src/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_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_ElecCSTab_VdwCSTab_GeomW4P1_VF_avx_128_fma_double
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_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              vdwioffset3;
  73     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  74     int              vdwjidx0A,vdwjidx0B;
  75     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  76     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  77     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  78     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  79     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  80     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     int              nvdwtype;
  83     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  84     int              *vdwtype;
  85     real             *vdwparam;
  86     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  87     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  88     __m128i          vfitab;
  89     __m128i          ifour       = _mm_set1_epi32(4);
  90     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
  91     real             *vftab;
  92     __m128d          dummy_mask,cutoff_mask;
  93     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  94     __m128d          one     = _mm_set1_pd(1.0);
  95     __m128d          two     = _mm_set1_pd(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_pd(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_pd(kernel_data->table_elec_vdw->scale);
 115
 116     /* Setup water-specific parameters */
 117     inr              = nlist->iinr[0];
 118     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 119     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 120     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 121     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 122
 123     /* Avoid stupid compiler warnings */
 124     jnrA = jnrB = 0;
 125     j_coord_offsetA = 0;
 126     j_coord_offsetB = 0;
 127
 128     outeriter        = 0;
 129     inneriter        = 0;
 130
 131     /* Start outer loop over neighborlists */
 132     for(iidx=0; iidx<nri; iidx++)
 133     {
 134         /* Load shift vector for this list */
 135         i_shift_offset   = DIM*shiftidx[iidx];
 136
 137         /* Load limits for loop over neighbors */
 138         j_index_start    = jindex[iidx];
 139         j_index_end      = jindex[iidx+1];
 140
 141         /* Get outer coordinate index */
 142         inr              = iinr[iidx];
 143         i_coord_offset   = DIM*inr;
 144
 145         /* Load i particle coords and add shift vector */
 146         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 147                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 148
 149         fix0             = _mm_setzero_pd();
 150         fiy0             = _mm_setzero_pd();
 151         fiz0             = _mm_setzero_pd();
 152         fix1             = _mm_setzero_pd();
 153         fiy1             = _mm_setzero_pd();
 154         fiz1             = _mm_setzero_pd();
 155         fix2             = _mm_setzero_pd();
 156         fiy2             = _mm_setzero_pd();
 157         fiz2             = _mm_setzero_pd();
 158         fix3             = _mm_setzero_pd();
 159         fiy3             = _mm_setzero_pd();
 160         fiz3             = _mm_setzero_pd();
 161
 162         /* Reset potential sums */
 163         velecsum         = _mm_setzero_pd();
 164         vvdwsum          = _mm_setzero_pd();
 165
 166         /* Start inner kernel loop */
 167         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 168         {
 169
 170             /* Get j neighbor index, and coordinate index */
 171             jnrA             = jjnr[jidx];
 172             jnrB             = jjnr[jidx+1];
 173             j_coord_offsetA  = DIM*jnrA;
 174             j_coord_offsetB  = DIM*jnrB;
 175
 176             /* load j atom coordinates */
 177             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 178                                               &jx0,&jy0,&jz0);
 179
 180             /* Calculate displacement vector */
 181             dx00             = _mm_sub_pd(ix0,jx0);
 182             dy00             = _mm_sub_pd(iy0,jy0);
 183             dz00             = _mm_sub_pd(iz0,jz0);
 184             dx10             = _mm_sub_pd(ix1,jx0);
 185             dy10             = _mm_sub_pd(iy1,jy0);
 186             dz10             = _mm_sub_pd(iz1,jz0);
 187             dx20             = _mm_sub_pd(ix2,jx0);
 188             dy20             = _mm_sub_pd(iy2,jy0);
 189             dz20             = _mm_sub_pd(iz2,jz0);
 190             dx30             = _mm_sub_pd(ix3,jx0);
 191             dy30             = _mm_sub_pd(iy3,jy0);
 192             dz30             = _mm_sub_pd(iz3,jz0);
 193
 194             /* Calculate squared distance and things based on it */
 195             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 196             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 197             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 198             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 199
 200             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 201             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 202             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 203             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 204
 205             /* Load parameters for j particles */
 206             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 207             vdwjidx0A        = 2*vdwtype[jnrA+0];
 208             vdwjidx0B        = 2*vdwtype[jnrB+0];
 209
 210             fjx0             = _mm_setzero_pd();
 211             fjy0             = _mm_setzero_pd();
 212             fjz0             = _mm_setzero_pd();
 213
 214             /**************************
 215              * CALCULATE INTERACTIONS *
 216              **************************/
 217
 218             r00              = _mm_mul_pd(rsq00,rinv00);
 219
 220             /* Compute parameters for interactions between i and j atoms */
 221             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 222                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 223
 224             /* Calculate table index by multiplying r with table scale and truncate to integer */
 225             rt               = _mm_mul_pd(r00,vftabscale);
 226             vfitab           = _mm_cvttpd_epi32(rt);
 227 #ifdef __XOP__
 228             vfeps            = _mm_frcz_pd(rt);
 229 #else
 230             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 231 #endif
 232             twovfeps         = _mm_add_pd(vfeps,vfeps);
 233             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 234
 235             /* CUBIC SPLINE TABLE DISPERSION */
 236             vfitab           = _mm_add_epi32(vfitab,ifour);
 237             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 238             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 239             GMX_MM_TRANSPOSE2_PD(Y,F);
 240             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 241             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 242             GMX_MM_TRANSPOSE2_PD(G,H);
 243             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 244             VV               = _mm_macc_pd(vfeps,Fp,Y);
 245             vvdw6            = _mm_mul_pd(c6_00,VV);
 246             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 247             fvdw6            = _mm_mul_pd(c6_00,FF);
 248
 249             /* CUBIC SPLINE TABLE REPULSION */
 250             vfitab           = _mm_add_epi32(vfitab,ifour);
 251             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 252             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 253             GMX_MM_TRANSPOSE2_PD(Y,F);
 254             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 255             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 256             GMX_MM_TRANSPOSE2_PD(G,H);
 257             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 258             VV               = _mm_macc_pd(vfeps,Fp,Y);
 259             vvdw12           = _mm_mul_pd(c12_00,VV);
 260             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 261             fvdw12           = _mm_mul_pd(c12_00,FF);
 262             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 263             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 264
 265             /* Update potential sum for this i atom from the interaction with this j atom. */
 266             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 267
 268             fscal            = fvdw;
 269
 270             /* Update vectorial force */
 271             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 272             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 273             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 274
 275             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 276             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 277             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 278
 279             /**************************
 280              * CALCULATE INTERACTIONS *
 281              **************************/
 282
 283             r10              = _mm_mul_pd(rsq10,rinv10);
 284
 285             /* Compute parameters for interactions between i and j atoms */
 286             qq10             = _mm_mul_pd(iq1,jq0);
 287
 288             /* Calculate table index by multiplying r with table scale and truncate to integer */
 289             rt               = _mm_mul_pd(r10,vftabscale);
 290             vfitab           = _mm_cvttpd_epi32(rt);
 291 #ifdef __XOP__
 292             vfeps            = _mm_frcz_pd(rt);
 293 #else
 294             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 295 #endif
 296             twovfeps         = _mm_add_pd(vfeps,vfeps);
 297             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 298
 299             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 300             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 301             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 302             GMX_MM_TRANSPOSE2_PD(Y,F);
 303             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 304             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 305             GMX_MM_TRANSPOSE2_PD(G,H);
 306             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 307             VV               = _mm_macc_pd(vfeps,Fp,Y);
 308             velec            = _mm_mul_pd(qq10,VV);
 309             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 310             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 311
 312             /* Update potential sum for this i atom from the interaction with this j atom. */
 313             velecsum         = _mm_add_pd(velecsum,velec);
 314
 315             fscal            = felec;
 316
 317             /* Update vectorial force */
 318             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 319             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 320             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 321
 322             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 323             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 324             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 325
 326             /**************************
 327              * CALCULATE INTERACTIONS *
 328              **************************/
 329
 330             r20              = _mm_mul_pd(rsq20,rinv20);
 331
 332             /* Compute parameters for interactions between i and j atoms */
 333             qq20             = _mm_mul_pd(iq2,jq0);
 334
 335             /* Calculate table index by multiplying r with table scale and truncate to integer */
 336             rt               = _mm_mul_pd(r20,vftabscale);
 337             vfitab           = _mm_cvttpd_epi32(rt);
 338 #ifdef __XOP__
 339             vfeps            = _mm_frcz_pd(rt);
 340 #else
 341             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 342 #endif
 343             twovfeps         = _mm_add_pd(vfeps,vfeps);
 344             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 345
 346             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 347             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 348             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 349             GMX_MM_TRANSPOSE2_PD(Y,F);
 350             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 351             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 352             GMX_MM_TRANSPOSE2_PD(G,H);
 353             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 354             VV               = _mm_macc_pd(vfeps,Fp,Y);
 355             velec            = _mm_mul_pd(qq20,VV);
 356             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 357             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 358
 359             /* Update potential sum for this i atom from the interaction with this j atom. */
 360             velecsum         = _mm_add_pd(velecsum,velec);
 361
 362             fscal            = felec;
 363
 364             /* Update vectorial force */
 365             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 366             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 367             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 368
 369             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 370             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 371             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 372
 373             /**************************
 374              * CALCULATE INTERACTIONS *
 375              **************************/
 376
 377             r30              = _mm_mul_pd(rsq30,rinv30);
 378
 379             /* Compute parameters for interactions between i and j atoms */
 380             qq30             = _mm_mul_pd(iq3,jq0);
 381
 382             /* Calculate table index by multiplying r with table scale and truncate to integer */
 383             rt               = _mm_mul_pd(r30,vftabscale);
 384             vfitab           = _mm_cvttpd_epi32(rt);
 385 #ifdef __XOP__
 386             vfeps            = _mm_frcz_pd(rt);
 387 #else
 388             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 389 #endif
 390             twovfeps         = _mm_add_pd(vfeps,vfeps);
 391             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 392
 393             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 394             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 395             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 396             GMX_MM_TRANSPOSE2_PD(Y,F);
 397             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 398             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 399             GMX_MM_TRANSPOSE2_PD(G,H);
 400             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 401             VV               = _mm_macc_pd(vfeps,Fp,Y);
 402             velec            = _mm_mul_pd(qq30,VV);
 403             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 404             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
 405
 406             /* Update potential sum for this i atom from the interaction with this j atom. */
 407             velecsum         = _mm_add_pd(velecsum,velec);
 408
 409             fscal            = felec;
 410
 411             /* Update vectorial force */
 412             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 413             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 414             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 415
 416             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 417             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 418             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 419
 420             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 421
 422             /* Inner loop uses 200 flops */
 423         }
 424
 425         if(jidx<j_index_end)
 426         {
 427
 428             jnrA             = jjnr[jidx];
 429             j_coord_offsetA  = DIM*jnrA;
 430
 431             /* load j atom coordinates */
 432             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 433                                               &jx0,&jy0,&jz0);
 434
 435             /* Calculate displacement vector */
 436             dx00             = _mm_sub_pd(ix0,jx0);
 437             dy00             = _mm_sub_pd(iy0,jy0);
 438             dz00             = _mm_sub_pd(iz0,jz0);
 439             dx10             = _mm_sub_pd(ix1,jx0);
 440             dy10             = _mm_sub_pd(iy1,jy0);
 441             dz10             = _mm_sub_pd(iz1,jz0);
 442             dx20             = _mm_sub_pd(ix2,jx0);
 443             dy20             = _mm_sub_pd(iy2,jy0);
 444             dz20             = _mm_sub_pd(iz2,jz0);
 445             dx30             = _mm_sub_pd(ix3,jx0);
 446             dy30             = _mm_sub_pd(iy3,jy0);
 447             dz30             = _mm_sub_pd(iz3,jz0);
 448
 449             /* Calculate squared distance and things based on it */
 450             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 451             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 452             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 453             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 454
 455             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 456             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 457             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 458             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 459
 460             /* Load parameters for j particles */
 461             jq0              = _mm_load_sd(charge+jnrA+0);
 462             vdwjidx0A        = 2*vdwtype[jnrA+0];
 463
 464             fjx0             = _mm_setzero_pd();
 465             fjy0             = _mm_setzero_pd();
 466             fjz0             = _mm_setzero_pd();
 467
 468             /**************************
 469              * CALCULATE INTERACTIONS *
 470              **************************/
 471
 472             r00              = _mm_mul_pd(rsq00,rinv00);
 473
 474             /* Compute parameters for interactions between i and j atoms */
 475             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 476
 477             /* Calculate table index by multiplying r with table scale and truncate to integer */
 478             rt               = _mm_mul_pd(r00,vftabscale);
 479             vfitab           = _mm_cvttpd_epi32(rt);
 480 #ifdef __XOP__
 481             vfeps            = _mm_frcz_pd(rt);
 482 #else
 483             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 484 #endif
 485             twovfeps         = _mm_add_pd(vfeps,vfeps);
 486             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 487
 488             /* CUBIC SPLINE TABLE DISPERSION */
 489             vfitab           = _mm_add_epi32(vfitab,ifour);
 490             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 491             F                = _mm_setzero_pd();
 492             GMX_MM_TRANSPOSE2_PD(Y,F);
 493             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 494             H                = _mm_setzero_pd();
 495             GMX_MM_TRANSPOSE2_PD(G,H);
 496             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 497             VV               = _mm_macc_pd(vfeps,Fp,Y);
 498             vvdw6            = _mm_mul_pd(c6_00,VV);
 499             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 500             fvdw6            = _mm_mul_pd(c6_00,FF);
 501
 502             /* CUBIC SPLINE TABLE REPULSION */
 503             vfitab           = _mm_add_epi32(vfitab,ifour);
 504             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 505             F                = _mm_setzero_pd();
 506             GMX_MM_TRANSPOSE2_PD(Y,F);
 507             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 508             H                = _mm_setzero_pd();
 509             GMX_MM_TRANSPOSE2_PD(G,H);
 510             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 511             VV               = _mm_macc_pd(vfeps,Fp,Y);
 512             vvdw12           = _mm_mul_pd(c12_00,VV);
 513             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 514             fvdw12           = _mm_mul_pd(c12_00,FF);
 515             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 516             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 517
 518             /* Update potential sum for this i atom from the interaction with this j atom. */
 519             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 520             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 521
 522             fscal            = fvdw;
 523
 524             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 525
 526             /* Update vectorial force */
 527             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 528             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 529             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 530
 531             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 532             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 533             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 534
 535             /**************************
 536              * CALCULATE INTERACTIONS *
 537              **************************/
 538
 539             r10              = _mm_mul_pd(rsq10,rinv10);
 540
 541             /* Compute parameters for interactions between i and j atoms */
 542             qq10             = _mm_mul_pd(iq1,jq0);
 543
 544             /* Calculate table index by multiplying r with table scale and truncate to integer */
 545             rt               = _mm_mul_pd(r10,vftabscale);
 546             vfitab           = _mm_cvttpd_epi32(rt);
 547 #ifdef __XOP__
 548             vfeps            = _mm_frcz_pd(rt);
 549 #else
 550             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 551 #endif
 552             twovfeps         = _mm_add_pd(vfeps,vfeps);
 553             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 554
 555             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 556             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 557             F                = _mm_setzero_pd();
 558             GMX_MM_TRANSPOSE2_PD(Y,F);
 559             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 560             H                = _mm_setzero_pd();
 561             GMX_MM_TRANSPOSE2_PD(G,H);
 562             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 563             VV               = _mm_macc_pd(vfeps,Fp,Y);
 564             velec            = _mm_mul_pd(qq10,VV);
 565             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 566             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 567
 568             /* Update potential sum for this i atom from the interaction with this j atom. */
 569             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 570             velecsum         = _mm_add_pd(velecsum,velec);
 571
 572             fscal            = felec;
 573
 574             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 575
 576             /* Update vectorial force */
 577             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 578             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 579             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 580
 581             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 582             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 583             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 584
 585             /**************************
 586              * CALCULATE INTERACTIONS *
 587              **************************/
 588
 589             r20              = _mm_mul_pd(rsq20,rinv20);
 590
 591             /* Compute parameters for interactions between i and j atoms */
 592             qq20             = _mm_mul_pd(iq2,jq0);
 593
 594             /* Calculate table index by multiplying r with table scale and truncate to integer */
 595             rt               = _mm_mul_pd(r20,vftabscale);
 596             vfitab           = _mm_cvttpd_epi32(rt);
 597 #ifdef __XOP__
 598             vfeps            = _mm_frcz_pd(rt);
 599 #else
 600             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 601 #endif
 602             twovfeps         = _mm_add_pd(vfeps,vfeps);
 603             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 604
 605             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 606             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 607             F                = _mm_setzero_pd();
 608             GMX_MM_TRANSPOSE2_PD(Y,F);
 609             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 610             H                = _mm_setzero_pd();
 611             GMX_MM_TRANSPOSE2_PD(G,H);
 612             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 613             VV               = _mm_macc_pd(vfeps,Fp,Y);
 614             velec            = _mm_mul_pd(qq20,VV);
 615             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 616             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 617
 618             /* Update potential sum for this i atom from the interaction with this j atom. */
 619             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 620             velecsum         = _mm_add_pd(velecsum,velec);
 621
 622             fscal            = felec;
 623
 624             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 625
 626             /* Update vectorial force */
 627             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 628             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 629             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 630
 631             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 632             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 633             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 634
 635             /**************************
 636              * CALCULATE INTERACTIONS *
 637              **************************/
 638
 639             r30              = _mm_mul_pd(rsq30,rinv30);
 640
 641             /* Compute parameters for interactions between i and j atoms */
 642             qq30             = _mm_mul_pd(iq3,jq0);
 643
 644             /* Calculate table index by multiplying r with table scale and truncate to integer */
 645             rt               = _mm_mul_pd(r30,vftabscale);
 646             vfitab           = _mm_cvttpd_epi32(rt);
 647 #ifdef __XOP__
 648             vfeps            = _mm_frcz_pd(rt);
 649 #else
 650             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 651 #endif
 652             twovfeps         = _mm_add_pd(vfeps,vfeps);
 653             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 654
 655             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 656             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 657             F                = _mm_setzero_pd();
 658             GMX_MM_TRANSPOSE2_PD(Y,F);
 659             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 660             H                = _mm_setzero_pd();
 661             GMX_MM_TRANSPOSE2_PD(G,H);
 662             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 663             VV               = _mm_macc_pd(vfeps,Fp,Y);
 664             velec            = _mm_mul_pd(qq30,VV);
 665             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 666             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
 667
 668             /* Update potential sum for this i atom from the interaction with this j atom. */
 669             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 670             velecsum         = _mm_add_pd(velecsum,velec);
 671
 672             fscal            = felec;
 673
 674             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 675
 676             /* Update vectorial force */
 677             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 678             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 679             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 680
 681             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 682             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 683             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 684
 685             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 686
 687             /* Inner loop uses 200 flops */
 688         }
 689
 690         /* End of innermost loop */
 691
 692         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 693                                               f+i_coord_offset,fshift+i_shift_offset);
 694
 695         ggid                        = gid[iidx];
 696         /* Update potential energies */
 697         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 698         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 699
 700         /* Increment number of inner iterations */
 701         inneriter                  += j_index_end - j_index_start;
 702
 703         /* Outer loop uses 26 flops */
 704     }
 705
 706     /* Increment number of outer iterations */
 707     outeriter        += nri;
 708
 709     /* Update outer/inner flops */
 710
 711     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*200);
 712 }
 713 /*
 714  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_F_avx_128_fma_double
 715  * Electrostatics interaction: CubicSplineTable
 716  * VdW interaction:            CubicSplineTable
 717  * Geometry:                   Water4-Particle
 718  * Calculate force/pot:        Force
 719  */
 720 void
 721 nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_F_avx_128_fma_double
 722                     (t_nblist * gmx_restrict                nlist,
 723                      rvec * gmx_restrict                    xx,
 724                      rvec * gmx_restrict                    ff,
 725                      t_forcerec * gmx_restrict              fr,
 726                      t_mdatoms * gmx_restrict               mdatoms,
 727                      nb_kernel_data_t * gmx_restrict        kernel_data,
 728                      t_nrnb * gmx_restrict                  nrnb)
 729 {
 730     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 731      * just 0 for non-waters.
 732      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 733      * jnr indices corresponding to data put in the four positions in the SIMD register.
 734      */
 735     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 736     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 737     int              jnrA,jnrB;
 738     int              j_coord_offsetA,j_coord_offsetB;
 739     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 740     real             rcutoff_scalar;
 741     real             *shiftvec,*fshift,*x,*f;
 742     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 743     int              vdwioffset0;
 744     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 745     int              vdwioffset1;
 746     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 747     int              vdwioffset2;
 748     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 749     int              vdwioffset3;
 750     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 751     int              vdwjidx0A,vdwjidx0B;
 752     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 753     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 754     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 755     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 756     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 757     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 758     real             *charge;
 759     int              nvdwtype;
 760     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 761     int              *vdwtype;
 762     real             *vdwparam;
 763     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 764     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 765     __m128i          vfitab;
 766     __m128i          ifour       = _mm_set1_epi32(4);
 767     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 768     real             *vftab;
 769     __m128d          dummy_mask,cutoff_mask;
 770     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 771     __m128d          one     = _mm_set1_pd(1.0);
 772     __m128d          two     = _mm_set1_pd(2.0);
 773     x                = xx[0];
 774     f                = ff[0];
 775
 776     nri              = nlist->nri;
 777     iinr             = nlist->iinr;
 778     jindex           = nlist->jindex;
 779     jjnr             = nlist->jjnr;
 780     shiftidx         = nlist->shift;
 781     gid              = nlist->gid;
 782     shiftvec         = fr->shift_vec[0];
 783     fshift           = fr->fshift[0];
 784     facel            = _mm_set1_pd(fr->epsfac);
 785     charge           = mdatoms->chargeA;
 786     nvdwtype         = fr->ntype;
 787     vdwparam         = fr->nbfp;
 788     vdwtype          = mdatoms->typeA;
 789
 790     vftab            = kernel_data->table_elec_vdw->data;
 791     vftabscale       = _mm_set1_pd(kernel_data->table_elec_vdw->scale);
 792
 793     /* Setup water-specific parameters */
 794     inr              = nlist->iinr[0];
 795     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 796     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 797     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 798     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 799
 800     /* Avoid stupid compiler warnings */
 801     jnrA = jnrB = 0;
 802     j_coord_offsetA = 0;
 803     j_coord_offsetB = 0;
 804
 805     outeriter        = 0;
 806     inneriter        = 0;
 807
 808     /* Start outer loop over neighborlists */
 809     for(iidx=0; iidx<nri; iidx++)
 810     {
 811         /* Load shift vector for this list */
 812         i_shift_offset   = DIM*shiftidx[iidx];
 813
 814         /* Load limits for loop over neighbors */
 815         j_index_start    = jindex[iidx];
 816         j_index_end      = jindex[iidx+1];
 817
 818         /* Get outer coordinate index */
 819         inr              = iinr[iidx];
 820         i_coord_offset   = DIM*inr;
 821
 822         /* Load i particle coords and add shift vector */
 823         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 824                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 825
 826         fix0             = _mm_setzero_pd();
 827         fiy0             = _mm_setzero_pd();
 828         fiz0             = _mm_setzero_pd();
 829         fix1             = _mm_setzero_pd();
 830         fiy1             = _mm_setzero_pd();
 831         fiz1             = _mm_setzero_pd();
 832         fix2             = _mm_setzero_pd();
 833         fiy2             = _mm_setzero_pd();
 834         fiz2             = _mm_setzero_pd();
 835         fix3             = _mm_setzero_pd();
 836         fiy3             = _mm_setzero_pd();
 837         fiz3             = _mm_setzero_pd();
 838
 839         /* Start inner kernel loop */
 840         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 841         {
 842
 843             /* Get j neighbor index, and coordinate index */
 844             jnrA             = jjnr[jidx];
 845             jnrB             = jjnr[jidx+1];
 846             j_coord_offsetA  = DIM*jnrA;
 847             j_coord_offsetB  = DIM*jnrB;
 848
 849             /* load j atom coordinates */
 850             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 851                                               &jx0,&jy0,&jz0);
 852
 853             /* Calculate displacement vector */
 854             dx00             = _mm_sub_pd(ix0,jx0);
 855             dy00             = _mm_sub_pd(iy0,jy0);
 856             dz00             = _mm_sub_pd(iz0,jz0);
 857             dx10             = _mm_sub_pd(ix1,jx0);
 858             dy10             = _mm_sub_pd(iy1,jy0);
 859             dz10             = _mm_sub_pd(iz1,jz0);
 860             dx20             = _mm_sub_pd(ix2,jx0);
 861             dy20             = _mm_sub_pd(iy2,jy0);
 862             dz20             = _mm_sub_pd(iz2,jz0);
 863             dx30             = _mm_sub_pd(ix3,jx0);
 864             dy30             = _mm_sub_pd(iy3,jy0);
 865             dz30             = _mm_sub_pd(iz3,jz0);
 866
 867             /* Calculate squared distance and things based on it */
 868             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 869             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 870             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 871             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 872
 873             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 874             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 875             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 876             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 877
 878             /* Load parameters for j particles */
 879             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 880             vdwjidx0A        = 2*vdwtype[jnrA+0];
 881             vdwjidx0B        = 2*vdwtype[jnrB+0];
 882
 883             fjx0             = _mm_setzero_pd();
 884             fjy0             = _mm_setzero_pd();
 885             fjz0             = _mm_setzero_pd();
 886
 887             /**************************
 888              * CALCULATE INTERACTIONS *
 889              **************************/
 890
 891             r00              = _mm_mul_pd(rsq00,rinv00);
 892
 893             /* Compute parameters for interactions between i and j atoms */
 894             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 895                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 896
 897             /* Calculate table index by multiplying r with table scale and truncate to integer */
 898             rt               = _mm_mul_pd(r00,vftabscale);
 899             vfitab           = _mm_cvttpd_epi32(rt);
 900 #ifdef __XOP__
 901             vfeps            = _mm_frcz_pd(rt);
 902 #else
 903             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 904 #endif
 905             twovfeps         = _mm_add_pd(vfeps,vfeps);
 906             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 907
 908             /* CUBIC SPLINE TABLE DISPERSION */
 909             vfitab           = _mm_add_epi32(vfitab,ifour);
 910             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 911             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 912             GMX_MM_TRANSPOSE2_PD(Y,F);
 913             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 914             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 915             GMX_MM_TRANSPOSE2_PD(G,H);
 916             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 917             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 918             fvdw6            = _mm_mul_pd(c6_00,FF);
 919
 920             /* CUBIC SPLINE TABLE REPULSION */
 921             vfitab           = _mm_add_epi32(vfitab,ifour);
 922             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 923             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 924             GMX_MM_TRANSPOSE2_PD(Y,F);
 925             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 926             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 927             GMX_MM_TRANSPOSE2_PD(G,H);
 928             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 929             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 930             fvdw12           = _mm_mul_pd(c12_00,FF);
 931             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 932
 933             fscal            = fvdw;
 934
 935             /* Update vectorial force */
 936             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 937             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 938             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 939
 940             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 941             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 942             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 943
 944             /**************************
 945              * CALCULATE INTERACTIONS *
 946              **************************/
 947
 948             r10              = _mm_mul_pd(rsq10,rinv10);
 949
 950             /* Compute parameters for interactions between i and j atoms */
 951             qq10             = _mm_mul_pd(iq1,jq0);
 952
 953             /* Calculate table index by multiplying r with table scale and truncate to integer */
 954             rt               = _mm_mul_pd(r10,vftabscale);
 955             vfitab           = _mm_cvttpd_epi32(rt);
 956 #ifdef __XOP__
 957             vfeps            = _mm_frcz_pd(rt);
 958 #else
 959             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 960 #endif
 961             twovfeps         = _mm_add_pd(vfeps,vfeps);
 962             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 963
 964             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 965             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 966             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 967             GMX_MM_TRANSPOSE2_PD(Y,F);
 968             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 969             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 970             GMX_MM_TRANSPOSE2_PD(G,H);
 971             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 972             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 973             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 974
 975             fscal            = felec;
 976
 977             /* Update vectorial force */
 978             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 979             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 980             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 981
 982             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 983             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 984             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 985
 986             /**************************
 987              * CALCULATE INTERACTIONS *
 988              **************************/
 989
 990             r20              = _mm_mul_pd(rsq20,rinv20);
 991
 992             /* Compute parameters for interactions between i and j atoms */
 993             qq20             = _mm_mul_pd(iq2,jq0);
 994
 995             /* Calculate table index by multiplying r with table scale and truncate to integer */
 996             rt               = _mm_mul_pd(r20,vftabscale);
 997             vfitab           = _mm_cvttpd_epi32(rt);
 998 #ifdef __XOP__
 999             vfeps            = _mm_frcz_pd(rt);
1000 #else
1001             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1002 #endif
1003             twovfeps         = _mm_add_pd(vfeps,vfeps);
1004             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1005
1006             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1007             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1008             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1009             GMX_MM_TRANSPOSE2_PD(Y,F);
1010             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1011             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1012             GMX_MM_TRANSPOSE2_PD(G,H);
1013             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1014             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1015             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
1016
1017             fscal            = felec;
1018
1019             /* Update vectorial force */
1020             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1021             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1022             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1023
1024             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1025             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1026             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1027
1028             /**************************
1029              * CALCULATE INTERACTIONS *
1030              **************************/
1031
1032             r30              = _mm_mul_pd(rsq30,rinv30);
1033
1034             /* Compute parameters for interactions between i and j atoms */
1035             qq30             = _mm_mul_pd(iq3,jq0);
1036
1037             /* Calculate table index by multiplying r with table scale and truncate to integer */
1038             rt               = _mm_mul_pd(r30,vftabscale);
1039             vfitab           = _mm_cvttpd_epi32(rt);
1040 #ifdef __XOP__
1041             vfeps            = _mm_frcz_pd(rt);
1042 #else
1043             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1044 #endif
1045             twovfeps         = _mm_add_pd(vfeps,vfeps);
1046             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1047
1048             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1049             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1050             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1051             GMX_MM_TRANSPOSE2_PD(Y,F);
1052             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1053             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1054             GMX_MM_TRANSPOSE2_PD(G,H);
1055             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1056             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1057             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
1058
1059             fscal            = felec;
1060
1061             /* Update vectorial force */
1062             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1063             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1064             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1065
1066             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1067             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1068             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1069
1070             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1071
1072             /* Inner loop uses 180 flops */
1073         }
1074
1075         if(jidx<j_index_end)
1076         {
1077
1078             jnrA             = jjnr[jidx];
1079             j_coord_offsetA  = DIM*jnrA;
1080
1081             /* load j atom coordinates */
1082             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1083                                               &jx0,&jy0,&jz0);
1084
1085             /* Calculate displacement vector */
1086             dx00             = _mm_sub_pd(ix0,jx0);
1087             dy00             = _mm_sub_pd(iy0,jy0);
1088             dz00             = _mm_sub_pd(iz0,jz0);
1089             dx10             = _mm_sub_pd(ix1,jx0);
1090             dy10             = _mm_sub_pd(iy1,jy0);
1091             dz10             = _mm_sub_pd(iz1,jz0);
1092             dx20             = _mm_sub_pd(ix2,jx0);
1093             dy20             = _mm_sub_pd(iy2,jy0);
1094             dz20             = _mm_sub_pd(iz2,jz0);
1095             dx30             = _mm_sub_pd(ix3,jx0);
1096             dy30             = _mm_sub_pd(iy3,jy0);
1097             dz30             = _mm_sub_pd(iz3,jz0);
1098
1099             /* Calculate squared distance and things based on it */
1100             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1101             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1102             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1103             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1104
1105             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1106             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1107             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1108             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1109
1110             /* Load parameters for j particles */
1111             jq0              = _mm_load_sd(charge+jnrA+0);
1112             vdwjidx0A        = 2*vdwtype[jnrA+0];
1113
1114             fjx0             = _mm_setzero_pd();
1115             fjy0             = _mm_setzero_pd();
1116             fjz0             = _mm_setzero_pd();
1117
1118             /**************************
1119              * CALCULATE INTERACTIONS *
1120              **************************/
1121
1122             r00              = _mm_mul_pd(rsq00,rinv00);
1123
1124             /* Compute parameters for interactions between i and j atoms */
1125             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1126
1127             /* Calculate table index by multiplying r with table scale and truncate to integer */
1128             rt               = _mm_mul_pd(r00,vftabscale);
1129             vfitab           = _mm_cvttpd_epi32(rt);
1130 #ifdef __XOP__
1131             vfeps            = _mm_frcz_pd(rt);
1132 #else
1133             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1134 #endif
1135             twovfeps         = _mm_add_pd(vfeps,vfeps);
1136             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1137
1138             /* CUBIC SPLINE TABLE DISPERSION */
1139             vfitab           = _mm_add_epi32(vfitab,ifour);
1140             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1141             F                = _mm_setzero_pd();
1142             GMX_MM_TRANSPOSE2_PD(Y,F);
1143             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1144             H                = _mm_setzero_pd();
1145             GMX_MM_TRANSPOSE2_PD(G,H);
1146             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1147             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1148             fvdw6            = _mm_mul_pd(c6_00,FF);
1149
1150             /* CUBIC SPLINE TABLE REPULSION */
1151             vfitab           = _mm_add_epi32(vfitab,ifour);
1152             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1153             F                = _mm_setzero_pd();
1154             GMX_MM_TRANSPOSE2_PD(Y,F);
1155             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1156             H                = _mm_setzero_pd();
1157             GMX_MM_TRANSPOSE2_PD(G,H);
1158             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1159             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1160             fvdw12           = _mm_mul_pd(c12_00,FF);
1161             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1162
1163             fscal            = fvdw;
1164
1165             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1166
1167             /* Update vectorial force */
1168             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1169             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1170             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1171
1172             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1173             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1174             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1175
1176             /**************************
1177              * CALCULATE INTERACTIONS *
1178              **************************/
1179
1180             r10              = _mm_mul_pd(rsq10,rinv10);
1181
1182             /* Compute parameters for interactions between i and j atoms */
1183             qq10             = _mm_mul_pd(iq1,jq0);
1184
1185             /* Calculate table index by multiplying r with table scale and truncate to integer */
1186             rt               = _mm_mul_pd(r10,vftabscale);
1187             vfitab           = _mm_cvttpd_epi32(rt);
1188 #ifdef __XOP__
1189             vfeps            = _mm_frcz_pd(rt);
1190 #else
1191             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1192 #endif
1193             twovfeps         = _mm_add_pd(vfeps,vfeps);
1194             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1195
1196             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1197             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1198             F                = _mm_setzero_pd();
1199             GMX_MM_TRANSPOSE2_PD(Y,F);
1200             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1201             H                = _mm_setzero_pd();
1202             GMX_MM_TRANSPOSE2_PD(G,H);
1203             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1204             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1205             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
1206
1207             fscal            = felec;
1208
1209             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1210
1211             /* Update vectorial force */
1212             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1213             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1214             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1215
1216             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1217             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1218             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1219
1220             /**************************
1221              * CALCULATE INTERACTIONS *
1222              **************************/
1223
1224             r20              = _mm_mul_pd(rsq20,rinv20);
1225
1226             /* Compute parameters for interactions between i and j atoms */
1227             qq20             = _mm_mul_pd(iq2,jq0);
1228
1229             /* Calculate table index by multiplying r with table scale and truncate to integer */
1230             rt               = _mm_mul_pd(r20,vftabscale);
1231             vfitab           = _mm_cvttpd_epi32(rt);
1232 #ifdef __XOP__
1233             vfeps            = _mm_frcz_pd(rt);
1234 #else
1235             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1236 #endif
1237             twovfeps         = _mm_add_pd(vfeps,vfeps);
1238             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1239
1240             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1241             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1242             F                = _mm_setzero_pd();
1243             GMX_MM_TRANSPOSE2_PD(Y,F);
1244             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1245             H                = _mm_setzero_pd();
1246             GMX_MM_TRANSPOSE2_PD(G,H);
1247             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1248             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1249             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
1250
1251             fscal            = felec;
1252
1253             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1254
1255             /* Update vectorial force */
1256             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1257             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1258             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1259
1260             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1261             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1262             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1263
1264             /**************************
1265              * CALCULATE INTERACTIONS *
1266              **************************/
1267
1268             r30              = _mm_mul_pd(rsq30,rinv30);
1269
1270             /* Compute parameters for interactions between i and j atoms */
1271             qq30             = _mm_mul_pd(iq3,jq0);
1272
1273             /* Calculate table index by multiplying r with table scale and truncate to integer */
1274             rt               = _mm_mul_pd(r30,vftabscale);
1275             vfitab           = _mm_cvttpd_epi32(rt);
1276 #ifdef __XOP__
1277             vfeps            = _mm_frcz_pd(rt);
1278 #else
1279             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1280 #endif
1281             twovfeps         = _mm_add_pd(vfeps,vfeps);
1282             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1283
1284             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1285             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1286             F                = _mm_setzero_pd();
1287             GMX_MM_TRANSPOSE2_PD(Y,F);
1288             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1289             H                = _mm_setzero_pd();
1290             GMX_MM_TRANSPOSE2_PD(G,H);
1291             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1292             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1293             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
1294
1295             fscal            = felec;
1296
1297             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1298
1299             /* Update vectorial force */
1300             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1301             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1302             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1303
1304             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1305             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1306             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1307
1308             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1309
1310             /* Inner loop uses 180 flops */
1311         }
1312
1313         /* End of innermost loop */
1314
1315         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1316                                               f+i_coord_offset,fshift+i_shift_offset);
1317
1318         /* Increment number of inner iterations */
1319         inneriter                  += j_index_end - j_index_start;
1320
1321         /* Outer loop uses 24 flops */
1322     }
1323
1324     /* Increment number of outer iterations */
1325     outeriter        += nri;
1326
1327     /* Update outer/inner flops */
1328
1329     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*180);
1330 }