src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_sse2_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse2_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_sse2_single.h"
  34 #include "kernelutil_x86_sse2_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_VF_sse2_single
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_VF_sse2_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              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  76     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  77     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  78     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  79     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  80     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     int              nvdwtype;
  83     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  84     int              *vdwtype;
  85     real             *vdwparam;
  86     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  87     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  88     __m128i          vfitab;
  89     __m128i          ifour       = _mm_set1_epi32(4);
  90     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  91     real             *vftab;
  92     __m128           dummy_mask,cutoff_mask;
  93     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  94     __m128           one     = _mm_set1_ps(1.0);
  95     __m128           two     = _mm_set1_ps(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_ps(fr->epsfac);
 108     charge           = mdatoms->chargeA;
 109     nvdwtype         = fr->ntype;
 110     vdwparam         = fr->nbfp;
 111     vdwtype          = mdatoms->typeA;
 112
 113     vftab            = kernel_data->table_elec->data;
 114     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 115
 116     /* Setup water-specific parameters */
 117     inr              = nlist->iinr[0];
 118     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 119     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 120     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 121     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 122
 123     /* Avoid stupid compiler warnings */
 124     jnrA = jnrB = jnrC = jnrD = 0;
 125     j_coord_offsetA = 0;
 126     j_coord_offsetB = 0;
 127     j_coord_offsetC = 0;
 128     j_coord_offsetD = 0;
 129
 130     outeriter        = 0;
 131     inneriter        = 0;
 132
 133     for(iidx=0;iidx<4*DIM;iidx++)
 134     {
 135         scratch[iidx] = 0.0;
 136     }
 137
 138     /* Start outer loop over neighborlists */
 139     for(iidx=0; iidx<nri; iidx++)
 140     {
 141         /* Load shift vector for this list */
 142         i_shift_offset   = DIM*shiftidx[iidx];
 143
 144         /* Load limits for loop over neighbors */
 145         j_index_start    = jindex[iidx];
 146         j_index_end      = jindex[iidx+1];
 147
 148         /* Get outer coordinate index */
 149         inr              = iinr[iidx];
 150         i_coord_offset   = DIM*inr;
 151
 152         /* Load i particle coords and add shift vector */
 153         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 154                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 155
 156         fix0             = _mm_setzero_ps();
 157         fiy0             = _mm_setzero_ps();
 158         fiz0             = _mm_setzero_ps();
 159         fix1             = _mm_setzero_ps();
 160         fiy1             = _mm_setzero_ps();
 161         fiz1             = _mm_setzero_ps();
 162         fix2             = _mm_setzero_ps();
 163         fiy2             = _mm_setzero_ps();
 164         fiz2             = _mm_setzero_ps();
 165
 166         /* Reset potential sums */
 167         velecsum         = _mm_setzero_ps();
 168         vvdwsum          = _mm_setzero_ps();
 169
 170         /* Start inner kernel loop */
 171         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 172         {
 173
 174             /* Get j neighbor index, and coordinate index */
 175             jnrA             = jjnr[jidx];
 176             jnrB             = jjnr[jidx+1];
 177             jnrC             = jjnr[jidx+2];
 178             jnrD             = jjnr[jidx+3];
 179             j_coord_offsetA  = DIM*jnrA;
 180             j_coord_offsetB  = DIM*jnrB;
 181             j_coord_offsetC  = DIM*jnrC;
 182             j_coord_offsetD  = DIM*jnrD;
 183
 184             /* load j atom coordinates */
 185             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 186                                               x+j_coord_offsetC,x+j_coord_offsetD,
 187                                               &jx0,&jy0,&jz0);
 188
 189             /* Calculate displacement vector */
 190             dx00             = _mm_sub_ps(ix0,jx0);
 191             dy00             = _mm_sub_ps(iy0,jy0);
 192             dz00             = _mm_sub_ps(iz0,jz0);
 193             dx10             = _mm_sub_ps(ix1,jx0);
 194             dy10             = _mm_sub_ps(iy1,jy0);
 195             dz10             = _mm_sub_ps(iz1,jz0);
 196             dx20             = _mm_sub_ps(ix2,jx0);
 197             dy20             = _mm_sub_ps(iy2,jy0);
 198             dz20             = _mm_sub_ps(iz2,jz0);
 199
 200             /* Calculate squared distance and things based on it */
 201             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 202             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 203             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 204
 205             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 206             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 207             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 208
 209             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 210
 211             /* Load parameters for j particles */
 212             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 213                                                               charge+jnrC+0,charge+jnrD+0);
 214             vdwjidx0A        = 2*vdwtype[jnrA+0];
 215             vdwjidx0B        = 2*vdwtype[jnrB+0];
 216             vdwjidx0C        = 2*vdwtype[jnrC+0];
 217             vdwjidx0D        = 2*vdwtype[jnrD+0];
 218
 219             fjx0             = _mm_setzero_ps();
 220             fjy0             = _mm_setzero_ps();
 221             fjz0             = _mm_setzero_ps();
 222
 223             /**************************
 224              * CALCULATE INTERACTIONS *
 225              **************************/
 226
 227             r00              = _mm_mul_ps(rsq00,rinv00);
 228
 229             /* Compute parameters for interactions between i and j atoms */
 230             qq00             = _mm_mul_ps(iq0,jq0);
 231             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 232                                          vdwparam+vdwioffset0+vdwjidx0B,
 233                                          vdwparam+vdwioffset0+vdwjidx0C,
 234                                          vdwparam+vdwioffset0+vdwjidx0D,
 235                                          &c6_00,&c12_00);
 236
 237             /* Calculate table index by multiplying r with table scale and truncate to integer */
 238             rt               = _mm_mul_ps(r00,vftabscale);
 239             vfitab           = _mm_cvttps_epi32(rt);
 240             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 241             vfitab           = _mm_slli_epi32(vfitab,2);
 242
 243             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 244             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 245             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 246             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 247             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 248             _MM_TRANSPOSE4_PS(Y,F,G,H);
 249             Heps             = _mm_mul_ps(vfeps,H);
 250             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 251             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 252             velec            = _mm_mul_ps(qq00,VV);
 253             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 254             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 255
 256             /* LENNARD-JONES DISPERSION/REPULSION */
 257
 258             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 259             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 260             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 261             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 262             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 263
 264             /* Update potential sum for this i atom from the interaction with this j atom. */
 265             velecsum         = _mm_add_ps(velecsum,velec);
 266             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 267
 268             fscal            = _mm_add_ps(felec,fvdw);
 269
 270             /* Calculate temporary vectorial force */
 271             tx               = _mm_mul_ps(fscal,dx00);
 272             ty               = _mm_mul_ps(fscal,dy00);
 273             tz               = _mm_mul_ps(fscal,dz00);
 274
 275             /* Update vectorial force */
 276             fix0             = _mm_add_ps(fix0,tx);
 277             fiy0             = _mm_add_ps(fiy0,ty);
 278             fiz0             = _mm_add_ps(fiz0,tz);
 279
 280             fjx0             = _mm_add_ps(fjx0,tx);
 281             fjy0             = _mm_add_ps(fjy0,ty);
 282             fjz0             = _mm_add_ps(fjz0,tz);
 283
 284             /**************************
 285              * CALCULATE INTERACTIONS *
 286              **************************/
 287
 288             r10              = _mm_mul_ps(rsq10,rinv10);
 289
 290             /* Compute parameters for interactions between i and j atoms */
 291             qq10             = _mm_mul_ps(iq1,jq0);
 292
 293             /* Calculate table index by multiplying r with table scale and truncate to integer */
 294             rt               = _mm_mul_ps(r10,vftabscale);
 295             vfitab           = _mm_cvttps_epi32(rt);
 296             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 297             vfitab           = _mm_slli_epi32(vfitab,2);
 298
 299             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 300             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 301             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 302             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 303             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 304             _MM_TRANSPOSE4_PS(Y,F,G,H);
 305             Heps             = _mm_mul_ps(vfeps,H);
 306             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 307             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 308             velec            = _mm_mul_ps(qq10,VV);
 309             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 310             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 311
 312             /* Update potential sum for this i atom from the interaction with this j atom. */
 313             velecsum         = _mm_add_ps(velecsum,velec);
 314
 315             fscal            = felec;
 316
 317             /* Calculate temporary vectorial force */
 318             tx               = _mm_mul_ps(fscal,dx10);
 319             ty               = _mm_mul_ps(fscal,dy10);
 320             tz               = _mm_mul_ps(fscal,dz10);
 321
 322             /* Update vectorial force */
 323             fix1             = _mm_add_ps(fix1,tx);
 324             fiy1             = _mm_add_ps(fiy1,ty);
 325             fiz1             = _mm_add_ps(fiz1,tz);
 326
 327             fjx0             = _mm_add_ps(fjx0,tx);
 328             fjy0             = _mm_add_ps(fjy0,ty);
 329             fjz0             = _mm_add_ps(fjz0,tz);
 330
 331             /**************************
 332              * CALCULATE INTERACTIONS *
 333              **************************/
 334
 335             r20              = _mm_mul_ps(rsq20,rinv20);
 336
 337             /* Compute parameters for interactions between i and j atoms */
 338             qq20             = _mm_mul_ps(iq2,jq0);
 339
 340             /* Calculate table index by multiplying r with table scale and truncate to integer */
 341             rt               = _mm_mul_ps(r20,vftabscale);
 342             vfitab           = _mm_cvttps_epi32(rt);
 343             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 344             vfitab           = _mm_slli_epi32(vfitab,2);
 345
 346             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 347             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 348             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 349             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 350             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 351             _MM_TRANSPOSE4_PS(Y,F,G,H);
 352             Heps             = _mm_mul_ps(vfeps,H);
 353             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 354             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 355             velec            = _mm_mul_ps(qq20,VV);
 356             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 357             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 358
 359             /* Update potential sum for this i atom from the interaction with this j atom. */
 360             velecsum         = _mm_add_ps(velecsum,velec);
 361
 362             fscal            = felec;
 363
 364             /* Calculate temporary vectorial force */
 365             tx               = _mm_mul_ps(fscal,dx20);
 366             ty               = _mm_mul_ps(fscal,dy20);
 367             tz               = _mm_mul_ps(fscal,dz20);
 368
 369             /* Update vectorial force */
 370             fix2             = _mm_add_ps(fix2,tx);
 371             fiy2             = _mm_add_ps(fiy2,ty);
 372             fiz2             = _mm_add_ps(fiz2,tz);
 373
 374             fjx0             = _mm_add_ps(fjx0,tx);
 375             fjy0             = _mm_add_ps(fjy0,ty);
 376             fjz0             = _mm_add_ps(fjz0,tz);
 377
 378             fjptrA             = f+j_coord_offsetA;
 379             fjptrB             = f+j_coord_offsetB;
 380             fjptrC             = f+j_coord_offsetC;
 381             fjptrD             = f+j_coord_offsetD;
 382
 383             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 384
 385             /* Inner loop uses 142 flops */
 386         }
 387
 388         if(jidx<j_index_end)
 389         {
 390
 391             /* Get j neighbor index, and coordinate index */
 392             jnrlistA         = jjnr[jidx];
 393             jnrlistB         = jjnr[jidx+1];
 394             jnrlistC         = jjnr[jidx+2];
 395             jnrlistD         = jjnr[jidx+3];
 396             /* Sign of each element will be negative for non-real atoms.
 397              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 398              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 399              */
 400             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 401             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 402             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 403             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 404             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 405             j_coord_offsetA  = DIM*jnrA;
 406             j_coord_offsetB  = DIM*jnrB;
 407             j_coord_offsetC  = DIM*jnrC;
 408             j_coord_offsetD  = DIM*jnrD;
 409
 410             /* load j atom coordinates */
 411             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 412                                               x+j_coord_offsetC,x+j_coord_offsetD,
 413                                               &jx0,&jy0,&jz0);
 414
 415             /* Calculate displacement vector */
 416             dx00             = _mm_sub_ps(ix0,jx0);
 417             dy00             = _mm_sub_ps(iy0,jy0);
 418             dz00             = _mm_sub_ps(iz0,jz0);
 419             dx10             = _mm_sub_ps(ix1,jx0);
 420             dy10             = _mm_sub_ps(iy1,jy0);
 421             dz10             = _mm_sub_ps(iz1,jz0);
 422             dx20             = _mm_sub_ps(ix2,jx0);
 423             dy20             = _mm_sub_ps(iy2,jy0);
 424             dz20             = _mm_sub_ps(iz2,jz0);
 425
 426             /* Calculate squared distance and things based on it */
 427             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 428             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 429             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 430
 431             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 432             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 433             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 434
 435             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 436
 437             /* Load parameters for j particles */
 438             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 439                                                               charge+jnrC+0,charge+jnrD+0);
 440             vdwjidx0A        = 2*vdwtype[jnrA+0];
 441             vdwjidx0B        = 2*vdwtype[jnrB+0];
 442             vdwjidx0C        = 2*vdwtype[jnrC+0];
 443             vdwjidx0D        = 2*vdwtype[jnrD+0];
 444
 445             fjx0             = _mm_setzero_ps();
 446             fjy0             = _mm_setzero_ps();
 447             fjz0             = _mm_setzero_ps();
 448
 449             /**************************
 450              * CALCULATE INTERACTIONS *
 451              **************************/
 452
 453             r00              = _mm_mul_ps(rsq00,rinv00);
 454             r00              = _mm_andnot_ps(dummy_mask,r00);
 455
 456             /* Compute parameters for interactions between i and j atoms */
 457             qq00             = _mm_mul_ps(iq0,jq0);
 458             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 459                                          vdwparam+vdwioffset0+vdwjidx0B,
 460                                          vdwparam+vdwioffset0+vdwjidx0C,
 461                                          vdwparam+vdwioffset0+vdwjidx0D,
 462                                          &c6_00,&c12_00);
 463
 464             /* Calculate table index by multiplying r with table scale and truncate to integer */
 465             rt               = _mm_mul_ps(r00,vftabscale);
 466             vfitab           = _mm_cvttps_epi32(rt);
 467             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 468             vfitab           = _mm_slli_epi32(vfitab,2);
 469
 470             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 471             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 472             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 473             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 474             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 475             _MM_TRANSPOSE4_PS(Y,F,G,H);
 476             Heps             = _mm_mul_ps(vfeps,H);
 477             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 478             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 479             velec            = _mm_mul_ps(qq00,VV);
 480             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 481             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 482
 483             /* LENNARD-JONES DISPERSION/REPULSION */
 484
 485             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 486             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 487             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 488             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 489             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 490
 491             /* Update potential sum for this i atom from the interaction with this j atom. */
 492             velec            = _mm_andnot_ps(dummy_mask,velec);
 493             velecsum         = _mm_add_ps(velecsum,velec);
 494             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 495             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 496
 497             fscal            = _mm_add_ps(felec,fvdw);
 498
 499             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 500
 501             /* Calculate temporary vectorial force */
 502             tx               = _mm_mul_ps(fscal,dx00);
 503             ty               = _mm_mul_ps(fscal,dy00);
 504             tz               = _mm_mul_ps(fscal,dz00);
 505
 506             /* Update vectorial force */
 507             fix0             = _mm_add_ps(fix0,tx);
 508             fiy0             = _mm_add_ps(fiy0,ty);
 509             fiz0             = _mm_add_ps(fiz0,tz);
 510
 511             fjx0             = _mm_add_ps(fjx0,tx);
 512             fjy0             = _mm_add_ps(fjy0,ty);
 513             fjz0             = _mm_add_ps(fjz0,tz);
 514
 515             /**************************
 516              * CALCULATE INTERACTIONS *
 517              **************************/
 518
 519             r10              = _mm_mul_ps(rsq10,rinv10);
 520             r10              = _mm_andnot_ps(dummy_mask,r10);
 521
 522             /* Compute parameters for interactions between i and j atoms */
 523             qq10             = _mm_mul_ps(iq1,jq0);
 524
 525             /* Calculate table index by multiplying r with table scale and truncate to integer */
 526             rt               = _mm_mul_ps(r10,vftabscale);
 527             vfitab           = _mm_cvttps_epi32(rt);
 528             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 529             vfitab           = _mm_slli_epi32(vfitab,2);
 530
 531             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 532             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 533             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 534             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 535             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 536             _MM_TRANSPOSE4_PS(Y,F,G,H);
 537             Heps             = _mm_mul_ps(vfeps,H);
 538             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 539             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 540             velec            = _mm_mul_ps(qq10,VV);
 541             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 542             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 543
 544             /* Update potential sum for this i atom from the interaction with this j atom. */
 545             velec            = _mm_andnot_ps(dummy_mask,velec);
 546             velecsum         = _mm_add_ps(velecsum,velec);
 547
 548             fscal            = felec;
 549
 550             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 551
 552             /* Calculate temporary vectorial force */
 553             tx               = _mm_mul_ps(fscal,dx10);
 554             ty               = _mm_mul_ps(fscal,dy10);
 555             tz               = _mm_mul_ps(fscal,dz10);
 556
 557             /* Update vectorial force */
 558             fix1             = _mm_add_ps(fix1,tx);
 559             fiy1             = _mm_add_ps(fiy1,ty);
 560             fiz1             = _mm_add_ps(fiz1,tz);
 561
 562             fjx0             = _mm_add_ps(fjx0,tx);
 563             fjy0             = _mm_add_ps(fjy0,ty);
 564             fjz0             = _mm_add_ps(fjz0,tz);
 565
 566             /**************************
 567              * CALCULATE INTERACTIONS *
 568              **************************/
 569
 570             r20              = _mm_mul_ps(rsq20,rinv20);
 571             r20              = _mm_andnot_ps(dummy_mask,r20);
 572
 573             /* Compute parameters for interactions between i and j atoms */
 574             qq20             = _mm_mul_ps(iq2,jq0);
 575
 576             /* Calculate table index by multiplying r with table scale and truncate to integer */
 577             rt               = _mm_mul_ps(r20,vftabscale);
 578             vfitab           = _mm_cvttps_epi32(rt);
 579             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 580             vfitab           = _mm_slli_epi32(vfitab,2);
 581
 582             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 583             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 584             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 585             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 586             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 587             _MM_TRANSPOSE4_PS(Y,F,G,H);
 588             Heps             = _mm_mul_ps(vfeps,H);
 589             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 590             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 591             velec            = _mm_mul_ps(qq20,VV);
 592             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 593             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 594
 595             /* Update potential sum for this i atom from the interaction with this j atom. */
 596             velec            = _mm_andnot_ps(dummy_mask,velec);
 597             velecsum         = _mm_add_ps(velecsum,velec);
 598
 599             fscal            = felec;
 600
 601             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 602
 603             /* Calculate temporary vectorial force */
 604             tx               = _mm_mul_ps(fscal,dx20);
 605             ty               = _mm_mul_ps(fscal,dy20);
 606             tz               = _mm_mul_ps(fscal,dz20);
 607
 608             /* Update vectorial force */
 609             fix2             = _mm_add_ps(fix2,tx);
 610             fiy2             = _mm_add_ps(fiy2,ty);
 611             fiz2             = _mm_add_ps(fiz2,tz);
 612
 613             fjx0             = _mm_add_ps(fjx0,tx);
 614             fjy0             = _mm_add_ps(fjy0,ty);
 615             fjz0             = _mm_add_ps(fjz0,tz);
 616
 617             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 618             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 619             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 620             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 621
 622             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 623
 624             /* Inner loop uses 145 flops */
 625         }
 626
 627         /* End of innermost loop */
 628
 629         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 630                                               f+i_coord_offset,fshift+i_shift_offset);
 631
 632         ggid                        = gid[iidx];
 633         /* Update potential energies */
 634         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 635         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 636
 637         /* Increment number of inner iterations */
 638         inneriter                  += j_index_end - j_index_start;
 639
 640         /* Outer loop uses 20 flops */
 641     }
 642
 643     /* Increment number of outer iterations */
 644     outeriter        += nri;
 645
 646     /* Update outer/inner flops */
 647
 648     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*145);
 649 }
 650 /*
 651  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_sse2_single
 652  * Electrostatics interaction: CubicSplineTable
 653  * VdW interaction:            LennardJones
 654  * Geometry:                   Water3-Particle
 655  * Calculate force/pot:        Force
 656  */
 657 void
 658 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_sse2_single
 659                     (t_nblist * gmx_restrict                nlist,
 660                      rvec * gmx_restrict                    xx,
 661                      rvec * gmx_restrict                    ff,
 662                      t_forcerec * gmx_restrict              fr,
 663                      t_mdatoms * gmx_restrict               mdatoms,
 664                      nb_kernel_data_t * gmx_restrict        kernel_data,
 665                      t_nrnb * gmx_restrict                  nrnb)
 666 {
 667     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 668      * just 0 for non-waters.
 669      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 670      * jnr indices corresponding to data put in the four positions in the SIMD register.
 671      */
 672     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 673     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 674     int              jnrA,jnrB,jnrC,jnrD;
 675     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 676     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 677     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 678     real             rcutoff_scalar;
 679     real             *shiftvec,*fshift,*x,*f;
 680     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 681     real             scratch[4*DIM];
 682     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 683     int              vdwioffset0;
 684     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 685     int              vdwioffset1;
 686     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 687     int              vdwioffset2;
 688     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 689     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 690     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 691     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 692     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 693     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 694     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 695     real             *charge;
 696     int              nvdwtype;
 697     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 698     int              *vdwtype;
 699     real             *vdwparam;
 700     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 701     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 702     __m128i          vfitab;
 703     __m128i          ifour       = _mm_set1_epi32(4);
 704     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 705     real             *vftab;
 706     __m128           dummy_mask,cutoff_mask;
 707     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 708     __m128           one     = _mm_set1_ps(1.0);
 709     __m128           two     = _mm_set1_ps(2.0);
 710     x                = xx[0];
 711     f                = ff[0];
 712
 713     nri              = nlist->nri;
 714     iinr             = nlist->iinr;
 715     jindex           = nlist->jindex;
 716     jjnr             = nlist->jjnr;
 717     shiftidx         = nlist->shift;
 718     gid              = nlist->gid;
 719     shiftvec         = fr->shift_vec[0];
 720     fshift           = fr->fshift[0];
 721     facel            = _mm_set1_ps(fr->epsfac);
 722     charge           = mdatoms->chargeA;
 723     nvdwtype         = fr->ntype;
 724     vdwparam         = fr->nbfp;
 725     vdwtype          = mdatoms->typeA;
 726
 727     vftab            = kernel_data->table_elec->data;
 728     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 729
 730     /* Setup water-specific parameters */
 731     inr              = nlist->iinr[0];
 732     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 733     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 734     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 735     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 736
 737     /* Avoid stupid compiler warnings */
 738     jnrA = jnrB = jnrC = jnrD = 0;
 739     j_coord_offsetA = 0;
 740     j_coord_offsetB = 0;
 741     j_coord_offsetC = 0;
 742     j_coord_offsetD = 0;
 743
 744     outeriter        = 0;
 745     inneriter        = 0;
 746
 747     for(iidx=0;iidx<4*DIM;iidx++)
 748     {
 749         scratch[iidx] = 0.0;
 750     }
 751
 752     /* Start outer loop over neighborlists */
 753     for(iidx=0; iidx<nri; iidx++)
 754     {
 755         /* Load shift vector for this list */
 756         i_shift_offset   = DIM*shiftidx[iidx];
 757
 758         /* Load limits for loop over neighbors */
 759         j_index_start    = jindex[iidx];
 760         j_index_end      = jindex[iidx+1];
 761
 762         /* Get outer coordinate index */
 763         inr              = iinr[iidx];
 764         i_coord_offset   = DIM*inr;
 765
 766         /* Load i particle coords and add shift vector */
 767         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 768                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 769
 770         fix0             = _mm_setzero_ps();
 771         fiy0             = _mm_setzero_ps();
 772         fiz0             = _mm_setzero_ps();
 773         fix1             = _mm_setzero_ps();
 774         fiy1             = _mm_setzero_ps();
 775         fiz1             = _mm_setzero_ps();
 776         fix2             = _mm_setzero_ps();
 777         fiy2             = _mm_setzero_ps();
 778         fiz2             = _mm_setzero_ps();
 779
 780         /* Start inner kernel loop */
 781         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 782         {
 783
 784             /* Get j neighbor index, and coordinate index */
 785             jnrA             = jjnr[jidx];
 786             jnrB             = jjnr[jidx+1];
 787             jnrC             = jjnr[jidx+2];
 788             jnrD             = jjnr[jidx+3];
 789             j_coord_offsetA  = DIM*jnrA;
 790             j_coord_offsetB  = DIM*jnrB;
 791             j_coord_offsetC  = DIM*jnrC;
 792             j_coord_offsetD  = DIM*jnrD;
 793
 794             /* load j atom coordinates */
 795             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 796                                               x+j_coord_offsetC,x+j_coord_offsetD,
 797                                               &jx0,&jy0,&jz0);
 798
 799             /* Calculate displacement vector */
 800             dx00             = _mm_sub_ps(ix0,jx0);
 801             dy00             = _mm_sub_ps(iy0,jy0);
 802             dz00             = _mm_sub_ps(iz0,jz0);
 803             dx10             = _mm_sub_ps(ix1,jx0);
 804             dy10             = _mm_sub_ps(iy1,jy0);
 805             dz10             = _mm_sub_ps(iz1,jz0);
 806             dx20             = _mm_sub_ps(ix2,jx0);
 807             dy20             = _mm_sub_ps(iy2,jy0);
 808             dz20             = _mm_sub_ps(iz2,jz0);
 809
 810             /* Calculate squared distance and things based on it */
 811             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 812             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 813             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 814
 815             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 816             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 817             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 818
 819             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 820
 821             /* Load parameters for j particles */
 822             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 823                                                               charge+jnrC+0,charge+jnrD+0);
 824             vdwjidx0A        = 2*vdwtype[jnrA+0];
 825             vdwjidx0B        = 2*vdwtype[jnrB+0];
 826             vdwjidx0C        = 2*vdwtype[jnrC+0];
 827             vdwjidx0D        = 2*vdwtype[jnrD+0];
 828
 829             fjx0             = _mm_setzero_ps();
 830             fjy0             = _mm_setzero_ps();
 831             fjz0             = _mm_setzero_ps();
 832
 833             /**************************
 834              * CALCULATE INTERACTIONS *
 835              **************************/
 836
 837             r00              = _mm_mul_ps(rsq00,rinv00);
 838
 839             /* Compute parameters for interactions between i and j atoms */
 840             qq00             = _mm_mul_ps(iq0,jq0);
 841             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 842                                          vdwparam+vdwioffset0+vdwjidx0B,
 843                                          vdwparam+vdwioffset0+vdwjidx0C,
 844                                          vdwparam+vdwioffset0+vdwjidx0D,
 845                                          &c6_00,&c12_00);
 846
 847             /* Calculate table index by multiplying r with table scale and truncate to integer */
 848             rt               = _mm_mul_ps(r00,vftabscale);
 849             vfitab           = _mm_cvttps_epi32(rt);
 850             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 851             vfitab           = _mm_slli_epi32(vfitab,2);
 852
 853             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 854             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 855             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 856             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 857             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 858             _MM_TRANSPOSE4_PS(Y,F,G,H);
 859             Heps             = _mm_mul_ps(vfeps,H);
 860             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 861             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 862             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 863
 864             /* LENNARD-JONES DISPERSION/REPULSION */
 865
 866             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 867             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 868
 869             fscal            = _mm_add_ps(felec,fvdw);
 870
 871             /* Calculate temporary vectorial force */
 872             tx               = _mm_mul_ps(fscal,dx00);
 873             ty               = _mm_mul_ps(fscal,dy00);
 874             tz               = _mm_mul_ps(fscal,dz00);
 875
 876             /* Update vectorial force */
 877             fix0             = _mm_add_ps(fix0,tx);
 878             fiy0             = _mm_add_ps(fiy0,ty);
 879             fiz0             = _mm_add_ps(fiz0,tz);
 880
 881             fjx0             = _mm_add_ps(fjx0,tx);
 882             fjy0             = _mm_add_ps(fjy0,ty);
 883             fjz0             = _mm_add_ps(fjz0,tz);
 884
 885             /**************************
 886              * CALCULATE INTERACTIONS *
 887              **************************/
 888
 889             r10              = _mm_mul_ps(rsq10,rinv10);
 890
 891             /* Compute parameters for interactions between i and j atoms */
 892             qq10             = _mm_mul_ps(iq1,jq0);
 893
 894             /* Calculate table index by multiplying r with table scale and truncate to integer */
 895             rt               = _mm_mul_ps(r10,vftabscale);
 896             vfitab           = _mm_cvttps_epi32(rt);
 897             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 898             vfitab           = _mm_slli_epi32(vfitab,2);
 899
 900             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 901             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 902             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 903             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 904             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 905             _MM_TRANSPOSE4_PS(Y,F,G,H);
 906             Heps             = _mm_mul_ps(vfeps,H);
 907             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 908             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 909             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 910
 911             fscal            = felec;
 912
 913             /* Calculate temporary vectorial force */
 914             tx               = _mm_mul_ps(fscal,dx10);
 915             ty               = _mm_mul_ps(fscal,dy10);
 916             tz               = _mm_mul_ps(fscal,dz10);
 917
 918             /* Update vectorial force */
 919             fix1             = _mm_add_ps(fix1,tx);
 920             fiy1             = _mm_add_ps(fiy1,ty);
 921             fiz1             = _mm_add_ps(fiz1,tz);
 922
 923             fjx0             = _mm_add_ps(fjx0,tx);
 924             fjy0             = _mm_add_ps(fjy0,ty);
 925             fjz0             = _mm_add_ps(fjz0,tz);
 926
 927             /**************************
 928              * CALCULATE INTERACTIONS *
 929              **************************/
 930
 931             r20              = _mm_mul_ps(rsq20,rinv20);
 932
 933             /* Compute parameters for interactions between i and j atoms */
 934             qq20             = _mm_mul_ps(iq2,jq0);
 935
 936             /* Calculate table index by multiplying r with table scale and truncate to integer */
 937             rt               = _mm_mul_ps(r20,vftabscale);
 938             vfitab           = _mm_cvttps_epi32(rt);
 939             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 940             vfitab           = _mm_slli_epi32(vfitab,2);
 941
 942             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 943             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 944             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 945             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 946             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 947             _MM_TRANSPOSE4_PS(Y,F,G,H);
 948             Heps             = _mm_mul_ps(vfeps,H);
 949             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 950             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 951             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 952
 953             fscal            = felec;
 954
 955             /* Calculate temporary vectorial force */
 956             tx               = _mm_mul_ps(fscal,dx20);
 957             ty               = _mm_mul_ps(fscal,dy20);
 958             tz               = _mm_mul_ps(fscal,dz20);
 959
 960             /* Update vectorial force */
 961             fix2             = _mm_add_ps(fix2,tx);
 962             fiy2             = _mm_add_ps(fiy2,ty);
 963             fiz2             = _mm_add_ps(fiz2,tz);
 964
 965             fjx0             = _mm_add_ps(fjx0,tx);
 966             fjy0             = _mm_add_ps(fjy0,ty);
 967             fjz0             = _mm_add_ps(fjz0,tz);
 968
 969             fjptrA             = f+j_coord_offsetA;
 970             fjptrB             = f+j_coord_offsetB;
 971             fjptrC             = f+j_coord_offsetC;
 972             fjptrD             = f+j_coord_offsetD;
 973
 974             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 975
 976             /* Inner loop uses 125 flops */
 977         }
 978
 979         if(jidx<j_index_end)
 980         {
 981
 982             /* Get j neighbor index, and coordinate index */
 983             jnrlistA         = jjnr[jidx];
 984             jnrlistB         = jjnr[jidx+1];
 985             jnrlistC         = jjnr[jidx+2];
 986             jnrlistD         = jjnr[jidx+3];
 987             /* Sign of each element will be negative for non-real atoms.
 988              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 989              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 990              */
 991             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 992             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 993             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 994             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 995             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 996             j_coord_offsetA  = DIM*jnrA;
 997             j_coord_offsetB  = DIM*jnrB;
 998             j_coord_offsetC  = DIM*jnrC;
 999             j_coord_offsetD  = DIM*jnrD;
1000
1001             /* load j atom coordinates */
1002             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1003                                               x+j_coord_offsetC,x+j_coord_offsetD,
1004                                               &jx0,&jy0,&jz0);
1005
1006             /* Calculate displacement vector */
1007             dx00             = _mm_sub_ps(ix0,jx0);
1008             dy00             = _mm_sub_ps(iy0,jy0);
1009             dz00             = _mm_sub_ps(iz0,jz0);
1010             dx10             = _mm_sub_ps(ix1,jx0);
1011             dy10             = _mm_sub_ps(iy1,jy0);
1012             dz10             = _mm_sub_ps(iz1,jz0);
1013             dx20             = _mm_sub_ps(ix2,jx0);
1014             dy20             = _mm_sub_ps(iy2,jy0);
1015             dz20             = _mm_sub_ps(iz2,jz0);
1016
1017             /* Calculate squared distance and things based on it */
1018             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1019             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1020             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1021
1022             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1023             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1024             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1025
1026             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1027
1028             /* Load parameters for j particles */
1029             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1030                                                               charge+jnrC+0,charge+jnrD+0);
1031             vdwjidx0A        = 2*vdwtype[jnrA+0];
1032             vdwjidx0B        = 2*vdwtype[jnrB+0];
1033             vdwjidx0C        = 2*vdwtype[jnrC+0];
1034             vdwjidx0D        = 2*vdwtype[jnrD+0];
1035
1036             fjx0             = _mm_setzero_ps();
1037             fjy0             = _mm_setzero_ps();
1038             fjz0             = _mm_setzero_ps();
1039
1040             /**************************
1041              * CALCULATE INTERACTIONS *
1042              **************************/
1043
1044             r00              = _mm_mul_ps(rsq00,rinv00);
1045             r00              = _mm_andnot_ps(dummy_mask,r00);
1046
1047             /* Compute parameters for interactions between i and j atoms */
1048             qq00             = _mm_mul_ps(iq0,jq0);
1049             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1050                                          vdwparam+vdwioffset0+vdwjidx0B,
1051                                          vdwparam+vdwioffset0+vdwjidx0C,
1052                                          vdwparam+vdwioffset0+vdwjidx0D,
1053                                          &c6_00,&c12_00);
1054
1055             /* Calculate table index by multiplying r with table scale and truncate to integer */
1056             rt               = _mm_mul_ps(r00,vftabscale);
1057             vfitab           = _mm_cvttps_epi32(rt);
1058             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1059             vfitab           = _mm_slli_epi32(vfitab,2);
1060
1061             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1062             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1063             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1064             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1065             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1066             _MM_TRANSPOSE4_PS(Y,F,G,H);
1067             Heps             = _mm_mul_ps(vfeps,H);
1068             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1069             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1070             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
1071
1072             /* LENNARD-JONES DISPERSION/REPULSION */
1073
1074             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1075             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1076
1077             fscal            = _mm_add_ps(felec,fvdw);
1078
1079             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1080
1081             /* Calculate temporary vectorial force */
1082             tx               = _mm_mul_ps(fscal,dx00);
1083             ty               = _mm_mul_ps(fscal,dy00);
1084             tz               = _mm_mul_ps(fscal,dz00);
1085
1086             /* Update vectorial force */
1087             fix0             = _mm_add_ps(fix0,tx);
1088             fiy0             = _mm_add_ps(fiy0,ty);
1089             fiz0             = _mm_add_ps(fiz0,tz);
1090
1091             fjx0             = _mm_add_ps(fjx0,tx);
1092             fjy0             = _mm_add_ps(fjy0,ty);
1093             fjz0             = _mm_add_ps(fjz0,tz);
1094
1095             /**************************
1096              * CALCULATE INTERACTIONS *
1097              **************************/
1098
1099             r10              = _mm_mul_ps(rsq10,rinv10);
1100             r10              = _mm_andnot_ps(dummy_mask,r10);
1101
1102             /* Compute parameters for interactions between i and j atoms */
1103             qq10             = _mm_mul_ps(iq1,jq0);
1104
1105             /* Calculate table index by multiplying r with table scale and truncate to integer */
1106             rt               = _mm_mul_ps(r10,vftabscale);
1107             vfitab           = _mm_cvttps_epi32(rt);
1108             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1109             vfitab           = _mm_slli_epi32(vfitab,2);
1110
1111             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1112             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1113             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1114             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1115             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1116             _MM_TRANSPOSE4_PS(Y,F,G,H);
1117             Heps             = _mm_mul_ps(vfeps,H);
1118             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1119             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1120             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1121
1122             fscal            = felec;
1123
1124             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1125
1126             /* Calculate temporary vectorial force */
1127             tx               = _mm_mul_ps(fscal,dx10);
1128             ty               = _mm_mul_ps(fscal,dy10);
1129             tz               = _mm_mul_ps(fscal,dz10);
1130
1131             /* Update vectorial force */
1132             fix1             = _mm_add_ps(fix1,tx);
1133             fiy1             = _mm_add_ps(fiy1,ty);
1134             fiz1             = _mm_add_ps(fiz1,tz);
1135
1136             fjx0             = _mm_add_ps(fjx0,tx);
1137             fjy0             = _mm_add_ps(fjy0,ty);
1138             fjz0             = _mm_add_ps(fjz0,tz);
1139
1140             /**************************
1141              * CALCULATE INTERACTIONS *
1142              **************************/
1143
1144             r20              = _mm_mul_ps(rsq20,rinv20);
1145             r20              = _mm_andnot_ps(dummy_mask,r20);
1146
1147             /* Compute parameters for interactions between i and j atoms */
1148             qq20             = _mm_mul_ps(iq2,jq0);
1149
1150             /* Calculate table index by multiplying r with table scale and truncate to integer */
1151             rt               = _mm_mul_ps(r20,vftabscale);
1152             vfitab           = _mm_cvttps_epi32(rt);
1153             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1154             vfitab           = _mm_slli_epi32(vfitab,2);
1155
1156             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1157             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1158             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1159             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1160             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1161             _MM_TRANSPOSE4_PS(Y,F,G,H);
1162             Heps             = _mm_mul_ps(vfeps,H);
1163             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1164             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1165             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1166
1167             fscal            = felec;
1168
1169             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1170
1171             /* Calculate temporary vectorial force */
1172             tx               = _mm_mul_ps(fscal,dx20);
1173             ty               = _mm_mul_ps(fscal,dy20);
1174             tz               = _mm_mul_ps(fscal,dz20);
1175
1176             /* Update vectorial force */
1177             fix2             = _mm_add_ps(fix2,tx);
1178             fiy2             = _mm_add_ps(fiy2,ty);
1179             fiz2             = _mm_add_ps(fiz2,tz);
1180
1181             fjx0             = _mm_add_ps(fjx0,tx);
1182             fjy0             = _mm_add_ps(fjy0,ty);
1183             fjz0             = _mm_add_ps(fjz0,tz);
1184
1185             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1186             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1187             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1188             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1189
1190             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1191
1192             /* Inner loop uses 128 flops */
1193         }
1194
1195         /* End of innermost loop */
1196
1197         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1198                                               f+i_coord_offset,fshift+i_shift_offset);
1199
1200         /* Increment number of inner iterations */
1201         inneriter                  += j_index_end - j_index_start;
1202
1203         /* Outer loop uses 18 flops */
1204     }
1205
1206     /* Increment number of outer iterations */
1207     outeriter        += nri;
1208
1209     /* Update outer/inner flops */
1210
1211     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*128);
1212 }