src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRF_VdwCSTab_GeomW4P1_sse2_single.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, 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 sse2_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse2_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW4P1_VF_sse2_single
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRF_VdwCSTab_GeomW4P1_VF_sse2_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwioffset3;
  89     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  90     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  91     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  96     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 103     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 104     __m128i          vfitab;
 105     __m128i          ifour       = _mm_set1_epi32(4);
 106     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 107     real             *vftab;
 108     __m128           dummy_mask,cutoff_mask;
 109     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 110     __m128           one     = _mm_set1_ps(1.0);
 111     __m128           two     = _mm_set1_ps(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm_set1_ps(fr->ic->epsfac);
 124     charge           = mdatoms->chargeA;
 125     krf              = _mm_set1_ps(fr->ic->k_rf);
 126     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 127     crf              = _mm_set1_ps(fr->ic->c_rf);
 128     nvdwtype         = fr->ntype;
 129     vdwparam         = fr->nbfp;
 130     vdwtype          = mdatoms->typeA;
 131
 132     vftab            = kernel_data->table_vdw->data;
 133     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 134
 135     /* Setup water-specific parameters */
 136     inr              = nlist->iinr[0];
 137     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 138     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 139     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 140     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 141
 142     /* Avoid stupid compiler warnings */
 143     jnrA = jnrB = jnrC = jnrD = 0;
 144     j_coord_offsetA = 0;
 145     j_coord_offsetB = 0;
 146     j_coord_offsetC = 0;
 147     j_coord_offsetD = 0;
 148
 149     outeriter        = 0;
 150     inneriter        = 0;
 151
 152     for(iidx=0;iidx<4*DIM;iidx++)
 153     {
 154         scratch[iidx] = 0.0;
 155     }
 156
 157     /* Start outer loop over neighborlists */
 158     for(iidx=0; iidx<nri; iidx++)
 159     {
 160         /* Load shift vector for this list */
 161         i_shift_offset   = DIM*shiftidx[iidx];
 162
 163         /* Load limits for loop over neighbors */
 164         j_index_start    = jindex[iidx];
 165         j_index_end      = jindex[iidx+1];
 166
 167         /* Get outer coordinate index */
 168         inr              = iinr[iidx];
 169         i_coord_offset   = DIM*inr;
 170
 171         /* Load i particle coords and add shift vector */
 172         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 173                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 174
 175         fix0             = _mm_setzero_ps();
 176         fiy0             = _mm_setzero_ps();
 177         fiz0             = _mm_setzero_ps();
 178         fix1             = _mm_setzero_ps();
 179         fiy1             = _mm_setzero_ps();
 180         fiz1             = _mm_setzero_ps();
 181         fix2             = _mm_setzero_ps();
 182         fiy2             = _mm_setzero_ps();
 183         fiz2             = _mm_setzero_ps();
 184         fix3             = _mm_setzero_ps();
 185         fiy3             = _mm_setzero_ps();
 186         fiz3             = _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             dx30             = _mm_sub_ps(ix3,jx0);
 222             dy30             = _mm_sub_ps(iy3,jy0);
 223             dz30             = _mm_sub_ps(iz3,jz0);
 224
 225             /* Calculate squared distance and things based on it */
 226             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 227             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 228             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 229             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 230
 231             rinv00           = sse2_invsqrt_f(rsq00);
 232             rinv10           = sse2_invsqrt_f(rsq10);
 233             rinv20           = sse2_invsqrt_f(rsq20);
 234             rinv30           = sse2_invsqrt_f(rsq30);
 235
 236             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 237             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 238             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 239
 240             /* Load parameters for j particles */
 241             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 242                                                               charge+jnrC+0,charge+jnrD+0);
 243             vdwjidx0A        = 2*vdwtype[jnrA+0];
 244             vdwjidx0B        = 2*vdwtype[jnrB+0];
 245             vdwjidx0C        = 2*vdwtype[jnrC+0];
 246             vdwjidx0D        = 2*vdwtype[jnrD+0];
 247
 248             fjx0             = _mm_setzero_ps();
 249             fjy0             = _mm_setzero_ps();
 250             fjz0             = _mm_setzero_ps();
 251
 252             /**************************
 253              * CALCULATE INTERACTIONS *
 254              **************************/
 255
 256             r00              = _mm_mul_ps(rsq00,rinv00);
 257
 258             /* Compute parameters for interactions between i and j atoms */
 259             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 260                                          vdwparam+vdwioffset0+vdwjidx0B,
 261                                          vdwparam+vdwioffset0+vdwjidx0C,
 262                                          vdwparam+vdwioffset0+vdwjidx0D,
 263                                          &c6_00,&c12_00);
 264
 265             /* Calculate table index by multiplying r with table scale and truncate to integer */
 266             rt               = _mm_mul_ps(r00,vftabscale);
 267             vfitab           = _mm_cvttps_epi32(rt);
 268             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 269             vfitab           = _mm_slli_epi32(vfitab,3);
 270
 271             /* CUBIC SPLINE TABLE DISPERSION */
 272             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 273             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 274             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 275             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 276             _MM_TRANSPOSE4_PS(Y,F,G,H);
 277             Heps             = _mm_mul_ps(vfeps,H);
 278             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 279             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 280             vvdw6            = _mm_mul_ps(c6_00,VV);
 281             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 282             fvdw6            = _mm_mul_ps(c6_00,FF);
 283
 284             /* CUBIC SPLINE TABLE REPULSION */
 285             vfitab           = _mm_add_epi32(vfitab,ifour);
 286             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 287             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 288             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 289             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 290             _MM_TRANSPOSE4_PS(Y,F,G,H);
 291             Heps             = _mm_mul_ps(vfeps,H);
 292             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 293             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 294             vvdw12           = _mm_mul_ps(c12_00,VV);
 295             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 296             fvdw12           = _mm_mul_ps(c12_00,FF);
 297             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 298             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 299
 300             /* Update potential sum for this i atom from the interaction with this j atom. */
 301             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 302
 303             fscal            = fvdw;
 304
 305             /* Calculate temporary vectorial force */
 306             tx               = _mm_mul_ps(fscal,dx00);
 307             ty               = _mm_mul_ps(fscal,dy00);
 308             tz               = _mm_mul_ps(fscal,dz00);
 309
 310             /* Update vectorial force */
 311             fix0             = _mm_add_ps(fix0,tx);
 312             fiy0             = _mm_add_ps(fiy0,ty);
 313             fiz0             = _mm_add_ps(fiz0,tz);
 314
 315             fjx0             = _mm_add_ps(fjx0,tx);
 316             fjy0             = _mm_add_ps(fjy0,ty);
 317             fjz0             = _mm_add_ps(fjz0,tz);
 318
 319             /**************************
 320              * CALCULATE INTERACTIONS *
 321              **************************/
 322
 323             /* Compute parameters for interactions between i and j atoms */
 324             qq10             = _mm_mul_ps(iq1,jq0);
 325
 326             /* REACTION-FIELD ELECTROSTATICS */
 327             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 328             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 329
 330             /* Update potential sum for this i atom from the interaction with this j atom. */
 331             velecsum         = _mm_add_ps(velecsum,velec);
 332
 333             fscal            = felec;
 334
 335             /* Calculate temporary vectorial force */
 336             tx               = _mm_mul_ps(fscal,dx10);
 337             ty               = _mm_mul_ps(fscal,dy10);
 338             tz               = _mm_mul_ps(fscal,dz10);
 339
 340             /* Update vectorial force */
 341             fix1             = _mm_add_ps(fix1,tx);
 342             fiy1             = _mm_add_ps(fiy1,ty);
 343             fiz1             = _mm_add_ps(fiz1,tz);
 344
 345             fjx0             = _mm_add_ps(fjx0,tx);
 346             fjy0             = _mm_add_ps(fjy0,ty);
 347             fjz0             = _mm_add_ps(fjz0,tz);
 348
 349             /**************************
 350              * CALCULATE INTERACTIONS *
 351              **************************/
 352
 353             /* Compute parameters for interactions between i and j atoms */
 354             qq20             = _mm_mul_ps(iq2,jq0);
 355
 356             /* REACTION-FIELD ELECTROSTATICS */
 357             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 358             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 359
 360             /* Update potential sum for this i atom from the interaction with this j atom. */
 361             velecsum         = _mm_add_ps(velecsum,velec);
 362
 363             fscal            = felec;
 364
 365             /* Calculate temporary vectorial force */
 366             tx               = _mm_mul_ps(fscal,dx20);
 367             ty               = _mm_mul_ps(fscal,dy20);
 368             tz               = _mm_mul_ps(fscal,dz20);
 369
 370             /* Update vectorial force */
 371             fix2             = _mm_add_ps(fix2,tx);
 372             fiy2             = _mm_add_ps(fiy2,ty);
 373             fiz2             = _mm_add_ps(fiz2,tz);
 374
 375             fjx0             = _mm_add_ps(fjx0,tx);
 376             fjy0             = _mm_add_ps(fjy0,ty);
 377             fjz0             = _mm_add_ps(fjz0,tz);
 378
 379             /**************************
 380              * CALCULATE INTERACTIONS *
 381              **************************/
 382
 383             /* Compute parameters for interactions between i and j atoms */
 384             qq30             = _mm_mul_ps(iq3,jq0);
 385
 386             /* REACTION-FIELD ELECTROSTATICS */
 387             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 388             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 389
 390             /* Update potential sum for this i atom from the interaction with this j atom. */
 391             velecsum         = _mm_add_ps(velecsum,velec);
 392
 393             fscal            = felec;
 394
 395             /* Calculate temporary vectorial force */
 396             tx               = _mm_mul_ps(fscal,dx30);
 397             ty               = _mm_mul_ps(fscal,dy30);
 398             tz               = _mm_mul_ps(fscal,dz30);
 399
 400             /* Update vectorial force */
 401             fix3             = _mm_add_ps(fix3,tx);
 402             fiy3             = _mm_add_ps(fiy3,ty);
 403             fiz3             = _mm_add_ps(fiz3,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             fjptrA             = f+j_coord_offsetA;
 410             fjptrB             = f+j_coord_offsetB;
 411             fjptrC             = f+j_coord_offsetC;
 412             fjptrD             = f+j_coord_offsetD;
 413
 414             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 415
 416             /* Inner loop uses 152 flops */
 417         }
 418
 419         if(jidx<j_index_end)
 420         {
 421
 422             /* Get j neighbor index, and coordinate index */
 423             jnrlistA         = jjnr[jidx];
 424             jnrlistB         = jjnr[jidx+1];
 425             jnrlistC         = jjnr[jidx+2];
 426             jnrlistD         = jjnr[jidx+3];
 427             /* Sign of each element will be negative for non-real atoms.
 428              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 429              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 430              */
 431             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 432             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 433             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 434             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 435             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 436             j_coord_offsetA  = DIM*jnrA;
 437             j_coord_offsetB  = DIM*jnrB;
 438             j_coord_offsetC  = DIM*jnrC;
 439             j_coord_offsetD  = DIM*jnrD;
 440
 441             /* load j atom coordinates */
 442             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 443                                               x+j_coord_offsetC,x+j_coord_offsetD,
 444                                               &jx0,&jy0,&jz0);
 445
 446             /* Calculate displacement vector */
 447             dx00             = _mm_sub_ps(ix0,jx0);
 448             dy00             = _mm_sub_ps(iy0,jy0);
 449             dz00             = _mm_sub_ps(iz0,jz0);
 450             dx10             = _mm_sub_ps(ix1,jx0);
 451             dy10             = _mm_sub_ps(iy1,jy0);
 452             dz10             = _mm_sub_ps(iz1,jz0);
 453             dx20             = _mm_sub_ps(ix2,jx0);
 454             dy20             = _mm_sub_ps(iy2,jy0);
 455             dz20             = _mm_sub_ps(iz2,jz0);
 456             dx30             = _mm_sub_ps(ix3,jx0);
 457             dy30             = _mm_sub_ps(iy3,jy0);
 458             dz30             = _mm_sub_ps(iz3,jz0);
 459
 460             /* Calculate squared distance and things based on it */
 461             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 462             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 463             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 464             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 465
 466             rinv00           = sse2_invsqrt_f(rsq00);
 467             rinv10           = sse2_invsqrt_f(rsq10);
 468             rinv20           = sse2_invsqrt_f(rsq20);
 469             rinv30           = sse2_invsqrt_f(rsq30);
 470
 471             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 472             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 473             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 474
 475             /* Load parameters for j particles */
 476             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 477                                                               charge+jnrC+0,charge+jnrD+0);
 478             vdwjidx0A        = 2*vdwtype[jnrA+0];
 479             vdwjidx0B        = 2*vdwtype[jnrB+0];
 480             vdwjidx0C        = 2*vdwtype[jnrC+0];
 481             vdwjidx0D        = 2*vdwtype[jnrD+0];
 482
 483             fjx0             = _mm_setzero_ps();
 484             fjy0             = _mm_setzero_ps();
 485             fjz0             = _mm_setzero_ps();
 486
 487             /**************************
 488              * CALCULATE INTERACTIONS *
 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             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 496                                          vdwparam+vdwioffset0+vdwjidx0B,
 497                                          vdwparam+vdwioffset0+vdwjidx0C,
 498                                          vdwparam+vdwioffset0+vdwjidx0D,
 499                                          &c6_00,&c12_00);
 500
 501             /* Calculate table index by multiplying r with table scale and truncate to integer */
 502             rt               = _mm_mul_ps(r00,vftabscale);
 503             vfitab           = _mm_cvttps_epi32(rt);
 504             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 505             vfitab           = _mm_slli_epi32(vfitab,3);
 506
 507             /* CUBIC SPLINE TABLE DISPERSION */
 508             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 509             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 510             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 511             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 512             _MM_TRANSPOSE4_PS(Y,F,G,H);
 513             Heps             = _mm_mul_ps(vfeps,H);
 514             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 515             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 516             vvdw6            = _mm_mul_ps(c6_00,VV);
 517             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 518             fvdw6            = _mm_mul_ps(c6_00,FF);
 519
 520             /* CUBIC SPLINE TABLE REPULSION */
 521             vfitab           = _mm_add_epi32(vfitab,ifour);
 522             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 523             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 524             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 525             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 526             _MM_TRANSPOSE4_PS(Y,F,G,H);
 527             Heps             = _mm_mul_ps(vfeps,H);
 528             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 529             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 530             vvdw12           = _mm_mul_ps(c12_00,VV);
 531             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 532             fvdw12           = _mm_mul_ps(c12_00,FF);
 533             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 534             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 535
 536             /* Update potential sum for this i atom from the interaction with this j atom. */
 537             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 538             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 539
 540             fscal            = fvdw;
 541
 542             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 543
 544             /* Calculate temporary vectorial force */
 545             tx               = _mm_mul_ps(fscal,dx00);
 546             ty               = _mm_mul_ps(fscal,dy00);
 547             tz               = _mm_mul_ps(fscal,dz00);
 548
 549             /* Update vectorial force */
 550             fix0             = _mm_add_ps(fix0,tx);
 551             fiy0             = _mm_add_ps(fiy0,ty);
 552             fiz0             = _mm_add_ps(fiz0,tz);
 553
 554             fjx0             = _mm_add_ps(fjx0,tx);
 555             fjy0             = _mm_add_ps(fjy0,ty);
 556             fjz0             = _mm_add_ps(fjz0,tz);
 557
 558             /**************************
 559              * CALCULATE INTERACTIONS *
 560              **************************/
 561
 562             /* Compute parameters for interactions between i and j atoms */
 563             qq10             = _mm_mul_ps(iq1,jq0);
 564
 565             /* REACTION-FIELD ELECTROSTATICS */
 566             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 567             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 568
 569             /* Update potential sum for this i atom from the interaction with this j atom. */
 570             velec            = _mm_andnot_ps(dummy_mask,velec);
 571             velecsum         = _mm_add_ps(velecsum,velec);
 572
 573             fscal            = felec;
 574
 575             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 576
 577             /* Calculate temporary vectorial force */
 578             tx               = _mm_mul_ps(fscal,dx10);
 579             ty               = _mm_mul_ps(fscal,dy10);
 580             tz               = _mm_mul_ps(fscal,dz10);
 581
 582             /* Update vectorial force */
 583             fix1             = _mm_add_ps(fix1,tx);
 584             fiy1             = _mm_add_ps(fiy1,ty);
 585             fiz1             = _mm_add_ps(fiz1,tz);
 586
 587             fjx0             = _mm_add_ps(fjx0,tx);
 588             fjy0             = _mm_add_ps(fjy0,ty);
 589             fjz0             = _mm_add_ps(fjz0,tz);
 590
 591             /**************************
 592              * CALCULATE INTERACTIONS *
 593              **************************/
 594
 595             /* Compute parameters for interactions between i and j atoms */
 596             qq20             = _mm_mul_ps(iq2,jq0);
 597
 598             /* REACTION-FIELD ELECTROSTATICS */
 599             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 600             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 601
 602             /* Update potential sum for this i atom from the interaction with this j atom. */
 603             velec            = _mm_andnot_ps(dummy_mask,velec);
 604             velecsum         = _mm_add_ps(velecsum,velec);
 605
 606             fscal            = felec;
 607
 608             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 609
 610             /* Calculate temporary vectorial force */
 611             tx               = _mm_mul_ps(fscal,dx20);
 612             ty               = _mm_mul_ps(fscal,dy20);
 613             tz               = _mm_mul_ps(fscal,dz20);
 614
 615             /* Update vectorial force */
 616             fix2             = _mm_add_ps(fix2,tx);
 617             fiy2             = _mm_add_ps(fiy2,ty);
 618             fiz2             = _mm_add_ps(fiz2,tz);
 619
 620             fjx0             = _mm_add_ps(fjx0,tx);
 621             fjy0             = _mm_add_ps(fjy0,ty);
 622             fjz0             = _mm_add_ps(fjz0,tz);
 623
 624             /**************************
 625              * CALCULATE INTERACTIONS *
 626              **************************/
 627
 628             /* Compute parameters for interactions between i and j atoms */
 629             qq30             = _mm_mul_ps(iq3,jq0);
 630
 631             /* REACTION-FIELD ELECTROSTATICS */
 632             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 633             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 634
 635             /* Update potential sum for this i atom from the interaction with this j atom. */
 636             velec            = _mm_andnot_ps(dummy_mask,velec);
 637             velecsum         = _mm_add_ps(velecsum,velec);
 638
 639             fscal            = felec;
 640
 641             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 642
 643             /* Calculate temporary vectorial force */
 644             tx               = _mm_mul_ps(fscal,dx30);
 645             ty               = _mm_mul_ps(fscal,dy30);
 646             tz               = _mm_mul_ps(fscal,dz30);
 647
 648             /* Update vectorial force */
 649             fix3             = _mm_add_ps(fix3,tx);
 650             fiy3             = _mm_add_ps(fiy3,ty);
 651             fiz3             = _mm_add_ps(fiz3,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             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 658             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 659             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 660             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 661
 662             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 663
 664             /* Inner loop uses 153 flops */
 665         }
 666
 667         /* End of innermost loop */
 668
 669         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 670                                               f+i_coord_offset,fshift+i_shift_offset);
 671
 672         ggid                        = gid[iidx];
 673         /* Update potential energies */
 674         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 675         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 676
 677         /* Increment number of inner iterations */
 678         inneriter                  += j_index_end - j_index_start;
 679
 680         /* Outer loop uses 26 flops */
 681     }
 682
 683     /* Increment number of outer iterations */
 684     outeriter        += nri;
 685
 686     /* Update outer/inner flops */
 687
 688     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*153);
 689 }
 690 /*
 691  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_sse2_single
 692  * Electrostatics interaction: ReactionField
 693  * VdW interaction:            CubicSplineTable
 694  * Geometry:                   Water4-Particle
 695  * Calculate force/pot:        Force
 696  */
 697 void
 698 nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_sse2_single
 699                     (t_nblist                    * gmx_restrict       nlist,
 700                      rvec                        * gmx_restrict          xx,
 701                      rvec                        * gmx_restrict          ff,
 702                      struct t_forcerec           * gmx_restrict          fr,
 703                      t_mdatoms                   * gmx_restrict     mdatoms,
 704                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 705                      t_nrnb                      * gmx_restrict        nrnb)
 706 {
 707     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 708      * just 0 for non-waters.
 709      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 710      * jnr indices corresponding to data put in the four positions in the SIMD register.
 711      */
 712     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 713     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 714     int              jnrA,jnrB,jnrC,jnrD;
 715     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 716     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 717     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 718     real             rcutoff_scalar;
 719     real             *shiftvec,*fshift,*x,*f;
 720     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 721     real             scratch[4*DIM];
 722     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 723     int              vdwioffset0;
 724     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 725     int              vdwioffset1;
 726     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 727     int              vdwioffset2;
 728     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 729     int              vdwioffset3;
 730     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 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           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 737     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 738     real             *charge;
 739     int              nvdwtype;
 740     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 741     int              *vdwtype;
 742     real             *vdwparam;
 743     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 744     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 745     __m128i          vfitab;
 746     __m128i          ifour       = _mm_set1_epi32(4);
 747     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 748     real             *vftab;
 749     __m128           dummy_mask,cutoff_mask;
 750     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 751     __m128           one     = _mm_set1_ps(1.0);
 752     __m128           two     = _mm_set1_ps(2.0);
 753     x                = xx[0];
 754     f                = ff[0];
 755
 756     nri              = nlist->nri;
 757     iinr             = nlist->iinr;
 758     jindex           = nlist->jindex;
 759     jjnr             = nlist->jjnr;
 760     shiftidx         = nlist->shift;
 761     gid              = nlist->gid;
 762     shiftvec         = fr->shift_vec[0];
 763     fshift           = fr->fshift[0];
 764     facel            = _mm_set1_ps(fr->ic->epsfac);
 765     charge           = mdatoms->chargeA;
 766     krf              = _mm_set1_ps(fr->ic->k_rf);
 767     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 768     crf              = _mm_set1_ps(fr->ic->c_rf);
 769     nvdwtype         = fr->ntype;
 770     vdwparam         = fr->nbfp;
 771     vdwtype          = mdatoms->typeA;
 772
 773     vftab            = kernel_data->table_vdw->data;
 774     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 775
 776     /* Setup water-specific parameters */
 777     inr              = nlist->iinr[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     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 781     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 782
 783     /* Avoid stupid compiler warnings */
 784     jnrA = jnrB = jnrC = jnrD = 0;
 785     j_coord_offsetA = 0;
 786     j_coord_offsetB = 0;
 787     j_coord_offsetC = 0;
 788     j_coord_offsetD = 0;
 789
 790     outeriter        = 0;
 791     inneriter        = 0;
 792
 793     for(iidx=0;iidx<4*DIM;iidx++)
 794     {
 795         scratch[iidx] = 0.0;
 796     }
 797
 798     /* Start outer loop over neighborlists */
 799     for(iidx=0; iidx<nri; iidx++)
 800     {
 801         /* Load shift vector for this list */
 802         i_shift_offset   = DIM*shiftidx[iidx];
 803
 804         /* Load limits for loop over neighbors */
 805         j_index_start    = jindex[iidx];
 806         j_index_end      = jindex[iidx+1];
 807
 808         /* Get outer coordinate index */
 809         inr              = iinr[iidx];
 810         i_coord_offset   = DIM*inr;
 811
 812         /* Load i particle coords and add shift vector */
 813         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 814                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 815
 816         fix0             = _mm_setzero_ps();
 817         fiy0             = _mm_setzero_ps();
 818         fiz0             = _mm_setzero_ps();
 819         fix1             = _mm_setzero_ps();
 820         fiy1             = _mm_setzero_ps();
 821         fiz1             = _mm_setzero_ps();
 822         fix2             = _mm_setzero_ps();
 823         fiy2             = _mm_setzero_ps();
 824         fiz2             = _mm_setzero_ps();
 825         fix3             = _mm_setzero_ps();
 826         fiy3             = _mm_setzero_ps();
 827         fiz3             = _mm_setzero_ps();
 828
 829         /* Start inner kernel loop */
 830         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 831         {
 832
 833             /* Get j neighbor index, and coordinate index */
 834             jnrA             = jjnr[jidx];
 835             jnrB             = jjnr[jidx+1];
 836             jnrC             = jjnr[jidx+2];
 837             jnrD             = jjnr[jidx+3];
 838             j_coord_offsetA  = DIM*jnrA;
 839             j_coord_offsetB  = DIM*jnrB;
 840             j_coord_offsetC  = DIM*jnrC;
 841             j_coord_offsetD  = DIM*jnrD;
 842
 843             /* load j atom coordinates */
 844             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 845                                               x+j_coord_offsetC,x+j_coord_offsetD,
 846                                               &jx0,&jy0,&jz0);
 847
 848             /* Calculate displacement vector */
 849             dx00             = _mm_sub_ps(ix0,jx0);
 850             dy00             = _mm_sub_ps(iy0,jy0);
 851             dz00             = _mm_sub_ps(iz0,jz0);
 852             dx10             = _mm_sub_ps(ix1,jx0);
 853             dy10             = _mm_sub_ps(iy1,jy0);
 854             dz10             = _mm_sub_ps(iz1,jz0);
 855             dx20             = _mm_sub_ps(ix2,jx0);
 856             dy20             = _mm_sub_ps(iy2,jy0);
 857             dz20             = _mm_sub_ps(iz2,jz0);
 858             dx30             = _mm_sub_ps(ix3,jx0);
 859             dy30             = _mm_sub_ps(iy3,jy0);
 860             dz30             = _mm_sub_ps(iz3,jz0);
 861
 862             /* Calculate squared distance and things based on it */
 863             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 864             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 865             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 866             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 867
 868             rinv00           = sse2_invsqrt_f(rsq00);
 869             rinv10           = sse2_invsqrt_f(rsq10);
 870             rinv20           = sse2_invsqrt_f(rsq20);
 871             rinv30           = sse2_invsqrt_f(rsq30);
 872
 873             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 874             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 875             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 876
 877             /* Load parameters for j particles */
 878             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 879                                                               charge+jnrC+0,charge+jnrD+0);
 880             vdwjidx0A        = 2*vdwtype[jnrA+0];
 881             vdwjidx0B        = 2*vdwtype[jnrB+0];
 882             vdwjidx0C        = 2*vdwtype[jnrC+0];
 883             vdwjidx0D        = 2*vdwtype[jnrD+0];
 884
 885             fjx0             = _mm_setzero_ps();
 886             fjy0             = _mm_setzero_ps();
 887             fjz0             = _mm_setzero_ps();
 888
 889             /**************************
 890              * CALCULATE INTERACTIONS *
 891              **************************/
 892
 893             r00              = _mm_mul_ps(rsq00,rinv00);
 894
 895             /* Compute parameters for interactions between i and j atoms */
 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_cvtepi32_ps(vfitab));
 906             vfitab           = _mm_slli_epi32(vfitab,3);
 907
 908             /* CUBIC SPLINE TABLE DISPERSION */
 909             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 910             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 911             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 912             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 913             _MM_TRANSPOSE4_PS(Y,F,G,H);
 914             Heps             = _mm_mul_ps(vfeps,H);
 915             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 916             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 917             fvdw6            = _mm_mul_ps(c6_00,FF);
 918
 919             /* CUBIC SPLINE TABLE REPULSION */
 920             vfitab           = _mm_add_epi32(vfitab,ifour);
 921             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 922             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 923             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 924             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 925             _MM_TRANSPOSE4_PS(Y,F,G,H);
 926             Heps             = _mm_mul_ps(vfeps,H);
 927             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 928             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 929             fvdw12           = _mm_mul_ps(c12_00,FF);
 930             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 931
 932             fscal            = fvdw;
 933
 934             /* Calculate temporary vectorial force */
 935             tx               = _mm_mul_ps(fscal,dx00);
 936             ty               = _mm_mul_ps(fscal,dy00);
 937             tz               = _mm_mul_ps(fscal,dz00);
 938
 939             /* Update vectorial force */
 940             fix0             = _mm_add_ps(fix0,tx);
 941             fiy0             = _mm_add_ps(fiy0,ty);
 942             fiz0             = _mm_add_ps(fiz0,tz);
 943
 944             fjx0             = _mm_add_ps(fjx0,tx);
 945             fjy0             = _mm_add_ps(fjy0,ty);
 946             fjz0             = _mm_add_ps(fjz0,tz);
 947
 948             /**************************
 949              * CALCULATE INTERACTIONS *
 950              **************************/
 951
 952             /* Compute parameters for interactions between i and j atoms */
 953             qq10             = _mm_mul_ps(iq1,jq0);
 954
 955             /* REACTION-FIELD ELECTROSTATICS */
 956             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 957
 958             fscal            = felec;
 959
 960             /* Calculate temporary vectorial force */
 961             tx               = _mm_mul_ps(fscal,dx10);
 962             ty               = _mm_mul_ps(fscal,dy10);
 963             tz               = _mm_mul_ps(fscal,dz10);
 964
 965             /* Update vectorial force */
 966             fix1             = _mm_add_ps(fix1,tx);
 967             fiy1             = _mm_add_ps(fiy1,ty);
 968             fiz1             = _mm_add_ps(fiz1,tz);
 969
 970             fjx0             = _mm_add_ps(fjx0,tx);
 971             fjy0             = _mm_add_ps(fjy0,ty);
 972             fjz0             = _mm_add_ps(fjz0,tz);
 973
 974             /**************************
 975              * CALCULATE INTERACTIONS *
 976              **************************/
 977
 978             /* Compute parameters for interactions between i and j atoms */
 979             qq20             = _mm_mul_ps(iq2,jq0);
 980
 981             /* REACTION-FIELD ELECTROSTATICS */
 982             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 983
 984             fscal            = felec;
 985
 986             /* Calculate temporary vectorial force */
 987             tx               = _mm_mul_ps(fscal,dx20);
 988             ty               = _mm_mul_ps(fscal,dy20);
 989             tz               = _mm_mul_ps(fscal,dz20);
 990
 991             /* Update vectorial force */
 992             fix2             = _mm_add_ps(fix2,tx);
 993             fiy2             = _mm_add_ps(fiy2,ty);
 994             fiz2             = _mm_add_ps(fiz2,tz);
 995
 996             fjx0             = _mm_add_ps(fjx0,tx);
 997             fjy0             = _mm_add_ps(fjy0,ty);
 998             fjz0             = _mm_add_ps(fjz0,tz);
 999
1000             /**************************
1001              * CALCULATE INTERACTIONS *
1002              **************************/
1003
1004             /* Compute parameters for interactions between i and j atoms */
1005             qq30             = _mm_mul_ps(iq3,jq0);
1006
1007             /* REACTION-FIELD ELECTROSTATICS */
1008             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1009
1010             fscal            = felec;
1011
1012             /* Calculate temporary vectorial force */
1013             tx               = _mm_mul_ps(fscal,dx30);
1014             ty               = _mm_mul_ps(fscal,dy30);
1015             tz               = _mm_mul_ps(fscal,dz30);
1016
1017             /* Update vectorial force */
1018             fix3             = _mm_add_ps(fix3,tx);
1019             fiy3             = _mm_add_ps(fiy3,ty);
1020             fiz3             = _mm_add_ps(fiz3,tz);
1021
1022             fjx0             = _mm_add_ps(fjx0,tx);
1023             fjy0             = _mm_add_ps(fjy0,ty);
1024             fjz0             = _mm_add_ps(fjz0,tz);
1025
1026             fjptrA             = f+j_coord_offsetA;
1027             fjptrB             = f+j_coord_offsetB;
1028             fjptrC             = f+j_coord_offsetC;
1029             fjptrD             = f+j_coord_offsetD;
1030
1031             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1032
1033             /* Inner loop uses 129 flops */
1034         }
1035
1036         if(jidx<j_index_end)
1037         {
1038
1039             /* Get j neighbor index, and coordinate index */
1040             jnrlistA         = jjnr[jidx];
1041             jnrlistB         = jjnr[jidx+1];
1042             jnrlistC         = jjnr[jidx+2];
1043             jnrlistD         = jjnr[jidx+3];
1044             /* Sign of each element will be negative for non-real atoms.
1045              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1046              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1047              */
1048             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1049             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1050             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1051             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1052             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1053             j_coord_offsetA  = DIM*jnrA;
1054             j_coord_offsetB  = DIM*jnrB;
1055             j_coord_offsetC  = DIM*jnrC;
1056             j_coord_offsetD  = DIM*jnrD;
1057
1058             /* load j atom coordinates */
1059             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1060                                               x+j_coord_offsetC,x+j_coord_offsetD,
1061                                               &jx0,&jy0,&jz0);
1062
1063             /* Calculate displacement vector */
1064             dx00             = _mm_sub_ps(ix0,jx0);
1065             dy00             = _mm_sub_ps(iy0,jy0);
1066             dz00             = _mm_sub_ps(iz0,jz0);
1067             dx10             = _mm_sub_ps(ix1,jx0);
1068             dy10             = _mm_sub_ps(iy1,jy0);
1069             dz10             = _mm_sub_ps(iz1,jz0);
1070             dx20             = _mm_sub_ps(ix2,jx0);
1071             dy20             = _mm_sub_ps(iy2,jy0);
1072             dz20             = _mm_sub_ps(iz2,jz0);
1073             dx30             = _mm_sub_ps(ix3,jx0);
1074             dy30             = _mm_sub_ps(iy3,jy0);
1075             dz30             = _mm_sub_ps(iz3,jz0);
1076
1077             /* Calculate squared distance and things based on it */
1078             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1079             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1080             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1081             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1082
1083             rinv00           = sse2_invsqrt_f(rsq00);
1084             rinv10           = sse2_invsqrt_f(rsq10);
1085             rinv20           = sse2_invsqrt_f(rsq20);
1086             rinv30           = sse2_invsqrt_f(rsq30);
1087
1088             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1089             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1090             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1091
1092             /* Load parameters for j particles */
1093             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1094                                                               charge+jnrC+0,charge+jnrD+0);
1095             vdwjidx0A        = 2*vdwtype[jnrA+0];
1096             vdwjidx0B        = 2*vdwtype[jnrB+0];
1097             vdwjidx0C        = 2*vdwtype[jnrC+0];
1098             vdwjidx0D        = 2*vdwtype[jnrD+0];
1099
1100             fjx0             = _mm_setzero_ps();
1101             fjy0             = _mm_setzero_ps();
1102             fjz0             = _mm_setzero_ps();
1103
1104             /**************************
1105              * CALCULATE INTERACTIONS *
1106              **************************/
1107
1108             r00              = _mm_mul_ps(rsq00,rinv00);
1109             r00              = _mm_andnot_ps(dummy_mask,r00);
1110
1111             /* Compute parameters for interactions between i and j atoms */
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_cvtepi32_ps(vfitab));
1122             vfitab           = _mm_slli_epi32(vfitab,3);
1123
1124             /* CUBIC SPLINE TABLE DISPERSION */
1125             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1126             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1127             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1128             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1129             _MM_TRANSPOSE4_PS(Y,F,G,H);
1130             Heps             = _mm_mul_ps(vfeps,H);
1131             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1132             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1133             fvdw6            = _mm_mul_ps(c6_00,FF);
1134
1135             /* CUBIC SPLINE TABLE REPULSION */
1136             vfitab           = _mm_add_epi32(vfitab,ifour);
1137             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1138             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1139             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1140             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1141             _MM_TRANSPOSE4_PS(Y,F,G,H);
1142             Heps             = _mm_mul_ps(vfeps,H);
1143             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1144             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1145             fvdw12           = _mm_mul_ps(c12_00,FF);
1146             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1147
1148             fscal            = fvdw;
1149
1150             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1151
1152             /* Calculate temporary vectorial force */
1153             tx               = _mm_mul_ps(fscal,dx00);
1154             ty               = _mm_mul_ps(fscal,dy00);
1155             tz               = _mm_mul_ps(fscal,dz00);
1156
1157             /* Update vectorial force */
1158             fix0             = _mm_add_ps(fix0,tx);
1159             fiy0             = _mm_add_ps(fiy0,ty);
1160             fiz0             = _mm_add_ps(fiz0,tz);
1161
1162             fjx0             = _mm_add_ps(fjx0,tx);
1163             fjy0             = _mm_add_ps(fjy0,ty);
1164             fjz0             = _mm_add_ps(fjz0,tz);
1165
1166             /**************************
1167              * CALCULATE INTERACTIONS *
1168              **************************/
1169
1170             /* Compute parameters for interactions between i and j atoms */
1171             qq10             = _mm_mul_ps(iq1,jq0);
1172
1173             /* REACTION-FIELD ELECTROSTATICS */
1174             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1175
1176             fscal            = felec;
1177
1178             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1179
1180             /* Calculate temporary vectorial force */
1181             tx               = _mm_mul_ps(fscal,dx10);
1182             ty               = _mm_mul_ps(fscal,dy10);
1183             tz               = _mm_mul_ps(fscal,dz10);
1184
1185             /* Update vectorial force */
1186             fix1             = _mm_add_ps(fix1,tx);
1187             fiy1             = _mm_add_ps(fiy1,ty);
1188             fiz1             = _mm_add_ps(fiz1,tz);
1189
1190             fjx0             = _mm_add_ps(fjx0,tx);
1191             fjy0             = _mm_add_ps(fjy0,ty);
1192             fjz0             = _mm_add_ps(fjz0,tz);
1193
1194             /**************************
1195              * CALCULATE INTERACTIONS *
1196              **************************/
1197
1198             /* Compute parameters for interactions between i and j atoms */
1199             qq20             = _mm_mul_ps(iq2,jq0);
1200
1201             /* REACTION-FIELD ELECTROSTATICS */
1202             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1203
1204             fscal            = felec;
1205
1206             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1207
1208             /* Calculate temporary vectorial force */
1209             tx               = _mm_mul_ps(fscal,dx20);
1210             ty               = _mm_mul_ps(fscal,dy20);
1211             tz               = _mm_mul_ps(fscal,dz20);
1212
1213             /* Update vectorial force */
1214             fix2             = _mm_add_ps(fix2,tx);
1215             fiy2             = _mm_add_ps(fiy2,ty);
1216             fiz2             = _mm_add_ps(fiz2,tz);
1217
1218             fjx0             = _mm_add_ps(fjx0,tx);
1219             fjy0             = _mm_add_ps(fjy0,ty);
1220             fjz0             = _mm_add_ps(fjz0,tz);
1221
1222             /**************************
1223              * CALCULATE INTERACTIONS *
1224              **************************/
1225
1226             /* Compute parameters for interactions between i and j atoms */
1227             qq30             = _mm_mul_ps(iq3,jq0);
1228
1229             /* REACTION-FIELD ELECTROSTATICS */
1230             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1231
1232             fscal            = felec;
1233
1234             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1235
1236             /* Calculate temporary vectorial force */
1237             tx               = _mm_mul_ps(fscal,dx30);
1238             ty               = _mm_mul_ps(fscal,dy30);
1239             tz               = _mm_mul_ps(fscal,dz30);
1240
1241             /* Update vectorial force */
1242             fix3             = _mm_add_ps(fix3,tx);
1243             fiy3             = _mm_add_ps(fiy3,ty);
1244             fiz3             = _mm_add_ps(fiz3,tz);
1245
1246             fjx0             = _mm_add_ps(fjx0,tx);
1247             fjy0             = _mm_add_ps(fjy0,ty);
1248             fjz0             = _mm_add_ps(fjz0,tz);
1249
1250             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1251             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1252             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1253             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1254
1255             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1256
1257             /* Inner loop uses 130 flops */
1258         }
1259
1260         /* End of innermost loop */
1261
1262         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1263                                               f+i_coord_offset,fshift+i_shift_offset);
1264
1265         /* Increment number of inner iterations */
1266         inneriter                  += j_index_end - j_index_start;
1267
1268         /* Outer loop uses 24 flops */
1269     }
1270
1271     /* Increment number of outer iterations */
1272     outeriter        += nri;
1273
1274     /* Update outer/inner flops */
1275
1276     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*130);
1277 }