src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecRFCut_VdwLJSh_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_ElecRFCut_VdwLJSh_GeomW3P1_VF_avx_128_fma_single
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_VdwLJSh_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     __m128           dummy_mask,cutoff_mask;
 103     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 104     __m128           one     = _mm_set1_ps(1.0);
 105     __m128           two     = _mm_set1_ps(2.0);
 106     x                = xx[0];
 107     f                = ff[0];
 108
 109     nri              = nlist->nri;
 110     iinr             = nlist->iinr;
 111     jindex           = nlist->jindex;
 112     jjnr             = nlist->jjnr;
 113     shiftidx         = nlist->shift;
 114     gid              = nlist->gid;
 115     shiftvec         = fr->shift_vec[0];
 116     fshift           = fr->fshift[0];
 117     facel            = _mm_set1_ps(fr->epsfac);
 118     charge           = mdatoms->chargeA;
 119     krf              = _mm_set1_ps(fr->ic->k_rf);
 120     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 121     crf              = _mm_set1_ps(fr->ic->c_rf);
 122     nvdwtype         = fr->ntype;
 123     vdwparam         = fr->nbfp;
 124     vdwtype          = mdatoms->typeA;
 125
 126     /* Setup water-specific parameters */
 127     inr              = nlist->iinr[0];
 128     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 129     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 130     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 131     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 132
 133     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 134     rcutoff_scalar   = fr->rcoulomb;
 135     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 136     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 137
 138     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 139     rvdw             = _mm_set1_ps(fr->rvdw);
 140
 141     /* Avoid stupid compiler warnings */
 142     jnrA = jnrB = jnrC = jnrD = 0;
 143     j_coord_offsetA = 0;
 144     j_coord_offsetB = 0;
 145     j_coord_offsetC = 0;
 146     j_coord_offsetD = 0;
 147
 148     outeriter        = 0;
 149     inneriter        = 0;
 150
 151     for(iidx=0;iidx<4*DIM;iidx++)
 152     {
 153         scratch[iidx] = 0.0;
 154     }
 155
 156     /* Start outer loop over neighborlists */
 157     for(iidx=0; iidx<nri; iidx++)
 158     {
 159         /* Load shift vector for this list */
 160         i_shift_offset   = DIM*shiftidx[iidx];
 161
 162         /* Load limits for loop over neighbors */
 163         j_index_start    = jindex[iidx];
 164         j_index_end      = jindex[iidx+1];
 165
 166         /* Get outer coordinate index */
 167         inr              = iinr[iidx];
 168         i_coord_offset   = DIM*inr;
 169
 170         /* Load i particle coords and add shift vector */
 171         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 172                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 173
 174         fix0             = _mm_setzero_ps();
 175         fiy0             = _mm_setzero_ps();
 176         fiz0             = _mm_setzero_ps();
 177         fix1             = _mm_setzero_ps();
 178         fiy1             = _mm_setzero_ps();
 179         fiz1             = _mm_setzero_ps();
 180         fix2             = _mm_setzero_ps();
 181         fiy2             = _mm_setzero_ps();
 182         fiz2             = _mm_setzero_ps();
 183
 184         /* Reset potential sums */
 185         velecsum         = _mm_setzero_ps();
 186         vvdwsum          = _mm_setzero_ps();
 187
 188         /* Start inner kernel loop */
 189         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 190         {
 191
 192             /* Get j neighbor index, and coordinate index */
 193             jnrA             = jjnr[jidx];
 194             jnrB             = jjnr[jidx+1];
 195             jnrC             = jjnr[jidx+2];
 196             jnrD             = jjnr[jidx+3];
 197             j_coord_offsetA  = DIM*jnrA;
 198             j_coord_offsetB  = DIM*jnrB;
 199             j_coord_offsetC  = DIM*jnrC;
 200             j_coord_offsetD  = DIM*jnrD;
 201
 202             /* load j atom coordinates */
 203             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 204                                               x+j_coord_offsetC,x+j_coord_offsetD,
 205                                               &jx0,&jy0,&jz0);
 206
 207             /* Calculate displacement vector */
 208             dx00             = _mm_sub_ps(ix0,jx0);
 209             dy00             = _mm_sub_ps(iy0,jy0);
 210             dz00             = _mm_sub_ps(iz0,jz0);
 211             dx10             = _mm_sub_ps(ix1,jx0);
 212             dy10             = _mm_sub_ps(iy1,jy0);
 213             dz10             = _mm_sub_ps(iz1,jz0);
 214             dx20             = _mm_sub_ps(ix2,jx0);
 215             dy20             = _mm_sub_ps(iy2,jy0);
 216             dz20             = _mm_sub_ps(iz2,jz0);
 217
 218             /* Calculate squared distance and things based on it */
 219             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 220             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 221             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 222
 223             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 224             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 225             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 226
 227             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 228             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 229             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 230
 231             /* Load parameters for j particles */
 232             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 233                                                               charge+jnrC+0,charge+jnrD+0);
 234             vdwjidx0A        = 2*vdwtype[jnrA+0];
 235             vdwjidx0B        = 2*vdwtype[jnrB+0];
 236             vdwjidx0C        = 2*vdwtype[jnrC+0];
 237             vdwjidx0D        = 2*vdwtype[jnrD+0];
 238
 239             fjx0             = _mm_setzero_ps();
 240             fjy0             = _mm_setzero_ps();
 241             fjz0             = _mm_setzero_ps();
 242
 243             /**************************
 244              * CALCULATE INTERACTIONS *
 245              **************************/
 246
 247             if (gmx_mm_any_lt(rsq00,rcutoff2))
 248             {
 249
 250             /* Compute parameters for interactions between i and j atoms */
 251             qq00             = _mm_mul_ps(iq0,jq0);
 252             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 253                                          vdwparam+vdwioffset0+vdwjidx0B,
 254                                          vdwparam+vdwioffset0+vdwjidx0C,
 255                                          vdwparam+vdwioffset0+vdwjidx0D,
 256                                          &c6_00,&c12_00);
 257
 258             /* REACTION-FIELD ELECTROSTATICS */
 259             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 260             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 261
 262             /* LENNARD-JONES DISPERSION/REPULSION */
 263
 264             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 265             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 266             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 267             vvdw             = _mm_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 268                                           _mm_mul_ps( _mm_nmacc_ps(c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 269             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 270
 271             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 272
 273             /* Update potential sum for this i atom from the interaction with this j atom. */
 274             velec            = _mm_and_ps(velec,cutoff_mask);
 275             velecsum         = _mm_add_ps(velecsum,velec);
 276             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 277             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 278
 279             fscal            = _mm_add_ps(felec,fvdw);
 280
 281             fscal            = _mm_and_ps(fscal,cutoff_mask);
 282
 283              /* Update vectorial force */
 284             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 285             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 286             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 287
 288             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 289             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 290             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 291
 292             }
 293
 294             /**************************
 295              * CALCULATE INTERACTIONS *
 296              **************************/
 297
 298             if (gmx_mm_any_lt(rsq10,rcutoff2))
 299             {
 300
 301             /* Compute parameters for interactions between i and j atoms */
 302             qq10             = _mm_mul_ps(iq1,jq0);
 303
 304             /* REACTION-FIELD ELECTROSTATICS */
 305             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 306             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 307
 308             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 309
 310             /* Update potential sum for this i atom from the interaction with this j atom. */
 311             velec            = _mm_and_ps(velec,cutoff_mask);
 312             velecsum         = _mm_add_ps(velecsum,velec);
 313
 314             fscal            = felec;
 315
 316             fscal            = _mm_and_ps(fscal,cutoff_mask);
 317
 318              /* Update vectorial force */
 319             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 320             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 321             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 322
 323             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 324             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 325             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 326
 327             }
 328
 329             /**************************
 330              * CALCULATE INTERACTIONS *
 331              **************************/
 332
 333             if (gmx_mm_any_lt(rsq20,rcutoff2))
 334             {
 335
 336             /* Compute parameters for interactions between i and j atoms */
 337             qq20             = _mm_mul_ps(iq2,jq0);
 338
 339             /* REACTION-FIELD ELECTROSTATICS */
 340             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 341             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 342
 343             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 344
 345             /* Update potential sum for this i atom from the interaction with this j atom. */
 346             velec            = _mm_and_ps(velec,cutoff_mask);
 347             velecsum         = _mm_add_ps(velecsum,velec);
 348
 349             fscal            = felec;
 350
 351             fscal            = _mm_and_ps(fscal,cutoff_mask);
 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             }
 363
 364             fjptrA             = f+j_coord_offsetA;
 365             fjptrB             = f+j_coord_offsetB;
 366             fjptrC             = f+j_coord_offsetC;
 367             fjptrD             = f+j_coord_offsetD;
 368
 369             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 370
 371             /* Inner loop uses 135 flops */
 372         }
 373
 374         if(jidx<j_index_end)
 375         {
 376
 377             /* Get j neighbor index, and coordinate index */
 378             jnrlistA         = jjnr[jidx];
 379             jnrlistB         = jjnr[jidx+1];
 380             jnrlistC         = jjnr[jidx+2];
 381             jnrlistD         = jjnr[jidx+3];
 382             /* Sign of each element will be negative for non-real atoms.
 383              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 384              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 385              */
 386             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 387             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 388             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 389             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 390             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 391             j_coord_offsetA  = DIM*jnrA;
 392             j_coord_offsetB  = DIM*jnrB;
 393             j_coord_offsetC  = DIM*jnrC;
 394             j_coord_offsetD  = DIM*jnrD;
 395
 396             /* load j atom coordinates */
 397             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 398                                               x+j_coord_offsetC,x+j_coord_offsetD,
 399                                               &jx0,&jy0,&jz0);
 400
 401             /* Calculate displacement vector */
 402             dx00             = _mm_sub_ps(ix0,jx0);
 403             dy00             = _mm_sub_ps(iy0,jy0);
 404             dz00             = _mm_sub_ps(iz0,jz0);
 405             dx10             = _mm_sub_ps(ix1,jx0);
 406             dy10             = _mm_sub_ps(iy1,jy0);
 407             dz10             = _mm_sub_ps(iz1,jz0);
 408             dx20             = _mm_sub_ps(ix2,jx0);
 409             dy20             = _mm_sub_ps(iy2,jy0);
 410             dz20             = _mm_sub_ps(iz2,jz0);
 411
 412             /* Calculate squared distance and things based on it */
 413             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 414             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 415             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 416
 417             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 418             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 419             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 420
 421             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 422             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 423             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 424
 425             /* Load parameters for j particles */
 426             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 427                                                               charge+jnrC+0,charge+jnrD+0);
 428             vdwjidx0A        = 2*vdwtype[jnrA+0];
 429             vdwjidx0B        = 2*vdwtype[jnrB+0];
 430             vdwjidx0C        = 2*vdwtype[jnrC+0];
 431             vdwjidx0D        = 2*vdwtype[jnrD+0];
 432
 433             fjx0             = _mm_setzero_ps();
 434             fjy0             = _mm_setzero_ps();
 435             fjz0             = _mm_setzero_ps();
 436
 437             /**************************
 438              * CALCULATE INTERACTIONS *
 439              **************************/
 440
 441             if (gmx_mm_any_lt(rsq00,rcutoff2))
 442             {
 443
 444             /* Compute parameters for interactions between i and j atoms */
 445             qq00             = _mm_mul_ps(iq0,jq0);
 446             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 447                                          vdwparam+vdwioffset0+vdwjidx0B,
 448                                          vdwparam+vdwioffset0+vdwjidx0C,
 449                                          vdwparam+vdwioffset0+vdwjidx0D,
 450                                          &c6_00,&c12_00);
 451
 452             /* REACTION-FIELD ELECTROSTATICS */
 453             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 454             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 455
 456             /* LENNARD-JONES DISPERSION/REPULSION */
 457
 458             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 459             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 460             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 461             vvdw             = _mm_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 462                                           _mm_mul_ps( _mm_nmacc_ps(c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 463             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 464
 465             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 466
 467             /* Update potential sum for this i atom from the interaction with this j atom. */
 468             velec            = _mm_and_ps(velec,cutoff_mask);
 469             velec            = _mm_andnot_ps(dummy_mask,velec);
 470             velecsum         = _mm_add_ps(velecsum,velec);
 471             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 472             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 473             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 474
 475             fscal            = _mm_add_ps(felec,fvdw);
 476
 477             fscal            = _mm_and_ps(fscal,cutoff_mask);
 478
 479             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 480
 481              /* Update vectorial force */
 482             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 483             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 484             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 485
 486             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 487             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 488             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 489
 490             }
 491
 492             /**************************
 493              * CALCULATE INTERACTIONS *
 494              **************************/
 495
 496             if (gmx_mm_any_lt(rsq10,rcutoff2))
 497             {
 498
 499             /* Compute parameters for interactions between i and j atoms */
 500             qq10             = _mm_mul_ps(iq1,jq0);
 501
 502             /* REACTION-FIELD ELECTROSTATICS */
 503             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 504             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 505
 506             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 507
 508             /* Update potential sum for this i atom from the interaction with this j atom. */
 509             velec            = _mm_and_ps(velec,cutoff_mask);
 510             velec            = _mm_andnot_ps(dummy_mask,velec);
 511             velecsum         = _mm_add_ps(velecsum,velec);
 512
 513             fscal            = felec;
 514
 515             fscal            = _mm_and_ps(fscal,cutoff_mask);
 516
 517             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 518
 519              /* Update vectorial force */
 520             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 521             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 522             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 523
 524             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 525             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 526             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 527
 528             }
 529
 530             /**************************
 531              * CALCULATE INTERACTIONS *
 532              **************************/
 533
 534             if (gmx_mm_any_lt(rsq20,rcutoff2))
 535             {
 536
 537             /* Compute parameters for interactions between i and j atoms */
 538             qq20             = _mm_mul_ps(iq2,jq0);
 539
 540             /* REACTION-FIELD ELECTROSTATICS */
 541             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 542             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 543
 544             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 545
 546             /* Update potential sum for this i atom from the interaction with this j atom. */
 547             velec            = _mm_and_ps(velec,cutoff_mask);
 548             velec            = _mm_andnot_ps(dummy_mask,velec);
 549             velecsum         = _mm_add_ps(velecsum,velec);
 550
 551             fscal            = felec;
 552
 553             fscal            = _mm_and_ps(fscal,cutoff_mask);
 554
 555             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 556
 557              /* Update vectorial force */
 558             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 559             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 560             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 561
 562             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 563             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 564             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 565
 566             }
 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 135 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*135);
 600 }
 601 /*
 602  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_avx_128_fma_single
 603  * Electrostatics interaction: ReactionField
 604  * VdW interaction:            LennardJones
 605  * Geometry:                   Water3-Particle
 606  * Calculate force/pot:        Force
 607  */
 608 void
 609 nb_kernel_ElecRFCut_VdwLJSh_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     __m128           dummy_mask,cutoff_mask;
 654     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 655     __m128           one     = _mm_set1_ps(1.0);
 656     __m128           two     = _mm_set1_ps(2.0);
 657     x                = xx[0];
 658     f                = ff[0];
 659
 660     nri              = nlist->nri;
 661     iinr             = nlist->iinr;
 662     jindex           = nlist->jindex;
 663     jjnr             = nlist->jjnr;
 664     shiftidx         = nlist->shift;
 665     gid              = nlist->gid;
 666     shiftvec         = fr->shift_vec[0];
 667     fshift           = fr->fshift[0];
 668     facel            = _mm_set1_ps(fr->epsfac);
 669     charge           = mdatoms->chargeA;
 670     krf              = _mm_set1_ps(fr->ic->k_rf);
 671     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 672     crf              = _mm_set1_ps(fr->ic->c_rf);
 673     nvdwtype         = fr->ntype;
 674     vdwparam         = fr->nbfp;
 675     vdwtype          = mdatoms->typeA;
 676
 677     /* Setup water-specific parameters */
 678     inr              = nlist->iinr[0];
 679     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 680     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 681     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 682     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 683
 684     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 685     rcutoff_scalar   = fr->rcoulomb;
 686     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 687     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 688
 689     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 690     rvdw             = _mm_set1_ps(fr->rvdw);
 691
 692     /* Avoid stupid compiler warnings */
 693     jnrA = jnrB = jnrC = jnrD = 0;
 694     j_coord_offsetA = 0;
 695     j_coord_offsetB = 0;
 696     j_coord_offsetC = 0;
 697     j_coord_offsetD = 0;
 698
 699     outeriter        = 0;
 700     inneriter        = 0;
 701
 702     for(iidx=0;iidx<4*DIM;iidx++)
 703     {
 704         scratch[iidx] = 0.0;
 705     }
 706
 707     /* Start outer loop over neighborlists */
 708     for(iidx=0; iidx<nri; iidx++)
 709     {
 710         /* Load shift vector for this list */
 711         i_shift_offset   = DIM*shiftidx[iidx];
 712
 713         /* Load limits for loop over neighbors */
 714         j_index_start    = jindex[iidx];
 715         j_index_end      = jindex[iidx+1];
 716
 717         /* Get outer coordinate index */
 718         inr              = iinr[iidx];
 719         i_coord_offset   = DIM*inr;
 720
 721         /* Load i particle coords and add shift vector */
 722         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 723                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 724
 725         fix0             = _mm_setzero_ps();
 726         fiy0             = _mm_setzero_ps();
 727         fiz0             = _mm_setzero_ps();
 728         fix1             = _mm_setzero_ps();
 729         fiy1             = _mm_setzero_ps();
 730         fiz1             = _mm_setzero_ps();
 731         fix2             = _mm_setzero_ps();
 732         fiy2             = _mm_setzero_ps();
 733         fiz2             = _mm_setzero_ps();
 734
 735         /* Start inner kernel loop */
 736         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 737         {
 738
 739             /* Get j neighbor index, and coordinate index */
 740             jnrA             = jjnr[jidx];
 741             jnrB             = jjnr[jidx+1];
 742             jnrC             = jjnr[jidx+2];
 743             jnrD             = jjnr[jidx+3];
 744             j_coord_offsetA  = DIM*jnrA;
 745             j_coord_offsetB  = DIM*jnrB;
 746             j_coord_offsetC  = DIM*jnrC;
 747             j_coord_offsetD  = DIM*jnrD;
 748
 749             /* load j atom coordinates */
 750             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 751                                               x+j_coord_offsetC,x+j_coord_offsetD,
 752                                               &jx0,&jy0,&jz0);
 753
 754             /* Calculate displacement vector */
 755             dx00             = _mm_sub_ps(ix0,jx0);
 756             dy00             = _mm_sub_ps(iy0,jy0);
 757             dz00             = _mm_sub_ps(iz0,jz0);
 758             dx10             = _mm_sub_ps(ix1,jx0);
 759             dy10             = _mm_sub_ps(iy1,jy0);
 760             dz10             = _mm_sub_ps(iz1,jz0);
 761             dx20             = _mm_sub_ps(ix2,jx0);
 762             dy20             = _mm_sub_ps(iy2,jy0);
 763             dz20             = _mm_sub_ps(iz2,jz0);
 764
 765             /* Calculate squared distance and things based on it */
 766             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 767             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 768             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 769
 770             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 771             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 772             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 773
 774             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 775             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 776             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 777
 778             /* Load parameters for j particles */
 779             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 780                                                               charge+jnrC+0,charge+jnrD+0);
 781             vdwjidx0A        = 2*vdwtype[jnrA+0];
 782             vdwjidx0B        = 2*vdwtype[jnrB+0];
 783             vdwjidx0C        = 2*vdwtype[jnrC+0];
 784             vdwjidx0D        = 2*vdwtype[jnrD+0];
 785
 786             fjx0             = _mm_setzero_ps();
 787             fjy0             = _mm_setzero_ps();
 788             fjz0             = _mm_setzero_ps();
 789
 790             /**************************
 791              * CALCULATE INTERACTIONS *
 792              **************************/
 793
 794             if (gmx_mm_any_lt(rsq00,rcutoff2))
 795             {
 796
 797             /* Compute parameters for interactions between i and j atoms */
 798             qq00             = _mm_mul_ps(iq0,jq0);
 799             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 800                                          vdwparam+vdwioffset0+vdwjidx0B,
 801                                          vdwparam+vdwioffset0+vdwjidx0C,
 802                                          vdwparam+vdwioffset0+vdwjidx0D,
 803                                          &c6_00,&c12_00);
 804
 805             /* REACTION-FIELD ELECTROSTATICS */
 806             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 807
 808             /* LENNARD-JONES DISPERSION/REPULSION */
 809
 810             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 811             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 812
 813             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 814
 815             fscal            = _mm_add_ps(felec,fvdw);
 816
 817             fscal            = _mm_and_ps(fscal,cutoff_mask);
 818
 819              /* Update vectorial force */
 820             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 821             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 822             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 823
 824             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 825             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 826             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 827
 828             }
 829
 830             /**************************
 831              * CALCULATE INTERACTIONS *
 832              **************************/
 833
 834             if (gmx_mm_any_lt(rsq10,rcutoff2))
 835             {
 836
 837             /* Compute parameters for interactions between i and j atoms */
 838             qq10             = _mm_mul_ps(iq1,jq0);
 839
 840             /* REACTION-FIELD ELECTROSTATICS */
 841             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 842
 843             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 844
 845             fscal            = felec;
 846
 847             fscal            = _mm_and_ps(fscal,cutoff_mask);
 848
 849              /* Update vectorial force */
 850             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 851             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 852             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 853
 854             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 855             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 856             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 857
 858             }
 859
 860             /**************************
 861              * CALCULATE INTERACTIONS *
 862              **************************/
 863
 864             if (gmx_mm_any_lt(rsq20,rcutoff2))
 865             {
 866
 867             /* Compute parameters for interactions between i and j atoms */
 868             qq20             = _mm_mul_ps(iq2,jq0);
 869
 870             /* REACTION-FIELD ELECTROSTATICS */
 871             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 872
 873             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 874
 875             fscal            = felec;
 876
 877             fscal            = _mm_and_ps(fscal,cutoff_mask);
 878
 879              /* Update vectorial force */
 880             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 881             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 882             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 883
 884             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 885             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 886             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 887
 888             }
 889
 890             fjptrA             = f+j_coord_offsetA;
 891             fjptrB             = f+j_coord_offsetB;
 892             fjptrC             = f+j_coord_offsetC;
 893             fjptrD             = f+j_coord_offsetD;
 894
 895             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 896
 897             /* Inner loop uses 106 flops */
 898         }
 899
 900         if(jidx<j_index_end)
 901         {
 902
 903             /* Get j neighbor index, and coordinate index */
 904             jnrlistA         = jjnr[jidx];
 905             jnrlistB         = jjnr[jidx+1];
 906             jnrlistC         = jjnr[jidx+2];
 907             jnrlistD         = jjnr[jidx+3];
 908             /* Sign of each element will be negative for non-real atoms.
 909              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 910              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 911              */
 912             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 913             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 914             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 915             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 916             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 917             j_coord_offsetA  = DIM*jnrA;
 918             j_coord_offsetB  = DIM*jnrB;
 919             j_coord_offsetC  = DIM*jnrC;
 920             j_coord_offsetD  = DIM*jnrD;
 921
 922             /* load j atom coordinates */
 923             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 924                                               x+j_coord_offsetC,x+j_coord_offsetD,
 925                                               &jx0,&jy0,&jz0);
 926
 927             /* Calculate displacement vector */
 928             dx00             = _mm_sub_ps(ix0,jx0);
 929             dy00             = _mm_sub_ps(iy0,jy0);
 930             dz00             = _mm_sub_ps(iz0,jz0);
 931             dx10             = _mm_sub_ps(ix1,jx0);
 932             dy10             = _mm_sub_ps(iy1,jy0);
 933             dz10             = _mm_sub_ps(iz1,jz0);
 934             dx20             = _mm_sub_ps(ix2,jx0);
 935             dy20             = _mm_sub_ps(iy2,jy0);
 936             dz20             = _mm_sub_ps(iz2,jz0);
 937
 938             /* Calculate squared distance and things based on it */
 939             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 940             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 941             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 942
 943             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 944             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 945             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 946
 947             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 948             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 949             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 950
 951             /* Load parameters for j particles */
 952             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 953                                                               charge+jnrC+0,charge+jnrD+0);
 954             vdwjidx0A        = 2*vdwtype[jnrA+0];
 955             vdwjidx0B        = 2*vdwtype[jnrB+0];
 956             vdwjidx0C        = 2*vdwtype[jnrC+0];
 957             vdwjidx0D        = 2*vdwtype[jnrD+0];
 958
 959             fjx0             = _mm_setzero_ps();
 960             fjy0             = _mm_setzero_ps();
 961             fjz0             = _mm_setzero_ps();
 962
 963             /**************************
 964              * CALCULATE INTERACTIONS *
 965              **************************/
 966
 967             if (gmx_mm_any_lt(rsq00,rcutoff2))
 968             {
 969
 970             /* Compute parameters for interactions between i and j atoms */
 971             qq00             = _mm_mul_ps(iq0,jq0);
 972             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 973                                          vdwparam+vdwioffset0+vdwjidx0B,
 974                                          vdwparam+vdwioffset0+vdwjidx0C,
 975                                          vdwparam+vdwioffset0+vdwjidx0D,
 976                                          &c6_00,&c12_00);
 977
 978             /* REACTION-FIELD ELECTROSTATICS */
 979             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 980
 981             /* LENNARD-JONES DISPERSION/REPULSION */
 982
 983             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 984             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 985
 986             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 987
 988             fscal            = _mm_add_ps(felec,fvdw);
 989
 990             fscal            = _mm_and_ps(fscal,cutoff_mask);
 991
 992             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 993
 994              /* Update vectorial force */
 995             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 996             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 997             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 998
 999             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1000             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1001             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1002
1003             }
1004
1005             /**************************
1006              * CALCULATE INTERACTIONS *
1007              **************************/
1008
1009             if (gmx_mm_any_lt(rsq10,rcutoff2))
1010             {
1011
1012             /* Compute parameters for interactions between i and j atoms */
1013             qq10             = _mm_mul_ps(iq1,jq0);
1014
1015             /* REACTION-FIELD ELECTROSTATICS */
1016             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
1017
1018             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1019
1020             fscal            = felec;
1021
1022             fscal            = _mm_and_ps(fscal,cutoff_mask);
1023
1024             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1025
1026              /* Update vectorial force */
1027             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1028             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1029             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1030
1031             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1032             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1033             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1034
1035             }
1036
1037             /**************************
1038              * CALCULATE INTERACTIONS *
1039              **************************/
1040
1041             if (gmx_mm_any_lt(rsq20,rcutoff2))
1042             {
1043
1044             /* Compute parameters for interactions between i and j atoms */
1045             qq20             = _mm_mul_ps(iq2,jq0);
1046
1047             /* REACTION-FIELD ELECTROSTATICS */
1048             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1049
1050             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1051
1052             fscal            = felec;
1053
1054             fscal            = _mm_and_ps(fscal,cutoff_mask);
1055
1056             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1057
1058              /* Update vectorial force */
1059             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1060             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1061             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1062
1063             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1064             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1065             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1066
1067             }
1068
1069             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1070             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1071             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1072             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1073
1074             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1075
1076             /* Inner loop uses 106 flops */
1077         }
1078
1079         /* End of innermost loop */
1080
1081         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1082                                               f+i_coord_offset,fshift+i_shift_offset);
1083
1084         /* Increment number of inner iterations */
1085         inneriter                  += j_index_end - j_index_start;
1086
1087         /* Outer loop uses 18 flops */
1088     }
1089
1090     /* Increment number of outer iterations */
1091     outeriter        += nri;
1092
1093     /* Update outer/inner flops */
1094
1095     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*106);
1096 }