src/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecEw_VdwLJ_GeomW3P1_sse2_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse2_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_sse2_double.h"
  34 #include "kernelutil_x86_sse2_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3P1_VF_sse2_double
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEw_VdwLJ_GeomW3P1_VF_sse2_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     /* Avoid stupid compiler warnings */
 122     jnrA = jnrB = 0;
 123     j_coord_offsetA = 0;
 124     j_coord_offsetB = 0;
 125
 126     outeriter        = 0;
 127     inneriter        = 0;
 128
 129     /* Start outer loop over neighborlists */
 130     for(iidx=0; iidx<nri; iidx++)
 131     {
 132         /* Load shift vector for this list */
 133         i_shift_offset   = DIM*shiftidx[iidx];
 134
 135         /* Load limits for loop over neighbors */
 136         j_index_start    = jindex[iidx];
 137         j_index_end      = jindex[iidx+1];
 138
 139         /* Get outer coordinate index */
 140         inr              = iinr[iidx];
 141         i_coord_offset   = DIM*inr;
 142
 143         /* Load i particle coords and add shift vector */
 144         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 145                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 146
 147         fix0             = _mm_setzero_pd();
 148         fiy0             = _mm_setzero_pd();
 149         fiz0             = _mm_setzero_pd();
 150         fix1             = _mm_setzero_pd();
 151         fiy1             = _mm_setzero_pd();
 152         fiz1             = _mm_setzero_pd();
 153         fix2             = _mm_setzero_pd();
 154         fiy2             = _mm_setzero_pd();
 155         fiz2             = _mm_setzero_pd();
 156
 157         /* Reset potential sums */
 158         velecsum         = _mm_setzero_pd();
 159         vvdwsum          = _mm_setzero_pd();
 160
 161         /* Start inner kernel loop */
 162         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 163         {
 164
 165             /* Get j neighbor index, and coordinate index */
 166             jnrA             = jjnr[jidx];
 167             jnrB             = jjnr[jidx+1];
 168             j_coord_offsetA  = DIM*jnrA;
 169             j_coord_offsetB  = DIM*jnrB;
 170
 171             /* load j atom coordinates */
 172             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 173                                               &jx0,&jy0,&jz0);
 174
 175             /* Calculate displacement vector */
 176             dx00             = _mm_sub_pd(ix0,jx0);
 177             dy00             = _mm_sub_pd(iy0,jy0);
 178             dz00             = _mm_sub_pd(iz0,jz0);
 179             dx10             = _mm_sub_pd(ix1,jx0);
 180             dy10             = _mm_sub_pd(iy1,jy0);
 181             dz10             = _mm_sub_pd(iz1,jz0);
 182             dx20             = _mm_sub_pd(ix2,jx0);
 183             dy20             = _mm_sub_pd(iy2,jy0);
 184             dz20             = _mm_sub_pd(iz2,jz0);
 185
 186             /* Calculate squared distance and things based on it */
 187             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 188             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 189             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 190
 191             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 192             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 193             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 194
 195             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 196             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 197             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 198
 199             /* Load parameters for j particles */
 200             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 201             vdwjidx0A        = 2*vdwtype[jnrA+0];
 202             vdwjidx0B        = 2*vdwtype[jnrB+0];
 203
 204             fjx0             = _mm_setzero_pd();
 205             fjy0             = _mm_setzero_pd();
 206             fjz0             = _mm_setzero_pd();
 207
 208             /**************************
 209              * CALCULATE INTERACTIONS *
 210              **************************/
 211
 212             r00              = _mm_mul_pd(rsq00,rinv00);
 213
 214             /* Compute parameters for interactions between i and j atoms */
 215             qq00             = _mm_mul_pd(iq0,jq0);
 216             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 217                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 218
 219             /* EWALD ELECTROSTATICS */
 220
 221             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 222             ewrt             = _mm_mul_pd(r00,ewtabscale);
 223             ewitab           = _mm_cvttpd_epi32(ewrt);
 224             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 225             ewitab           = _mm_slli_epi32(ewitab,2);
 226             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 227             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 228             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 229             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 230             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 231             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 232             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 233             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 234             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 235             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 236
 237             /* LENNARD-JONES DISPERSION/REPULSION */
 238
 239             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 240             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 241             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 242             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 243             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 244
 245             /* Update potential sum for this i atom from the interaction with this j atom. */
 246             velecsum         = _mm_add_pd(velecsum,velec);
 247             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 248
 249             fscal            = _mm_add_pd(felec,fvdw);
 250
 251             /* Calculate temporary vectorial force */
 252             tx               = _mm_mul_pd(fscal,dx00);
 253             ty               = _mm_mul_pd(fscal,dy00);
 254             tz               = _mm_mul_pd(fscal,dz00);
 255
 256             /* Update vectorial force */
 257             fix0             = _mm_add_pd(fix0,tx);
 258             fiy0             = _mm_add_pd(fiy0,ty);
 259             fiz0             = _mm_add_pd(fiz0,tz);
 260
 261             fjx0             = _mm_add_pd(fjx0,tx);
 262             fjy0             = _mm_add_pd(fjy0,ty);
 263             fjz0             = _mm_add_pd(fjz0,tz);
 264
 265             /**************************
 266              * CALCULATE INTERACTIONS *
 267              **************************/
 268
 269             r10              = _mm_mul_pd(rsq10,rinv10);
 270
 271             /* Compute parameters for interactions between i and j atoms */
 272             qq10             = _mm_mul_pd(iq1,jq0);
 273
 274             /* EWALD ELECTROSTATICS */
 275
 276             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 277             ewrt             = _mm_mul_pd(r10,ewtabscale);
 278             ewitab           = _mm_cvttpd_epi32(ewrt);
 279             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 280             ewitab           = _mm_slli_epi32(ewitab,2);
 281             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 282             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 283             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 284             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 285             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 286             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 287             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 288             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 289             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 290             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 291
 292             /* Update potential sum for this i atom from the interaction with this j atom. */
 293             velecsum         = _mm_add_pd(velecsum,velec);
 294
 295             fscal            = felec;
 296
 297             /* Calculate temporary vectorial force */
 298             tx               = _mm_mul_pd(fscal,dx10);
 299             ty               = _mm_mul_pd(fscal,dy10);
 300             tz               = _mm_mul_pd(fscal,dz10);
 301
 302             /* Update vectorial force */
 303             fix1             = _mm_add_pd(fix1,tx);
 304             fiy1             = _mm_add_pd(fiy1,ty);
 305             fiz1             = _mm_add_pd(fiz1,tz);
 306
 307             fjx0             = _mm_add_pd(fjx0,tx);
 308             fjy0             = _mm_add_pd(fjy0,ty);
 309             fjz0             = _mm_add_pd(fjz0,tz);
 310
 311             /**************************
 312              * CALCULATE INTERACTIONS *
 313              **************************/
 314
 315             r20              = _mm_mul_pd(rsq20,rinv20);
 316
 317             /* Compute parameters for interactions between i and j atoms */
 318             qq20             = _mm_mul_pd(iq2,jq0);
 319
 320             /* EWALD ELECTROSTATICS */
 321
 322             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 323             ewrt             = _mm_mul_pd(r20,ewtabscale);
 324             ewitab           = _mm_cvttpd_epi32(ewrt);
 325             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 326             ewitab           = _mm_slli_epi32(ewitab,2);
 327             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 328             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 329             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 330             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 331             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 332             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 333             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 334             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 335             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 336             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 337
 338             /* Update potential sum for this i atom from the interaction with this j atom. */
 339             velecsum         = _mm_add_pd(velecsum,velec);
 340
 341             fscal            = felec;
 342
 343             /* Calculate temporary vectorial force */
 344             tx               = _mm_mul_pd(fscal,dx20);
 345             ty               = _mm_mul_pd(fscal,dy20);
 346             tz               = _mm_mul_pd(fscal,dz20);
 347
 348             /* Update vectorial force */
 349             fix2             = _mm_add_pd(fix2,tx);
 350             fiy2             = _mm_add_pd(fiy2,ty);
 351             fiz2             = _mm_add_pd(fiz2,tz);
 352
 353             fjx0             = _mm_add_pd(fjx0,tx);
 354             fjy0             = _mm_add_pd(fjy0,ty);
 355             fjz0             = _mm_add_pd(fjz0,tz);
 356
 357             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 358
 359             /* Inner loop uses 138 flops */
 360         }
 361
 362         if(jidx<j_index_end)
 363         {
 364
 365             jnrA             = jjnr[jidx];
 366             j_coord_offsetA  = DIM*jnrA;
 367
 368             /* load j atom coordinates */
 369             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 370                                               &jx0,&jy0,&jz0);
 371
 372             /* Calculate displacement vector */
 373             dx00             = _mm_sub_pd(ix0,jx0);
 374             dy00             = _mm_sub_pd(iy0,jy0);
 375             dz00             = _mm_sub_pd(iz0,jz0);
 376             dx10             = _mm_sub_pd(ix1,jx0);
 377             dy10             = _mm_sub_pd(iy1,jy0);
 378             dz10             = _mm_sub_pd(iz1,jz0);
 379             dx20             = _mm_sub_pd(ix2,jx0);
 380             dy20             = _mm_sub_pd(iy2,jy0);
 381             dz20             = _mm_sub_pd(iz2,jz0);
 382
 383             /* Calculate squared distance and things based on it */
 384             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 385             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 386             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 387
 388             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 389             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 390             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 391
 392             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 393             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 394             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 395
 396             /* Load parameters for j particles */
 397             jq0              = _mm_load_sd(charge+jnrA+0);
 398             vdwjidx0A        = 2*vdwtype[jnrA+0];
 399
 400             fjx0             = _mm_setzero_pd();
 401             fjy0             = _mm_setzero_pd();
 402             fjz0             = _mm_setzero_pd();
 403
 404             /**************************
 405              * CALCULATE INTERACTIONS *
 406              **************************/
 407
 408             r00              = _mm_mul_pd(rsq00,rinv00);
 409
 410             /* Compute parameters for interactions between i and j atoms */
 411             qq00             = _mm_mul_pd(iq0,jq0);
 412             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 413
 414             /* EWALD ELECTROSTATICS */
 415
 416             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 417             ewrt             = _mm_mul_pd(r00,ewtabscale);
 418             ewitab           = _mm_cvttpd_epi32(ewrt);
 419             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 420             ewitab           = _mm_slli_epi32(ewitab,2);
 421             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 422             ewtabD           = _mm_setzero_pd();
 423             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 424             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 425             ewtabFn          = _mm_setzero_pd();
 426             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 427             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 428             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 429             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 430             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 431
 432             /* LENNARD-JONES DISPERSION/REPULSION */
 433
 434             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 435             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 436             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 437             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 438             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 439
 440             /* Update potential sum for this i atom from the interaction with this j atom. */
 441             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 442             velecsum         = _mm_add_pd(velecsum,velec);
 443             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 444             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 445
 446             fscal            = _mm_add_pd(felec,fvdw);
 447
 448             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 449
 450             /* Calculate temporary vectorial force */
 451             tx               = _mm_mul_pd(fscal,dx00);
 452             ty               = _mm_mul_pd(fscal,dy00);
 453             tz               = _mm_mul_pd(fscal,dz00);
 454
 455             /* Update vectorial force */
 456             fix0             = _mm_add_pd(fix0,tx);
 457             fiy0             = _mm_add_pd(fiy0,ty);
 458             fiz0             = _mm_add_pd(fiz0,tz);
 459
 460             fjx0             = _mm_add_pd(fjx0,tx);
 461             fjy0             = _mm_add_pd(fjy0,ty);
 462             fjz0             = _mm_add_pd(fjz0,tz);
 463
 464             /**************************
 465              * CALCULATE INTERACTIONS *
 466              **************************/
 467
 468             r10              = _mm_mul_pd(rsq10,rinv10);
 469
 470             /* Compute parameters for interactions between i and j atoms */
 471             qq10             = _mm_mul_pd(iq1,jq0);
 472
 473             /* EWALD ELECTROSTATICS */
 474
 475             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 476             ewrt             = _mm_mul_pd(r10,ewtabscale);
 477             ewitab           = _mm_cvttpd_epi32(ewrt);
 478             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 479             ewitab           = _mm_slli_epi32(ewitab,2);
 480             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 481             ewtabD           = _mm_setzero_pd();
 482             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 483             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 484             ewtabFn          = _mm_setzero_pd();
 485             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 486             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 487             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 488             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 489             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 490
 491             /* Update potential sum for this i atom from the interaction with this j atom. */
 492             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 493             velecsum         = _mm_add_pd(velecsum,velec);
 494
 495             fscal            = felec;
 496
 497             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 498
 499             /* Calculate temporary vectorial force */
 500             tx               = _mm_mul_pd(fscal,dx10);
 501             ty               = _mm_mul_pd(fscal,dy10);
 502             tz               = _mm_mul_pd(fscal,dz10);
 503
 504             /* Update vectorial force */
 505             fix1             = _mm_add_pd(fix1,tx);
 506             fiy1             = _mm_add_pd(fiy1,ty);
 507             fiz1             = _mm_add_pd(fiz1,tz);
 508
 509             fjx0             = _mm_add_pd(fjx0,tx);
 510             fjy0             = _mm_add_pd(fjy0,ty);
 511             fjz0             = _mm_add_pd(fjz0,tz);
 512
 513             /**************************
 514              * CALCULATE INTERACTIONS *
 515              **************************/
 516
 517             r20              = _mm_mul_pd(rsq20,rinv20);
 518
 519             /* Compute parameters for interactions between i and j atoms */
 520             qq20             = _mm_mul_pd(iq2,jq0);
 521
 522             /* EWALD ELECTROSTATICS */
 523
 524             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 525             ewrt             = _mm_mul_pd(r20,ewtabscale);
 526             ewitab           = _mm_cvttpd_epi32(ewrt);
 527             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 528             ewitab           = _mm_slli_epi32(ewitab,2);
 529             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 530             ewtabD           = _mm_setzero_pd();
 531             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 532             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 533             ewtabFn          = _mm_setzero_pd();
 534             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 535             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 536             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 537             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 538             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 539
 540             /* Update potential sum for this i atom from the interaction with this j atom. */
 541             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 542             velecsum         = _mm_add_pd(velecsum,velec);
 543
 544             fscal            = felec;
 545
 546             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 547
 548             /* Calculate temporary vectorial force */
 549             tx               = _mm_mul_pd(fscal,dx20);
 550             ty               = _mm_mul_pd(fscal,dy20);
 551             tz               = _mm_mul_pd(fscal,dz20);
 552
 553             /* Update vectorial force */
 554             fix2             = _mm_add_pd(fix2,tx);
 555             fiy2             = _mm_add_pd(fiy2,ty);
 556             fiz2             = _mm_add_pd(fiz2,tz);
 557
 558             fjx0             = _mm_add_pd(fjx0,tx);
 559             fjy0             = _mm_add_pd(fjy0,ty);
 560             fjz0             = _mm_add_pd(fjz0,tz);
 561
 562             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 563
 564             /* Inner loop uses 138 flops */
 565         }
 566
 567         /* End of innermost loop */
 568
 569         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 570                                               f+i_coord_offset,fshift+i_shift_offset);
 571
 572         ggid                        = gid[iidx];
 573         /* Update potential energies */
 574         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 575         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 576
 577         /* Increment number of inner iterations */
 578         inneriter                  += j_index_end - j_index_start;
 579
 580         /* Outer loop uses 20 flops */
 581     }
 582
 583     /* Increment number of outer iterations */
 584     outeriter        += nri;
 585
 586     /* Update outer/inner flops */
 587
 588     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*138);
 589 }
 590 /*
 591  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_sse2_double
 592  * Electrostatics interaction: Ewald
 593  * VdW interaction:            LennardJones
 594  * Geometry:                   Water3-Particle
 595  * Calculate force/pot:        Force
 596  */
 597 void
 598 nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_sse2_double
 599                     (t_nblist * gmx_restrict                nlist,
 600                      rvec * gmx_restrict                    xx,
 601                      rvec * gmx_restrict                    ff,
 602                      t_forcerec * gmx_restrict              fr,
 603                      t_mdatoms * gmx_restrict               mdatoms,
 604                      nb_kernel_data_t * gmx_restrict        kernel_data,
 605                      t_nrnb * gmx_restrict                  nrnb)
 606 {
 607     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 608      * just 0 for non-waters.
 609      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 610      * jnr indices corresponding to data put in the four positions in the SIMD register.
 611      */
 612     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 613     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 614     int              jnrA,jnrB;
 615     int              j_coord_offsetA,j_coord_offsetB;
 616     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 617     real             rcutoff_scalar;
 618     real             *shiftvec,*fshift,*x,*f;
 619     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 620     int              vdwioffset0;
 621     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 622     int              vdwioffset1;
 623     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 624     int              vdwioffset2;
 625     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 626     int              vdwjidx0A,vdwjidx0B;
 627     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 628     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 629     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 630     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 631     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 632     real             *charge;
 633     int              nvdwtype;
 634     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 635     int              *vdwtype;
 636     real             *vdwparam;
 637     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 638     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 639     __m128i          ewitab;
 640     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 641     real             *ewtab;
 642     __m128d          dummy_mask,cutoff_mask;
 643     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 644     __m128d          one     = _mm_set1_pd(1.0);
 645     __m128d          two     = _mm_set1_pd(2.0);
 646     x                = xx[0];
 647     f                = ff[0];
 648
 649     nri              = nlist->nri;
 650     iinr             = nlist->iinr;
 651     jindex           = nlist->jindex;
 652     jjnr             = nlist->jjnr;
 653     shiftidx         = nlist->shift;
 654     gid              = nlist->gid;
 655     shiftvec         = fr->shift_vec[0];
 656     fshift           = fr->fshift[0];
 657     facel            = _mm_set1_pd(fr->epsfac);
 658     charge           = mdatoms->chargeA;
 659     nvdwtype         = fr->ntype;
 660     vdwparam         = fr->nbfp;
 661     vdwtype          = mdatoms->typeA;
 662
 663     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 664     ewtab            = fr->ic->tabq_coul_F;
 665     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 666     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 667
 668     /* Setup water-specific parameters */
 669     inr              = nlist->iinr[0];
 670     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 671     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 672     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 673     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 674
 675     /* Avoid stupid compiler warnings */
 676     jnrA = jnrB = 0;
 677     j_coord_offsetA = 0;
 678     j_coord_offsetB = 0;
 679
 680     outeriter        = 0;
 681     inneriter        = 0;
 682
 683     /* Start outer loop over neighborlists */
 684     for(iidx=0; iidx<nri; iidx++)
 685     {
 686         /* Load shift vector for this list */
 687         i_shift_offset   = DIM*shiftidx[iidx];
 688
 689         /* Load limits for loop over neighbors */
 690         j_index_start    = jindex[iidx];
 691         j_index_end      = jindex[iidx+1];
 692
 693         /* Get outer coordinate index */
 694         inr              = iinr[iidx];
 695         i_coord_offset   = DIM*inr;
 696
 697         /* Load i particle coords and add shift vector */
 698         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 699                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 700
 701         fix0             = _mm_setzero_pd();
 702         fiy0             = _mm_setzero_pd();
 703         fiz0             = _mm_setzero_pd();
 704         fix1             = _mm_setzero_pd();
 705         fiy1             = _mm_setzero_pd();
 706         fiz1             = _mm_setzero_pd();
 707         fix2             = _mm_setzero_pd();
 708         fiy2             = _mm_setzero_pd();
 709         fiz2             = _mm_setzero_pd();
 710
 711         /* Start inner kernel loop */
 712         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 713         {
 714
 715             /* Get j neighbor index, and coordinate index */
 716             jnrA             = jjnr[jidx];
 717             jnrB             = jjnr[jidx+1];
 718             j_coord_offsetA  = DIM*jnrA;
 719             j_coord_offsetB  = DIM*jnrB;
 720
 721             /* load j atom coordinates */
 722             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 723                                               &jx0,&jy0,&jz0);
 724
 725             /* Calculate displacement vector */
 726             dx00             = _mm_sub_pd(ix0,jx0);
 727             dy00             = _mm_sub_pd(iy0,jy0);
 728             dz00             = _mm_sub_pd(iz0,jz0);
 729             dx10             = _mm_sub_pd(ix1,jx0);
 730             dy10             = _mm_sub_pd(iy1,jy0);
 731             dz10             = _mm_sub_pd(iz1,jz0);
 732             dx20             = _mm_sub_pd(ix2,jx0);
 733             dy20             = _mm_sub_pd(iy2,jy0);
 734             dz20             = _mm_sub_pd(iz2,jz0);
 735
 736             /* Calculate squared distance and things based on it */
 737             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 738             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 739             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 740
 741             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 742             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 743             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 744
 745             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 746             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 747             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 748
 749             /* Load parameters for j particles */
 750             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 751             vdwjidx0A        = 2*vdwtype[jnrA+0];
 752             vdwjidx0B        = 2*vdwtype[jnrB+0];
 753
 754             fjx0             = _mm_setzero_pd();
 755             fjy0             = _mm_setzero_pd();
 756             fjz0             = _mm_setzero_pd();
 757
 758             /**************************
 759              * CALCULATE INTERACTIONS *
 760              **************************/
 761
 762             r00              = _mm_mul_pd(rsq00,rinv00);
 763
 764             /* Compute parameters for interactions between i and j atoms */
 765             qq00             = _mm_mul_pd(iq0,jq0);
 766             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 767                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 768
 769             /* EWALD ELECTROSTATICS */
 770
 771             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 772             ewrt             = _mm_mul_pd(r00,ewtabscale);
 773             ewitab           = _mm_cvttpd_epi32(ewrt);
 774             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 775             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 776                                          &ewtabF,&ewtabFn);
 777             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 778             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 779
 780             /* LENNARD-JONES DISPERSION/REPULSION */
 781
 782             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 783             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 784
 785             fscal            = _mm_add_pd(felec,fvdw);
 786
 787             /* Calculate temporary vectorial force */
 788             tx               = _mm_mul_pd(fscal,dx00);
 789             ty               = _mm_mul_pd(fscal,dy00);
 790             tz               = _mm_mul_pd(fscal,dz00);
 791
 792             /* Update vectorial force */
 793             fix0             = _mm_add_pd(fix0,tx);
 794             fiy0             = _mm_add_pd(fiy0,ty);
 795             fiz0             = _mm_add_pd(fiz0,tz);
 796
 797             fjx0             = _mm_add_pd(fjx0,tx);
 798             fjy0             = _mm_add_pd(fjy0,ty);
 799             fjz0             = _mm_add_pd(fjz0,tz);
 800
 801             /**************************
 802              * CALCULATE INTERACTIONS *
 803              **************************/
 804
 805             r10              = _mm_mul_pd(rsq10,rinv10);
 806
 807             /* Compute parameters for interactions between i and j atoms */
 808             qq10             = _mm_mul_pd(iq1,jq0);
 809
 810             /* EWALD ELECTROSTATICS */
 811
 812             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 813             ewrt             = _mm_mul_pd(r10,ewtabscale);
 814             ewitab           = _mm_cvttpd_epi32(ewrt);
 815             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 816             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 817                                          &ewtabF,&ewtabFn);
 818             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 819             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 820
 821             fscal            = felec;
 822
 823             /* Calculate temporary vectorial force */
 824             tx               = _mm_mul_pd(fscal,dx10);
 825             ty               = _mm_mul_pd(fscal,dy10);
 826             tz               = _mm_mul_pd(fscal,dz10);
 827
 828             /* Update vectorial force */
 829             fix1             = _mm_add_pd(fix1,tx);
 830             fiy1             = _mm_add_pd(fiy1,ty);
 831             fiz1             = _mm_add_pd(fiz1,tz);
 832
 833             fjx0             = _mm_add_pd(fjx0,tx);
 834             fjy0             = _mm_add_pd(fjy0,ty);
 835             fjz0             = _mm_add_pd(fjz0,tz);
 836
 837             /**************************
 838              * CALCULATE INTERACTIONS *
 839              **************************/
 840
 841             r20              = _mm_mul_pd(rsq20,rinv20);
 842
 843             /* Compute parameters for interactions between i and j atoms */
 844             qq20             = _mm_mul_pd(iq2,jq0);
 845
 846             /* EWALD ELECTROSTATICS */
 847
 848             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 849             ewrt             = _mm_mul_pd(r20,ewtabscale);
 850             ewitab           = _mm_cvttpd_epi32(ewrt);
 851             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 852             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 853                                          &ewtabF,&ewtabFn);
 854             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 855             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 856
 857             fscal            = felec;
 858
 859             /* Calculate temporary vectorial force */
 860             tx               = _mm_mul_pd(fscal,dx20);
 861             ty               = _mm_mul_pd(fscal,dy20);
 862             tz               = _mm_mul_pd(fscal,dz20);
 863
 864             /* Update vectorial force */
 865             fix2             = _mm_add_pd(fix2,tx);
 866             fiy2             = _mm_add_pd(fiy2,ty);
 867             fiz2             = _mm_add_pd(fiz2,tz);
 868
 869             fjx0             = _mm_add_pd(fjx0,tx);
 870             fjy0             = _mm_add_pd(fjy0,ty);
 871             fjz0             = _mm_add_pd(fjz0,tz);
 872
 873             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 874
 875             /* Inner loop uses 118 flops */
 876         }
 877
 878         if(jidx<j_index_end)
 879         {
 880
 881             jnrA             = jjnr[jidx];
 882             j_coord_offsetA  = DIM*jnrA;
 883
 884             /* load j atom coordinates */
 885             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 886                                               &jx0,&jy0,&jz0);
 887
 888             /* Calculate displacement vector */
 889             dx00             = _mm_sub_pd(ix0,jx0);
 890             dy00             = _mm_sub_pd(iy0,jy0);
 891             dz00             = _mm_sub_pd(iz0,jz0);
 892             dx10             = _mm_sub_pd(ix1,jx0);
 893             dy10             = _mm_sub_pd(iy1,jy0);
 894             dz10             = _mm_sub_pd(iz1,jz0);
 895             dx20             = _mm_sub_pd(ix2,jx0);
 896             dy20             = _mm_sub_pd(iy2,jy0);
 897             dz20             = _mm_sub_pd(iz2,jz0);
 898
 899             /* Calculate squared distance and things based on it */
 900             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 901             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 902             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 903
 904             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 905             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 906             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 907
 908             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 909             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 910             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 911
 912             /* Load parameters for j particles */
 913             jq0              = _mm_load_sd(charge+jnrA+0);
 914             vdwjidx0A        = 2*vdwtype[jnrA+0];
 915
 916             fjx0             = _mm_setzero_pd();
 917             fjy0             = _mm_setzero_pd();
 918             fjz0             = _mm_setzero_pd();
 919
 920             /**************************
 921              * CALCULATE INTERACTIONS *
 922              **************************/
 923
 924             r00              = _mm_mul_pd(rsq00,rinv00);
 925
 926             /* Compute parameters for interactions between i and j atoms */
 927             qq00             = _mm_mul_pd(iq0,jq0);
 928             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 929
 930             /* EWALD ELECTROSTATICS */
 931
 932             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 933             ewrt             = _mm_mul_pd(r00,ewtabscale);
 934             ewitab           = _mm_cvttpd_epi32(ewrt);
 935             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 936             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 937             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 938             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 939
 940             /* LENNARD-JONES DISPERSION/REPULSION */
 941
 942             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 943             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 944
 945             fscal            = _mm_add_pd(felec,fvdw);
 946
 947             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 948
 949             /* Calculate temporary vectorial force */
 950             tx               = _mm_mul_pd(fscal,dx00);
 951             ty               = _mm_mul_pd(fscal,dy00);
 952             tz               = _mm_mul_pd(fscal,dz00);
 953
 954             /* Update vectorial force */
 955             fix0             = _mm_add_pd(fix0,tx);
 956             fiy0             = _mm_add_pd(fiy0,ty);
 957             fiz0             = _mm_add_pd(fiz0,tz);
 958
 959             fjx0             = _mm_add_pd(fjx0,tx);
 960             fjy0             = _mm_add_pd(fjy0,ty);
 961             fjz0             = _mm_add_pd(fjz0,tz);
 962
 963             /**************************
 964              * CALCULATE INTERACTIONS *
 965              **************************/
 966
 967             r10              = _mm_mul_pd(rsq10,rinv10);
 968
 969             /* Compute parameters for interactions between i and j atoms */
 970             qq10             = _mm_mul_pd(iq1,jq0);
 971
 972             /* EWALD ELECTROSTATICS */
 973
 974             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 975             ewrt             = _mm_mul_pd(r10,ewtabscale);
 976             ewitab           = _mm_cvttpd_epi32(ewrt);
 977             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 978             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 979             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 980             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 981
 982             fscal            = felec;
 983
 984             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 985
 986             /* Calculate temporary vectorial force */
 987             tx               = _mm_mul_pd(fscal,dx10);
 988             ty               = _mm_mul_pd(fscal,dy10);
 989             tz               = _mm_mul_pd(fscal,dz10);
 990
 991             /* Update vectorial force */
 992             fix1             = _mm_add_pd(fix1,tx);
 993             fiy1             = _mm_add_pd(fiy1,ty);
 994             fiz1             = _mm_add_pd(fiz1,tz);
 995
 996             fjx0             = _mm_add_pd(fjx0,tx);
 997             fjy0             = _mm_add_pd(fjy0,ty);
 998             fjz0             = _mm_add_pd(fjz0,tz);
 999
1000             /**************************
1001              * CALCULATE INTERACTIONS *
1002              **************************/
1003
1004             r20              = _mm_mul_pd(rsq20,rinv20);
1005
1006             /* Compute parameters for interactions between i and j atoms */
1007             qq20             = _mm_mul_pd(iq2,jq0);
1008
1009             /* EWALD ELECTROSTATICS */
1010
1011             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1012             ewrt             = _mm_mul_pd(r20,ewtabscale);
1013             ewitab           = _mm_cvttpd_epi32(ewrt);
1014             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1015             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1016             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1017             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1018
1019             fscal            = felec;
1020
1021             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1022
1023             /* Calculate temporary vectorial force */
1024             tx               = _mm_mul_pd(fscal,dx20);
1025             ty               = _mm_mul_pd(fscal,dy20);
1026             tz               = _mm_mul_pd(fscal,dz20);
1027
1028             /* Update vectorial force */
1029             fix2             = _mm_add_pd(fix2,tx);
1030             fiy2             = _mm_add_pd(fiy2,ty);
1031             fiz2             = _mm_add_pd(fiz2,tz);
1032
1033             fjx0             = _mm_add_pd(fjx0,tx);
1034             fjy0             = _mm_add_pd(fjy0,ty);
1035             fjz0             = _mm_add_pd(fjz0,tz);
1036
1037             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1038
1039             /* Inner loop uses 118 flops */
1040         }
1041
1042         /* End of innermost loop */
1043
1044         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1045                                               f+i_coord_offset,fshift+i_shift_offset);
1046
1047         /* Increment number of inner iterations */
1048         inneriter                  += j_index_end - j_index_start;
1049
1050         /* Outer loop uses 18 flops */
1051     }
1052
1053     /* Increment number of outer iterations */
1054     outeriter        += nri;
1055
1056     /* Update outer/inner flops */
1057
1058     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*118);
1059 }