src/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_sse4_1_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse4_1_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_sse4_1_double.h"
  34 #include "kernelutil_x86_sse4_1_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_VF_sse4_1_double
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_VF_sse4_1_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;
  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_elec->data;
 111     vftabscale       = _mm_set1_pd(kernel_data->table_elec->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
 196             /* Load parameters for j particles */
 197             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 198             vdwjidx0A        = 2*vdwtype[jnrA+0];
 199             vdwjidx0B        = 2*vdwtype[jnrB+0];
 200
 201             fjx0             = _mm_setzero_pd();
 202             fjy0             = _mm_setzero_pd();
 203             fjz0             = _mm_setzero_pd();
 204
 205             /**************************
 206              * CALCULATE INTERACTIONS *
 207              **************************/
 208
 209             r00              = _mm_mul_pd(rsq00,rinv00);
 210
 211             /* Compute parameters for interactions between i and j atoms */
 212             qq00             = _mm_mul_pd(iq0,jq0);
 213             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 214                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 215
 216             /* Calculate table index by multiplying r with table scale and truncate to integer */
 217             rt               = _mm_mul_pd(r00,vftabscale);
 218             vfitab           = _mm_cvttpd_epi32(rt);
 219             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 220             vfitab           = _mm_slli_epi32(vfitab,2);
 221
 222             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 223             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 224             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 225             GMX_MM_TRANSPOSE2_PD(Y,F);
 226             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 227             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 228             GMX_MM_TRANSPOSE2_PD(G,H);
 229             Heps             = _mm_mul_pd(vfeps,H);
 230             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 231             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 232             velec            = _mm_mul_pd(qq00,VV);
 233             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 234             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 235
 236             /* LENNARD-JONES DISPERSION/REPULSION */
 237
 238             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 239             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 240             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 241             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 242             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 243
 244             /* Update potential sum for this i atom from the interaction with this j atom. */
 245             velecsum         = _mm_add_pd(velecsum,velec);
 246             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 247
 248             fscal            = _mm_add_pd(felec,fvdw);
 249
 250             /* Calculate temporary vectorial force */
 251             tx               = _mm_mul_pd(fscal,dx00);
 252             ty               = _mm_mul_pd(fscal,dy00);
 253             tz               = _mm_mul_pd(fscal,dz00);
 254
 255             /* Update vectorial force */
 256             fix0             = _mm_add_pd(fix0,tx);
 257             fiy0             = _mm_add_pd(fiy0,ty);
 258             fiz0             = _mm_add_pd(fiz0,tz);
 259
 260             fjx0             = _mm_add_pd(fjx0,tx);
 261             fjy0             = _mm_add_pd(fjy0,ty);
 262             fjz0             = _mm_add_pd(fjz0,tz);
 263
 264             /**************************
 265              * CALCULATE INTERACTIONS *
 266              **************************/
 267
 268             r10              = _mm_mul_pd(rsq10,rinv10);
 269
 270             /* Compute parameters for interactions between i and j atoms */
 271             qq10             = _mm_mul_pd(iq1,jq0);
 272
 273             /* Calculate table index by multiplying r with table scale and truncate to integer */
 274             rt               = _mm_mul_pd(r10,vftabscale);
 275             vfitab           = _mm_cvttpd_epi32(rt);
 276             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 277             vfitab           = _mm_slli_epi32(vfitab,2);
 278
 279             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 280             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 281             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 282             GMX_MM_TRANSPOSE2_PD(Y,F);
 283             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 284             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 285             GMX_MM_TRANSPOSE2_PD(G,H);
 286             Heps             = _mm_mul_pd(vfeps,H);
 287             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 288             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 289             velec            = _mm_mul_pd(qq10,VV);
 290             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 291             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 292
 293             /* Update potential sum for this i atom from the interaction with this j atom. */
 294             velecsum         = _mm_add_pd(velecsum,velec);
 295
 296             fscal            = felec;
 297
 298             /* Calculate temporary vectorial force */
 299             tx               = _mm_mul_pd(fscal,dx10);
 300             ty               = _mm_mul_pd(fscal,dy10);
 301             tz               = _mm_mul_pd(fscal,dz10);
 302
 303             /* Update vectorial force */
 304             fix1             = _mm_add_pd(fix1,tx);
 305             fiy1             = _mm_add_pd(fiy1,ty);
 306             fiz1             = _mm_add_pd(fiz1,tz);
 307
 308             fjx0             = _mm_add_pd(fjx0,tx);
 309             fjy0             = _mm_add_pd(fjy0,ty);
 310             fjz0             = _mm_add_pd(fjz0,tz);
 311
 312             /**************************
 313              * CALCULATE INTERACTIONS *
 314              **************************/
 315
 316             r20              = _mm_mul_pd(rsq20,rinv20);
 317
 318             /* Compute parameters for interactions between i and j atoms */
 319             qq20             = _mm_mul_pd(iq2,jq0);
 320
 321             /* Calculate table index by multiplying r with table scale and truncate to integer */
 322             rt               = _mm_mul_pd(r20,vftabscale);
 323             vfitab           = _mm_cvttpd_epi32(rt);
 324             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 325             vfitab           = _mm_slli_epi32(vfitab,2);
 326
 327             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 328             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 329             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 330             GMX_MM_TRANSPOSE2_PD(Y,F);
 331             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 332             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 333             GMX_MM_TRANSPOSE2_PD(G,H);
 334             Heps             = _mm_mul_pd(vfeps,H);
 335             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 336             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 337             velec            = _mm_mul_pd(qq20,VV);
 338             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 339             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 340
 341             /* Update potential sum for this i atom from the interaction with this j atom. */
 342             velecsum         = _mm_add_pd(velecsum,velec);
 343
 344             fscal            = felec;
 345
 346             /* Calculate temporary vectorial force */
 347             tx               = _mm_mul_pd(fscal,dx20);
 348             ty               = _mm_mul_pd(fscal,dy20);
 349             tz               = _mm_mul_pd(fscal,dz20);
 350
 351             /* Update vectorial force */
 352             fix2             = _mm_add_pd(fix2,tx);
 353             fiy2             = _mm_add_pd(fiy2,ty);
 354             fiz2             = _mm_add_pd(fiz2,tz);
 355
 356             fjx0             = _mm_add_pd(fjx0,tx);
 357             fjy0             = _mm_add_pd(fjy0,ty);
 358             fjz0             = _mm_add_pd(fjz0,tz);
 359
 360             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 361
 362             /* Inner loop uses 145 flops */
 363         }
 364
 365         if(jidx<j_index_end)
 366         {
 367
 368             jnrA             = jjnr[jidx];
 369             j_coord_offsetA  = DIM*jnrA;
 370
 371             /* load j atom coordinates */
 372             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 373                                               &jx0,&jy0,&jz0);
 374
 375             /* Calculate displacement vector */
 376             dx00             = _mm_sub_pd(ix0,jx0);
 377             dy00             = _mm_sub_pd(iy0,jy0);
 378             dz00             = _mm_sub_pd(iz0,jz0);
 379             dx10             = _mm_sub_pd(ix1,jx0);
 380             dy10             = _mm_sub_pd(iy1,jy0);
 381             dz10             = _mm_sub_pd(iz1,jz0);
 382             dx20             = _mm_sub_pd(ix2,jx0);
 383             dy20             = _mm_sub_pd(iy2,jy0);
 384             dz20             = _mm_sub_pd(iz2,jz0);
 385
 386             /* Calculate squared distance and things based on it */
 387             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 388             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 389             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 390
 391             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 392             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 393             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 394
 395             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 396
 397             /* Load parameters for j particles */
 398             jq0              = _mm_load_sd(charge+jnrA+0);
 399             vdwjidx0A        = 2*vdwtype[jnrA+0];
 400
 401             fjx0             = _mm_setzero_pd();
 402             fjy0             = _mm_setzero_pd();
 403             fjz0             = _mm_setzero_pd();
 404
 405             /**************************
 406              * CALCULATE INTERACTIONS *
 407              **************************/
 408
 409             r00              = _mm_mul_pd(rsq00,rinv00);
 410
 411             /* Compute parameters for interactions between i and j atoms */
 412             qq00             = _mm_mul_pd(iq0,jq0);
 413             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 414
 415             /* Calculate table index by multiplying r with table scale and truncate to integer */
 416             rt               = _mm_mul_pd(r00,vftabscale);
 417             vfitab           = _mm_cvttpd_epi32(rt);
 418             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 419             vfitab           = _mm_slli_epi32(vfitab,2);
 420
 421             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 422             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 423             F                = _mm_setzero_pd();
 424             GMX_MM_TRANSPOSE2_PD(Y,F);
 425             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 426             H                = _mm_setzero_pd();
 427             GMX_MM_TRANSPOSE2_PD(G,H);
 428             Heps             = _mm_mul_pd(vfeps,H);
 429             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 430             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 431             velec            = _mm_mul_pd(qq00,VV);
 432             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 433             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 434
 435             /* LENNARD-JONES DISPERSION/REPULSION */
 436
 437             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 438             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 439             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 440             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 441             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 442
 443             /* Update potential sum for this i atom from the interaction with this j atom. */
 444             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 445             velecsum         = _mm_add_pd(velecsum,velec);
 446             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 447             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 448
 449             fscal            = _mm_add_pd(felec,fvdw);
 450
 451             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 452
 453             /* Calculate temporary vectorial force */
 454             tx               = _mm_mul_pd(fscal,dx00);
 455             ty               = _mm_mul_pd(fscal,dy00);
 456             tz               = _mm_mul_pd(fscal,dz00);
 457
 458             /* Update vectorial force */
 459             fix0             = _mm_add_pd(fix0,tx);
 460             fiy0             = _mm_add_pd(fiy0,ty);
 461             fiz0             = _mm_add_pd(fiz0,tz);
 462
 463             fjx0             = _mm_add_pd(fjx0,tx);
 464             fjy0             = _mm_add_pd(fjy0,ty);
 465             fjz0             = _mm_add_pd(fjz0,tz);
 466
 467             /**************************
 468              * CALCULATE INTERACTIONS *
 469              **************************/
 470
 471             r10              = _mm_mul_pd(rsq10,rinv10);
 472
 473             /* Compute parameters for interactions between i and j atoms */
 474             qq10             = _mm_mul_pd(iq1,jq0);
 475
 476             /* Calculate table index by multiplying r with table scale and truncate to integer */
 477             rt               = _mm_mul_pd(r10,vftabscale);
 478             vfitab           = _mm_cvttpd_epi32(rt);
 479             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 480             vfitab           = _mm_slli_epi32(vfitab,2);
 481
 482             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 483             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 484             F                = _mm_setzero_pd();
 485             GMX_MM_TRANSPOSE2_PD(Y,F);
 486             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 487             H                = _mm_setzero_pd();
 488             GMX_MM_TRANSPOSE2_PD(G,H);
 489             Heps             = _mm_mul_pd(vfeps,H);
 490             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 491             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 492             velec            = _mm_mul_pd(qq10,VV);
 493             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 494             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 495
 496             /* Update potential sum for this i atom from the interaction with this j atom. */
 497             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 498             velecsum         = _mm_add_pd(velecsum,velec);
 499
 500             fscal            = felec;
 501
 502             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 503
 504             /* Calculate temporary vectorial force */
 505             tx               = _mm_mul_pd(fscal,dx10);
 506             ty               = _mm_mul_pd(fscal,dy10);
 507             tz               = _mm_mul_pd(fscal,dz10);
 508
 509             /* Update vectorial force */
 510             fix1             = _mm_add_pd(fix1,tx);
 511             fiy1             = _mm_add_pd(fiy1,ty);
 512             fiz1             = _mm_add_pd(fiz1,tz);
 513
 514             fjx0             = _mm_add_pd(fjx0,tx);
 515             fjy0             = _mm_add_pd(fjy0,ty);
 516             fjz0             = _mm_add_pd(fjz0,tz);
 517
 518             /**************************
 519              * CALCULATE INTERACTIONS *
 520              **************************/
 521
 522             r20              = _mm_mul_pd(rsq20,rinv20);
 523
 524             /* Compute parameters for interactions between i and j atoms */
 525             qq20             = _mm_mul_pd(iq2,jq0);
 526
 527             /* Calculate table index by multiplying r with table scale and truncate to integer */
 528             rt               = _mm_mul_pd(r20,vftabscale);
 529             vfitab           = _mm_cvttpd_epi32(rt);
 530             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 531             vfitab           = _mm_slli_epi32(vfitab,2);
 532
 533             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 534             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 535             F                = _mm_setzero_pd();
 536             GMX_MM_TRANSPOSE2_PD(Y,F);
 537             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 538             H                = _mm_setzero_pd();
 539             GMX_MM_TRANSPOSE2_PD(G,H);
 540             Heps             = _mm_mul_pd(vfeps,H);
 541             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 542             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 543             velec            = _mm_mul_pd(qq20,VV);
 544             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 545             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 546
 547             /* Update potential sum for this i atom from the interaction with this j atom. */
 548             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 549             velecsum         = _mm_add_pd(velecsum,velec);
 550
 551             fscal            = felec;
 552
 553             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 554
 555             /* Calculate temporary vectorial force */
 556             tx               = _mm_mul_pd(fscal,dx20);
 557             ty               = _mm_mul_pd(fscal,dy20);
 558             tz               = _mm_mul_pd(fscal,dz20);
 559
 560             /* Update vectorial force */
 561             fix2             = _mm_add_pd(fix2,tx);
 562             fiy2             = _mm_add_pd(fiy2,ty);
 563             fiz2             = _mm_add_pd(fiz2,tz);
 564
 565             fjx0             = _mm_add_pd(fjx0,tx);
 566             fjy0             = _mm_add_pd(fjy0,ty);
 567             fjz0             = _mm_add_pd(fjz0,tz);
 568
 569             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 570
 571             /* Inner loop uses 145 flops */
 572         }
 573
 574         /* End of innermost loop */
 575
 576         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 577                                               f+i_coord_offset,fshift+i_shift_offset);
 578
 579         ggid                        = gid[iidx];
 580         /* Update potential energies */
 581         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 582         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 583
 584         /* Increment number of inner iterations */
 585         inneriter                  += j_index_end - j_index_start;
 586
 587         /* Outer loop uses 20 flops */
 588     }
 589
 590     /* Increment number of outer iterations */
 591     outeriter        += nri;
 592
 593     /* Update outer/inner flops */
 594
 595     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*145);
 596 }
 597 /*
 598  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_sse4_1_double
 599  * Electrostatics interaction: CubicSplineTable
 600  * VdW interaction:            LennardJones
 601  * Geometry:                   Water3-Particle
 602  * Calculate force/pot:        Force
 603  */
 604 void
 605 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_sse4_1_double
 606                     (t_nblist * gmx_restrict                nlist,
 607                      rvec * gmx_restrict                    xx,
 608                      rvec * gmx_restrict                    ff,
 609                      t_forcerec * gmx_restrict              fr,
 610                      t_mdatoms * gmx_restrict               mdatoms,
 611                      nb_kernel_data_t * gmx_restrict        kernel_data,
 612                      t_nrnb * gmx_restrict                  nrnb)
 613 {
 614     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 615      * just 0 for non-waters.
 616      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 617      * jnr indices corresponding to data put in the four positions in the SIMD register.
 618      */
 619     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 620     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 621     int              jnrA,jnrB;
 622     int              j_coord_offsetA,j_coord_offsetB;
 623     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 624     real             rcutoff_scalar;
 625     real             *shiftvec,*fshift,*x,*f;
 626     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 627     int              vdwioffset0;
 628     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 629     int              vdwioffset1;
 630     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 631     int              vdwioffset2;
 632     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 633     int              vdwjidx0A,vdwjidx0B;
 634     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 635     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 636     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 637     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 638     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 639     real             *charge;
 640     int              nvdwtype;
 641     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 642     int              *vdwtype;
 643     real             *vdwparam;
 644     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 645     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 646     __m128i          vfitab;
 647     __m128i          ifour       = _mm_set1_epi32(4);
 648     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 649     real             *vftab;
 650     __m128d          dummy_mask,cutoff_mask;
 651     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 652     __m128d          one     = _mm_set1_pd(1.0);
 653     __m128d          two     = _mm_set1_pd(2.0);
 654     x                = xx[0];
 655     f                = ff[0];
 656
 657     nri              = nlist->nri;
 658     iinr             = nlist->iinr;
 659     jindex           = nlist->jindex;
 660     jjnr             = nlist->jjnr;
 661     shiftidx         = nlist->shift;
 662     gid              = nlist->gid;
 663     shiftvec         = fr->shift_vec[0];
 664     fshift           = fr->fshift[0];
 665     facel            = _mm_set1_pd(fr->epsfac);
 666     charge           = mdatoms->chargeA;
 667     nvdwtype         = fr->ntype;
 668     vdwparam         = fr->nbfp;
 669     vdwtype          = mdatoms->typeA;
 670
 671     vftab            = kernel_data->table_elec->data;
 672     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 673
 674     /* Setup water-specific parameters */
 675     inr              = nlist->iinr[0];
 676     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 677     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 678     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 679     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 680
 681     /* Avoid stupid compiler warnings */
 682     jnrA = jnrB = 0;
 683     j_coord_offsetA = 0;
 684     j_coord_offsetB = 0;
 685
 686     outeriter        = 0;
 687     inneriter        = 0;
 688
 689     /* Start outer loop over neighborlists */
 690     for(iidx=0; iidx<nri; iidx++)
 691     {
 692         /* Load shift vector for this list */
 693         i_shift_offset   = DIM*shiftidx[iidx];
 694
 695         /* Load limits for loop over neighbors */
 696         j_index_start    = jindex[iidx];
 697         j_index_end      = jindex[iidx+1];
 698
 699         /* Get outer coordinate index */
 700         inr              = iinr[iidx];
 701         i_coord_offset   = DIM*inr;
 702
 703         /* Load i particle coords and add shift vector */
 704         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 705                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 706
 707         fix0             = _mm_setzero_pd();
 708         fiy0             = _mm_setzero_pd();
 709         fiz0             = _mm_setzero_pd();
 710         fix1             = _mm_setzero_pd();
 711         fiy1             = _mm_setzero_pd();
 712         fiz1             = _mm_setzero_pd();
 713         fix2             = _mm_setzero_pd();
 714         fiy2             = _mm_setzero_pd();
 715         fiz2             = _mm_setzero_pd();
 716
 717         /* Start inner kernel loop */
 718         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 719         {
 720
 721             /* Get j neighbor index, and coordinate index */
 722             jnrA             = jjnr[jidx];
 723             jnrB             = jjnr[jidx+1];
 724             j_coord_offsetA  = DIM*jnrA;
 725             j_coord_offsetB  = DIM*jnrB;
 726
 727             /* load j atom coordinates */
 728             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 729                                               &jx0,&jy0,&jz0);
 730
 731             /* Calculate displacement vector */
 732             dx00             = _mm_sub_pd(ix0,jx0);
 733             dy00             = _mm_sub_pd(iy0,jy0);
 734             dz00             = _mm_sub_pd(iz0,jz0);
 735             dx10             = _mm_sub_pd(ix1,jx0);
 736             dy10             = _mm_sub_pd(iy1,jy0);
 737             dz10             = _mm_sub_pd(iz1,jz0);
 738             dx20             = _mm_sub_pd(ix2,jx0);
 739             dy20             = _mm_sub_pd(iy2,jy0);
 740             dz20             = _mm_sub_pd(iz2,jz0);
 741
 742             /* Calculate squared distance and things based on it */
 743             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 744             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 745             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 746
 747             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 748             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 749             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 750
 751             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 752
 753             /* Load parameters for j particles */
 754             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 755             vdwjidx0A        = 2*vdwtype[jnrA+0];
 756             vdwjidx0B        = 2*vdwtype[jnrB+0];
 757
 758             fjx0             = _mm_setzero_pd();
 759             fjy0             = _mm_setzero_pd();
 760             fjz0             = _mm_setzero_pd();
 761
 762             /**************************
 763              * CALCULATE INTERACTIONS *
 764              **************************/
 765
 766             r00              = _mm_mul_pd(rsq00,rinv00);
 767
 768             /* Compute parameters for interactions between i and j atoms */
 769             qq00             = _mm_mul_pd(iq0,jq0);
 770             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 771                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 772
 773             /* Calculate table index by multiplying r with table scale and truncate to integer */
 774             rt               = _mm_mul_pd(r00,vftabscale);
 775             vfitab           = _mm_cvttpd_epi32(rt);
 776             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 777             vfitab           = _mm_slli_epi32(vfitab,2);
 778
 779             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 780             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 781             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 782             GMX_MM_TRANSPOSE2_PD(Y,F);
 783             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 784             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 785             GMX_MM_TRANSPOSE2_PD(G,H);
 786             Heps             = _mm_mul_pd(vfeps,H);
 787             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 788             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 789             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 790
 791             /* LENNARD-JONES DISPERSION/REPULSION */
 792
 793             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 794             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 795
 796             fscal            = _mm_add_pd(felec,fvdw);
 797
 798             /* Calculate temporary vectorial force */
 799             tx               = _mm_mul_pd(fscal,dx00);
 800             ty               = _mm_mul_pd(fscal,dy00);
 801             tz               = _mm_mul_pd(fscal,dz00);
 802
 803             /* Update vectorial force */
 804             fix0             = _mm_add_pd(fix0,tx);
 805             fiy0             = _mm_add_pd(fiy0,ty);
 806             fiz0             = _mm_add_pd(fiz0,tz);
 807
 808             fjx0             = _mm_add_pd(fjx0,tx);
 809             fjy0             = _mm_add_pd(fjy0,ty);
 810             fjz0             = _mm_add_pd(fjz0,tz);
 811
 812             /**************************
 813              * CALCULATE INTERACTIONS *
 814              **************************/
 815
 816             r10              = _mm_mul_pd(rsq10,rinv10);
 817
 818             /* Compute parameters for interactions between i and j atoms */
 819             qq10             = _mm_mul_pd(iq1,jq0);
 820
 821             /* Calculate table index by multiplying r with table scale and truncate to integer */
 822             rt               = _mm_mul_pd(r10,vftabscale);
 823             vfitab           = _mm_cvttpd_epi32(rt);
 824             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 825             vfitab           = _mm_slli_epi32(vfitab,2);
 826
 827             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 828             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 829             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 830             GMX_MM_TRANSPOSE2_PD(Y,F);
 831             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 832             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 833             GMX_MM_TRANSPOSE2_PD(G,H);
 834             Heps             = _mm_mul_pd(vfeps,H);
 835             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 836             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 837             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 838
 839             fscal            = felec;
 840
 841             /* Calculate temporary vectorial force */
 842             tx               = _mm_mul_pd(fscal,dx10);
 843             ty               = _mm_mul_pd(fscal,dy10);
 844             tz               = _mm_mul_pd(fscal,dz10);
 845
 846             /* Update vectorial force */
 847             fix1             = _mm_add_pd(fix1,tx);
 848             fiy1             = _mm_add_pd(fiy1,ty);
 849             fiz1             = _mm_add_pd(fiz1,tz);
 850
 851             fjx0             = _mm_add_pd(fjx0,tx);
 852             fjy0             = _mm_add_pd(fjy0,ty);
 853             fjz0             = _mm_add_pd(fjz0,tz);
 854
 855             /**************************
 856              * CALCULATE INTERACTIONS *
 857              **************************/
 858
 859             r20              = _mm_mul_pd(rsq20,rinv20);
 860
 861             /* Compute parameters for interactions between i and j atoms */
 862             qq20             = _mm_mul_pd(iq2,jq0);
 863
 864             /* Calculate table index by multiplying r with table scale and truncate to integer */
 865             rt               = _mm_mul_pd(r20,vftabscale);
 866             vfitab           = _mm_cvttpd_epi32(rt);
 867             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 868             vfitab           = _mm_slli_epi32(vfitab,2);
 869
 870             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 871             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 872             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 873             GMX_MM_TRANSPOSE2_PD(Y,F);
 874             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 875             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 876             GMX_MM_TRANSPOSE2_PD(G,H);
 877             Heps             = _mm_mul_pd(vfeps,H);
 878             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 879             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 880             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 881
 882             fscal            = felec;
 883
 884             /* Calculate temporary vectorial force */
 885             tx               = _mm_mul_pd(fscal,dx20);
 886             ty               = _mm_mul_pd(fscal,dy20);
 887             tz               = _mm_mul_pd(fscal,dz20);
 888
 889             /* Update vectorial force */
 890             fix2             = _mm_add_pd(fix2,tx);
 891             fiy2             = _mm_add_pd(fiy2,ty);
 892             fiz2             = _mm_add_pd(fiz2,tz);
 893
 894             fjx0             = _mm_add_pd(fjx0,tx);
 895             fjy0             = _mm_add_pd(fjy0,ty);
 896             fjz0             = _mm_add_pd(fjz0,tz);
 897
 898             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 899
 900             /* Inner loop uses 128 flops */
 901         }
 902
 903         if(jidx<j_index_end)
 904         {
 905
 906             jnrA             = jjnr[jidx];
 907             j_coord_offsetA  = DIM*jnrA;
 908
 909             /* load j atom coordinates */
 910             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 911                                               &jx0,&jy0,&jz0);
 912
 913             /* Calculate displacement vector */
 914             dx00             = _mm_sub_pd(ix0,jx0);
 915             dy00             = _mm_sub_pd(iy0,jy0);
 916             dz00             = _mm_sub_pd(iz0,jz0);
 917             dx10             = _mm_sub_pd(ix1,jx0);
 918             dy10             = _mm_sub_pd(iy1,jy0);
 919             dz10             = _mm_sub_pd(iz1,jz0);
 920             dx20             = _mm_sub_pd(ix2,jx0);
 921             dy20             = _mm_sub_pd(iy2,jy0);
 922             dz20             = _mm_sub_pd(iz2,jz0);
 923
 924             /* Calculate squared distance and things based on it */
 925             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 926             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 927             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 928
 929             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 930             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 931             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 932
 933             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 934
 935             /* Load parameters for j particles */
 936             jq0              = _mm_load_sd(charge+jnrA+0);
 937             vdwjidx0A        = 2*vdwtype[jnrA+0];
 938
 939             fjx0             = _mm_setzero_pd();
 940             fjy0             = _mm_setzero_pd();
 941             fjz0             = _mm_setzero_pd();
 942
 943             /**************************
 944              * CALCULATE INTERACTIONS *
 945              **************************/
 946
 947             r00              = _mm_mul_pd(rsq00,rinv00);
 948
 949             /* Compute parameters for interactions between i and j atoms */
 950             qq00             = _mm_mul_pd(iq0,jq0);
 951             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 952
 953             /* Calculate table index by multiplying r with table scale and truncate to integer */
 954             rt               = _mm_mul_pd(r00,vftabscale);
 955             vfitab           = _mm_cvttpd_epi32(rt);
 956             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 957             vfitab           = _mm_slli_epi32(vfitab,2);
 958
 959             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 960             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 961             F                = _mm_setzero_pd();
 962             GMX_MM_TRANSPOSE2_PD(Y,F);
 963             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 964             H                = _mm_setzero_pd();
 965             GMX_MM_TRANSPOSE2_PD(G,H);
 966             Heps             = _mm_mul_pd(vfeps,H);
 967             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 968             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 969             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 970
 971             /* LENNARD-JONES DISPERSION/REPULSION */
 972
 973             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 974             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 975
 976             fscal            = _mm_add_pd(felec,fvdw);
 977
 978             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 979
 980             /* Calculate temporary vectorial force */
 981             tx               = _mm_mul_pd(fscal,dx00);
 982             ty               = _mm_mul_pd(fscal,dy00);
 983             tz               = _mm_mul_pd(fscal,dz00);
 984
 985             /* Update vectorial force */
 986             fix0             = _mm_add_pd(fix0,tx);
 987             fiy0             = _mm_add_pd(fiy0,ty);
 988             fiz0             = _mm_add_pd(fiz0,tz);
 989
 990             fjx0             = _mm_add_pd(fjx0,tx);
 991             fjy0             = _mm_add_pd(fjy0,ty);
 992             fjz0             = _mm_add_pd(fjz0,tz);
 993
 994             /**************************
 995              * CALCULATE INTERACTIONS *
 996              **************************/
 997
 998             r10              = _mm_mul_pd(rsq10,rinv10);
 999
1000             /* Compute parameters for interactions between i and j atoms */
1001             qq10             = _mm_mul_pd(iq1,jq0);
1002
1003             /* Calculate table index by multiplying r with table scale and truncate to integer */
1004             rt               = _mm_mul_pd(r10,vftabscale);
1005             vfitab           = _mm_cvttpd_epi32(rt);
1006             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1007             vfitab           = _mm_slli_epi32(vfitab,2);
1008
1009             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1010             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1011             F                = _mm_setzero_pd();
1012             GMX_MM_TRANSPOSE2_PD(Y,F);
1013             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1014             H                = _mm_setzero_pd();
1015             GMX_MM_TRANSPOSE2_PD(G,H);
1016             Heps             = _mm_mul_pd(vfeps,H);
1017             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1018             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1019             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
1020
1021             fscal            = felec;
1022
1023             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1024
1025             /* Calculate temporary vectorial force */
1026             tx               = _mm_mul_pd(fscal,dx10);
1027             ty               = _mm_mul_pd(fscal,dy10);
1028             tz               = _mm_mul_pd(fscal,dz10);
1029
1030             /* Update vectorial force */
1031             fix1             = _mm_add_pd(fix1,tx);
1032             fiy1             = _mm_add_pd(fiy1,ty);
1033             fiz1             = _mm_add_pd(fiz1,tz);
1034
1035             fjx0             = _mm_add_pd(fjx0,tx);
1036             fjy0             = _mm_add_pd(fjy0,ty);
1037             fjz0             = _mm_add_pd(fjz0,tz);
1038
1039             /**************************
1040              * CALCULATE INTERACTIONS *
1041              **************************/
1042
1043             r20              = _mm_mul_pd(rsq20,rinv20);
1044
1045             /* Compute parameters for interactions between i and j atoms */
1046             qq20             = _mm_mul_pd(iq2,jq0);
1047
1048             /* Calculate table index by multiplying r with table scale and truncate to integer */
1049             rt               = _mm_mul_pd(r20,vftabscale);
1050             vfitab           = _mm_cvttpd_epi32(rt);
1051             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1052             vfitab           = _mm_slli_epi32(vfitab,2);
1053
1054             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1055             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1056             F                = _mm_setzero_pd();
1057             GMX_MM_TRANSPOSE2_PD(Y,F);
1058             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1059             H                = _mm_setzero_pd();
1060             GMX_MM_TRANSPOSE2_PD(G,H);
1061             Heps             = _mm_mul_pd(vfeps,H);
1062             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1063             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1064             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
1065
1066             fscal            = felec;
1067
1068             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1069
1070             /* Calculate temporary vectorial force */
1071             tx               = _mm_mul_pd(fscal,dx20);
1072             ty               = _mm_mul_pd(fscal,dy20);
1073             tz               = _mm_mul_pd(fscal,dz20);
1074
1075             /* Update vectorial force */
1076             fix2             = _mm_add_pd(fix2,tx);
1077             fiy2             = _mm_add_pd(fiy2,ty);
1078             fiz2             = _mm_add_pd(fiz2,tz);
1079
1080             fjx0             = _mm_add_pd(fjx0,tx);
1081             fjy0             = _mm_add_pd(fjy0,ty);
1082             fjz0             = _mm_add_pd(fjz0,tz);
1083
1084             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1085
1086             /* Inner loop uses 128 flops */
1087         }
1088
1089         /* End of innermost loop */
1090
1091         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1092                                               f+i_coord_offset,fshift+i_shift_offset);
1093
1094         /* Increment number of inner iterations */
1095         inneriter                  += j_index_end - j_index_start;
1096
1097         /* Outer loop uses 18 flops */
1098     }
1099
1100     /* Increment number of outer iterations */
1101     outeriter        += nri;
1102
1103     /* Update outer/inner flops */
1104
1105     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*128);
1106 }