src/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_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_ElecCoul_VdwCSTab_GeomW4P1_VF_sse4_1_double
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_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              vdwioffset3;
  73     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  74     int              vdwjidx0A,vdwjidx0B;
  75     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  76     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  77     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  78     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  79     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  80     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     int              nvdwtype;
  83     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  84     int              *vdwtype;
  85     real             *vdwparam;
  86     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  87     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  88     __m128i          vfitab;
  89     __m128i          ifour       = _mm_set1_epi32(4);
  90     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  91     real             *vftab;
  92     __m128d          dummy_mask,cutoff_mask;
  93     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  94     __m128d          one     = _mm_set1_pd(1.0);
  95     __m128d          two     = _mm_set1_pd(2.0);
  96     x                = xx[0];
  97     f                = ff[0];
  98
  99     nri              = nlist->nri;
 100     iinr             = nlist->iinr;
 101     jindex           = nlist->jindex;
 102     jjnr             = nlist->jjnr;
 103     shiftidx         = nlist->shift;
 104     gid              = nlist->gid;
 105     shiftvec         = fr->shift_vec[0];
 106     fshift           = fr->fshift[0];
 107     facel            = _mm_set1_pd(fr->epsfac);
 108     charge           = mdatoms->chargeA;
 109     nvdwtype         = fr->ntype;
 110     vdwparam         = fr->nbfp;
 111     vdwtype          = mdatoms->typeA;
 112
 113     vftab            = kernel_data->table_vdw->data;
 114     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 115
 116     /* Setup water-specific parameters */
 117     inr              = nlist->iinr[0];
 118     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 119     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 120     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 121     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 122
 123     /* Avoid stupid compiler warnings */
 124     jnrA = jnrB = 0;
 125     j_coord_offsetA = 0;
 126     j_coord_offsetB = 0;
 127
 128     outeriter        = 0;
 129     inneriter        = 0;
 130
 131     /* Start outer loop over neighborlists */
 132     for(iidx=0; iidx<nri; iidx++)
 133     {
 134         /* Load shift vector for this list */
 135         i_shift_offset   = DIM*shiftidx[iidx];
 136
 137         /* Load limits for loop over neighbors */
 138         j_index_start    = jindex[iidx];
 139         j_index_end      = jindex[iidx+1];
 140
 141         /* Get outer coordinate index */
 142         inr              = iinr[iidx];
 143         i_coord_offset   = DIM*inr;
 144
 145         /* Load i particle coords and add shift vector */
 146         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 147                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 148
 149         fix0             = _mm_setzero_pd();
 150         fiy0             = _mm_setzero_pd();
 151         fiz0             = _mm_setzero_pd();
 152         fix1             = _mm_setzero_pd();
 153         fiy1             = _mm_setzero_pd();
 154         fiz1             = _mm_setzero_pd();
 155         fix2             = _mm_setzero_pd();
 156         fiy2             = _mm_setzero_pd();
 157         fiz2             = _mm_setzero_pd();
 158         fix3             = _mm_setzero_pd();
 159         fiy3             = _mm_setzero_pd();
 160         fiz3             = _mm_setzero_pd();
 161
 162         /* Reset potential sums */
 163         velecsum         = _mm_setzero_pd();
 164         vvdwsum          = _mm_setzero_pd();
 165
 166         /* Start inner kernel loop */
 167         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 168         {
 169
 170             /* Get j neighbor index, and coordinate index */
 171             jnrA             = jjnr[jidx];
 172             jnrB             = jjnr[jidx+1];
 173             j_coord_offsetA  = DIM*jnrA;
 174             j_coord_offsetB  = DIM*jnrB;
 175
 176             /* load j atom coordinates */
 177             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 178                                               &jx0,&jy0,&jz0);
 179
 180             /* Calculate displacement vector */
 181             dx00             = _mm_sub_pd(ix0,jx0);
 182             dy00             = _mm_sub_pd(iy0,jy0);
 183             dz00             = _mm_sub_pd(iz0,jz0);
 184             dx10             = _mm_sub_pd(ix1,jx0);
 185             dy10             = _mm_sub_pd(iy1,jy0);
 186             dz10             = _mm_sub_pd(iz1,jz0);
 187             dx20             = _mm_sub_pd(ix2,jx0);
 188             dy20             = _mm_sub_pd(iy2,jy0);
 189             dz20             = _mm_sub_pd(iz2,jz0);
 190             dx30             = _mm_sub_pd(ix3,jx0);
 191             dy30             = _mm_sub_pd(iy3,jy0);
 192             dz30             = _mm_sub_pd(iz3,jz0);
 193
 194             /* Calculate squared distance and things based on it */
 195             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 196             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 197             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 198             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 199
 200             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 201             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 202             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 203             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 204
 205             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 206             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 207             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 208
 209             /* Load parameters for j particles */
 210             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 211             vdwjidx0A        = 2*vdwtype[jnrA+0];
 212             vdwjidx0B        = 2*vdwtype[jnrB+0];
 213
 214             fjx0             = _mm_setzero_pd();
 215             fjy0             = _mm_setzero_pd();
 216             fjz0             = _mm_setzero_pd();
 217
 218             /**************************
 219              * CALCULATE INTERACTIONS *
 220              **************************/
 221
 222             r00              = _mm_mul_pd(rsq00,rinv00);
 223
 224             /* Compute parameters for interactions between i and j atoms */
 225             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 226                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 227
 228             /* Calculate table index by multiplying r with table scale and truncate to integer */
 229             rt               = _mm_mul_pd(r00,vftabscale);
 230             vfitab           = _mm_cvttpd_epi32(rt);
 231             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 232             vfitab           = _mm_slli_epi32(vfitab,3);
 233
 234             /* CUBIC SPLINE TABLE DISPERSION */
 235             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 236             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 237             GMX_MM_TRANSPOSE2_PD(Y,F);
 238             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 239             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 240             GMX_MM_TRANSPOSE2_PD(G,H);
 241             Heps             = _mm_mul_pd(vfeps,H);
 242             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 243             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 244             vvdw6            = _mm_mul_pd(c6_00,VV);
 245             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 246             fvdw6            = _mm_mul_pd(c6_00,FF);
 247
 248             /* CUBIC SPLINE TABLE REPULSION */
 249             vfitab           = _mm_add_epi32(vfitab,ifour);
 250             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 251             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 252             GMX_MM_TRANSPOSE2_PD(Y,F);
 253             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 254             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 255             GMX_MM_TRANSPOSE2_PD(G,H);
 256             Heps             = _mm_mul_pd(vfeps,H);
 257             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 258             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 259             vvdw12           = _mm_mul_pd(c12_00,VV);
 260             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 261             fvdw12           = _mm_mul_pd(c12_00,FF);
 262             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 263             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 264
 265             /* Update potential sum for this i atom from the interaction with this j atom. */
 266             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 267
 268             fscal            = fvdw;
 269
 270             /* Calculate temporary vectorial force */
 271             tx               = _mm_mul_pd(fscal,dx00);
 272             ty               = _mm_mul_pd(fscal,dy00);
 273             tz               = _mm_mul_pd(fscal,dz00);
 274
 275             /* Update vectorial force */
 276             fix0             = _mm_add_pd(fix0,tx);
 277             fiy0             = _mm_add_pd(fiy0,ty);
 278             fiz0             = _mm_add_pd(fiz0,tz);
 279
 280             fjx0             = _mm_add_pd(fjx0,tx);
 281             fjy0             = _mm_add_pd(fjy0,ty);
 282             fjz0             = _mm_add_pd(fjz0,tz);
 283
 284             /**************************
 285              * CALCULATE INTERACTIONS *
 286              **************************/
 287
 288             /* Compute parameters for interactions between i and j atoms */
 289             qq10             = _mm_mul_pd(iq1,jq0);
 290
 291             /* COULOMB ELECTROSTATICS */
 292             velec            = _mm_mul_pd(qq10,rinv10);
 293             felec            = _mm_mul_pd(velec,rinvsq10);
 294
 295             /* Update potential sum for this i atom from the interaction with this j atom. */
 296             velecsum         = _mm_add_pd(velecsum,velec);
 297
 298             fscal            = felec;
 299
 300             /* Calculate temporary vectorial force */
 301             tx               = _mm_mul_pd(fscal,dx10);
 302             ty               = _mm_mul_pd(fscal,dy10);
 303             tz               = _mm_mul_pd(fscal,dz10);
 304
 305             /* Update vectorial force */
 306             fix1             = _mm_add_pd(fix1,tx);
 307             fiy1             = _mm_add_pd(fiy1,ty);
 308             fiz1             = _mm_add_pd(fiz1,tz);
 309
 310             fjx0             = _mm_add_pd(fjx0,tx);
 311             fjy0             = _mm_add_pd(fjy0,ty);
 312             fjz0             = _mm_add_pd(fjz0,tz);
 313
 314             /**************************
 315              * CALCULATE INTERACTIONS *
 316              **************************/
 317
 318             /* Compute parameters for interactions between i and j atoms */
 319             qq20             = _mm_mul_pd(iq2,jq0);
 320
 321             /* COULOMB ELECTROSTATICS */
 322             velec            = _mm_mul_pd(qq20,rinv20);
 323             felec            = _mm_mul_pd(velec,rinvsq20);
 324
 325             /* Update potential sum for this i atom from the interaction with this j atom. */
 326             velecsum         = _mm_add_pd(velecsum,velec);
 327
 328             fscal            = felec;
 329
 330             /* Calculate temporary vectorial force */
 331             tx               = _mm_mul_pd(fscal,dx20);
 332             ty               = _mm_mul_pd(fscal,dy20);
 333             tz               = _mm_mul_pd(fscal,dz20);
 334
 335             /* Update vectorial force */
 336             fix2             = _mm_add_pd(fix2,tx);
 337             fiy2             = _mm_add_pd(fiy2,ty);
 338             fiz2             = _mm_add_pd(fiz2,tz);
 339
 340             fjx0             = _mm_add_pd(fjx0,tx);
 341             fjy0             = _mm_add_pd(fjy0,ty);
 342             fjz0             = _mm_add_pd(fjz0,tz);
 343
 344             /**************************
 345              * CALCULATE INTERACTIONS *
 346              **************************/
 347
 348             /* Compute parameters for interactions between i and j atoms */
 349             qq30             = _mm_mul_pd(iq3,jq0);
 350
 351             /* COULOMB ELECTROSTATICS */
 352             velec            = _mm_mul_pd(qq30,rinv30);
 353             felec            = _mm_mul_pd(velec,rinvsq30);
 354
 355             /* Update potential sum for this i atom from the interaction with this j atom. */
 356             velecsum         = _mm_add_pd(velecsum,velec);
 357
 358             fscal            = felec;
 359
 360             /* Calculate temporary vectorial force */
 361             tx               = _mm_mul_pd(fscal,dx30);
 362             ty               = _mm_mul_pd(fscal,dy30);
 363             tz               = _mm_mul_pd(fscal,dz30);
 364
 365             /* Update vectorial force */
 366             fix3             = _mm_add_pd(fix3,tx);
 367             fiy3             = _mm_add_pd(fiy3,ty);
 368             fiz3             = _mm_add_pd(fiz3,tz);
 369
 370             fjx0             = _mm_add_pd(fjx0,tx);
 371             fjy0             = _mm_add_pd(fjy0,ty);
 372             fjz0             = _mm_add_pd(fjz0,tz);
 373
 374             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 375
 376             /* Inner loop uses 143 flops */
 377         }
 378
 379         if(jidx<j_index_end)
 380         {
 381
 382             jnrA             = jjnr[jidx];
 383             j_coord_offsetA  = DIM*jnrA;
 384
 385             /* load j atom coordinates */
 386             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 387                                               &jx0,&jy0,&jz0);
 388
 389             /* Calculate displacement vector */
 390             dx00             = _mm_sub_pd(ix0,jx0);
 391             dy00             = _mm_sub_pd(iy0,jy0);
 392             dz00             = _mm_sub_pd(iz0,jz0);
 393             dx10             = _mm_sub_pd(ix1,jx0);
 394             dy10             = _mm_sub_pd(iy1,jy0);
 395             dz10             = _mm_sub_pd(iz1,jz0);
 396             dx20             = _mm_sub_pd(ix2,jx0);
 397             dy20             = _mm_sub_pd(iy2,jy0);
 398             dz20             = _mm_sub_pd(iz2,jz0);
 399             dx30             = _mm_sub_pd(ix3,jx0);
 400             dy30             = _mm_sub_pd(iy3,jy0);
 401             dz30             = _mm_sub_pd(iz3,jz0);
 402
 403             /* Calculate squared distance and things based on it */
 404             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 405             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 406             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 407             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 408
 409             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 410             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 411             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 412             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 413
 414             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 415             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 416             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 417
 418             /* Load parameters for j particles */
 419             jq0              = _mm_load_sd(charge+jnrA+0);
 420             vdwjidx0A        = 2*vdwtype[jnrA+0];
 421
 422             fjx0             = _mm_setzero_pd();
 423             fjy0             = _mm_setzero_pd();
 424             fjz0             = _mm_setzero_pd();
 425
 426             /**************************
 427              * CALCULATE INTERACTIONS *
 428              **************************/
 429
 430             r00              = _mm_mul_pd(rsq00,rinv00);
 431
 432             /* Compute parameters for interactions between i and j atoms */
 433             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 434
 435             /* Calculate table index by multiplying r with table scale and truncate to integer */
 436             rt               = _mm_mul_pd(r00,vftabscale);
 437             vfitab           = _mm_cvttpd_epi32(rt);
 438             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 439             vfitab           = _mm_slli_epi32(vfitab,3);
 440
 441             /* CUBIC SPLINE TABLE DISPERSION */
 442             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 443             F                = _mm_setzero_pd();
 444             GMX_MM_TRANSPOSE2_PD(Y,F);
 445             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 446             H                = _mm_setzero_pd();
 447             GMX_MM_TRANSPOSE2_PD(G,H);
 448             Heps             = _mm_mul_pd(vfeps,H);
 449             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 450             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 451             vvdw6            = _mm_mul_pd(c6_00,VV);
 452             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 453             fvdw6            = _mm_mul_pd(c6_00,FF);
 454
 455             /* CUBIC SPLINE TABLE REPULSION */
 456             vfitab           = _mm_add_epi32(vfitab,ifour);
 457             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 458             F                = _mm_setzero_pd();
 459             GMX_MM_TRANSPOSE2_PD(Y,F);
 460             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 461             H                = _mm_setzero_pd();
 462             GMX_MM_TRANSPOSE2_PD(G,H);
 463             Heps             = _mm_mul_pd(vfeps,H);
 464             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 465             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 466             vvdw12           = _mm_mul_pd(c12_00,VV);
 467             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 468             fvdw12           = _mm_mul_pd(c12_00,FF);
 469             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 470             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 471
 472             /* Update potential sum for this i atom from the interaction with this j atom. */
 473             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 474             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 475
 476             fscal            = fvdw;
 477
 478             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 479
 480             /* Calculate temporary vectorial force */
 481             tx               = _mm_mul_pd(fscal,dx00);
 482             ty               = _mm_mul_pd(fscal,dy00);
 483             tz               = _mm_mul_pd(fscal,dz00);
 484
 485             /* Update vectorial force */
 486             fix0             = _mm_add_pd(fix0,tx);
 487             fiy0             = _mm_add_pd(fiy0,ty);
 488             fiz0             = _mm_add_pd(fiz0,tz);
 489
 490             fjx0             = _mm_add_pd(fjx0,tx);
 491             fjy0             = _mm_add_pd(fjy0,ty);
 492             fjz0             = _mm_add_pd(fjz0,tz);
 493
 494             /**************************
 495              * CALCULATE INTERACTIONS *
 496              **************************/
 497
 498             /* Compute parameters for interactions between i and j atoms */
 499             qq10             = _mm_mul_pd(iq1,jq0);
 500
 501             /* COULOMB ELECTROSTATICS */
 502             velec            = _mm_mul_pd(qq10,rinv10);
 503             felec            = _mm_mul_pd(velec,rinvsq10);
 504
 505             /* Update potential sum for this i atom from the interaction with this j atom. */
 506             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 507             velecsum         = _mm_add_pd(velecsum,velec);
 508
 509             fscal            = felec;
 510
 511             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 512
 513             /* Calculate temporary vectorial force */
 514             tx               = _mm_mul_pd(fscal,dx10);
 515             ty               = _mm_mul_pd(fscal,dy10);
 516             tz               = _mm_mul_pd(fscal,dz10);
 517
 518             /* Update vectorial force */
 519             fix1             = _mm_add_pd(fix1,tx);
 520             fiy1             = _mm_add_pd(fiy1,ty);
 521             fiz1             = _mm_add_pd(fiz1,tz);
 522
 523             fjx0             = _mm_add_pd(fjx0,tx);
 524             fjy0             = _mm_add_pd(fjy0,ty);
 525             fjz0             = _mm_add_pd(fjz0,tz);
 526
 527             /**************************
 528              * CALCULATE INTERACTIONS *
 529              **************************/
 530
 531             /* Compute parameters for interactions between i and j atoms */
 532             qq20             = _mm_mul_pd(iq2,jq0);
 533
 534             /* COULOMB ELECTROSTATICS */
 535             velec            = _mm_mul_pd(qq20,rinv20);
 536             felec            = _mm_mul_pd(velec,rinvsq20);
 537
 538             /* Update potential sum for this i atom from the interaction with this j atom. */
 539             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 540             velecsum         = _mm_add_pd(velecsum,velec);
 541
 542             fscal            = felec;
 543
 544             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 545
 546             /* Calculate temporary vectorial force */
 547             tx               = _mm_mul_pd(fscal,dx20);
 548             ty               = _mm_mul_pd(fscal,dy20);
 549             tz               = _mm_mul_pd(fscal,dz20);
 550
 551             /* Update vectorial force */
 552             fix2             = _mm_add_pd(fix2,tx);
 553             fiy2             = _mm_add_pd(fiy2,ty);
 554             fiz2             = _mm_add_pd(fiz2,tz);
 555
 556             fjx0             = _mm_add_pd(fjx0,tx);
 557             fjy0             = _mm_add_pd(fjy0,ty);
 558             fjz0             = _mm_add_pd(fjz0,tz);
 559
 560             /**************************
 561              * CALCULATE INTERACTIONS *
 562              **************************/
 563
 564             /* Compute parameters for interactions between i and j atoms */
 565             qq30             = _mm_mul_pd(iq3,jq0);
 566
 567             /* COULOMB ELECTROSTATICS */
 568             velec            = _mm_mul_pd(qq30,rinv30);
 569             felec            = _mm_mul_pd(velec,rinvsq30);
 570
 571             /* Update potential sum for this i atom from the interaction with this j atom. */
 572             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 573             velecsum         = _mm_add_pd(velecsum,velec);
 574
 575             fscal            = felec;
 576
 577             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 578
 579             /* Calculate temporary vectorial force */
 580             tx               = _mm_mul_pd(fscal,dx30);
 581             ty               = _mm_mul_pd(fscal,dy30);
 582             tz               = _mm_mul_pd(fscal,dz30);
 583
 584             /* Update vectorial force */
 585             fix3             = _mm_add_pd(fix3,tx);
 586             fiy3             = _mm_add_pd(fiy3,ty);
 587             fiz3             = _mm_add_pd(fiz3,tz);
 588
 589             fjx0             = _mm_add_pd(fjx0,tx);
 590             fjy0             = _mm_add_pd(fjy0,ty);
 591             fjz0             = _mm_add_pd(fjz0,tz);
 592
 593             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 594
 595             /* Inner loop uses 143 flops */
 596         }
 597
 598         /* End of innermost loop */
 599
 600         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 601                                               f+i_coord_offset,fshift+i_shift_offset);
 602
 603         ggid                        = gid[iidx];
 604         /* Update potential energies */
 605         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 606         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 607
 608         /* Increment number of inner iterations */
 609         inneriter                  += j_index_end - j_index_start;
 610
 611         /* Outer loop uses 26 flops */
 612     }
 613
 614     /* Increment number of outer iterations */
 615     outeriter        += nri;
 616
 617     /* Update outer/inner flops */
 618
 619     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*143);
 620 }
 621 /*
 622  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_sse4_1_double
 623  * Electrostatics interaction: Coulomb
 624  * VdW interaction:            CubicSplineTable
 625  * Geometry:                   Water4-Particle
 626  * Calculate force/pot:        Force
 627  */
 628 void
 629 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_sse4_1_double
 630                     (t_nblist * gmx_restrict                nlist,
 631                      rvec * gmx_restrict                    xx,
 632                      rvec * gmx_restrict                    ff,
 633                      t_forcerec * gmx_restrict              fr,
 634                      t_mdatoms * gmx_restrict               mdatoms,
 635                      nb_kernel_data_t * gmx_restrict        kernel_data,
 636                      t_nrnb * gmx_restrict                  nrnb)
 637 {
 638     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 639      * just 0 for non-waters.
 640      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 641      * jnr indices corresponding to data put in the four positions in the SIMD register.
 642      */
 643     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 644     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 645     int              jnrA,jnrB;
 646     int              j_coord_offsetA,j_coord_offsetB;
 647     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 648     real             rcutoff_scalar;
 649     real             *shiftvec,*fshift,*x,*f;
 650     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 651     int              vdwioffset0;
 652     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 653     int              vdwioffset1;
 654     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 655     int              vdwioffset2;
 656     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 657     int              vdwioffset3;
 658     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 659     int              vdwjidx0A,vdwjidx0B;
 660     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 661     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 662     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 663     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 664     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 665     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 666     real             *charge;
 667     int              nvdwtype;
 668     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 669     int              *vdwtype;
 670     real             *vdwparam;
 671     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 672     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 673     __m128i          vfitab;
 674     __m128i          ifour       = _mm_set1_epi32(4);
 675     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 676     real             *vftab;
 677     __m128d          dummy_mask,cutoff_mask;
 678     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 679     __m128d          one     = _mm_set1_pd(1.0);
 680     __m128d          two     = _mm_set1_pd(2.0);
 681     x                = xx[0];
 682     f                = ff[0];
 683
 684     nri              = nlist->nri;
 685     iinr             = nlist->iinr;
 686     jindex           = nlist->jindex;
 687     jjnr             = nlist->jjnr;
 688     shiftidx         = nlist->shift;
 689     gid              = nlist->gid;
 690     shiftvec         = fr->shift_vec[0];
 691     fshift           = fr->fshift[0];
 692     facel            = _mm_set1_pd(fr->epsfac);
 693     charge           = mdatoms->chargeA;
 694     nvdwtype         = fr->ntype;
 695     vdwparam         = fr->nbfp;
 696     vdwtype          = mdatoms->typeA;
 697
 698     vftab            = kernel_data->table_vdw->data;
 699     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 700
 701     /* Setup water-specific parameters */
 702     inr              = nlist->iinr[0];
 703     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 704     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 705     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 706     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 707
 708     /* Avoid stupid compiler warnings */
 709     jnrA = jnrB = 0;
 710     j_coord_offsetA = 0;
 711     j_coord_offsetB = 0;
 712
 713     outeriter        = 0;
 714     inneriter        = 0;
 715
 716     /* Start outer loop over neighborlists */
 717     for(iidx=0; iidx<nri; iidx++)
 718     {
 719         /* Load shift vector for this list */
 720         i_shift_offset   = DIM*shiftidx[iidx];
 721
 722         /* Load limits for loop over neighbors */
 723         j_index_start    = jindex[iidx];
 724         j_index_end      = jindex[iidx+1];
 725
 726         /* Get outer coordinate index */
 727         inr              = iinr[iidx];
 728         i_coord_offset   = DIM*inr;
 729
 730         /* Load i particle coords and add shift vector */
 731         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 732                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 733
 734         fix0             = _mm_setzero_pd();
 735         fiy0             = _mm_setzero_pd();
 736         fiz0             = _mm_setzero_pd();
 737         fix1             = _mm_setzero_pd();
 738         fiy1             = _mm_setzero_pd();
 739         fiz1             = _mm_setzero_pd();
 740         fix2             = _mm_setzero_pd();
 741         fiy2             = _mm_setzero_pd();
 742         fiz2             = _mm_setzero_pd();
 743         fix3             = _mm_setzero_pd();
 744         fiy3             = _mm_setzero_pd();
 745         fiz3             = _mm_setzero_pd();
 746
 747         /* Start inner kernel loop */
 748         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 749         {
 750
 751             /* Get j neighbor index, and coordinate index */
 752             jnrA             = jjnr[jidx];
 753             jnrB             = jjnr[jidx+1];
 754             j_coord_offsetA  = DIM*jnrA;
 755             j_coord_offsetB  = DIM*jnrB;
 756
 757             /* load j atom coordinates */
 758             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 759                                               &jx0,&jy0,&jz0);
 760
 761             /* Calculate displacement vector */
 762             dx00             = _mm_sub_pd(ix0,jx0);
 763             dy00             = _mm_sub_pd(iy0,jy0);
 764             dz00             = _mm_sub_pd(iz0,jz0);
 765             dx10             = _mm_sub_pd(ix1,jx0);
 766             dy10             = _mm_sub_pd(iy1,jy0);
 767             dz10             = _mm_sub_pd(iz1,jz0);
 768             dx20             = _mm_sub_pd(ix2,jx0);
 769             dy20             = _mm_sub_pd(iy2,jy0);
 770             dz20             = _mm_sub_pd(iz2,jz0);
 771             dx30             = _mm_sub_pd(ix3,jx0);
 772             dy30             = _mm_sub_pd(iy3,jy0);
 773             dz30             = _mm_sub_pd(iz3,jz0);
 774
 775             /* Calculate squared distance and things based on it */
 776             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 777             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 778             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 779             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 780
 781             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 782             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 783             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 784             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 785
 786             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 787             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 788             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 789
 790             /* Load parameters for j particles */
 791             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 792             vdwjidx0A        = 2*vdwtype[jnrA+0];
 793             vdwjidx0B        = 2*vdwtype[jnrB+0];
 794
 795             fjx0             = _mm_setzero_pd();
 796             fjy0             = _mm_setzero_pd();
 797             fjz0             = _mm_setzero_pd();
 798
 799             /**************************
 800              * CALCULATE INTERACTIONS *
 801              **************************/
 802
 803             r00              = _mm_mul_pd(rsq00,rinv00);
 804
 805             /* Compute parameters for interactions between i and j atoms */
 806             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 807                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 808
 809             /* Calculate table index by multiplying r with table scale and truncate to integer */
 810             rt               = _mm_mul_pd(r00,vftabscale);
 811             vfitab           = _mm_cvttpd_epi32(rt);
 812             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 813             vfitab           = _mm_slli_epi32(vfitab,3);
 814
 815             /* CUBIC SPLINE TABLE DISPERSION */
 816             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 817             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 818             GMX_MM_TRANSPOSE2_PD(Y,F);
 819             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 820             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 821             GMX_MM_TRANSPOSE2_PD(G,H);
 822             Heps             = _mm_mul_pd(vfeps,H);
 823             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 824             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 825             fvdw6            = _mm_mul_pd(c6_00,FF);
 826
 827             /* CUBIC SPLINE TABLE REPULSION */
 828             vfitab           = _mm_add_epi32(vfitab,ifour);
 829             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 830             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 831             GMX_MM_TRANSPOSE2_PD(Y,F);
 832             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 833             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 834             GMX_MM_TRANSPOSE2_PD(G,H);
 835             Heps             = _mm_mul_pd(vfeps,H);
 836             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 837             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 838             fvdw12           = _mm_mul_pd(c12_00,FF);
 839             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 840
 841             fscal            = fvdw;
 842
 843             /* Calculate temporary vectorial force */
 844             tx               = _mm_mul_pd(fscal,dx00);
 845             ty               = _mm_mul_pd(fscal,dy00);
 846             tz               = _mm_mul_pd(fscal,dz00);
 847
 848             /* Update vectorial force */
 849             fix0             = _mm_add_pd(fix0,tx);
 850             fiy0             = _mm_add_pd(fiy0,ty);
 851             fiz0             = _mm_add_pd(fiz0,tz);
 852
 853             fjx0             = _mm_add_pd(fjx0,tx);
 854             fjy0             = _mm_add_pd(fjy0,ty);
 855             fjz0             = _mm_add_pd(fjz0,tz);
 856
 857             /**************************
 858              * CALCULATE INTERACTIONS *
 859              **************************/
 860
 861             /* Compute parameters for interactions between i and j atoms */
 862             qq10             = _mm_mul_pd(iq1,jq0);
 863
 864             /* COULOMB ELECTROSTATICS */
 865             velec            = _mm_mul_pd(qq10,rinv10);
 866             felec            = _mm_mul_pd(velec,rinvsq10);
 867
 868             fscal            = felec;
 869
 870             /* Calculate temporary vectorial force */
 871             tx               = _mm_mul_pd(fscal,dx10);
 872             ty               = _mm_mul_pd(fscal,dy10);
 873             tz               = _mm_mul_pd(fscal,dz10);
 874
 875             /* Update vectorial force */
 876             fix1             = _mm_add_pd(fix1,tx);
 877             fiy1             = _mm_add_pd(fiy1,ty);
 878             fiz1             = _mm_add_pd(fiz1,tz);
 879
 880             fjx0             = _mm_add_pd(fjx0,tx);
 881             fjy0             = _mm_add_pd(fjy0,ty);
 882             fjz0             = _mm_add_pd(fjz0,tz);
 883
 884             /**************************
 885              * CALCULATE INTERACTIONS *
 886              **************************/
 887
 888             /* Compute parameters for interactions between i and j atoms */
 889             qq20             = _mm_mul_pd(iq2,jq0);
 890
 891             /* COULOMB ELECTROSTATICS */
 892             velec            = _mm_mul_pd(qq20,rinv20);
 893             felec            = _mm_mul_pd(velec,rinvsq20);
 894
 895             fscal            = felec;
 896
 897             /* Calculate temporary vectorial force */
 898             tx               = _mm_mul_pd(fscal,dx20);
 899             ty               = _mm_mul_pd(fscal,dy20);
 900             tz               = _mm_mul_pd(fscal,dz20);
 901
 902             /* Update vectorial force */
 903             fix2             = _mm_add_pd(fix2,tx);
 904             fiy2             = _mm_add_pd(fiy2,ty);
 905             fiz2             = _mm_add_pd(fiz2,tz);
 906
 907             fjx0             = _mm_add_pd(fjx0,tx);
 908             fjy0             = _mm_add_pd(fjy0,ty);
 909             fjz0             = _mm_add_pd(fjz0,tz);
 910
 911             /**************************
 912              * CALCULATE INTERACTIONS *
 913              **************************/
 914
 915             /* Compute parameters for interactions between i and j atoms */
 916             qq30             = _mm_mul_pd(iq3,jq0);
 917
 918             /* COULOMB ELECTROSTATICS */
 919             velec            = _mm_mul_pd(qq30,rinv30);
 920             felec            = _mm_mul_pd(velec,rinvsq30);
 921
 922             fscal            = felec;
 923
 924             /* Calculate temporary vectorial force */
 925             tx               = _mm_mul_pd(fscal,dx30);
 926             ty               = _mm_mul_pd(fscal,dy30);
 927             tz               = _mm_mul_pd(fscal,dz30);
 928
 929             /* Update vectorial force */
 930             fix3             = _mm_add_pd(fix3,tx);
 931             fiy3             = _mm_add_pd(fiy3,ty);
 932             fiz3             = _mm_add_pd(fiz3,tz);
 933
 934             fjx0             = _mm_add_pd(fjx0,tx);
 935             fjy0             = _mm_add_pd(fjy0,ty);
 936             fjz0             = _mm_add_pd(fjz0,tz);
 937
 938             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 939
 940             /* Inner loop uses 132 flops */
 941         }
 942
 943         if(jidx<j_index_end)
 944         {
 945
 946             jnrA             = jjnr[jidx];
 947             j_coord_offsetA  = DIM*jnrA;
 948
 949             /* load j atom coordinates */
 950             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 951                                               &jx0,&jy0,&jz0);
 952
 953             /* Calculate displacement vector */
 954             dx00             = _mm_sub_pd(ix0,jx0);
 955             dy00             = _mm_sub_pd(iy0,jy0);
 956             dz00             = _mm_sub_pd(iz0,jz0);
 957             dx10             = _mm_sub_pd(ix1,jx0);
 958             dy10             = _mm_sub_pd(iy1,jy0);
 959             dz10             = _mm_sub_pd(iz1,jz0);
 960             dx20             = _mm_sub_pd(ix2,jx0);
 961             dy20             = _mm_sub_pd(iy2,jy0);
 962             dz20             = _mm_sub_pd(iz2,jz0);
 963             dx30             = _mm_sub_pd(ix3,jx0);
 964             dy30             = _mm_sub_pd(iy3,jy0);
 965             dz30             = _mm_sub_pd(iz3,jz0);
 966
 967             /* Calculate squared distance and things based on it */
 968             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 969             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 970             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 971             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 972
 973             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 974             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 975             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 976             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 977
 978             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 979             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 980             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 981
 982             /* Load parameters for j particles */
 983             jq0              = _mm_load_sd(charge+jnrA+0);
 984             vdwjidx0A        = 2*vdwtype[jnrA+0];
 985
 986             fjx0             = _mm_setzero_pd();
 987             fjy0             = _mm_setzero_pd();
 988             fjz0             = _mm_setzero_pd();
 989
 990             /**************************
 991              * CALCULATE INTERACTIONS *
 992              **************************/
 993
 994             r00              = _mm_mul_pd(rsq00,rinv00);
 995
 996             /* Compute parameters for interactions between i and j atoms */
 997             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 998
 999             /* Calculate table index by multiplying r with table scale and truncate to integer */
1000             rt               = _mm_mul_pd(r00,vftabscale);
1001             vfitab           = _mm_cvttpd_epi32(rt);
1002             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1003             vfitab           = _mm_slli_epi32(vfitab,3);
1004
1005             /* CUBIC SPLINE TABLE DISPERSION */
1006             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1007             F                = _mm_setzero_pd();
1008             GMX_MM_TRANSPOSE2_PD(Y,F);
1009             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1010             H                = _mm_setzero_pd();
1011             GMX_MM_TRANSPOSE2_PD(G,H);
1012             Heps             = _mm_mul_pd(vfeps,H);
1013             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1014             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1015             fvdw6            = _mm_mul_pd(c6_00,FF);
1016
1017             /* CUBIC SPLINE TABLE REPULSION */
1018             vfitab           = _mm_add_epi32(vfitab,ifour);
1019             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1020             F                = _mm_setzero_pd();
1021             GMX_MM_TRANSPOSE2_PD(Y,F);
1022             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1023             H                = _mm_setzero_pd();
1024             GMX_MM_TRANSPOSE2_PD(G,H);
1025             Heps             = _mm_mul_pd(vfeps,H);
1026             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1027             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1028             fvdw12           = _mm_mul_pd(c12_00,FF);
1029             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1030
1031             fscal            = fvdw;
1032
1033             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1034
1035             /* Calculate temporary vectorial force */
1036             tx               = _mm_mul_pd(fscal,dx00);
1037             ty               = _mm_mul_pd(fscal,dy00);
1038             tz               = _mm_mul_pd(fscal,dz00);
1039
1040             /* Update vectorial force */
1041             fix0             = _mm_add_pd(fix0,tx);
1042             fiy0             = _mm_add_pd(fiy0,ty);
1043             fiz0             = _mm_add_pd(fiz0,tz);
1044
1045             fjx0             = _mm_add_pd(fjx0,tx);
1046             fjy0             = _mm_add_pd(fjy0,ty);
1047             fjz0             = _mm_add_pd(fjz0,tz);
1048
1049             /**************************
1050              * CALCULATE INTERACTIONS *
1051              **************************/
1052
1053             /* Compute parameters for interactions between i and j atoms */
1054             qq10             = _mm_mul_pd(iq1,jq0);
1055
1056             /* COULOMB ELECTROSTATICS */
1057             velec            = _mm_mul_pd(qq10,rinv10);
1058             felec            = _mm_mul_pd(velec,rinvsq10);
1059
1060             fscal            = felec;
1061
1062             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1063
1064             /* Calculate temporary vectorial force */
1065             tx               = _mm_mul_pd(fscal,dx10);
1066             ty               = _mm_mul_pd(fscal,dy10);
1067             tz               = _mm_mul_pd(fscal,dz10);
1068
1069             /* Update vectorial force */
1070             fix1             = _mm_add_pd(fix1,tx);
1071             fiy1             = _mm_add_pd(fiy1,ty);
1072             fiz1             = _mm_add_pd(fiz1,tz);
1073
1074             fjx0             = _mm_add_pd(fjx0,tx);
1075             fjy0             = _mm_add_pd(fjy0,ty);
1076             fjz0             = _mm_add_pd(fjz0,tz);
1077
1078             /**************************
1079              * CALCULATE INTERACTIONS *
1080              **************************/
1081
1082             /* Compute parameters for interactions between i and j atoms */
1083             qq20             = _mm_mul_pd(iq2,jq0);
1084
1085             /* COULOMB ELECTROSTATICS */
1086             velec            = _mm_mul_pd(qq20,rinv20);
1087             felec            = _mm_mul_pd(velec,rinvsq20);
1088
1089             fscal            = felec;
1090
1091             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1092
1093             /* Calculate temporary vectorial force */
1094             tx               = _mm_mul_pd(fscal,dx20);
1095             ty               = _mm_mul_pd(fscal,dy20);
1096             tz               = _mm_mul_pd(fscal,dz20);
1097
1098             /* Update vectorial force */
1099             fix2             = _mm_add_pd(fix2,tx);
1100             fiy2             = _mm_add_pd(fiy2,ty);
1101             fiz2             = _mm_add_pd(fiz2,tz);
1102
1103             fjx0             = _mm_add_pd(fjx0,tx);
1104             fjy0             = _mm_add_pd(fjy0,ty);
1105             fjz0             = _mm_add_pd(fjz0,tz);
1106
1107             /**************************
1108              * CALCULATE INTERACTIONS *
1109              **************************/
1110
1111             /* Compute parameters for interactions between i and j atoms */
1112             qq30             = _mm_mul_pd(iq3,jq0);
1113
1114             /* COULOMB ELECTROSTATICS */
1115             velec            = _mm_mul_pd(qq30,rinv30);
1116             felec            = _mm_mul_pd(velec,rinvsq30);
1117
1118             fscal            = felec;
1119
1120             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1121
1122             /* Calculate temporary vectorial force */
1123             tx               = _mm_mul_pd(fscal,dx30);
1124             ty               = _mm_mul_pd(fscal,dy30);
1125             tz               = _mm_mul_pd(fscal,dz30);
1126
1127             /* Update vectorial force */
1128             fix3             = _mm_add_pd(fix3,tx);
1129             fiy3             = _mm_add_pd(fiy3,ty);
1130             fiz3             = _mm_add_pd(fiz3,tz);
1131
1132             fjx0             = _mm_add_pd(fjx0,tx);
1133             fjy0             = _mm_add_pd(fjy0,ty);
1134             fjz0             = _mm_add_pd(fjz0,tz);
1135
1136             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1137
1138             /* Inner loop uses 132 flops */
1139         }
1140
1141         /* End of innermost loop */
1142
1143         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1144                                               f+i_coord_offset,fshift+i_shift_offset);
1145
1146         /* Increment number of inner iterations */
1147         inneriter                  += j_index_end - j_index_start;
1148
1149         /* Outer loop uses 24 flops */
1150     }
1151
1152     /* Increment number of outer iterations */
1153     outeriter        += nri;
1154
1155     /* Update outer/inner flops */
1156
1157     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*132);
1158 }