src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecRF_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, 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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  50 #include "kernelutil_x86_avx_128_fma_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW3P1_VF_avx_128_fma_single
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_VF_avx_128_fma_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 103     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 104     __m128i          vfitab;
 105     __m128i          ifour       = _mm_set1_epi32(4);
 106     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 107     real             *vftab;
 108     __m128           dummy_mask,cutoff_mask;
 109     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 110     __m128           one     = _mm_set1_ps(1.0);
 111     __m128           two     = _mm_set1_ps(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm_set1_ps(fr->epsfac);
 124     charge           = mdatoms->chargeA;
 125     krf              = _mm_set1_ps(fr->ic->k_rf);
 126     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 127     crf              = _mm_set1_ps(fr->ic->c_rf);
 128     nvdwtype         = fr->ntype;
 129     vdwparam         = fr->nbfp;
 130     vdwtype          = mdatoms->typeA;
 131
 132     vftab            = kernel_data->table_vdw->data;
 133     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 134
 135     /* Setup water-specific parameters */
 136     inr              = nlist->iinr[0];
 137     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 138     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 139     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 140     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 141
 142     /* Avoid stupid compiler warnings */
 143     jnrA = jnrB = jnrC = jnrD = 0;
 144     j_coord_offsetA = 0;
 145     j_coord_offsetB = 0;
 146     j_coord_offsetC = 0;
 147     j_coord_offsetD = 0;
 148
 149     outeriter        = 0;
 150     inneriter        = 0;
 151
 152     for(iidx=0;iidx<4*DIM;iidx++)
 153     {
 154         scratch[iidx] = 0.0;
 155     }
 156
 157     /* Start outer loop over neighborlists */
 158     for(iidx=0; iidx<nri; iidx++)
 159     {
 160         /* Load shift vector for this list */
 161         i_shift_offset   = DIM*shiftidx[iidx];
 162
 163         /* Load limits for loop over neighbors */
 164         j_index_start    = jindex[iidx];
 165         j_index_end      = jindex[iidx+1];
 166
 167         /* Get outer coordinate index */
 168         inr              = iinr[iidx];
 169         i_coord_offset   = DIM*inr;
 170
 171         /* Load i particle coords and add shift vector */
 172         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 173                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 174
 175         fix0             = _mm_setzero_ps();
 176         fiy0             = _mm_setzero_ps();
 177         fiz0             = _mm_setzero_ps();
 178         fix1             = _mm_setzero_ps();
 179         fiy1             = _mm_setzero_ps();
 180         fiz1             = _mm_setzero_ps();
 181         fix2             = _mm_setzero_ps();
 182         fiy2             = _mm_setzero_ps();
 183         fiz2             = _mm_setzero_ps();
 184
 185         /* Reset potential sums */
 186         velecsum         = _mm_setzero_ps();
 187         vvdwsum          = _mm_setzero_ps();
 188
 189         /* Start inner kernel loop */
 190         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 191         {
 192
 193             /* Get j neighbor index, and coordinate index */
 194             jnrA             = jjnr[jidx];
 195             jnrB             = jjnr[jidx+1];
 196             jnrC             = jjnr[jidx+2];
 197             jnrD             = jjnr[jidx+3];
 198             j_coord_offsetA  = DIM*jnrA;
 199             j_coord_offsetB  = DIM*jnrB;
 200             j_coord_offsetC  = DIM*jnrC;
 201             j_coord_offsetD  = DIM*jnrD;
 202
 203             /* load j atom coordinates */
 204             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 205                                               x+j_coord_offsetC,x+j_coord_offsetD,
 206                                               &jx0,&jy0,&jz0);
 207
 208             /* Calculate displacement vector */
 209             dx00             = _mm_sub_ps(ix0,jx0);
 210             dy00             = _mm_sub_ps(iy0,jy0);
 211             dz00             = _mm_sub_ps(iz0,jz0);
 212             dx10             = _mm_sub_ps(ix1,jx0);
 213             dy10             = _mm_sub_ps(iy1,jy0);
 214             dz10             = _mm_sub_ps(iz1,jz0);
 215             dx20             = _mm_sub_ps(ix2,jx0);
 216             dy20             = _mm_sub_ps(iy2,jy0);
 217             dz20             = _mm_sub_ps(iz2,jz0);
 218
 219             /* Calculate squared distance and things based on it */
 220             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 221             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 222             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 223
 224             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 225             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 226             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 227
 228             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 229             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 230             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 231
 232             /* Load parameters for j particles */
 233             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 234                                                               charge+jnrC+0,charge+jnrD+0);
 235             vdwjidx0A        = 2*vdwtype[jnrA+0];
 236             vdwjidx0B        = 2*vdwtype[jnrB+0];
 237             vdwjidx0C        = 2*vdwtype[jnrC+0];
 238             vdwjidx0D        = 2*vdwtype[jnrD+0];
 239
 240             fjx0             = _mm_setzero_ps();
 241             fjy0             = _mm_setzero_ps();
 242             fjz0             = _mm_setzero_ps();
 243
 244             /**************************
 245              * CALCULATE INTERACTIONS *
 246              **************************/
 247
 248             r00              = _mm_mul_ps(rsq00,rinv00);
 249
 250             /* Compute parameters for interactions between i and j atoms */
 251             qq00             = _mm_mul_ps(iq0,jq0);
 252             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 253                                          vdwparam+vdwioffset0+vdwjidx0B,
 254                                          vdwparam+vdwioffset0+vdwjidx0C,
 255                                          vdwparam+vdwioffset0+vdwjidx0D,
 256                                          &c6_00,&c12_00);
 257
 258             /* Calculate table index by multiplying r with table scale and truncate to integer */
 259             rt               = _mm_mul_ps(r00,vftabscale);
 260             vfitab           = _mm_cvttps_epi32(rt);
 261 #ifdef __XOP__
 262             vfeps            = _mm_frcz_ps(rt);
 263 #else
 264             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 265 #endif
 266             twovfeps         = _mm_add_ps(vfeps,vfeps);
 267             vfitab           = _mm_slli_epi32(vfitab,3);
 268
 269             /* REACTION-FIELD ELECTROSTATICS */
 270             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 271             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 272
 273             /* CUBIC SPLINE TABLE DISPERSION */
 274             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 275             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 276             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 277             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 278             _MM_TRANSPOSE4_PS(Y,F,G,H);
 279             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 280             VV               = _mm_macc_ps(vfeps,Fp,Y);
 281             vvdw6            = _mm_mul_ps(c6_00,VV);
 282             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 283             fvdw6            = _mm_mul_ps(c6_00,FF);
 284
 285             /* CUBIC SPLINE TABLE REPULSION */
 286             vfitab           = _mm_add_epi32(vfitab,ifour);
 287             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 288             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 289             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 290             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 291             _MM_TRANSPOSE4_PS(Y,F,G,H);
 292             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 293             VV               = _mm_macc_ps(vfeps,Fp,Y);
 294             vvdw12           = _mm_mul_ps(c12_00,VV);
 295             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 296             fvdw12           = _mm_mul_ps(c12_00,FF);
 297             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 298             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 299
 300             /* Update potential sum for this i atom from the interaction with this j atom. */
 301             velecsum         = _mm_add_ps(velecsum,velec);
 302             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 303
 304             fscal            = _mm_add_ps(felec,fvdw);
 305
 306              /* Update vectorial force */
 307             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 308             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 309             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 310
 311             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 312             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 313             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 314
 315             /**************************
 316              * CALCULATE INTERACTIONS *
 317              **************************/
 318
 319             /* Compute parameters for interactions between i and j atoms */
 320             qq10             = _mm_mul_ps(iq1,jq0);
 321
 322             /* REACTION-FIELD ELECTROSTATICS */
 323             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 324             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 325
 326             /* Update potential sum for this i atom from the interaction with this j atom. */
 327             velecsum         = _mm_add_ps(velecsum,velec);
 328
 329             fscal            = felec;
 330
 331              /* Update vectorial force */
 332             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 333             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 334             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 335
 336             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 337             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 338             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 339
 340             /**************************
 341              * CALCULATE INTERACTIONS *
 342              **************************/
 343
 344             /* Compute parameters for interactions between i and j atoms */
 345             qq20             = _mm_mul_ps(iq2,jq0);
 346
 347             /* REACTION-FIELD ELECTROSTATICS */
 348             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 349             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 350
 351             /* Update potential sum for this i atom from the interaction with this j atom. */
 352             velecsum         = _mm_add_ps(velecsum,velec);
 353
 354             fscal            = felec;
 355
 356              /* Update vectorial force */
 357             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 358             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 359             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 360
 361             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 362             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 363             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 364
 365             fjptrA             = f+j_coord_offsetA;
 366             fjptrB             = f+j_coord_offsetB;
 367             fjptrC             = f+j_coord_offsetC;
 368             fjptrD             = f+j_coord_offsetD;
 369
 370             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 371
 372             /* Inner loop uses 140 flops */
 373         }
 374
 375         if(jidx<j_index_end)
 376         {
 377
 378             /* Get j neighbor index, and coordinate index */
 379             jnrlistA         = jjnr[jidx];
 380             jnrlistB         = jjnr[jidx+1];
 381             jnrlistC         = jjnr[jidx+2];
 382             jnrlistD         = jjnr[jidx+3];
 383             /* Sign of each element will be negative for non-real atoms.
 384              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 385              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 386              */
 387             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 388             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 389             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 390             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 391             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 392             j_coord_offsetA  = DIM*jnrA;
 393             j_coord_offsetB  = DIM*jnrB;
 394             j_coord_offsetC  = DIM*jnrC;
 395             j_coord_offsetD  = DIM*jnrD;
 396
 397             /* load j atom coordinates */
 398             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 399                                               x+j_coord_offsetC,x+j_coord_offsetD,
 400                                               &jx0,&jy0,&jz0);
 401
 402             /* Calculate displacement vector */
 403             dx00             = _mm_sub_ps(ix0,jx0);
 404             dy00             = _mm_sub_ps(iy0,jy0);
 405             dz00             = _mm_sub_ps(iz0,jz0);
 406             dx10             = _mm_sub_ps(ix1,jx0);
 407             dy10             = _mm_sub_ps(iy1,jy0);
 408             dz10             = _mm_sub_ps(iz1,jz0);
 409             dx20             = _mm_sub_ps(ix2,jx0);
 410             dy20             = _mm_sub_ps(iy2,jy0);
 411             dz20             = _mm_sub_ps(iz2,jz0);
 412
 413             /* Calculate squared distance and things based on it */
 414             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 415             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 416             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 417
 418             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 419             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 420             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 421
 422             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 423             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 424             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 425
 426             /* Load parameters for j particles */
 427             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 428                                                               charge+jnrC+0,charge+jnrD+0);
 429             vdwjidx0A        = 2*vdwtype[jnrA+0];
 430             vdwjidx0B        = 2*vdwtype[jnrB+0];
 431             vdwjidx0C        = 2*vdwtype[jnrC+0];
 432             vdwjidx0D        = 2*vdwtype[jnrD+0];
 433
 434             fjx0             = _mm_setzero_ps();
 435             fjy0             = _mm_setzero_ps();
 436             fjz0             = _mm_setzero_ps();
 437
 438             /**************************
 439              * CALCULATE INTERACTIONS *
 440              **************************/
 441
 442             r00              = _mm_mul_ps(rsq00,rinv00);
 443             r00              = _mm_andnot_ps(dummy_mask,r00);
 444
 445             /* Compute parameters for interactions between i and j atoms */
 446             qq00             = _mm_mul_ps(iq0,jq0);
 447             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 448                                          vdwparam+vdwioffset0+vdwjidx0B,
 449                                          vdwparam+vdwioffset0+vdwjidx0C,
 450                                          vdwparam+vdwioffset0+vdwjidx0D,
 451                                          &c6_00,&c12_00);
 452
 453             /* Calculate table index by multiplying r with table scale and truncate to integer */
 454             rt               = _mm_mul_ps(r00,vftabscale);
 455             vfitab           = _mm_cvttps_epi32(rt);
 456 #ifdef __XOP__
 457             vfeps            = _mm_frcz_ps(rt);
 458 #else
 459             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 460 #endif
 461             twovfeps         = _mm_add_ps(vfeps,vfeps);
 462             vfitab           = _mm_slli_epi32(vfitab,3);
 463
 464             /* REACTION-FIELD ELECTROSTATICS */
 465             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 466             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 467
 468             /* CUBIC SPLINE TABLE DISPERSION */
 469             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 470             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 471             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 472             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 473             _MM_TRANSPOSE4_PS(Y,F,G,H);
 474             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 475             VV               = _mm_macc_ps(vfeps,Fp,Y);
 476             vvdw6            = _mm_mul_ps(c6_00,VV);
 477             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 478             fvdw6            = _mm_mul_ps(c6_00,FF);
 479
 480             /* CUBIC SPLINE TABLE REPULSION */
 481             vfitab           = _mm_add_epi32(vfitab,ifour);
 482             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 483             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 484             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 485             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 486             _MM_TRANSPOSE4_PS(Y,F,G,H);
 487             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 488             VV               = _mm_macc_ps(vfeps,Fp,Y);
 489             vvdw12           = _mm_mul_ps(c12_00,VV);
 490             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 491             fvdw12           = _mm_mul_ps(c12_00,FF);
 492             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 493             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 494
 495             /* Update potential sum for this i atom from the interaction with this j atom. */
 496             velec            = _mm_andnot_ps(dummy_mask,velec);
 497             velecsum         = _mm_add_ps(velecsum,velec);
 498             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 499             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 500
 501             fscal            = _mm_add_ps(felec,fvdw);
 502
 503             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 504
 505              /* Update vectorial force */
 506             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 507             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 508             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 509
 510             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 511             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 512             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 513
 514             /**************************
 515              * CALCULATE INTERACTIONS *
 516              **************************/
 517
 518             /* Compute parameters for interactions between i and j atoms */
 519             qq10             = _mm_mul_ps(iq1,jq0);
 520
 521             /* REACTION-FIELD ELECTROSTATICS */
 522             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 523             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 524
 525             /* Update potential sum for this i atom from the interaction with this j atom. */
 526             velec            = _mm_andnot_ps(dummy_mask,velec);
 527             velecsum         = _mm_add_ps(velecsum,velec);
 528
 529             fscal            = felec;
 530
 531             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 532
 533              /* Update vectorial force */
 534             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 535             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 536             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 537
 538             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 539             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 540             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 541
 542             /**************************
 543              * CALCULATE INTERACTIONS *
 544              **************************/
 545
 546             /* Compute parameters for interactions between i and j atoms */
 547             qq20             = _mm_mul_ps(iq2,jq0);
 548
 549             /* REACTION-FIELD ELECTROSTATICS */
 550             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 551             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 552
 553             /* Update potential sum for this i atom from the interaction with this j atom. */
 554             velec            = _mm_andnot_ps(dummy_mask,velec);
 555             velecsum         = _mm_add_ps(velecsum,velec);
 556
 557             fscal            = felec;
 558
 559             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 560
 561              /* Update vectorial force */
 562             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 563             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 564             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 565
 566             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 567             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 568             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 569
 570             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 571             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 572             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 573             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 574
 575             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 576
 577             /* Inner loop uses 141 flops */
 578         }
 579
 580         /* End of innermost loop */
 581
 582         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 583                                               f+i_coord_offset,fshift+i_shift_offset);
 584
 585         ggid                        = gid[iidx];
 586         /* Update potential energies */
 587         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 588         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 589
 590         /* Increment number of inner iterations */
 591         inneriter                  += j_index_end - j_index_start;
 592
 593         /* Outer loop uses 20 flops */
 594     }
 595
 596     /* Increment number of outer iterations */
 597     outeriter        += nri;
 598
 599     /* Update outer/inner flops */
 600
 601     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*141);
 602 }
 603 /*
 604  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_avx_128_fma_single
 605  * Electrostatics interaction: ReactionField
 606  * VdW interaction:            CubicSplineTable
 607  * Geometry:                   Water3-Particle
 608  * Calculate force/pot:        Force
 609  */
 610 void
 611 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_avx_128_fma_single
 612                     (t_nblist                    * gmx_restrict       nlist,
 613                      rvec                        * gmx_restrict          xx,
 614                      rvec                        * gmx_restrict          ff,
 615                      t_forcerec                  * gmx_restrict          fr,
 616                      t_mdatoms                   * gmx_restrict     mdatoms,
 617                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 618                      t_nrnb                      * gmx_restrict        nrnb)
 619 {
 620     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 621      * just 0 for non-waters.
 622      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 623      * jnr indices corresponding to data put in the four positions in the SIMD register.
 624      */
 625     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 626     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 627     int              jnrA,jnrB,jnrC,jnrD;
 628     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 629     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 630     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 631     real             rcutoff_scalar;
 632     real             *shiftvec,*fshift,*x,*f;
 633     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 634     real             scratch[4*DIM];
 635     __m128           fscal,rcutoff,rcutoff2,jidxall;
 636     int              vdwioffset0;
 637     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 638     int              vdwioffset1;
 639     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 640     int              vdwioffset2;
 641     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 642     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 643     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 644     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 645     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 646     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 647     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 648     real             *charge;
 649     int              nvdwtype;
 650     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 651     int              *vdwtype;
 652     real             *vdwparam;
 653     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 654     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 655     __m128i          vfitab;
 656     __m128i          ifour       = _mm_set1_epi32(4);
 657     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 658     real             *vftab;
 659     __m128           dummy_mask,cutoff_mask;
 660     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 661     __m128           one     = _mm_set1_ps(1.0);
 662     __m128           two     = _mm_set1_ps(2.0);
 663     x                = xx[0];
 664     f                = ff[0];
 665
 666     nri              = nlist->nri;
 667     iinr             = nlist->iinr;
 668     jindex           = nlist->jindex;
 669     jjnr             = nlist->jjnr;
 670     shiftidx         = nlist->shift;
 671     gid              = nlist->gid;
 672     shiftvec         = fr->shift_vec[0];
 673     fshift           = fr->fshift[0];
 674     facel            = _mm_set1_ps(fr->epsfac);
 675     charge           = mdatoms->chargeA;
 676     krf              = _mm_set1_ps(fr->ic->k_rf);
 677     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 678     crf              = _mm_set1_ps(fr->ic->c_rf);
 679     nvdwtype         = fr->ntype;
 680     vdwparam         = fr->nbfp;
 681     vdwtype          = mdatoms->typeA;
 682
 683     vftab            = kernel_data->table_vdw->data;
 684     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 685
 686     /* Setup water-specific parameters */
 687     inr              = nlist->iinr[0];
 688     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 689     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 690     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 691     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 692
 693     /* Avoid stupid compiler warnings */
 694     jnrA = jnrB = jnrC = jnrD = 0;
 695     j_coord_offsetA = 0;
 696     j_coord_offsetB = 0;
 697     j_coord_offsetC = 0;
 698     j_coord_offsetD = 0;
 699
 700     outeriter        = 0;
 701     inneriter        = 0;
 702
 703     for(iidx=0;iidx<4*DIM;iidx++)
 704     {
 705         scratch[iidx] = 0.0;
 706     }
 707
 708     /* Start outer loop over neighborlists */
 709     for(iidx=0; iidx<nri; iidx++)
 710     {
 711         /* Load shift vector for this list */
 712         i_shift_offset   = DIM*shiftidx[iidx];
 713
 714         /* Load limits for loop over neighbors */
 715         j_index_start    = jindex[iidx];
 716         j_index_end      = jindex[iidx+1];
 717
 718         /* Get outer coordinate index */
 719         inr              = iinr[iidx];
 720         i_coord_offset   = DIM*inr;
 721
 722         /* Load i particle coords and add shift vector */
 723         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 724                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 725
 726         fix0             = _mm_setzero_ps();
 727         fiy0             = _mm_setzero_ps();
 728         fiz0             = _mm_setzero_ps();
 729         fix1             = _mm_setzero_ps();
 730         fiy1             = _mm_setzero_ps();
 731         fiz1             = _mm_setzero_ps();
 732         fix2             = _mm_setzero_ps();
 733         fiy2             = _mm_setzero_ps();
 734         fiz2             = _mm_setzero_ps();
 735
 736         /* Start inner kernel loop */
 737         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 738         {
 739
 740             /* Get j neighbor index, and coordinate index */
 741             jnrA             = jjnr[jidx];
 742             jnrB             = jjnr[jidx+1];
 743             jnrC             = jjnr[jidx+2];
 744             jnrD             = jjnr[jidx+3];
 745             j_coord_offsetA  = DIM*jnrA;
 746             j_coord_offsetB  = DIM*jnrB;
 747             j_coord_offsetC  = DIM*jnrC;
 748             j_coord_offsetD  = DIM*jnrD;
 749
 750             /* load j atom coordinates */
 751             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 752                                               x+j_coord_offsetC,x+j_coord_offsetD,
 753                                               &jx0,&jy0,&jz0);
 754
 755             /* Calculate displacement vector */
 756             dx00             = _mm_sub_ps(ix0,jx0);
 757             dy00             = _mm_sub_ps(iy0,jy0);
 758             dz00             = _mm_sub_ps(iz0,jz0);
 759             dx10             = _mm_sub_ps(ix1,jx0);
 760             dy10             = _mm_sub_ps(iy1,jy0);
 761             dz10             = _mm_sub_ps(iz1,jz0);
 762             dx20             = _mm_sub_ps(ix2,jx0);
 763             dy20             = _mm_sub_ps(iy2,jy0);
 764             dz20             = _mm_sub_ps(iz2,jz0);
 765
 766             /* Calculate squared distance and things based on it */
 767             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 768             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 769             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 770
 771             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 772             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 773             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 774
 775             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 776             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 777             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 778
 779             /* Load parameters for j particles */
 780             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 781                                                               charge+jnrC+0,charge+jnrD+0);
 782             vdwjidx0A        = 2*vdwtype[jnrA+0];
 783             vdwjidx0B        = 2*vdwtype[jnrB+0];
 784             vdwjidx0C        = 2*vdwtype[jnrC+0];
 785             vdwjidx0D        = 2*vdwtype[jnrD+0];
 786
 787             fjx0             = _mm_setzero_ps();
 788             fjy0             = _mm_setzero_ps();
 789             fjz0             = _mm_setzero_ps();
 790
 791             /**************************
 792              * CALCULATE INTERACTIONS *
 793              **************************/
 794
 795             r00              = _mm_mul_ps(rsq00,rinv00);
 796
 797             /* Compute parameters for interactions between i and j atoms */
 798             qq00             = _mm_mul_ps(iq0,jq0);
 799             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 800                                          vdwparam+vdwioffset0+vdwjidx0B,
 801                                          vdwparam+vdwioffset0+vdwjidx0C,
 802                                          vdwparam+vdwioffset0+vdwjidx0D,
 803                                          &c6_00,&c12_00);
 804
 805             /* Calculate table index by multiplying r with table scale and truncate to integer */
 806             rt               = _mm_mul_ps(r00,vftabscale);
 807             vfitab           = _mm_cvttps_epi32(rt);
 808 #ifdef __XOP__
 809             vfeps            = _mm_frcz_ps(rt);
 810 #else
 811             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 812 #endif
 813             twovfeps         = _mm_add_ps(vfeps,vfeps);
 814             vfitab           = _mm_slli_epi32(vfitab,3);
 815
 816             /* REACTION-FIELD ELECTROSTATICS */
 817             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 818
 819             /* CUBIC SPLINE TABLE DISPERSION */
 820             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 821             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 822             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 823             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 824             _MM_TRANSPOSE4_PS(Y,F,G,H);
 825             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 826             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 827             fvdw6            = _mm_mul_ps(c6_00,FF);
 828
 829             /* CUBIC SPLINE TABLE REPULSION */
 830             vfitab           = _mm_add_epi32(vfitab,ifour);
 831             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 832             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 833             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 834             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 835             _MM_TRANSPOSE4_PS(Y,F,G,H);
 836             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 837             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 838             fvdw12           = _mm_mul_ps(c12_00,FF);
 839             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 840
 841             fscal            = _mm_add_ps(felec,fvdw);
 842
 843              /* Update vectorial force */
 844             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 845             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 846             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 847
 848             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 849             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 850             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 851
 852             /**************************
 853              * CALCULATE INTERACTIONS *
 854              **************************/
 855
 856             /* Compute parameters for interactions between i and j atoms */
 857             qq10             = _mm_mul_ps(iq1,jq0);
 858
 859             /* REACTION-FIELD ELECTROSTATICS */
 860             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 861
 862             fscal            = felec;
 863
 864              /* Update vectorial force */
 865             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 866             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 867             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 868
 869             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 870             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 871             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 872
 873             /**************************
 874              * CALCULATE INTERACTIONS *
 875              **************************/
 876
 877             /* Compute parameters for interactions between i and j atoms */
 878             qq20             = _mm_mul_ps(iq2,jq0);
 879
 880             /* REACTION-FIELD ELECTROSTATICS */
 881             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 882
 883             fscal            = felec;
 884
 885              /* Update vectorial force */
 886             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 887             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 888             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 889
 890             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 891             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 892             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 893
 894             fjptrA             = f+j_coord_offsetA;
 895             fjptrB             = f+j_coord_offsetB;
 896             fjptrC             = f+j_coord_offsetC;
 897             fjptrD             = f+j_coord_offsetD;
 898
 899             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 900
 901             /* Inner loop uses 117 flops */
 902         }
 903
 904         if(jidx<j_index_end)
 905         {
 906
 907             /* Get j neighbor index, and coordinate index */
 908             jnrlistA         = jjnr[jidx];
 909             jnrlistB         = jjnr[jidx+1];
 910             jnrlistC         = jjnr[jidx+2];
 911             jnrlistD         = jjnr[jidx+3];
 912             /* Sign of each element will be negative for non-real atoms.
 913              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 914              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 915              */
 916             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 917             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 918             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 919             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 920             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 921             j_coord_offsetA  = DIM*jnrA;
 922             j_coord_offsetB  = DIM*jnrB;
 923             j_coord_offsetC  = DIM*jnrC;
 924             j_coord_offsetD  = DIM*jnrD;
 925
 926             /* load j atom coordinates */
 927             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 928                                               x+j_coord_offsetC,x+j_coord_offsetD,
 929                                               &jx0,&jy0,&jz0);
 930
 931             /* Calculate displacement vector */
 932             dx00             = _mm_sub_ps(ix0,jx0);
 933             dy00             = _mm_sub_ps(iy0,jy0);
 934             dz00             = _mm_sub_ps(iz0,jz0);
 935             dx10             = _mm_sub_ps(ix1,jx0);
 936             dy10             = _mm_sub_ps(iy1,jy0);
 937             dz10             = _mm_sub_ps(iz1,jz0);
 938             dx20             = _mm_sub_ps(ix2,jx0);
 939             dy20             = _mm_sub_ps(iy2,jy0);
 940             dz20             = _mm_sub_ps(iz2,jz0);
 941
 942             /* Calculate squared distance and things based on it */
 943             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 944             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 945             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 946
 947             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 948             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 949             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 950
 951             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 952             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 953             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 954
 955             /* Load parameters for j particles */
 956             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 957                                                               charge+jnrC+0,charge+jnrD+0);
 958             vdwjidx0A        = 2*vdwtype[jnrA+0];
 959             vdwjidx0B        = 2*vdwtype[jnrB+0];
 960             vdwjidx0C        = 2*vdwtype[jnrC+0];
 961             vdwjidx0D        = 2*vdwtype[jnrD+0];
 962
 963             fjx0             = _mm_setzero_ps();
 964             fjy0             = _mm_setzero_ps();
 965             fjz0             = _mm_setzero_ps();
 966
 967             /**************************
 968              * CALCULATE INTERACTIONS *
 969              **************************/
 970
 971             r00              = _mm_mul_ps(rsq00,rinv00);
 972             r00              = _mm_andnot_ps(dummy_mask,r00);
 973
 974             /* Compute parameters for interactions between i and j atoms */
 975             qq00             = _mm_mul_ps(iq0,jq0);
 976             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 977                                          vdwparam+vdwioffset0+vdwjidx0B,
 978                                          vdwparam+vdwioffset0+vdwjidx0C,
 979                                          vdwparam+vdwioffset0+vdwjidx0D,
 980                                          &c6_00,&c12_00);
 981
 982             /* Calculate table index by multiplying r with table scale and truncate to integer */
 983             rt               = _mm_mul_ps(r00,vftabscale);
 984             vfitab           = _mm_cvttps_epi32(rt);
 985 #ifdef __XOP__
 986             vfeps            = _mm_frcz_ps(rt);
 987 #else
 988             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 989 #endif
 990             twovfeps         = _mm_add_ps(vfeps,vfeps);
 991             vfitab           = _mm_slli_epi32(vfitab,3);
 992
 993             /* REACTION-FIELD ELECTROSTATICS */
 994             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 995
 996             /* CUBIC SPLINE TABLE DISPERSION */
 997             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 998             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 999             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1000             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1001             _MM_TRANSPOSE4_PS(Y,F,G,H);
1002             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1003             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1004             fvdw6            = _mm_mul_ps(c6_00,FF);
1005
1006             /* CUBIC SPLINE TABLE REPULSION */
1007             vfitab           = _mm_add_epi32(vfitab,ifour);
1008             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1009             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1010             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1011             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1012             _MM_TRANSPOSE4_PS(Y,F,G,H);
1013             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1014             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1015             fvdw12           = _mm_mul_ps(c12_00,FF);
1016             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1017
1018             fscal            = _mm_add_ps(felec,fvdw);
1019
1020             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1021
1022              /* Update vectorial force */
1023             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1024             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1025             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1026
1027             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1028             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1029             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1030
1031             /**************************
1032              * CALCULATE INTERACTIONS *
1033              **************************/
1034
1035             /* Compute parameters for interactions between i and j atoms */
1036             qq10             = _mm_mul_ps(iq1,jq0);
1037
1038             /* REACTION-FIELD ELECTROSTATICS */
1039             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
1040
1041             fscal            = felec;
1042
1043             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1044
1045              /* Update vectorial force */
1046             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1047             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1048             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1049
1050             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1051             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1052             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1053
1054             /**************************
1055              * CALCULATE INTERACTIONS *
1056              **************************/
1057
1058             /* Compute parameters for interactions between i and j atoms */
1059             qq20             = _mm_mul_ps(iq2,jq0);
1060
1061             /* REACTION-FIELD ELECTROSTATICS */
1062             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1063
1064             fscal            = felec;
1065
1066             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1067
1068              /* Update vectorial force */
1069             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1070             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1071             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1072
1073             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1074             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1075             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1076
1077             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1078             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1079             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1080             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1081
1082             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1083
1084             /* Inner loop uses 118 flops */
1085         }
1086
1087         /* End of innermost loop */
1088
1089         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1090                                               f+i_coord_offset,fshift+i_shift_offset);
1091
1092         /* Increment number of inner iterations */
1093         inneriter                  += j_index_end - j_index_start;
1094
1095         /* Outer loop uses 18 flops */
1096     }
1097
1098     /* Increment number of outer iterations */
1099     outeriter        += nri;
1100
1101     /* Update outer/inner flops */
1102
1103     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*118);
1104 }