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) 2012,2013,2014,2015,2017, 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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_128_fma_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_VF_avx_128_fma_single
  51  * Electrostatics interaction: Coulomb
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_VF_avx_128_fma_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  89     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 100     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 101     __m128i          vfitab;
 102     __m128i          ifour       = _mm_set1_epi32(4);
 103     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 104     real             *vftab;
 105     __m128           dummy_mask,cutoff_mask;
 106     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 107     __m128           one     = _mm_set1_ps(1.0);
 108     __m128           two     = _mm_set1_ps(2.0);
 109     x                = xx[0];
 110     f                = ff[0];
 111
 112     nri              = nlist->nri;
 113     iinr             = nlist->iinr;
 114     jindex           = nlist->jindex;
 115     jjnr             = nlist->jjnr;
 116     shiftidx         = nlist->shift;
 117     gid              = nlist->gid;
 118     shiftvec         = fr->shift_vec[0];
 119     fshift           = fr->fshift[0];
 120     facel            = _mm_set1_ps(fr->ic->epsfac);
 121     charge           = mdatoms->chargeA;
 122     nvdwtype         = fr->ntype;
 123     vdwparam         = fr->nbfp;
 124     vdwtype          = mdatoms->typeA;
 125
 126     vftab            = kernel_data->table_vdw->data;
 127     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 128
 129     /* Setup water-specific parameters */
 130     inr              = nlist->iinr[0];
 131     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 132     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 133     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 134     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 135
 136     /* Avoid stupid compiler warnings */
 137     jnrA = jnrB = jnrC = jnrD = 0;
 138     j_coord_offsetA = 0;
 139     j_coord_offsetB = 0;
 140     j_coord_offsetC = 0;
 141     j_coord_offsetD = 0;
 142
 143     outeriter        = 0;
 144     inneriter        = 0;
 145
 146     for(iidx=0;iidx<4*DIM;iidx++)
 147     {
 148         scratch[iidx] = 0.0;
 149     }
 150
 151     /* Start outer loop over neighborlists */
 152     for(iidx=0; iidx<nri; iidx++)
 153     {
 154         /* Load shift vector for this list */
 155         i_shift_offset   = DIM*shiftidx[iidx];
 156
 157         /* Load limits for loop over neighbors */
 158         j_index_start    = jindex[iidx];
 159         j_index_end      = jindex[iidx+1];
 160
 161         /* Get outer coordinate index */
 162         inr              = iinr[iidx];
 163         i_coord_offset   = DIM*inr;
 164
 165         /* Load i particle coords and add shift vector */
 166         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 167                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 168
 169         fix0             = _mm_setzero_ps();
 170         fiy0             = _mm_setzero_ps();
 171         fiz0             = _mm_setzero_ps();
 172         fix1             = _mm_setzero_ps();
 173         fiy1             = _mm_setzero_ps();
 174         fiz1             = _mm_setzero_ps();
 175         fix2             = _mm_setzero_ps();
 176         fiy2             = _mm_setzero_ps();
 177         fiz2             = _mm_setzero_ps();
 178
 179         /* Reset potential sums */
 180         velecsum         = _mm_setzero_ps();
 181         vvdwsum          = _mm_setzero_ps();
 182
 183         /* Start inner kernel loop */
 184         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 185         {
 186
 187             /* Get j neighbor index, and coordinate index */
 188             jnrA             = jjnr[jidx];
 189             jnrB             = jjnr[jidx+1];
 190             jnrC             = jjnr[jidx+2];
 191             jnrD             = jjnr[jidx+3];
 192             j_coord_offsetA  = DIM*jnrA;
 193             j_coord_offsetB  = DIM*jnrB;
 194             j_coord_offsetC  = DIM*jnrC;
 195             j_coord_offsetD  = DIM*jnrD;
 196
 197             /* load j atom coordinates */
 198             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 199                                               x+j_coord_offsetC,x+j_coord_offsetD,
 200                                               &jx0,&jy0,&jz0);
 201
 202             /* Calculate displacement vector */
 203             dx00             = _mm_sub_ps(ix0,jx0);
 204             dy00             = _mm_sub_ps(iy0,jy0);
 205             dz00             = _mm_sub_ps(iz0,jz0);
 206             dx10             = _mm_sub_ps(ix1,jx0);
 207             dy10             = _mm_sub_ps(iy1,jy0);
 208             dz10             = _mm_sub_ps(iz1,jz0);
 209             dx20             = _mm_sub_ps(ix2,jx0);
 210             dy20             = _mm_sub_ps(iy2,jy0);
 211             dz20             = _mm_sub_ps(iz2,jz0);
 212
 213             /* Calculate squared distance and things based on it */
 214             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 215             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 216             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 217
 218             rinv00           = avx128fma_invsqrt_f(rsq00);
 219             rinv10           = avx128fma_invsqrt_f(rsq10);
 220             rinv20           = avx128fma_invsqrt_f(rsq20);
 221
 222             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 223             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 224             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 225
 226             /* Load parameters for j particles */
 227             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 228                                                               charge+jnrC+0,charge+jnrD+0);
 229             vdwjidx0A        = 2*vdwtype[jnrA+0];
 230             vdwjidx0B        = 2*vdwtype[jnrB+0];
 231             vdwjidx0C        = 2*vdwtype[jnrC+0];
 232             vdwjidx0D        = 2*vdwtype[jnrD+0];
 233
 234             fjx0             = _mm_setzero_ps();
 235             fjy0             = _mm_setzero_ps();
 236             fjz0             = _mm_setzero_ps();
 237
 238             /**************************
 239              * CALCULATE INTERACTIONS *
 240              **************************/
 241
 242             r00              = _mm_mul_ps(rsq00,rinv00);
 243
 244             /* Compute parameters for interactions between i and j atoms */
 245             qq00             = _mm_mul_ps(iq0,jq0);
 246             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 247                                          vdwparam+vdwioffset0+vdwjidx0B,
 248                                          vdwparam+vdwioffset0+vdwjidx0C,
 249                                          vdwparam+vdwioffset0+vdwjidx0D,
 250                                          &c6_00,&c12_00);
 251
 252             /* Calculate table index by multiplying r with table scale and truncate to integer */
 253             rt               = _mm_mul_ps(r00,vftabscale);
 254             vfitab           = _mm_cvttps_epi32(rt);
 255 #ifdef __XOP__
 256             vfeps            = _mm_frcz_ps(rt);
 257 #else
 258             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 259 #endif
 260             twovfeps         = _mm_add_ps(vfeps,vfeps);
 261             vfitab           = _mm_slli_epi32(vfitab,3);
 262
 263             /* COULOMB ELECTROSTATICS */
 264             velec            = _mm_mul_ps(qq00,rinv00);
 265             felec            = _mm_mul_ps(velec,rinvsq00);
 266
 267             /* CUBIC SPLINE TABLE DISPERSION */
 268             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 269             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 270             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 271             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 272             _MM_TRANSPOSE4_PS(Y,F,G,H);
 273             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 274             VV               = _mm_macc_ps(vfeps,Fp,Y);
 275             vvdw6            = _mm_mul_ps(c6_00,VV);
 276             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 277             fvdw6            = _mm_mul_ps(c6_00,FF);
 278
 279             /* CUBIC SPLINE TABLE REPULSION */
 280             vfitab           = _mm_add_epi32(vfitab,ifour);
 281             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 282             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 283             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 284             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 285             _MM_TRANSPOSE4_PS(Y,F,G,H);
 286             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 287             VV               = _mm_macc_ps(vfeps,Fp,Y);
 288             vvdw12           = _mm_mul_ps(c12_00,VV);
 289             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 290             fvdw12           = _mm_mul_ps(c12_00,FF);
 291             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 292             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 293
 294             /* Update potential sum for this i atom from the interaction with this j atom. */
 295             velecsum         = _mm_add_ps(velecsum,velec);
 296             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 297
 298             fscal            = _mm_add_ps(felec,fvdw);
 299
 300              /* Update vectorial force */
 301             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 302             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 303             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 304
 305             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 306             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 307             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 308
 309             /**************************
 310              * CALCULATE INTERACTIONS *
 311              **************************/
 312
 313             /* Compute parameters for interactions between i and j atoms */
 314             qq10             = _mm_mul_ps(iq1,jq0);
 315
 316             /* COULOMB ELECTROSTATICS */
 317             velec            = _mm_mul_ps(qq10,rinv10);
 318             felec            = _mm_mul_ps(velec,rinvsq10);
 319
 320             /* Update potential sum for this i atom from the interaction with this j atom. */
 321             velecsum         = _mm_add_ps(velecsum,velec);
 322
 323             fscal            = felec;
 324
 325              /* Update vectorial force */
 326             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 327             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 328             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 329
 330             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 331             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 332             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 333
 334             /**************************
 335              * CALCULATE INTERACTIONS *
 336              **************************/
 337
 338             /* Compute parameters for interactions between i and j atoms */
 339             qq20             = _mm_mul_ps(iq2,jq0);
 340
 341             /* COULOMB ELECTROSTATICS */
 342             velec            = _mm_mul_ps(qq20,rinv20);
 343             felec            = _mm_mul_ps(velec,rinvsq20);
 344
 345             /* Update potential sum for this i atom from the interaction with this j atom. */
 346             velecsum         = _mm_add_ps(velecsum,velec);
 347
 348             fscal            = felec;
 349
 350              /* Update vectorial force */
 351             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 352             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 353             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 354
 355             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 356             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 357             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 358
 359             fjptrA             = f+j_coord_offsetA;
 360             fjptrB             = f+j_coord_offsetB;
 361             fjptrC             = f+j_coord_offsetC;
 362             fjptrD             = f+j_coord_offsetD;
 363
 364             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 365
 366             /* Inner loop uses 128 flops */
 367         }
 368
 369         if(jidx<j_index_end)
 370         {
 371
 372             /* Get j neighbor index, and coordinate index */
 373             jnrlistA         = jjnr[jidx];
 374             jnrlistB         = jjnr[jidx+1];
 375             jnrlistC         = jjnr[jidx+2];
 376             jnrlistD         = jjnr[jidx+3];
 377             /* Sign of each element will be negative for non-real atoms.
 378              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 379              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 380              */
 381             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 382             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 383             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 384             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 385             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 386             j_coord_offsetA  = DIM*jnrA;
 387             j_coord_offsetB  = DIM*jnrB;
 388             j_coord_offsetC  = DIM*jnrC;
 389             j_coord_offsetD  = DIM*jnrD;
 390
 391             /* load j atom coordinates */
 392             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 393                                               x+j_coord_offsetC,x+j_coord_offsetD,
 394                                               &jx0,&jy0,&jz0);
 395
 396             /* Calculate displacement vector */
 397             dx00             = _mm_sub_ps(ix0,jx0);
 398             dy00             = _mm_sub_ps(iy0,jy0);
 399             dz00             = _mm_sub_ps(iz0,jz0);
 400             dx10             = _mm_sub_ps(ix1,jx0);
 401             dy10             = _mm_sub_ps(iy1,jy0);
 402             dz10             = _mm_sub_ps(iz1,jz0);
 403             dx20             = _mm_sub_ps(ix2,jx0);
 404             dy20             = _mm_sub_ps(iy2,jy0);
 405             dz20             = _mm_sub_ps(iz2,jz0);
 406
 407             /* Calculate squared distance and things based on it */
 408             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 409             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 410             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 411
 412             rinv00           = avx128fma_invsqrt_f(rsq00);
 413             rinv10           = avx128fma_invsqrt_f(rsq10);
 414             rinv20           = avx128fma_invsqrt_f(rsq20);
 415
 416             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 417             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 418             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 419
 420             /* Load parameters for j particles */
 421             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 422                                                               charge+jnrC+0,charge+jnrD+0);
 423             vdwjidx0A        = 2*vdwtype[jnrA+0];
 424             vdwjidx0B        = 2*vdwtype[jnrB+0];
 425             vdwjidx0C        = 2*vdwtype[jnrC+0];
 426             vdwjidx0D        = 2*vdwtype[jnrD+0];
 427
 428             fjx0             = _mm_setzero_ps();
 429             fjy0             = _mm_setzero_ps();
 430             fjz0             = _mm_setzero_ps();
 431
 432             /**************************
 433              * CALCULATE INTERACTIONS *
 434              **************************/
 435
 436             r00              = _mm_mul_ps(rsq00,rinv00);
 437             r00              = _mm_andnot_ps(dummy_mask,r00);
 438
 439             /* Compute parameters for interactions between i and j atoms */
 440             qq00             = _mm_mul_ps(iq0,jq0);
 441             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 442                                          vdwparam+vdwioffset0+vdwjidx0B,
 443                                          vdwparam+vdwioffset0+vdwjidx0C,
 444                                          vdwparam+vdwioffset0+vdwjidx0D,
 445                                          &c6_00,&c12_00);
 446
 447             /* Calculate table index by multiplying r with table scale and truncate to integer */
 448             rt               = _mm_mul_ps(r00,vftabscale);
 449             vfitab           = _mm_cvttps_epi32(rt);
 450 #ifdef __XOP__
 451             vfeps            = _mm_frcz_ps(rt);
 452 #else
 453             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 454 #endif
 455             twovfeps         = _mm_add_ps(vfeps,vfeps);
 456             vfitab           = _mm_slli_epi32(vfitab,3);
 457
 458             /* COULOMB ELECTROSTATICS */
 459             velec            = _mm_mul_ps(qq00,rinv00);
 460             felec            = _mm_mul_ps(velec,rinvsq00);
 461
 462             /* CUBIC SPLINE TABLE DISPERSION */
 463             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 464             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 465             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 466             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 467             _MM_TRANSPOSE4_PS(Y,F,G,H);
 468             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 469             VV               = _mm_macc_ps(vfeps,Fp,Y);
 470             vvdw6            = _mm_mul_ps(c6_00,VV);
 471             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 472             fvdw6            = _mm_mul_ps(c6_00,FF);
 473
 474             /* CUBIC SPLINE TABLE REPULSION */
 475             vfitab           = _mm_add_epi32(vfitab,ifour);
 476             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 477             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 478             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 479             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 480             _MM_TRANSPOSE4_PS(Y,F,G,H);
 481             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 482             VV               = _mm_macc_ps(vfeps,Fp,Y);
 483             vvdw12           = _mm_mul_ps(c12_00,VV);
 484             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 485             fvdw12           = _mm_mul_ps(c12_00,FF);
 486             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 487             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 488
 489             /* Update potential sum for this i atom from the interaction with this j atom. */
 490             velec            = _mm_andnot_ps(dummy_mask,velec);
 491             velecsum         = _mm_add_ps(velecsum,velec);
 492             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 493             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 494
 495             fscal            = _mm_add_ps(felec,fvdw);
 496
 497             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 498
 499              /* Update vectorial force */
 500             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 501             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 502             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 503
 504             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 505             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 506             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 507
 508             /**************************
 509              * CALCULATE INTERACTIONS *
 510              **************************/
 511
 512             /* Compute parameters for interactions between i and j atoms */
 513             qq10             = _mm_mul_ps(iq1,jq0);
 514
 515             /* COULOMB ELECTROSTATICS */
 516             velec            = _mm_mul_ps(qq10,rinv10);
 517             felec            = _mm_mul_ps(velec,rinvsq10);
 518
 519             /* Update potential sum for this i atom from the interaction with this j atom. */
 520             velec            = _mm_andnot_ps(dummy_mask,velec);
 521             velecsum         = _mm_add_ps(velecsum,velec);
 522
 523             fscal            = felec;
 524
 525             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 526
 527              /* Update vectorial force */
 528             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 529             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 530             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 531
 532             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 533             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 534             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 535
 536             /**************************
 537              * CALCULATE INTERACTIONS *
 538              **************************/
 539
 540             /* Compute parameters for interactions between i and j atoms */
 541             qq20             = _mm_mul_ps(iq2,jq0);
 542
 543             /* COULOMB ELECTROSTATICS */
 544             velec            = _mm_mul_ps(qq20,rinv20);
 545             felec            = _mm_mul_ps(velec,rinvsq20);
 546
 547             /* Update potential sum for this i atom from the interaction with this j atom. */
 548             velec            = _mm_andnot_ps(dummy_mask,velec);
 549             velecsum         = _mm_add_ps(velecsum,velec);
 550
 551             fscal            = felec;
 552
 553             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 554
 555              /* Update vectorial force */
 556             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 557             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 558             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 559
 560             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 561             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 562             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 563
 564             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 565             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 566             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 567             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 568
 569             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 570
 571             /* Inner loop uses 129 flops */
 572         }
 573
 574         /* End of innermost loop */
 575
 576         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 577                                               f+i_coord_offset,fshift+i_shift_offset);
 578
 579         ggid                        = gid[iidx];
 580         /* Update potential energies */
 581         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 582         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 583
 584         /* Increment number of inner iterations */
 585         inneriter                  += j_index_end - j_index_start;
 586
 587         /* Outer loop uses 20 flops */
 588     }
 589
 590     /* Increment number of outer iterations */
 591     outeriter        += nri;
 592
 593     /* Update outer/inner flops */
 594
 595     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*129);
 596 }
 597 /*
 598  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_avx_128_fma_single
 599  * Electrostatics interaction: Coulomb
 600  * VdW interaction:            CubicSplineTable
 601  * Geometry:                   Water3-Particle
 602  * Calculate force/pot:        Force
 603  */
 604 void
 605 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_avx_128_fma_single
 606                     (t_nblist                    * gmx_restrict       nlist,
 607                      rvec                        * gmx_restrict          xx,
 608                      rvec                        * gmx_restrict          ff,
 609                      struct t_forcerec           * gmx_restrict          fr,
 610                      t_mdatoms                   * gmx_restrict     mdatoms,
 611                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 612                      t_nrnb                      * gmx_restrict        nrnb)
 613 {
 614     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 615      * just 0 for non-waters.
 616      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 617      * jnr indices corresponding to data put in the four positions in the SIMD register.
 618      */
 619     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 620     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 621     int              jnrA,jnrB,jnrC,jnrD;
 622     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 623     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 624     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 625     real             rcutoff_scalar;
 626     real             *shiftvec,*fshift,*x,*f;
 627     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 628     real             scratch[4*DIM];
 629     __m128           fscal,rcutoff,rcutoff2,jidxall;
 630     int              vdwioffset0;
 631     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 632     int              vdwioffset1;
 633     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 634     int              vdwioffset2;
 635     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 636     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 637     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 638     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 639     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 640     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 641     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 642     real             *charge;
 643     int              nvdwtype;
 644     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 645     int              *vdwtype;
 646     real             *vdwparam;
 647     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 648     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 649     __m128i          vfitab;
 650     __m128i          ifour       = _mm_set1_epi32(4);
 651     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 652     real             *vftab;
 653     __m128           dummy_mask,cutoff_mask;
 654     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 655     __m128           one     = _mm_set1_ps(1.0);
 656     __m128           two     = _mm_set1_ps(2.0);
 657     x                = xx[0];
 658     f                = ff[0];
 659
 660     nri              = nlist->nri;
 661     iinr             = nlist->iinr;
 662     jindex           = nlist->jindex;
 663     jjnr             = nlist->jjnr;
 664     shiftidx         = nlist->shift;
 665     gid              = nlist->gid;
 666     shiftvec         = fr->shift_vec[0];
 667     fshift           = fr->fshift[0];
 668     facel            = _mm_set1_ps(fr->ic->epsfac);
 669     charge           = mdatoms->chargeA;
 670     nvdwtype         = fr->ntype;
 671     vdwparam         = fr->nbfp;
 672     vdwtype          = mdatoms->typeA;
 673
 674     vftab            = kernel_data->table_vdw->data;
 675     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 676
 677     /* Setup water-specific parameters */
 678     inr              = nlist->iinr[0];
 679     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 680     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 681     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 682     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 683
 684     /* Avoid stupid compiler warnings */
 685     jnrA = jnrB = jnrC = jnrD = 0;
 686     j_coord_offsetA = 0;
 687     j_coord_offsetB = 0;
 688     j_coord_offsetC = 0;
 689     j_coord_offsetD = 0;
 690
 691     outeriter        = 0;
 692     inneriter        = 0;
 693
 694     for(iidx=0;iidx<4*DIM;iidx++)
 695     {
 696         scratch[iidx] = 0.0;
 697     }
 698
 699     /* Start outer loop over neighborlists */
 700     for(iidx=0; iidx<nri; iidx++)
 701     {
 702         /* Load shift vector for this list */
 703         i_shift_offset   = DIM*shiftidx[iidx];
 704
 705         /* Load limits for loop over neighbors */
 706         j_index_start    = jindex[iidx];
 707         j_index_end      = jindex[iidx+1];
 708
 709         /* Get outer coordinate index */
 710         inr              = iinr[iidx];
 711         i_coord_offset   = DIM*inr;
 712
 713         /* Load i particle coords and add shift vector */
 714         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 715                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 716
 717         fix0             = _mm_setzero_ps();
 718         fiy0             = _mm_setzero_ps();
 719         fiz0             = _mm_setzero_ps();
 720         fix1             = _mm_setzero_ps();
 721         fiy1             = _mm_setzero_ps();
 722         fiz1             = _mm_setzero_ps();
 723         fix2             = _mm_setzero_ps();
 724         fiy2             = _mm_setzero_ps();
 725         fiz2             = _mm_setzero_ps();
 726
 727         /* Start inner kernel loop */
 728         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 729         {
 730
 731             /* Get j neighbor index, and coordinate index */
 732             jnrA             = jjnr[jidx];
 733             jnrB             = jjnr[jidx+1];
 734             jnrC             = jjnr[jidx+2];
 735             jnrD             = jjnr[jidx+3];
 736             j_coord_offsetA  = DIM*jnrA;
 737             j_coord_offsetB  = DIM*jnrB;
 738             j_coord_offsetC  = DIM*jnrC;
 739             j_coord_offsetD  = DIM*jnrD;
 740
 741             /* load j atom coordinates */
 742             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 743                                               x+j_coord_offsetC,x+j_coord_offsetD,
 744                                               &jx0,&jy0,&jz0);
 745
 746             /* Calculate displacement vector */
 747             dx00             = _mm_sub_ps(ix0,jx0);
 748             dy00             = _mm_sub_ps(iy0,jy0);
 749             dz00             = _mm_sub_ps(iz0,jz0);
 750             dx10             = _mm_sub_ps(ix1,jx0);
 751             dy10             = _mm_sub_ps(iy1,jy0);
 752             dz10             = _mm_sub_ps(iz1,jz0);
 753             dx20             = _mm_sub_ps(ix2,jx0);
 754             dy20             = _mm_sub_ps(iy2,jy0);
 755             dz20             = _mm_sub_ps(iz2,jz0);
 756
 757             /* Calculate squared distance and things based on it */
 758             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 759             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 760             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 761
 762             rinv00           = avx128fma_invsqrt_f(rsq00);
 763             rinv10           = avx128fma_invsqrt_f(rsq10);
 764             rinv20           = avx128fma_invsqrt_f(rsq20);
 765
 766             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 767             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 768             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 769
 770             /* Load parameters for j particles */
 771             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 772                                                               charge+jnrC+0,charge+jnrD+0);
 773             vdwjidx0A        = 2*vdwtype[jnrA+0];
 774             vdwjidx0B        = 2*vdwtype[jnrB+0];
 775             vdwjidx0C        = 2*vdwtype[jnrC+0];
 776             vdwjidx0D        = 2*vdwtype[jnrD+0];
 777
 778             fjx0             = _mm_setzero_ps();
 779             fjy0             = _mm_setzero_ps();
 780             fjz0             = _mm_setzero_ps();
 781
 782             /**************************
 783              * CALCULATE INTERACTIONS *
 784              **************************/
 785
 786             r00              = _mm_mul_ps(rsq00,rinv00);
 787
 788             /* Compute parameters for interactions between i and j atoms */
 789             qq00             = _mm_mul_ps(iq0,jq0);
 790             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 791                                          vdwparam+vdwioffset0+vdwjidx0B,
 792                                          vdwparam+vdwioffset0+vdwjidx0C,
 793                                          vdwparam+vdwioffset0+vdwjidx0D,
 794                                          &c6_00,&c12_00);
 795
 796             /* Calculate table index by multiplying r with table scale and truncate to integer */
 797             rt               = _mm_mul_ps(r00,vftabscale);
 798             vfitab           = _mm_cvttps_epi32(rt);
 799 #ifdef __XOP__
 800             vfeps            = _mm_frcz_ps(rt);
 801 #else
 802             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 803 #endif
 804             twovfeps         = _mm_add_ps(vfeps,vfeps);
 805             vfitab           = _mm_slli_epi32(vfitab,3);
 806
 807             /* COULOMB ELECTROSTATICS */
 808             velec            = _mm_mul_ps(qq00,rinv00);
 809             felec            = _mm_mul_ps(velec,rinvsq00);
 810
 811             /* CUBIC SPLINE TABLE DISPERSION */
 812             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 813             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 814             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 815             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 816             _MM_TRANSPOSE4_PS(Y,F,G,H);
 817             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 818             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 819             fvdw6            = _mm_mul_ps(c6_00,FF);
 820
 821             /* CUBIC SPLINE TABLE REPULSION */
 822             vfitab           = _mm_add_epi32(vfitab,ifour);
 823             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 824             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 825             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 826             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 827             _MM_TRANSPOSE4_PS(Y,F,G,H);
 828             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 829             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 830             fvdw12           = _mm_mul_ps(c12_00,FF);
 831             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 832
 833             fscal            = _mm_add_ps(felec,fvdw);
 834
 835              /* Update vectorial force */
 836             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 837             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 838             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 839
 840             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 841             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 842             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 843
 844             /**************************
 845              * CALCULATE INTERACTIONS *
 846              **************************/
 847
 848             /* Compute parameters for interactions between i and j atoms */
 849             qq10             = _mm_mul_ps(iq1,jq0);
 850
 851             /* COULOMB ELECTROSTATICS */
 852             velec            = _mm_mul_ps(qq10,rinv10);
 853             felec            = _mm_mul_ps(velec,rinvsq10);
 854
 855             fscal            = felec;
 856
 857              /* Update vectorial force */
 858             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 859             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 860             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 861
 862             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 863             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 864             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 865
 866             /**************************
 867              * CALCULATE INTERACTIONS *
 868              **************************/
 869
 870             /* Compute parameters for interactions between i and j atoms */
 871             qq20             = _mm_mul_ps(iq2,jq0);
 872
 873             /* COULOMB ELECTROSTATICS */
 874             velec            = _mm_mul_ps(qq20,rinv20);
 875             felec            = _mm_mul_ps(velec,rinvsq20);
 876
 877             fscal            = felec;
 878
 879              /* Update vectorial force */
 880             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 881             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 882             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 883
 884             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 885             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 886             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 887
 888             fjptrA             = f+j_coord_offsetA;
 889             fjptrB             = f+j_coord_offsetB;
 890             fjptrC             = f+j_coord_offsetC;
 891             fjptrD             = f+j_coord_offsetD;
 892
 893             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 894
 895             /* Inner loop uses 117 flops */
 896         }
 897
 898         if(jidx<j_index_end)
 899         {
 900
 901             /* Get j neighbor index, and coordinate index */
 902             jnrlistA         = jjnr[jidx];
 903             jnrlistB         = jjnr[jidx+1];
 904             jnrlistC         = jjnr[jidx+2];
 905             jnrlistD         = jjnr[jidx+3];
 906             /* Sign of each element will be negative for non-real atoms.
 907              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 908              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 909              */
 910             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 911             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 912             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 913             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 914             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 915             j_coord_offsetA  = DIM*jnrA;
 916             j_coord_offsetB  = DIM*jnrB;
 917             j_coord_offsetC  = DIM*jnrC;
 918             j_coord_offsetD  = DIM*jnrD;
 919
 920             /* load j atom coordinates */
 921             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 922                                               x+j_coord_offsetC,x+j_coord_offsetD,
 923                                               &jx0,&jy0,&jz0);
 924
 925             /* Calculate displacement vector */
 926             dx00             = _mm_sub_ps(ix0,jx0);
 927             dy00             = _mm_sub_ps(iy0,jy0);
 928             dz00             = _mm_sub_ps(iz0,jz0);
 929             dx10             = _mm_sub_ps(ix1,jx0);
 930             dy10             = _mm_sub_ps(iy1,jy0);
 931             dz10             = _mm_sub_ps(iz1,jz0);
 932             dx20             = _mm_sub_ps(ix2,jx0);
 933             dy20             = _mm_sub_ps(iy2,jy0);
 934             dz20             = _mm_sub_ps(iz2,jz0);
 935
 936             /* Calculate squared distance and things based on it */
 937             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 938             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 939             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 940
 941             rinv00           = avx128fma_invsqrt_f(rsq00);
 942             rinv10           = avx128fma_invsqrt_f(rsq10);
 943             rinv20           = avx128fma_invsqrt_f(rsq20);
 944
 945             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 946             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 947             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 948
 949             /* Load parameters for j particles */
 950             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 951                                                               charge+jnrC+0,charge+jnrD+0);
 952             vdwjidx0A        = 2*vdwtype[jnrA+0];
 953             vdwjidx0B        = 2*vdwtype[jnrB+0];
 954             vdwjidx0C        = 2*vdwtype[jnrC+0];
 955             vdwjidx0D        = 2*vdwtype[jnrD+0];
 956
 957             fjx0             = _mm_setzero_ps();
 958             fjy0             = _mm_setzero_ps();
 959             fjz0             = _mm_setzero_ps();
 960
 961             /**************************
 962              * CALCULATE INTERACTIONS *
 963              **************************/
 964
 965             r00              = _mm_mul_ps(rsq00,rinv00);
 966             r00              = _mm_andnot_ps(dummy_mask,r00);
 967
 968             /* Compute parameters for interactions between i and j atoms */
 969             qq00             = _mm_mul_ps(iq0,jq0);
 970             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 971                                          vdwparam+vdwioffset0+vdwjidx0B,
 972                                          vdwparam+vdwioffset0+vdwjidx0C,
 973                                          vdwparam+vdwioffset0+vdwjidx0D,
 974                                          &c6_00,&c12_00);
 975
 976             /* Calculate table index by multiplying r with table scale and truncate to integer */
 977             rt               = _mm_mul_ps(r00,vftabscale);
 978             vfitab           = _mm_cvttps_epi32(rt);
 979 #ifdef __XOP__
 980             vfeps            = _mm_frcz_ps(rt);
 981 #else
 982             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 983 #endif
 984             twovfeps         = _mm_add_ps(vfeps,vfeps);
 985             vfitab           = _mm_slli_epi32(vfitab,3);
 986
 987             /* COULOMB ELECTROSTATICS */
 988             velec            = _mm_mul_ps(qq00,rinv00);
 989             felec            = _mm_mul_ps(velec,rinvsq00);
 990
 991             /* CUBIC SPLINE TABLE DISPERSION */
 992             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 993             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 994             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 995             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 996             _MM_TRANSPOSE4_PS(Y,F,G,H);
 997             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 998             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 999             fvdw6            = _mm_mul_ps(c6_00,FF);
1000
1001             /* CUBIC SPLINE TABLE REPULSION */
1002             vfitab           = _mm_add_epi32(vfitab,ifour);
1003             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1004             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1005             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1006             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1007             _MM_TRANSPOSE4_PS(Y,F,G,H);
1008             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1009             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1010             fvdw12           = _mm_mul_ps(c12_00,FF);
1011             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1012
1013             fscal            = _mm_add_ps(felec,fvdw);
1014
1015             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1016
1017              /* Update vectorial force */
1018             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1019             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1020             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1021
1022             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1023             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1024             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1025
1026             /**************************
1027              * CALCULATE INTERACTIONS *
1028              **************************/
1029
1030             /* Compute parameters for interactions between i and j atoms */
1031             qq10             = _mm_mul_ps(iq1,jq0);
1032
1033             /* COULOMB ELECTROSTATICS */
1034             velec            = _mm_mul_ps(qq10,rinv10);
1035             felec            = _mm_mul_ps(velec,rinvsq10);
1036
1037             fscal            = felec;
1038
1039             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1040
1041              /* Update vectorial force */
1042             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1043             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1044             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1045
1046             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1047             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1048             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1049
1050             /**************************
1051              * CALCULATE INTERACTIONS *
1052              **************************/
1053
1054             /* Compute parameters for interactions between i and j atoms */
1055             qq20             = _mm_mul_ps(iq2,jq0);
1056
1057             /* COULOMB ELECTROSTATICS */
1058             velec            = _mm_mul_ps(qq20,rinv20);
1059             felec            = _mm_mul_ps(velec,rinvsq20);
1060
1061             fscal            = felec;
1062
1063             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1064
1065              /* Update vectorial force */
1066             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1067             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1068             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1069
1070             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1071             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1072             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1073
1074             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1075             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1076             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1077             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1078
1079             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1080
1081             /* Inner loop uses 118 flops */
1082         }
1083
1084         /* End of innermost loop */
1085
1086         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1087                                               f+i_coord_offset,fshift+i_shift_offset);
1088
1089         /* Increment number of inner iterations */
1090         inneriter                  += j_index_end - j_index_start;
1091
1092         /* Outer loop uses 18 flops */
1093     }
1094
1095     /* Increment number of outer iterations */
1096     outeriter        += nri;
1097
1098     /* Update outer/inner flops */
1099
1100     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*118);
1101 }