src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_avx_128_fma_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 avx_128_fma_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  48 #include "kernelutil_x86_avx_128_fma_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_VF_avx_128_fma_single
  52  * Electrostatics interaction: Coulomb
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_VF_avx_128_fma_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwioffset1;
  86     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     int              vdwioffset2;
  88     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  92     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  93     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  94     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     int              nvdwtype;
  97     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 101     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 102     __m128i          vfitab;
 103     __m128i          ifour       = _mm_set1_epi32(4);
 104     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 105     real             *vftab;
 106     __m128           dummy_mask,cutoff_mask;
 107     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 108     __m128           one     = _mm_set1_ps(1.0);
 109     __m128           two     = _mm_set1_ps(2.0);
 110     x                = xx[0];
 111     f                = ff[0];
 112
 113     nri              = nlist->nri;
 114     iinr             = nlist->iinr;
 115     jindex           = nlist->jindex;
 116     jjnr             = nlist->jjnr;
 117     shiftidx         = nlist->shift;
 118     gid              = nlist->gid;
 119     shiftvec         = fr->shift_vec[0];
 120     fshift           = fr->fshift[0];
 121     facel            = _mm_set1_ps(fr->epsfac);
 122     charge           = mdatoms->chargeA;
 123     nvdwtype         = fr->ntype;
 124     vdwparam         = fr->nbfp;
 125     vdwtype          = mdatoms->typeA;
 126
 127     vftab            = kernel_data->table_vdw->data;
 128     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 129
 130     /* Setup water-specific parameters */
 131     inr              = nlist->iinr[0];
 132     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 133     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 134     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 135     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 136
 137     /* Avoid stupid compiler warnings */
 138     jnrA = jnrB = jnrC = jnrD = 0;
 139     j_coord_offsetA = 0;
 140     j_coord_offsetB = 0;
 141     j_coord_offsetC = 0;
 142     j_coord_offsetD = 0;
 143
 144     outeriter        = 0;
 145     inneriter        = 0;
 146
 147     for(iidx=0;iidx<4*DIM;iidx++)
 148     {
 149         scratch[iidx] = 0.0;
 150     }
 151
 152     /* Start outer loop over neighborlists */
 153     for(iidx=0; iidx<nri; iidx++)
 154     {
 155         /* Load shift vector for this list */
 156         i_shift_offset   = DIM*shiftidx[iidx];
 157
 158         /* Load limits for loop over neighbors */
 159         j_index_start    = jindex[iidx];
 160         j_index_end      = jindex[iidx+1];
 161
 162         /* Get outer coordinate index */
 163         inr              = iinr[iidx];
 164         i_coord_offset   = DIM*inr;
 165
 166         /* Load i particle coords and add shift vector */
 167         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 168                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 169
 170         fix0             = _mm_setzero_ps();
 171         fiy0             = _mm_setzero_ps();
 172         fiz0             = _mm_setzero_ps();
 173         fix1             = _mm_setzero_ps();
 174         fiy1             = _mm_setzero_ps();
 175         fiz1             = _mm_setzero_ps();
 176         fix2             = _mm_setzero_ps();
 177         fiy2             = _mm_setzero_ps();
 178         fiz2             = _mm_setzero_ps();
 179
 180         /* Reset potential sums */
 181         velecsum         = _mm_setzero_ps();
 182         vvdwsum          = _mm_setzero_ps();
 183
 184         /* Start inner kernel loop */
 185         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 186         {
 187
 188             /* Get j neighbor index, and coordinate index */
 189             jnrA             = jjnr[jidx];
 190             jnrB             = jjnr[jidx+1];
 191             jnrC             = jjnr[jidx+2];
 192             jnrD             = jjnr[jidx+3];
 193             j_coord_offsetA  = DIM*jnrA;
 194             j_coord_offsetB  = DIM*jnrB;
 195             j_coord_offsetC  = DIM*jnrC;
 196             j_coord_offsetD  = DIM*jnrD;
 197
 198             /* load j atom coordinates */
 199             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 200                                               x+j_coord_offsetC,x+j_coord_offsetD,
 201                                               &jx0,&jy0,&jz0);
 202
 203             /* Calculate displacement vector */
 204             dx00             = _mm_sub_ps(ix0,jx0);
 205             dy00             = _mm_sub_ps(iy0,jy0);
 206             dz00             = _mm_sub_ps(iz0,jz0);
 207             dx10             = _mm_sub_ps(ix1,jx0);
 208             dy10             = _mm_sub_ps(iy1,jy0);
 209             dz10             = _mm_sub_ps(iz1,jz0);
 210             dx20             = _mm_sub_ps(ix2,jx0);
 211             dy20             = _mm_sub_ps(iy2,jy0);
 212             dz20             = _mm_sub_ps(iz2,jz0);
 213
 214             /* Calculate squared distance and things based on it */
 215             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 216             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 217             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 218
 219             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 220             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 221             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 222
 223             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 224             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 225             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 226
 227             /* Load parameters for j particles */
 228             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 229                                                               charge+jnrC+0,charge+jnrD+0);
 230             vdwjidx0A        = 2*vdwtype[jnrA+0];
 231             vdwjidx0B        = 2*vdwtype[jnrB+0];
 232             vdwjidx0C        = 2*vdwtype[jnrC+0];
 233             vdwjidx0D        = 2*vdwtype[jnrD+0];
 234
 235             fjx0             = _mm_setzero_ps();
 236             fjy0             = _mm_setzero_ps();
 237             fjz0             = _mm_setzero_ps();
 238
 239             /**************************
 240              * CALCULATE INTERACTIONS *
 241              **************************/
 242
 243             r00              = _mm_mul_ps(rsq00,rinv00);
 244
 245             /* Compute parameters for interactions between i and j atoms */
 246             qq00             = _mm_mul_ps(iq0,jq0);
 247             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 248                                          vdwparam+vdwioffset0+vdwjidx0B,
 249                                          vdwparam+vdwioffset0+vdwjidx0C,
 250                                          vdwparam+vdwioffset0+vdwjidx0D,
 251                                          &c6_00,&c12_00);
 252
 253             /* Calculate table index by multiplying r with table scale and truncate to integer */
 254             rt               = _mm_mul_ps(r00,vftabscale);
 255             vfitab           = _mm_cvttps_epi32(rt);
 256 #ifdef __XOP__
 257             vfeps            = _mm_frcz_ps(rt);
 258 #else
 259             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 260 #endif
 261             twovfeps         = _mm_add_ps(vfeps,vfeps);
 262             vfitab           = _mm_slli_epi32(vfitab,3);
 263
 264             /* COULOMB ELECTROSTATICS */
 265             velec            = _mm_mul_ps(qq00,rinv00);
 266             felec            = _mm_mul_ps(velec,rinvsq00);
 267
 268             /* CUBIC SPLINE TABLE DISPERSION */
 269             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 270             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 271             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 272             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 273             _MM_TRANSPOSE4_PS(Y,F,G,H);
 274             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 275             VV               = _mm_macc_ps(vfeps,Fp,Y);
 276             vvdw6            = _mm_mul_ps(c6_00,VV);
 277             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 278             fvdw6            = _mm_mul_ps(c6_00,FF);
 279
 280             /* CUBIC SPLINE TABLE REPULSION */
 281             vfitab           = _mm_add_epi32(vfitab,ifour);
 282             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 283             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 284             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 285             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 286             _MM_TRANSPOSE4_PS(Y,F,G,H);
 287             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 288             VV               = _mm_macc_ps(vfeps,Fp,Y);
 289             vvdw12           = _mm_mul_ps(c12_00,VV);
 290             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 291             fvdw12           = _mm_mul_ps(c12_00,FF);
 292             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 293             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 294
 295             /* Update potential sum for this i atom from the interaction with this j atom. */
 296             velecsum         = _mm_add_ps(velecsum,velec);
 297             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 298
 299             fscal            = _mm_add_ps(felec,fvdw);
 300
 301              /* Update vectorial force */
 302             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 303             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 304             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 305
 306             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 307             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 308             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 309
 310             /**************************
 311              * CALCULATE INTERACTIONS *
 312              **************************/
 313
 314             /* Compute parameters for interactions between i and j atoms */
 315             qq10             = _mm_mul_ps(iq1,jq0);
 316
 317             /* COULOMB ELECTROSTATICS */
 318             velec            = _mm_mul_ps(qq10,rinv10);
 319             felec            = _mm_mul_ps(velec,rinvsq10);
 320
 321             /* Update potential sum for this i atom from the interaction with this j atom. */
 322             velecsum         = _mm_add_ps(velecsum,velec);
 323
 324             fscal            = felec;
 325
 326              /* Update vectorial force */
 327             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 328             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 329             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 330
 331             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 332             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 333             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 334
 335             /**************************
 336              * CALCULATE INTERACTIONS *
 337              **************************/
 338
 339             /* Compute parameters for interactions between i and j atoms */
 340             qq20             = _mm_mul_ps(iq2,jq0);
 341
 342             /* COULOMB ELECTROSTATICS */
 343             velec            = _mm_mul_ps(qq20,rinv20);
 344             felec            = _mm_mul_ps(velec,rinvsq20);
 345
 346             /* Update potential sum for this i atom from the interaction with this j atom. */
 347             velecsum         = _mm_add_ps(velecsum,velec);
 348
 349             fscal            = felec;
 350
 351              /* Update vectorial force */
 352             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 353             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 354             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 355
 356             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 357             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 358             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 359
 360             fjptrA             = f+j_coord_offsetA;
 361             fjptrB             = f+j_coord_offsetB;
 362             fjptrC             = f+j_coord_offsetC;
 363             fjptrD             = f+j_coord_offsetD;
 364
 365             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 366
 367             /* Inner loop uses 128 flops */
 368         }
 369
 370         if(jidx<j_index_end)
 371         {
 372
 373             /* Get j neighbor index, and coordinate index */
 374             jnrlistA         = jjnr[jidx];
 375             jnrlistB         = jjnr[jidx+1];
 376             jnrlistC         = jjnr[jidx+2];
 377             jnrlistD         = jjnr[jidx+3];
 378             /* Sign of each element will be negative for non-real atoms.
 379              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 380              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 381              */
 382             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 383             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 384             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 385             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 386             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 387             j_coord_offsetA  = DIM*jnrA;
 388             j_coord_offsetB  = DIM*jnrB;
 389             j_coord_offsetC  = DIM*jnrC;
 390             j_coord_offsetD  = DIM*jnrD;
 391
 392             /* load j atom coordinates */
 393             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 394                                               x+j_coord_offsetC,x+j_coord_offsetD,
 395                                               &jx0,&jy0,&jz0);
 396
 397             /* Calculate displacement vector */
 398             dx00             = _mm_sub_ps(ix0,jx0);
 399             dy00             = _mm_sub_ps(iy0,jy0);
 400             dz00             = _mm_sub_ps(iz0,jz0);
 401             dx10             = _mm_sub_ps(ix1,jx0);
 402             dy10             = _mm_sub_ps(iy1,jy0);
 403             dz10             = _mm_sub_ps(iz1,jz0);
 404             dx20             = _mm_sub_ps(ix2,jx0);
 405             dy20             = _mm_sub_ps(iy2,jy0);
 406             dz20             = _mm_sub_ps(iz2,jz0);
 407
 408             /* Calculate squared distance and things based on it */
 409             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 410             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 411             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 412
 413             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 414             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 415             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 416
 417             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 418             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 419             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 420
 421             /* Load parameters for j particles */
 422             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 423                                                               charge+jnrC+0,charge+jnrD+0);
 424             vdwjidx0A        = 2*vdwtype[jnrA+0];
 425             vdwjidx0B        = 2*vdwtype[jnrB+0];
 426             vdwjidx0C        = 2*vdwtype[jnrC+0];
 427             vdwjidx0D        = 2*vdwtype[jnrD+0];
 428
 429             fjx0             = _mm_setzero_ps();
 430             fjy0             = _mm_setzero_ps();
 431             fjz0             = _mm_setzero_ps();
 432
 433             /**************************
 434              * CALCULATE INTERACTIONS *
 435              **************************/
 436
 437             r00              = _mm_mul_ps(rsq00,rinv00);
 438             r00              = _mm_andnot_ps(dummy_mask,r00);
 439
 440             /* Compute parameters for interactions between i and j atoms */
 441             qq00             = _mm_mul_ps(iq0,jq0);
 442             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 443                                          vdwparam+vdwioffset0+vdwjidx0B,
 444                                          vdwparam+vdwioffset0+vdwjidx0C,
 445                                          vdwparam+vdwioffset0+vdwjidx0D,
 446                                          &c6_00,&c12_00);
 447
 448             /* Calculate table index by multiplying r with table scale and truncate to integer */
 449             rt               = _mm_mul_ps(r00,vftabscale);
 450             vfitab           = _mm_cvttps_epi32(rt);
 451 #ifdef __XOP__
 452             vfeps            = _mm_frcz_ps(rt);
 453 #else
 454             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 455 #endif
 456             twovfeps         = _mm_add_ps(vfeps,vfeps);
 457             vfitab           = _mm_slli_epi32(vfitab,3);
 458
 459             /* COULOMB ELECTROSTATICS */
 460             velec            = _mm_mul_ps(qq00,rinv00);
 461             felec            = _mm_mul_ps(velec,rinvsq00);
 462
 463             /* CUBIC SPLINE TABLE DISPERSION */
 464             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 465             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 466             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 467             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 468             _MM_TRANSPOSE4_PS(Y,F,G,H);
 469             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 470             VV               = _mm_macc_ps(vfeps,Fp,Y);
 471             vvdw6            = _mm_mul_ps(c6_00,VV);
 472             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 473             fvdw6            = _mm_mul_ps(c6_00,FF);
 474
 475             /* CUBIC SPLINE TABLE REPULSION */
 476             vfitab           = _mm_add_epi32(vfitab,ifour);
 477             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 478             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 479             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 480             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 481             _MM_TRANSPOSE4_PS(Y,F,G,H);
 482             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 483             VV               = _mm_macc_ps(vfeps,Fp,Y);
 484             vvdw12           = _mm_mul_ps(c12_00,VV);
 485             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 486             fvdw12           = _mm_mul_ps(c12_00,FF);
 487             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 488             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 489
 490             /* Update potential sum for this i atom from the interaction with this j atom. */
 491             velec            = _mm_andnot_ps(dummy_mask,velec);
 492             velecsum         = _mm_add_ps(velecsum,velec);
 493             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 494             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 495
 496             fscal            = _mm_add_ps(felec,fvdw);
 497
 498             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 499
 500              /* Update vectorial force */
 501             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 502             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 503             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 504
 505             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 506             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 507             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 508
 509             /**************************
 510              * CALCULATE INTERACTIONS *
 511              **************************/
 512
 513             /* Compute parameters for interactions between i and j atoms */
 514             qq10             = _mm_mul_ps(iq1,jq0);
 515
 516             /* COULOMB ELECTROSTATICS */
 517             velec            = _mm_mul_ps(qq10,rinv10);
 518             felec            = _mm_mul_ps(velec,rinvsq10);
 519
 520             /* Update potential sum for this i atom from the interaction with this j atom. */
 521             velec            = _mm_andnot_ps(dummy_mask,velec);
 522             velecsum         = _mm_add_ps(velecsum,velec);
 523
 524             fscal            = felec;
 525
 526             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 527
 528              /* Update vectorial force */
 529             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 530             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 531             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 532
 533             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 534             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 535             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 536
 537             /**************************
 538              * CALCULATE INTERACTIONS *
 539              **************************/
 540
 541             /* Compute parameters for interactions between i and j atoms */
 542             qq20             = _mm_mul_ps(iq2,jq0);
 543
 544             /* COULOMB ELECTROSTATICS */
 545             velec            = _mm_mul_ps(qq20,rinv20);
 546             felec            = _mm_mul_ps(velec,rinvsq20);
 547
 548             /* Update potential sum for this i atom from the interaction with this j atom. */
 549             velec            = _mm_andnot_ps(dummy_mask,velec);
 550             velecsum         = _mm_add_ps(velecsum,velec);
 551
 552             fscal            = felec;
 553
 554             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 555
 556              /* Update vectorial force */
 557             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 558             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 559             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 560
 561             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 562             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 563             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 564
 565             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 566             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 567             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 568             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 569
 570             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 571
 572             /* Inner loop uses 129 flops */
 573         }
 574
 575         /* End of innermost loop */
 576
 577         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 578                                               f+i_coord_offset,fshift+i_shift_offset);
 579
 580         ggid                        = gid[iidx];
 581         /* Update potential energies */
 582         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 583         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 584
 585         /* Increment number of inner iterations */
 586         inneriter                  += j_index_end - j_index_start;
 587
 588         /* Outer loop uses 20 flops */
 589     }
 590
 591     /* Increment number of outer iterations */
 592     outeriter        += nri;
 593
 594     /* Update outer/inner flops */
 595
 596     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*129);
 597 }
 598 /*
 599  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_avx_128_fma_single
 600  * Electrostatics interaction: Coulomb
 601  * VdW interaction:            CubicSplineTable
 602  * Geometry:                   Water3-Particle
 603  * Calculate force/pot:        Force
 604  */
 605 void
 606 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_avx_128_fma_single
 607                     (t_nblist                    * gmx_restrict       nlist,
 608                      rvec                        * gmx_restrict          xx,
 609                      rvec                        * gmx_restrict          ff,
 610                      t_forcerec                  * gmx_restrict          fr,
 611                      t_mdatoms                   * gmx_restrict     mdatoms,
 612                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 613                      t_nrnb                      * gmx_restrict        nrnb)
 614 {
 615     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 616      * just 0 for non-waters.
 617      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 618      * jnr indices corresponding to data put in the four positions in the SIMD register.
 619      */
 620     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 621     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 622     int              jnrA,jnrB,jnrC,jnrD;
 623     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 624     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 625     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 626     real             rcutoff_scalar;
 627     real             *shiftvec,*fshift,*x,*f;
 628     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 629     real             scratch[4*DIM];
 630     __m128           fscal,rcutoff,rcutoff2,jidxall;
 631     int              vdwioffset0;
 632     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 633     int              vdwioffset1;
 634     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 635     int              vdwioffset2;
 636     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 637     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 638     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 639     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 640     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 641     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 642     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 643     real             *charge;
 644     int              nvdwtype;
 645     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 646     int              *vdwtype;
 647     real             *vdwparam;
 648     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 649     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 650     __m128i          vfitab;
 651     __m128i          ifour       = _mm_set1_epi32(4);
 652     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 653     real             *vftab;
 654     __m128           dummy_mask,cutoff_mask;
 655     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 656     __m128           one     = _mm_set1_ps(1.0);
 657     __m128           two     = _mm_set1_ps(2.0);
 658     x                = xx[0];
 659     f                = ff[0];
 660
 661     nri              = nlist->nri;
 662     iinr             = nlist->iinr;
 663     jindex           = nlist->jindex;
 664     jjnr             = nlist->jjnr;
 665     shiftidx         = nlist->shift;
 666     gid              = nlist->gid;
 667     shiftvec         = fr->shift_vec[0];
 668     fshift           = fr->fshift[0];
 669     facel            = _mm_set1_ps(fr->epsfac);
 670     charge           = mdatoms->chargeA;
 671     nvdwtype         = fr->ntype;
 672     vdwparam         = fr->nbfp;
 673     vdwtype          = mdatoms->typeA;
 674
 675     vftab            = kernel_data->table_vdw->data;
 676     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 677
 678     /* Setup water-specific parameters */
 679     inr              = nlist->iinr[0];
 680     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 681     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 682     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 683     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 684
 685     /* Avoid stupid compiler warnings */
 686     jnrA = jnrB = jnrC = jnrD = 0;
 687     j_coord_offsetA = 0;
 688     j_coord_offsetB = 0;
 689     j_coord_offsetC = 0;
 690     j_coord_offsetD = 0;
 691
 692     outeriter        = 0;
 693     inneriter        = 0;
 694
 695     for(iidx=0;iidx<4*DIM;iidx++)
 696     {
 697         scratch[iidx] = 0.0;
 698     }
 699
 700     /* Start outer loop over neighborlists */
 701     for(iidx=0; iidx<nri; iidx++)
 702     {
 703         /* Load shift vector for this list */
 704         i_shift_offset   = DIM*shiftidx[iidx];
 705
 706         /* Load limits for loop over neighbors */
 707         j_index_start    = jindex[iidx];
 708         j_index_end      = jindex[iidx+1];
 709
 710         /* Get outer coordinate index */
 711         inr              = iinr[iidx];
 712         i_coord_offset   = DIM*inr;
 713
 714         /* Load i particle coords and add shift vector */
 715         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 716                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 717
 718         fix0             = _mm_setzero_ps();
 719         fiy0             = _mm_setzero_ps();
 720         fiz0             = _mm_setzero_ps();
 721         fix1             = _mm_setzero_ps();
 722         fiy1             = _mm_setzero_ps();
 723         fiz1             = _mm_setzero_ps();
 724         fix2             = _mm_setzero_ps();
 725         fiy2             = _mm_setzero_ps();
 726         fiz2             = _mm_setzero_ps();
 727
 728         /* Start inner kernel loop */
 729         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 730         {
 731
 732             /* Get j neighbor index, and coordinate index */
 733             jnrA             = jjnr[jidx];
 734             jnrB             = jjnr[jidx+1];
 735             jnrC             = jjnr[jidx+2];
 736             jnrD             = jjnr[jidx+3];
 737             j_coord_offsetA  = DIM*jnrA;
 738             j_coord_offsetB  = DIM*jnrB;
 739             j_coord_offsetC  = DIM*jnrC;
 740             j_coord_offsetD  = DIM*jnrD;
 741
 742             /* load j atom coordinates */
 743             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 744                                               x+j_coord_offsetC,x+j_coord_offsetD,
 745                                               &jx0,&jy0,&jz0);
 746
 747             /* Calculate displacement vector */
 748             dx00             = _mm_sub_ps(ix0,jx0);
 749             dy00             = _mm_sub_ps(iy0,jy0);
 750             dz00             = _mm_sub_ps(iz0,jz0);
 751             dx10             = _mm_sub_ps(ix1,jx0);
 752             dy10             = _mm_sub_ps(iy1,jy0);
 753             dz10             = _mm_sub_ps(iz1,jz0);
 754             dx20             = _mm_sub_ps(ix2,jx0);
 755             dy20             = _mm_sub_ps(iy2,jy0);
 756             dz20             = _mm_sub_ps(iz2,jz0);
 757
 758             /* Calculate squared distance and things based on it */
 759             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 760             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 761             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 762
 763             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 764             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 765             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 766
 767             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 768             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 769             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 770
 771             /* Load parameters for j particles */
 772             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 773                                                               charge+jnrC+0,charge+jnrD+0);
 774             vdwjidx0A        = 2*vdwtype[jnrA+0];
 775             vdwjidx0B        = 2*vdwtype[jnrB+0];
 776             vdwjidx0C        = 2*vdwtype[jnrC+0];
 777             vdwjidx0D        = 2*vdwtype[jnrD+0];
 778
 779             fjx0             = _mm_setzero_ps();
 780             fjy0             = _mm_setzero_ps();
 781             fjz0             = _mm_setzero_ps();
 782
 783             /**************************
 784              * CALCULATE INTERACTIONS *
 785              **************************/
 786
 787             r00              = _mm_mul_ps(rsq00,rinv00);
 788
 789             /* Compute parameters for interactions between i and j atoms */
 790             qq00             = _mm_mul_ps(iq0,jq0);
 791             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 792                                          vdwparam+vdwioffset0+vdwjidx0B,
 793                                          vdwparam+vdwioffset0+vdwjidx0C,
 794                                          vdwparam+vdwioffset0+vdwjidx0D,
 795                                          &c6_00,&c12_00);
 796
 797             /* Calculate table index by multiplying r with table scale and truncate to integer */
 798             rt               = _mm_mul_ps(r00,vftabscale);
 799             vfitab           = _mm_cvttps_epi32(rt);
 800 #ifdef __XOP__
 801             vfeps            = _mm_frcz_ps(rt);
 802 #else
 803             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 804 #endif
 805             twovfeps         = _mm_add_ps(vfeps,vfeps);
 806             vfitab           = _mm_slli_epi32(vfitab,3);
 807
 808             /* COULOMB ELECTROSTATICS */
 809             velec            = _mm_mul_ps(qq00,rinv00);
 810             felec            = _mm_mul_ps(velec,rinvsq00);
 811
 812             /* CUBIC SPLINE TABLE DISPERSION */
 813             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 814             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 815             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 816             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 817             _MM_TRANSPOSE4_PS(Y,F,G,H);
 818             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 819             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 820             fvdw6            = _mm_mul_ps(c6_00,FF);
 821
 822             /* CUBIC SPLINE TABLE REPULSION */
 823             vfitab           = _mm_add_epi32(vfitab,ifour);
 824             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 825             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 826             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 827             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 828             _MM_TRANSPOSE4_PS(Y,F,G,H);
 829             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 830             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 831             fvdw12           = _mm_mul_ps(c12_00,FF);
 832             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 833
 834             fscal            = _mm_add_ps(felec,fvdw);
 835
 836              /* Update vectorial force */
 837             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 838             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 839             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 840
 841             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 842             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 843             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 844
 845             /**************************
 846              * CALCULATE INTERACTIONS *
 847              **************************/
 848
 849             /* Compute parameters for interactions between i and j atoms */
 850             qq10             = _mm_mul_ps(iq1,jq0);
 851
 852             /* COULOMB ELECTROSTATICS */
 853             velec            = _mm_mul_ps(qq10,rinv10);
 854             felec            = _mm_mul_ps(velec,rinvsq10);
 855
 856             fscal            = felec;
 857
 858              /* Update vectorial force */
 859             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 860             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 861             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 862
 863             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 864             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 865             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 866
 867             /**************************
 868              * CALCULATE INTERACTIONS *
 869              **************************/
 870
 871             /* Compute parameters for interactions between i and j atoms */
 872             qq20             = _mm_mul_ps(iq2,jq0);
 873
 874             /* COULOMB ELECTROSTATICS */
 875             velec            = _mm_mul_ps(qq20,rinv20);
 876             felec            = _mm_mul_ps(velec,rinvsq20);
 877
 878             fscal            = felec;
 879
 880              /* Update vectorial force */
 881             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 882             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 883             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 884
 885             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 886             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 887             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 888
 889             fjptrA             = f+j_coord_offsetA;
 890             fjptrB             = f+j_coord_offsetB;
 891             fjptrC             = f+j_coord_offsetC;
 892             fjptrD             = f+j_coord_offsetD;
 893
 894             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 895
 896             /* Inner loop uses 117 flops */
 897         }
 898
 899         if(jidx<j_index_end)
 900         {
 901
 902             /* Get j neighbor index, and coordinate index */
 903             jnrlistA         = jjnr[jidx];
 904             jnrlistB         = jjnr[jidx+1];
 905             jnrlistC         = jjnr[jidx+2];
 906             jnrlistD         = jjnr[jidx+3];
 907             /* Sign of each element will be negative for non-real atoms.
 908              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 909              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 910              */
 911             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 912             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 913             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 914             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 915             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 916             j_coord_offsetA  = DIM*jnrA;
 917             j_coord_offsetB  = DIM*jnrB;
 918             j_coord_offsetC  = DIM*jnrC;
 919             j_coord_offsetD  = DIM*jnrD;
 920
 921             /* load j atom coordinates */
 922             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 923                                               x+j_coord_offsetC,x+j_coord_offsetD,
 924                                               &jx0,&jy0,&jz0);
 925
 926             /* Calculate displacement vector */
 927             dx00             = _mm_sub_ps(ix0,jx0);
 928             dy00             = _mm_sub_ps(iy0,jy0);
 929             dz00             = _mm_sub_ps(iz0,jz0);
 930             dx10             = _mm_sub_ps(ix1,jx0);
 931             dy10             = _mm_sub_ps(iy1,jy0);
 932             dz10             = _mm_sub_ps(iz1,jz0);
 933             dx20             = _mm_sub_ps(ix2,jx0);
 934             dy20             = _mm_sub_ps(iy2,jy0);
 935             dz20             = _mm_sub_ps(iz2,jz0);
 936
 937             /* Calculate squared distance and things based on it */
 938             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 939             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 940             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 941
 942             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 943             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 944             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 945
 946             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 947             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 948             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 949
 950             /* Load parameters for j particles */
 951             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 952                                                               charge+jnrC+0,charge+jnrD+0);
 953             vdwjidx0A        = 2*vdwtype[jnrA+0];
 954             vdwjidx0B        = 2*vdwtype[jnrB+0];
 955             vdwjidx0C        = 2*vdwtype[jnrC+0];
 956             vdwjidx0D        = 2*vdwtype[jnrD+0];
 957
 958             fjx0             = _mm_setzero_ps();
 959             fjy0             = _mm_setzero_ps();
 960             fjz0             = _mm_setzero_ps();
 961
 962             /**************************
 963              * CALCULATE INTERACTIONS *
 964              **************************/
 965
 966             r00              = _mm_mul_ps(rsq00,rinv00);
 967             r00              = _mm_andnot_ps(dummy_mask,r00);
 968
 969             /* Compute parameters for interactions between i and j atoms */
 970             qq00             = _mm_mul_ps(iq0,jq0);
 971             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 972                                          vdwparam+vdwioffset0+vdwjidx0B,
 973                                          vdwparam+vdwioffset0+vdwjidx0C,
 974                                          vdwparam+vdwioffset0+vdwjidx0D,
 975                                          &c6_00,&c12_00);
 976
 977             /* Calculate table index by multiplying r with table scale and truncate to integer */
 978             rt               = _mm_mul_ps(r00,vftabscale);
 979             vfitab           = _mm_cvttps_epi32(rt);
 980 #ifdef __XOP__
 981             vfeps            = _mm_frcz_ps(rt);
 982 #else
 983             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 984 #endif
 985             twovfeps         = _mm_add_ps(vfeps,vfeps);
 986             vfitab           = _mm_slli_epi32(vfitab,3);
 987
 988             /* COULOMB ELECTROSTATICS */
 989             velec            = _mm_mul_ps(qq00,rinv00);
 990             felec            = _mm_mul_ps(velec,rinvsq00);
 991
 992             /* CUBIC SPLINE TABLE DISPERSION */
 993             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 994             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 995             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 996             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 997             _MM_TRANSPOSE4_PS(Y,F,G,H);
 998             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 999             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1000             fvdw6            = _mm_mul_ps(c6_00,FF);
1001
1002             /* CUBIC SPLINE TABLE REPULSION */
1003             vfitab           = _mm_add_epi32(vfitab,ifour);
1004             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1005             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1006             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1007             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1008             _MM_TRANSPOSE4_PS(Y,F,G,H);
1009             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1010             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1011             fvdw12           = _mm_mul_ps(c12_00,FF);
1012             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1013
1014             fscal            = _mm_add_ps(felec,fvdw);
1015
1016             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1017
1018              /* Update vectorial force */
1019             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1020             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1021             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1022
1023             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1024             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1025             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1026
1027             /**************************
1028              * CALCULATE INTERACTIONS *
1029              **************************/
1030
1031             /* Compute parameters for interactions between i and j atoms */
1032             qq10             = _mm_mul_ps(iq1,jq0);
1033
1034             /* COULOMB ELECTROSTATICS */
1035             velec            = _mm_mul_ps(qq10,rinv10);
1036             felec            = _mm_mul_ps(velec,rinvsq10);
1037
1038             fscal            = felec;
1039
1040             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1041
1042              /* Update vectorial force */
1043             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1044             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1045             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1046
1047             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1048             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1049             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1050
1051             /**************************
1052              * CALCULATE INTERACTIONS *
1053              **************************/
1054
1055             /* Compute parameters for interactions between i and j atoms */
1056             qq20             = _mm_mul_ps(iq2,jq0);
1057
1058             /* COULOMB ELECTROSTATICS */
1059             velec            = _mm_mul_ps(qq20,rinv20);
1060             felec            = _mm_mul_ps(velec,rinvsq20);
1061
1062             fscal            = felec;
1063
1064             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1065
1066              /* Update vectorial force */
1067             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1068             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1069             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1070
1071             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1072             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1073             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1074
1075             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1076             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1077             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1078             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1079
1080             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1081
1082             /* Inner loop uses 118 flops */
1083         }
1084
1085         /* End of innermost loop */
1086
1087         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1088                                               f+i_coord_offset,fshift+i_shift_offset);
1089
1090         /* Increment number of inner iterations */
1091         inneriter                  += j_index_end - j_index_start;
1092
1093         /* Outer loop uses 18 flops */
1094     }
1095
1096     /* Increment number of outer iterations */
1097     outeriter        += nri;
1098
1099     /* Update outer/inner flops */
1100
1101     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*118);
1102 }