src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecEwSh_VdwLJSh_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_ElecEwSh_VdwLJSh_GeomW3P1_VF_sse4_1_double
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEwSh_VdwLJSh_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          ewitab;
  86     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  87     real             *ewtab;
  88     __m128d          dummy_mask,cutoff_mask;
  89     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  90     __m128d          one     = _mm_set1_pd(1.0);
  91     __m128d          two     = _mm_set1_pd(2.0);
  92     x                = xx[0];
  93     f                = ff[0];
  94
  95     nri              = nlist->nri;
  96     iinr             = nlist->iinr;
  97     jindex           = nlist->jindex;
  98     jjnr             = nlist->jjnr;
  99     shiftidx         = nlist->shift;
 100     gid              = nlist->gid;
 101     shiftvec         = fr->shift_vec[0];
 102     fshift           = fr->fshift[0];
 103     facel            = _mm_set1_pd(fr->epsfac);
 104     charge           = mdatoms->chargeA;
 105     nvdwtype         = fr->ntype;
 106     vdwparam         = fr->nbfp;
 107     vdwtype          = mdatoms->typeA;
 108
 109     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 110     ewtab            = fr->ic->tabq_coul_FDV0;
 111     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 112     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 113
 114     /* Setup water-specific parameters */
 115     inr              = nlist->iinr[0];
 116     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 117     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 118     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 119     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 120
 121     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 122     rcutoff_scalar   = fr->rcoulomb;
 123     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 124     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 125
 126     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 127     rvdw             = _mm_set1_pd(fr->rvdw);
 128
 129     /* Avoid stupid compiler warnings */
 130     jnrA = jnrB = 0;
 131     j_coord_offsetA = 0;
 132     j_coord_offsetB = 0;
 133
 134     outeriter        = 0;
 135     inneriter        = 0;
 136
 137     /* Start outer loop over neighborlists */
 138     for(iidx=0; iidx<nri; iidx++)
 139     {
 140         /* Load shift vector for this list */
 141         i_shift_offset   = DIM*shiftidx[iidx];
 142
 143         /* Load limits for loop over neighbors */
 144         j_index_start    = jindex[iidx];
 145         j_index_end      = jindex[iidx+1];
 146
 147         /* Get outer coordinate index */
 148         inr              = iinr[iidx];
 149         i_coord_offset   = DIM*inr;
 150
 151         /* Load i particle coords and add shift vector */
 152         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 153                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 154
 155         fix0             = _mm_setzero_pd();
 156         fiy0             = _mm_setzero_pd();
 157         fiz0             = _mm_setzero_pd();
 158         fix1             = _mm_setzero_pd();
 159         fiy1             = _mm_setzero_pd();
 160         fiz1             = _mm_setzero_pd();
 161         fix2             = _mm_setzero_pd();
 162         fiy2             = _mm_setzero_pd();
 163         fiz2             = _mm_setzero_pd();
 164
 165         /* Reset potential sums */
 166         velecsum         = _mm_setzero_pd();
 167         vvdwsum          = _mm_setzero_pd();
 168
 169         /* Start inner kernel loop */
 170         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 171         {
 172
 173             /* Get j neighbor index, and coordinate index */
 174             jnrA             = jjnr[jidx];
 175             jnrB             = jjnr[jidx+1];
 176             j_coord_offsetA  = DIM*jnrA;
 177             j_coord_offsetB  = DIM*jnrB;
 178
 179             /* load j atom coordinates */
 180             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 181                                               &jx0,&jy0,&jz0);
 182
 183             /* Calculate displacement vector */
 184             dx00             = _mm_sub_pd(ix0,jx0);
 185             dy00             = _mm_sub_pd(iy0,jy0);
 186             dz00             = _mm_sub_pd(iz0,jz0);
 187             dx10             = _mm_sub_pd(ix1,jx0);
 188             dy10             = _mm_sub_pd(iy1,jy0);
 189             dz10             = _mm_sub_pd(iz1,jz0);
 190             dx20             = _mm_sub_pd(ix2,jx0);
 191             dy20             = _mm_sub_pd(iy2,jy0);
 192             dz20             = _mm_sub_pd(iz2,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
 199             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 200             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 201             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 202
 203             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 204             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 205             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 206
 207             /* Load parameters for j particles */
 208             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 209             vdwjidx0A        = 2*vdwtype[jnrA+0];
 210             vdwjidx0B        = 2*vdwtype[jnrB+0];
 211
 212             fjx0             = _mm_setzero_pd();
 213             fjy0             = _mm_setzero_pd();
 214             fjz0             = _mm_setzero_pd();
 215
 216             /**************************
 217              * CALCULATE INTERACTIONS *
 218              **************************/
 219
 220             if (gmx_mm_any_lt(rsq00,rcutoff2))
 221             {
 222
 223             r00              = _mm_mul_pd(rsq00,rinv00);
 224
 225             /* Compute parameters for interactions between i and j atoms */
 226             qq00             = _mm_mul_pd(iq0,jq0);
 227             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 228                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 229
 230             /* EWALD ELECTROSTATICS */
 231
 232             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 233             ewrt             = _mm_mul_pd(r00,ewtabscale);
 234             ewitab           = _mm_cvttpd_epi32(ewrt);
 235             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 236             ewitab           = _mm_slli_epi32(ewitab,2);
 237             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 238             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 239             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 240             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 241             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 242             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 243             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 244             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 245             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 246             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 247
 248             /* LENNARD-JONES DISPERSION/REPULSION */
 249
 250             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 251             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 252             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 253             vvdw             = _mm_sub_pd(_mm_mul_pd( _mm_sub_pd(vvdw12 , _mm_mul_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 254                                           _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 255             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 256
 257             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 258
 259             /* Update potential sum for this i atom from the interaction with this j atom. */
 260             velec            = _mm_and_pd(velec,cutoff_mask);
 261             velecsum         = _mm_add_pd(velecsum,velec);
 262             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 263             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 264
 265             fscal            = _mm_add_pd(felec,fvdw);
 266
 267             fscal            = _mm_and_pd(fscal,cutoff_mask);
 268
 269             /* Calculate temporary vectorial force */
 270             tx               = _mm_mul_pd(fscal,dx00);
 271             ty               = _mm_mul_pd(fscal,dy00);
 272             tz               = _mm_mul_pd(fscal,dz00);
 273
 274             /* Update vectorial force */
 275             fix0             = _mm_add_pd(fix0,tx);
 276             fiy0             = _mm_add_pd(fiy0,ty);
 277             fiz0             = _mm_add_pd(fiz0,tz);
 278
 279             fjx0             = _mm_add_pd(fjx0,tx);
 280             fjy0             = _mm_add_pd(fjy0,ty);
 281             fjz0             = _mm_add_pd(fjz0,tz);
 282
 283             }
 284
 285             /**************************
 286              * CALCULATE INTERACTIONS *
 287              **************************/
 288
 289             if (gmx_mm_any_lt(rsq10,rcutoff2))
 290             {
 291
 292             r10              = _mm_mul_pd(rsq10,rinv10);
 293
 294             /* Compute parameters for interactions between i and j atoms */
 295             qq10             = _mm_mul_pd(iq1,jq0);
 296
 297             /* EWALD ELECTROSTATICS */
 298
 299             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 300             ewrt             = _mm_mul_pd(r10,ewtabscale);
 301             ewitab           = _mm_cvttpd_epi32(ewrt);
 302             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 303             ewitab           = _mm_slli_epi32(ewitab,2);
 304             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 305             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 306             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 307             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 308             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 309             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 310             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 311             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 312             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 313             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 314
 315             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 316
 317             /* Update potential sum for this i atom from the interaction with this j atom. */
 318             velec            = _mm_and_pd(velec,cutoff_mask);
 319             velecsum         = _mm_add_pd(velecsum,velec);
 320
 321             fscal            = felec;
 322
 323             fscal            = _mm_and_pd(fscal,cutoff_mask);
 324
 325             /* Calculate temporary vectorial force */
 326             tx               = _mm_mul_pd(fscal,dx10);
 327             ty               = _mm_mul_pd(fscal,dy10);
 328             tz               = _mm_mul_pd(fscal,dz10);
 329
 330             /* Update vectorial force */
 331             fix1             = _mm_add_pd(fix1,tx);
 332             fiy1             = _mm_add_pd(fiy1,ty);
 333             fiz1             = _mm_add_pd(fiz1,tz);
 334
 335             fjx0             = _mm_add_pd(fjx0,tx);
 336             fjy0             = _mm_add_pd(fjy0,ty);
 337             fjz0             = _mm_add_pd(fjz0,tz);
 338
 339             }
 340
 341             /**************************
 342              * CALCULATE INTERACTIONS *
 343              **************************/
 344
 345             if (gmx_mm_any_lt(rsq20,rcutoff2))
 346             {
 347
 348             r20              = _mm_mul_pd(rsq20,rinv20);
 349
 350             /* Compute parameters for interactions between i and j atoms */
 351             qq20             = _mm_mul_pd(iq2,jq0);
 352
 353             /* EWALD ELECTROSTATICS */
 354
 355             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 356             ewrt             = _mm_mul_pd(r20,ewtabscale);
 357             ewitab           = _mm_cvttpd_epi32(ewrt);
 358             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 359             ewitab           = _mm_slli_epi32(ewitab,2);
 360             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 361             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 362             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 363             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 364             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 365             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 366             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 367             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 368             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 369             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 370
 371             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 372
 373             /* Update potential sum for this i atom from the interaction with this j atom. */
 374             velec            = _mm_and_pd(velec,cutoff_mask);
 375             velecsum         = _mm_add_pd(velecsum,velec);
 376
 377             fscal            = felec;
 378
 379             fscal            = _mm_and_pd(fscal,cutoff_mask);
 380
 381             /* Calculate temporary vectorial force */
 382             tx               = _mm_mul_pd(fscal,dx20);
 383             ty               = _mm_mul_pd(fscal,dy20);
 384             tz               = _mm_mul_pd(fscal,dz20);
 385
 386             /* Update vectorial force */
 387             fix2             = _mm_add_pd(fix2,tx);
 388             fiy2             = _mm_add_pd(fiy2,ty);
 389             fiz2             = _mm_add_pd(fiz2,tz);
 390
 391             fjx0             = _mm_add_pd(fjx0,tx);
 392             fjy0             = _mm_add_pd(fjy0,ty);
 393             fjz0             = _mm_add_pd(fjz0,tz);
 394
 395             }
 396
 397             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 398
 399             /* Inner loop uses 159 flops */
 400         }
 401
 402         if(jidx<j_index_end)
 403         {
 404
 405             jnrA             = jjnr[jidx];
 406             j_coord_offsetA  = DIM*jnrA;
 407
 408             /* load j atom coordinates */
 409             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 410                                               &jx0,&jy0,&jz0);
 411
 412             /* Calculate displacement vector */
 413             dx00             = _mm_sub_pd(ix0,jx0);
 414             dy00             = _mm_sub_pd(iy0,jy0);
 415             dz00             = _mm_sub_pd(iz0,jz0);
 416             dx10             = _mm_sub_pd(ix1,jx0);
 417             dy10             = _mm_sub_pd(iy1,jy0);
 418             dz10             = _mm_sub_pd(iz1,jz0);
 419             dx20             = _mm_sub_pd(ix2,jx0);
 420             dy20             = _mm_sub_pd(iy2,jy0);
 421             dz20             = _mm_sub_pd(iz2,jz0);
 422
 423             /* Calculate squared distance and things based on it */
 424             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 425             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 426             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 427
 428             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 429             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 430             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 431
 432             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 433             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 434             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 435
 436             /* Load parameters for j particles */
 437             jq0              = _mm_load_sd(charge+jnrA+0);
 438             vdwjidx0A        = 2*vdwtype[jnrA+0];
 439
 440             fjx0             = _mm_setzero_pd();
 441             fjy0             = _mm_setzero_pd();
 442             fjz0             = _mm_setzero_pd();
 443
 444             /**************************
 445              * CALCULATE INTERACTIONS *
 446              **************************/
 447
 448             if (gmx_mm_any_lt(rsq00,rcutoff2))
 449             {
 450
 451             r00              = _mm_mul_pd(rsq00,rinv00);
 452
 453             /* Compute parameters for interactions between i and j atoms */
 454             qq00             = _mm_mul_pd(iq0,jq0);
 455             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 456
 457             /* EWALD ELECTROSTATICS */
 458
 459             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 460             ewrt             = _mm_mul_pd(r00,ewtabscale);
 461             ewitab           = _mm_cvttpd_epi32(ewrt);
 462             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 463             ewitab           = _mm_slli_epi32(ewitab,2);
 464             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 465             ewtabD           = _mm_setzero_pd();
 466             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 467             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 468             ewtabFn          = _mm_setzero_pd();
 469             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 470             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 471             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 472             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 473             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 474
 475             /* LENNARD-JONES DISPERSION/REPULSION */
 476
 477             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 478             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 479             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 480             vvdw             = _mm_sub_pd(_mm_mul_pd( _mm_sub_pd(vvdw12 , _mm_mul_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 481                                           _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 482             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 483
 484             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 485
 486             /* Update potential sum for this i atom from the interaction with this j atom. */
 487             velec            = _mm_and_pd(velec,cutoff_mask);
 488             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 489             velecsum         = _mm_add_pd(velecsum,velec);
 490             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 491             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 492             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 493
 494             fscal            = _mm_add_pd(felec,fvdw);
 495
 496             fscal            = _mm_and_pd(fscal,cutoff_mask);
 497
 498             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 499
 500             /* Calculate temporary vectorial force */
 501             tx               = _mm_mul_pd(fscal,dx00);
 502             ty               = _mm_mul_pd(fscal,dy00);
 503             tz               = _mm_mul_pd(fscal,dz00);
 504
 505             /* Update vectorial force */
 506             fix0             = _mm_add_pd(fix0,tx);
 507             fiy0             = _mm_add_pd(fiy0,ty);
 508             fiz0             = _mm_add_pd(fiz0,tz);
 509
 510             fjx0             = _mm_add_pd(fjx0,tx);
 511             fjy0             = _mm_add_pd(fjy0,ty);
 512             fjz0             = _mm_add_pd(fjz0,tz);
 513
 514             }
 515
 516             /**************************
 517              * CALCULATE INTERACTIONS *
 518              **************************/
 519
 520             if (gmx_mm_any_lt(rsq10,rcutoff2))
 521             {
 522
 523             r10              = _mm_mul_pd(rsq10,rinv10);
 524
 525             /* Compute parameters for interactions between i and j atoms */
 526             qq10             = _mm_mul_pd(iq1,jq0);
 527
 528             /* EWALD ELECTROSTATICS */
 529
 530             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 531             ewrt             = _mm_mul_pd(r10,ewtabscale);
 532             ewitab           = _mm_cvttpd_epi32(ewrt);
 533             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 534             ewitab           = _mm_slli_epi32(ewitab,2);
 535             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 536             ewtabD           = _mm_setzero_pd();
 537             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 538             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 539             ewtabFn          = _mm_setzero_pd();
 540             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 541             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 542             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 543             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 544             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 545
 546             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 547
 548             /* Update potential sum for this i atom from the interaction with this j atom. */
 549             velec            = _mm_and_pd(velec,cutoff_mask);
 550             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 551             velecsum         = _mm_add_pd(velecsum,velec);
 552
 553             fscal            = felec;
 554
 555             fscal            = _mm_and_pd(fscal,cutoff_mask);
 556
 557             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 558
 559             /* Calculate temporary vectorial force */
 560             tx               = _mm_mul_pd(fscal,dx10);
 561             ty               = _mm_mul_pd(fscal,dy10);
 562             tz               = _mm_mul_pd(fscal,dz10);
 563
 564             /* Update vectorial force */
 565             fix1             = _mm_add_pd(fix1,tx);
 566             fiy1             = _mm_add_pd(fiy1,ty);
 567             fiz1             = _mm_add_pd(fiz1,tz);
 568
 569             fjx0             = _mm_add_pd(fjx0,tx);
 570             fjy0             = _mm_add_pd(fjy0,ty);
 571             fjz0             = _mm_add_pd(fjz0,tz);
 572
 573             }
 574
 575             /**************************
 576              * CALCULATE INTERACTIONS *
 577              **************************/
 578
 579             if (gmx_mm_any_lt(rsq20,rcutoff2))
 580             {
 581
 582             r20              = _mm_mul_pd(rsq20,rinv20);
 583
 584             /* Compute parameters for interactions between i and j atoms */
 585             qq20             = _mm_mul_pd(iq2,jq0);
 586
 587             /* EWALD ELECTROSTATICS */
 588
 589             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 590             ewrt             = _mm_mul_pd(r20,ewtabscale);
 591             ewitab           = _mm_cvttpd_epi32(ewrt);
 592             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 593             ewitab           = _mm_slli_epi32(ewitab,2);
 594             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 595             ewtabD           = _mm_setzero_pd();
 596             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 597             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 598             ewtabFn          = _mm_setzero_pd();
 599             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 600             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 601             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 602             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 603             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 604
 605             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 606
 607             /* Update potential sum for this i atom from the interaction with this j atom. */
 608             velec            = _mm_and_pd(velec,cutoff_mask);
 609             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 610             velecsum         = _mm_add_pd(velecsum,velec);
 611
 612             fscal            = felec;
 613
 614             fscal            = _mm_and_pd(fscal,cutoff_mask);
 615
 616             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 617
 618             /* Calculate temporary vectorial force */
 619             tx               = _mm_mul_pd(fscal,dx20);
 620             ty               = _mm_mul_pd(fscal,dy20);
 621             tz               = _mm_mul_pd(fscal,dz20);
 622
 623             /* Update vectorial force */
 624             fix2             = _mm_add_pd(fix2,tx);
 625             fiy2             = _mm_add_pd(fiy2,ty);
 626             fiz2             = _mm_add_pd(fiz2,tz);
 627
 628             fjx0             = _mm_add_pd(fjx0,tx);
 629             fjy0             = _mm_add_pd(fjy0,ty);
 630             fjz0             = _mm_add_pd(fjz0,tz);
 631
 632             }
 633
 634             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 635
 636             /* Inner loop uses 159 flops */
 637         }
 638
 639         /* End of innermost loop */
 640
 641         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 642                                               f+i_coord_offset,fshift+i_shift_offset);
 643
 644         ggid                        = gid[iidx];
 645         /* Update potential energies */
 646         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 647         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 648
 649         /* Increment number of inner iterations */
 650         inneriter                  += j_index_end - j_index_start;
 651
 652         /* Outer loop uses 20 flops */
 653     }
 654
 655     /* Increment number of outer iterations */
 656     outeriter        += nri;
 657
 658     /* Update outer/inner flops */
 659
 660     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*159);
 661 }
 662 /*
 663  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_sse4_1_double
 664  * Electrostatics interaction: Ewald
 665  * VdW interaction:            LennardJones
 666  * Geometry:                   Water3-Particle
 667  * Calculate force/pot:        Force
 668  */
 669 void
 670 nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_sse4_1_double
 671                     (t_nblist * gmx_restrict                nlist,
 672                      rvec * gmx_restrict                    xx,
 673                      rvec * gmx_restrict                    ff,
 674                      t_forcerec * gmx_restrict              fr,
 675                      t_mdatoms * gmx_restrict               mdatoms,
 676                      nb_kernel_data_t * gmx_restrict        kernel_data,
 677                      t_nrnb * gmx_restrict                  nrnb)
 678 {
 679     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 680      * just 0 for non-waters.
 681      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 682      * jnr indices corresponding to data put in the four positions in the SIMD register.
 683      */
 684     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 685     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 686     int              jnrA,jnrB;
 687     int              j_coord_offsetA,j_coord_offsetB;
 688     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 689     real             rcutoff_scalar;
 690     real             *shiftvec,*fshift,*x,*f;
 691     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 692     int              vdwioffset0;
 693     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 694     int              vdwioffset1;
 695     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 696     int              vdwioffset2;
 697     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 698     int              vdwjidx0A,vdwjidx0B;
 699     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 700     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 701     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 702     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 703     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 704     real             *charge;
 705     int              nvdwtype;
 706     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 707     int              *vdwtype;
 708     real             *vdwparam;
 709     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 710     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 711     __m128i          ewitab;
 712     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 713     real             *ewtab;
 714     __m128d          dummy_mask,cutoff_mask;
 715     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 716     __m128d          one     = _mm_set1_pd(1.0);
 717     __m128d          two     = _mm_set1_pd(2.0);
 718     x                = xx[0];
 719     f                = ff[0];
 720
 721     nri              = nlist->nri;
 722     iinr             = nlist->iinr;
 723     jindex           = nlist->jindex;
 724     jjnr             = nlist->jjnr;
 725     shiftidx         = nlist->shift;
 726     gid              = nlist->gid;
 727     shiftvec         = fr->shift_vec[0];
 728     fshift           = fr->fshift[0];
 729     facel            = _mm_set1_pd(fr->epsfac);
 730     charge           = mdatoms->chargeA;
 731     nvdwtype         = fr->ntype;
 732     vdwparam         = fr->nbfp;
 733     vdwtype          = mdatoms->typeA;
 734
 735     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 736     ewtab            = fr->ic->tabq_coul_F;
 737     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 738     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 739
 740     /* Setup water-specific parameters */
 741     inr              = nlist->iinr[0];
 742     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 743     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 744     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 745     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 746
 747     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 748     rcutoff_scalar   = fr->rcoulomb;
 749     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 750     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 751
 752     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 753     rvdw             = _mm_set1_pd(fr->rvdw);
 754
 755     /* Avoid stupid compiler warnings */
 756     jnrA = jnrB = 0;
 757     j_coord_offsetA = 0;
 758     j_coord_offsetB = 0;
 759
 760     outeriter        = 0;
 761     inneriter        = 0;
 762
 763     /* Start outer loop over neighborlists */
 764     for(iidx=0; iidx<nri; iidx++)
 765     {
 766         /* Load shift vector for this list */
 767         i_shift_offset   = DIM*shiftidx[iidx];
 768
 769         /* Load limits for loop over neighbors */
 770         j_index_start    = jindex[iidx];
 771         j_index_end      = jindex[iidx+1];
 772
 773         /* Get outer coordinate index */
 774         inr              = iinr[iidx];
 775         i_coord_offset   = DIM*inr;
 776
 777         /* Load i particle coords and add shift vector */
 778         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 779                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 780
 781         fix0             = _mm_setzero_pd();
 782         fiy0             = _mm_setzero_pd();
 783         fiz0             = _mm_setzero_pd();
 784         fix1             = _mm_setzero_pd();
 785         fiy1             = _mm_setzero_pd();
 786         fiz1             = _mm_setzero_pd();
 787         fix2             = _mm_setzero_pd();
 788         fiy2             = _mm_setzero_pd();
 789         fiz2             = _mm_setzero_pd();
 790
 791         /* Start inner kernel loop */
 792         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 793         {
 794
 795             /* Get j neighbor index, and coordinate index */
 796             jnrA             = jjnr[jidx];
 797             jnrB             = jjnr[jidx+1];
 798             j_coord_offsetA  = DIM*jnrA;
 799             j_coord_offsetB  = DIM*jnrB;
 800
 801             /* load j atom coordinates */
 802             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 803                                               &jx0,&jy0,&jz0);
 804
 805             /* Calculate displacement vector */
 806             dx00             = _mm_sub_pd(ix0,jx0);
 807             dy00             = _mm_sub_pd(iy0,jy0);
 808             dz00             = _mm_sub_pd(iz0,jz0);
 809             dx10             = _mm_sub_pd(ix1,jx0);
 810             dy10             = _mm_sub_pd(iy1,jy0);
 811             dz10             = _mm_sub_pd(iz1,jz0);
 812             dx20             = _mm_sub_pd(ix2,jx0);
 813             dy20             = _mm_sub_pd(iy2,jy0);
 814             dz20             = _mm_sub_pd(iz2,jz0);
 815
 816             /* Calculate squared distance and things based on it */
 817             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 818             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 819             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 820
 821             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 822             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 823             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 824
 825             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 826             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 827             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 828
 829             /* Load parameters for j particles */
 830             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 831             vdwjidx0A        = 2*vdwtype[jnrA+0];
 832             vdwjidx0B        = 2*vdwtype[jnrB+0];
 833
 834             fjx0             = _mm_setzero_pd();
 835             fjy0             = _mm_setzero_pd();
 836             fjz0             = _mm_setzero_pd();
 837
 838             /**************************
 839              * CALCULATE INTERACTIONS *
 840              **************************/
 841
 842             if (gmx_mm_any_lt(rsq00,rcutoff2))
 843             {
 844
 845             r00              = _mm_mul_pd(rsq00,rinv00);
 846
 847             /* Compute parameters for interactions between i and j atoms */
 848             qq00             = _mm_mul_pd(iq0,jq0);
 849             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 850                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 851
 852             /* EWALD ELECTROSTATICS */
 853
 854             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 855             ewrt             = _mm_mul_pd(r00,ewtabscale);
 856             ewitab           = _mm_cvttpd_epi32(ewrt);
 857             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 858             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 859                                          &ewtabF,&ewtabFn);
 860             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 861             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 862
 863             /* LENNARD-JONES DISPERSION/REPULSION */
 864
 865             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 866             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 867
 868             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 869
 870             fscal            = _mm_add_pd(felec,fvdw);
 871
 872             fscal            = _mm_and_pd(fscal,cutoff_mask);
 873
 874             /* Calculate temporary vectorial force */
 875             tx               = _mm_mul_pd(fscal,dx00);
 876             ty               = _mm_mul_pd(fscal,dy00);
 877             tz               = _mm_mul_pd(fscal,dz00);
 878
 879             /* Update vectorial force */
 880             fix0             = _mm_add_pd(fix0,tx);
 881             fiy0             = _mm_add_pd(fiy0,ty);
 882             fiz0             = _mm_add_pd(fiz0,tz);
 883
 884             fjx0             = _mm_add_pd(fjx0,tx);
 885             fjy0             = _mm_add_pd(fjy0,ty);
 886             fjz0             = _mm_add_pd(fjz0,tz);
 887
 888             }
 889
 890             /**************************
 891              * CALCULATE INTERACTIONS *
 892              **************************/
 893
 894             if (gmx_mm_any_lt(rsq10,rcutoff2))
 895             {
 896
 897             r10              = _mm_mul_pd(rsq10,rinv10);
 898
 899             /* Compute parameters for interactions between i and j atoms */
 900             qq10             = _mm_mul_pd(iq1,jq0);
 901
 902             /* EWALD ELECTROSTATICS */
 903
 904             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 905             ewrt             = _mm_mul_pd(r10,ewtabscale);
 906             ewitab           = _mm_cvttpd_epi32(ewrt);
 907             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 908             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 909                                          &ewtabF,&ewtabFn);
 910             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 911             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 912
 913             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 914
 915             fscal            = felec;
 916
 917             fscal            = _mm_and_pd(fscal,cutoff_mask);
 918
 919             /* Calculate temporary vectorial force */
 920             tx               = _mm_mul_pd(fscal,dx10);
 921             ty               = _mm_mul_pd(fscal,dy10);
 922             tz               = _mm_mul_pd(fscal,dz10);
 923
 924             /* Update vectorial force */
 925             fix1             = _mm_add_pd(fix1,tx);
 926             fiy1             = _mm_add_pd(fiy1,ty);
 927             fiz1             = _mm_add_pd(fiz1,tz);
 928
 929             fjx0             = _mm_add_pd(fjx0,tx);
 930             fjy0             = _mm_add_pd(fjy0,ty);
 931             fjz0             = _mm_add_pd(fjz0,tz);
 932
 933             }
 934
 935             /**************************
 936              * CALCULATE INTERACTIONS *
 937              **************************/
 938
 939             if (gmx_mm_any_lt(rsq20,rcutoff2))
 940             {
 941
 942             r20              = _mm_mul_pd(rsq20,rinv20);
 943
 944             /* Compute parameters for interactions between i and j atoms */
 945             qq20             = _mm_mul_pd(iq2,jq0);
 946
 947             /* EWALD ELECTROSTATICS */
 948
 949             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 950             ewrt             = _mm_mul_pd(r20,ewtabscale);
 951             ewitab           = _mm_cvttpd_epi32(ewrt);
 952             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 953             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 954                                          &ewtabF,&ewtabFn);
 955             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 956             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 957
 958             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 959
 960             fscal            = felec;
 961
 962             fscal            = _mm_and_pd(fscal,cutoff_mask);
 963
 964             /* Calculate temporary vectorial force */
 965             tx               = _mm_mul_pd(fscal,dx20);
 966             ty               = _mm_mul_pd(fscal,dy20);
 967             tz               = _mm_mul_pd(fscal,dz20);
 968
 969             /* Update vectorial force */
 970             fix2             = _mm_add_pd(fix2,tx);
 971             fiy2             = _mm_add_pd(fiy2,ty);
 972             fiz2             = _mm_add_pd(fiz2,tz);
 973
 974             fjx0             = _mm_add_pd(fjx0,tx);
 975             fjy0             = _mm_add_pd(fjy0,ty);
 976             fjz0             = _mm_add_pd(fjz0,tz);
 977
 978             }
 979
 980             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 981
 982             /* Inner loop uses 127 flops */
 983         }
 984
 985         if(jidx<j_index_end)
 986         {
 987
 988             jnrA             = jjnr[jidx];
 989             j_coord_offsetA  = DIM*jnrA;
 990
 991             /* load j atom coordinates */
 992             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 993                                               &jx0,&jy0,&jz0);
 994
 995             /* Calculate displacement vector */
 996             dx00             = _mm_sub_pd(ix0,jx0);
 997             dy00             = _mm_sub_pd(iy0,jy0);
 998             dz00             = _mm_sub_pd(iz0,jz0);
 999             dx10             = _mm_sub_pd(ix1,jx0);
1000             dy10             = _mm_sub_pd(iy1,jy0);
1001             dz10             = _mm_sub_pd(iz1,jz0);
1002             dx20             = _mm_sub_pd(ix2,jx0);
1003             dy20             = _mm_sub_pd(iy2,jy0);
1004             dz20             = _mm_sub_pd(iz2,jz0);
1005
1006             /* Calculate squared distance and things based on it */
1007             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1008             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1009             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1010
1011             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1012             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1013             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1014
1015             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1016             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1017             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1018
1019             /* Load parameters for j particles */
1020             jq0              = _mm_load_sd(charge+jnrA+0);
1021             vdwjidx0A        = 2*vdwtype[jnrA+0];
1022
1023             fjx0             = _mm_setzero_pd();
1024             fjy0             = _mm_setzero_pd();
1025             fjz0             = _mm_setzero_pd();
1026
1027             /**************************
1028              * CALCULATE INTERACTIONS *
1029              **************************/
1030
1031             if (gmx_mm_any_lt(rsq00,rcutoff2))
1032             {
1033
1034             r00              = _mm_mul_pd(rsq00,rinv00);
1035
1036             /* Compute parameters for interactions between i and j atoms */
1037             qq00             = _mm_mul_pd(iq0,jq0);
1038             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1039
1040             /* EWALD ELECTROSTATICS */
1041
1042             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1043             ewrt             = _mm_mul_pd(r00,ewtabscale);
1044             ewitab           = _mm_cvttpd_epi32(ewrt);
1045             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1046             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1047             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1048             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1049
1050             /* LENNARD-JONES DISPERSION/REPULSION */
1051
1052             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1053             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1054
1055             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1056
1057             fscal            = _mm_add_pd(felec,fvdw);
1058
1059             fscal            = _mm_and_pd(fscal,cutoff_mask);
1060
1061             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1062
1063             /* Calculate temporary vectorial force */
1064             tx               = _mm_mul_pd(fscal,dx00);
1065             ty               = _mm_mul_pd(fscal,dy00);
1066             tz               = _mm_mul_pd(fscal,dz00);
1067
1068             /* Update vectorial force */
1069             fix0             = _mm_add_pd(fix0,tx);
1070             fiy0             = _mm_add_pd(fiy0,ty);
1071             fiz0             = _mm_add_pd(fiz0,tz);
1072
1073             fjx0             = _mm_add_pd(fjx0,tx);
1074             fjy0             = _mm_add_pd(fjy0,ty);
1075             fjz0             = _mm_add_pd(fjz0,tz);
1076
1077             }
1078
1079             /**************************
1080              * CALCULATE INTERACTIONS *
1081              **************************/
1082
1083             if (gmx_mm_any_lt(rsq10,rcutoff2))
1084             {
1085
1086             r10              = _mm_mul_pd(rsq10,rinv10);
1087
1088             /* Compute parameters for interactions between i and j atoms */
1089             qq10             = _mm_mul_pd(iq1,jq0);
1090
1091             /* EWALD ELECTROSTATICS */
1092
1093             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1094             ewrt             = _mm_mul_pd(r10,ewtabscale);
1095             ewitab           = _mm_cvttpd_epi32(ewrt);
1096             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1097             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1098             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1099             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1100
1101             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1102
1103             fscal            = felec;
1104
1105             fscal            = _mm_and_pd(fscal,cutoff_mask);
1106
1107             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1108
1109             /* Calculate temporary vectorial force */
1110             tx               = _mm_mul_pd(fscal,dx10);
1111             ty               = _mm_mul_pd(fscal,dy10);
1112             tz               = _mm_mul_pd(fscal,dz10);
1113
1114             /* Update vectorial force */
1115             fix1             = _mm_add_pd(fix1,tx);
1116             fiy1             = _mm_add_pd(fiy1,ty);
1117             fiz1             = _mm_add_pd(fiz1,tz);
1118
1119             fjx0             = _mm_add_pd(fjx0,tx);
1120             fjy0             = _mm_add_pd(fjy0,ty);
1121             fjz0             = _mm_add_pd(fjz0,tz);
1122
1123             }
1124
1125             /**************************
1126              * CALCULATE INTERACTIONS *
1127              **************************/
1128
1129             if (gmx_mm_any_lt(rsq20,rcutoff2))
1130             {
1131
1132             r20              = _mm_mul_pd(rsq20,rinv20);
1133
1134             /* Compute parameters for interactions between i and j atoms */
1135             qq20             = _mm_mul_pd(iq2,jq0);
1136
1137             /* EWALD ELECTROSTATICS */
1138
1139             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1140             ewrt             = _mm_mul_pd(r20,ewtabscale);
1141             ewitab           = _mm_cvttpd_epi32(ewrt);
1142             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1143             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1144             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1145             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1146
1147             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1148
1149             fscal            = felec;
1150
1151             fscal            = _mm_and_pd(fscal,cutoff_mask);
1152
1153             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1154
1155             /* Calculate temporary vectorial force */
1156             tx               = _mm_mul_pd(fscal,dx20);
1157             ty               = _mm_mul_pd(fscal,dy20);
1158             tz               = _mm_mul_pd(fscal,dz20);
1159
1160             /* Update vectorial force */
1161             fix2             = _mm_add_pd(fix2,tx);
1162             fiy2             = _mm_add_pd(fiy2,ty);
1163             fiz2             = _mm_add_pd(fiz2,tz);
1164
1165             fjx0             = _mm_add_pd(fjx0,tx);
1166             fjy0             = _mm_add_pd(fjy0,ty);
1167             fjz0             = _mm_add_pd(fjz0,tz);
1168
1169             }
1170
1171             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1172
1173             /* Inner loop uses 127 flops */
1174         }
1175
1176         /* End of innermost loop */
1177
1178         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1179                                               f+i_coord_offset,fshift+i_shift_offset);
1180
1181         /* Increment number of inner iterations */
1182         inneriter                  += j_index_end - j_index_start;
1183
1184         /* Outer loop uses 18 flops */
1185     }
1186
1187     /* Increment number of outer iterations */
1188     outeriter        += nri;
1189
1190     /* Update outer/inner flops */
1191
1192     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*127);
1193 }