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) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  48 #include "kernelutil_x86_avx_128_fma_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW3P1_VF_avx_128_fma_single
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_VF_avx_128_fma_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwioffset1;
  86     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     int              vdwioffset2;
  88     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  92     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  93     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  94     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     int              nvdwtype;
  97     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 101     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 102     __m128i          vfitab;
 103     __m128i          ifour       = _mm_set1_epi32(4);
 104     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 105     real             *vftab;
 106     __m128           dummy_mask,cutoff_mask;
 107     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 108     __m128           one     = _mm_set1_ps(1.0);
 109     __m128           two     = _mm_set1_ps(2.0);
 110     x                = xx[0];
 111     f                = ff[0];
 112
 113     nri              = nlist->nri;
 114     iinr             = nlist->iinr;
 115     jindex           = nlist->jindex;
 116     jjnr             = nlist->jjnr;
 117     shiftidx         = nlist->shift;
 118     gid              = nlist->gid;
 119     shiftvec         = fr->shift_vec[0];
 120     fshift           = fr->fshift[0];
 121     facel            = _mm_set1_ps(fr->epsfac);
 122     charge           = mdatoms->chargeA;
 123     krf              = _mm_set1_ps(fr->ic->k_rf);
 124     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 125     crf              = _mm_set1_ps(fr->ic->c_rf);
 126     nvdwtype         = fr->ntype;
 127     vdwparam         = fr->nbfp;
 128     vdwtype          = mdatoms->typeA;
 129
 130     vftab            = kernel_data->table_vdw->data;
 131     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 132
 133     /* Setup water-specific parameters */
 134     inr              = nlist->iinr[0];
 135     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 136     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 137     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 138     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 139
 140     /* Avoid stupid compiler warnings */
 141     jnrA = jnrB = jnrC = jnrD = 0;
 142     j_coord_offsetA = 0;
 143     j_coord_offsetB = 0;
 144     j_coord_offsetC = 0;
 145     j_coord_offsetD = 0;
 146
 147     outeriter        = 0;
 148     inneriter        = 0;
 149
 150     for(iidx=0;iidx<4*DIM;iidx++)
 151     {
 152         scratch[iidx] = 0.0;
 153     }
 154
 155     /* Start outer loop over neighborlists */
 156     for(iidx=0; iidx<nri; iidx++)
 157     {
 158         /* Load shift vector for this list */
 159         i_shift_offset   = DIM*shiftidx[iidx];
 160
 161         /* Load limits for loop over neighbors */
 162         j_index_start    = jindex[iidx];
 163         j_index_end      = jindex[iidx+1];
 164
 165         /* Get outer coordinate index */
 166         inr              = iinr[iidx];
 167         i_coord_offset   = DIM*inr;
 168
 169         /* Load i particle coords and add shift vector */
 170         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 171                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 172
 173         fix0             = _mm_setzero_ps();
 174         fiy0             = _mm_setzero_ps();
 175         fiz0             = _mm_setzero_ps();
 176         fix1             = _mm_setzero_ps();
 177         fiy1             = _mm_setzero_ps();
 178         fiz1             = _mm_setzero_ps();
 179         fix2             = _mm_setzero_ps();
 180         fiy2             = _mm_setzero_ps();
 181         fiz2             = _mm_setzero_ps();
 182
 183         /* Reset potential sums */
 184         velecsum         = _mm_setzero_ps();
 185         vvdwsum          = _mm_setzero_ps();
 186
 187         /* Start inner kernel loop */
 188         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 189         {
 190
 191             /* Get j neighbor index, and coordinate index */
 192             jnrA             = jjnr[jidx];
 193             jnrB             = jjnr[jidx+1];
 194             jnrC             = jjnr[jidx+2];
 195             jnrD             = jjnr[jidx+3];
 196             j_coord_offsetA  = DIM*jnrA;
 197             j_coord_offsetB  = DIM*jnrB;
 198             j_coord_offsetC  = DIM*jnrC;
 199             j_coord_offsetD  = DIM*jnrD;
 200
 201             /* load j atom coordinates */
 202             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 203                                               x+j_coord_offsetC,x+j_coord_offsetD,
 204                                               &jx0,&jy0,&jz0);
 205
 206             /* Calculate displacement vector */
 207             dx00             = _mm_sub_ps(ix0,jx0);
 208             dy00             = _mm_sub_ps(iy0,jy0);
 209             dz00             = _mm_sub_ps(iz0,jz0);
 210             dx10             = _mm_sub_ps(ix1,jx0);
 211             dy10             = _mm_sub_ps(iy1,jy0);
 212             dz10             = _mm_sub_ps(iz1,jz0);
 213             dx20             = _mm_sub_ps(ix2,jx0);
 214             dy20             = _mm_sub_ps(iy2,jy0);
 215             dz20             = _mm_sub_ps(iz2,jz0);
 216
 217             /* Calculate squared distance and things based on it */
 218             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 219             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 220             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 221
 222             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 223             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 224             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 225
 226             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 227             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 228             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 229
 230             /* Load parameters for j particles */
 231             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 232                                                               charge+jnrC+0,charge+jnrD+0);
 233             vdwjidx0A        = 2*vdwtype[jnrA+0];
 234             vdwjidx0B        = 2*vdwtype[jnrB+0];
 235             vdwjidx0C        = 2*vdwtype[jnrC+0];
 236             vdwjidx0D        = 2*vdwtype[jnrD+0];
 237
 238             fjx0             = _mm_setzero_ps();
 239             fjy0             = _mm_setzero_ps();
 240             fjz0             = _mm_setzero_ps();
 241
 242             /**************************
 243              * CALCULATE INTERACTIONS *
 244              **************************/
 245
 246             r00              = _mm_mul_ps(rsq00,rinv00);
 247
 248             /* Compute parameters for interactions between i and j atoms */
 249             qq00             = _mm_mul_ps(iq0,jq0);
 250             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 251                                          vdwparam+vdwioffset0+vdwjidx0B,
 252                                          vdwparam+vdwioffset0+vdwjidx0C,
 253                                          vdwparam+vdwioffset0+vdwjidx0D,
 254                                          &c6_00,&c12_00);
 255
 256             /* Calculate table index by multiplying r with table scale and truncate to integer */
 257             rt               = _mm_mul_ps(r00,vftabscale);
 258             vfitab           = _mm_cvttps_epi32(rt);
 259 #ifdef __XOP__
 260             vfeps            = _mm_frcz_ps(rt);
 261 #else
 262             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 263 #endif
 264             twovfeps         = _mm_add_ps(vfeps,vfeps);
 265             vfitab           = _mm_slli_epi32(vfitab,3);
 266
 267             /* REACTION-FIELD ELECTROSTATICS */
 268             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 269             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 270
 271             /* CUBIC SPLINE TABLE DISPERSION */
 272             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 273             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 274             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 275             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 276             _MM_TRANSPOSE4_PS(Y,F,G,H);
 277             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 278             VV               = _mm_macc_ps(vfeps,Fp,Y);
 279             vvdw6            = _mm_mul_ps(c6_00,VV);
 280             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 281             fvdw6            = _mm_mul_ps(c6_00,FF);
 282
 283             /* CUBIC SPLINE TABLE REPULSION */
 284             vfitab           = _mm_add_epi32(vfitab,ifour);
 285             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 286             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 287             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 288             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 289             _MM_TRANSPOSE4_PS(Y,F,G,H);
 290             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 291             VV               = _mm_macc_ps(vfeps,Fp,Y);
 292             vvdw12           = _mm_mul_ps(c12_00,VV);
 293             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 294             fvdw12           = _mm_mul_ps(c12_00,FF);
 295             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 296             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 297
 298             /* Update potential sum for this i atom from the interaction with this j atom. */
 299             velecsum         = _mm_add_ps(velecsum,velec);
 300             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 301
 302             fscal            = _mm_add_ps(felec,fvdw);
 303
 304              /* Update vectorial force */
 305             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 306             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 307             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 308
 309             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 310             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 311             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 312
 313             /**************************
 314              * CALCULATE INTERACTIONS *
 315              **************************/
 316
 317             /* Compute parameters for interactions between i and j atoms */
 318             qq10             = _mm_mul_ps(iq1,jq0);
 319
 320             /* REACTION-FIELD ELECTROSTATICS */
 321             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 322             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 323
 324             /* Update potential sum for this i atom from the interaction with this j atom. */
 325             velecsum         = _mm_add_ps(velecsum,velec);
 326
 327             fscal            = felec;
 328
 329              /* Update vectorial force */
 330             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 331             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 332             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 333
 334             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 335             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 336             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 337
 338             /**************************
 339              * CALCULATE INTERACTIONS *
 340              **************************/
 341
 342             /* Compute parameters for interactions between i and j atoms */
 343             qq20             = _mm_mul_ps(iq2,jq0);
 344
 345             /* REACTION-FIELD ELECTROSTATICS */
 346             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 347             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 348
 349             /* Update potential sum for this i atom from the interaction with this j atom. */
 350             velecsum         = _mm_add_ps(velecsum,velec);
 351
 352             fscal            = felec;
 353
 354              /* Update vectorial force */
 355             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 356             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 357             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 358
 359             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 360             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 361             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 362
 363             fjptrA             = f+j_coord_offsetA;
 364             fjptrB             = f+j_coord_offsetB;
 365             fjptrC             = f+j_coord_offsetC;
 366             fjptrD             = f+j_coord_offsetD;
 367
 368             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 369
 370             /* Inner loop uses 140 flops */
 371         }
 372
 373         if(jidx<j_index_end)
 374         {
 375
 376             /* Get j neighbor index, and coordinate index */
 377             jnrlistA         = jjnr[jidx];
 378             jnrlistB         = jjnr[jidx+1];
 379             jnrlistC         = jjnr[jidx+2];
 380             jnrlistD         = jjnr[jidx+3];
 381             /* Sign of each element will be negative for non-real atoms.
 382              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 383              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 384              */
 385             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 386             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 387             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 388             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 389             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 390             j_coord_offsetA  = DIM*jnrA;
 391             j_coord_offsetB  = DIM*jnrB;
 392             j_coord_offsetC  = DIM*jnrC;
 393             j_coord_offsetD  = DIM*jnrD;
 394
 395             /* load j atom coordinates */
 396             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 397                                               x+j_coord_offsetC,x+j_coord_offsetD,
 398                                               &jx0,&jy0,&jz0);
 399
 400             /* Calculate displacement vector */
 401             dx00             = _mm_sub_ps(ix0,jx0);
 402             dy00             = _mm_sub_ps(iy0,jy0);
 403             dz00             = _mm_sub_ps(iz0,jz0);
 404             dx10             = _mm_sub_ps(ix1,jx0);
 405             dy10             = _mm_sub_ps(iy1,jy0);
 406             dz10             = _mm_sub_ps(iz1,jz0);
 407             dx20             = _mm_sub_ps(ix2,jx0);
 408             dy20             = _mm_sub_ps(iy2,jy0);
 409             dz20             = _mm_sub_ps(iz2,jz0);
 410
 411             /* Calculate squared distance and things based on it */
 412             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 413             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 414             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 415
 416             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 417             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 418             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 419
 420             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 421             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 422             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 423
 424             /* Load parameters for j particles */
 425             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 426                                                               charge+jnrC+0,charge+jnrD+0);
 427             vdwjidx0A        = 2*vdwtype[jnrA+0];
 428             vdwjidx0B        = 2*vdwtype[jnrB+0];
 429             vdwjidx0C        = 2*vdwtype[jnrC+0];
 430             vdwjidx0D        = 2*vdwtype[jnrD+0];
 431
 432             fjx0             = _mm_setzero_ps();
 433             fjy0             = _mm_setzero_ps();
 434             fjz0             = _mm_setzero_ps();
 435
 436             /**************************
 437              * CALCULATE INTERACTIONS *
 438              **************************/
 439
 440             r00              = _mm_mul_ps(rsq00,rinv00);
 441             r00              = _mm_andnot_ps(dummy_mask,r00);
 442
 443             /* Compute parameters for interactions between i and j atoms */
 444             qq00             = _mm_mul_ps(iq0,jq0);
 445             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 446                                          vdwparam+vdwioffset0+vdwjidx0B,
 447                                          vdwparam+vdwioffset0+vdwjidx0C,
 448                                          vdwparam+vdwioffset0+vdwjidx0D,
 449                                          &c6_00,&c12_00);
 450
 451             /* Calculate table index by multiplying r with table scale and truncate to integer */
 452             rt               = _mm_mul_ps(r00,vftabscale);
 453             vfitab           = _mm_cvttps_epi32(rt);
 454 #ifdef __XOP__
 455             vfeps            = _mm_frcz_ps(rt);
 456 #else
 457             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 458 #endif
 459             twovfeps         = _mm_add_ps(vfeps,vfeps);
 460             vfitab           = _mm_slli_epi32(vfitab,3);
 461
 462             /* REACTION-FIELD ELECTROSTATICS */
 463             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 464             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 465
 466             /* CUBIC SPLINE TABLE DISPERSION */
 467             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 468             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 469             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 470             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 471             _MM_TRANSPOSE4_PS(Y,F,G,H);
 472             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 473             VV               = _mm_macc_ps(vfeps,Fp,Y);
 474             vvdw6            = _mm_mul_ps(c6_00,VV);
 475             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 476             fvdw6            = _mm_mul_ps(c6_00,FF);
 477
 478             /* CUBIC SPLINE TABLE REPULSION */
 479             vfitab           = _mm_add_epi32(vfitab,ifour);
 480             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 481             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 482             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 483             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 484             _MM_TRANSPOSE4_PS(Y,F,G,H);
 485             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 486             VV               = _mm_macc_ps(vfeps,Fp,Y);
 487             vvdw12           = _mm_mul_ps(c12_00,VV);
 488             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 489             fvdw12           = _mm_mul_ps(c12_00,FF);
 490             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 491             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 492
 493             /* Update potential sum for this i atom from the interaction with this j atom. */
 494             velec            = _mm_andnot_ps(dummy_mask,velec);
 495             velecsum         = _mm_add_ps(velecsum,velec);
 496             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 497             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 498
 499             fscal            = _mm_add_ps(felec,fvdw);
 500
 501             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 502
 503              /* Update vectorial force */
 504             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 505             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 506             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 507
 508             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 509             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 510             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 511
 512             /**************************
 513              * CALCULATE INTERACTIONS *
 514              **************************/
 515
 516             /* Compute parameters for interactions between i and j atoms */
 517             qq10             = _mm_mul_ps(iq1,jq0);
 518
 519             /* REACTION-FIELD ELECTROSTATICS */
 520             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 521             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 522
 523             /* Update potential sum for this i atom from the interaction with this j atom. */
 524             velec            = _mm_andnot_ps(dummy_mask,velec);
 525             velecsum         = _mm_add_ps(velecsum,velec);
 526
 527             fscal            = felec;
 528
 529             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 530
 531              /* Update vectorial force */
 532             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 533             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 534             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 535
 536             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 537             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 538             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 539
 540             /**************************
 541              * CALCULATE INTERACTIONS *
 542              **************************/
 543
 544             /* Compute parameters for interactions between i and j atoms */
 545             qq20             = _mm_mul_ps(iq2,jq0);
 546
 547             /* REACTION-FIELD ELECTROSTATICS */
 548             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 549             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 550
 551             /* Update potential sum for this i atom from the interaction with this j atom. */
 552             velec            = _mm_andnot_ps(dummy_mask,velec);
 553             velecsum         = _mm_add_ps(velecsum,velec);
 554
 555             fscal            = felec;
 556
 557             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 558
 559              /* Update vectorial force */
 560             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 561             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 562             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 563
 564             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 565             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 566             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 567
 568             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 569             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 570             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 571             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 572
 573             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 574
 575             /* Inner loop uses 141 flops */
 576         }
 577
 578         /* End of innermost loop */
 579
 580         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 581                                               f+i_coord_offset,fshift+i_shift_offset);
 582
 583         ggid                        = gid[iidx];
 584         /* Update potential energies */
 585         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 586         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 587
 588         /* Increment number of inner iterations */
 589         inneriter                  += j_index_end - j_index_start;
 590
 591         /* Outer loop uses 20 flops */
 592     }
 593
 594     /* Increment number of outer iterations */
 595     outeriter        += nri;
 596
 597     /* Update outer/inner flops */
 598
 599     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*141);
 600 }
 601 /*
 602  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_avx_128_fma_single
 603  * Electrostatics interaction: ReactionField
 604  * VdW interaction:            CubicSplineTable
 605  * Geometry:                   Water3-Particle
 606  * Calculate force/pot:        Force
 607  */
 608 void
 609 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_avx_128_fma_single
 610                     (t_nblist                    * gmx_restrict       nlist,
 611                      rvec                        * gmx_restrict          xx,
 612                      rvec                        * gmx_restrict          ff,
 613                      t_forcerec                  * gmx_restrict          fr,
 614                      t_mdatoms                   * gmx_restrict     mdatoms,
 615                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 616                      t_nrnb                      * gmx_restrict        nrnb)
 617 {
 618     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 619      * just 0 for non-waters.
 620      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 621      * jnr indices corresponding to data put in the four positions in the SIMD register.
 622      */
 623     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 624     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 625     int              jnrA,jnrB,jnrC,jnrD;
 626     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 627     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 628     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 629     real             rcutoff_scalar;
 630     real             *shiftvec,*fshift,*x,*f;
 631     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 632     real             scratch[4*DIM];
 633     __m128           fscal,rcutoff,rcutoff2,jidxall;
 634     int              vdwioffset0;
 635     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 636     int              vdwioffset1;
 637     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 638     int              vdwioffset2;
 639     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 640     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 641     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 642     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 643     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 644     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 645     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 646     real             *charge;
 647     int              nvdwtype;
 648     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 649     int              *vdwtype;
 650     real             *vdwparam;
 651     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 652     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 653     __m128i          vfitab;
 654     __m128i          ifour       = _mm_set1_epi32(4);
 655     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 656     real             *vftab;
 657     __m128           dummy_mask,cutoff_mask;
 658     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 659     __m128           one     = _mm_set1_ps(1.0);
 660     __m128           two     = _mm_set1_ps(2.0);
 661     x                = xx[0];
 662     f                = ff[0];
 663
 664     nri              = nlist->nri;
 665     iinr             = nlist->iinr;
 666     jindex           = nlist->jindex;
 667     jjnr             = nlist->jjnr;
 668     shiftidx         = nlist->shift;
 669     gid              = nlist->gid;
 670     shiftvec         = fr->shift_vec[0];
 671     fshift           = fr->fshift[0];
 672     facel            = _mm_set1_ps(fr->epsfac);
 673     charge           = mdatoms->chargeA;
 674     krf              = _mm_set1_ps(fr->ic->k_rf);
 675     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 676     crf              = _mm_set1_ps(fr->ic->c_rf);
 677     nvdwtype         = fr->ntype;
 678     vdwparam         = fr->nbfp;
 679     vdwtype          = mdatoms->typeA;
 680
 681     vftab            = kernel_data->table_vdw->data;
 682     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 683
 684     /* Setup water-specific parameters */
 685     inr              = nlist->iinr[0];
 686     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 687     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 688     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 689     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 690
 691     /* Avoid stupid compiler warnings */
 692     jnrA = jnrB = jnrC = jnrD = 0;
 693     j_coord_offsetA = 0;
 694     j_coord_offsetB = 0;
 695     j_coord_offsetC = 0;
 696     j_coord_offsetD = 0;
 697
 698     outeriter        = 0;
 699     inneriter        = 0;
 700
 701     for(iidx=0;iidx<4*DIM;iidx++)
 702     {
 703         scratch[iidx] = 0.0;
 704     }
 705
 706     /* Start outer loop over neighborlists */
 707     for(iidx=0; iidx<nri; iidx++)
 708     {
 709         /* Load shift vector for this list */
 710         i_shift_offset   = DIM*shiftidx[iidx];
 711
 712         /* Load limits for loop over neighbors */
 713         j_index_start    = jindex[iidx];
 714         j_index_end      = jindex[iidx+1];
 715
 716         /* Get outer coordinate index */
 717         inr              = iinr[iidx];
 718         i_coord_offset   = DIM*inr;
 719
 720         /* Load i particle coords and add shift vector */
 721         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 722                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 723
 724         fix0             = _mm_setzero_ps();
 725         fiy0             = _mm_setzero_ps();
 726         fiz0             = _mm_setzero_ps();
 727         fix1             = _mm_setzero_ps();
 728         fiy1             = _mm_setzero_ps();
 729         fiz1             = _mm_setzero_ps();
 730         fix2             = _mm_setzero_ps();
 731         fiy2             = _mm_setzero_ps();
 732         fiz2             = _mm_setzero_ps();
 733
 734         /* Start inner kernel loop */
 735         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 736         {
 737
 738             /* Get j neighbor index, and coordinate index */
 739             jnrA             = jjnr[jidx];
 740             jnrB             = jjnr[jidx+1];
 741             jnrC             = jjnr[jidx+2];
 742             jnrD             = jjnr[jidx+3];
 743             j_coord_offsetA  = DIM*jnrA;
 744             j_coord_offsetB  = DIM*jnrB;
 745             j_coord_offsetC  = DIM*jnrC;
 746             j_coord_offsetD  = DIM*jnrD;
 747
 748             /* load j atom coordinates */
 749             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 750                                               x+j_coord_offsetC,x+j_coord_offsetD,
 751                                               &jx0,&jy0,&jz0);
 752
 753             /* Calculate displacement vector */
 754             dx00             = _mm_sub_ps(ix0,jx0);
 755             dy00             = _mm_sub_ps(iy0,jy0);
 756             dz00             = _mm_sub_ps(iz0,jz0);
 757             dx10             = _mm_sub_ps(ix1,jx0);
 758             dy10             = _mm_sub_ps(iy1,jy0);
 759             dz10             = _mm_sub_ps(iz1,jz0);
 760             dx20             = _mm_sub_ps(ix2,jx0);
 761             dy20             = _mm_sub_ps(iy2,jy0);
 762             dz20             = _mm_sub_ps(iz2,jz0);
 763
 764             /* Calculate squared distance and things based on it */
 765             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 766             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 767             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 768
 769             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 770             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 771             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 772
 773             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 774             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 775             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 776
 777             /* Load parameters for j particles */
 778             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 779                                                               charge+jnrC+0,charge+jnrD+0);
 780             vdwjidx0A        = 2*vdwtype[jnrA+0];
 781             vdwjidx0B        = 2*vdwtype[jnrB+0];
 782             vdwjidx0C        = 2*vdwtype[jnrC+0];
 783             vdwjidx0D        = 2*vdwtype[jnrD+0];
 784
 785             fjx0             = _mm_setzero_ps();
 786             fjy0             = _mm_setzero_ps();
 787             fjz0             = _mm_setzero_ps();
 788
 789             /**************************
 790              * CALCULATE INTERACTIONS *
 791              **************************/
 792
 793             r00              = _mm_mul_ps(rsq00,rinv00);
 794
 795             /* Compute parameters for interactions between i and j atoms */
 796             qq00             = _mm_mul_ps(iq0,jq0);
 797             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 798                                          vdwparam+vdwioffset0+vdwjidx0B,
 799                                          vdwparam+vdwioffset0+vdwjidx0C,
 800                                          vdwparam+vdwioffset0+vdwjidx0D,
 801                                          &c6_00,&c12_00);
 802
 803             /* Calculate table index by multiplying r with table scale and truncate to integer */
 804             rt               = _mm_mul_ps(r00,vftabscale);
 805             vfitab           = _mm_cvttps_epi32(rt);
 806 #ifdef __XOP__
 807             vfeps            = _mm_frcz_ps(rt);
 808 #else
 809             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 810 #endif
 811             twovfeps         = _mm_add_ps(vfeps,vfeps);
 812             vfitab           = _mm_slli_epi32(vfitab,3);
 813
 814             /* REACTION-FIELD ELECTROSTATICS */
 815             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 816
 817             /* CUBIC SPLINE TABLE DISPERSION */
 818             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 819             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 820             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 821             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 822             _MM_TRANSPOSE4_PS(Y,F,G,H);
 823             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 824             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 825             fvdw6            = _mm_mul_ps(c6_00,FF);
 826
 827             /* CUBIC SPLINE TABLE REPULSION */
 828             vfitab           = _mm_add_epi32(vfitab,ifour);
 829             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 830             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 831             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 832             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 833             _MM_TRANSPOSE4_PS(Y,F,G,H);
 834             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 835             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 836             fvdw12           = _mm_mul_ps(c12_00,FF);
 837             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 838
 839             fscal            = _mm_add_ps(felec,fvdw);
 840
 841              /* Update vectorial force */
 842             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 843             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 844             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 845
 846             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 847             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 848             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 849
 850             /**************************
 851              * CALCULATE INTERACTIONS *
 852              **************************/
 853
 854             /* Compute parameters for interactions between i and j atoms */
 855             qq10             = _mm_mul_ps(iq1,jq0);
 856
 857             /* REACTION-FIELD ELECTROSTATICS */
 858             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 859
 860             fscal            = felec;
 861
 862              /* Update vectorial force */
 863             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 864             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 865             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 866
 867             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 868             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 869             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 870
 871             /**************************
 872              * CALCULATE INTERACTIONS *
 873              **************************/
 874
 875             /* Compute parameters for interactions between i and j atoms */
 876             qq20             = _mm_mul_ps(iq2,jq0);
 877
 878             /* REACTION-FIELD ELECTROSTATICS */
 879             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 880
 881             fscal            = felec;
 882
 883              /* Update vectorial force */
 884             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 885             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 886             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 887
 888             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 889             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 890             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 891
 892             fjptrA             = f+j_coord_offsetA;
 893             fjptrB             = f+j_coord_offsetB;
 894             fjptrC             = f+j_coord_offsetC;
 895             fjptrD             = f+j_coord_offsetD;
 896
 897             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 898
 899             /* Inner loop uses 117 flops */
 900         }
 901
 902         if(jidx<j_index_end)
 903         {
 904
 905             /* Get j neighbor index, and coordinate index */
 906             jnrlistA         = jjnr[jidx];
 907             jnrlistB         = jjnr[jidx+1];
 908             jnrlistC         = jjnr[jidx+2];
 909             jnrlistD         = jjnr[jidx+3];
 910             /* Sign of each element will be negative for non-real atoms.
 911              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 912              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 913              */
 914             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 915             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 916             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 917             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 918             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 919             j_coord_offsetA  = DIM*jnrA;
 920             j_coord_offsetB  = DIM*jnrB;
 921             j_coord_offsetC  = DIM*jnrC;
 922             j_coord_offsetD  = DIM*jnrD;
 923
 924             /* load j atom coordinates */
 925             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 926                                               x+j_coord_offsetC,x+j_coord_offsetD,
 927                                               &jx0,&jy0,&jz0);
 928
 929             /* Calculate displacement vector */
 930             dx00             = _mm_sub_ps(ix0,jx0);
 931             dy00             = _mm_sub_ps(iy0,jy0);
 932             dz00             = _mm_sub_ps(iz0,jz0);
 933             dx10             = _mm_sub_ps(ix1,jx0);
 934             dy10             = _mm_sub_ps(iy1,jy0);
 935             dz10             = _mm_sub_ps(iz1,jz0);
 936             dx20             = _mm_sub_ps(ix2,jx0);
 937             dy20             = _mm_sub_ps(iy2,jy0);
 938             dz20             = _mm_sub_ps(iz2,jz0);
 939
 940             /* Calculate squared distance and things based on it */
 941             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 942             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 943             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 944
 945             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 946             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 947             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 948
 949             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 950             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 951             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 952
 953             /* Load parameters for j particles */
 954             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 955                                                               charge+jnrC+0,charge+jnrD+0);
 956             vdwjidx0A        = 2*vdwtype[jnrA+0];
 957             vdwjidx0B        = 2*vdwtype[jnrB+0];
 958             vdwjidx0C        = 2*vdwtype[jnrC+0];
 959             vdwjidx0D        = 2*vdwtype[jnrD+0];
 960
 961             fjx0             = _mm_setzero_ps();
 962             fjy0             = _mm_setzero_ps();
 963             fjz0             = _mm_setzero_ps();
 964
 965             /**************************
 966              * CALCULATE INTERACTIONS *
 967              **************************/
 968
 969             r00              = _mm_mul_ps(rsq00,rinv00);
 970             r00              = _mm_andnot_ps(dummy_mask,r00);
 971
 972             /* Compute parameters for interactions between i and j atoms */
 973             qq00             = _mm_mul_ps(iq0,jq0);
 974             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 975                                          vdwparam+vdwioffset0+vdwjidx0B,
 976                                          vdwparam+vdwioffset0+vdwjidx0C,
 977                                          vdwparam+vdwioffset0+vdwjidx0D,
 978                                          &c6_00,&c12_00);
 979
 980             /* Calculate table index by multiplying r with table scale and truncate to integer */
 981             rt               = _mm_mul_ps(r00,vftabscale);
 982             vfitab           = _mm_cvttps_epi32(rt);
 983 #ifdef __XOP__
 984             vfeps            = _mm_frcz_ps(rt);
 985 #else
 986             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 987 #endif
 988             twovfeps         = _mm_add_ps(vfeps,vfeps);
 989             vfitab           = _mm_slli_epi32(vfitab,3);
 990
 991             /* REACTION-FIELD ELECTROSTATICS */
 992             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 993
 994             /* CUBIC SPLINE TABLE DISPERSION */
 995             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 996             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 997             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 998             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 999             _MM_TRANSPOSE4_PS(Y,F,G,H);
1000             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1001             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1002             fvdw6            = _mm_mul_ps(c6_00,FF);
1003
1004             /* CUBIC SPLINE TABLE REPULSION */
1005             vfitab           = _mm_add_epi32(vfitab,ifour);
1006             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1007             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1008             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1009             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1010             _MM_TRANSPOSE4_PS(Y,F,G,H);
1011             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1012             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1013             fvdw12           = _mm_mul_ps(c12_00,FF);
1014             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1015
1016             fscal            = _mm_add_ps(felec,fvdw);
1017
1018             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1019
1020              /* Update vectorial force */
1021             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1022             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1023             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1024
1025             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1026             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1027             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1028
1029             /**************************
1030              * CALCULATE INTERACTIONS *
1031              **************************/
1032
1033             /* Compute parameters for interactions between i and j atoms */
1034             qq10             = _mm_mul_ps(iq1,jq0);
1035
1036             /* REACTION-FIELD ELECTROSTATICS */
1037             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
1038
1039             fscal            = felec;
1040
1041             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1042
1043              /* Update vectorial force */
1044             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1045             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1046             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1047
1048             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1049             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1050             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1051
1052             /**************************
1053              * CALCULATE INTERACTIONS *
1054              **************************/
1055
1056             /* Compute parameters for interactions between i and j atoms */
1057             qq20             = _mm_mul_ps(iq2,jq0);
1058
1059             /* REACTION-FIELD ELECTROSTATICS */
1060             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1061
1062             fscal            = felec;
1063
1064             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1065
1066              /* Update vectorial force */
1067             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1068             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1069             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1070
1071             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1072             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1073             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1074
1075             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1076             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1077             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1078             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1079
1080             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1081
1082             /* Inner loop uses 118 flops */
1083         }
1084
1085         /* End of innermost loop */
1086
1087         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1088                                               f+i_coord_offset,fshift+i_shift_offset);
1089
1090         /* Increment number of inner iterations */
1091         inneriter                  += j_index_end - j_index_start;
1092
1093         /* Outer loop uses 18 flops */
1094     }
1095
1096     /* Increment number of outer iterations */
1097     outeriter        += nri;
1098
1099     /* Update outer/inner flops */
1100
1101     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*118);
1102 }