src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecRF_VdwCSTab_GeomW3P1_avx_128_fma_single.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS 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_ElecRF_VdwCSTab_GeomW3P1_VF_avx_128_fma_single
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRF_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     krf              = _mm_set1_ps(fr->ic->k_rf);
 123     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 124     crf              = _mm_set1_ps(fr->ic->c_rf);
 125     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128
 129     vftab            = kernel_data->table_vdw->data;
 130     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 131
 132     /* Setup water-specific parameters */
 133     inr              = nlist->iinr[0];
 134     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 135     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 136     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 137     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 138
 139     /* Avoid stupid compiler warnings */
 140     jnrA = jnrB = jnrC = jnrD = 0;
 141     j_coord_offsetA = 0;
 142     j_coord_offsetB = 0;
 143     j_coord_offsetC = 0;
 144     j_coord_offsetD = 0;
 145
 146     outeriter        = 0;
 147     inneriter        = 0;
 148
 149     for(iidx=0;iidx<4*DIM;iidx++)
 150     {
 151         scratch[iidx] = 0.0;
 152     }
 153
 154     /* Start outer loop over neighborlists */
 155     for(iidx=0; iidx<nri; iidx++)
 156     {
 157         /* Load shift vector for this list */
 158         i_shift_offset   = DIM*shiftidx[iidx];
 159
 160         /* Load limits for loop over neighbors */
 161         j_index_start    = jindex[iidx];
 162         j_index_end      = jindex[iidx+1];
 163
 164         /* Get outer coordinate index */
 165         inr              = iinr[iidx];
 166         i_coord_offset   = DIM*inr;
 167
 168         /* Load i particle coords and add shift vector */
 169         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 170                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 171
 172         fix0             = _mm_setzero_ps();
 173         fiy0             = _mm_setzero_ps();
 174         fiz0             = _mm_setzero_ps();
 175         fix1             = _mm_setzero_ps();
 176         fiy1             = _mm_setzero_ps();
 177         fiz1             = _mm_setzero_ps();
 178         fix2             = _mm_setzero_ps();
 179         fiy2             = _mm_setzero_ps();
 180         fiz2             = _mm_setzero_ps();
 181
 182         /* Reset potential sums */
 183         velecsum         = _mm_setzero_ps();
 184         vvdwsum          = _mm_setzero_ps();
 185
 186         /* Start inner kernel loop */
 187         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 188         {
 189
 190             /* Get j neighbor index, and coordinate index */
 191             jnrA             = jjnr[jidx];
 192             jnrB             = jjnr[jidx+1];
 193             jnrC             = jjnr[jidx+2];
 194             jnrD             = jjnr[jidx+3];
 195             j_coord_offsetA  = DIM*jnrA;
 196             j_coord_offsetB  = DIM*jnrB;
 197             j_coord_offsetC  = DIM*jnrC;
 198             j_coord_offsetD  = DIM*jnrD;
 199
 200             /* load j atom coordinates */
 201             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 202                                               x+j_coord_offsetC,x+j_coord_offsetD,
 203                                               &jx0,&jy0,&jz0);
 204
 205             /* Calculate displacement vector */
 206             dx00             = _mm_sub_ps(ix0,jx0);
 207             dy00             = _mm_sub_ps(iy0,jy0);
 208             dz00             = _mm_sub_ps(iz0,jz0);
 209             dx10             = _mm_sub_ps(ix1,jx0);
 210             dy10             = _mm_sub_ps(iy1,jy0);
 211             dz10             = _mm_sub_ps(iz1,jz0);
 212             dx20             = _mm_sub_ps(ix2,jx0);
 213             dy20             = _mm_sub_ps(iy2,jy0);
 214             dz20             = _mm_sub_ps(iz2,jz0);
 215
 216             /* Calculate squared distance and things based on it */
 217             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 218             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 219             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 220
 221             rinv00           = avx128fma_invsqrt_f(rsq00);
 222             rinv10           = avx128fma_invsqrt_f(rsq10);
 223             rinv20           = avx128fma_invsqrt_f(rsq20);
 224
 225             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 226             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 227             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 228
 229             /* Load parameters for j particles */
 230             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 231                                                               charge+jnrC+0,charge+jnrD+0);
 232             vdwjidx0A        = 2*vdwtype[jnrA+0];
 233             vdwjidx0B        = 2*vdwtype[jnrB+0];
 234             vdwjidx0C        = 2*vdwtype[jnrC+0];
 235             vdwjidx0D        = 2*vdwtype[jnrD+0];
 236
 237             fjx0             = _mm_setzero_ps();
 238             fjy0             = _mm_setzero_ps();
 239             fjz0             = _mm_setzero_ps();
 240
 241             /**************************
 242              * CALCULATE INTERACTIONS *
 243              **************************/
 244
 245             r00              = _mm_mul_ps(rsq00,rinv00);
 246
 247             /* Compute parameters for interactions between i and j atoms */
 248             qq00             = _mm_mul_ps(iq0,jq0);
 249             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 250                                          vdwparam+vdwioffset0+vdwjidx0B,
 251                                          vdwparam+vdwioffset0+vdwjidx0C,
 252                                          vdwparam+vdwioffset0+vdwjidx0D,
 253                                          &c6_00,&c12_00);
 254
 255             /* Calculate table index by multiplying r with table scale and truncate to integer */
 256             rt               = _mm_mul_ps(r00,vftabscale);
 257             vfitab           = _mm_cvttps_epi32(rt);
 258 #ifdef __XOP__
 259             vfeps            = _mm_frcz_ps(rt);
 260 #else
 261             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 262 #endif
 263             twovfeps         = _mm_add_ps(vfeps,vfeps);
 264             vfitab           = _mm_slli_epi32(vfitab,3);
 265
 266             /* REACTION-FIELD ELECTROSTATICS */
 267             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 268             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 269
 270             /* CUBIC SPLINE TABLE DISPERSION */
 271             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 272             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 273             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 274             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 275             _MM_TRANSPOSE4_PS(Y,F,G,H);
 276             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 277             VV               = _mm_macc_ps(vfeps,Fp,Y);
 278             vvdw6            = _mm_mul_ps(c6_00,VV);
 279             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 280             fvdw6            = _mm_mul_ps(c6_00,FF);
 281
 282             /* CUBIC SPLINE TABLE REPULSION */
 283             vfitab           = _mm_add_epi32(vfitab,ifour);
 284             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 285             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 286             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 287             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 288             _MM_TRANSPOSE4_PS(Y,F,G,H);
 289             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 290             VV               = _mm_macc_ps(vfeps,Fp,Y);
 291             vvdw12           = _mm_mul_ps(c12_00,VV);
 292             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 293             fvdw12           = _mm_mul_ps(c12_00,FF);
 294             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 295             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 296
 297             /* Update potential sum for this i atom from the interaction with this j atom. */
 298             velecsum         = _mm_add_ps(velecsum,velec);
 299             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 300
 301             fscal            = _mm_add_ps(felec,fvdw);
 302
 303              /* Update vectorial force */
 304             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 305             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 306             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 307
 308             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 309             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 310             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 311
 312             /**************************
 313              * CALCULATE INTERACTIONS *
 314              **************************/
 315
 316             /* Compute parameters for interactions between i and j atoms */
 317             qq10             = _mm_mul_ps(iq1,jq0);
 318
 319             /* REACTION-FIELD ELECTROSTATICS */
 320             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 321             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 322
 323             /* Update potential sum for this i atom from the interaction with this j atom. */
 324             velecsum         = _mm_add_ps(velecsum,velec);
 325
 326             fscal            = felec;
 327
 328              /* Update vectorial force */
 329             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 330             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 331             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 332
 333             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 334             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 335             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 336
 337             /**************************
 338              * CALCULATE INTERACTIONS *
 339              **************************/
 340
 341             /* Compute parameters for interactions between i and j atoms */
 342             qq20             = _mm_mul_ps(iq2,jq0);
 343
 344             /* REACTION-FIELD ELECTROSTATICS */
 345             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 346             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 347
 348             /* Update potential sum for this i atom from the interaction with this j atom. */
 349             velecsum         = _mm_add_ps(velecsum,velec);
 350
 351             fscal            = felec;
 352
 353              /* Update vectorial force */
 354             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 355             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 356             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 357
 358             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 359             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 360             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 361
 362             fjptrA             = f+j_coord_offsetA;
 363             fjptrB             = f+j_coord_offsetB;
 364             fjptrC             = f+j_coord_offsetC;
 365             fjptrD             = f+j_coord_offsetD;
 366
 367             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 368
 369             /* Inner loop uses 140 flops */
 370         }
 371
 372         if(jidx<j_index_end)
 373         {
 374
 375             /* Get j neighbor index, and coordinate index */
 376             jnrlistA         = jjnr[jidx];
 377             jnrlistB         = jjnr[jidx+1];
 378             jnrlistC         = jjnr[jidx+2];
 379             jnrlistD         = jjnr[jidx+3];
 380             /* Sign of each element will be negative for non-real atoms.
 381              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 382              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 383              */
 384             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 385             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 386             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 387             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 388             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 389             j_coord_offsetA  = DIM*jnrA;
 390             j_coord_offsetB  = DIM*jnrB;
 391             j_coord_offsetC  = DIM*jnrC;
 392             j_coord_offsetD  = DIM*jnrD;
 393
 394             /* load j atom coordinates */
 395             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 396                                               x+j_coord_offsetC,x+j_coord_offsetD,
 397                                               &jx0,&jy0,&jz0);
 398
 399             /* Calculate displacement vector */
 400             dx00             = _mm_sub_ps(ix0,jx0);
 401             dy00             = _mm_sub_ps(iy0,jy0);
 402             dz00             = _mm_sub_ps(iz0,jz0);
 403             dx10             = _mm_sub_ps(ix1,jx0);
 404             dy10             = _mm_sub_ps(iy1,jy0);
 405             dz10             = _mm_sub_ps(iz1,jz0);
 406             dx20             = _mm_sub_ps(ix2,jx0);
 407             dy20             = _mm_sub_ps(iy2,jy0);
 408             dz20             = _mm_sub_ps(iz2,jz0);
 409
 410             /* Calculate squared distance and things based on it */
 411             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 412             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 413             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 414
 415             rinv00           = avx128fma_invsqrt_f(rsq00);
 416             rinv10           = avx128fma_invsqrt_f(rsq10);
 417             rinv20           = avx128fma_invsqrt_f(rsq20);
 418
 419             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 420             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 421             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 422
 423             /* Load parameters for j particles */
 424             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 425                                                               charge+jnrC+0,charge+jnrD+0);
 426             vdwjidx0A        = 2*vdwtype[jnrA+0];
 427             vdwjidx0B        = 2*vdwtype[jnrB+0];
 428             vdwjidx0C        = 2*vdwtype[jnrC+0];
 429             vdwjidx0D        = 2*vdwtype[jnrD+0];
 430
 431             fjx0             = _mm_setzero_ps();
 432             fjy0             = _mm_setzero_ps();
 433             fjz0             = _mm_setzero_ps();
 434
 435             /**************************
 436              * CALCULATE INTERACTIONS *
 437              **************************/
 438
 439             r00              = _mm_mul_ps(rsq00,rinv00);
 440             r00              = _mm_andnot_ps(dummy_mask,r00);
 441
 442             /* Compute parameters for interactions between i and j atoms */
 443             qq00             = _mm_mul_ps(iq0,jq0);
 444             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 445                                          vdwparam+vdwioffset0+vdwjidx0B,
 446                                          vdwparam+vdwioffset0+vdwjidx0C,
 447                                          vdwparam+vdwioffset0+vdwjidx0D,
 448                                          &c6_00,&c12_00);
 449
 450             /* Calculate table index by multiplying r with table scale and truncate to integer */
 451             rt               = _mm_mul_ps(r00,vftabscale);
 452             vfitab           = _mm_cvttps_epi32(rt);
 453 #ifdef __XOP__
 454             vfeps            = _mm_frcz_ps(rt);
 455 #else
 456             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 457 #endif
 458             twovfeps         = _mm_add_ps(vfeps,vfeps);
 459             vfitab           = _mm_slli_epi32(vfitab,3);
 460
 461             /* REACTION-FIELD ELECTROSTATICS */
 462             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 463             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 464
 465             /* CUBIC SPLINE TABLE DISPERSION */
 466             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 467             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 468             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 469             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 470             _MM_TRANSPOSE4_PS(Y,F,G,H);
 471             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 472             VV               = _mm_macc_ps(vfeps,Fp,Y);
 473             vvdw6            = _mm_mul_ps(c6_00,VV);
 474             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 475             fvdw6            = _mm_mul_ps(c6_00,FF);
 476
 477             /* CUBIC SPLINE TABLE REPULSION */
 478             vfitab           = _mm_add_epi32(vfitab,ifour);
 479             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 480             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 481             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 482             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 483             _MM_TRANSPOSE4_PS(Y,F,G,H);
 484             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 485             VV               = _mm_macc_ps(vfeps,Fp,Y);
 486             vvdw12           = _mm_mul_ps(c12_00,VV);
 487             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 488             fvdw12           = _mm_mul_ps(c12_00,FF);
 489             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 490             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 491
 492             /* Update potential sum for this i atom from the interaction with this j atom. */
 493             velec            = _mm_andnot_ps(dummy_mask,velec);
 494             velecsum         = _mm_add_ps(velecsum,velec);
 495             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 496             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 497
 498             fscal            = _mm_add_ps(felec,fvdw);
 499
 500             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 501
 502              /* Update vectorial force */
 503             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 504             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 505             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 506
 507             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 508             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 509             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 510
 511             /**************************
 512              * CALCULATE INTERACTIONS *
 513              **************************/
 514
 515             /* Compute parameters for interactions between i and j atoms */
 516             qq10             = _mm_mul_ps(iq1,jq0);
 517
 518             /* REACTION-FIELD ELECTROSTATICS */
 519             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 520             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 521
 522             /* Update potential sum for this i atom from the interaction with this j atom. */
 523             velec            = _mm_andnot_ps(dummy_mask,velec);
 524             velecsum         = _mm_add_ps(velecsum,velec);
 525
 526             fscal            = felec;
 527
 528             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 529
 530              /* Update vectorial force */
 531             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 532             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 533             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 534
 535             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 536             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 537             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 538
 539             /**************************
 540              * CALCULATE INTERACTIONS *
 541              **************************/
 542
 543             /* Compute parameters for interactions between i and j atoms */
 544             qq20             = _mm_mul_ps(iq2,jq0);
 545
 546             /* REACTION-FIELD ELECTROSTATICS */
 547             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 548             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 549
 550             /* Update potential sum for this i atom from the interaction with this j atom. */
 551             velec            = _mm_andnot_ps(dummy_mask,velec);
 552             velecsum         = _mm_add_ps(velecsum,velec);
 553
 554             fscal            = felec;
 555
 556             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 557
 558              /* Update vectorial force */
 559             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 560             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 561             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 562
 563             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 564             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 565             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 566
 567             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 568             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 569             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 570             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 571
 572             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 573
 574             /* Inner loop uses 141 flops */
 575         }
 576
 577         /* End of innermost loop */
 578
 579         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 580                                               f+i_coord_offset,fshift+i_shift_offset);
 581
 582         ggid                        = gid[iidx];
 583         /* Update potential energies */
 584         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 585         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 586
 587         /* Increment number of inner iterations */
 588         inneriter                  += j_index_end - j_index_start;
 589
 590         /* Outer loop uses 20 flops */
 591     }
 592
 593     /* Increment number of outer iterations */
 594     outeriter        += nri;
 595
 596     /* Update outer/inner flops */
 597
 598     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*141);
 599 }
 600 /*
 601  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_avx_128_fma_single
 602  * Electrostatics interaction: ReactionField
 603  * VdW interaction:            CubicSplineTable
 604  * Geometry:                   Water3-Particle
 605  * Calculate force/pot:        Force
 606  */
 607 void
 608 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_avx_128_fma_single
 609                     (t_nblist                    * gmx_restrict       nlist,
 610                      rvec                        * gmx_restrict          xx,
 611                      rvec                        * gmx_restrict          ff,
 612                      struct t_forcerec           * gmx_restrict          fr,
 613                      t_mdatoms                   * gmx_restrict     mdatoms,
 614                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 615                      t_nrnb                      * gmx_restrict        nrnb)
 616 {
 617     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 618      * just 0 for non-waters.
 619      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 620      * jnr indices corresponding to data put in the four positions in the SIMD register.
 621      */
 622     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 623     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 624     int              jnrA,jnrB,jnrC,jnrD;
 625     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 626     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 627     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 628     real             rcutoff_scalar;
 629     real             *shiftvec,*fshift,*x,*f;
 630     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 631     real             scratch[4*DIM];
 632     __m128           fscal,rcutoff,rcutoff2,jidxall;
 633     int              vdwioffset0;
 634     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 635     int              vdwioffset1;
 636     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 637     int              vdwioffset2;
 638     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 639     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 640     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 641     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 642     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 643     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 644     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 645     real             *charge;
 646     int              nvdwtype;
 647     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 648     int              *vdwtype;
 649     real             *vdwparam;
 650     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 651     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 652     __m128i          vfitab;
 653     __m128i          ifour       = _mm_set1_epi32(4);
 654     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 655     real             *vftab;
 656     __m128           dummy_mask,cutoff_mask;
 657     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 658     __m128           one     = _mm_set1_ps(1.0);
 659     __m128           two     = _mm_set1_ps(2.0);
 660     x                = xx[0];
 661     f                = ff[0];
 662
 663     nri              = nlist->nri;
 664     iinr             = nlist->iinr;
 665     jindex           = nlist->jindex;
 666     jjnr             = nlist->jjnr;
 667     shiftidx         = nlist->shift;
 668     gid              = nlist->gid;
 669     shiftvec         = fr->shift_vec[0];
 670     fshift           = fr->fshift[0];
 671     facel            = _mm_set1_ps(fr->ic->epsfac);
 672     charge           = mdatoms->chargeA;
 673     krf              = _mm_set1_ps(fr->ic->k_rf);
 674     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 675     crf              = _mm_set1_ps(fr->ic->c_rf);
 676     nvdwtype         = fr->ntype;
 677     vdwparam         = fr->nbfp;
 678     vdwtype          = mdatoms->typeA;
 679
 680     vftab            = kernel_data->table_vdw->data;
 681     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 682
 683     /* Setup water-specific parameters */
 684     inr              = nlist->iinr[0];
 685     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 686     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 687     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 688     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 689
 690     /* Avoid stupid compiler warnings */
 691     jnrA = jnrB = jnrC = jnrD = 0;
 692     j_coord_offsetA = 0;
 693     j_coord_offsetB = 0;
 694     j_coord_offsetC = 0;
 695     j_coord_offsetD = 0;
 696
 697     outeriter        = 0;
 698     inneriter        = 0;
 699
 700     for(iidx=0;iidx<4*DIM;iidx++)
 701     {
 702         scratch[iidx] = 0.0;
 703     }
 704
 705     /* Start outer loop over neighborlists */
 706     for(iidx=0; iidx<nri; iidx++)
 707     {
 708         /* Load shift vector for this list */
 709         i_shift_offset   = DIM*shiftidx[iidx];
 710
 711         /* Load limits for loop over neighbors */
 712         j_index_start    = jindex[iidx];
 713         j_index_end      = jindex[iidx+1];
 714
 715         /* Get outer coordinate index */
 716         inr              = iinr[iidx];
 717         i_coord_offset   = DIM*inr;
 718
 719         /* Load i particle coords and add shift vector */
 720         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 721                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 722
 723         fix0             = _mm_setzero_ps();
 724         fiy0             = _mm_setzero_ps();
 725         fiz0             = _mm_setzero_ps();
 726         fix1             = _mm_setzero_ps();
 727         fiy1             = _mm_setzero_ps();
 728         fiz1             = _mm_setzero_ps();
 729         fix2             = _mm_setzero_ps();
 730         fiy2             = _mm_setzero_ps();
 731         fiz2             = _mm_setzero_ps();
 732
 733         /* Start inner kernel loop */
 734         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 735         {
 736
 737             /* Get j neighbor index, and coordinate index */
 738             jnrA             = jjnr[jidx];
 739             jnrB             = jjnr[jidx+1];
 740             jnrC             = jjnr[jidx+2];
 741             jnrD             = jjnr[jidx+3];
 742             j_coord_offsetA  = DIM*jnrA;
 743             j_coord_offsetB  = DIM*jnrB;
 744             j_coord_offsetC  = DIM*jnrC;
 745             j_coord_offsetD  = DIM*jnrD;
 746
 747             /* load j atom coordinates */
 748             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 749                                               x+j_coord_offsetC,x+j_coord_offsetD,
 750                                               &jx0,&jy0,&jz0);
 751
 752             /* Calculate displacement vector */
 753             dx00             = _mm_sub_ps(ix0,jx0);
 754             dy00             = _mm_sub_ps(iy0,jy0);
 755             dz00             = _mm_sub_ps(iz0,jz0);
 756             dx10             = _mm_sub_ps(ix1,jx0);
 757             dy10             = _mm_sub_ps(iy1,jy0);
 758             dz10             = _mm_sub_ps(iz1,jz0);
 759             dx20             = _mm_sub_ps(ix2,jx0);
 760             dy20             = _mm_sub_ps(iy2,jy0);
 761             dz20             = _mm_sub_ps(iz2,jz0);
 762
 763             /* Calculate squared distance and things based on it */
 764             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 765             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 766             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 767
 768             rinv00           = avx128fma_invsqrt_f(rsq00);
 769             rinv10           = avx128fma_invsqrt_f(rsq10);
 770             rinv20           = avx128fma_invsqrt_f(rsq20);
 771
 772             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 773             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 774             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 775
 776             /* Load parameters for j particles */
 777             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 778                                                               charge+jnrC+0,charge+jnrD+0);
 779             vdwjidx0A        = 2*vdwtype[jnrA+0];
 780             vdwjidx0B        = 2*vdwtype[jnrB+0];
 781             vdwjidx0C        = 2*vdwtype[jnrC+0];
 782             vdwjidx0D        = 2*vdwtype[jnrD+0];
 783
 784             fjx0             = _mm_setzero_ps();
 785             fjy0             = _mm_setzero_ps();
 786             fjz0             = _mm_setzero_ps();
 787
 788             /**************************
 789              * CALCULATE INTERACTIONS *
 790              **************************/
 791
 792             r00              = _mm_mul_ps(rsq00,rinv00);
 793
 794             /* Compute parameters for interactions between i and j atoms */
 795             qq00             = _mm_mul_ps(iq0,jq0);
 796             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 797                                          vdwparam+vdwioffset0+vdwjidx0B,
 798                                          vdwparam+vdwioffset0+vdwjidx0C,
 799                                          vdwparam+vdwioffset0+vdwjidx0D,
 800                                          &c6_00,&c12_00);
 801
 802             /* Calculate table index by multiplying r with table scale and truncate to integer */
 803             rt               = _mm_mul_ps(r00,vftabscale);
 804             vfitab           = _mm_cvttps_epi32(rt);
 805 #ifdef __XOP__
 806             vfeps            = _mm_frcz_ps(rt);
 807 #else
 808             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 809 #endif
 810             twovfeps         = _mm_add_ps(vfeps,vfeps);
 811             vfitab           = _mm_slli_epi32(vfitab,3);
 812
 813             /* REACTION-FIELD ELECTROSTATICS */
 814             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 815
 816             /* CUBIC SPLINE TABLE DISPERSION */
 817             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 818             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 819             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 820             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 821             _MM_TRANSPOSE4_PS(Y,F,G,H);
 822             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 823             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 824             fvdw6            = _mm_mul_ps(c6_00,FF);
 825
 826             /* CUBIC SPLINE TABLE REPULSION */
 827             vfitab           = _mm_add_epi32(vfitab,ifour);
 828             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 829             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 830             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 831             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 832             _MM_TRANSPOSE4_PS(Y,F,G,H);
 833             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 834             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 835             fvdw12           = _mm_mul_ps(c12_00,FF);
 836             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 837
 838             fscal            = _mm_add_ps(felec,fvdw);
 839
 840              /* Update vectorial force */
 841             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 842             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 843             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 844
 845             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 846             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 847             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 848
 849             /**************************
 850              * CALCULATE INTERACTIONS *
 851              **************************/
 852
 853             /* Compute parameters for interactions between i and j atoms */
 854             qq10             = _mm_mul_ps(iq1,jq0);
 855
 856             /* REACTION-FIELD ELECTROSTATICS */
 857             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 858
 859             fscal            = felec;
 860
 861              /* Update vectorial force */
 862             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 863             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 864             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 865
 866             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 867             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 868             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 869
 870             /**************************
 871              * CALCULATE INTERACTIONS *
 872              **************************/
 873
 874             /* Compute parameters for interactions between i and j atoms */
 875             qq20             = _mm_mul_ps(iq2,jq0);
 876
 877             /* REACTION-FIELD ELECTROSTATICS */
 878             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 879
 880             fscal            = felec;
 881
 882              /* Update vectorial force */
 883             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 884             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 885             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 886
 887             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 888             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 889             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 890
 891             fjptrA             = f+j_coord_offsetA;
 892             fjptrB             = f+j_coord_offsetB;
 893             fjptrC             = f+j_coord_offsetC;
 894             fjptrD             = f+j_coord_offsetD;
 895
 896             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 897
 898             /* Inner loop uses 117 flops */
 899         }
 900
 901         if(jidx<j_index_end)
 902         {
 903
 904             /* Get j neighbor index, and coordinate index */
 905             jnrlistA         = jjnr[jidx];
 906             jnrlistB         = jjnr[jidx+1];
 907             jnrlistC         = jjnr[jidx+2];
 908             jnrlistD         = jjnr[jidx+3];
 909             /* Sign of each element will be negative for non-real atoms.
 910              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 911              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 912              */
 913             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 914             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 915             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 916             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 917             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 918             j_coord_offsetA  = DIM*jnrA;
 919             j_coord_offsetB  = DIM*jnrB;
 920             j_coord_offsetC  = DIM*jnrC;
 921             j_coord_offsetD  = DIM*jnrD;
 922
 923             /* load j atom coordinates */
 924             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 925                                               x+j_coord_offsetC,x+j_coord_offsetD,
 926                                               &jx0,&jy0,&jz0);
 927
 928             /* Calculate displacement vector */
 929             dx00             = _mm_sub_ps(ix0,jx0);
 930             dy00             = _mm_sub_ps(iy0,jy0);
 931             dz00             = _mm_sub_ps(iz0,jz0);
 932             dx10             = _mm_sub_ps(ix1,jx0);
 933             dy10             = _mm_sub_ps(iy1,jy0);
 934             dz10             = _mm_sub_ps(iz1,jz0);
 935             dx20             = _mm_sub_ps(ix2,jx0);
 936             dy20             = _mm_sub_ps(iy2,jy0);
 937             dz20             = _mm_sub_ps(iz2,jz0);
 938
 939             /* Calculate squared distance and things based on it */
 940             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 941             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 942             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 943
 944             rinv00           = avx128fma_invsqrt_f(rsq00);
 945             rinv10           = avx128fma_invsqrt_f(rsq10);
 946             rinv20           = avx128fma_invsqrt_f(rsq20);
 947
 948             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 949             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 950             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 951
 952             /* Load parameters for j particles */
 953             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 954                                                               charge+jnrC+0,charge+jnrD+0);
 955             vdwjidx0A        = 2*vdwtype[jnrA+0];
 956             vdwjidx0B        = 2*vdwtype[jnrB+0];
 957             vdwjidx0C        = 2*vdwtype[jnrC+0];
 958             vdwjidx0D        = 2*vdwtype[jnrD+0];
 959
 960             fjx0             = _mm_setzero_ps();
 961             fjy0             = _mm_setzero_ps();
 962             fjz0             = _mm_setzero_ps();
 963
 964             /**************************
 965              * CALCULATE INTERACTIONS *
 966              **************************/
 967
 968             r00              = _mm_mul_ps(rsq00,rinv00);
 969             r00              = _mm_andnot_ps(dummy_mask,r00);
 970
 971             /* Compute parameters for interactions between i and j atoms */
 972             qq00             = _mm_mul_ps(iq0,jq0);
 973             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 974                                          vdwparam+vdwioffset0+vdwjidx0B,
 975                                          vdwparam+vdwioffset0+vdwjidx0C,
 976                                          vdwparam+vdwioffset0+vdwjidx0D,
 977                                          &c6_00,&c12_00);
 978
 979             /* Calculate table index by multiplying r with table scale and truncate to integer */
 980             rt               = _mm_mul_ps(r00,vftabscale);
 981             vfitab           = _mm_cvttps_epi32(rt);
 982 #ifdef __XOP__
 983             vfeps            = _mm_frcz_ps(rt);
 984 #else
 985             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 986 #endif
 987             twovfeps         = _mm_add_ps(vfeps,vfeps);
 988             vfitab           = _mm_slli_epi32(vfitab,3);
 989
 990             /* REACTION-FIELD ELECTROSTATICS */
 991             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 992
 993             /* CUBIC SPLINE TABLE DISPERSION */
 994             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 995             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 996             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 997             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 998             _MM_TRANSPOSE4_PS(Y,F,G,H);
 999             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1000             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1001             fvdw6            = _mm_mul_ps(c6_00,FF);
1002
1003             /* CUBIC SPLINE TABLE REPULSION */
1004             vfitab           = _mm_add_epi32(vfitab,ifour);
1005             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1006             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1007             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1008             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1009             _MM_TRANSPOSE4_PS(Y,F,G,H);
1010             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1011             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1012             fvdw12           = _mm_mul_ps(c12_00,FF);
1013             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1014
1015             fscal            = _mm_add_ps(felec,fvdw);
1016
1017             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1018
1019              /* Update vectorial force */
1020             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1021             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1022             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1023
1024             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1025             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1026             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1027
1028             /**************************
1029              * CALCULATE INTERACTIONS *
1030              **************************/
1031
1032             /* Compute parameters for interactions between i and j atoms */
1033             qq10             = _mm_mul_ps(iq1,jq0);
1034
1035             /* REACTION-FIELD ELECTROSTATICS */
1036             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
1037
1038             fscal            = felec;
1039
1040             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1041
1042              /* Update vectorial force */
1043             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1044             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1045             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1046
1047             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1048             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1049             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1050
1051             /**************************
1052              * CALCULATE INTERACTIONS *
1053              **************************/
1054
1055             /* Compute parameters for interactions between i and j atoms */
1056             qq20             = _mm_mul_ps(iq2,jq0);
1057
1058             /* REACTION-FIELD ELECTROSTATICS */
1059             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
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 }