src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_sse4_1_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse4_1_single.h"
  48 #include "kernelutil_x86_sse4_1_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_VF_sse4_1_single
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_VF_sse4_1_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwioffset1;
  86     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     int              vdwioffset2;
  88     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  92     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  93     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  94     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     int              nvdwtype;
  97     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 101     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 102     __m128i          vfitab;
 103     __m128i          ifour       = _mm_set1_epi32(4);
 104     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 105     real             *vftab;
 106     __m128           dummy_mask,cutoff_mask;
 107     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 108     __m128           one     = _mm_set1_ps(1.0);
 109     __m128           two     = _mm_set1_ps(2.0);
 110     x                = xx[0];
 111     f                = ff[0];
 112
 113     nri              = nlist->nri;
 114     iinr             = nlist->iinr;
 115     jindex           = nlist->jindex;
 116     jjnr             = nlist->jjnr;
 117     shiftidx         = nlist->shift;
 118     gid              = nlist->gid;
 119     shiftvec         = fr->shift_vec[0];
 120     fshift           = fr->fshift[0];
 121     facel            = _mm_set1_ps(fr->epsfac);
 122     charge           = mdatoms->chargeA;
 123     krf              = _mm_set1_ps(fr->ic->k_rf);
 124     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 125     crf              = _mm_set1_ps(fr->ic->c_rf);
 126     nvdwtype         = fr->ntype;
 127     vdwparam         = fr->nbfp;
 128     vdwtype          = mdatoms->typeA;
 129
 130     vftab            = kernel_data->table_vdw->data;
 131     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 132
 133     /* Setup water-specific parameters */
 134     inr              = nlist->iinr[0];
 135     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 136     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 137     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 138     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 139
 140     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 141     rcutoff_scalar   = fr->rcoulomb;
 142     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 143     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 144
 145     /* Avoid stupid compiler warnings */
 146     jnrA = jnrB = jnrC = jnrD = 0;
 147     j_coord_offsetA = 0;
 148     j_coord_offsetB = 0;
 149     j_coord_offsetC = 0;
 150     j_coord_offsetD = 0;
 151
 152     outeriter        = 0;
 153     inneriter        = 0;
 154
 155     for(iidx=0;iidx<4*DIM;iidx++)
 156     {
 157         scratch[iidx] = 0.0;
 158     }
 159
 160     /* Start outer loop over neighborlists */
 161     for(iidx=0; iidx<nri; iidx++)
 162     {
 163         /* Load shift vector for this list */
 164         i_shift_offset   = DIM*shiftidx[iidx];
 165
 166         /* Load limits for loop over neighbors */
 167         j_index_start    = jindex[iidx];
 168         j_index_end      = jindex[iidx+1];
 169
 170         /* Get outer coordinate index */
 171         inr              = iinr[iidx];
 172         i_coord_offset   = DIM*inr;
 173
 174         /* Load i particle coords and add shift vector */
 175         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 176                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 177
 178         fix0             = _mm_setzero_ps();
 179         fiy0             = _mm_setzero_ps();
 180         fiz0             = _mm_setzero_ps();
 181         fix1             = _mm_setzero_ps();
 182         fiy1             = _mm_setzero_ps();
 183         fiz1             = _mm_setzero_ps();
 184         fix2             = _mm_setzero_ps();
 185         fiy2             = _mm_setzero_ps();
 186         fiz2             = _mm_setzero_ps();
 187
 188         /* Reset potential sums */
 189         velecsum         = _mm_setzero_ps();
 190         vvdwsum          = _mm_setzero_ps();
 191
 192         /* Start inner kernel loop */
 193         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 194         {
 195
 196             /* Get j neighbor index, and coordinate index */
 197             jnrA             = jjnr[jidx];
 198             jnrB             = jjnr[jidx+1];
 199             jnrC             = jjnr[jidx+2];
 200             jnrD             = jjnr[jidx+3];
 201             j_coord_offsetA  = DIM*jnrA;
 202             j_coord_offsetB  = DIM*jnrB;
 203             j_coord_offsetC  = DIM*jnrC;
 204             j_coord_offsetD  = DIM*jnrD;
 205
 206             /* load j atom coordinates */
 207             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 208                                               x+j_coord_offsetC,x+j_coord_offsetD,
 209                                               &jx0,&jy0,&jz0);
 210
 211             /* Calculate displacement vector */
 212             dx00             = _mm_sub_ps(ix0,jx0);
 213             dy00             = _mm_sub_ps(iy0,jy0);
 214             dz00             = _mm_sub_ps(iz0,jz0);
 215             dx10             = _mm_sub_ps(ix1,jx0);
 216             dy10             = _mm_sub_ps(iy1,jy0);
 217             dz10             = _mm_sub_ps(iz1,jz0);
 218             dx20             = _mm_sub_ps(ix2,jx0);
 219             dy20             = _mm_sub_ps(iy2,jy0);
 220             dz20             = _mm_sub_ps(iz2,jz0);
 221
 222             /* Calculate squared distance and things based on it */
 223             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 224             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 225             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 226
 227             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 228             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 229             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 230
 231             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 232             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 233             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 234
 235             /* Load parameters for j particles */
 236             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 237                                                               charge+jnrC+0,charge+jnrD+0);
 238             vdwjidx0A        = 2*vdwtype[jnrA+0];
 239             vdwjidx0B        = 2*vdwtype[jnrB+0];
 240             vdwjidx0C        = 2*vdwtype[jnrC+0];
 241             vdwjidx0D        = 2*vdwtype[jnrD+0];
 242
 243             fjx0             = _mm_setzero_ps();
 244             fjy0             = _mm_setzero_ps();
 245             fjz0             = _mm_setzero_ps();
 246
 247             /**************************
 248              * CALCULATE INTERACTIONS *
 249              **************************/
 250
 251             if (gmx_mm_any_lt(rsq00,rcutoff2))
 252             {
 253
 254             r00              = _mm_mul_ps(rsq00,rinv00);
 255
 256             /* Compute parameters for interactions between i and j atoms */
 257             qq00             = _mm_mul_ps(iq0,jq0);
 258             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 259                                          vdwparam+vdwioffset0+vdwjidx0B,
 260                                          vdwparam+vdwioffset0+vdwjidx0C,
 261                                          vdwparam+vdwioffset0+vdwjidx0D,
 262                                          &c6_00,&c12_00);
 263
 264             /* Calculate table index by multiplying r with table scale and truncate to integer */
 265             rt               = _mm_mul_ps(r00,vftabscale);
 266             vfitab           = _mm_cvttps_epi32(rt);
 267             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 268             vfitab           = _mm_slli_epi32(vfitab,3);
 269
 270             /* REACTION-FIELD ELECTROSTATICS */
 271             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 272             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 273
 274             /* CUBIC SPLINE TABLE DISPERSION */
 275             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 276             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 277             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 278             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 279             _MM_TRANSPOSE4_PS(Y,F,G,H);
 280             Heps             = _mm_mul_ps(vfeps,H);
 281             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 282             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 283             vvdw6            = _mm_mul_ps(c6_00,VV);
 284             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 285             fvdw6            = _mm_mul_ps(c6_00,FF);
 286
 287             /* CUBIC SPLINE TABLE REPULSION */
 288             vfitab           = _mm_add_epi32(vfitab,ifour);
 289             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 290             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 291             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 292             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 293             _MM_TRANSPOSE4_PS(Y,F,G,H);
 294             Heps             = _mm_mul_ps(vfeps,H);
 295             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 296             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 297             vvdw12           = _mm_mul_ps(c12_00,VV);
 298             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 299             fvdw12           = _mm_mul_ps(c12_00,FF);
 300             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 301             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 302
 303             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 304
 305             /* Update potential sum for this i atom from the interaction with this j atom. */
 306             velec            = _mm_and_ps(velec,cutoff_mask);
 307             velecsum         = _mm_add_ps(velecsum,velec);
 308             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 309             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 310
 311             fscal            = _mm_add_ps(felec,fvdw);
 312
 313             fscal            = _mm_and_ps(fscal,cutoff_mask);
 314
 315             /* Calculate temporary vectorial force */
 316             tx               = _mm_mul_ps(fscal,dx00);
 317             ty               = _mm_mul_ps(fscal,dy00);
 318             tz               = _mm_mul_ps(fscal,dz00);
 319
 320             /* Update vectorial force */
 321             fix0             = _mm_add_ps(fix0,tx);
 322             fiy0             = _mm_add_ps(fiy0,ty);
 323             fiz0             = _mm_add_ps(fiz0,tz);
 324
 325             fjx0             = _mm_add_ps(fjx0,tx);
 326             fjy0             = _mm_add_ps(fjy0,ty);
 327             fjz0             = _mm_add_ps(fjz0,tz);
 328
 329             }
 330
 331             /**************************
 332              * CALCULATE INTERACTIONS *
 333              **************************/
 334
 335             if (gmx_mm_any_lt(rsq10,rcutoff2))
 336             {
 337
 338             /* Compute parameters for interactions between i and j atoms */
 339             qq10             = _mm_mul_ps(iq1,jq0);
 340
 341             /* REACTION-FIELD ELECTROSTATICS */
 342             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 343             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 344
 345             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 346
 347             /* Update potential sum for this i atom from the interaction with this j atom. */
 348             velec            = _mm_and_ps(velec,cutoff_mask);
 349             velecsum         = _mm_add_ps(velecsum,velec);
 350
 351             fscal            = felec;
 352
 353             fscal            = _mm_and_ps(fscal,cutoff_mask);
 354
 355             /* Calculate temporary vectorial force */
 356             tx               = _mm_mul_ps(fscal,dx10);
 357             ty               = _mm_mul_ps(fscal,dy10);
 358             tz               = _mm_mul_ps(fscal,dz10);
 359
 360             /* Update vectorial force */
 361             fix1             = _mm_add_ps(fix1,tx);
 362             fiy1             = _mm_add_ps(fiy1,ty);
 363             fiz1             = _mm_add_ps(fiz1,tz);
 364
 365             fjx0             = _mm_add_ps(fjx0,tx);
 366             fjy0             = _mm_add_ps(fjy0,ty);
 367             fjz0             = _mm_add_ps(fjz0,tz);
 368
 369             }
 370
 371             /**************************
 372              * CALCULATE INTERACTIONS *
 373              **************************/
 374
 375             if (gmx_mm_any_lt(rsq20,rcutoff2))
 376             {
 377
 378             /* Compute parameters for interactions between i and j atoms */
 379             qq20             = _mm_mul_ps(iq2,jq0);
 380
 381             /* REACTION-FIELD ELECTROSTATICS */
 382             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 383             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 384
 385             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 386
 387             /* Update potential sum for this i atom from the interaction with this j atom. */
 388             velec            = _mm_and_ps(velec,cutoff_mask);
 389             velecsum         = _mm_add_ps(velecsum,velec);
 390
 391             fscal            = felec;
 392
 393             fscal            = _mm_and_ps(fscal,cutoff_mask);
 394
 395             /* Calculate temporary vectorial force */
 396             tx               = _mm_mul_ps(fscal,dx20);
 397             ty               = _mm_mul_ps(fscal,dy20);
 398             tz               = _mm_mul_ps(fscal,dz20);
 399
 400             /* Update vectorial force */
 401             fix2             = _mm_add_ps(fix2,tx);
 402             fiy2             = _mm_add_ps(fiy2,ty);
 403             fiz2             = _mm_add_ps(fiz2,tz);
 404
 405             fjx0             = _mm_add_ps(fjx0,tx);
 406             fjy0             = _mm_add_ps(fjy0,ty);
 407             fjz0             = _mm_add_ps(fjz0,tz);
 408
 409             }
 410
 411             fjptrA             = f+j_coord_offsetA;
 412             fjptrB             = f+j_coord_offsetB;
 413             fjptrC             = f+j_coord_offsetC;
 414             fjptrD             = f+j_coord_offsetD;
 415
 416             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 417
 418             /* Inner loop uses 144 flops */
 419         }
 420
 421         if(jidx<j_index_end)
 422         {
 423
 424             /* Get j neighbor index, and coordinate index */
 425             jnrlistA         = jjnr[jidx];
 426             jnrlistB         = jjnr[jidx+1];
 427             jnrlistC         = jjnr[jidx+2];
 428             jnrlistD         = jjnr[jidx+3];
 429             /* Sign of each element will be negative for non-real atoms.
 430              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 431              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 432              */
 433             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 434             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 435             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 436             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 437             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 438             j_coord_offsetA  = DIM*jnrA;
 439             j_coord_offsetB  = DIM*jnrB;
 440             j_coord_offsetC  = DIM*jnrC;
 441             j_coord_offsetD  = DIM*jnrD;
 442
 443             /* load j atom coordinates */
 444             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 445                                               x+j_coord_offsetC,x+j_coord_offsetD,
 446                                               &jx0,&jy0,&jz0);
 447
 448             /* Calculate displacement vector */
 449             dx00             = _mm_sub_ps(ix0,jx0);
 450             dy00             = _mm_sub_ps(iy0,jy0);
 451             dz00             = _mm_sub_ps(iz0,jz0);
 452             dx10             = _mm_sub_ps(ix1,jx0);
 453             dy10             = _mm_sub_ps(iy1,jy0);
 454             dz10             = _mm_sub_ps(iz1,jz0);
 455             dx20             = _mm_sub_ps(ix2,jx0);
 456             dy20             = _mm_sub_ps(iy2,jy0);
 457             dz20             = _mm_sub_ps(iz2,jz0);
 458
 459             /* Calculate squared distance and things based on it */
 460             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 461             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 462             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 463
 464             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 465             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 466             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 467
 468             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 469             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 470             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 471
 472             /* Load parameters for j particles */
 473             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 474                                                               charge+jnrC+0,charge+jnrD+0);
 475             vdwjidx0A        = 2*vdwtype[jnrA+0];
 476             vdwjidx0B        = 2*vdwtype[jnrB+0];
 477             vdwjidx0C        = 2*vdwtype[jnrC+0];
 478             vdwjidx0D        = 2*vdwtype[jnrD+0];
 479
 480             fjx0             = _mm_setzero_ps();
 481             fjy0             = _mm_setzero_ps();
 482             fjz0             = _mm_setzero_ps();
 483
 484             /**************************
 485              * CALCULATE INTERACTIONS *
 486              **************************/
 487
 488             if (gmx_mm_any_lt(rsq00,rcutoff2))
 489             {
 490
 491             r00              = _mm_mul_ps(rsq00,rinv00);
 492             r00              = _mm_andnot_ps(dummy_mask,r00);
 493
 494             /* Compute parameters for interactions between i and j atoms */
 495             qq00             = _mm_mul_ps(iq0,jq0);
 496             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 497                                          vdwparam+vdwioffset0+vdwjidx0B,
 498                                          vdwparam+vdwioffset0+vdwjidx0C,
 499                                          vdwparam+vdwioffset0+vdwjidx0D,
 500                                          &c6_00,&c12_00);
 501
 502             /* Calculate table index by multiplying r with table scale and truncate to integer */
 503             rt               = _mm_mul_ps(r00,vftabscale);
 504             vfitab           = _mm_cvttps_epi32(rt);
 505             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 506             vfitab           = _mm_slli_epi32(vfitab,3);
 507
 508             /* REACTION-FIELD ELECTROSTATICS */
 509             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 510             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 511
 512             /* CUBIC SPLINE TABLE DISPERSION */
 513             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 514             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 515             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 516             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 517             _MM_TRANSPOSE4_PS(Y,F,G,H);
 518             Heps             = _mm_mul_ps(vfeps,H);
 519             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 520             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 521             vvdw6            = _mm_mul_ps(c6_00,VV);
 522             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 523             fvdw6            = _mm_mul_ps(c6_00,FF);
 524
 525             /* CUBIC SPLINE TABLE REPULSION */
 526             vfitab           = _mm_add_epi32(vfitab,ifour);
 527             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 528             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 529             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 530             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 531             _MM_TRANSPOSE4_PS(Y,F,G,H);
 532             Heps             = _mm_mul_ps(vfeps,H);
 533             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 534             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 535             vvdw12           = _mm_mul_ps(c12_00,VV);
 536             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 537             fvdw12           = _mm_mul_ps(c12_00,FF);
 538             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 539             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 540
 541             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 542
 543             /* Update potential sum for this i atom from the interaction with this j atom. */
 544             velec            = _mm_and_ps(velec,cutoff_mask);
 545             velec            = _mm_andnot_ps(dummy_mask,velec);
 546             velecsum         = _mm_add_ps(velecsum,velec);
 547             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 548             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 549             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 550
 551             fscal            = _mm_add_ps(felec,fvdw);
 552
 553             fscal            = _mm_and_ps(fscal,cutoff_mask);
 554
 555             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 556
 557             /* Calculate temporary vectorial force */
 558             tx               = _mm_mul_ps(fscal,dx00);
 559             ty               = _mm_mul_ps(fscal,dy00);
 560             tz               = _mm_mul_ps(fscal,dz00);
 561
 562             /* Update vectorial force */
 563             fix0             = _mm_add_ps(fix0,tx);
 564             fiy0             = _mm_add_ps(fiy0,ty);
 565             fiz0             = _mm_add_ps(fiz0,tz);
 566
 567             fjx0             = _mm_add_ps(fjx0,tx);
 568             fjy0             = _mm_add_ps(fjy0,ty);
 569             fjz0             = _mm_add_ps(fjz0,tz);
 570
 571             }
 572
 573             /**************************
 574              * CALCULATE INTERACTIONS *
 575              **************************/
 576
 577             if (gmx_mm_any_lt(rsq10,rcutoff2))
 578             {
 579
 580             /* Compute parameters for interactions between i and j atoms */
 581             qq10             = _mm_mul_ps(iq1,jq0);
 582
 583             /* REACTION-FIELD ELECTROSTATICS */
 584             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 585             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 586
 587             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 588
 589             /* Update potential sum for this i atom from the interaction with this j atom. */
 590             velec            = _mm_and_ps(velec,cutoff_mask);
 591             velec            = _mm_andnot_ps(dummy_mask,velec);
 592             velecsum         = _mm_add_ps(velecsum,velec);
 593
 594             fscal            = felec;
 595
 596             fscal            = _mm_and_ps(fscal,cutoff_mask);
 597
 598             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 599
 600             /* Calculate temporary vectorial force */
 601             tx               = _mm_mul_ps(fscal,dx10);
 602             ty               = _mm_mul_ps(fscal,dy10);
 603             tz               = _mm_mul_ps(fscal,dz10);
 604
 605             /* Update vectorial force */
 606             fix1             = _mm_add_ps(fix1,tx);
 607             fiy1             = _mm_add_ps(fiy1,ty);
 608             fiz1             = _mm_add_ps(fiz1,tz);
 609
 610             fjx0             = _mm_add_ps(fjx0,tx);
 611             fjy0             = _mm_add_ps(fjy0,ty);
 612             fjz0             = _mm_add_ps(fjz0,tz);
 613
 614             }
 615
 616             /**************************
 617              * CALCULATE INTERACTIONS *
 618              **************************/
 619
 620             if (gmx_mm_any_lt(rsq20,rcutoff2))
 621             {
 622
 623             /* Compute parameters for interactions between i and j atoms */
 624             qq20             = _mm_mul_ps(iq2,jq0);
 625
 626             /* REACTION-FIELD ELECTROSTATICS */
 627             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 628             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 629
 630             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 631
 632             /* Update potential sum for this i atom from the interaction with this j atom. */
 633             velec            = _mm_and_ps(velec,cutoff_mask);
 634             velec            = _mm_andnot_ps(dummy_mask,velec);
 635             velecsum         = _mm_add_ps(velecsum,velec);
 636
 637             fscal            = felec;
 638
 639             fscal            = _mm_and_ps(fscal,cutoff_mask);
 640
 641             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 642
 643             /* Calculate temporary vectorial force */
 644             tx               = _mm_mul_ps(fscal,dx20);
 645             ty               = _mm_mul_ps(fscal,dy20);
 646             tz               = _mm_mul_ps(fscal,dz20);
 647
 648             /* Update vectorial force */
 649             fix2             = _mm_add_ps(fix2,tx);
 650             fiy2             = _mm_add_ps(fiy2,ty);
 651             fiz2             = _mm_add_ps(fiz2,tz);
 652
 653             fjx0             = _mm_add_ps(fjx0,tx);
 654             fjy0             = _mm_add_ps(fjy0,ty);
 655             fjz0             = _mm_add_ps(fjz0,tz);
 656
 657             }
 658
 659             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 660             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 661             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 662             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 663
 664             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 665
 666             /* Inner loop uses 145 flops */
 667         }
 668
 669         /* End of innermost loop */
 670
 671         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 672                                               f+i_coord_offset,fshift+i_shift_offset);
 673
 674         ggid                        = gid[iidx];
 675         /* Update potential energies */
 676         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 677         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 678
 679         /* Increment number of inner iterations */
 680         inneriter                  += j_index_end - j_index_start;
 681
 682         /* Outer loop uses 20 flops */
 683     }
 684
 685     /* Increment number of outer iterations */
 686     outeriter        += nri;
 687
 688     /* Update outer/inner flops */
 689
 690     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*145);
 691 }
 692 /*
 693  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_sse4_1_single
 694  * Electrostatics interaction: ReactionField
 695  * VdW interaction:            CubicSplineTable
 696  * Geometry:                   Water3-Particle
 697  * Calculate force/pot:        Force
 698  */
 699 void
 700 nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_sse4_1_single
 701                     (t_nblist                    * gmx_restrict       nlist,
 702                      rvec                        * gmx_restrict          xx,
 703                      rvec                        * gmx_restrict          ff,
 704                      t_forcerec                  * gmx_restrict          fr,
 705                      t_mdatoms                   * gmx_restrict     mdatoms,
 706                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 707                      t_nrnb                      * gmx_restrict        nrnb)
 708 {
 709     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 710      * just 0 for non-waters.
 711      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 712      * jnr indices corresponding to data put in the four positions in the SIMD register.
 713      */
 714     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 715     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 716     int              jnrA,jnrB,jnrC,jnrD;
 717     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 718     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 719     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 720     real             rcutoff_scalar;
 721     real             *shiftvec,*fshift,*x,*f;
 722     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 723     real             scratch[4*DIM];
 724     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 725     int              vdwioffset0;
 726     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 727     int              vdwioffset1;
 728     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 729     int              vdwioffset2;
 730     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 731     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 732     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 733     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 734     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 735     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 736     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 737     real             *charge;
 738     int              nvdwtype;
 739     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 740     int              *vdwtype;
 741     real             *vdwparam;
 742     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 743     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 744     __m128i          vfitab;
 745     __m128i          ifour       = _mm_set1_epi32(4);
 746     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 747     real             *vftab;
 748     __m128           dummy_mask,cutoff_mask;
 749     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 750     __m128           one     = _mm_set1_ps(1.0);
 751     __m128           two     = _mm_set1_ps(2.0);
 752     x                = xx[0];
 753     f                = ff[0];
 754
 755     nri              = nlist->nri;
 756     iinr             = nlist->iinr;
 757     jindex           = nlist->jindex;
 758     jjnr             = nlist->jjnr;
 759     shiftidx         = nlist->shift;
 760     gid              = nlist->gid;
 761     shiftvec         = fr->shift_vec[0];
 762     fshift           = fr->fshift[0];
 763     facel            = _mm_set1_ps(fr->epsfac);
 764     charge           = mdatoms->chargeA;
 765     krf              = _mm_set1_ps(fr->ic->k_rf);
 766     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 767     crf              = _mm_set1_ps(fr->ic->c_rf);
 768     nvdwtype         = fr->ntype;
 769     vdwparam         = fr->nbfp;
 770     vdwtype          = mdatoms->typeA;
 771
 772     vftab            = kernel_data->table_vdw->data;
 773     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 774
 775     /* Setup water-specific parameters */
 776     inr              = nlist->iinr[0];
 777     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 778     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 779     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 780     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 781
 782     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 783     rcutoff_scalar   = fr->rcoulomb;
 784     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 785     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 786
 787     /* Avoid stupid compiler warnings */
 788     jnrA = jnrB = jnrC = jnrD = 0;
 789     j_coord_offsetA = 0;
 790     j_coord_offsetB = 0;
 791     j_coord_offsetC = 0;
 792     j_coord_offsetD = 0;
 793
 794     outeriter        = 0;
 795     inneriter        = 0;
 796
 797     for(iidx=0;iidx<4*DIM;iidx++)
 798     {
 799         scratch[iidx] = 0.0;
 800     }
 801
 802     /* Start outer loop over neighborlists */
 803     for(iidx=0; iidx<nri; iidx++)
 804     {
 805         /* Load shift vector for this list */
 806         i_shift_offset   = DIM*shiftidx[iidx];
 807
 808         /* Load limits for loop over neighbors */
 809         j_index_start    = jindex[iidx];
 810         j_index_end      = jindex[iidx+1];
 811
 812         /* Get outer coordinate index */
 813         inr              = iinr[iidx];
 814         i_coord_offset   = DIM*inr;
 815
 816         /* Load i particle coords and add shift vector */
 817         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 818                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 819
 820         fix0             = _mm_setzero_ps();
 821         fiy0             = _mm_setzero_ps();
 822         fiz0             = _mm_setzero_ps();
 823         fix1             = _mm_setzero_ps();
 824         fiy1             = _mm_setzero_ps();
 825         fiz1             = _mm_setzero_ps();
 826         fix2             = _mm_setzero_ps();
 827         fiy2             = _mm_setzero_ps();
 828         fiz2             = _mm_setzero_ps();
 829
 830         /* Start inner kernel loop */
 831         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 832         {
 833
 834             /* Get j neighbor index, and coordinate index */
 835             jnrA             = jjnr[jidx];
 836             jnrB             = jjnr[jidx+1];
 837             jnrC             = jjnr[jidx+2];
 838             jnrD             = jjnr[jidx+3];
 839             j_coord_offsetA  = DIM*jnrA;
 840             j_coord_offsetB  = DIM*jnrB;
 841             j_coord_offsetC  = DIM*jnrC;
 842             j_coord_offsetD  = DIM*jnrD;
 843
 844             /* load j atom coordinates */
 845             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 846                                               x+j_coord_offsetC,x+j_coord_offsetD,
 847                                               &jx0,&jy0,&jz0);
 848
 849             /* Calculate displacement vector */
 850             dx00             = _mm_sub_ps(ix0,jx0);
 851             dy00             = _mm_sub_ps(iy0,jy0);
 852             dz00             = _mm_sub_ps(iz0,jz0);
 853             dx10             = _mm_sub_ps(ix1,jx0);
 854             dy10             = _mm_sub_ps(iy1,jy0);
 855             dz10             = _mm_sub_ps(iz1,jz0);
 856             dx20             = _mm_sub_ps(ix2,jx0);
 857             dy20             = _mm_sub_ps(iy2,jy0);
 858             dz20             = _mm_sub_ps(iz2,jz0);
 859
 860             /* Calculate squared distance and things based on it */
 861             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 862             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 863             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 864
 865             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 866             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 867             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 868
 869             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 870             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 871             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 872
 873             /* Load parameters for j particles */
 874             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 875                                                               charge+jnrC+0,charge+jnrD+0);
 876             vdwjidx0A        = 2*vdwtype[jnrA+0];
 877             vdwjidx0B        = 2*vdwtype[jnrB+0];
 878             vdwjidx0C        = 2*vdwtype[jnrC+0];
 879             vdwjidx0D        = 2*vdwtype[jnrD+0];
 880
 881             fjx0             = _mm_setzero_ps();
 882             fjy0             = _mm_setzero_ps();
 883             fjz0             = _mm_setzero_ps();
 884
 885             /**************************
 886              * CALCULATE INTERACTIONS *
 887              **************************/
 888
 889             if (gmx_mm_any_lt(rsq00,rcutoff2))
 890             {
 891
 892             r00              = _mm_mul_ps(rsq00,rinv00);
 893
 894             /* Compute parameters for interactions between i and j atoms */
 895             qq00             = _mm_mul_ps(iq0,jq0);
 896             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 897                                          vdwparam+vdwioffset0+vdwjidx0B,
 898                                          vdwparam+vdwioffset0+vdwjidx0C,
 899                                          vdwparam+vdwioffset0+vdwjidx0D,
 900                                          &c6_00,&c12_00);
 901
 902             /* Calculate table index by multiplying r with table scale and truncate to integer */
 903             rt               = _mm_mul_ps(r00,vftabscale);
 904             vfitab           = _mm_cvttps_epi32(rt);
 905             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 906             vfitab           = _mm_slli_epi32(vfitab,3);
 907
 908             /* REACTION-FIELD ELECTROSTATICS */
 909             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 910
 911             /* CUBIC SPLINE TABLE DISPERSION */
 912             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 913             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 914             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 915             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 916             _MM_TRANSPOSE4_PS(Y,F,G,H);
 917             Heps             = _mm_mul_ps(vfeps,H);
 918             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 919             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 920             fvdw6            = _mm_mul_ps(c6_00,FF);
 921
 922             /* CUBIC SPLINE TABLE REPULSION */
 923             vfitab           = _mm_add_epi32(vfitab,ifour);
 924             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 925             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 926             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 927             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 928             _MM_TRANSPOSE4_PS(Y,F,G,H);
 929             Heps             = _mm_mul_ps(vfeps,H);
 930             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 931             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 932             fvdw12           = _mm_mul_ps(c12_00,FF);
 933             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 934
 935             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 936
 937             fscal            = _mm_add_ps(felec,fvdw);
 938
 939             fscal            = _mm_and_ps(fscal,cutoff_mask);
 940
 941             /* Calculate temporary vectorial force */
 942             tx               = _mm_mul_ps(fscal,dx00);
 943             ty               = _mm_mul_ps(fscal,dy00);
 944             tz               = _mm_mul_ps(fscal,dz00);
 945
 946             /* Update vectorial force */
 947             fix0             = _mm_add_ps(fix0,tx);
 948             fiy0             = _mm_add_ps(fiy0,ty);
 949             fiz0             = _mm_add_ps(fiz0,tz);
 950
 951             fjx0             = _mm_add_ps(fjx0,tx);
 952             fjy0             = _mm_add_ps(fjy0,ty);
 953             fjz0             = _mm_add_ps(fjz0,tz);
 954
 955             }
 956
 957             /**************************
 958              * CALCULATE INTERACTIONS *
 959              **************************/
 960
 961             if (gmx_mm_any_lt(rsq10,rcutoff2))
 962             {
 963
 964             /* Compute parameters for interactions between i and j atoms */
 965             qq10             = _mm_mul_ps(iq1,jq0);
 966
 967             /* REACTION-FIELD ELECTROSTATICS */
 968             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 969
 970             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 971
 972             fscal            = felec;
 973
 974             fscal            = _mm_and_ps(fscal,cutoff_mask);
 975
 976             /* Calculate temporary vectorial force */
 977             tx               = _mm_mul_ps(fscal,dx10);
 978             ty               = _mm_mul_ps(fscal,dy10);
 979             tz               = _mm_mul_ps(fscal,dz10);
 980
 981             /* Update vectorial force */
 982             fix1             = _mm_add_ps(fix1,tx);
 983             fiy1             = _mm_add_ps(fiy1,ty);
 984             fiz1             = _mm_add_ps(fiz1,tz);
 985
 986             fjx0             = _mm_add_ps(fjx0,tx);
 987             fjy0             = _mm_add_ps(fjy0,ty);
 988             fjz0             = _mm_add_ps(fjz0,tz);
 989
 990             }
 991
 992             /**************************
 993              * CALCULATE INTERACTIONS *
 994              **************************/
 995
 996             if (gmx_mm_any_lt(rsq20,rcutoff2))
 997             {
 998
 999             /* Compute parameters for interactions between i and j atoms */
1000             qq20             = _mm_mul_ps(iq2,jq0);
1001
1002             /* REACTION-FIELD ELECTROSTATICS */
1003             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1004
1005             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1006
1007             fscal            = felec;
1008
1009             fscal            = _mm_and_ps(fscal,cutoff_mask);
1010
1011             /* Calculate temporary vectorial force */
1012             tx               = _mm_mul_ps(fscal,dx20);
1013             ty               = _mm_mul_ps(fscal,dy20);
1014             tz               = _mm_mul_ps(fscal,dz20);
1015
1016             /* Update vectorial force */
1017             fix2             = _mm_add_ps(fix2,tx);
1018             fiy2             = _mm_add_ps(fiy2,ty);
1019             fiz2             = _mm_add_ps(fiz2,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
1027             fjptrA             = f+j_coord_offsetA;
1028             fjptrB             = f+j_coord_offsetB;
1029             fjptrC             = f+j_coord_offsetC;
1030             fjptrD             = f+j_coord_offsetD;
1031
1032             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1033
1034             /* Inner loop uses 117 flops */
1035         }
1036
1037         if(jidx<j_index_end)
1038         {
1039
1040             /* Get j neighbor index, and coordinate index */
1041             jnrlistA         = jjnr[jidx];
1042             jnrlistB         = jjnr[jidx+1];
1043             jnrlistC         = jjnr[jidx+2];
1044             jnrlistD         = jjnr[jidx+3];
1045             /* Sign of each element will be negative for non-real atoms.
1046              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1047              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1048              */
1049             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1050             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1051             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1052             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1053             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1054             j_coord_offsetA  = DIM*jnrA;
1055             j_coord_offsetB  = DIM*jnrB;
1056             j_coord_offsetC  = DIM*jnrC;
1057             j_coord_offsetD  = DIM*jnrD;
1058
1059             /* load j atom coordinates */
1060             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1061                                               x+j_coord_offsetC,x+j_coord_offsetD,
1062                                               &jx0,&jy0,&jz0);
1063
1064             /* Calculate displacement vector */
1065             dx00             = _mm_sub_ps(ix0,jx0);
1066             dy00             = _mm_sub_ps(iy0,jy0);
1067             dz00             = _mm_sub_ps(iz0,jz0);
1068             dx10             = _mm_sub_ps(ix1,jx0);
1069             dy10             = _mm_sub_ps(iy1,jy0);
1070             dz10             = _mm_sub_ps(iz1,jz0);
1071             dx20             = _mm_sub_ps(ix2,jx0);
1072             dy20             = _mm_sub_ps(iy2,jy0);
1073             dz20             = _mm_sub_ps(iz2,jz0);
1074
1075             /* Calculate squared distance and things based on it */
1076             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1077             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1078             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1079
1080             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1081             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1082             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1083
1084             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1085             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1086             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1087
1088             /* Load parameters for j particles */
1089             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1090                                                               charge+jnrC+0,charge+jnrD+0);
1091             vdwjidx0A        = 2*vdwtype[jnrA+0];
1092             vdwjidx0B        = 2*vdwtype[jnrB+0];
1093             vdwjidx0C        = 2*vdwtype[jnrC+0];
1094             vdwjidx0D        = 2*vdwtype[jnrD+0];
1095
1096             fjx0             = _mm_setzero_ps();
1097             fjy0             = _mm_setzero_ps();
1098             fjz0             = _mm_setzero_ps();
1099
1100             /**************************
1101              * CALCULATE INTERACTIONS *
1102              **************************/
1103
1104             if (gmx_mm_any_lt(rsq00,rcutoff2))
1105             {
1106
1107             r00              = _mm_mul_ps(rsq00,rinv00);
1108             r00              = _mm_andnot_ps(dummy_mask,r00);
1109
1110             /* Compute parameters for interactions between i and j atoms */
1111             qq00             = _mm_mul_ps(iq0,jq0);
1112             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1113                                          vdwparam+vdwioffset0+vdwjidx0B,
1114                                          vdwparam+vdwioffset0+vdwjidx0C,
1115                                          vdwparam+vdwioffset0+vdwjidx0D,
1116                                          &c6_00,&c12_00);
1117
1118             /* Calculate table index by multiplying r with table scale and truncate to integer */
1119             rt               = _mm_mul_ps(r00,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(vfitab,3);
1123
1124             /* REACTION-FIELD ELECTROSTATICS */
1125             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
1126
1127             /* CUBIC SPLINE TABLE DISPERSION */
1128             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1129             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1130             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1131             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1132             _MM_TRANSPOSE4_PS(Y,F,G,H);
1133             Heps             = _mm_mul_ps(vfeps,H);
1134             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1135             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1136             fvdw6            = _mm_mul_ps(c6_00,FF);
1137
1138             /* CUBIC SPLINE TABLE REPULSION */
1139             vfitab           = _mm_add_epi32(vfitab,ifour);
1140             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1141             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1142             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1143             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1144             _MM_TRANSPOSE4_PS(Y,F,G,H);
1145             Heps             = _mm_mul_ps(vfeps,H);
1146             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1147             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1148             fvdw12           = _mm_mul_ps(c12_00,FF);
1149             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1150
1151             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1152
1153             fscal            = _mm_add_ps(felec,fvdw);
1154
1155             fscal            = _mm_and_ps(fscal,cutoff_mask);
1156
1157             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1158
1159             /* Calculate temporary vectorial force */
1160             tx               = _mm_mul_ps(fscal,dx00);
1161             ty               = _mm_mul_ps(fscal,dy00);
1162             tz               = _mm_mul_ps(fscal,dz00);
1163
1164             /* Update vectorial force */
1165             fix0             = _mm_add_ps(fix0,tx);
1166             fiy0             = _mm_add_ps(fiy0,ty);
1167             fiz0             = _mm_add_ps(fiz0,tz);
1168
1169             fjx0             = _mm_add_ps(fjx0,tx);
1170             fjy0             = _mm_add_ps(fjy0,ty);
1171             fjz0             = _mm_add_ps(fjz0,tz);
1172
1173             }
1174
1175             /**************************
1176              * CALCULATE INTERACTIONS *
1177              **************************/
1178
1179             if (gmx_mm_any_lt(rsq10,rcutoff2))
1180             {
1181
1182             /* Compute parameters for interactions between i and j atoms */
1183             qq10             = _mm_mul_ps(iq1,jq0);
1184
1185             /* REACTION-FIELD ELECTROSTATICS */
1186             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1187
1188             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1189
1190             fscal            = felec;
1191
1192             fscal            = _mm_and_ps(fscal,cutoff_mask);
1193
1194             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1195
1196             /* Calculate temporary vectorial force */
1197             tx               = _mm_mul_ps(fscal,dx10);
1198             ty               = _mm_mul_ps(fscal,dy10);
1199             tz               = _mm_mul_ps(fscal,dz10);
1200
1201             /* Update vectorial force */
1202             fix1             = _mm_add_ps(fix1,tx);
1203             fiy1             = _mm_add_ps(fiy1,ty);
1204             fiz1             = _mm_add_ps(fiz1,tz);
1205
1206             fjx0             = _mm_add_ps(fjx0,tx);
1207             fjy0             = _mm_add_ps(fjy0,ty);
1208             fjz0             = _mm_add_ps(fjz0,tz);
1209
1210             }
1211
1212             /**************************
1213              * CALCULATE INTERACTIONS *
1214              **************************/
1215
1216             if (gmx_mm_any_lt(rsq20,rcutoff2))
1217             {
1218
1219             /* Compute parameters for interactions between i and j atoms */
1220             qq20             = _mm_mul_ps(iq2,jq0);
1221
1222             /* REACTION-FIELD ELECTROSTATICS */
1223             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1224
1225             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1226
1227             fscal            = felec;
1228
1229             fscal            = _mm_and_ps(fscal,cutoff_mask);
1230
1231             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1232
1233             /* Calculate temporary vectorial force */
1234             tx               = _mm_mul_ps(fscal,dx20);
1235             ty               = _mm_mul_ps(fscal,dy20);
1236             tz               = _mm_mul_ps(fscal,dz20);
1237
1238             /* Update vectorial force */
1239             fix2             = _mm_add_ps(fix2,tx);
1240             fiy2             = _mm_add_ps(fiy2,ty);
1241             fiz2             = _mm_add_ps(fiz2,tz);
1242
1243             fjx0             = _mm_add_ps(fjx0,tx);
1244             fjy0             = _mm_add_ps(fjy0,ty);
1245             fjz0             = _mm_add_ps(fjz0,tz);
1246
1247             }
1248
1249             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1250             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1251             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1252             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1253
1254             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1255
1256             /* Inner loop uses 118 flops */
1257         }
1258
1259         /* End of innermost loop */
1260
1261         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1262                                               f+i_coord_offset,fshift+i_shift_offset);
1263
1264         /* Increment number of inner iterations */
1265         inneriter                  += j_index_end - j_index_start;
1266
1267         /* Outer loop uses 18 flops */
1268     }
1269
1270     /* Increment number of outer iterations */
1271     outeriter        += nri;
1272
1273     /* Update outer/inner flops */
1274
1275     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*118);
1276 }