src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_avx_128_fma_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_128_fma_double 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_double.h"
  34 #include "kernelutil_x86_avx_128_fma_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_VF_avx_128_fma_double
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_VF_avx_128_fma_double
  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 refer to j loop unrolling done with SSE double precision, e.g. for the two 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;
  61     int              j_coord_offsetA,j_coord_offsetB;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwioffset1;
  69     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  70     int              vdwioffset2;
  71     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  72     int              vdwjidx0A,vdwjidx0B;
  73     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  74     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  75     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  76     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  77     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  78     real             *charge;
  79     int              nvdwtype;
  80     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  81     int              *vdwtype;
  82     real             *vdwparam;
  83     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  84     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  85     __m128i          vfitab;
  86     __m128i          ifour       = _mm_set1_epi32(4);
  87     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
  88     real             *vftab;
  89     __m128d          dummy_mask,cutoff_mask;
  90     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  91     __m128d          one     = _mm_set1_pd(1.0);
  92     __m128d          two     = _mm_set1_pd(2.0);
  93     x                = xx[0];
  94     f                = ff[0];
  95
  96     nri              = nlist->nri;
  97     iinr             = nlist->iinr;
  98     jindex           = nlist->jindex;
  99     jjnr             = nlist->jjnr;
 100     shiftidx         = nlist->shift;
 101     gid              = nlist->gid;
 102     shiftvec         = fr->shift_vec[0];
 103     fshift           = fr->fshift[0];
 104     facel            = _mm_set1_pd(fr->epsfac);
 105     charge           = mdatoms->chargeA;
 106     nvdwtype         = fr->ntype;
 107     vdwparam         = fr->nbfp;
 108     vdwtype          = mdatoms->typeA;
 109
 110     vftab            = kernel_data->table_vdw->data;
 111     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 112
 113     /* Setup water-specific parameters */
 114     inr              = nlist->iinr[0];
 115     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 116     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 117     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 118     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 119
 120     /* Avoid stupid compiler warnings */
 121     jnrA = jnrB = 0;
 122     j_coord_offsetA = 0;
 123     j_coord_offsetB = 0;
 124
 125     outeriter        = 0;
 126     inneriter        = 0;
 127
 128     /* Start outer loop over neighborlists */
 129     for(iidx=0; iidx<nri; iidx++)
 130     {
 131         /* Load shift vector for this list */
 132         i_shift_offset   = DIM*shiftidx[iidx];
 133
 134         /* Load limits for loop over neighbors */
 135         j_index_start    = jindex[iidx];
 136         j_index_end      = jindex[iidx+1];
 137
 138         /* Get outer coordinate index */
 139         inr              = iinr[iidx];
 140         i_coord_offset   = DIM*inr;
 141
 142         /* Load i particle coords and add shift vector */
 143         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 144                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 145
 146         fix0             = _mm_setzero_pd();
 147         fiy0             = _mm_setzero_pd();
 148         fiz0             = _mm_setzero_pd();
 149         fix1             = _mm_setzero_pd();
 150         fiy1             = _mm_setzero_pd();
 151         fiz1             = _mm_setzero_pd();
 152         fix2             = _mm_setzero_pd();
 153         fiy2             = _mm_setzero_pd();
 154         fiz2             = _mm_setzero_pd();
 155
 156         /* Reset potential sums */
 157         velecsum         = _mm_setzero_pd();
 158         vvdwsum          = _mm_setzero_pd();
 159
 160         /* Start inner kernel loop */
 161         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 162         {
 163
 164             /* Get j neighbor index, and coordinate index */
 165             jnrA             = jjnr[jidx];
 166             jnrB             = jjnr[jidx+1];
 167             j_coord_offsetA  = DIM*jnrA;
 168             j_coord_offsetB  = DIM*jnrB;
 169
 170             /* load j atom coordinates */
 171             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 172                                               &jx0,&jy0,&jz0);
 173
 174             /* Calculate displacement vector */
 175             dx00             = _mm_sub_pd(ix0,jx0);
 176             dy00             = _mm_sub_pd(iy0,jy0);
 177             dz00             = _mm_sub_pd(iz0,jz0);
 178             dx10             = _mm_sub_pd(ix1,jx0);
 179             dy10             = _mm_sub_pd(iy1,jy0);
 180             dz10             = _mm_sub_pd(iz1,jz0);
 181             dx20             = _mm_sub_pd(ix2,jx0);
 182             dy20             = _mm_sub_pd(iy2,jy0);
 183             dz20             = _mm_sub_pd(iz2,jz0);
 184
 185             /* Calculate squared distance and things based on it */
 186             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 187             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 188             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 189
 190             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 191             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 192             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 193
 194             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 195             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 196             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 197
 198             /* Load parameters for j particles */
 199             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 200             vdwjidx0A        = 2*vdwtype[jnrA+0];
 201             vdwjidx0B        = 2*vdwtype[jnrB+0];
 202
 203             fjx0             = _mm_setzero_pd();
 204             fjy0             = _mm_setzero_pd();
 205             fjz0             = _mm_setzero_pd();
 206
 207             /**************************
 208              * CALCULATE INTERACTIONS *
 209              **************************/
 210
 211             r00              = _mm_mul_pd(rsq00,rinv00);
 212
 213             /* Compute parameters for interactions between i and j atoms */
 214             qq00             = _mm_mul_pd(iq0,jq0);
 215             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 216                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 217
 218             /* Calculate table index by multiplying r with table scale and truncate to integer */
 219             rt               = _mm_mul_pd(r00,vftabscale);
 220             vfitab           = _mm_cvttpd_epi32(rt);
 221 #ifdef __XOP__
 222             vfeps            = _mm_frcz_pd(rt);
 223 #else
 224             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 225 #endif
 226             twovfeps         = _mm_add_pd(vfeps,vfeps);
 227             vfitab           = _mm_slli_epi32(vfitab,3);
 228
 229             /* COULOMB ELECTROSTATICS */
 230             velec            = _mm_mul_pd(qq00,rinv00);
 231             felec            = _mm_mul_pd(velec,rinvsq00);
 232
 233             /* CUBIC SPLINE TABLE DISPERSION */
 234             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 235             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 236             GMX_MM_TRANSPOSE2_PD(Y,F);
 237             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 238             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 239             GMX_MM_TRANSPOSE2_PD(G,H);
 240             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 241             VV               = _mm_macc_pd(vfeps,Fp,Y);
 242             vvdw6            = _mm_mul_pd(c6_00,VV);
 243             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 244             fvdw6            = _mm_mul_pd(c6_00,FF);
 245
 246             /* CUBIC SPLINE TABLE REPULSION */
 247             vfitab           = _mm_add_epi32(vfitab,ifour);
 248             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 249             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 250             GMX_MM_TRANSPOSE2_PD(Y,F);
 251             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 252             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 253             GMX_MM_TRANSPOSE2_PD(G,H);
 254             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 255             VV               = _mm_macc_pd(vfeps,Fp,Y);
 256             vvdw12           = _mm_mul_pd(c12_00,VV);
 257             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 258             fvdw12           = _mm_mul_pd(c12_00,FF);
 259             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 260             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 261
 262             /* Update potential sum for this i atom from the interaction with this j atom. */
 263             velecsum         = _mm_add_pd(velecsum,velec);
 264             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 265
 266             fscal            = _mm_add_pd(felec,fvdw);
 267
 268             /* Update vectorial force */
 269             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 270             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 271             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 272
 273             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 274             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 275             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 276
 277             /**************************
 278              * CALCULATE INTERACTIONS *
 279              **************************/
 280
 281             /* Compute parameters for interactions between i and j atoms */
 282             qq10             = _mm_mul_pd(iq1,jq0);
 283
 284             /* COULOMB ELECTROSTATICS */
 285             velec            = _mm_mul_pd(qq10,rinv10);
 286             felec            = _mm_mul_pd(velec,rinvsq10);
 287
 288             /* Update potential sum for this i atom from the interaction with this j atom. */
 289             velecsum         = _mm_add_pd(velecsum,velec);
 290
 291             fscal            = felec;
 292
 293             /* Update vectorial force */
 294             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 295             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 296             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 297
 298             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 299             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 300             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 301
 302             /**************************
 303              * CALCULATE INTERACTIONS *
 304              **************************/
 305
 306             /* Compute parameters for interactions between i and j atoms */
 307             qq20             = _mm_mul_pd(iq2,jq0);
 308
 309             /* COULOMB ELECTROSTATICS */
 310             velec            = _mm_mul_pd(qq20,rinv20);
 311             felec            = _mm_mul_pd(velec,rinvsq20);
 312
 313             /* Update potential sum for this i atom from the interaction with this j atom. */
 314             velecsum         = _mm_add_pd(velecsum,velec);
 315
 316             fscal            = felec;
 317
 318             /* Update vectorial force */
 319             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 320             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 321             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 322
 323             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 324             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 325             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 326
 327             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 328
 329             /* Inner loop uses 131 flops */
 330         }
 331
 332         if(jidx<j_index_end)
 333         {
 334
 335             jnrA             = jjnr[jidx];
 336             j_coord_offsetA  = DIM*jnrA;
 337
 338             /* load j atom coordinates */
 339             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 340                                               &jx0,&jy0,&jz0);
 341
 342             /* Calculate displacement vector */
 343             dx00             = _mm_sub_pd(ix0,jx0);
 344             dy00             = _mm_sub_pd(iy0,jy0);
 345             dz00             = _mm_sub_pd(iz0,jz0);
 346             dx10             = _mm_sub_pd(ix1,jx0);
 347             dy10             = _mm_sub_pd(iy1,jy0);
 348             dz10             = _mm_sub_pd(iz1,jz0);
 349             dx20             = _mm_sub_pd(ix2,jx0);
 350             dy20             = _mm_sub_pd(iy2,jy0);
 351             dz20             = _mm_sub_pd(iz2,jz0);
 352
 353             /* Calculate squared distance and things based on it */
 354             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 355             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 356             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 357
 358             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 359             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 360             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 361
 362             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 363             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 364             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 365
 366             /* Load parameters for j particles */
 367             jq0              = _mm_load_sd(charge+jnrA+0);
 368             vdwjidx0A        = 2*vdwtype[jnrA+0];
 369
 370             fjx0             = _mm_setzero_pd();
 371             fjy0             = _mm_setzero_pd();
 372             fjz0             = _mm_setzero_pd();
 373
 374             /**************************
 375              * CALCULATE INTERACTIONS *
 376              **************************/
 377
 378             r00              = _mm_mul_pd(rsq00,rinv00);
 379
 380             /* Compute parameters for interactions between i and j atoms */
 381             qq00             = _mm_mul_pd(iq0,jq0);
 382             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 383
 384             /* Calculate table index by multiplying r with table scale and truncate to integer */
 385             rt               = _mm_mul_pd(r00,vftabscale);
 386             vfitab           = _mm_cvttpd_epi32(rt);
 387 #ifdef __XOP__
 388             vfeps            = _mm_frcz_pd(rt);
 389 #else
 390             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 391 #endif
 392             twovfeps         = _mm_add_pd(vfeps,vfeps);
 393             vfitab           = _mm_slli_epi32(vfitab,3);
 394
 395             /* COULOMB ELECTROSTATICS */
 396             velec            = _mm_mul_pd(qq00,rinv00);
 397             felec            = _mm_mul_pd(velec,rinvsq00);
 398
 399             /* CUBIC SPLINE TABLE DISPERSION */
 400             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 401             F                = _mm_setzero_pd();
 402             GMX_MM_TRANSPOSE2_PD(Y,F);
 403             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 404             H                = _mm_setzero_pd();
 405             GMX_MM_TRANSPOSE2_PD(G,H);
 406             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 407             VV               = _mm_macc_pd(vfeps,Fp,Y);
 408             vvdw6            = _mm_mul_pd(c6_00,VV);
 409             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 410             fvdw6            = _mm_mul_pd(c6_00,FF);
 411
 412             /* CUBIC SPLINE TABLE REPULSION */
 413             vfitab           = _mm_add_epi32(vfitab,ifour);
 414             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 415             F                = _mm_setzero_pd();
 416             GMX_MM_TRANSPOSE2_PD(Y,F);
 417             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 418             H                = _mm_setzero_pd();
 419             GMX_MM_TRANSPOSE2_PD(G,H);
 420             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 421             VV               = _mm_macc_pd(vfeps,Fp,Y);
 422             vvdw12           = _mm_mul_pd(c12_00,VV);
 423             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 424             fvdw12           = _mm_mul_pd(c12_00,FF);
 425             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 426             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 427
 428             /* Update potential sum for this i atom from the interaction with this j atom. */
 429             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 430             velecsum         = _mm_add_pd(velecsum,velec);
 431             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 432             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 433
 434             fscal            = _mm_add_pd(felec,fvdw);
 435
 436             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 437
 438             /* Update vectorial force */
 439             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 440             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 441             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 442
 443             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 444             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 445             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 446
 447             /**************************
 448              * CALCULATE INTERACTIONS *
 449              **************************/
 450
 451             /* Compute parameters for interactions between i and j atoms */
 452             qq10             = _mm_mul_pd(iq1,jq0);
 453
 454             /* COULOMB ELECTROSTATICS */
 455             velec            = _mm_mul_pd(qq10,rinv10);
 456             felec            = _mm_mul_pd(velec,rinvsq10);
 457
 458             /* Update potential sum for this i atom from the interaction with this j atom. */
 459             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 460             velecsum         = _mm_add_pd(velecsum,velec);
 461
 462             fscal            = felec;
 463
 464             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 465
 466             /* Update vectorial force */
 467             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 468             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 469             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 470
 471             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 472             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 473             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 474
 475             /**************************
 476              * CALCULATE INTERACTIONS *
 477              **************************/
 478
 479             /* Compute parameters for interactions between i and j atoms */
 480             qq20             = _mm_mul_pd(iq2,jq0);
 481
 482             /* COULOMB ELECTROSTATICS */
 483             velec            = _mm_mul_pd(qq20,rinv20);
 484             felec            = _mm_mul_pd(velec,rinvsq20);
 485
 486             /* Update potential sum for this i atom from the interaction with this j atom. */
 487             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 488             velecsum         = _mm_add_pd(velecsum,velec);
 489
 490             fscal            = felec;
 491
 492             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 493
 494             /* Update vectorial force */
 495             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 496             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 497             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 498
 499             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 500             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 501             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 502
 503             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 504
 505             /* Inner loop uses 131 flops */
 506         }
 507
 508         /* End of innermost loop */
 509
 510         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 511                                               f+i_coord_offset,fshift+i_shift_offset);
 512
 513         ggid                        = gid[iidx];
 514         /* Update potential energies */
 515         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 516         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 517
 518         /* Increment number of inner iterations */
 519         inneriter                  += j_index_end - j_index_start;
 520
 521         /* Outer loop uses 20 flops */
 522     }
 523
 524     /* Increment number of outer iterations */
 525     outeriter        += nri;
 526
 527     /* Update outer/inner flops */
 528
 529     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*131);
 530 }
 531 /*
 532  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_avx_128_fma_double
 533  * Electrostatics interaction: Coulomb
 534  * VdW interaction:            CubicSplineTable
 535  * Geometry:                   Water3-Particle
 536  * Calculate force/pot:        Force
 537  */
 538 void
 539 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_avx_128_fma_double
 540                     (t_nblist * gmx_restrict                nlist,
 541                      rvec * gmx_restrict                    xx,
 542                      rvec * gmx_restrict                    ff,
 543                      t_forcerec * gmx_restrict              fr,
 544                      t_mdatoms * gmx_restrict               mdatoms,
 545                      nb_kernel_data_t * gmx_restrict        kernel_data,
 546                      t_nrnb * gmx_restrict                  nrnb)
 547 {
 548     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 549      * just 0 for non-waters.
 550      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 551      * jnr indices corresponding to data put in the four positions in the SIMD register.
 552      */
 553     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 554     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 555     int              jnrA,jnrB;
 556     int              j_coord_offsetA,j_coord_offsetB;
 557     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 558     real             rcutoff_scalar;
 559     real             *shiftvec,*fshift,*x,*f;
 560     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 561     int              vdwioffset0;
 562     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 563     int              vdwioffset1;
 564     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 565     int              vdwioffset2;
 566     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 567     int              vdwjidx0A,vdwjidx0B;
 568     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 569     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 570     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 571     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 572     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 573     real             *charge;
 574     int              nvdwtype;
 575     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 576     int              *vdwtype;
 577     real             *vdwparam;
 578     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 579     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 580     __m128i          vfitab;
 581     __m128i          ifour       = _mm_set1_epi32(4);
 582     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 583     real             *vftab;
 584     __m128d          dummy_mask,cutoff_mask;
 585     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 586     __m128d          one     = _mm_set1_pd(1.0);
 587     __m128d          two     = _mm_set1_pd(2.0);
 588     x                = xx[0];
 589     f                = ff[0];
 590
 591     nri              = nlist->nri;
 592     iinr             = nlist->iinr;
 593     jindex           = nlist->jindex;
 594     jjnr             = nlist->jjnr;
 595     shiftidx         = nlist->shift;
 596     gid              = nlist->gid;
 597     shiftvec         = fr->shift_vec[0];
 598     fshift           = fr->fshift[0];
 599     facel            = _mm_set1_pd(fr->epsfac);
 600     charge           = mdatoms->chargeA;
 601     nvdwtype         = fr->ntype;
 602     vdwparam         = fr->nbfp;
 603     vdwtype          = mdatoms->typeA;
 604
 605     vftab            = kernel_data->table_vdw->data;
 606     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 607
 608     /* Setup water-specific parameters */
 609     inr              = nlist->iinr[0];
 610     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 611     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 612     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 613     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 614
 615     /* Avoid stupid compiler warnings */
 616     jnrA = jnrB = 0;
 617     j_coord_offsetA = 0;
 618     j_coord_offsetB = 0;
 619
 620     outeriter        = 0;
 621     inneriter        = 0;
 622
 623     /* Start outer loop over neighborlists */
 624     for(iidx=0; iidx<nri; iidx++)
 625     {
 626         /* Load shift vector for this list */
 627         i_shift_offset   = DIM*shiftidx[iidx];
 628
 629         /* Load limits for loop over neighbors */
 630         j_index_start    = jindex[iidx];
 631         j_index_end      = jindex[iidx+1];
 632
 633         /* Get outer coordinate index */
 634         inr              = iinr[iidx];
 635         i_coord_offset   = DIM*inr;
 636
 637         /* Load i particle coords and add shift vector */
 638         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 639                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 640
 641         fix0             = _mm_setzero_pd();
 642         fiy0             = _mm_setzero_pd();
 643         fiz0             = _mm_setzero_pd();
 644         fix1             = _mm_setzero_pd();
 645         fiy1             = _mm_setzero_pd();
 646         fiz1             = _mm_setzero_pd();
 647         fix2             = _mm_setzero_pd();
 648         fiy2             = _mm_setzero_pd();
 649         fiz2             = _mm_setzero_pd();
 650
 651         /* Start inner kernel loop */
 652         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 653         {
 654
 655             /* Get j neighbor index, and coordinate index */
 656             jnrA             = jjnr[jidx];
 657             jnrB             = jjnr[jidx+1];
 658             j_coord_offsetA  = DIM*jnrA;
 659             j_coord_offsetB  = DIM*jnrB;
 660
 661             /* load j atom coordinates */
 662             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 663                                               &jx0,&jy0,&jz0);
 664
 665             /* Calculate displacement vector */
 666             dx00             = _mm_sub_pd(ix0,jx0);
 667             dy00             = _mm_sub_pd(iy0,jy0);
 668             dz00             = _mm_sub_pd(iz0,jz0);
 669             dx10             = _mm_sub_pd(ix1,jx0);
 670             dy10             = _mm_sub_pd(iy1,jy0);
 671             dz10             = _mm_sub_pd(iz1,jz0);
 672             dx20             = _mm_sub_pd(ix2,jx0);
 673             dy20             = _mm_sub_pd(iy2,jy0);
 674             dz20             = _mm_sub_pd(iz2,jz0);
 675
 676             /* Calculate squared distance and things based on it */
 677             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 678             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 679             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 680
 681             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 682             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 683             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 684
 685             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 686             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 687             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 688
 689             /* Load parameters for j particles */
 690             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 691             vdwjidx0A        = 2*vdwtype[jnrA+0];
 692             vdwjidx0B        = 2*vdwtype[jnrB+0];
 693
 694             fjx0             = _mm_setzero_pd();
 695             fjy0             = _mm_setzero_pd();
 696             fjz0             = _mm_setzero_pd();
 697
 698             /**************************
 699              * CALCULATE INTERACTIONS *
 700              **************************/
 701
 702             r00              = _mm_mul_pd(rsq00,rinv00);
 703
 704             /* Compute parameters for interactions between i and j atoms */
 705             qq00             = _mm_mul_pd(iq0,jq0);
 706             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 707                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 708
 709             /* Calculate table index by multiplying r with table scale and truncate to integer */
 710             rt               = _mm_mul_pd(r00,vftabscale);
 711             vfitab           = _mm_cvttpd_epi32(rt);
 712 #ifdef __XOP__
 713             vfeps            = _mm_frcz_pd(rt);
 714 #else
 715             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 716 #endif
 717             twovfeps         = _mm_add_pd(vfeps,vfeps);
 718             vfitab           = _mm_slli_epi32(vfitab,3);
 719
 720             /* COULOMB ELECTROSTATICS */
 721             velec            = _mm_mul_pd(qq00,rinv00);
 722             felec            = _mm_mul_pd(velec,rinvsq00);
 723
 724             /* CUBIC SPLINE TABLE DISPERSION */
 725             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 726             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 727             GMX_MM_TRANSPOSE2_PD(Y,F);
 728             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 729             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 730             GMX_MM_TRANSPOSE2_PD(G,H);
 731             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 732             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 733             fvdw6            = _mm_mul_pd(c6_00,FF);
 734
 735             /* CUBIC SPLINE TABLE REPULSION */
 736             vfitab           = _mm_add_epi32(vfitab,ifour);
 737             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 738             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 739             GMX_MM_TRANSPOSE2_PD(Y,F);
 740             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 741             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 742             GMX_MM_TRANSPOSE2_PD(G,H);
 743             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 744             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 745             fvdw12           = _mm_mul_pd(c12_00,FF);
 746             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 747
 748             fscal            = _mm_add_pd(felec,fvdw);
 749
 750             /* Update vectorial force */
 751             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 752             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 753             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 754
 755             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 756             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 757             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 758
 759             /**************************
 760              * CALCULATE INTERACTIONS *
 761              **************************/
 762
 763             /* Compute parameters for interactions between i and j atoms */
 764             qq10             = _mm_mul_pd(iq1,jq0);
 765
 766             /* COULOMB ELECTROSTATICS */
 767             velec            = _mm_mul_pd(qq10,rinv10);
 768             felec            = _mm_mul_pd(velec,rinvsq10);
 769
 770             fscal            = felec;
 771
 772             /* Update vectorial force */
 773             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 774             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 775             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 776
 777             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 778             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 779             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 780
 781             /**************************
 782              * CALCULATE INTERACTIONS *
 783              **************************/
 784
 785             /* Compute parameters for interactions between i and j atoms */
 786             qq20             = _mm_mul_pd(iq2,jq0);
 787
 788             /* COULOMB ELECTROSTATICS */
 789             velec            = _mm_mul_pd(qq20,rinv20);
 790             felec            = _mm_mul_pd(velec,rinvsq20);
 791
 792             fscal            = felec;
 793
 794             /* Update vectorial force */
 795             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 796             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 797             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 798
 799             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 800             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 801             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 802
 803             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 804
 805             /* Inner loop uses 120 flops */
 806         }
 807
 808         if(jidx<j_index_end)
 809         {
 810
 811             jnrA             = jjnr[jidx];
 812             j_coord_offsetA  = DIM*jnrA;
 813
 814             /* load j atom coordinates */
 815             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 816                                               &jx0,&jy0,&jz0);
 817
 818             /* Calculate displacement vector */
 819             dx00             = _mm_sub_pd(ix0,jx0);
 820             dy00             = _mm_sub_pd(iy0,jy0);
 821             dz00             = _mm_sub_pd(iz0,jz0);
 822             dx10             = _mm_sub_pd(ix1,jx0);
 823             dy10             = _mm_sub_pd(iy1,jy0);
 824             dz10             = _mm_sub_pd(iz1,jz0);
 825             dx20             = _mm_sub_pd(ix2,jx0);
 826             dy20             = _mm_sub_pd(iy2,jy0);
 827             dz20             = _mm_sub_pd(iz2,jz0);
 828
 829             /* Calculate squared distance and things based on it */
 830             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 831             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 832             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 833
 834             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 835             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 836             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 837
 838             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 839             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 840             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 841
 842             /* Load parameters for j particles */
 843             jq0              = _mm_load_sd(charge+jnrA+0);
 844             vdwjidx0A        = 2*vdwtype[jnrA+0];
 845
 846             fjx0             = _mm_setzero_pd();
 847             fjy0             = _mm_setzero_pd();
 848             fjz0             = _mm_setzero_pd();
 849
 850             /**************************
 851              * CALCULATE INTERACTIONS *
 852              **************************/
 853
 854             r00              = _mm_mul_pd(rsq00,rinv00);
 855
 856             /* Compute parameters for interactions between i and j atoms */
 857             qq00             = _mm_mul_pd(iq0,jq0);
 858             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 859
 860             /* Calculate table index by multiplying r with table scale and truncate to integer */
 861             rt               = _mm_mul_pd(r00,vftabscale);
 862             vfitab           = _mm_cvttpd_epi32(rt);
 863 #ifdef __XOP__
 864             vfeps            = _mm_frcz_pd(rt);
 865 #else
 866             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 867 #endif
 868             twovfeps         = _mm_add_pd(vfeps,vfeps);
 869             vfitab           = _mm_slli_epi32(vfitab,3);
 870
 871             /* COULOMB ELECTROSTATICS */
 872             velec            = _mm_mul_pd(qq00,rinv00);
 873             felec            = _mm_mul_pd(velec,rinvsq00);
 874
 875             /* CUBIC SPLINE TABLE DISPERSION */
 876             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 877             F                = _mm_setzero_pd();
 878             GMX_MM_TRANSPOSE2_PD(Y,F);
 879             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 880             H                = _mm_setzero_pd();
 881             GMX_MM_TRANSPOSE2_PD(G,H);
 882             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 883             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 884             fvdw6            = _mm_mul_pd(c6_00,FF);
 885
 886             /* CUBIC SPLINE TABLE REPULSION */
 887             vfitab           = _mm_add_epi32(vfitab,ifour);
 888             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 889             F                = _mm_setzero_pd();
 890             GMX_MM_TRANSPOSE2_PD(Y,F);
 891             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 892             H                = _mm_setzero_pd();
 893             GMX_MM_TRANSPOSE2_PD(G,H);
 894             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 895             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 896             fvdw12           = _mm_mul_pd(c12_00,FF);
 897             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 898
 899             fscal            = _mm_add_pd(felec,fvdw);
 900
 901             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 902
 903             /* Update vectorial force */
 904             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 905             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 906             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 907
 908             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 909             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 910             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 911
 912             /**************************
 913              * CALCULATE INTERACTIONS *
 914              **************************/
 915
 916             /* Compute parameters for interactions between i and j atoms */
 917             qq10             = _mm_mul_pd(iq1,jq0);
 918
 919             /* COULOMB ELECTROSTATICS */
 920             velec            = _mm_mul_pd(qq10,rinv10);
 921             felec            = _mm_mul_pd(velec,rinvsq10);
 922
 923             fscal            = felec;
 924
 925             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 926
 927             /* Update vectorial force */
 928             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 929             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 930             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 931
 932             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 933             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 934             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 935
 936             /**************************
 937              * CALCULATE INTERACTIONS *
 938              **************************/
 939
 940             /* Compute parameters for interactions between i and j atoms */
 941             qq20             = _mm_mul_pd(iq2,jq0);
 942
 943             /* COULOMB ELECTROSTATICS */
 944             velec            = _mm_mul_pd(qq20,rinv20);
 945             felec            = _mm_mul_pd(velec,rinvsq20);
 946
 947             fscal            = felec;
 948
 949             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 950
 951             /* Update vectorial force */
 952             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 953             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 954             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 955
 956             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 957             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 958             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 959
 960             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 961
 962             /* Inner loop uses 120 flops */
 963         }
 964
 965         /* End of innermost loop */
 966
 967         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 968                                               f+i_coord_offset,fshift+i_shift_offset);
 969
 970         /* Increment number of inner iterations */
 971         inneriter                  += j_index_end - j_index_start;
 972
 973         /* Outer loop uses 18 flops */
 974     }
 975
 976     /* Increment number of outer iterations */
 977     outeriter        += nri;
 978
 979     /* Update outer/inner flops */
 980
 981     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*120);
 982 }