src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRF_VdwCSTab_GeomW3P1_sse2_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 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/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse2_single.h"
  50 #include "kernelutil_x86_sse2_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW3P1_VF_sse2_single
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_VF_sse2_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  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->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     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 138     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 139     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 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_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 173                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 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
 185         /* Reset potential sums */
 186         velecsum         = _mm_setzero_ps();
 187         vvdwsum          = _mm_setzero_ps();
 188
 189         /* Start inner kernel loop */
 190         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 191         {
 192
 193             /* Get j neighbor index, and coordinate index */
 194             jnrA             = jjnr[jidx];
 195             jnrB             = jjnr[jidx+1];
 196             jnrC             = jjnr[jidx+2];
 197             jnrD             = jjnr[jidx+3];
 198             j_coord_offsetA  = DIM*jnrA;
 199             j_coord_offsetB  = DIM*jnrB;
 200             j_coord_offsetC  = DIM*jnrC;
 201             j_coord_offsetD  = DIM*jnrD;
 202
 203             /* load j atom coordinates */
 204             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 205                                               x+j_coord_offsetC,x+j_coord_offsetD,
 206                                               &jx0,&jy0,&jz0);
 207
 208             /* Calculate displacement vector */
 209             dx00             = _mm_sub_ps(ix0,jx0);
 210             dy00             = _mm_sub_ps(iy0,jy0);
 211             dz00             = _mm_sub_ps(iz0,jz0);
 212             dx10             = _mm_sub_ps(ix1,jx0);
 213             dy10             = _mm_sub_ps(iy1,jy0);
 214             dz10             = _mm_sub_ps(iz1,jz0);
 215             dx20             = _mm_sub_ps(ix2,jx0);
 216             dy20             = _mm_sub_ps(iy2,jy0);
 217             dz20             = _mm_sub_ps(iz2,jz0);
 218
 219             /* Calculate squared distance and things based on it */
 220             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 221             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 222             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 223
 224             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 225             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 226             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 227
 228             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 229             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 230             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 231
 232             /* Load parameters for j particles */
 233             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 234                                                               charge+jnrC+0,charge+jnrD+0);
 235             vdwjidx0A        = 2*vdwtype[jnrA+0];
 236             vdwjidx0B        = 2*vdwtype[jnrB+0];
 237             vdwjidx0C        = 2*vdwtype[jnrC+0];
 238             vdwjidx0D        = 2*vdwtype[jnrD+0];
 239
 240             fjx0             = _mm_setzero_ps();
 241             fjy0             = _mm_setzero_ps();
 242             fjz0             = _mm_setzero_ps();
 243
 244             /**************************
 245              * CALCULATE INTERACTIONS *
 246              **************************/
 247
 248             r00              = _mm_mul_ps(rsq00,rinv00);
 249
 250             /* Compute parameters for interactions between i and j atoms */
 251             qq00             = _mm_mul_ps(iq0,jq0);
 252             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 253                                          vdwparam+vdwioffset0+vdwjidx0B,
 254                                          vdwparam+vdwioffset0+vdwjidx0C,
 255                                          vdwparam+vdwioffset0+vdwjidx0D,
 256                                          &c6_00,&c12_00);
 257
 258             /* Calculate table index by multiplying r with table scale and truncate to integer */
 259             rt               = _mm_mul_ps(r00,vftabscale);
 260             vfitab           = _mm_cvttps_epi32(rt);
 261             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 262             vfitab           = _mm_slli_epi32(vfitab,3);
 263
 264             /* REACTION-FIELD ELECTROSTATICS */
 265             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 266             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 267
 268             /* CUBIC SPLINE TABLE DISPERSION */
 269             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 270             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 271             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 272             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 273             _MM_TRANSPOSE4_PS(Y,F,G,H);
 274             Heps             = _mm_mul_ps(vfeps,H);
 275             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 276             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 277             vvdw6            = _mm_mul_ps(c6_00,VV);
 278             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 279             fvdw6            = _mm_mul_ps(c6_00,FF);
 280
 281             /* CUBIC SPLINE TABLE REPULSION */
 282             vfitab           = _mm_add_epi32(vfitab,ifour);
 283             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 284             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 285             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 286             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 287             _MM_TRANSPOSE4_PS(Y,F,G,H);
 288             Heps             = _mm_mul_ps(vfeps,H);
 289             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 290             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 291             vvdw12           = _mm_mul_ps(c12_00,VV);
 292             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 293             fvdw12           = _mm_mul_ps(c12_00,FF);
 294             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 295             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 296
 297             /* Update potential sum for this i atom from the interaction with this j atom. */
 298             velecsum         = _mm_add_ps(velecsum,velec);
 299             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 300
 301             fscal            = _mm_add_ps(felec,fvdw);
 302
 303             /* Calculate temporary vectorial force */
 304             tx               = _mm_mul_ps(fscal,dx00);
 305             ty               = _mm_mul_ps(fscal,dy00);
 306             tz               = _mm_mul_ps(fscal,dz00);
 307
 308             /* Update vectorial force */
 309             fix0             = _mm_add_ps(fix0,tx);
 310             fiy0             = _mm_add_ps(fiy0,ty);
 311             fiz0             = _mm_add_ps(fiz0,tz);
 312
 313             fjx0             = _mm_add_ps(fjx0,tx);
 314             fjy0             = _mm_add_ps(fjy0,ty);
 315             fjz0             = _mm_add_ps(fjz0,tz);
 316
 317             /**************************
 318              * CALCULATE INTERACTIONS *
 319              **************************/
 320
 321             /* Compute parameters for interactions between i and j atoms */
 322             qq10             = _mm_mul_ps(iq1,jq0);
 323
 324             /* REACTION-FIELD ELECTROSTATICS */
 325             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 326             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 327
 328             /* Update potential sum for this i atom from the interaction with this j atom. */
 329             velecsum         = _mm_add_ps(velecsum,velec);
 330
 331             fscal            = felec;
 332
 333             /* Calculate temporary vectorial force */
 334             tx               = _mm_mul_ps(fscal,dx10);
 335             ty               = _mm_mul_ps(fscal,dy10);
 336             tz               = _mm_mul_ps(fscal,dz10);
 337
 338             /* Update vectorial force */
 339             fix1             = _mm_add_ps(fix1,tx);
 340             fiy1             = _mm_add_ps(fiy1,ty);
 341             fiz1             = _mm_add_ps(fiz1,tz);
 342
 343             fjx0             = _mm_add_ps(fjx0,tx);
 344             fjy0             = _mm_add_ps(fjy0,ty);
 345             fjz0             = _mm_add_ps(fjz0,tz);
 346
 347             /**************************
 348              * CALCULATE INTERACTIONS *
 349              **************************/
 350
 351             /* Compute parameters for interactions between i and j atoms */
 352             qq20             = _mm_mul_ps(iq2,jq0);
 353
 354             /* REACTION-FIELD ELECTROSTATICS */
 355             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 356             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 357
 358             /* Update potential sum for this i atom from the interaction with this j atom. */
 359             velecsum         = _mm_add_ps(velecsum,velec);
 360
 361             fscal            = felec;
 362
 363             /* Calculate temporary vectorial force */
 364             tx               = _mm_mul_ps(fscal,dx20);
 365             ty               = _mm_mul_ps(fscal,dy20);
 366             tz               = _mm_mul_ps(fscal,dz20);
 367
 368             /* Update vectorial force */
 369             fix2             = _mm_add_ps(fix2,tx);
 370             fiy2             = _mm_add_ps(fiy2,ty);
 371             fiz2             = _mm_add_ps(fiz2,tz);
 372
 373             fjx0             = _mm_add_ps(fjx0,tx);
 374             fjy0             = _mm_add_ps(fjy0,ty);
 375             fjz0             = _mm_add_ps(fjz0,tz);
 376
 377             fjptrA             = f+j_coord_offsetA;
 378             fjptrB             = f+j_coord_offsetB;
 379             fjptrC             = f+j_coord_offsetC;
 380             fjptrD             = f+j_coord_offsetD;
 381
 382             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 383
 384             /* Inner loop uses 131 flops */
 385         }
 386
 387         if(jidx<j_index_end)
 388         {
 389
 390             /* Get j neighbor index, and coordinate index */
 391             jnrlistA         = jjnr[jidx];
 392             jnrlistB         = jjnr[jidx+1];
 393             jnrlistC         = jjnr[jidx+2];
 394             jnrlistD         = jjnr[jidx+3];
 395             /* Sign of each element will be negative for non-real atoms.
 396              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 397              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 398              */
 399             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 400             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 401             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 402             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 403             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 404             j_coord_offsetA  = DIM*jnrA;
 405             j_coord_offsetB  = DIM*jnrB;
 406             j_coord_offsetC  = DIM*jnrC;
 407             j_coord_offsetD  = DIM*jnrD;
 408
 409             /* load j atom coordinates */
 410             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 411                                               x+j_coord_offsetC,x+j_coord_offsetD,
 412                                               &jx0,&jy0,&jz0);
 413
 414             /* Calculate displacement vector */
 415             dx00             = _mm_sub_ps(ix0,jx0);
 416             dy00             = _mm_sub_ps(iy0,jy0);
 417             dz00             = _mm_sub_ps(iz0,jz0);
 418             dx10             = _mm_sub_ps(ix1,jx0);
 419             dy10             = _mm_sub_ps(iy1,jy0);
 420             dz10             = _mm_sub_ps(iz1,jz0);
 421             dx20             = _mm_sub_ps(ix2,jx0);
 422             dy20             = _mm_sub_ps(iy2,jy0);
 423             dz20             = _mm_sub_ps(iz2,jz0);
 424
 425             /* Calculate squared distance and things based on it */
 426             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 427             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 428             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 429
 430             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 431             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 432             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 433
 434             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 435             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 436             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 437
 438             /* Load parameters for j particles */
 439             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 440                                                               charge+jnrC+0,charge+jnrD+0);
 441             vdwjidx0A        = 2*vdwtype[jnrA+0];
 442             vdwjidx0B        = 2*vdwtype[jnrB+0];
 443             vdwjidx0C        = 2*vdwtype[jnrC+0];
 444             vdwjidx0D        = 2*vdwtype[jnrD+0];
 445
 446             fjx0             = _mm_setzero_ps();
 447             fjy0             = _mm_setzero_ps();
 448             fjz0             = _mm_setzero_ps();
 449
 450             /**************************
 451              * CALCULATE INTERACTIONS *
 452              **************************/
 453
 454             r00              = _mm_mul_ps(rsq00,rinv00);
 455             r00              = _mm_andnot_ps(dummy_mask,r00);
 456
 457             /* Compute parameters for interactions between i and j atoms */
 458             qq00             = _mm_mul_ps(iq0,jq0);
 459             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 460                                          vdwparam+vdwioffset0+vdwjidx0B,
 461                                          vdwparam+vdwioffset0+vdwjidx0C,
 462                                          vdwparam+vdwioffset0+vdwjidx0D,
 463                                          &c6_00,&c12_00);
 464
 465             /* Calculate table index by multiplying r with table scale and truncate to integer */
 466             rt               = _mm_mul_ps(r00,vftabscale);
 467             vfitab           = _mm_cvttps_epi32(rt);
 468             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 469             vfitab           = _mm_slli_epi32(vfitab,3);
 470
 471             /* REACTION-FIELD ELECTROSTATICS */
 472             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 473             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 474
 475             /* CUBIC SPLINE TABLE DISPERSION */
 476             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 477             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 478             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 479             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 480             _MM_TRANSPOSE4_PS(Y,F,G,H);
 481             Heps             = _mm_mul_ps(vfeps,H);
 482             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 483             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 484             vvdw6            = _mm_mul_ps(c6_00,VV);
 485             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 486             fvdw6            = _mm_mul_ps(c6_00,FF);
 487
 488             /* CUBIC SPLINE TABLE REPULSION */
 489             vfitab           = _mm_add_epi32(vfitab,ifour);
 490             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 491             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 492             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 493             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 494             _MM_TRANSPOSE4_PS(Y,F,G,H);
 495             Heps             = _mm_mul_ps(vfeps,H);
 496             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 497             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 498             vvdw12           = _mm_mul_ps(c12_00,VV);
 499             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 500             fvdw12           = _mm_mul_ps(c12_00,FF);
 501             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 502             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 503
 504             /* Update potential sum for this i atom from the interaction with this j atom. */
 505             velec            = _mm_andnot_ps(dummy_mask,velec);
 506             velecsum         = _mm_add_ps(velecsum,velec);
 507             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 508             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 509
 510             fscal            = _mm_add_ps(felec,fvdw);
 511
 512             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 513
 514             /* Calculate temporary vectorial force */
 515             tx               = _mm_mul_ps(fscal,dx00);
 516             ty               = _mm_mul_ps(fscal,dy00);
 517             tz               = _mm_mul_ps(fscal,dz00);
 518
 519             /* Update vectorial force */
 520             fix0             = _mm_add_ps(fix0,tx);
 521             fiy0             = _mm_add_ps(fiy0,ty);
 522             fiz0             = _mm_add_ps(fiz0,tz);
 523
 524             fjx0             = _mm_add_ps(fjx0,tx);
 525             fjy0             = _mm_add_ps(fjy0,ty);
 526             fjz0             = _mm_add_ps(fjz0,tz);
 527
 528             /**************************
 529              * CALCULATE INTERACTIONS *
 530              **************************/
 531
 532             /* Compute parameters for interactions between i and j atoms */
 533             qq10             = _mm_mul_ps(iq1,jq0);
 534
 535             /* REACTION-FIELD ELECTROSTATICS */
 536             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 537             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 538
 539             /* Update potential sum for this i atom from the interaction with this j atom. */
 540             velec            = _mm_andnot_ps(dummy_mask,velec);
 541             velecsum         = _mm_add_ps(velecsum,velec);
 542
 543             fscal            = felec;
 544
 545             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 546
 547             /* Calculate temporary vectorial force */
 548             tx               = _mm_mul_ps(fscal,dx10);
 549             ty               = _mm_mul_ps(fscal,dy10);
 550             tz               = _mm_mul_ps(fscal,dz10);
 551
 552             /* Update vectorial force */
 553             fix1             = _mm_add_ps(fix1,tx);
 554             fiy1             = _mm_add_ps(fiy1,ty);
 555             fiz1             = _mm_add_ps(fiz1,tz);
 556
 557             fjx0             = _mm_add_ps(fjx0,tx);
 558             fjy0             = _mm_add_ps(fjy0,ty);
 559             fjz0             = _mm_add_ps(fjz0,tz);
 560
 561             /**************************
 562              * CALCULATE INTERACTIONS *
 563              **************************/
 564
 565             /* Compute parameters for interactions between i and j atoms */
 566             qq20             = _mm_mul_ps(iq2,jq0);
 567
 568             /* REACTION-FIELD ELECTROSTATICS */
 569             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 570             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 571
 572             /* Update potential sum for this i atom from the interaction with this j atom. */
 573             velec            = _mm_andnot_ps(dummy_mask,velec);
 574             velecsum         = _mm_add_ps(velecsum,velec);
 575
 576             fscal            = felec;
 577
 578             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 579
 580             /* Calculate temporary vectorial force */
 581             tx               = _mm_mul_ps(fscal,dx20);
 582             ty               = _mm_mul_ps(fscal,dy20);
 583             tz               = _mm_mul_ps(fscal,dz20);
 584
 585             /* Update vectorial force */
 586             fix2             = _mm_add_ps(fix2,tx);
 587             fiy2             = _mm_add_ps(fiy2,ty);
 588             fiz2             = _mm_add_ps(fiz2,tz);
 589
 590             fjx0             = _mm_add_ps(fjx0,tx);
 591             fjy0             = _mm_add_ps(fjy0,ty);
 592             fjz0             = _mm_add_ps(fjz0,tz);
 593
 594             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 595             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 596             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 597             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 598
 599             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 600
 601             /* Inner loop uses 132 flops */
 602         }
 603
 604         /* End of innermost loop */
 605
 606         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 607                                               f+i_coord_offset,fshift+i_shift_offset);
 608
 609         ggid                        = gid[iidx];
 610         /* Update potential energies */
 611         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 612         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 613
 614         /* Increment number of inner iterations */
 615         inneriter                  += j_index_end - j_index_start;
 616
 617         /* Outer loop uses 20 flops */
 618     }
 619
 620     /* Increment number of outer iterations */
 621     outeriter        += nri;
 622
 623     /* Update outer/inner flops */
 624
 625     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*132);
 626 }
 627 /*
 628  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_sse2_single
 629  * Electrostatics interaction: ReactionField
 630  * VdW interaction:            CubicSplineTable
 631  * Geometry:                   Water3-Particle
 632  * Calculate force/pot:        Force
 633  */
 634 void
 635 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_sse2_single
 636                     (t_nblist                    * gmx_restrict       nlist,
 637                      rvec                        * gmx_restrict          xx,
 638                      rvec                        * gmx_restrict          ff,
 639                      t_forcerec                  * gmx_restrict          fr,
 640                      t_mdatoms                   * gmx_restrict     mdatoms,
 641                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 642                      t_nrnb                      * gmx_restrict        nrnb)
 643 {
 644     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 645      * just 0 for non-waters.
 646      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 647      * jnr indices corresponding to data put in the four positions in the SIMD register.
 648      */
 649     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 650     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 651     int              jnrA,jnrB,jnrC,jnrD;
 652     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 653     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 654     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 655     real             rcutoff_scalar;
 656     real             *shiftvec,*fshift,*x,*f;
 657     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 658     real             scratch[4*DIM];
 659     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 660     int              vdwioffset0;
 661     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 662     int              vdwioffset1;
 663     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 664     int              vdwioffset2;
 665     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 666     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 667     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 668     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 669     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 670     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 671     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 672     real             *charge;
 673     int              nvdwtype;
 674     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 675     int              *vdwtype;
 676     real             *vdwparam;
 677     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 678     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 679     __m128i          vfitab;
 680     __m128i          ifour       = _mm_set1_epi32(4);
 681     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 682     real             *vftab;
 683     __m128           dummy_mask,cutoff_mask;
 684     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 685     __m128           one     = _mm_set1_ps(1.0);
 686     __m128           two     = _mm_set1_ps(2.0);
 687     x                = xx[0];
 688     f                = ff[0];
 689
 690     nri              = nlist->nri;
 691     iinr             = nlist->iinr;
 692     jindex           = nlist->jindex;
 693     jjnr             = nlist->jjnr;
 694     shiftidx         = nlist->shift;
 695     gid              = nlist->gid;
 696     shiftvec         = fr->shift_vec[0];
 697     fshift           = fr->fshift[0];
 698     facel            = _mm_set1_ps(fr->epsfac);
 699     charge           = mdatoms->chargeA;
 700     krf              = _mm_set1_ps(fr->ic->k_rf);
 701     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 702     crf              = _mm_set1_ps(fr->ic->c_rf);
 703     nvdwtype         = fr->ntype;
 704     vdwparam         = fr->nbfp;
 705     vdwtype          = mdatoms->typeA;
 706
 707     vftab            = kernel_data->table_vdw->data;
 708     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 709
 710     /* Setup water-specific parameters */
 711     inr              = nlist->iinr[0];
 712     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 713     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 714     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 715     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 716
 717     /* Avoid stupid compiler warnings */
 718     jnrA = jnrB = jnrC = jnrD = 0;
 719     j_coord_offsetA = 0;
 720     j_coord_offsetB = 0;
 721     j_coord_offsetC = 0;
 722     j_coord_offsetD = 0;
 723
 724     outeriter        = 0;
 725     inneriter        = 0;
 726
 727     for(iidx=0;iidx<4*DIM;iidx++)
 728     {
 729         scratch[iidx] = 0.0;
 730     }
 731
 732     /* Start outer loop over neighborlists */
 733     for(iidx=0; iidx<nri; iidx++)
 734     {
 735         /* Load shift vector for this list */
 736         i_shift_offset   = DIM*shiftidx[iidx];
 737
 738         /* Load limits for loop over neighbors */
 739         j_index_start    = jindex[iidx];
 740         j_index_end      = jindex[iidx+1];
 741
 742         /* Get outer coordinate index */
 743         inr              = iinr[iidx];
 744         i_coord_offset   = DIM*inr;
 745
 746         /* Load i particle coords and add shift vector */
 747         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 748                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 749
 750         fix0             = _mm_setzero_ps();
 751         fiy0             = _mm_setzero_ps();
 752         fiz0             = _mm_setzero_ps();
 753         fix1             = _mm_setzero_ps();
 754         fiy1             = _mm_setzero_ps();
 755         fiz1             = _mm_setzero_ps();
 756         fix2             = _mm_setzero_ps();
 757         fiy2             = _mm_setzero_ps();
 758         fiz2             = _mm_setzero_ps();
 759
 760         /* Start inner kernel loop */
 761         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 762         {
 763
 764             /* Get j neighbor index, and coordinate index */
 765             jnrA             = jjnr[jidx];
 766             jnrB             = jjnr[jidx+1];
 767             jnrC             = jjnr[jidx+2];
 768             jnrD             = jjnr[jidx+3];
 769             j_coord_offsetA  = DIM*jnrA;
 770             j_coord_offsetB  = DIM*jnrB;
 771             j_coord_offsetC  = DIM*jnrC;
 772             j_coord_offsetD  = DIM*jnrD;
 773
 774             /* load j atom coordinates */
 775             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 776                                               x+j_coord_offsetC,x+j_coord_offsetD,
 777                                               &jx0,&jy0,&jz0);
 778
 779             /* Calculate displacement vector */
 780             dx00             = _mm_sub_ps(ix0,jx0);
 781             dy00             = _mm_sub_ps(iy0,jy0);
 782             dz00             = _mm_sub_ps(iz0,jz0);
 783             dx10             = _mm_sub_ps(ix1,jx0);
 784             dy10             = _mm_sub_ps(iy1,jy0);
 785             dz10             = _mm_sub_ps(iz1,jz0);
 786             dx20             = _mm_sub_ps(ix2,jx0);
 787             dy20             = _mm_sub_ps(iy2,jy0);
 788             dz20             = _mm_sub_ps(iz2,jz0);
 789
 790             /* Calculate squared distance and things based on it */
 791             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 792             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 793             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 794
 795             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 796             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 797             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 798
 799             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 800             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 801             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 802
 803             /* Load parameters for j particles */
 804             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 805                                                               charge+jnrC+0,charge+jnrD+0);
 806             vdwjidx0A        = 2*vdwtype[jnrA+0];
 807             vdwjidx0B        = 2*vdwtype[jnrB+0];
 808             vdwjidx0C        = 2*vdwtype[jnrC+0];
 809             vdwjidx0D        = 2*vdwtype[jnrD+0];
 810
 811             fjx0             = _mm_setzero_ps();
 812             fjy0             = _mm_setzero_ps();
 813             fjz0             = _mm_setzero_ps();
 814
 815             /**************************
 816              * CALCULATE INTERACTIONS *
 817              **************************/
 818
 819             r00              = _mm_mul_ps(rsq00,rinv00);
 820
 821             /* Compute parameters for interactions between i and j atoms */
 822             qq00             = _mm_mul_ps(iq0,jq0);
 823             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 824                                          vdwparam+vdwioffset0+vdwjidx0B,
 825                                          vdwparam+vdwioffset0+vdwjidx0C,
 826                                          vdwparam+vdwioffset0+vdwjidx0D,
 827                                          &c6_00,&c12_00);
 828
 829             /* Calculate table index by multiplying r with table scale and truncate to integer */
 830             rt               = _mm_mul_ps(r00,vftabscale);
 831             vfitab           = _mm_cvttps_epi32(rt);
 832             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 833             vfitab           = _mm_slli_epi32(vfitab,3);
 834
 835             /* REACTION-FIELD ELECTROSTATICS */
 836             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 837
 838             /* CUBIC SPLINE TABLE DISPERSION */
 839             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 840             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 841             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 842             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 843             _MM_TRANSPOSE4_PS(Y,F,G,H);
 844             Heps             = _mm_mul_ps(vfeps,H);
 845             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 846             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 847             fvdw6            = _mm_mul_ps(c6_00,FF);
 848
 849             /* CUBIC SPLINE TABLE REPULSION */
 850             vfitab           = _mm_add_epi32(vfitab,ifour);
 851             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 852             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 853             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 854             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 855             _MM_TRANSPOSE4_PS(Y,F,G,H);
 856             Heps             = _mm_mul_ps(vfeps,H);
 857             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 858             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 859             fvdw12           = _mm_mul_ps(c12_00,FF);
 860             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 861
 862             fscal            = _mm_add_ps(felec,fvdw);
 863
 864             /* Calculate temporary vectorial force */
 865             tx               = _mm_mul_ps(fscal,dx00);
 866             ty               = _mm_mul_ps(fscal,dy00);
 867             tz               = _mm_mul_ps(fscal,dz00);
 868
 869             /* Update vectorial force */
 870             fix0             = _mm_add_ps(fix0,tx);
 871             fiy0             = _mm_add_ps(fiy0,ty);
 872             fiz0             = _mm_add_ps(fiz0,tz);
 873
 874             fjx0             = _mm_add_ps(fjx0,tx);
 875             fjy0             = _mm_add_ps(fjy0,ty);
 876             fjz0             = _mm_add_ps(fjz0,tz);
 877
 878             /**************************
 879              * CALCULATE INTERACTIONS *
 880              **************************/
 881
 882             /* Compute parameters for interactions between i and j atoms */
 883             qq10             = _mm_mul_ps(iq1,jq0);
 884
 885             /* REACTION-FIELD ELECTROSTATICS */
 886             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 887
 888             fscal            = felec;
 889
 890             /* Calculate temporary vectorial force */
 891             tx               = _mm_mul_ps(fscal,dx10);
 892             ty               = _mm_mul_ps(fscal,dy10);
 893             tz               = _mm_mul_ps(fscal,dz10);
 894
 895             /* Update vectorial force */
 896             fix1             = _mm_add_ps(fix1,tx);
 897             fiy1             = _mm_add_ps(fiy1,ty);
 898             fiz1             = _mm_add_ps(fiz1,tz);
 899
 900             fjx0             = _mm_add_ps(fjx0,tx);
 901             fjy0             = _mm_add_ps(fjy0,ty);
 902             fjz0             = _mm_add_ps(fjz0,tz);
 903
 904             /**************************
 905              * CALCULATE INTERACTIONS *
 906              **************************/
 907
 908             /* Compute parameters for interactions between i and j atoms */
 909             qq20             = _mm_mul_ps(iq2,jq0);
 910
 911             /* REACTION-FIELD ELECTROSTATICS */
 912             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 913
 914             fscal            = felec;
 915
 916             /* Calculate temporary vectorial force */
 917             tx               = _mm_mul_ps(fscal,dx20);
 918             ty               = _mm_mul_ps(fscal,dy20);
 919             tz               = _mm_mul_ps(fscal,dz20);
 920
 921             /* Update vectorial force */
 922             fix2             = _mm_add_ps(fix2,tx);
 923             fiy2             = _mm_add_ps(fiy2,ty);
 924             fiz2             = _mm_add_ps(fiz2,tz);
 925
 926             fjx0             = _mm_add_ps(fjx0,tx);
 927             fjy0             = _mm_add_ps(fjy0,ty);
 928             fjz0             = _mm_add_ps(fjz0,tz);
 929
 930             fjptrA             = f+j_coord_offsetA;
 931             fjptrB             = f+j_coord_offsetB;
 932             fjptrC             = f+j_coord_offsetC;
 933             fjptrD             = f+j_coord_offsetD;
 934
 935             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 936
 937             /* Inner loop uses 108 flops */
 938         }
 939
 940         if(jidx<j_index_end)
 941         {
 942
 943             /* Get j neighbor index, and coordinate index */
 944             jnrlistA         = jjnr[jidx];
 945             jnrlistB         = jjnr[jidx+1];
 946             jnrlistC         = jjnr[jidx+2];
 947             jnrlistD         = jjnr[jidx+3];
 948             /* Sign of each element will be negative for non-real atoms.
 949              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 950              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 951              */
 952             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 953             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 954             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 955             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 956             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 957             j_coord_offsetA  = DIM*jnrA;
 958             j_coord_offsetB  = DIM*jnrB;
 959             j_coord_offsetC  = DIM*jnrC;
 960             j_coord_offsetD  = DIM*jnrD;
 961
 962             /* load j atom coordinates */
 963             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 964                                               x+j_coord_offsetC,x+j_coord_offsetD,
 965                                               &jx0,&jy0,&jz0);
 966
 967             /* Calculate displacement vector */
 968             dx00             = _mm_sub_ps(ix0,jx0);
 969             dy00             = _mm_sub_ps(iy0,jy0);
 970             dz00             = _mm_sub_ps(iz0,jz0);
 971             dx10             = _mm_sub_ps(ix1,jx0);
 972             dy10             = _mm_sub_ps(iy1,jy0);
 973             dz10             = _mm_sub_ps(iz1,jz0);
 974             dx20             = _mm_sub_ps(ix2,jx0);
 975             dy20             = _mm_sub_ps(iy2,jy0);
 976             dz20             = _mm_sub_ps(iz2,jz0);
 977
 978             /* Calculate squared distance and things based on it */
 979             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 980             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 981             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 982
 983             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 984             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 985             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 986
 987             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 988             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 989             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 990
 991             /* Load parameters for j particles */
 992             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 993                                                               charge+jnrC+0,charge+jnrD+0);
 994             vdwjidx0A        = 2*vdwtype[jnrA+0];
 995             vdwjidx0B        = 2*vdwtype[jnrB+0];
 996             vdwjidx0C        = 2*vdwtype[jnrC+0];
 997             vdwjidx0D        = 2*vdwtype[jnrD+0];
 998
 999             fjx0             = _mm_setzero_ps();
1000             fjy0             = _mm_setzero_ps();
1001             fjz0             = _mm_setzero_ps();
1002
1003             /**************************
1004              * CALCULATE INTERACTIONS *
1005              **************************/
1006
1007             r00              = _mm_mul_ps(rsq00,rinv00);
1008             r00              = _mm_andnot_ps(dummy_mask,r00);
1009
1010             /* Compute parameters for interactions between i and j atoms */
1011             qq00             = _mm_mul_ps(iq0,jq0);
1012             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1013                                          vdwparam+vdwioffset0+vdwjidx0B,
1014                                          vdwparam+vdwioffset0+vdwjidx0C,
1015                                          vdwparam+vdwioffset0+vdwjidx0D,
1016                                          &c6_00,&c12_00);
1017
1018             /* Calculate table index by multiplying r with table scale and truncate to integer */
1019             rt               = _mm_mul_ps(r00,vftabscale);
1020             vfitab           = _mm_cvttps_epi32(rt);
1021             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1022             vfitab           = _mm_slli_epi32(vfitab,3);
1023
1024             /* REACTION-FIELD ELECTROSTATICS */
1025             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
1026
1027             /* CUBIC SPLINE TABLE DISPERSION */
1028             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1029             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1030             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1031             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1032             _MM_TRANSPOSE4_PS(Y,F,G,H);
1033             Heps             = _mm_mul_ps(vfeps,H);
1034             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1035             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1036             fvdw6            = _mm_mul_ps(c6_00,FF);
1037
1038             /* CUBIC SPLINE TABLE REPULSION */
1039             vfitab           = _mm_add_epi32(vfitab,ifour);
1040             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1041             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1042             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1043             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1044             _MM_TRANSPOSE4_PS(Y,F,G,H);
1045             Heps             = _mm_mul_ps(vfeps,H);
1046             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1047             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1048             fvdw12           = _mm_mul_ps(c12_00,FF);
1049             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1050
1051             fscal            = _mm_add_ps(felec,fvdw);
1052
1053             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1054
1055             /* Calculate temporary vectorial force */
1056             tx               = _mm_mul_ps(fscal,dx00);
1057             ty               = _mm_mul_ps(fscal,dy00);
1058             tz               = _mm_mul_ps(fscal,dz00);
1059
1060             /* Update vectorial force */
1061             fix0             = _mm_add_ps(fix0,tx);
1062             fiy0             = _mm_add_ps(fiy0,ty);
1063             fiz0             = _mm_add_ps(fiz0,tz);
1064
1065             fjx0             = _mm_add_ps(fjx0,tx);
1066             fjy0             = _mm_add_ps(fjy0,ty);
1067             fjz0             = _mm_add_ps(fjz0,tz);
1068
1069             /**************************
1070              * CALCULATE INTERACTIONS *
1071              **************************/
1072
1073             /* Compute parameters for interactions between i and j atoms */
1074             qq10             = _mm_mul_ps(iq1,jq0);
1075
1076             /* REACTION-FIELD ELECTROSTATICS */
1077             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1078
1079             fscal            = felec;
1080
1081             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1082
1083             /* Calculate temporary vectorial force */
1084             tx               = _mm_mul_ps(fscal,dx10);
1085             ty               = _mm_mul_ps(fscal,dy10);
1086             tz               = _mm_mul_ps(fscal,dz10);
1087
1088             /* Update vectorial force */
1089             fix1             = _mm_add_ps(fix1,tx);
1090             fiy1             = _mm_add_ps(fiy1,ty);
1091             fiz1             = _mm_add_ps(fiz1,tz);
1092
1093             fjx0             = _mm_add_ps(fjx0,tx);
1094             fjy0             = _mm_add_ps(fjy0,ty);
1095             fjz0             = _mm_add_ps(fjz0,tz);
1096
1097             /**************************
1098              * CALCULATE INTERACTIONS *
1099              **************************/
1100
1101             /* Compute parameters for interactions between i and j atoms */
1102             qq20             = _mm_mul_ps(iq2,jq0);
1103
1104             /* REACTION-FIELD ELECTROSTATICS */
1105             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1106
1107             fscal            = felec;
1108
1109             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1110
1111             /* Calculate temporary vectorial force */
1112             tx               = _mm_mul_ps(fscal,dx20);
1113             ty               = _mm_mul_ps(fscal,dy20);
1114             tz               = _mm_mul_ps(fscal,dz20);
1115
1116             /* Update vectorial force */
1117             fix2             = _mm_add_ps(fix2,tx);
1118             fiy2             = _mm_add_ps(fiy2,ty);
1119             fiz2             = _mm_add_ps(fiz2,tz);
1120
1121             fjx0             = _mm_add_ps(fjx0,tx);
1122             fjy0             = _mm_add_ps(fjy0,ty);
1123             fjz0             = _mm_add_ps(fjz0,tz);
1124
1125             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1126             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1127             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1128             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1129
1130             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1131
1132             /* Inner loop uses 109 flops */
1133         }
1134
1135         /* End of innermost loop */
1136
1137         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1138                                               f+i_coord_offset,fshift+i_shift_offset);
1139
1140         /* Increment number of inner iterations */
1141         inneriter                  += j_index_end - j_index_start;
1142
1143         /* Outer loop uses 18 flops */
1144     }
1145
1146     /* Increment number of outer iterations */
1147     outeriter        += nri;
1148
1149     /* Update outer/inner flops */
1150
1151     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*109);
1152 }