src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_sse4_1_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse4_1_single 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_single.h"
  34 #include "kernelutil_x86_sse4_1_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_VF_sse4_1_single
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_VF_sse4_1_single
  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,C,D refer to j loop unrolling done with SSE, e.g. for the four 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,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwioffset1;
  72     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  73     int              vdwioffset2;
  74     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  75     int              vdwioffset3;
  76     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  77     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  78     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  79     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  80     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  81     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  82     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  83     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  84     real             *charge;
  85     int              nvdwtype;
  86     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  87     int              *vdwtype;
  88     real             *vdwparam;
  89     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  90     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  91     __m128i          vfitab;
  92     __m128i          ifour       = _mm_set1_epi32(4);
  93     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  94     real             *vftab;
  95     __m128           dummy_mask,cutoff_mask;
  96     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  97     __m128           one     = _mm_set1_ps(1.0);
  98     __m128           two     = _mm_set1_ps(2.0);
  99     x                = xx[0];
 100     f                = ff[0];
 101
 102     nri              = nlist->nri;
 103     iinr             = nlist->iinr;
 104     jindex           = nlist->jindex;
 105     jjnr             = nlist->jjnr;
 106     shiftidx         = nlist->shift;
 107     gid              = nlist->gid;
 108     shiftvec         = fr->shift_vec[0];
 109     fshift           = fr->fshift[0];
 110     facel            = _mm_set1_ps(fr->epsfac);
 111     charge           = mdatoms->chargeA;
 112     nvdwtype         = fr->ntype;
 113     vdwparam         = fr->nbfp;
 114     vdwtype          = mdatoms->typeA;
 115
 116     vftab            = kernel_data->table_elec_vdw->data;
 117     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 118
 119     /* Setup water-specific parameters */
 120     inr              = nlist->iinr[0];
 121     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 122     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 123     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 124     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 125
 126     /* Avoid stupid compiler warnings */
 127     jnrA = jnrB = jnrC = jnrD = 0;
 128     j_coord_offsetA = 0;
 129     j_coord_offsetB = 0;
 130     j_coord_offsetC = 0;
 131     j_coord_offsetD = 0;
 132
 133     outeriter        = 0;
 134     inneriter        = 0;
 135
 136     for(iidx=0;iidx<4*DIM;iidx++)
 137     {
 138         scratch[iidx] = 0.0;
 139     }
 140
 141     /* Start outer loop over neighborlists */
 142     for(iidx=0; iidx<nri; iidx++)
 143     {
 144         /* Load shift vector for this list */
 145         i_shift_offset   = DIM*shiftidx[iidx];
 146
 147         /* Load limits for loop over neighbors */
 148         j_index_start    = jindex[iidx];
 149         j_index_end      = jindex[iidx+1];
 150
 151         /* Get outer coordinate index */
 152         inr              = iinr[iidx];
 153         i_coord_offset   = DIM*inr;
 154
 155         /* Load i particle coords and add shift vector */
 156         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 157                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 158
 159         fix0             = _mm_setzero_ps();
 160         fiy0             = _mm_setzero_ps();
 161         fiz0             = _mm_setzero_ps();
 162         fix1             = _mm_setzero_ps();
 163         fiy1             = _mm_setzero_ps();
 164         fiz1             = _mm_setzero_ps();
 165         fix2             = _mm_setzero_ps();
 166         fiy2             = _mm_setzero_ps();
 167         fiz2             = _mm_setzero_ps();
 168         fix3             = _mm_setzero_ps();
 169         fiy3             = _mm_setzero_ps();
 170         fiz3             = _mm_setzero_ps();
 171
 172         /* Reset potential sums */
 173         velecsum         = _mm_setzero_ps();
 174         vvdwsum          = _mm_setzero_ps();
 175
 176         /* Start inner kernel loop */
 177         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 178         {
 179
 180             /* Get j neighbor index, and coordinate index */
 181             jnrA             = jjnr[jidx];
 182             jnrB             = jjnr[jidx+1];
 183             jnrC             = jjnr[jidx+2];
 184             jnrD             = jjnr[jidx+3];
 185             j_coord_offsetA  = DIM*jnrA;
 186             j_coord_offsetB  = DIM*jnrB;
 187             j_coord_offsetC  = DIM*jnrC;
 188             j_coord_offsetD  = DIM*jnrD;
 189
 190             /* load j atom coordinates */
 191             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 192                                               x+j_coord_offsetC,x+j_coord_offsetD,
 193                                               &jx0,&jy0,&jz0);
 194
 195             /* Calculate displacement vector */
 196             dx00             = _mm_sub_ps(ix0,jx0);
 197             dy00             = _mm_sub_ps(iy0,jy0);
 198             dz00             = _mm_sub_ps(iz0,jz0);
 199             dx10             = _mm_sub_ps(ix1,jx0);
 200             dy10             = _mm_sub_ps(iy1,jy0);
 201             dz10             = _mm_sub_ps(iz1,jz0);
 202             dx20             = _mm_sub_ps(ix2,jx0);
 203             dy20             = _mm_sub_ps(iy2,jy0);
 204             dz20             = _mm_sub_ps(iz2,jz0);
 205             dx30             = _mm_sub_ps(ix3,jx0);
 206             dy30             = _mm_sub_ps(iy3,jy0);
 207             dz30             = _mm_sub_ps(iz3,jz0);
 208
 209             /* Calculate squared distance and things based on it */
 210             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 211             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 212             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 213             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 214
 215             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 216             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 217             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 218             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 219
 220             /* Load parameters for j particles */
 221             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 222                                                               charge+jnrC+0,charge+jnrD+0);
 223             vdwjidx0A        = 2*vdwtype[jnrA+0];
 224             vdwjidx0B        = 2*vdwtype[jnrB+0];
 225             vdwjidx0C        = 2*vdwtype[jnrC+0];
 226             vdwjidx0D        = 2*vdwtype[jnrD+0];
 227
 228             fjx0             = _mm_setzero_ps();
 229             fjy0             = _mm_setzero_ps();
 230             fjz0             = _mm_setzero_ps();
 231
 232             /**************************
 233              * CALCULATE INTERACTIONS *
 234              **************************/
 235
 236             r00              = _mm_mul_ps(rsq00,rinv00);
 237
 238             /* Compute parameters for interactions between i and j atoms */
 239             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 240                                          vdwparam+vdwioffset0+vdwjidx0B,
 241                                          vdwparam+vdwioffset0+vdwjidx0C,
 242                                          vdwparam+vdwioffset0+vdwjidx0D,
 243                                          &c6_00,&c12_00);
 244
 245             /* Calculate table index by multiplying r with table scale and truncate to integer */
 246             rt               = _mm_mul_ps(r00,vftabscale);
 247             vfitab           = _mm_cvttps_epi32(rt);
 248             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 249             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 250
 251             /* CUBIC SPLINE TABLE DISPERSION */
 252             vfitab           = _mm_add_epi32(vfitab,ifour);
 253             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 254             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 255             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 256             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 257             _MM_TRANSPOSE4_PS(Y,F,G,H);
 258             Heps             = _mm_mul_ps(vfeps,H);
 259             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 260             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 261             vvdw6            = _mm_mul_ps(c6_00,VV);
 262             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 263             fvdw6            = _mm_mul_ps(c6_00,FF);
 264
 265             /* CUBIC SPLINE TABLE REPULSION */
 266             vfitab           = _mm_add_epi32(vfitab,ifour);
 267             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 268             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 269             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 270             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 271             _MM_TRANSPOSE4_PS(Y,F,G,H);
 272             Heps             = _mm_mul_ps(vfeps,H);
 273             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 274             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 275             vvdw12           = _mm_mul_ps(c12_00,VV);
 276             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 277             fvdw12           = _mm_mul_ps(c12_00,FF);
 278             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 279             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 280
 281             /* Update potential sum for this i atom from the interaction with this j atom. */
 282             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 283
 284             fscal            = fvdw;
 285
 286             /* Calculate temporary vectorial force */
 287             tx               = _mm_mul_ps(fscal,dx00);
 288             ty               = _mm_mul_ps(fscal,dy00);
 289             tz               = _mm_mul_ps(fscal,dz00);
 290
 291             /* Update vectorial force */
 292             fix0             = _mm_add_ps(fix0,tx);
 293             fiy0             = _mm_add_ps(fiy0,ty);
 294             fiz0             = _mm_add_ps(fiz0,tz);
 295
 296             fjx0             = _mm_add_ps(fjx0,tx);
 297             fjy0             = _mm_add_ps(fjy0,ty);
 298             fjz0             = _mm_add_ps(fjz0,tz);
 299
 300             /**************************
 301              * CALCULATE INTERACTIONS *
 302              **************************/
 303
 304             r10              = _mm_mul_ps(rsq10,rinv10);
 305
 306             /* Compute parameters for interactions between i and j atoms */
 307             qq10             = _mm_mul_ps(iq1,jq0);
 308
 309             /* Calculate table index by multiplying r with table scale and truncate to integer */
 310             rt               = _mm_mul_ps(r10,vftabscale);
 311             vfitab           = _mm_cvttps_epi32(rt);
 312             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 313             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 314
 315             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 316             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 317             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 318             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 319             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 320             _MM_TRANSPOSE4_PS(Y,F,G,H);
 321             Heps             = _mm_mul_ps(vfeps,H);
 322             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 323             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 324             velec            = _mm_mul_ps(qq10,VV);
 325             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 326             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 327
 328             /* Update potential sum for this i atom from the interaction with this j atom. */
 329             velecsum         = _mm_add_ps(velecsum,velec);
 330
 331             fscal            = felec;
 332
 333             /* Calculate temporary vectorial force */
 334             tx               = _mm_mul_ps(fscal,dx10);
 335             ty               = _mm_mul_ps(fscal,dy10);
 336             tz               = _mm_mul_ps(fscal,dz10);
 337
 338             /* Update vectorial force */
 339             fix1             = _mm_add_ps(fix1,tx);
 340             fiy1             = _mm_add_ps(fiy1,ty);
 341             fiz1             = _mm_add_ps(fiz1,tz);
 342
 343             fjx0             = _mm_add_ps(fjx0,tx);
 344             fjy0             = _mm_add_ps(fjy0,ty);
 345             fjz0             = _mm_add_ps(fjz0,tz);
 346
 347             /**************************
 348              * CALCULATE INTERACTIONS *
 349              **************************/
 350
 351             r20              = _mm_mul_ps(rsq20,rinv20);
 352
 353             /* Compute parameters for interactions between i and j atoms */
 354             qq20             = _mm_mul_ps(iq2,jq0);
 355
 356             /* Calculate table index by multiplying r with table scale and truncate to integer */
 357             rt               = _mm_mul_ps(r20,vftabscale);
 358             vfitab           = _mm_cvttps_epi32(rt);
 359             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 360             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 361
 362             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 363             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 364             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 365             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 366             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 367             _MM_TRANSPOSE4_PS(Y,F,G,H);
 368             Heps             = _mm_mul_ps(vfeps,H);
 369             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 370             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 371             velec            = _mm_mul_ps(qq20,VV);
 372             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 373             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 374
 375             /* Update potential sum for this i atom from the interaction with this j atom. */
 376             velecsum         = _mm_add_ps(velecsum,velec);
 377
 378             fscal            = felec;
 379
 380             /* Calculate temporary vectorial force */
 381             tx               = _mm_mul_ps(fscal,dx20);
 382             ty               = _mm_mul_ps(fscal,dy20);
 383             tz               = _mm_mul_ps(fscal,dz20);
 384
 385             /* Update vectorial force */
 386             fix2             = _mm_add_ps(fix2,tx);
 387             fiy2             = _mm_add_ps(fiy2,ty);
 388             fiz2             = _mm_add_ps(fiz2,tz);
 389
 390             fjx0             = _mm_add_ps(fjx0,tx);
 391             fjy0             = _mm_add_ps(fjy0,ty);
 392             fjz0             = _mm_add_ps(fjz0,tz);
 393
 394             /**************************
 395              * CALCULATE INTERACTIONS *
 396              **************************/
 397
 398             r30              = _mm_mul_ps(rsq30,rinv30);
 399
 400             /* Compute parameters for interactions between i and j atoms */
 401             qq30             = _mm_mul_ps(iq3,jq0);
 402
 403             /* Calculate table index by multiplying r with table scale and truncate to integer */
 404             rt               = _mm_mul_ps(r30,vftabscale);
 405             vfitab           = _mm_cvttps_epi32(rt);
 406             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 407             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 408
 409             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 410             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 411             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 412             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 413             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 414             _MM_TRANSPOSE4_PS(Y,F,G,H);
 415             Heps             = _mm_mul_ps(vfeps,H);
 416             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 417             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 418             velec            = _mm_mul_ps(qq30,VV);
 419             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 420             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
 421
 422             /* Update potential sum for this i atom from the interaction with this j atom. */
 423             velecsum         = _mm_add_ps(velecsum,velec);
 424
 425             fscal            = felec;
 426
 427             /* Calculate temporary vectorial force */
 428             tx               = _mm_mul_ps(fscal,dx30);
 429             ty               = _mm_mul_ps(fscal,dy30);
 430             tz               = _mm_mul_ps(fscal,dz30);
 431
 432             /* Update vectorial force */
 433             fix3             = _mm_add_ps(fix3,tx);
 434             fiy3             = _mm_add_ps(fiy3,ty);
 435             fiz3             = _mm_add_ps(fiz3,tz);
 436
 437             fjx0             = _mm_add_ps(fjx0,tx);
 438             fjy0             = _mm_add_ps(fjy0,ty);
 439             fjz0             = _mm_add_ps(fjz0,tz);
 440
 441             fjptrA             = f+j_coord_offsetA;
 442             fjptrB             = f+j_coord_offsetB;
 443             fjptrC             = f+j_coord_offsetC;
 444             fjptrD             = f+j_coord_offsetD;
 445
 446             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 447
 448             /* Inner loop uses 185 flops */
 449         }
 450
 451         if(jidx<j_index_end)
 452         {
 453
 454             /* Get j neighbor index, and coordinate index */
 455             jnrlistA         = jjnr[jidx];
 456             jnrlistB         = jjnr[jidx+1];
 457             jnrlistC         = jjnr[jidx+2];
 458             jnrlistD         = jjnr[jidx+3];
 459             /* Sign of each element will be negative for non-real atoms.
 460              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 461              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 462              */
 463             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 464             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 465             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 466             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 467             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 468             j_coord_offsetA  = DIM*jnrA;
 469             j_coord_offsetB  = DIM*jnrB;
 470             j_coord_offsetC  = DIM*jnrC;
 471             j_coord_offsetD  = DIM*jnrD;
 472
 473             /* load j atom coordinates */
 474             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 475                                               x+j_coord_offsetC,x+j_coord_offsetD,
 476                                               &jx0,&jy0,&jz0);
 477
 478             /* Calculate displacement vector */
 479             dx00             = _mm_sub_ps(ix0,jx0);
 480             dy00             = _mm_sub_ps(iy0,jy0);
 481             dz00             = _mm_sub_ps(iz0,jz0);
 482             dx10             = _mm_sub_ps(ix1,jx0);
 483             dy10             = _mm_sub_ps(iy1,jy0);
 484             dz10             = _mm_sub_ps(iz1,jz0);
 485             dx20             = _mm_sub_ps(ix2,jx0);
 486             dy20             = _mm_sub_ps(iy2,jy0);
 487             dz20             = _mm_sub_ps(iz2,jz0);
 488             dx30             = _mm_sub_ps(ix3,jx0);
 489             dy30             = _mm_sub_ps(iy3,jy0);
 490             dz30             = _mm_sub_ps(iz3,jz0);
 491
 492             /* Calculate squared distance and things based on it */
 493             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 494             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 495             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 496             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 497
 498             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 499             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 500             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 501             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 502
 503             /* Load parameters for j particles */
 504             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 505                                                               charge+jnrC+0,charge+jnrD+0);
 506             vdwjidx0A        = 2*vdwtype[jnrA+0];
 507             vdwjidx0B        = 2*vdwtype[jnrB+0];
 508             vdwjidx0C        = 2*vdwtype[jnrC+0];
 509             vdwjidx0D        = 2*vdwtype[jnrD+0];
 510
 511             fjx0             = _mm_setzero_ps();
 512             fjy0             = _mm_setzero_ps();
 513             fjz0             = _mm_setzero_ps();
 514
 515             /**************************
 516              * CALCULATE INTERACTIONS *
 517              **************************/
 518
 519             r00              = _mm_mul_ps(rsq00,rinv00);
 520             r00              = _mm_andnot_ps(dummy_mask,r00);
 521
 522             /* Compute parameters for interactions between i and j atoms */
 523             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 524                                          vdwparam+vdwioffset0+vdwjidx0B,
 525                                          vdwparam+vdwioffset0+vdwjidx0C,
 526                                          vdwparam+vdwioffset0+vdwjidx0D,
 527                                          &c6_00,&c12_00);
 528
 529             /* Calculate table index by multiplying r with table scale and truncate to integer */
 530             rt               = _mm_mul_ps(r00,vftabscale);
 531             vfitab           = _mm_cvttps_epi32(rt);
 532             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 533             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 534
 535             /* CUBIC SPLINE TABLE DISPERSION */
 536             vfitab           = _mm_add_epi32(vfitab,ifour);
 537             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 538             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 539             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 540             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 541             _MM_TRANSPOSE4_PS(Y,F,G,H);
 542             Heps             = _mm_mul_ps(vfeps,H);
 543             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 544             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 545             vvdw6            = _mm_mul_ps(c6_00,VV);
 546             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 547             fvdw6            = _mm_mul_ps(c6_00,FF);
 548
 549             /* CUBIC SPLINE TABLE REPULSION */
 550             vfitab           = _mm_add_epi32(vfitab,ifour);
 551             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 552             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 553             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 554             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 555             _MM_TRANSPOSE4_PS(Y,F,G,H);
 556             Heps             = _mm_mul_ps(vfeps,H);
 557             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 558             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 559             vvdw12           = _mm_mul_ps(c12_00,VV);
 560             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 561             fvdw12           = _mm_mul_ps(c12_00,FF);
 562             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 563             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 564
 565             /* Update potential sum for this i atom from the interaction with this j atom. */
 566             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 567             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 568
 569             fscal            = fvdw;
 570
 571             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 572
 573             /* Calculate temporary vectorial force */
 574             tx               = _mm_mul_ps(fscal,dx00);
 575             ty               = _mm_mul_ps(fscal,dy00);
 576             tz               = _mm_mul_ps(fscal,dz00);
 577
 578             /* Update vectorial force */
 579             fix0             = _mm_add_ps(fix0,tx);
 580             fiy0             = _mm_add_ps(fiy0,ty);
 581             fiz0             = _mm_add_ps(fiz0,tz);
 582
 583             fjx0             = _mm_add_ps(fjx0,tx);
 584             fjy0             = _mm_add_ps(fjy0,ty);
 585             fjz0             = _mm_add_ps(fjz0,tz);
 586
 587             /**************************
 588              * CALCULATE INTERACTIONS *
 589              **************************/
 590
 591             r10              = _mm_mul_ps(rsq10,rinv10);
 592             r10              = _mm_andnot_ps(dummy_mask,r10);
 593
 594             /* Compute parameters for interactions between i and j atoms */
 595             qq10             = _mm_mul_ps(iq1,jq0);
 596
 597             /* Calculate table index by multiplying r with table scale and truncate to integer */
 598             rt               = _mm_mul_ps(r10,vftabscale);
 599             vfitab           = _mm_cvttps_epi32(rt);
 600             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 601             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 602
 603             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 604             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 605             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 606             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 607             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 608             _MM_TRANSPOSE4_PS(Y,F,G,H);
 609             Heps             = _mm_mul_ps(vfeps,H);
 610             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 611             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 612             velec            = _mm_mul_ps(qq10,VV);
 613             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 614             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 615
 616             /* Update potential sum for this i atom from the interaction with this j atom. */
 617             velec            = _mm_andnot_ps(dummy_mask,velec);
 618             velecsum         = _mm_add_ps(velecsum,velec);
 619
 620             fscal            = felec;
 621
 622             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 623
 624             /* Calculate temporary vectorial force */
 625             tx               = _mm_mul_ps(fscal,dx10);
 626             ty               = _mm_mul_ps(fscal,dy10);
 627             tz               = _mm_mul_ps(fscal,dz10);
 628
 629             /* Update vectorial force */
 630             fix1             = _mm_add_ps(fix1,tx);
 631             fiy1             = _mm_add_ps(fiy1,ty);
 632             fiz1             = _mm_add_ps(fiz1,tz);
 633
 634             fjx0             = _mm_add_ps(fjx0,tx);
 635             fjy0             = _mm_add_ps(fjy0,ty);
 636             fjz0             = _mm_add_ps(fjz0,tz);
 637
 638             /**************************
 639              * CALCULATE INTERACTIONS *
 640              **************************/
 641
 642             r20              = _mm_mul_ps(rsq20,rinv20);
 643             r20              = _mm_andnot_ps(dummy_mask,r20);
 644
 645             /* Compute parameters for interactions between i and j atoms */
 646             qq20             = _mm_mul_ps(iq2,jq0);
 647
 648             /* Calculate table index by multiplying r with table scale and truncate to integer */
 649             rt               = _mm_mul_ps(r20,vftabscale);
 650             vfitab           = _mm_cvttps_epi32(rt);
 651             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 652             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 653
 654             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 655             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 656             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 657             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 658             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 659             _MM_TRANSPOSE4_PS(Y,F,G,H);
 660             Heps             = _mm_mul_ps(vfeps,H);
 661             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 662             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 663             velec            = _mm_mul_ps(qq20,VV);
 664             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 665             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 666
 667             /* Update potential sum for this i atom from the interaction with this j atom. */
 668             velec            = _mm_andnot_ps(dummy_mask,velec);
 669             velecsum         = _mm_add_ps(velecsum,velec);
 670
 671             fscal            = felec;
 672
 673             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 674
 675             /* Calculate temporary vectorial force */
 676             tx               = _mm_mul_ps(fscal,dx20);
 677             ty               = _mm_mul_ps(fscal,dy20);
 678             tz               = _mm_mul_ps(fscal,dz20);
 679
 680             /* Update vectorial force */
 681             fix2             = _mm_add_ps(fix2,tx);
 682             fiy2             = _mm_add_ps(fiy2,ty);
 683             fiz2             = _mm_add_ps(fiz2,tz);
 684
 685             fjx0             = _mm_add_ps(fjx0,tx);
 686             fjy0             = _mm_add_ps(fjy0,ty);
 687             fjz0             = _mm_add_ps(fjz0,tz);
 688
 689             /**************************
 690              * CALCULATE INTERACTIONS *
 691              **************************/
 692
 693             r30              = _mm_mul_ps(rsq30,rinv30);
 694             r30              = _mm_andnot_ps(dummy_mask,r30);
 695
 696             /* Compute parameters for interactions between i and j atoms */
 697             qq30             = _mm_mul_ps(iq3,jq0);
 698
 699             /* Calculate table index by multiplying r with table scale and truncate to integer */
 700             rt               = _mm_mul_ps(r30,vftabscale);
 701             vfitab           = _mm_cvttps_epi32(rt);
 702             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 703             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 704
 705             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 706             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 707             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 708             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 709             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 710             _MM_TRANSPOSE4_PS(Y,F,G,H);
 711             Heps             = _mm_mul_ps(vfeps,H);
 712             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 713             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 714             velec            = _mm_mul_ps(qq30,VV);
 715             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 716             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
 717
 718             /* Update potential sum for this i atom from the interaction with this j atom. */
 719             velec            = _mm_andnot_ps(dummy_mask,velec);
 720             velecsum         = _mm_add_ps(velecsum,velec);
 721
 722             fscal            = felec;
 723
 724             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 725
 726             /* Calculate temporary vectorial force */
 727             tx               = _mm_mul_ps(fscal,dx30);
 728             ty               = _mm_mul_ps(fscal,dy30);
 729             tz               = _mm_mul_ps(fscal,dz30);
 730
 731             /* Update vectorial force */
 732             fix3             = _mm_add_ps(fix3,tx);
 733             fiy3             = _mm_add_ps(fiy3,ty);
 734             fiz3             = _mm_add_ps(fiz3,tz);
 735
 736             fjx0             = _mm_add_ps(fjx0,tx);
 737             fjy0             = _mm_add_ps(fjy0,ty);
 738             fjz0             = _mm_add_ps(fjz0,tz);
 739
 740             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 741             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 742             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 743             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 744
 745             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 746
 747             /* Inner loop uses 189 flops */
 748         }
 749
 750         /* End of innermost loop */
 751
 752         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 753                                               f+i_coord_offset,fshift+i_shift_offset);
 754
 755         ggid                        = gid[iidx];
 756         /* Update potential energies */
 757         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 758         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 759
 760         /* Increment number of inner iterations */
 761         inneriter                  += j_index_end - j_index_start;
 762
 763         /* Outer loop uses 26 flops */
 764     }
 765
 766     /* Increment number of outer iterations */
 767     outeriter        += nri;
 768
 769     /* Update outer/inner flops */
 770
 771     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*189);
 772 }
 773 /*
 774  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_F_sse4_1_single
 775  * Electrostatics interaction: CubicSplineTable
 776  * VdW interaction:            CubicSplineTable
 777  * Geometry:                   Water4-Particle
 778  * Calculate force/pot:        Force
 779  */
 780 void
 781 nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_F_sse4_1_single
 782                     (t_nblist * gmx_restrict                nlist,
 783                      rvec * gmx_restrict                    xx,
 784                      rvec * gmx_restrict                    ff,
 785                      t_forcerec * gmx_restrict              fr,
 786                      t_mdatoms * gmx_restrict               mdatoms,
 787                      nb_kernel_data_t * gmx_restrict        kernel_data,
 788                      t_nrnb * gmx_restrict                  nrnb)
 789 {
 790     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 791      * just 0 for non-waters.
 792      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 793      * jnr indices corresponding to data put in the four positions in the SIMD register.
 794      */
 795     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 796     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 797     int              jnrA,jnrB,jnrC,jnrD;
 798     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 799     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 800     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 801     real             rcutoff_scalar;
 802     real             *shiftvec,*fshift,*x,*f;
 803     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 804     real             scratch[4*DIM];
 805     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 806     int              vdwioffset0;
 807     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 808     int              vdwioffset1;
 809     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 810     int              vdwioffset2;
 811     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 812     int              vdwioffset3;
 813     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 814     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 815     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 816     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 817     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 818     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 819     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 820     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 821     real             *charge;
 822     int              nvdwtype;
 823     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 824     int              *vdwtype;
 825     real             *vdwparam;
 826     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 827     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 828     __m128i          vfitab;
 829     __m128i          ifour       = _mm_set1_epi32(4);
 830     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 831     real             *vftab;
 832     __m128           dummy_mask,cutoff_mask;
 833     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 834     __m128           one     = _mm_set1_ps(1.0);
 835     __m128           two     = _mm_set1_ps(2.0);
 836     x                = xx[0];
 837     f                = ff[0];
 838
 839     nri              = nlist->nri;
 840     iinr             = nlist->iinr;
 841     jindex           = nlist->jindex;
 842     jjnr             = nlist->jjnr;
 843     shiftidx         = nlist->shift;
 844     gid              = nlist->gid;
 845     shiftvec         = fr->shift_vec[0];
 846     fshift           = fr->fshift[0];
 847     facel            = _mm_set1_ps(fr->epsfac);
 848     charge           = mdatoms->chargeA;
 849     nvdwtype         = fr->ntype;
 850     vdwparam         = fr->nbfp;
 851     vdwtype          = mdatoms->typeA;
 852
 853     vftab            = kernel_data->table_elec_vdw->data;
 854     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 855
 856     /* Setup water-specific parameters */
 857     inr              = nlist->iinr[0];
 858     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 859     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 860     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 861     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 862
 863     /* Avoid stupid compiler warnings */
 864     jnrA = jnrB = jnrC = jnrD = 0;
 865     j_coord_offsetA = 0;
 866     j_coord_offsetB = 0;
 867     j_coord_offsetC = 0;
 868     j_coord_offsetD = 0;
 869
 870     outeriter        = 0;
 871     inneriter        = 0;
 872
 873     for(iidx=0;iidx<4*DIM;iidx++)
 874     {
 875         scratch[iidx] = 0.0;
 876     }
 877
 878     /* Start outer loop over neighborlists */
 879     for(iidx=0; iidx<nri; iidx++)
 880     {
 881         /* Load shift vector for this list */
 882         i_shift_offset   = DIM*shiftidx[iidx];
 883
 884         /* Load limits for loop over neighbors */
 885         j_index_start    = jindex[iidx];
 886         j_index_end      = jindex[iidx+1];
 887
 888         /* Get outer coordinate index */
 889         inr              = iinr[iidx];
 890         i_coord_offset   = DIM*inr;
 891
 892         /* Load i particle coords and add shift vector */
 893         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 894                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 895
 896         fix0             = _mm_setzero_ps();
 897         fiy0             = _mm_setzero_ps();
 898         fiz0             = _mm_setzero_ps();
 899         fix1             = _mm_setzero_ps();
 900         fiy1             = _mm_setzero_ps();
 901         fiz1             = _mm_setzero_ps();
 902         fix2             = _mm_setzero_ps();
 903         fiy2             = _mm_setzero_ps();
 904         fiz2             = _mm_setzero_ps();
 905         fix3             = _mm_setzero_ps();
 906         fiy3             = _mm_setzero_ps();
 907         fiz3             = _mm_setzero_ps();
 908
 909         /* Start inner kernel loop */
 910         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 911         {
 912
 913             /* Get j neighbor index, and coordinate index */
 914             jnrA             = jjnr[jidx];
 915             jnrB             = jjnr[jidx+1];
 916             jnrC             = jjnr[jidx+2];
 917             jnrD             = jjnr[jidx+3];
 918             j_coord_offsetA  = DIM*jnrA;
 919             j_coord_offsetB  = DIM*jnrB;
 920             j_coord_offsetC  = DIM*jnrC;
 921             j_coord_offsetD  = DIM*jnrD;
 922
 923             /* load j atom coordinates */
 924             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 925                                               x+j_coord_offsetC,x+j_coord_offsetD,
 926                                               &jx0,&jy0,&jz0);
 927
 928             /* Calculate displacement vector */
 929             dx00             = _mm_sub_ps(ix0,jx0);
 930             dy00             = _mm_sub_ps(iy0,jy0);
 931             dz00             = _mm_sub_ps(iz0,jz0);
 932             dx10             = _mm_sub_ps(ix1,jx0);
 933             dy10             = _mm_sub_ps(iy1,jy0);
 934             dz10             = _mm_sub_ps(iz1,jz0);
 935             dx20             = _mm_sub_ps(ix2,jx0);
 936             dy20             = _mm_sub_ps(iy2,jy0);
 937             dz20             = _mm_sub_ps(iz2,jz0);
 938             dx30             = _mm_sub_ps(ix3,jx0);
 939             dy30             = _mm_sub_ps(iy3,jy0);
 940             dz30             = _mm_sub_ps(iz3,jz0);
 941
 942             /* Calculate squared distance and things based on it */
 943             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 944             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 945             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 946             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 947
 948             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 949             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 950             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 951             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 952
 953             /* Load parameters for j particles */
 954             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 955                                                               charge+jnrC+0,charge+jnrD+0);
 956             vdwjidx0A        = 2*vdwtype[jnrA+0];
 957             vdwjidx0B        = 2*vdwtype[jnrB+0];
 958             vdwjidx0C        = 2*vdwtype[jnrC+0];
 959             vdwjidx0D        = 2*vdwtype[jnrD+0];
 960
 961             fjx0             = _mm_setzero_ps();
 962             fjy0             = _mm_setzero_ps();
 963             fjz0             = _mm_setzero_ps();
 964
 965             /**************************
 966              * CALCULATE INTERACTIONS *
 967              **************************/
 968
 969             r00              = _mm_mul_ps(rsq00,rinv00);
 970
 971             /* Compute parameters for interactions between i and j atoms */
 972             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 973                                          vdwparam+vdwioffset0+vdwjidx0B,
 974                                          vdwparam+vdwioffset0+vdwjidx0C,
 975                                          vdwparam+vdwioffset0+vdwjidx0D,
 976                                          &c6_00,&c12_00);
 977
 978             /* Calculate table index by multiplying r with table scale and truncate to integer */
 979             rt               = _mm_mul_ps(r00,vftabscale);
 980             vfitab           = _mm_cvttps_epi32(rt);
 981             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 982             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 983
 984             /* CUBIC SPLINE TABLE DISPERSION */
 985             vfitab           = _mm_add_epi32(vfitab,ifour);
 986             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 987             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 988             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 989             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 990             _MM_TRANSPOSE4_PS(Y,F,G,H);
 991             Heps             = _mm_mul_ps(vfeps,H);
 992             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 993             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 994             fvdw6            = _mm_mul_ps(c6_00,FF);
 995
 996             /* CUBIC SPLINE TABLE REPULSION */
 997             vfitab           = _mm_add_epi32(vfitab,ifour);
 998             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 999             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1000             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1001             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1002             _MM_TRANSPOSE4_PS(Y,F,G,H);
1003             Heps             = _mm_mul_ps(vfeps,H);
1004             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1005             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1006             fvdw12           = _mm_mul_ps(c12_00,FF);
1007             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1008
1009             fscal            = fvdw;
1010
1011             /* Calculate temporary vectorial force */
1012             tx               = _mm_mul_ps(fscal,dx00);
1013             ty               = _mm_mul_ps(fscal,dy00);
1014             tz               = _mm_mul_ps(fscal,dz00);
1015
1016             /* Update vectorial force */
1017             fix0             = _mm_add_ps(fix0,tx);
1018             fiy0             = _mm_add_ps(fiy0,ty);
1019             fiz0             = _mm_add_ps(fiz0,tz);
1020
1021             fjx0             = _mm_add_ps(fjx0,tx);
1022             fjy0             = _mm_add_ps(fjy0,ty);
1023             fjz0             = _mm_add_ps(fjz0,tz);
1024
1025             /**************************
1026              * CALCULATE INTERACTIONS *
1027              **************************/
1028
1029             r10              = _mm_mul_ps(rsq10,rinv10);
1030
1031             /* Compute parameters for interactions between i and j atoms */
1032             qq10             = _mm_mul_ps(iq1,jq0);
1033
1034             /* Calculate table index by multiplying r with table scale and truncate to integer */
1035             rt               = _mm_mul_ps(r10,vftabscale);
1036             vfitab           = _mm_cvttps_epi32(rt);
1037             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1038             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1039
1040             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1041             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1042             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1043             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1044             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1045             _MM_TRANSPOSE4_PS(Y,F,G,H);
1046             Heps             = _mm_mul_ps(vfeps,H);
1047             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1048             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1049             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1050
1051             fscal            = felec;
1052
1053             /* Calculate temporary vectorial force */
1054             tx               = _mm_mul_ps(fscal,dx10);
1055             ty               = _mm_mul_ps(fscal,dy10);
1056             tz               = _mm_mul_ps(fscal,dz10);
1057
1058             /* Update vectorial force */
1059             fix1             = _mm_add_ps(fix1,tx);
1060             fiy1             = _mm_add_ps(fiy1,ty);
1061             fiz1             = _mm_add_ps(fiz1,tz);
1062
1063             fjx0             = _mm_add_ps(fjx0,tx);
1064             fjy0             = _mm_add_ps(fjy0,ty);
1065             fjz0             = _mm_add_ps(fjz0,tz);
1066
1067             /**************************
1068              * CALCULATE INTERACTIONS *
1069              **************************/
1070
1071             r20              = _mm_mul_ps(rsq20,rinv20);
1072
1073             /* Compute parameters for interactions between i and j atoms */
1074             qq20             = _mm_mul_ps(iq2,jq0);
1075
1076             /* Calculate table index by multiplying r with table scale and truncate to integer */
1077             rt               = _mm_mul_ps(r20,vftabscale);
1078             vfitab           = _mm_cvttps_epi32(rt);
1079             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1080             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1081
1082             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1083             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1084             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1085             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1086             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1087             _MM_TRANSPOSE4_PS(Y,F,G,H);
1088             Heps             = _mm_mul_ps(vfeps,H);
1089             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1090             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1091             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1092
1093             fscal            = felec;
1094
1095             /* Calculate temporary vectorial force */
1096             tx               = _mm_mul_ps(fscal,dx20);
1097             ty               = _mm_mul_ps(fscal,dy20);
1098             tz               = _mm_mul_ps(fscal,dz20);
1099
1100             /* Update vectorial force */
1101             fix2             = _mm_add_ps(fix2,tx);
1102             fiy2             = _mm_add_ps(fiy2,ty);
1103             fiz2             = _mm_add_ps(fiz2,tz);
1104
1105             fjx0             = _mm_add_ps(fjx0,tx);
1106             fjy0             = _mm_add_ps(fjy0,ty);
1107             fjz0             = _mm_add_ps(fjz0,tz);
1108
1109             /**************************
1110              * CALCULATE INTERACTIONS *
1111              **************************/
1112
1113             r30              = _mm_mul_ps(rsq30,rinv30);
1114
1115             /* Compute parameters for interactions between i and j atoms */
1116             qq30             = _mm_mul_ps(iq3,jq0);
1117
1118             /* Calculate table index by multiplying r with table scale and truncate to integer */
1119             rt               = _mm_mul_ps(r30,vftabscale);
1120             vfitab           = _mm_cvttps_epi32(rt);
1121             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1122             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1123
1124             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1125             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1126             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1127             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1128             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1129             _MM_TRANSPOSE4_PS(Y,F,G,H);
1130             Heps             = _mm_mul_ps(vfeps,H);
1131             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1132             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1133             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1134
1135             fscal            = felec;
1136
1137             /* Calculate temporary vectorial force */
1138             tx               = _mm_mul_ps(fscal,dx30);
1139             ty               = _mm_mul_ps(fscal,dy30);
1140             tz               = _mm_mul_ps(fscal,dz30);
1141
1142             /* Update vectorial force */
1143             fix3             = _mm_add_ps(fix3,tx);
1144             fiy3             = _mm_add_ps(fiy3,ty);
1145             fiz3             = _mm_add_ps(fiz3,tz);
1146
1147             fjx0             = _mm_add_ps(fjx0,tx);
1148             fjy0             = _mm_add_ps(fjy0,ty);
1149             fjz0             = _mm_add_ps(fjz0,tz);
1150
1151             fjptrA             = f+j_coord_offsetA;
1152             fjptrB             = f+j_coord_offsetB;
1153             fjptrC             = f+j_coord_offsetC;
1154             fjptrD             = f+j_coord_offsetD;
1155
1156             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1157
1158             /* Inner loop uses 165 flops */
1159         }
1160
1161         if(jidx<j_index_end)
1162         {
1163
1164             /* Get j neighbor index, and coordinate index */
1165             jnrlistA         = jjnr[jidx];
1166             jnrlistB         = jjnr[jidx+1];
1167             jnrlistC         = jjnr[jidx+2];
1168             jnrlistD         = jjnr[jidx+3];
1169             /* Sign of each element will be negative for non-real atoms.
1170              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1171              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1172              */
1173             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1174             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1175             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1176             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1177             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1178             j_coord_offsetA  = DIM*jnrA;
1179             j_coord_offsetB  = DIM*jnrB;
1180             j_coord_offsetC  = DIM*jnrC;
1181             j_coord_offsetD  = DIM*jnrD;
1182
1183             /* load j atom coordinates */
1184             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1185                                               x+j_coord_offsetC,x+j_coord_offsetD,
1186                                               &jx0,&jy0,&jz0);
1187
1188             /* Calculate displacement vector */
1189             dx00             = _mm_sub_ps(ix0,jx0);
1190             dy00             = _mm_sub_ps(iy0,jy0);
1191             dz00             = _mm_sub_ps(iz0,jz0);
1192             dx10             = _mm_sub_ps(ix1,jx0);
1193             dy10             = _mm_sub_ps(iy1,jy0);
1194             dz10             = _mm_sub_ps(iz1,jz0);
1195             dx20             = _mm_sub_ps(ix2,jx0);
1196             dy20             = _mm_sub_ps(iy2,jy0);
1197             dz20             = _mm_sub_ps(iz2,jz0);
1198             dx30             = _mm_sub_ps(ix3,jx0);
1199             dy30             = _mm_sub_ps(iy3,jy0);
1200             dz30             = _mm_sub_ps(iz3,jz0);
1201
1202             /* Calculate squared distance and things based on it */
1203             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1204             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1205             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1206             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1207
1208             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1209             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1210             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1211             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1212
1213             /* Load parameters for j particles */
1214             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1215                                                               charge+jnrC+0,charge+jnrD+0);
1216             vdwjidx0A        = 2*vdwtype[jnrA+0];
1217             vdwjidx0B        = 2*vdwtype[jnrB+0];
1218             vdwjidx0C        = 2*vdwtype[jnrC+0];
1219             vdwjidx0D        = 2*vdwtype[jnrD+0];
1220
1221             fjx0             = _mm_setzero_ps();
1222             fjy0             = _mm_setzero_ps();
1223             fjz0             = _mm_setzero_ps();
1224
1225             /**************************
1226              * CALCULATE INTERACTIONS *
1227              **************************/
1228
1229             r00              = _mm_mul_ps(rsq00,rinv00);
1230             r00              = _mm_andnot_ps(dummy_mask,r00);
1231
1232             /* Compute parameters for interactions between i and j atoms */
1233             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1234                                          vdwparam+vdwioffset0+vdwjidx0B,
1235                                          vdwparam+vdwioffset0+vdwjidx0C,
1236                                          vdwparam+vdwioffset0+vdwjidx0D,
1237                                          &c6_00,&c12_00);
1238
1239             /* Calculate table index by multiplying r with table scale and truncate to integer */
1240             rt               = _mm_mul_ps(r00,vftabscale);
1241             vfitab           = _mm_cvttps_epi32(rt);
1242             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1243             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1244
1245             /* CUBIC SPLINE TABLE DISPERSION */
1246             vfitab           = _mm_add_epi32(vfitab,ifour);
1247             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1248             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1249             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1250             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1251             _MM_TRANSPOSE4_PS(Y,F,G,H);
1252             Heps             = _mm_mul_ps(vfeps,H);
1253             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1254             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1255             fvdw6            = _mm_mul_ps(c6_00,FF);
1256
1257             /* CUBIC SPLINE TABLE REPULSION */
1258             vfitab           = _mm_add_epi32(vfitab,ifour);
1259             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1260             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1261             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1262             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1263             _MM_TRANSPOSE4_PS(Y,F,G,H);
1264             Heps             = _mm_mul_ps(vfeps,H);
1265             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1266             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1267             fvdw12           = _mm_mul_ps(c12_00,FF);
1268             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1269
1270             fscal            = fvdw;
1271
1272             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1273
1274             /* Calculate temporary vectorial force */
1275             tx               = _mm_mul_ps(fscal,dx00);
1276             ty               = _mm_mul_ps(fscal,dy00);
1277             tz               = _mm_mul_ps(fscal,dz00);
1278
1279             /* Update vectorial force */
1280             fix0             = _mm_add_ps(fix0,tx);
1281             fiy0             = _mm_add_ps(fiy0,ty);
1282             fiz0             = _mm_add_ps(fiz0,tz);
1283
1284             fjx0             = _mm_add_ps(fjx0,tx);
1285             fjy0             = _mm_add_ps(fjy0,ty);
1286             fjz0             = _mm_add_ps(fjz0,tz);
1287
1288             /**************************
1289              * CALCULATE INTERACTIONS *
1290              **************************/
1291
1292             r10              = _mm_mul_ps(rsq10,rinv10);
1293             r10              = _mm_andnot_ps(dummy_mask,r10);
1294
1295             /* Compute parameters for interactions between i and j atoms */
1296             qq10             = _mm_mul_ps(iq1,jq0);
1297
1298             /* Calculate table index by multiplying r with table scale and truncate to integer */
1299             rt               = _mm_mul_ps(r10,vftabscale);
1300             vfitab           = _mm_cvttps_epi32(rt);
1301             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1302             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1303
1304             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1305             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1306             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1307             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1308             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1309             _MM_TRANSPOSE4_PS(Y,F,G,H);
1310             Heps             = _mm_mul_ps(vfeps,H);
1311             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1312             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1313             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1314
1315             fscal            = felec;
1316
1317             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1318
1319             /* Calculate temporary vectorial force */
1320             tx               = _mm_mul_ps(fscal,dx10);
1321             ty               = _mm_mul_ps(fscal,dy10);
1322             tz               = _mm_mul_ps(fscal,dz10);
1323
1324             /* Update vectorial force */
1325             fix1             = _mm_add_ps(fix1,tx);
1326             fiy1             = _mm_add_ps(fiy1,ty);
1327             fiz1             = _mm_add_ps(fiz1,tz);
1328
1329             fjx0             = _mm_add_ps(fjx0,tx);
1330             fjy0             = _mm_add_ps(fjy0,ty);
1331             fjz0             = _mm_add_ps(fjz0,tz);
1332
1333             /**************************
1334              * CALCULATE INTERACTIONS *
1335              **************************/
1336
1337             r20              = _mm_mul_ps(rsq20,rinv20);
1338             r20              = _mm_andnot_ps(dummy_mask,r20);
1339
1340             /* Compute parameters for interactions between i and j atoms */
1341             qq20             = _mm_mul_ps(iq2,jq0);
1342
1343             /* Calculate table index by multiplying r with table scale and truncate to integer */
1344             rt               = _mm_mul_ps(r20,vftabscale);
1345             vfitab           = _mm_cvttps_epi32(rt);
1346             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1347             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1348
1349             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1350             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1351             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1352             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1353             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1354             _MM_TRANSPOSE4_PS(Y,F,G,H);
1355             Heps             = _mm_mul_ps(vfeps,H);
1356             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1357             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1358             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1359
1360             fscal            = felec;
1361
1362             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1363
1364             /* Calculate temporary vectorial force */
1365             tx               = _mm_mul_ps(fscal,dx20);
1366             ty               = _mm_mul_ps(fscal,dy20);
1367             tz               = _mm_mul_ps(fscal,dz20);
1368
1369             /* Update vectorial force */
1370             fix2             = _mm_add_ps(fix2,tx);
1371             fiy2             = _mm_add_ps(fiy2,ty);
1372             fiz2             = _mm_add_ps(fiz2,tz);
1373
1374             fjx0             = _mm_add_ps(fjx0,tx);
1375             fjy0             = _mm_add_ps(fjy0,ty);
1376             fjz0             = _mm_add_ps(fjz0,tz);
1377
1378             /**************************
1379              * CALCULATE INTERACTIONS *
1380              **************************/
1381
1382             r30              = _mm_mul_ps(rsq30,rinv30);
1383             r30              = _mm_andnot_ps(dummy_mask,r30);
1384
1385             /* Compute parameters for interactions between i and j atoms */
1386             qq30             = _mm_mul_ps(iq3,jq0);
1387
1388             /* Calculate table index by multiplying r with table scale and truncate to integer */
1389             rt               = _mm_mul_ps(r30,vftabscale);
1390             vfitab           = _mm_cvttps_epi32(rt);
1391             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1392             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1393
1394             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1395             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1396             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1397             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1398             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1399             _MM_TRANSPOSE4_PS(Y,F,G,H);
1400             Heps             = _mm_mul_ps(vfeps,H);
1401             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1402             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1403             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1404
1405             fscal            = felec;
1406
1407             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1408
1409             /* Calculate temporary vectorial force */
1410             tx               = _mm_mul_ps(fscal,dx30);
1411             ty               = _mm_mul_ps(fscal,dy30);
1412             tz               = _mm_mul_ps(fscal,dz30);
1413
1414             /* Update vectorial force */
1415             fix3             = _mm_add_ps(fix3,tx);
1416             fiy3             = _mm_add_ps(fiy3,ty);
1417             fiz3             = _mm_add_ps(fiz3,tz);
1418
1419             fjx0             = _mm_add_ps(fjx0,tx);
1420             fjy0             = _mm_add_ps(fjy0,ty);
1421             fjz0             = _mm_add_ps(fjz0,tz);
1422
1423             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1424             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1425             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1426             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1427
1428             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1429
1430             /* Inner loop uses 169 flops */
1431         }
1432
1433         /* End of innermost loop */
1434
1435         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1436                                               f+i_coord_offset,fshift+i_shift_offset);
1437
1438         /* Increment number of inner iterations */
1439         inneriter                  += j_index_end - j_index_start;
1440
1441         /* Outer loop uses 24 flops */
1442     }
1443
1444     /* Increment number of outer iterations */
1445     outeriter        += nri;
1446
1447     /* Update outer/inner flops */
1448
1449     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*169);
1450 }