0;
 769     j_coord_offsetA = 0;
 770     j_coord_offsetB = 0;
 771     j_coord_offsetC = 0;
 772     j_coord_offsetD = 0;
 773
 774     outeriter        = 0;
 775     inneriter        = 0;
 776
 777     for(iidx=0;iidx<4*DIM;iidx++)
 778     {
 779         scratch[iidx] = 0.0;
 780     }
 781
 782     /* Start outer loop over neighborlists */
 783     for(iidx=0; iidx<nri; iidx++)
 784     {
 785         /* Load shift vector for this list */
 786         i_shift_offset   = DIM*shiftidx[iidx];
 787
 788         /* Load limits for loop over neighbors */
 789         j_index_start    = jindex[iidx];
 790         j_index_end      = jindex[iidx+1];
 791
 792         /* Get outer coordinate index */
 793         inr              = iinr[iidx];
 794         i_coord_offset   = DIM*inr;
 795
 796         /* Load i particle coords and add shift vector */
 797         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 798                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 799
 800         fix0             = _mm_setzero_ps();
 801         fiy0             = _mm_setzero_ps();
 802         fiz0             = _mm_setzero_ps();
 803         fix1             = _mm_setzero_ps();
 804         fiy1             = _mm_setzero_ps();
 805         fiz1             = _mm_setzero_ps();
 806         fix2             = _mm_setzero_ps();
 807         fiy2             = _mm_setzero_ps();
 808         fiz2             = _mm_setzero_ps();
 809
 810         /* Start inner kernel loop */
 811         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 812         {
 813
 814             /* Get j neighbor index, and coordinate index */
 815             jnrA             = jjnr[jidx];
 816             jnrB             = jjnr[jidx+1];
 817             jnrC             = jjnr[jidx+2];
 818             jnrD             = jjnr[jidx+3];
 819             j_coord_offsetA  = DIM*jnrA;
 820             j_coord_offsetB  = DIM*jnrB;
 821             j_coord_offsetC  = DIM*jnrC;
 822             j_coord_offsetD  = DIM*jnrD;
 823
 824             /* load j atom coordinates */
 825             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 826                                               x+j_coord_offsetC,x+j_coord_offsetD,
 827                                               &jx0,&jy0,&jz0);
 828
 829             /* Calculate displacement vector */
 830             dx00             = _mm_sub_ps(ix0,jx0);
 831             dy00             = _mm_sub_ps(iy0,jy0);
 832             dz00             = _mm_sub_ps(iz0,jz0);
 833             dx10             = _mm_sub_ps(ix1,jx0);
 834             dy10             = _mm_sub_ps(iy1,jy0);
 835             dz10             = _mm_sub_ps(iz1,jz0);
 836             dx20             = _mm_sub_ps(ix2,jx0);
 837             dy20             = _mm_sub_ps(iy2,jy0);
 838             dz20             = _mm_sub_ps(iz2,jz0);
 839
 840             /* Calculate squared distance and things based on it */
 841             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 842             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 843             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 844
 845             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 846             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 847             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 848
 849             /* Load parameters for j particles */
 850             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 851                                                               charge+jnrC+0,charge+jnrD+0);
 852             vdwjidx0A        = 2*vdwtype[jnrA+0];
 853             vdwjidx0B        = 2*vdwtype[jnrB+0];
 854             vdwjidx0C        = 2*vdwtype[jnrC+0];
 855             vdwjidx0D        = 2*vdwtype[jnrD+0];
 856
 857             fjx0             = _mm_setzero_ps();
 858             fjy0             = _mm_setzero_ps();
 859             fjz0             = _mm_setzero_ps();
 860
 861             /**************************
 862              * CALCULATE INTERACTIONS *
 863              **************************/
 864
 865             r00              = _mm_mul_ps(rsq00,rinv00);
 866
 867             /* Compute parameters for interactions between i and j atoms */
 868             qq00             = _mm_mul_ps(iq0,jq0);
 869             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 870                                          vdwparam+vdwioffset0+vdwjidx0B,
 871                                          vdwparam+vdwioffset0+vdwjidx0C,
 872                                          vdwparam+vdwioffset0+vdwjidx0D,
 873                                          &c6_00,&c12_00);
 874
 875             /* Calculate table index by multiplying r with table scale and truncate to integer */
 876             rt               = _mm_mul_ps(r00,vftabscale);
 877             vfitab           = _mm_cvttps_epi32(rt);
 878 #ifdef __XOP__
 879             vfeps            = _mm_frcz_ps(rt);
 880 #else
 881             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 882 #endif
 883             twovfeps         = _mm_add_ps(vfeps,vfeps);
 884             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 885
 886             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 887             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 888             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 889             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 890             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 891             _MM_TRANSPOSE4_PS(Y,F,G,H);
 892             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 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(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 895
 896             /* CUBIC SPLINE TABLE DISPERSION */
 897             vfitab           = _mm_add_epi32(vfitab,ifour);
 898             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 899             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 900             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 901             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 902             _MM_TRANSPOSE4_PS(Y,F,G,H);
 903             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 904             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 905             fvdw6            = _mm_mul_ps(c6_00,FF);
 906
 907             /* CUBIC SPLINE TABLE REPULSION */
 908             vfitab           = _mm_add_epi32(vfitab,ifour);
 909             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 910             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 911             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 912             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 913             _MM_TRANSPOSE4_PS(Y,F,G,H);
 914             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 915             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 916             fvdw12           = _mm_mul_ps(c12_00,FF);
 917             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 918
 919             fscal            = _mm_add_ps(felec,fvdw);
 920
 921              /* Update vectorial force */
 922             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 923             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 924             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 925
 926             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 927             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 928             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 929
 930             /**************************
 931              * CALCULATE INTERACTIONS *
 932              **************************/
 933
 934             r10              = _mm_mul_ps(rsq10,rinv10);
 935
 936             /* Compute parameters for interactions between i and j atoms */
 937             qq10             = _mm_mul_ps(iq1,jq0);
 938
 939             /* Calculate table index by multiplying r with table scale and truncate to integer */
 940             rt               = _mm_mul_ps(r10,vftabscale);
 941             vfitab           = _mm_cvttps_epi32(rt);
 942 #ifdef __XOP__
 943             vfeps            = _mm_frcz_ps(rt);
 944 #else
 945             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 946 #endif
 947             twovfeps         = _mm_add_ps(vfeps,vfeps);
 948             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 949
 950             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 951             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 952             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 953             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 954             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 955             _MM_TRANSPOSE4_PS(Y,F,G,H);
 956             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 957             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 958             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 959
 960             fscal            = felec;
 961
 962              /* Update vectorial force */
 963             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 964             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 965             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 966
 967             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 968             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 969             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 970
 971             /**************************
 972              * CALCULATE INTERACTIONS *
 973              **************************/
 974
 975             r20              = _mm_mul_ps(rsq20,rinv20);
 976
 977             /* Compute parameters for interactions between i and j atoms */
 978             qq20             = _mm_mul_ps(iq2,jq0);
 979
 980             /* Calculate table index by multiplying r with table scale and truncate to integer */
 981             rt               = _mm_mul_ps(r20,vftabscale);
 982             vfitab           = _mm_cvttps_epi32(rt);
 983 #ifdef __XOP__
 984             vfeps            = _mm_frcz_ps(rt);
 985 #else
 986             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 987 #endif
 988             twovfeps         = _mm_add_ps(vfeps,vfeps);
 989             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 990
 991             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 992             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 993             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 994             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 995             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 996             _MM_TRANSPOSE4_PS(Y,F,G,H);
 997             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 998             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 999             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1000
1001             fscal            = felec;
1002
1003              /* Update vectorial force */
1004             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1005             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1006             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1007
1008             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1009             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1010             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1011
1012             fjptrA             = f+j_coord_offsetA;
1013             fjptrB             = f+j_coord_offsetB;
1014             fjptrC             = f+j_coord_offsetC;
1015             fjptrD             = f+j_coord_offsetD;
1016
1017             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1018
1019             /* Inner loop uses 148 flops */
1020         }
1021
1022         if(jidx<j_index_end)
1023         {
1024
1025             /* Get j neighbor index, and coordinate index */
1026             jnrlistA         = jjnr[jidx];
1027             jnrlistB         = jjnr[jidx+1];
1028             jnrlistC         = jjnr[jidx+2];
1029             jnrlistD         = jjnr[jidx+3];
1030             /* Sign of each element will be negative for non-real atoms.
1031              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1032              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1033              */
1034             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1035             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1036             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1037             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1038             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1039             j_coord_offsetA  = DIM*jnrA;
1040             j_coord_offsetB  = DIM*jnrB;
1041             j_coord_offsetC  = DIM*jnrC;
1042             j_coord_offsetD  = DIM*jnrD;
1043
1044             /* load j atom coordinates */
1045             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1046                                               x+j_coord_offsetC,x+j_coord_offsetD,
1047                                               &jx0,&jy0,&jz0);
1048
1049             /* Calculate displacement vector */
1050             dx00             = _mm_sub_ps(ix0,jx0);
1051             dy00             = _mm_sub_ps(iy0,jy0);
1052             dz00             = _mm_sub_ps(iz0,jz0);
1053             dx10             = _mm_sub_ps(ix1,jx0);
1054             dy10             = _mm_sub_ps(iy1,jy0);
1055             dz10             = _mm_sub_ps(iz1,jz0);
1056             dx20             = _mm_sub_ps(ix2,jx0);
1057             dy20             = _mm_sub_ps(iy2,jy0);
1058             dz20             = _mm_sub_ps(iz2,jz0);
1059
1060             /* Calculate squared distance and things based on it */
1061             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1062             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1063             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1064
1065             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1066             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1067             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1068
1069             /* Load parameters for j particles */
1070             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1071                                                               charge+jnrC+0,charge+jnrD+0);
1072             vdwjidx0A        = 2*vdwtype[jnrA+0];
1073             vdwjidx0B        = 2*vdwtype[jnrB+0];
1074             vdwjidx0C        = 2*vdwtype[jnrC+0];
1075             vdwjidx0D        = 2*vdwtype[jnrD+0];
1076
1077             fjx0             = _mm_setzero_ps();
1078             fjy0             = _mm_setzero_ps();
1079             fjz0             = _mm_setzero_ps();
1080
1081             /**************************
1082              * CALCULATE INTERACTIONS *
1083              **************************/
1084
1085             r00              = _mm_mul_ps(rsq00,rinv00);
1086             r00              = _mm_andnot_ps(dummy_mask,r00);
1087
1088             /* Compute parameters for interactions between i and j atoms */
1089             qq00             = _mm_mul_ps(iq0,jq0);
1090             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1091                                          vdwparam+vdwioffset0+vdwjidx0B,
1092                                          vdwparam+vdwioffset0+vdwjidx0C,
1093                                          vdwparam+vdwioffset0+vdwjidx0D,
1094                                          &c6_00,&c12_00);
1095
1096             /* Calculate table index by multiplying r with table scale and truncate to integer */
1097             rt               = _mm_mul_ps(r00,vftabscale);
1098             vfitab           = _mm_cvttps_epi32(rt);
1099 #ifdef __XOP__
1100             vfeps            = _mm_frcz_ps(rt);
1101 #else
1102             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1103 #endif
1104             twovfeps         = _mm_add_ps(vfeps,vfeps);
1105             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1106
1107             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1108             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1109             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1110             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1111             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1112             _MM_TRANSPOSE4_PS(Y,F,G,H);
1113             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1114             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1115             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
1116
1117             /* CUBIC SPLINE TABLE DISPERSION */
1118             vfitab           = _mm_add_epi32(vfitab,ifour);
1119             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1120             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1121             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1122             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1123             _MM_TRANSPOSE4_PS(Y,F,G,H);
1124             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1125             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1126             fvdw6            = _mm_mul_ps(c6_00,FF);
1127
1128             /* CUBIC SPLINE TABLE REPULSION */
1129             vfitab           = _mm_add_epi32(vfitab,ifour);
1130             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1131             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1132             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1133             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1134             _MM_TRANSPOSE4_PS(Y,F,G,H);
1135             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1136             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1137             fvdw12           = _mm_mul_ps(c12_00,FF);
1138             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1139
1140             fscal            = _mm_add_ps(felec,fvdw);
1141
1142             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1143
1144              /* Update vectorial force */
1145             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1146             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1147             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1148
1149             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1150             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1151             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1152
1153             /**************************
1154              * CALCULATE INTERACTIONS *
1155              **************************/
1156
1157             r10              = _mm_mul_ps(rsq10,rinv10);
1158             r10              = _mm_andnot_ps(dummy_mask,r10);
1159
1160             /* Compute parameters for interactions between i and j atoms */
1161             qq10             = _mm_mul_ps(iq1,jq0);
1162
1163             /* Calculate table index by multiplying r with table scale and truncate to integer */
1164             rt               = _mm_mul_ps(r10,vftabscale);
1165             vfitab           = _mm_cvttps_epi32(rt);
1166 #ifdef __XOP__
1167             vfeps            = _mm_frcz_ps(rt);
1168 #else
1169             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1170 #endif
1171             twovfeps         = _mm_add_ps(vfeps,vfeps);
1172             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1173
1174             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1175             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1176             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1177             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1178             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1179             _MM_TRANSPOSE4_PS(Y,F,G,H);
1180             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1181             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1182             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1183
1184             fscal            = felec;
1185
1186             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1187
1188              /* Update vectorial force */
1189             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1190             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1191             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1192
1193             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1194             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1195             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1196
1197             /**************************
1198              * CALCULATE INTERACTIONS *
1199              **************************/
1200
1201             r20              = _mm_mul_ps(rsq20,rinv20);
1202             r20              = _mm_andnot_ps(dummy_mask,r20);
1203
1204             /* Compute parameters for interactions between i and j atoms */
1205             qq20             = _mm_mul_ps(iq2,jq0);
1206
1207             /* Calculate table index by multiplying r with table scale and truncate to integer */
1208             rt               = _mm_mul_ps(r20,vftabscale);
1209             vfitab           = _mm_cvttps_epi32(rt);
1210 #ifdef __XOP__
1211             vfeps            = _mm_frcz_ps(rt);
1212 #else
1213             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1214 #endif
1215             twovfeps         = _mm_add_ps(vfeps,vfeps);
1216             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1217
1218             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1219             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1220             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1221             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1222             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1223             _MM_TRANSPOSE4_PS(Y,F,G,H);
1224             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1225             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1226             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1227
1228             fscal            = felec;
1229
1230             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1231
1232              /* Update vectorial force */
1233             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1234             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1235             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1236
1237             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1238             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1239             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1240
1241             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1242             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1243             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1244             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1245
1246             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1247
1248             /* Inner loop uses 151 flops */
1249         }
1250
1251         /* End of innermost loop */
1252
1253         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1254                                               f+i_coord_offset,fshift+i_shift_offset);
1255
1256         /* Increment number of inner iterations */
1257         inneriter                  += j_index_end - j_index_start;
1258
1259         /* Outer loop uses 18 flops */
1260     }
1261
1262     /* Increment number of outer iterations */
1263     outeriter        += nri;
1264
1265     /* Update outer/inner flops */
1266
1267     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*151);
1268 }